Applied numerical modeling of saturated / unsaturated flow and ...
Applied numerical modeling of saturated / unsaturated flow and ...
Applied numerical modeling of saturated / unsaturated flow and ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
<strong>Applied</strong> <strong>numerical</strong> <strong>modeling</strong> <strong>of</strong> <strong>saturated</strong> /<br />
un<strong>saturated</strong> <strong>flow</strong> <strong>and</strong> reactive contaminant transport<br />
-<br />
evaluation <strong>of</strong> site investigation strategies<br />
<strong>and</strong> assessment <strong>of</strong> environmental impact<br />
Dissertation<br />
zur Erlangung des Grades eines Doktors der Naturwissenschaften<br />
der Geowissenschaftlichen Fakultät<br />
der Eberhard-Karls-Universität Tübingen<br />
vorgelegt von<br />
Christ<strong>of</strong> Beyer<br />
aus Braunschweig<br />
2007<br />
1
2<br />
Tag der mündlichen Prüfung: 23.02.2007<br />
Dekan: Pr<strong>of</strong>. Dr. Peter Grathwohl<br />
1. Berichterstatter: Pr<strong>of</strong>. Dr.-Ing. Olaf Kolditz<br />
2. Berichterstatter: Priv. Doz. Dr. rer. nat. Sebastian Bauer<br />
3. Berichterstatter: Pr<strong>of</strong>. James F. Barker
Abstract<br />
<strong>Applied</strong> <strong>numerical</strong> <strong>modeling</strong> <strong>of</strong> <strong>saturated</strong> /<br />
un<strong>saturated</strong> <strong>flow</strong> <strong>and</strong> reactive contaminant transport -<br />
evaluation <strong>of</strong> site investigation strategies<br />
<strong>and</strong> assessment <strong>of</strong> environmental impact<br />
CHRISTOF BEYER<br />
In this thesis <strong>numerical</strong> models <strong>of</strong> variably <strong>saturated</strong> <strong>flow</strong> <strong>and</strong> reactive transport<br />
processes in porous media are employed as assessment tools in two different fields <strong>of</strong><br />
application.<br />
The first thematic complex studied is the computer based evaluation <strong>of</strong> investigation<br />
strategies for contaminated soils <strong>and</strong> aquifers. For this purpose the “Virtual Aquifer”<br />
(VA) concept is introduced <strong>and</strong> demonstrated by an assessment <strong>of</strong> the so called center<br />
line method for contaminant plume investigation. Errors <strong>and</strong> uncertainties in the<br />
estimation <strong>of</strong> contaminant degradation rates from center line data, which is <strong>of</strong>ten<br />
collected at field sites, are analysed. This is done by application at synthetic<br />
contaminated aquifers models, which are generated in the computer by <strong>numerical</strong><br />
simulation <strong>of</strong> contaminant spreading. Monte Carlo simulations <strong>and</strong> sensitivity studies<br />
are performed to quantify the influences <strong>of</strong> sampling error magnitude, aquifer<br />
heterogeneity <strong>and</strong> model parameterisation on the estimated rate constants. In a second<br />
application example, the VA concept is used for the development <strong>and</strong> testing <strong>of</strong> a new<br />
approach for biodegradation parameter estimation. The performance <strong>of</strong> this method is<br />
studied in heterogeneous synthetic aquifers. Also, the propagation <strong>of</strong> errors <strong>and</strong><br />
uncertainties from estimated rate parameters to a prognosis <strong>of</strong> the contaminant plume<br />
lengths is studied to obtain an indicator for the significance <strong>of</strong> the estimated degradation<br />
potential. The performance <strong>of</strong> the new parameter estimation method is assessed by<br />
comparison to frequently used approximations by first order kinetics.<br />
The second thematic complex addressed in this thesis is the evaluation <strong>of</strong> environmental<br />
impact <strong>of</strong> contaminant emissions from road constructions by type scenario modelling.<br />
Two general concepts for modelling <strong>of</strong> <strong>flow</strong> <strong>and</strong> transport, i.e. the Eulerian <strong>and</strong> the<br />
Lagrangian points <strong>of</strong> view, are combined in this study to assess the extent <strong>of</strong> leaching<br />
from contaminated demolition waste within road structures to the groundwater surface.<br />
Transport simulations are performed for a number <strong>of</strong> typical subsoil units <strong>of</strong> Germany<br />
to analyse the sensitivity <strong>of</strong> contaminant transport behaviour on subsoil properties. The<br />
relevant contaminant transport <strong>and</strong> attenuation processes in road constructions <strong>and</strong> the<br />
different subsoils are identified. The study allows to draw important conclusions on how<br />
these mechanisms could be used or enhanced to reduce contaminant leaching to groundwater.<br />
3
Angew<strong>and</strong>te numerische Modellierung der gesättigten / ungesättigten<br />
Strömung und des reaktiven Schadst<strong>of</strong>ftransports-<br />
Evaluierung von Strategien der St<strong>and</strong>orterkundung<br />
und Umweltwirkungsprognose<br />
Kurzfassung<br />
CHRISTOF BEYER<br />
Diese Arbeit stellt die Verwendung numerischer Modelle gesättigt / ungesättigter<br />
Strömungs- und reaktiver St<strong>of</strong>ftransportprozesse in porösen Medien als<br />
Bewertungsinstrument in zwei verschiedenen Anwendungsfeldern vor.<br />
Der erste hier betrachtete Themenkomplex beh<strong>and</strong>elt die computerbasierte Evaluierung<br />
von Erkundungsstrategien für kontaminierte Böden und Grundwasserleiter. Hierzu wird<br />
das Konzept der „Virtuellen Aquifere“ (VA) eingeführt und anh<strong>and</strong> einer Bewertung<br />
der sogenannten Center-Line Methodik zur Schadst<strong>of</strong>ffahnenerkundung demonstriert.<br />
Die Center-Line Methode wird in der Praxis häufig zur Erhebung von Daten<br />
angewendet, auf deren Grundlage Schadst<strong>of</strong>f-Abbauratenkonstanten abgeschätzt<br />
werden können. Die Bewertung dieser Vorgehensweise wird durch eine Analyse der<br />
dabei auftretenden Fehler und Unsicherheiten vorgenommen, indem die Methode bei<br />
synthetischen kontaminierten Aquifermodellen angewendet wird, die am Computer<br />
durch numerische Simulation der Schadst<strong>of</strong>fausbreitung generierten wurden. Durch<br />
Monte-Carlo Simulationen und Sensitivitätsanalysen werden die Einflüsse von<br />
Messfehlern, Aquifer-Heterogenität und Modellparametrisierung quantifiziert. In einem<br />
zweiten Anwendungsbeispiel wird die VA-Methodik zur Ableitung und Prüfung eines<br />
neuen Ansatzes zur Schätzung von Parametern mikrobieller Abbaukinetiken<br />
angewendet. Die Eignung des Verfahrens wird in synthetischen heterogenen Aquiferen<br />
geprüft. Darüber hinaus wird die Fehler- und Unsicherheitspropagation von den<br />
geschätzten Abbauparametern zu mit diesen prognostizierte Schadst<strong>of</strong>ffahnenlängen<br />
untersucht, um Indikatoren für die Aussagekraft der Parameter zu erhalten. Das<br />
Verfahren wird durch einen Vergleich mit häufig angewendeten Näherungen durch<br />
Kinetiken erster Ordnung bewertet.<br />
Der zweite im Rahmen dieser Arbeit beh<strong>and</strong>elte Themenkomplex ist die Bewertung der<br />
Umweltauswirkungen von Schadst<strong>of</strong>fausträgen aus Straßenbauten durch Typszenarienmodellierung.<br />
Hier werden zwei allgemeine Konzepte zur Modellierung von Strömung<br />
und Transport, der Eulerische und der Larangesche Ansatz, mitein<strong>and</strong>er kombiniert.<br />
Das Ausmaß des Schadst<strong>of</strong>feintrags aus belastetem im Straßenbau eingesetztem<br />
Recycling-Bauschutt ins Grundwasser wird durch Einsatz der mit ein<strong>and</strong>er gekoppelten<br />
numerischen Modelle GeoSys/Rock<strong>flow</strong> und SMART bewertet. Die Transportsimulationen<br />
werden für eine Reihe typischer Unterbodeneinheiten Deutschl<strong>and</strong>s<br />
durchgeführt, um die Sensitivität des Schadst<strong>of</strong>ftransportverhaltens auf die<br />
Unterbodeneigenschaften zu untersuchen. Die relevanten Schadst<strong>of</strong>ftransport- und<br />
Attenuierungsprozesse im Straßenaufbau und den Unterböden werden identifiziert. Die<br />
Studie ermöglicht wichtige Schlüsse im Hinblick eine Nutzung und Förderung dieser<br />
Mechanismen zu einer weiteren Reduktion der Schadst<strong>of</strong>feinträge ins Grundwasser.<br />
5
Vorwort<br />
Die zurückliegenden drei Jahre am ZAG in Tübingen waren für mich eine sehr spannende Zeit<br />
und nicht nur im Hinblick auf meine Promotion sehr lehrreich. Aus diesem Grund möchte ich<br />
hier die Gelegenheit nutzen, ein paar persönliche Worte zu verlieren und einigen Personen zu<br />
danken, ohne deren Beteiligung diese Arbeit so sicher nicht hätte entstehen können.<br />
Die im Rahmen dieser Arbeit vorgestellten Ergebnisse wurden innerhalb der beiden Projekte<br />
“Virtueller Aquifer” (Förderkennzeichen 033 05 12/033 05 13) und “Übertragung der Ergebnisse<br />
des BMBF - Förderschwerpunktes „Sickerwasserprognose“ auf repräsentative Fallbeispiele”<br />
(Förderkennzeichen 02WP0517) erarbeitet. Beide Projekte wurden durch das Bundesministerium<br />
für Bildung und Forschung (BMBF) finanziell gefördert, w<strong>of</strong>ür diesem hiermit<br />
gedankt sei.<br />
Besonders bedanken möchte ich mich bei den Betreuern dieser Arbeit. Herrn Pr<strong>of</strong>. Dr.-Ing. Olaf<br />
Kolditz danke ich sehr für die Ermöglichung meiner Promotion, das in mich gesetztes Vertrauen<br />
und die stete Unterstützung. Seine immer optimistische Sicht der Dinge hat sich im Laufe der<br />
Zeit auch ein Wenig auf mich übertragen. Herr Dr. Sebastian Bauer ist mit Sicherheit die<br />
Person, die die Richtung meiner Arbeit am stärksten mitgeprägt hat. Sein aufrichtiges Interesse<br />
(auch abseits der fachlichen Dinge), viele ausführliche Diskussionen und immer konstruktive<br />
Kritik haben maßgeblich zu ihrem Gelingen beigetragen. Eine bessere Betreuung für seine<br />
Doktorarbeit kann man sich eigentlich kaum wünschen.<br />
Herrn Pr<strong>of</strong>. Dr. James F. Barker danke ich für die Erstellung des dritten Gutachtens dieser<br />
Arbeit.<br />
Herzlich danken möchte ich Herrn Pr<strong>of</strong>. Dr. Peter Grathwohl und Herrn Pr<strong>of</strong>. Dr. Rudolf Liedl,<br />
die mir die Bearbeitung des Sickerwasserprognose-Projekts anvertrauten und mir so zu einem in<br />
jeder Hinsicht spannenden letzten Dreivierteljahr verhalfen.<br />
Ein herzlicher Dank geht an meine Kollegen Dr. Cui Chen, Dipl.-Geol. Jan Gronewold, Dr.<br />
Wilfried Konrad, Dr. Thomas Kalbacher, Dr. Chan Hee Park, Dr. Wenqing Wang, Dr. Chris<br />
McDermott, Dr. Martin Beinhorn, Robert Walsh, M.Sc., Dr. Dirk Schäfer von der Uni Kiel, Dr.<br />
Hermann Rügner und Dr. Peter Dietrich vom UFZ in Leipzig, sowie an zahlreiche weitere<br />
Kollegen für angenehm reibungslose Zusammenarbeit, viele hilfreiche Diskussionen und<br />
gemeinsame Pausen vom Zahlen hin und her schubsen.<br />
Mein größter Dank gilt meinen Eltern, da Ihr mir durch Eure Unterstützung diesen Weg erst<br />
ermöglicht habt, und Frauke für Deine große Geduld, Dein Verständnis, Deine Liebe und dafür,<br />
dass Du es trotz aller Umstände und Entfernungen gemeinsam mit mir bis hierhin geschafft<br />
hast.<br />
7
Table <strong>of</strong> Contents<br />
List <strong>of</strong> abbreviations <strong>and</strong> mathematical symbols<br />
1. Introduction 1<br />
2. Mathematical models 2<br />
2.1. Saturated / un<strong>saturated</strong> <strong>flow</strong> 2<br />
2.2. Transport processes 3<br />
2.3. Reactive processes 4<br />
2.4. Numerics <strong>and</strong> s<strong>of</strong>tware methods 6<br />
3. Modeling applications 9<br />
3.1. Evaluation <strong>of</strong> investigation strategies for contaminated aquifers using<br />
the Virtual Aquifer concept 9<br />
3.2. Development <strong>and</strong> testing <strong>of</strong> a new approach to estimating biodegradation<br />
parameters from field data 16<br />
3.3. Prognosis <strong>of</strong> long term contaminant leaching from recycling materials in<br />
road constructions 18<br />
4. Conclusions <strong>and</strong> outlook 21<br />
References<br />
Enclosed Publications<br />
23<br />
9
List <strong>of</strong> abbreviations <strong>and</strong> mathematical symbols<br />
a empirical sorption constant [-]<br />
b empirical sorption constant [-]<br />
C concentration [M L -3 ]<br />
C0 equilibrium concentration [M L -3 ]<br />
Cl liquid phase concentration [M L -3 ]<br />
Cs solid phase concentration [M M -1 ]<br />
Corg soil organic carbon content [%]<br />
Cw(�) water capacity function<br />
D tensor <strong>of</strong> hydrodynamic dispersion<br />
[L² T -1 ]<br />
Da aqueous molecular diffusion<br />
coefficient [L² T -1 ]<br />
Dae effective diffusion coefficient [L² T -1 ]<br />
Dap apparent diffusion coefficient [L² T -1 ]<br />
De tensor <strong>of</strong> effective hydrodynamic<br />
dispersion [L² T -1 ]<br />
Dm tensor <strong>of</strong> mechanical dispersion<br />
[L² T -1 ]<br />
DW demolition waste<br />
FE finite element<br />
FEM finite element method<br />
g travel time probability density<br />
function<br />
h hydraulic head [L]<br />
h´ erroneous head measurement [L]<br />
I identity tensor<br />
IC inhibition concentration [M L -3 ]<br />
K tensor <strong>of</strong> hydraulic conductivity<br />
[L T -1 ]<br />
kmax maximum degradation rate<br />
[M L -3 T -1 ]<br />
l pore connectivity parameter [-]<br />
m empirical Van Genuchten<br />
parameter [-]<br />
MC half-saturation concentration [M L -3 ]<br />
MM Michaelis-Menten<br />
n empirical Van Genuchten<br />
parameter [-]<br />
OOP object-oriented programming<br />
PDE partial differential equation<br />
pdf probability density function<br />
Q source or sink term [M L -3 T -1 ]<br />
r radial distance [L]<br />
REV representative elementary volume<br />
S specific storativity [L -1 ]<br />
Sr relative saturation [-]<br />
t time [T]<br />
V magnitude <strong>of</strong> velocity vector<br />
[L T -1 ]<br />
vmax maximum growth rate [T -1 ]<br />
v average linear velocity [L T -1 ]<br />
v vector <strong>of</strong> average linear velocity<br />
[L T -1 ]<br />
VA Virtual Aquifer<br />
X microbial population [M L -3 ]<br />
Y yield coefficient [-]<br />
z elevation [L]<br />
� empirical Van Genuchten<br />
parameter [L -1 ]<br />
�L<br />
�T<br />
longitudinal dispersivity [L]<br />
transverse dispersivity [L]<br />
� reaction function<br />
� tortuosity related coefficient [-]<br />
� r<strong>and</strong>om number<br />
� intraparticle porosity [-]<br />
��h maximum measurement error [L]<br />
� empirical sorption constant [-]<br />
�e effective porosity [-]<br />
� volumetric water content [-]<br />
�r residual water content [-]<br />
<strong>saturated</strong> water content [-]<br />
�s<br />
� empirical sorption constant<br />
[M 1-� L 3� M -1 ]<br />
� first order degradation rate<br />
constant [T -1 ]<br />
� microbial decay term<br />
� density [M L -3 ]<br />
2<br />
� Y ln(K) variance<br />
� travel time [T]<br />
tortuosity factor [-]<br />
�f<br />
� matric head [L]<br />
11
<strong>Applied</strong> <strong>numerical</strong> <strong>modeling</strong> <strong>of</strong> <strong>saturated</strong> / un<strong>saturated</strong><br />
<strong>flow</strong> <strong>and</strong> reactive contaminant transport –<br />
evaluation <strong>of</strong> site investigation strategies <strong>and</strong> assessment<br />
<strong>of</strong> environmental impact<br />
1. Introduction<br />
In the field <strong>of</strong> subsurface hydrology, mathematical<br />
models are used to simulate fluid<br />
<strong>flow</strong> <strong>and</strong> solute transport by translating<br />
physical <strong>and</strong> biogeochemical processes into<br />
mathematical equations, which can be<br />
solved by either analytical or <strong>numerical</strong><br />
methods. Underst<strong>and</strong>ing <strong>of</strong> individual processes<br />
in domains <strong>of</strong> simple geometry is<br />
already a challenging task for itself. In large<br />
scale applications, however, we are faced<br />
with heterogeneous environments <strong>and</strong> interactions<br />
<strong>of</strong> many different types <strong>of</strong> spatially<br />
<strong>and</strong> temporally variable processes, which<br />
leave the <strong>numerical</strong> treatment <strong>of</strong> such complex<br />
coupled problems <strong>of</strong>ten as the only<br />
way to reach meaningful conclusions<br />
(Zheng <strong>and</strong> Bennett, 1995). As a benefit <strong>of</strong><br />
the rapid development <strong>of</strong> computational capabilities<br />
(e.g. high performance parallel<br />
computing, specialized s<strong>of</strong>tware implementation<br />
methods, data pre- <strong>and</strong> post-processing<br />
tools, graphical display routines) the<br />
<strong>numerical</strong> simulation <strong>of</strong> complex coupled<br />
problems in subsurface hydrology is continuously<br />
advanced. In general, <strong>numerical</strong><br />
models <strong>of</strong> <strong>flow</strong> <strong>and</strong> transport in geosystems<br />
are used as tools for<br />
� qualitative <strong>and</strong> quantitative analysis <strong>of</strong><br />
single or coupled processes<br />
� identification <strong>of</strong> relevant parameters<br />
� parameter estimation / inverse <strong>modeling</strong><br />
� sensitivity <strong>and</strong> uncertainty analysis<br />
� prediction <strong>of</strong> system response to changes<br />
in initial or boundary conditions<br />
These capabilities as well as the spectrum<br />
<strong>of</strong> application would not have been<br />
achieved without the ever growing dem<strong>and</strong><br />
for groundwater resources <strong>and</strong> the concern<br />
about its quality. Groundwater is one <strong>of</strong> the<br />
main drinking water supplies <strong>and</strong> increasingly<br />
used for agricultural field irrigation<br />
(Morris et al., 2003). At the same time<br />
pollution from industrial activities, waste<br />
disposal, agricultural use <strong>of</strong> fertilizers or<br />
pesticides <strong>and</strong> urban waste waters (Scheidleder<br />
et al., 1999) poses a serious threat to<br />
our groundwater resources.<br />
In this thesis <strong>numerical</strong> <strong>modeling</strong> is used in<br />
three applications within the context <strong>of</strong> contaminant<br />
hydrology. The term applied <strong>numerical</strong><br />
<strong>modeling</strong> here emphasizes the field<br />
scale application <strong>of</strong> computational methods<br />
to obtain solutions to systems <strong>of</strong> partial differential<br />
equations describing <strong>flow</strong> <strong>and</strong><br />
reactive transport process interactions in<br />
porous media. Application 1 introduces the<br />
“Virtual Aquifer” (VA) concept, in which<br />
<strong>numerical</strong> <strong>modeling</strong> is used as a tool for the<br />
evaluation <strong>of</strong> investigation <strong>and</strong> remediation<br />
strategies for contaminated soils <strong>and</strong> aquifers.<br />
The concept is demonstrated by an<br />
assessment <strong>of</strong> the so called center line method<br />
for site investigation. In application 2<br />
the VA concept is applied for the development<br />
<strong>and</strong> testing <strong>of</strong> a new approach for biodegradation<br />
parameter estimation. Both<br />
applications make use <strong>of</strong> the GeoSys /<br />
Rock<strong>flow</strong> code (Kolditz et al., 2006) for the<br />
<strong>numerical</strong> simulations. In application 3 the<br />
Eulerian <strong>and</strong> Lagrangian concepts for contaminant<br />
transport <strong>modeling</strong> are combined.<br />
GeoSys / Rock<strong>flow</strong> is coupled with the<br />
SMART model (Finkel et al., 1998) <strong>and</strong><br />
used for type scenario <strong>modeling</strong> to assess<br />
the environmental impact <strong>of</strong> recycling<br />
materials in road constructions.<br />
This synthesis is organized as follows:<br />
Chapter 2 presents the mathematical process<br />
models <strong>and</strong> <strong>numerical</strong> schemes used<br />
1
for the application studies outlined. The<br />
chapter is based on fundamental publications<br />
in the field <strong>of</strong> computational hydrology.<br />
However, it is not intended to<br />
serve as a comprehensive review <strong>of</strong> processes<br />
<strong>and</strong> model concepts, as this would be<br />
beyond the scope <strong>of</strong> this synthesis. Chapter<br />
3 <strong>and</strong> its subsections present the results<br />
<strong>and</strong> conclusions <strong>of</strong> the three application<br />
examples. Chapter 4 closes this synthesis<br />
with general conclusions <strong>and</strong> an outlook.<br />
2. Mathematical models<br />
The three-dimensional structure <strong>of</strong> natural<br />
porous media is manifested in its composition<br />
<strong>of</strong> three constituting phases, i.e. the<br />
solid (mineral or biophase), water <strong>and</strong> gaseous<br />
phases. At scales larger than the pore<br />
scale, a description <strong>of</strong> porous medium geometry<br />
becomes very complex <strong>and</strong> thus infeasible<br />
for <strong>modeling</strong> applications. To<br />
underst<strong>and</strong> <strong>and</strong> formulate the dynamics <strong>of</strong><br />
fluids in the subsurface the so called representative<br />
elementary volume (REV) concept<br />
is introduced (Bear, 1972): In the<br />
transition from the microscale to a larger<br />
macroscale material, parameters are averaged<br />
over a volume which is sufficiently<br />
large to describe the porous medium at that<br />
larger scale (see Fig. 1).<br />
Fig. 1: Representative elementary volume<br />
concept (Bear <strong>and</strong> Bachmat, 1990).<br />
This may also require a reformulation <strong>of</strong> the<br />
mathematical process descriptions. The<br />
derivation <strong>of</strong> representative or effective<br />
new material parameters <strong>and</strong> the corresponding<br />
governing equations is termed<br />
2<br />
upscaling. Within the REV the detailed<br />
structure <strong>of</strong> the medium is lost <strong>and</strong> becomes<br />
a continuous field. Parameters like porosity,<br />
permeability or dispersivity are considered<br />
constant over the averaging volume. In the<br />
following sections material parameters <strong>and</strong><br />
governing equations are based on this<br />
continuum approach.<br />
2.1. Saturated / un<strong>saturated</strong> <strong>flow</strong><br />
The dynamics <strong>of</strong> the water in fully <strong>saturated</strong><br />
three dimensional porous media can be described<br />
by the combination <strong>of</strong> the mass<br />
balance for the water phase (eq. (1)) <strong>and</strong><br />
Darcy’s law (eq. (2)) as a constitutive equation<br />
(Bear, 1972)<br />
�h<br />
S � �� �q<br />
� Q<br />
(1)<br />
�t<br />
q � �K�h<br />
(2)<br />
where S [m -1 ] is specific storativity, h [m] is<br />
the hydraulic head, given as the sum <strong>of</strong> elevation<br />
z [m] <strong>and</strong> the pressure head � [m], t<br />
[s] is time, q [m s -1 ] is the Darcy flux vector,<br />
K [m s -1 ] is the tensor <strong>of</strong> hydraulic conductivity<br />
<strong>and</strong> Q [s -1 ] is a source or sink<br />
term. The governing equation for groundwater<br />
<strong>flow</strong> under transient conditions is<br />
thus given by (Bear, 1972)<br />
�h<br />
S � � � �K �h��Q<br />
(3)<br />
�t<br />
which at steady state converts to<br />
� �h���Q<br />
� � K (4).<br />
This model <strong>of</strong> steady state <strong>flow</strong> in <strong>saturated</strong><br />
porous media is employed in applications 1<br />
<strong>and</strong> 2 (sections 3.1, 3.2) <strong>of</strong> this synthesis.<br />
For un<strong>saturated</strong> conditions, which are prevalent<br />
in application 3 (section 3.3), a more<br />
general form <strong>of</strong> eq. (2), the Buckingham-<br />
Darcy-law, can be used (Jury et al., 1991)<br />
q � �K(<br />
� ) �h<br />
(5)<br />
where K is a function <strong>of</strong> the pressure (or<br />
matric) head �, which itself depends on the<br />
volumetric water content � [-] <strong>of</strong> the porous<br />
medium. As for <strong>saturated</strong> conditions, the
Buckingham-Darcy law is combined with<br />
mass balance principles to yield the governing<br />
equation <strong>of</strong> <strong>flow</strong> for un<strong>saturated</strong><br />
conditions, i.e. the Richards equation. This<br />
equation exists in three main forms with �,<br />
� or both quantities as dependent variables<br />
(Jury et al., 1991). In GeoSys / Rock<strong>flow</strong><br />
the �-based form (Freeze <strong>and</strong> Cherry,<br />
1979) is implemented (Du et al., 2005)<br />
C w<br />
��<br />
�t<br />
��� � � � �K ( � ) ��<br />
�<br />
�K<br />
�<br />
�z<br />
��� � Q<br />
(6)<br />
where Cw(�) is the water capacity function<br />
defined by d� /d� <strong>and</strong> with z positive in a<br />
downward direction. For the functional<br />
description <strong>of</strong> un<strong>saturated</strong> hydraulic properties<br />
different mathematical formulations<br />
have been proposed in literature. A frequently<br />
used constitutive relation is the<br />
Van-Genuchten-Mualem model (Van Genuchten,<br />
1980) which is based on the<br />
statistical pore space model <strong>of</strong> Mualem<br />
(1976) <strong>and</strong> is given by<br />
� � � �2 �m<br />
m<br />
1�<br />
1�<br />
Sr<br />
l<br />
K(<br />
S ) � KS<br />
(7)<br />
Sr<br />
r<br />
� ��<br />
r � �<br />
� ��<br />
s<br />
r<br />
r<br />
1<br />
� � � �m n<br />
1�<br />
��<br />
(8)<br />
m � 1� 1/<br />
n<br />
(9)<br />
where Sr [-] is defined as the relative<br />
saturation, l [-] is a pore connectivity parameter,<br />
�, �r <strong>and</strong> �s are the actual, residual<br />
<strong>and</strong> <strong>saturated</strong> volumetric water contents, �<br />
[m -1 ], n [-], <strong>and</strong> m [-] are empirical parameters.<br />
Other constitutive relationships comprise<br />
approaches such as the Brooks-Corey<br />
model (Brooks <strong>and</strong> Corey, 1966), the Haverkamp<br />
model (Haverkamp et al., 1977), potential<br />
functions as introduced by Huyakorn <strong>and</strong><br />
Pinder (1983) or the multimodal model <strong>of</strong><br />
Durner (1994). Recently, also free form parameterizations<br />
were suggested (Bitterlich et al.,<br />
2004). An overview on the prevalent approaches<br />
is given e.g. by Durner <strong>and</strong> Flühler (2005).<br />
2.2. Transport processes<br />
The most fundamental transport process <strong>of</strong><br />
dissolved substances in <strong>saturated</strong> porous<br />
media is advection. Advection is passive<br />
with the <strong>flow</strong>ing water. Purely advective<br />
transport <strong>of</strong> a solute plume is free <strong>of</strong> interference<br />
or mixing with the surrounding<br />
ambient water <strong>and</strong> is described with the advection<br />
equation (Zheng <strong>and</strong> Bennett, 1995)<br />
�C<br />
�t<br />
� �v<br />
�C<br />
� Q<br />
(10)<br />
where C [kg m -3 ] is the concentration <strong>of</strong> a<br />
dissolved species, v [m s -1 ] is the vector <strong>of</strong><br />
average linear velocity which is given by<br />
division <strong>of</strong> q with the effective porosity �e<br />
[-], <strong>and</strong> Q [kg m -3 s -1 ] is a source or sink<br />
term for species C.<br />
In natural aquifers or soils, purely advective<br />
transport is practically not established as<br />
dissolved molecules migrate from high to<br />
low concentration regions by Brownian<br />
motion. This concentration gradient driven<br />
mass transport is termed molecular diffusion<br />
<strong>and</strong> occurs even when the fluid itself is<br />
stagnant. For transient systems the diffusion<br />
process in water can be described using<br />
Fick’s 2 nd law, (Fetter, 1993)<br />
2<br />
�C<br />
� C<br />
� �Da<br />
2<br />
�t<br />
�x<br />
(11)<br />
where Da is the molecular diffusion coefficient<br />
in water [m 2 s -1 ]. In porous media the<br />
diffusion process is hindered by the presence<br />
<strong>of</strong> the solid phase matrix <strong>and</strong> the tortous<br />
nature <strong>of</strong> the pores. Thus an effective<br />
diffusion coefficient Dae is derived as<br />
(Grathwohl, 1998)<br />
D<br />
� �<br />
e<br />
ae a<br />
f<br />
D � (12)<br />
�<br />
where � [-] is the constrictivity <strong>and</strong> �f [-] the<br />
tortuosity <strong>of</strong> the porous medium. Under<br />
un<strong>saturated</strong> conditions Dae can also be<br />
related to � (e.g. Olsen <strong>and</strong> Kemper, 1968).<br />
As water moves through a porous medium,<br />
single streamline velocities can be greater<br />
or less than v. This effect is due to different<br />
3
path lengths <strong>of</strong> water molecules that bypass<br />
mineral grains <strong>of</strong> different size <strong>and</strong> shape,<br />
different pore diameters as well as inner<br />
pore friction which results in velocity contrasts<br />
along pore cross sections. The consequential<br />
divergence <strong>of</strong> transport velocities<br />
for dissolved solutes causes mixing with the<br />
ambient water along the <strong>flow</strong> path <strong>and</strong> thus<br />
results in solute spreading longitudinally<br />
<strong>and</strong> transversally to the main <strong>flow</strong> direction.<br />
This process is termed mechanical dispersion.<br />
For its mathematical description usually<br />
an analogy to the diffusion process is<br />
assumed. According to Bear (1972) the tensor<br />
<strong>of</strong> mechanical dispersion Dm is given by<br />
D m<br />
4<br />
� � I � �<br />
��L�� V<br />
TV<br />
T<br />
vv<br />
(13)<br />
where �L <strong>and</strong> �T [m] are longitudinal <strong>and</strong><br />
transverse dispersivities, V is the magnitude<br />
<strong>of</strong> the velocity vector, I is the identity<br />
tensor <strong>and</strong> vv is the dyadic <strong>of</strong> the velocity<br />
vector. The tensor <strong>of</strong> hydrodynamic dispersion<br />
D [m 2 s -1 ] combines the dispersion <strong>and</strong><br />
diffusion processes <strong>and</strong> is calculated by<br />
D m ae<br />
� D � D I<br />
(14).<br />
The mathematical formulation <strong>of</strong> advectivedispersive<br />
transport in fully <strong>saturated</strong> porous<br />
media assuming constant porosity is<br />
given by the sum <strong>of</strong> the advective <strong>and</strong><br />
dispersive fluxes, i.e. the advection-dispersion-equation<br />
(Zheng <strong>and</strong> Bennett, 1995)<br />
�C<br />
� �� � �v C�����D�C��Q<br />
(15).<br />
�t<br />
For un<strong>saturated</strong> conditions the total solute<br />
flux in the water phase is described by<br />
��C<br />
� �� � �q�C������De�C��Q(16) �t<br />
where the effective hydrodynamic disper�<br />
sion tensor De is used, as besides Dae also<br />
�L <strong>and</strong> �T depend on � (Bear, 1979).<br />
2.3. Reactive processes<br />
The source or sink terms Q in eq. (15) <strong>and</strong><br />
(16) represent a large variety <strong>of</strong> processes<br />
other than advection or hydrodynamic dispersion,<br />
which may cause temporal changes<br />
in the solute concentration C. Hence, Q<br />
may represent transfer <strong>of</strong> species between<br />
solid, water, gaseous or biophase (e.g. volatilization,<br />
non aqueous phase liquid dissolution,<br />
sorption), or equilibrium <strong>and</strong> kinetic<br />
reactions <strong>of</strong> (geo-)chemical or biochemical<br />
nature. According to Rubin (1983) (Fig. 2)<br />
reactive processes can be classified by<br />
� reaction velocity <strong>and</strong> reversibility (equilibrium<br />
or non-equilibrium; level A)<br />
� involvement <strong>of</strong> only a single or several<br />
phases (homogeneous / heterogeneous;<br />
level B)<br />
� reaction type: surface (e.g. sorption) or<br />
“classical” chemical reaction (level C)<br />
For the sake <strong>of</strong> brevity, here only process<br />
concepts relevant for the model applications<br />
<strong>of</strong> chapter 3 are explained in more detail.<br />
Fig. 2: Classification <strong>of</strong> reactions in porous<br />
media (Rubin, 1983).<br />
Kinetic sorption<br />
Transfer <strong>of</strong> dissolved species from the<br />
mobile phase to the solid matrix by<br />
physico-chemical processes is termed<br />
sorption. Sorbed species are immobilized<br />
<strong>and</strong> not transported with the water flux.<br />
Sorption is a reversible process, i.e. sorbed<br />
species can be remobilised by desorption.<br />
Temporary immobilisation by sorption results<br />
in lowered solute concentrations <strong>and</strong><br />
retarded transport velocities. The manifold<br />
processes contributing to sorption phenomena<br />
include physical as well as chemical<br />
mechanisms (e.g. ion exchange <strong>and</strong> surface<br />
complexation through Coulomb or van der<br />
Waals forces, hydrogen-, hydrophobic- or
covalent bonding). To mathematically describe<br />
the sorption / desorption behaviour <strong>of</strong><br />
a species so called sorption isotherms can<br />
be used, in which the sorbed amount <strong>of</strong> the<br />
species is a function <strong>of</strong> its dissolved<br />
concentration. A general nonlinear sorption<br />
model can be formulated as<br />
C<br />
s<br />
� �a�Cl� � � � abC �<br />
� 1<br />
(17)<br />
l<br />
where Cs [kg kg -1 ] is the sorbed solid phase<br />
<strong>and</strong> Cl [kg m -3 ] the liquid phase concentration,<br />
�, a,� b <strong>and</strong> � are empirical constants.<br />
For b = 0, a = 1 [-] <strong>and</strong> � = 1 [-] eq.<br />
(17) is the linear Henry isotherm, in this<br />
case � [m³ kg -1 ] is a simple equilibrium<br />
constant. For � = 1 <strong>and</strong> b = 1 eq. (17) is the<br />
Langmuir isotherm with � [kg kg -1 ] being<br />
the maximum amount <strong>of</strong> a species which<br />
can be sorbed to the solid phase <strong>and</strong> a<br />
[m³ kg -1 ] an adsorption constant. For b = 0,<br />
a = 1 [-] <strong>and</strong> � � 1 [-] (usually � < 1) the<br />
Freundlich isotherm is obtained, where �<br />
[kg 1-� m 3� kg -1 ] is the Freundlich coefficient<br />
<strong>and</strong> � [-] is the Freundlich exponent.<br />
In general, sorption processes may be treated<br />
as equilibrium reactions because sorption<br />
is fast compared to transport. However,<br />
there are exceptions to this rule because for<br />
some solute species as well as soils or<br />
aquifers, equilibration is a slow process.<br />
Possible mechanisms include slow diffusion<br />
into intraparticle pores accompanied by<br />
equilibrium sorption to surfaces within the<br />
pores or slow diffusion in organic matter<br />
(Ball <strong>and</strong> Roberts, 1991; Grathwohl, 1998;<br />
Rügner et al., 1999). Hence, from a<br />
macroscopic point <strong>of</strong> view sorption equilibrium<br />
may not be reached within the available<br />
contact time between mobile <strong>and</strong> solid<br />
phases. Assuming spherical particles, intraparticle<br />
diffusion kinetics can be described<br />
by Fick´s 2 nd law in radial coordinates (e.g.<br />
Grathwohl, 1998)<br />
2<br />
� C ��<br />
C 2 � C �<br />
� Dap�<br />
� 2<br />
� t<br />
�<br />
� � r r � r �<br />
(18)<br />
where r [m] is the radial distance from the<br />
particle center <strong>and</strong> Dap [m 2 s -1 ] is the appa-<br />
rent diffusion coefficient, which is calculated<br />
from Da by<br />
D<br />
D �<br />
a<br />
ap � (19)<br />
( � ��<br />
�)<br />
� f<br />
where � [-] is the intraparticle porosity, �<br />
the linear equilibrium sorption coefficient<br />
<strong>and</strong> � [kg m -3 ] the particle density. Other<br />
approaches to describe slow sorption / desorption<br />
kinetics comprise first- or secondorder,<br />
two-stage models (e.g. Brusseau <strong>and</strong><br />
Rao, 1989; Ma <strong>and</strong> Selim, 1994; Streck et<br />
al., 1995). A comparison <strong>of</strong> first-order <strong>and</strong><br />
diffusion limited approaches was recently<br />
published by Altfelder <strong>and</strong> Streck (2006).<br />
Kinetic degradation<br />
Degradation, whether biotic or abiotic, is<br />
the only process that reduces the overall<br />
mass <strong>of</strong> contaminants in natural porous<br />
media without transfer to other phases.<br />
Biological degradation mechanisms are numerous,<br />
complex <strong>and</strong> by far not completely<br />
understood nor even identified. The vast<br />
amount <strong>of</strong> different types <strong>of</strong> microorganisms<br />
in the subsurface provides many<br />
metabolic pathways for contaminant degradation<br />
under aerobic <strong>and</strong> anaerobic conditions.<br />
Through successive oxidation or<br />
reduction reactions contaminants can be<br />
transformed to innocuous compounds like<br />
methane, chloride, water or carbon dioxide<br />
(Wiedemeier et al. 1999). However, intermediate<br />
products can even be <strong>of</strong> significantly<br />
higher toxicity <strong>and</strong> persistence than<br />
their parent compounds (e.g., dechlorination<br />
<strong>of</strong> dichloromethane to vinyl chloride;<br />
Wiedemeier et al. 1999).<br />
Kinetic growth <strong>and</strong> decay <strong>of</strong> a microbial<br />
population X [kg m -3 ] can be described by a<br />
generalized Monod-type equation as given<br />
e.g. in Schäfer et al. (1998)<br />
�X<br />
� v<br />
�t<br />
�<br />
max<br />
X<br />
�<br />
I<br />
�<br />
I<br />
Ci<br />
M � C<br />
i Ci<br />
i<br />
C j<br />
� C<br />
j C j j<br />
��<br />
�X� (20)<br />
5
where vmax [s -1 ] is a maximum growth rate,<br />
Ci [kg m -3 ] is the concentration <strong>of</strong> the i th<br />
substrate, MCi [kg m -3 ] is the corresponding<br />
half velocity concentration [kg m -3 ], Cj<br />
[kg m -3 ] is the concentration <strong>of</strong> the j th substance<br />
inhibiting microbial growth, ICj is the<br />
corresponding inhibition concentration<br />
[kg m -3 ] <strong>and</strong> �(X) is a microbial decay term,<br />
which is <strong>of</strong>ten modeled as being <strong>of</strong> first<br />
order. Consumption <strong>of</strong> substrates or production<br />
<strong>of</strong> k metabolites Ck [kg m -3 ] is both<br />
coupled to microbial growth via (Schäfer et<br />
al., 1998)<br />
�Ck �1<br />
��X<br />
�<br />
�<br />
�t<br />
Y �<br />
� �t<br />
�<br />
�<br />
6<br />
(21)<br />
k growth<br />
where Yk [-] is the yield coefficient for<br />
substrate or metabolite Ck, <strong>and</strong> [·]growth<br />
refers to the growth term only in eq. (20).<br />
From the generalized Monod-type equation<br />
different more simple kinetic formulations<br />
can be derived. For a temporally constant<br />
microbial population, i.e. growth <strong>and</strong> decay<br />
terms are constant <strong>and</strong> <strong>of</strong> equal magnitude,<br />
no inhibition <strong>and</strong> dependence on only a<br />
single substrate, eq. (20) <strong>and</strong> (21) can be<br />
combined to yield the Michaelis-Menten<br />
(MM) kinetics model (Simkins <strong>and</strong><br />
Alex<strong>and</strong>er, 1984)<br />
dC<br />
dt<br />
C<br />
� �kmax<br />
(22)<br />
C � M C<br />
where k max is the maximum degradation rate<br />
[kg m -3 s -1 ] <strong>and</strong> M C is the MM half-saturation<br />
concentration [kg m -3 ]. This approximation<br />
may be applicable e.g. when aquifer<br />
sediments have been exposed to contaminants<br />
for several years (Bekins et al., 1998)<br />
<strong>and</strong> is used in application 2 (section 3.2).<br />
Often, contaminant degradation is also<br />
described by simple first order kinetics, e.g.<br />
to simulate abiotic degradation reactions<br />
like hydrolysis <strong>and</strong> dehydrohalogenation <strong>of</strong><br />
halogenated compounds (Wiedemeier et al.,<br />
1999). First order kinetics can be derived<br />
from eq. (22) for C > MC eq. (22)<br />
approaches zero order kinetics. The first<br />
order model is used in application 1 <strong>and</strong> 3<br />
(sections 3.1 <strong>and</strong> 3.3). Extensive reviews on<br />
kinetic models <strong>of</strong> biodegradation can be<br />
found e.g. in Baveye <strong>and</strong> Valocchi (1989),<br />
Rittmann <strong>and</strong> VanBriesen (1996) or Islam<br />
et al. (2001).<br />
2.4. Numerics <strong>and</strong> s<strong>of</strong>tware<br />
methods<br />
Numerical solution <strong>of</strong> the governing<br />
equations<br />
The governing equations for <strong>flow</strong> <strong>and</strong> reactive<br />
transport presented in sections 2.1 - 2.3<br />
belong to the group <strong>of</strong> partial differential<br />
equations (PDE), containing derivatives <strong>of</strong><br />
first order in time <strong>and</strong> <strong>of</strong> first as well as<br />
second order in space. The classification <strong>of</strong><br />
PDE can be based on mathematical aspects<br />
(highest order derivatives in the dependent<br />
variables) or on a physical point <strong>of</strong> view<br />
(problem type <strong>of</strong> physical process) (Kolditz,<br />
2002). Parabolic PDE are used for timedependent<br />
problems with dissipation process,<br />
such as diffusion (eq. (11)) or transient<br />
groundwater <strong>flow</strong> (eq. (3)), which<br />
convert to elliptic PDE for steady state<br />
conditions (eq. (4)). A third class <strong>of</strong> PDE<br />
are hyperbolic equations like the linear advection<br />
equation (eq. (10)), which are used<br />
to describe time-dependent problems without<br />
dissipation process. The transport equations<br />
(15) <strong>and</strong> (16) are <strong>of</strong> mixed type with a<br />
parabolic dispersion-diffusion term <strong>and</strong> a<br />
hyperbolic advection term. Their behaviour<br />
for a particular problem depends on the<br />
relative magnitudes <strong>of</strong> these flux components.<br />
In general the transport equations are<br />
<strong>of</strong> parabolic character which changes to<br />
hyperbolic for large ratios <strong>of</strong> v/�L as in this<br />
case the advective flux term is dominant<br />
(Kolditz, 2002).<br />
In general, PDE describing physical<br />
problems are well-posed when appropriate<br />
initial <strong>and</strong> boundary conditions are specified<br />
for the domain where a solution is<br />
required. While analytical solutions can be
found for a number <strong>of</strong> problems with<br />
simple geometries <strong>and</strong> boundary conditions<br />
(e.g. Bear, 1979; Van Genuchten <strong>and</strong> Alves,<br />
1982; Kinzelbach, 1983), for complex nonlinear<br />
problems exact solutions may not<br />
exist <strong>and</strong> thus approximate <strong>numerical</strong><br />
solutions must be obtained. Given the<br />
governing equations with appropriate initial<br />
<strong>and</strong> boundary conditions for a specified<br />
problem, the general strategy for a <strong>numerical</strong><br />
solution is to first convert the PDE into<br />
a system <strong>of</strong> discrete algebraic equations <strong>and</strong><br />
then to find the exact solution <strong>of</strong> the latter,<br />
which is the approximate solution <strong>of</strong> the<br />
PDE. An overview <strong>of</strong> approximation<br />
methods for the solution <strong>of</strong> PDE is given in<br />
Fig. 3.<br />
Fig. 3: Overview <strong>of</strong> approximation methods<br />
for PDE (Kolditz, 2002).<br />
Among the many possible approaches for<br />
the <strong>numerical</strong> simulation <strong>of</strong> fluid dynamics<br />
<strong>and</strong> transport, finite difference, finite element<br />
<strong>and</strong> finite volume methods are the<br />
most frequently used. Their theory <strong>and</strong> implementation<br />
is content <strong>of</strong> numerous text<br />
books (e.g. Pinder <strong>and</strong> Gray, 1977; Baker,<br />
1983; Zienkiewicz <strong>and</strong> Taylor, 1993;<br />
Helmig, 1997; Kolditz, 2002). Here, a short<br />
description <strong>of</strong> the finite element method<br />
(FEM) only is given, as it is the <strong>numerical</strong><br />
scheme which is implemented in the<br />
GeoSys / Rock<strong>flow</strong> code (Kolditz et al.,<br />
2006) which is used in all applications<br />
presented in chapter 3.<br />
The concept <strong>of</strong> the FEM is based on the<br />
spatial discretization (“meshing”) <strong>of</strong> continuous<br />
structures into discrete substructures,<br />
i.e. the finite elements (FE). In comparison<br />
to finite differences, the big advantage <strong>of</strong><br />
the FEM is its ability to also h<strong>and</strong>le com-<br />
plex geometries by unstructured or arbitrarily<br />
shaped grids (Anderson <strong>and</strong><br />
Woessner, 1992). The governing PDE are<br />
discretized by deriving integral formulations,<br />
e.g. by the method <strong>of</strong> weighted<br />
residuals. By introducing weighting functions<br />
the approximate solution is forced to<br />
satisfy the condition that the weighted<br />
average residual <strong>of</strong> the unknown true solution<br />
at the nodes over the computational<br />
domain is equal to zero. For any location<br />
within the domain the solution is obtained<br />
by linear combination <strong>of</strong> local interpolation<br />
functions <strong>and</strong> the solution at the nodes.<br />
With the st<strong>and</strong>ard Galerkin method<br />
(Huyakorn <strong>and</strong> Pinder, 1983) weighting <strong>and</strong><br />
interpolation functions are selected<br />
identically. The FEM is globally mass<br />
conservative, locally, however, mass<br />
conservation problems may occur. Therefore<br />
mixed FEM approaches draw increasing<br />
attention (e.g. Starke, 2000; Knabner<br />
<strong>and</strong> Schneid, 2002; Korsawe et al., 2006).<br />
The governing PDE <strong>of</strong> <strong>flow</strong> <strong>and</strong> reactive<br />
transport in porous media outlined in the<br />
previous sections are formulated from the<br />
point <strong>of</strong> view <strong>of</strong> an fixed observer with the<br />
fluid <strong>and</strong> solute moving on a fixed spatial<br />
grid. This approach is termed the Eulerian<br />
concept <strong>and</strong> is able to h<strong>and</strong>le dispersiondominated<br />
problems accurately <strong>and</strong> efficiently.<br />
For advection-dominated problems,<br />
however, the Eulerian concept is susceptible<br />
to <strong>numerical</strong> dispersion <strong>and</strong> artificial<br />
oscillations (Zheng <strong>and</strong> Bennett, 1995), For<br />
the price <strong>of</strong> increased computational effort<br />
this problem can be limited by sufficiently<br />
fine spatial <strong>and</strong> temporal discretization.<br />
By contrast, the Lagrangian concept is well<br />
suited for advection-dominated problems,<br />
but problematic when advection <strong>and</strong> dispersion<br />
must be solved together (Thorenz,<br />
2001). In the Lagrangian concept concentrations<br />
are not associated with fixed spatial<br />
points but rather with moving particles, that<br />
are transported with the prevailing <strong>flow</strong><br />
velocity. The <strong>numerical</strong> model SMART<br />
(Finkel et al., 1998) is based on the<br />
Langrangian concept <strong>of</strong> Cvetkovic <strong>and</strong><br />
Dagan (1996). In SMART the model<br />
7
domain is discretized along advective <strong>flow</strong><br />
paths <strong>of</strong> the Eulerian <strong>flow</strong> field by the travel<br />
time � [s] <strong>of</strong> inert particles between an<br />
injection plane <strong>and</strong> a control plane, both<br />
oriented normal to the mean <strong>flow</strong> direction.<br />
Each particle trajectory is regarded as a<br />
separate one-dimensional stream tube <strong>of</strong> the<br />
<strong>flow</strong> field with an infinitesimal cross<br />
section. The probability density function<br />
(pdf) g(�, x) <strong>of</strong> all particles travel times<br />
completely reflects all hydraulic heterogeneities<br />
<strong>of</strong> the model domain. Influences <strong>of</strong><br />
reactions (e.g. biodegradation, intraparticle<br />
diffusion, sorption, etc.) are quantified by<br />
means <strong>of</strong> the reaction function �(�, t),<br />
which is evaluated by the BESSY model<br />
(Jäger <strong>and</strong> Liedl, 2000) implemented in<br />
SMART. With given g <strong>and</strong> � the normalized<br />
breakthrough curve <strong>of</strong> a reactive<br />
solute at a control plane is calculated by<br />
(Finkel et al. 1998)<br />
8<br />
�<br />
� �<br />
0<br />
�x, t�<br />
g��,<br />
x����t�d�<br />
C ,<br />
(24).<br />
To overcome the limitations <strong>of</strong> both the<br />
Eulerian <strong>and</strong> the Lagrangian concepts,<br />
mixed Eulerian-Lagrangian methods can be<br />
used, which take advantage <strong>of</strong> the particular<br />
appropriateness <strong>of</strong> both concepts for <strong>modeling</strong><br />
advective <strong>and</strong> dispersive transport (e.g.<br />
Neumann, 1981; Thorenz, 2001; Park et al.,<br />
2006). In application 3 (section 3.3) a<br />
combination <strong>of</strong> the s<strong>of</strong>tware codes GeoSys /<br />
Rock<strong>flow</strong> <strong>and</strong> SMART is used for a combined<br />
application <strong>of</strong> the Eulerian <strong>and</strong> the<br />
Lagrangian concepts. GeoSys / Rock<strong>flow</strong> is<br />
used to model the Eulerian <strong>flow</strong> field in a<br />
heterogeneous two dimensional domain <strong>and</strong><br />
to derive the representative pdf g(�, x). The<br />
SMART model then utilizes the pdf for the<br />
simulation <strong>of</strong> reactive transport in the<br />
model domain.<br />
Object- <strong>and</strong> process-oriented methods<br />
The GeoSys / Rock<strong>flow</strong> code, which is used<br />
for most <strong>of</strong> the <strong>numerical</strong> simulations<br />
described in chapter 3, is written in the C++<br />
language <strong>and</strong> thus allows the implementation<br />
by object oriented programming (OOP)<br />
methods. The OOP concept is especially<br />
helpful for the development <strong>of</strong> complex<br />
s<strong>of</strong>tware in programmer teams, as encapsulation<br />
<strong>and</strong> class-structures render the code<br />
more stable <strong>and</strong> errors are easier to detect.<br />
In GeoSys / Rock<strong>flow</strong> the OOP concept is<br />
met by so called process orientation<br />
(Kolditz <strong>and</strong> Bauer, 2004), which allows<br />
the coupling <strong>of</strong> two-phase <strong>flow</strong>, heat transport,<br />
mass transport, chemical reactions <strong>and</strong><br />
deformation in an efficient way (Wang et<br />
al., 2006). The basic idea <strong>of</strong> process orientation<br />
is that between each physical process<br />
(e.g. single species transport) <strong>and</strong> its<br />
<strong>numerical</strong> approximation by an algebraic<br />
equation system (AES) exists a direct correspondence<br />
(Fig. 4). The AES originates<br />
from the temporal <strong>and</strong> spatial discretization<br />
<strong>of</strong> the PDE on the computational grid. For<br />
its solution the following steps are performed:<br />
� AES assemblage <strong>and</strong> incorporation <strong>of</strong><br />
initial conditions<br />
� determination <strong>of</strong> element matrices<br />
� incorporation <strong>of</strong> boundary conditions<br />
<strong>and</strong> source terms<br />
� solving the AES by appropriate solvers<br />
This procedure can be generalized for any<br />
physical process regardless <strong>of</strong> its specific<br />
type in an object oriented way by introducing<br />
the process object (Kolditz <strong>and</strong> Bauer,<br />
2004) (Fig. 4).<br />
Transport <strong>of</strong> non-reactive<br />
species in water<br />
„PROCESS“<br />
Multifield problem, if many<br />
mobile species are<br />
transported<br />
Solution <strong>of</strong> a PDE<br />
system<br />
PROCESS-OBJECT<br />
System <strong>of</strong> PDE<br />
solved by<br />
Multi-Process-Method<br />
Fig. 4: Process analogy <strong>and</strong> process object,<br />
shown for an instance <strong>of</strong> a transport<br />
process (Kolditz <strong>and</strong> Bauer, 2004).
The process object has access to all<br />
required data structures <strong>and</strong> functions, <strong>and</strong><br />
thus is self configuring, executing <strong>and</strong><br />
destructing.<br />
Contaminant transport problems usually<br />
involve a number <strong>of</strong> different processes as<br />
outlined in sections 2.1 – 2.3. The resulting<br />
multi-field problems can be approached by<br />
a multi-process algorithm, where one instance<br />
<strong>of</strong> the process object is created for<br />
each process considered. Solution <strong>of</strong> any<br />
number <strong>of</strong> <strong>flow</strong> or transport equations is<br />
fully automatic <strong>and</strong> encapsulated, guaranteeing<br />
high efficiency <strong>and</strong> flexibility.<br />
In GeoSys / Rock<strong>flow</strong> <strong>saturated</strong> <strong>and</strong> un<strong>saturated</strong><br />
<strong>flow</strong> as well as conservative transport<br />
are solved using st<strong>and</strong>ard Galerkin FE. A<br />
non-iterative operator splitting technique<br />
for the coupling <strong>of</strong> conservative transport<br />
<strong>and</strong> (bio-)chemical reaction processes is<br />
used (Xie et al., 2006; Bauer et al., 2006b).<br />
First, the <strong>flow</strong> field is solved followed by<br />
conservative transport for all species. In the<br />
third step the calculation <strong>of</strong> kinetic<br />
biochemical reactions is performed. Finally<br />
chemical equilibrium reactions are calculated.<br />
This approach allows an easy h<strong>and</strong>ling<br />
<strong>of</strong> any number <strong>of</strong> species <strong>and</strong> reaction<br />
processes as well as employing optimised<br />
mathematical methods for the solution <strong>of</strong><br />
the corresponding equation systems. The<br />
non-iterative approach, however, is limited<br />
to small time steps in order to avoid<br />
<strong>numerical</strong> instabilities. It is also known not<br />
to converge necessarily to the exact solution<br />
(Carrayrou et al., 2004). These limitations<br />
can be overcome using a computationally<br />
more dem<strong>and</strong>ing iterative operator splitting<br />
approach (e.g. Kinzelbach et al., 1991).<br />
3. Modeling applications<br />
In this chapter three application examples<br />
<strong>of</strong> <strong>numerical</strong> models for <strong>flow</strong> <strong>and</strong> reactive<br />
transport as established in chapter 2 are<br />
presented. Section 3.1 introduces the Virtual<br />
Aquifer (VA) concept, in which <strong>numerical</strong><br />
<strong>modeling</strong> is used for the evaluation <strong>of</strong><br />
investigation strategies for contaminated<br />
sites. In section 3.2 the VA concept is<br />
applied to derive <strong>and</strong> test a novel method<br />
for the estimation <strong>of</strong> biodegradation kinetic<br />
parameters from measured field data. Section<br />
3.3 uses <strong>numerical</strong> <strong>modeling</strong> as a tool<br />
to predict the environmental impact <strong>of</strong> demolition<br />
waste used in road constructions<br />
3.1. Evaluation <strong>of</strong> investigation<br />
strategies for contaminated<br />
aquifers using the Virtual<br />
Aquifer concept<br />
Due to the limited accessibility <strong>of</strong> the<br />
subsurface, measurements <strong>of</strong> piezometric<br />
heads <strong>and</strong> pollutant concentrations at contaminated<br />
sites are sparse <strong>and</strong> may not be<br />
representative <strong>of</strong> the heterogeneous hydrogeologic<br />
conditions. Any site investigation<br />
is thus subject to uncertainty, reflecting the<br />
limited knowledge on aquifer properties<br />
<strong>and</strong> the extent <strong>of</strong> the contamination. Three<br />
main sources <strong>of</strong> uncertainty can be identified<br />
for site investigation, which are illustrated<br />
in Fig. 5. Conceptual model errors<br />
result from an incorrect identification <strong>of</strong> the<br />
governing processes at a site. Heterogeneity<br />
<strong>of</strong> the site causes an incomplete or wrong<br />
description <strong>of</strong> the relevant parameter distributions.<br />
For variables measured at observation<br />
wells like heads or concentrations<br />
measurement errors are inevitable. Due to<br />
this uncertainty, field investigation methods<br />
for plume screening <strong>and</strong> measuring <strong>of</strong><br />
hydraulic conductivity or degradation rates<br />
can hardly be tested or verified in the field.<br />
The VA approach is particularly aimed to<br />
overcome this problem. Its basic idea is the<br />
computer based evaluation <strong>of</strong> the performance<br />
<strong>and</strong> reliability <strong>of</strong> field investigation<br />
methods by application in heterogeneous<br />
synthetic (i.e. virtual) aquifers. In this it<br />
resembles the concept <strong>of</strong> “virtual realities”<br />
(Schäfer et al., 2002) which are used e.g. in<br />
car industry (“virtual crash test”), education<br />
(flight simulators) or medicine (interactive<br />
operation planning).<br />
9
conceptional<br />
model error:<br />
wrong process<br />
description<br />
10<br />
measurement<br />
error<br />
resulting<br />
investigation error<br />
conceptual site model<br />
investigation<br />
virtual aquifer<br />
Fig. 5: Virtual aquifer concept <strong>and</strong> possible sources <strong>of</strong> investigation error.<br />
An application <strong>of</strong> the VA concept requires<br />
the definition <strong>of</strong> a synthetic site model <strong>and</strong><br />
its translation into a <strong>numerical</strong> model for<br />
the simulation <strong>of</strong> the identified relevant processes.<br />
Synthetic site models are generated<br />
based on statistical properties <strong>of</strong> natural<br />
aquifers. A defined contaminant source is<br />
then introduced <strong>and</strong> the evolution <strong>of</strong> aquifer<br />
contamination is simulated by <strong>numerical</strong><br />
<strong>modeling</strong>, thus generating a realistic<br />
contaminant distribution in the synthetic<br />
aquifer. In comparison to the "real world",<br />
the unique advantage <strong>of</strong> the synthetic aquifer<br />
is that the spatial distribution <strong>of</strong> all<br />
physical <strong>and</strong> geochemical properties <strong>and</strong><br />
parameters as well as contaminant concentrations<br />
are exactly known. Once the<br />
synthetic contaminated aquifer is generated,<br />
it can be studied by st<strong>and</strong>ard monitoring<br />
<strong>and</strong> investigation techniques, e.g. by emplacement<br />
<strong>of</strong> observation wells. Although<br />
the parameter distribution <strong>of</strong> the synthetic<br />
aquifer is known a priori, only the data<br />
“measured” at wells (i.e. hydraulic heads or<br />
concentrations) are used <strong>and</strong> interpreted.<br />
This is done because in a real site investigation<br />
also only a limited amount <strong>of</strong> measured<br />
data would be available. Finally, the results<br />
are compared to the “true” parameter distribution<br />
known from the synthetic aquifer,<br />
allowing an evaluation <strong>of</strong> the accuracy <strong>of</strong><br />
the investigation method used. Using the<br />
VA concept, sources <strong>of</strong> uncertainty or error<br />
can be considered individually <strong>and</strong> the<br />
heterogeneity:<br />
incomplete or<br />
wrong description<br />
<strong>of</strong> parameter<br />
distribution<br />
sensitivity <strong>of</strong> investigation results on these<br />
can be studied. Stochastic approaches like<br />
the Monte-Carlo method are applied to<br />
study the propagation <strong>of</strong> parameter variability<br />
<strong>and</strong> uncertainty into the investigation<br />
results. The VA has been first introduced<br />
by Schäfer et al. (2002) <strong>and</strong> was applied by<br />
Schäfer et al. (2004, 2006b), Bauer <strong>and</strong><br />
Kolditz (2006), Bauer et al. (2005 [EP 1];<br />
2006a [EP 2], 2007 [EP 4]) <strong>and</strong> Beyer et al.<br />
(2006 [EP 3], 2007a [EP 5]). An overview<br />
<strong>of</strong> VA applications is given in Bauer et al.<br />
(2006b) <strong>and</strong> Schäfer et al. (2006a).<br />
Fig. 6: Virtual investigation <strong>of</strong> a heterogeneous<br />
contaminant plume by the center<br />
line method (Bauer et al., 2005 [EP 1]).<br />
The VA concept is used here to study errors<br />
<strong>and</strong> uncertainties in degradation rate constants<br />
estimated from data typically collected<br />
by site investigation with the so called
center line method (Fig. 6). This method is<br />
frequently used in field studies when natural<br />
attenuation is considered as a remediation<br />
alternative <strong>and</strong> is based on observation<br />
wells that are placed along the presumed<br />
center line <strong>of</strong> the contaminant plume.<br />
Influence <strong>of</strong> measurement errors<br />
The first aspect studied here is the influence<br />
<strong>of</strong> measurement errors in hydraulic heads<br />
on degradation rate constant estimates<br />
(Bauer et al., 2007 [EP 4]). The VA concept<br />
used in this study is based on a two<br />
dimensional conceptual model <strong>of</strong> the<br />
groundwater body using a homogeneous<br />
distribution <strong>of</strong> hydraulic conductivity K.<br />
The development <strong>of</strong> the contaminant plume<br />
originating from a rectangular source zone<br />
is simulated until steady state conditions are<br />
established. The <strong>numerical</strong> simulations are<br />
performed using the GeoSys / Rock<strong>flow</strong><br />
code, which was introduced in the previous<br />
chapter. The contaminant is subject to a<br />
first order kinetics (eq. (23)) degradation<br />
reaction using a uniform degradation rate<br />
constant �. Additionally, a conservative<br />
tracer is emitted from the source zone. The<br />
contaminant plume thus generated is then<br />
investigated by the center line method.<br />
From the hydraulic heads measured at three<br />
initial observation wells (one being located<br />
directly in the center <strong>of</strong> the source zone)<br />
first the direction <strong>of</strong> groundwater <strong>flow</strong> is<br />
estimated by construction <strong>of</strong> a hydrogeological<br />
triangle. Head measurements are<br />
obtained by reading the model output at the<br />
respective well locations <strong>and</strong> adding a<br />
r<strong>and</strong>om measurement error by<br />
h � � h � ���<br />
(25)<br />
h<br />
where h´ <strong>and</strong> h are the erroneous <strong>and</strong> exact<br />
(i.e. simulated) heads, respectively, � is an<br />
evenly distributed r<strong>and</strong>om number from the<br />
interval [-1, 1] <strong>and</strong> ��h is the maximum<br />
measurement error. Along the estimated<br />
(<strong>and</strong> potentially erroneous) <strong>flow</strong> direction<br />
three new observation wells are installed,<br />
one at every 10 m. These wells were then<br />
used to measure local (erroneous) heads,<br />
contaminant concentrations <strong>and</strong> hydraulic<br />
conductivities along the presumed plume<br />
center line. From the hydraulic head difference,<br />
true porosity <strong>and</strong> well positions<br />
groundwater <strong>flow</strong> velocities are calculated.<br />
Together with the concentration data this<br />
allows the determination <strong>of</strong> � using any <strong>of</strong><br />
the analytical models presented in Tab. 1.<br />
As hydraulic conductivity K is distributed<br />
homogeneously <strong>and</strong> concentrations are<br />
assumed to be measured precisely, the only<br />
source <strong>of</strong> error here is the measured head.<br />
Tab. 1: Analytical models for the estimation <strong>of</strong> the first order degradation rate constant �.<br />
method formula description reference<br />
1<br />
2<br />
3<br />
4<br />
v � �<br />
a C ( x)<br />
� � � �<br />
�<br />
�<br />
�<br />
1 ln<br />
�x<br />
� C0<br />
�<br />
v � � �<br />
a � C ( x)<br />
C0<br />
� � �<br />
�<br />
2 ln<br />
�x<br />
� C � �<br />
� 0 C ( x)<br />
�<br />
with<br />
�C( x)<br />
C �<br />
2<br />
v �<br />
�<br />
a ��<br />
ln<br />
0 �<br />
�<br />
� �<br />
3 �<br />
�<br />
�1<br />
� 2�<br />
L<br />
� 1<br />
4�<br />
�<br />
L ��<br />
�x<br />
� �<br />
�C( x)<br />
( C � ) �<br />
2<br />
v �<br />
�<br />
a ��<br />
ln<br />
0 �<br />
�<br />
� �<br />
4 �<br />
�<br />
�1<br />
� 2�<br />
L<br />
� 1<br />
4�<br />
�<br />
L ��<br />
�x<br />
� �<br />
� �<br />
�<br />
W S<br />
� � erf<br />
�<br />
� �<br />
� 4 � T � x �<br />
analyt. solution <strong>of</strong> 1D advection<br />
equation with first order<br />
degradation<br />
same as method 1; concentrations<br />
normalized by conservative tracer<br />
analyt. solution <strong>of</strong> 1D advectiondispersion<br />
equation with first<br />
order degradation<br />
analyt. solution <strong>of</strong> 2D advectiondispersion<br />
equation with first<br />
order degradation <strong>and</strong> accounting<br />
for the source area width .<br />
Newell et al.<br />
(2002)<br />
Wiedemeyer<br />
et al. (1996)<br />
Buscheck<br />
<strong>and</strong> Alcantar<br />
(1995)<br />
Zhang <strong>and</strong><br />
Heathcote<br />
(2003)<br />
11
The conceptual model used is a rigorous<br />
simplification <strong>of</strong> the processes observed in<br />
natural aquifer systems, as K varies in space<br />
<strong>and</strong> contaminant degradation usually follows<br />
more complicated laws, depends on<br />
microorganism growth <strong>and</strong> may be limited<br />
or inhibited by other substances. The<br />
simplifications assumed here are however<br />
necessary to study the sole effects <strong>of</strong> measurement<br />
error on rate constant estimation<br />
under otherwise ideal conditions for the<br />
application <strong>of</strong> the center line method <strong>and</strong><br />
the analytical models <strong>of</strong> Tab. 1.<br />
In this synthesis only results for method 1<br />
<strong>of</strong> Tab. 1 are presented. The degradation<br />
rate estimated from evaluation <strong>of</strong> measured<br />
heads <strong>and</strong> concentrations is divided by the<br />
true rate constant used in the <strong>numerical</strong><br />
model yielding normalized overestimation<br />
factors. To assess the range <strong>of</strong> uncertainty<br />
resulting from the r<strong>and</strong>om measurement<br />
error a Monte Carlo analysis is conducted<br />
for five increasing values <strong>of</strong> ��h, each with<br />
a sample size <strong>of</strong> 100. Fig. 7 presents the<br />
rate constants thus estimated versus ��h.<br />
Without measurement error the correct rate<br />
constant is obtained. However, already for<br />
very small ��h < 1 cm, a significant<br />
overestimation can be observed. This has<br />
two main reasons: Firstly, an erroneous<br />
head yields an incorrect transport velocity<br />
<strong>and</strong> thus results in rate constant<br />
overestimation for too high velocities <strong>and</strong> in<br />
underestimation for too low velocities.<br />
Secondly, erroneous heads result in an incorrect<br />
derivation <strong>of</strong> the <strong>flow</strong> direction<br />
using the hydrogeologic triangle. Thus the<br />
observation wells installed downgradient <strong>of</strong><br />
the source may be placed <strong>of</strong>f the true center<br />
line position (compare Fig. 6). Hence, concentrations<br />
measured in <strong>of</strong>f center line wells<br />
are too low, indicating an overly high rate<br />
<strong>of</strong> degradation. Overestimation <strong>of</strong> the<br />
degradation rate increases with the maximum<br />
head error <strong>and</strong> reaches factors <strong>of</strong> more<br />
than 20 in the worst cases.<br />
In Bauer et al. (2007 [EP 4]) also the<br />
influence <strong>of</strong> concentration measurement<br />
errors was studied. It was found that also<br />
12<br />
this type <strong>of</strong> error results in overestimation<br />
<strong>of</strong> the rate constant on average. Erroneous<br />
heads, however, were found to have a larger<br />
impact on the rate constant estimates. For<br />
real field applications therefore much care<br />
has to be taken when measuring hydraulic<br />
heads for the derivation <strong>of</strong> the plume center<br />
line position.<br />
Fig. 7: Influence <strong>of</strong> head measurement<br />
error on rate constant estimates (Bauer et<br />
al., 2007 [EP 4]). The reference rate<br />
constant is indicated by the horizontal line,<br />
small symbols represent single estimates,<br />
big symbols are ensemble averages with<br />
st<strong>and</strong>ard deviation as error bars.<br />
Influence <strong>of</strong> aquifer heterogeneity<br />
The second aspect studied using the VA<br />
concept is the influence <strong>of</strong> spatially heterogeneous<br />
hydraulic conductivity distributions<br />
on the accuracy <strong>of</strong> rate constant<br />
estimation methods 1 – 4 <strong>of</strong> Tab. 1. For this<br />
end the conceptual model used so far is<br />
modified by assuming all head, concentration<br />
<strong>and</strong> K measurements to be free <strong>of</strong><br />
measurement error <strong>and</strong> regarding K as a<br />
spatial r<strong>and</strong>om variable. This is done to<br />
study the sole influence <strong>of</strong> heterogeneous<br />
conductivity. Multiple realizations <strong>of</strong> heterogeneous<br />
K fields for four degrees <strong>of</strong><br />
aquifer heterogeneity characterized by the<br />
ln-conductivity variance 2<br />
� Y were generated,<br />
using a Monte Carlo approach to study<br />
the range <strong>of</strong> investigation result uncertainty.<br />
� at least 100 different reali-<br />
For each 2<br />
Y
zations were generated. For all realizations<br />
the spreading <strong>of</strong> a conservative <strong>and</strong> a<br />
reactive contaminant plume subject to first<br />
order degradation was simulated. The resulting<br />
heterogeneous plumes were investigated<br />
as explained above. Fig. 8 presents<br />
normalized estimated rate constants for<br />
2<br />
methods 1 – 4 (Tab. 1) versus � Y (Bauer et<br />
al., 2005 [EP 1]). Clearly, most rate<br />
constants are larger than 1, i.e. the rate<br />
constant is generally overestimated. Single<br />
realizations show overestimation by several<br />
orders <strong>of</strong> magnitude. It is obvious that an<br />
increase in K heterogeneity causes higher<br />
overestimation. Also the spread <strong>of</strong> the 100<br />
realizations <strong>and</strong> the resulting ensemble<br />
st<strong>and</strong>ard deviations increase significantly,<br />
causing higher uncertainty in the rate constant<br />
estimate. The main reasons for the<br />
observed overestimation are identified as<br />
deviation <strong>of</strong> observation wells from the true<br />
plume center line position, an incorrect<br />
approximation <strong>of</strong> the transport velocity <strong>and</strong><br />
no or inadequate consideration <strong>of</strong> concen-<br />
normalized deg. rate constant [-]<br />
normalized deg. rate constant [-]<br />
1000<br />
100<br />
10<br />
1<br />
0.1<br />
0.01<br />
1000<br />
100<br />
10<br />
1<br />
0.1<br />
0.01<br />
(a) method 1 (b) method 2<br />
(c) method 3 (d) method 4<br />
0 1 2 3 4 5<br />
tration reduction by longitudinal <strong>and</strong><br />
transverse dispersion. Comparing the performance<br />
<strong>of</strong> methods 1 – 4 yields that<br />
method 2 is the most accurate <strong>and</strong> reliable<br />
among the four approaches. The superiority<br />
<strong>of</strong> method 2 follows from the correction <strong>of</strong><br />
contaminant concentrations by normalization<br />
to the concentrations <strong>of</strong> a conservative<br />
tracer, which is spread from the same<br />
source. The tracer correction successfully<br />
accounts for the effects <strong>of</strong> dispersion <strong>and</strong><br />
measuring <strong>of</strong>f the center line. Method 3 explicitly<br />
accounts for longitudinal, method 4<br />
for both, longitudinal <strong>and</strong> transverse dispersion.<br />
However, for each realization investigated,<br />
method 3 yields a higher rate constant<br />
estimate than methods 1 or 2. Longitudinal<br />
dispersion <strong>of</strong> a degrading contaminant<br />
results in a stronger spreading <strong>of</strong> the solute<br />
downstream <strong>and</strong> thus in higher concentrations<br />
along the center line <strong>of</strong> a steady state<br />
plume. To model an observed concentration<br />
reduction with a one-dimensional model<br />
like method 3 which accounts for �L only<br />
0 1 2 3 4 5<br />
�2 �2 Y Y<br />
Fig. 8: Estimated degradation rate constants versus aquifer heterogeneity for methods 1 (a),<br />
2 (b), 3 (c) <strong>and</strong> 4 (d) <strong>of</strong> Tab. 1 (Bauer et al., 2005 [EP 1]). Small symbols represent single<br />
realization results, large symbols ensemble averages <strong>of</strong> 100 realizations.<br />
13
equires therefore a higher degradation rate<br />
constant. Method 4 yields significantly<br />
lower rate constant estimates than<br />
method 1, when aquifer heterogeneity is<br />
low, but almost the same results as method<br />
3 for high heterogeneity. The rate constant<br />
underestimation for low heterogeneity is<br />
due to an “over correction” for transverse<br />
dispersion by the error function term � in<br />
the rate equation. Hence, both methods fail<br />
to yield closer rate constant estimates than<br />
the simpler approach <strong>of</strong> method 1, which<br />
completely neglects the dispersion process.<br />
Influence <strong>of</strong> dispersivity parameterization<br />
Since the results <strong>of</strong> methods 3 <strong>and</strong> 4 depend<br />
on longitudinal <strong>and</strong> transverse dispersivities,<br />
an adequate parameterization is crucial<br />
for their success. In this study, parameterization<br />
<strong>of</strong> �L as well as �T is based on the<br />
scale <strong>of</strong> the contamination problem, as<br />
common with many field applications.<br />
Clearly from the results in Fig. 8 derivation<br />
<strong>of</strong> �L <strong>and</strong> �T from the assumed length <strong>of</strong> the<br />
contaminant plume is not appropriate. As<br />
known from stochastic hydrogeology (e.g.<br />
Dagan, 1989) �L as well as �T strongly<br />
depend on travel time <strong>and</strong> distance as well<br />
as on the correlation structure <strong>of</strong> hydraulic<br />
conductivity <strong>and</strong> <strong>flow</strong> velocity. Therefore<br />
the influence <strong>of</strong> dispersivity parameterization<br />
on the performance <strong>of</strong> methods 3 <strong>and</strong> 4<br />
was analyzed (Bauer et al., 2006a [EP 2]).<br />
From the <strong>of</strong>ten very limited amount <strong>of</strong> data<br />
on the degree <strong>of</strong> aquifer heterogeneity <strong>and</strong><br />
the spatial correlation structure <strong>of</strong> hydraulic<br />
conductivity available for real field sites,<br />
the inference <strong>of</strong> dispersivities by methods<br />
<strong>of</strong> stochastic theory is rarely possible or at<br />
least afflicted by high uncertainty.<br />
Therefore a sensitivity analysis is performed<br />
to cover a wide range <strong>of</strong> values for<br />
�L <strong>and</strong> �T. Varying both parameters the<br />
heterogeneous plume realizations were reevaluated<br />
using the different parameterizations<br />
<strong>of</strong> method 4. Fig. 9 shows results <strong>of</strong><br />
method 4 in terms <strong>of</strong> ensemble medians <strong>of</strong><br />
the 100 estimated rate constants for each <strong>of</strong><br />
the four different degrees <strong>of</strong> heterogeneity<br />
14<br />
<strong>and</strong> all combinations <strong>of</strong> �L <strong>and</strong> �T considered.<br />
As method 4 converges with<br />
method 3 for �T = 0, results for �T = 0 are<br />
also representative for method 3.<br />
For all degrees <strong>of</strong> heterogeneity, decreasing<br />
median degradation rates are found with<br />
increasing �T. For low <strong>and</strong> medium heterogeneity<br />
(Fig. 9 (a) <strong>and</strong> (b)) combinations <strong>of</strong><br />
�L <strong>and</strong> �T can be found, which allow for an<br />
optimal estimation <strong>of</strong> the degradation rate<br />
constant, i.e. the results are <strong>of</strong> comparable<br />
accuracy as those <strong>of</strong> methods 1 or 2. For the<br />
highest degree <strong>of</strong> heterogeneity (Fig. 9 (d)),<br />
no such parameter combination is found<br />
within the range <strong>of</strong> dispersivities considered<br />
here. However, with an inadequate parameterization<br />
also for low heterogeneities<br />
severe over- as well as underestimation is<br />
possible. Moreover, with the exception <strong>of</strong><br />
the low heterogeneity case, only unphysical<br />
combinations <strong>of</strong> low �L <strong>and</strong> high �T yield<br />
close estimates <strong>of</strong> the rate constant. These<br />
parameters do not represent actual dispersivities,<br />
but must be considered as mere<br />
lumped fitting parameters, as they are used<br />
to correct for <strong>of</strong>f center line measurements<br />
<strong>and</strong> not the dispersion process only. As the<br />
magnitude <strong>of</strong> bias caused by <strong>of</strong>f center line<br />
measurements is not known <strong>and</strong> the K<br />
correlation structure <strong>and</strong> degree <strong>of</strong> heterogeneity<br />
are usually not properly characterized<br />
at many field sites, choosing<br />
dispersivities for an application <strong>of</strong> method 4<br />
would be highly uncertain <strong>and</strong> arbitrary.<br />
From the results presented here, it can be<br />
concluded that for contaminated sites,<br />
where the assumption <strong>of</strong> first order degradation<br />
kinetics can be justified, method 2<br />
should be preferably applied, if the degradation<br />
rate constant is to be derived with<br />
the center line approach. In situations,<br />
where a correction <strong>of</strong> concentrations by a<br />
conservative tracer is not possible,<br />
alternatively method 1 should be used, as<br />
its application implies the least amount <strong>of</strong><br />
parameterization uncertainty.
Fig. 9: Influence <strong>of</strong> dispersivity parameterization on estimated degradation rate constants for<br />
2<br />
method 4 (<strong>and</strong> method 3 with �T = 0) <strong>of</strong> Tab. 1 <strong>and</strong> four degrees <strong>of</strong> aquifer heterogeneity � Y<br />
(Bauer et al., 2006a [EP 2]). Symbols represent ensemble medians <strong>of</strong> 100 realizations.<br />
Extensive monitoring networks<br />
For the previous studies, linearly arranged<br />
center line observation wells have been<br />
used for the VA investigation (compare Fig.<br />
6). Contaminant plumes in natural heterogeneous<br />
aquifers may however show a<br />
substantial amount <strong>of</strong> me<strong>and</strong>ering (Wilson<br />
et al., 2004) resulting in non-linear center<br />
line orientations. As documented, an<br />
inappropriate assumption <strong>of</strong> a linear center<br />
line may result in misplacement <strong>of</strong> the<br />
observation wells <strong>and</strong> in significant overestimation,<br />
if degradation rate constants are to<br />
be derived. Moreover, only measurements<br />
in observation wells located on the plume<br />
axis are evaluated <strong>and</strong> additional information<br />
possibly at h<strong>and</strong> (i.e. well data downgradient<br />
from the source but not on the<br />
center line) is not explicitly accounted for<br />
in the rate estimation. Therefore the performance<br />
<strong>of</strong> methods 1, 3 <strong>and</strong> 4 <strong>of</strong> Tab. 1 was<br />
Fig. 10: Site investigation for degradation<br />
rate constant evaluation by (A) non-linearly<br />
positioned center line wells <strong>and</strong> (B) using<br />
all observation wells <strong>of</strong> the monitoring<br />
network (Beyer et al., 2007a [EP 5]). Small<br />
dark squares show observation well<br />
locations, larger squares show presumed<br />
center line well locations, contour lines are<br />
concentrations calculated analytically with<br />
the approach <strong>of</strong> Stenback et al. (2004).<br />
15
studied for heterogeneous sites with extensive<br />
monitoring networks allowing the estimation<br />
<strong>of</strong> the degradation rate constant<br />
based on freely positioned observation<br />
wells from which a center line is constructed<br />
(strategy A, Fig. 10 (a)). Results are<br />
compared to a two-dimensional inverse<br />
<strong>modeling</strong> approach <strong>of</strong> Stenback et al.<br />
(2004), which accounts for information<br />
from all observation wells <strong>of</strong> a monitoring<br />
network (strategy B, Fig. 10 (b)). Both rate<br />
constant estimation strategies are applied to<br />
a set <strong>of</strong> synthetic contaminated sites with<br />
independently designed extensive monitoring<br />
networks (Beyer et al., 2007a [EP 5]).<br />
a)<br />
b)<br />
16<br />
norm. degradation rate constant [-]<br />
norm. degradation rate constant [-]<br />
100<br />
10<br />
1<br />
0.1<br />
100<br />
10<br />
1<br />
0.1<br />
SW = 4 m<br />
SW = 16 m<br />
meth. 1<br />
(A)<br />
meth. 3<br />
(A)<br />
meth. 4<br />
(A)<br />
investigation strategy<br />
Stenback<br />
(B)<br />
Fig. 11: Rate constant estimates obtained<br />
for investigation strategy A <strong>and</strong> methods 1,<br />
3 <strong>and</strong> 4 as well as strategy (B) for a source<br />
width WS <strong>of</strong> 4 (a) <strong>and</strong> 16 m (b) (Beyer et al.,<br />
2007a [EP 5]). Small symbols represent<br />
single realization results, large symbols<br />
ensemble averages.<br />
Two different types <strong>of</strong> plumes are considered<br />
in this comparison: narrow contaminant<br />
plumes with a small source area width<br />
WS = 4 m (Fig. 11 (a)) <strong>and</strong> wider plumes<br />
with WS = 16 m (Fig. 11 (b)). For small<br />
source widths overestimation by strategy B<br />
is comparable to or at best slightly less than<br />
for approach A. For large source widths <strong>and</strong><br />
wider plumes, however, strategy B yields<br />
closer estimates <strong>of</strong> the degradation rate<br />
constant than strategy A on average. The<br />
results <strong>of</strong> this study suggest that incorporation<br />
<strong>of</strong> <strong>of</strong>f center line information for the<br />
estimation <strong>of</strong> the degradation rate constant<br />
can improve results <strong>of</strong> the plume investigation<br />
significantly.<br />
3.2. Development <strong>and</strong> testing <strong>of</strong> a<br />
new approach to estimating<br />
biodegradation parameters<br />
from field data<br />
In the previous section, the VA method was<br />
applied to evaluate the performance <strong>of</strong><br />
different analytical models for the derivation<br />
<strong>of</strong> first order degradation rate constants<br />
from center line investigation data in heterogeneous<br />
aquifers. From the literature,<br />
however, it is well known that the use <strong>of</strong><br />
first order kinetics may be problematic in<br />
some situations, as it is an inaccurate representation<br />
<strong>of</strong> the processes occurring in<br />
contaminated aquifers. Usage <strong>of</strong> a first<br />
order model outside its range <strong>of</strong> validity<br />
may result either in significant under- or<br />
overestimation <strong>of</strong> the attenuation potential<br />
at a site (Bekins et al., 1998). In an<br />
extension to this study, therefore a new<br />
approach for the estimation <strong>of</strong> degradation<br />
parameters kmax <strong>and</strong> MC for Michaelis-<br />
Menten (MM) kinetics (eq. 22) from the<br />
same plume investigation strategy was<br />
developed <strong>and</strong> tested in Beyer et al. (2006<br />
[EP 3]) using the VA method.<br />
In its integral form eq. (22) can be<br />
rearranged to
v<br />
�<br />
�C�Cx) �<br />
� C0<br />
�<br />
ln��<br />
C�x���<br />
M � � 1<br />
�<br />
�<br />
( k C � C(<br />
x)<br />
k<br />
x C<br />
(25).<br />
a 0<br />
max 0<br />
max<br />
With the same type <strong>of</strong> center line<br />
investigation data used for the estimation <strong>of</strong><br />
the first order degradation rate constant (i.e.<br />
local concentrations, heads, hydraulic conductivities),<br />
eq. (25) can be utilized to<br />
estimate the MM parameters kmax <strong>and</strong> MC<br />
by linear regression.<br />
The Monte Carlo scenario definition <strong>and</strong><br />
site investigation procedure for this study<br />
are similar to those explained in section 3.1.<br />
Numerical simulations <strong>of</strong> plume development<br />
in homogeneous <strong>and</strong> heterogeneous<br />
aquifers were performed with the GeoSys /<br />
Rock<strong>flow</strong> code. Here, however, the contaminant<br />
plumes investigated were generated<br />
using MM instead <strong>of</strong> first order degradation<br />
kinetics. The parameters k max <strong>and</strong> M C<br />
estimated with eq. (25) for the different<br />
plume realizations were normalized to the<br />
true values used in the <strong>numerical</strong><br />
simulations <strong>and</strong> are shown in a 3D-scatterplot<br />
versus aquifer heterogeneity (Fig. 12).<br />
Fig. 12: Normalized Michaelis-Menten parameters<br />
(given as overestimation factors)<br />
versus aquifer heterogeneity.<br />
In general an overestimation <strong>of</strong> both k max<br />
<strong>and</strong> M C is observed, which increases with<br />
heterogeneity. An overestimation <strong>of</strong> k max increases<br />
the velocity <strong>of</strong> contaminant degradation<br />
as long as concentrations are much<br />
higher than MC. The simultaneous overestimation<br />
<strong>of</strong> MC counterbalances this effect<br />
because the concentration threshold is<br />
raised at which the kinetic begins to show a<br />
dependence on concentration <strong>and</strong> transits<br />
from zero to first order <strong>and</strong> hence decreases<br />
the rate <strong>of</strong> degradation.<br />
To obtain an indicator for the significance<br />
<strong>of</strong> the estimated degradation potential, the<br />
MM parameters determined were used in an<br />
analytical transport model to estimate the<br />
contaminant plume lengths. These then<br />
were compared to the respective true plume<br />
lengths from the <strong>numerical</strong> simulations<br />
(Fig. 13 (a)). As a consequence <strong>of</strong> overestimating<br />
the degradation parameters, calculated<br />
plume lengths for high heterogeneities<br />
are estimated to about 75 % <strong>of</strong> the true<br />
length on average <strong>and</strong> thus are not conservative.<br />
For low heterogeneities, however,<br />
the suggested regression approach on average<br />
yields good estimates <strong>of</strong> the plume<br />
length <strong>and</strong> the degradation potential.<br />
In addition to the effect <strong>of</strong> aquifer heterogeneity<br />
on estimated MM parameters <strong>and</strong> the<br />
resultant plume length estimates, also the<br />
effect <strong>of</strong> a wrong process identification<br />
(compare Fig. 5) is studied in Beyer et al.<br />
(2006 [EP 3]). Although it is well known<br />
that contaminant degradation in natural<br />
aquifers is governed by complex processes<br />
<strong>and</strong> kinetic laws, simple first order models<br />
are routinely used at many field sites. This<br />
study therefore highlights some <strong>of</strong> the<br />
problems that result from an insufficient<br />
wrong process identification. For this end<br />
investigation <strong>of</strong> the plumes following MM<br />
degradation kinetics is repeated, assuming<br />
the appropriateness <strong>of</strong> a first order rate law<br />
to approximate the contaminant degradation<br />
behaviour. Hence the methods <strong>of</strong> Tab. 1<br />
were used to derive first order rate constants<br />
for the multiple plume realizations.<br />
As for the estimated MM parameters the<br />
estimated first order rate constants were<br />
evaluated by analytical transport models to<br />
yield estimates <strong>of</strong> the contaminant plume<br />
length. Results for method 1 (Tab. 1) are<br />
presented in Fig. 13 (b).<br />
17
Fig. 13: Plume length overestimation factors<br />
versus aquifer heterogeneity (Beyer et<br />
al., 2006 [EP 3]). Plume lengths were estimated<br />
for plumes following Michaelis-Menten<br />
degradation kinetics estimated using the<br />
Michaelis-Menten model (a) <strong>and</strong> assuming<br />
validity <strong>of</strong> a first order rate law (b).<br />
In comparison with Fig. 13 (a) an additional<br />
error is introduced which stems from the<br />
first order approximation. Uncertainty as<br />
well as bias increase significantly, as can be<br />
seen by the wider spread <strong>of</strong> single realization<br />
results around the ensemble medians.<br />
Estimated plume lengths here are found to<br />
be less than 40 % <strong>of</strong> the true length on<br />
average even for mildly heterogeneous<br />
aquifers. Plume lengths calculated using the<br />
MM parameters in general are significantly<br />
closer to the correct length compared to<br />
those obtained by a first order approximation.<br />
This approach is therefore recom-<br />
18<br />
mended, if field data collected along the<br />
center line <strong>of</strong> a plume give evidence <strong>of</strong> MM<br />
type degradation kinetics.<br />
3.3. Prognosis <strong>of</strong> long term contaminant<br />
leaching from recycling<br />
materials in road<br />
constructions<br />
The third application <strong>of</strong> the <strong>numerical</strong><br />
models <strong>and</strong> methods presented in section 2<br />
is focussed on the prognosis <strong>of</strong> contaminant<br />
leaching <strong>and</strong> transport by seepage water<br />
from pollutant loaded recycling materials,<br />
which are used in earthworks or road<br />
constructions. According to the German<br />
federal soil protection decree (BBodSchV,<br />
1999) such a prognosis is required for<br />
contaminated sites as well as for constructions<br />
or depositions <strong>of</strong> contaminated materials<br />
in order to assess the extent <strong>and</strong> environmental<br />
impact <strong>of</strong> potential contaminant<br />
leaching through the vadose zone to the<br />
groundwater. In such a prognosis, the<br />
relevant attenuation processes need to be<br />
considered <strong>and</strong> quantified, as significant<br />
contaminant attenuation could result in less<br />
restrictive utilization criteria without<br />
compromising the protection <strong>of</strong> groundwater<br />
resources. For this end, the application<br />
<strong>of</strong> process based <strong>numerical</strong> transport<br />
models is favorable, as complex geometries<br />
<strong>of</strong> the model scenarios <strong>and</strong> possible process<br />
interactions limit the applicability <strong>of</strong> analytical<br />
models or expertise founded “verbalargumentative”<br />
assessments.<br />
In this study, process based type scenario<br />
<strong>modeling</strong> is used as a tool to assess<br />
contaminant leaching from recycled demolition<br />
waste (DW) material. The type<br />
scenarios are based on three different case<br />
studies for the utilization <strong>of</strong> the DW, i.e.<br />
recycling as base <strong>and</strong> subbase layers <strong>of</strong> a<br />
parking lot, a noise protection dam <strong>and</strong> a<br />
road dam (Fig. 14) (Beyer et al., 2007b<br />
[EP 6]). Instead <strong>of</strong> regarding the full spectrum<br />
<strong>of</strong> contaminants typically embodied in<br />
DW three model substances are considered<br />
in the type scenarios: a conservative tracer
1.3m<br />
0.3m<br />
as a representative for highly soluble salts,<br />
naphthalene for moderately sorbing <strong>and</strong><br />
phenanthrene for strongly sorbing organic<br />
compounds. Contaminant leaching from the<br />
DW to the groundwater surface is studied<br />
with six different characteristic subsoil<br />
units <strong>of</strong> Germany (BGR, 2006) to be able to<br />
compare the influence <strong>of</strong> hydraulic <strong>and</strong><br />
basic physico-chemical soil properties on<br />
contaminant attenuation. Fig. 14 presents<br />
the conceptual model for the road dam.<br />
Here coarse grained DW is used as unbound<br />
base / subbase layers below the<br />
asphalt surface <strong>of</strong> the road. According to<br />
German road construction regulations, the<br />
base / subbase layers are covered by low<br />
<strong>and</strong> high permeable soil layers along the<br />
embankment (Fig. 14).<br />
RCB<br />
Körnung 0/32<br />
b horizon<br />
c horizon<br />
symmetry axis<br />
10m 1.5m<br />
asphalt layer<br />
impermeable low permeable soil<br />
3%<br />
4%<br />
12%<br />
1:1. 5<br />
2.3m<br />
10 cm high<br />
permeable soil<br />
2m<br />
groundwater surface<br />
(point <strong>of</strong> compliance)<br />
Fig. 14: Road construction with demolition<br />
waste recycled in base / subbase layers<br />
(Beyer et al., 2007b [EP 6]).<br />
The simulation strategy for this study<br />
combines the Eulerian <strong>and</strong> Lagrangian<br />
frameworks <strong>of</strong> transport <strong>modeling</strong>. Un<strong>saturated</strong><br />
<strong>flow</strong>, i.e. the hydraulics <strong>of</strong> the constructions,<br />
is modeled with GeoSys /<br />
Rock<strong>flow</strong> using st<strong>and</strong>ard FEM. Fig. 15<br />
shows that the two-dimensional scenarios<br />
exhibit complex <strong>flow</strong> patterns under un<strong>saturated</strong><br />
conditions. The velocity vectors at<br />
the element nodes <strong>of</strong> the FEM mesh for the<br />
road dam presented in Fig. 15 indicate<br />
unhindered infiltration <strong>of</strong> water from the<br />
low permeable soil on top <strong>of</strong> the embankment<br />
into the DW material. Along the<br />
sloped material boundary with high permeable<br />
soil on top <strong>of</strong> the coarse DW, however,<br />
strong capillary barrier effects are observed.<br />
These cause a concentration <strong>of</strong> the water<br />
flux on top <strong>of</strong> the DW <strong>and</strong> generation <strong>of</strong><br />
lateral run<strong>of</strong>f, almost completely bypassing<br />
the DW.<br />
Fig. 15: Element nodes <strong>and</strong> velocity vectors<br />
<strong>of</strong> un<strong>saturated</strong> water <strong>flow</strong> in a section <strong>of</strong> the<br />
model domain (Beyer et al., 2007b [EP 6]).<br />
Reactive transport <strong>of</strong> the model compounds<br />
is simulated using the stream tube concept<br />
<strong>of</strong> the SMART code, to take full advantage<br />
<strong>of</strong> the reactive process models implemented<br />
in SMART (degradation, intraparticle<br />
diffusion kinetics, sorption, etc.). Coupling<br />
between GeoSys / Rock<strong>flow</strong> <strong>and</strong> SMART is<br />
achieved through the travel time pdf. These<br />
are generated by simulation <strong>of</strong> conservative<br />
tracer breakthrough curves by GeoSys /<br />
Rock<strong>flow</strong> using the FEM. The pdf are used<br />
as input for the reactive transport simulations<br />
with SMART. The model output <strong>of</strong><br />
SMART for the three model substances at<br />
the groundwater surface represents concentrations<br />
integrated along the cross-section<br />
<strong>of</strong> the contaminant transport path only.<br />
These breakthrough curves for the tracer in<br />
the road dam are displayed in Fig. 16 as<br />
grey curves. The three black curves<br />
represent the same tracer breakthrough<br />
concentrations but are integrated along the<br />
groundwater surface <strong>of</strong> the overall model<br />
domain. Hence, they account for dilution by<br />
uncontaminated seepage water which<br />
bypasses the actual transport path due to the<br />
capillary barrier.<br />
Breakthrough curves in relative concentrations<br />
C/C0 [-] are given here for three <strong>of</strong><br />
the six soils regarded in this study, i.e. for a<br />
cambisol, a podzol <strong>and</strong> a chernozem. Comparing<br />
the concentration breakthrough for<br />
the transport path (grey curves), it is found<br />
that the earliest breakthrough time for the<br />
19
concentration maximum is for the cambisol,<br />
followed by the podzol <strong>and</strong> the chernozem.<br />
In general, however, breakthrough times are<br />
very similar for the three soils. Also the<br />
maximum concentrations observed are <strong>of</strong><br />
comparable magnitude. As the tracer is conservative,<br />
concentration reductions <strong>of</strong> about<br />
70 % can be attributed to the dispersion<br />
process. Integration <strong>of</strong> breakthrough concentrations<br />
along the overall lower model<br />
boundary (black curves) yields a further<br />
reduction <strong>of</strong> concentrations, as about 30 %<br />
<strong>of</strong> the infiltration bypasses the contaminant<br />
transport path.<br />
Fig. 16: Tracer breakthrough curves at the<br />
groundwater surface below the road dam<br />
for three different subsoil types (Beyer et<br />
al., 2007b [EP 6]).<br />
For naphthalene <strong>and</strong> phenanthrene, contaminant<br />
transport is regarded with <strong>and</strong><br />
without degradation. For the cases with<br />
degradation, a simple first order kinetics is<br />
used. Sorption <strong>of</strong> both compounds is<br />
modeled by a linear isotherm, where the<br />
equilibrium sorption coefficient is derived<br />
from the soil organic carbon content <strong>and</strong> the<br />
distribution coefficient between organic<br />
carbon <strong>and</strong> the aqueous phase (Beyer et al.,<br />
2007b [EP 6]). Sorption kinetics are quantified<br />
using the intraparticle diffusion model<br />
(eq. 18). Concentration breakthroughs are<br />
presented in Fig. 17. For the moderately<br />
sorbing naphthalene without decay the influence<br />
<strong>of</strong> soil organic carbon Corg is clearly<br />
20<br />
visible. The cambisol, which is almost free<br />
<strong>of</strong> Corg (0.01 %), shows the earliest breakthrough<br />
<strong>of</strong> the three soils. For the podzol<br />
with a little higher Corg (0.21 %), the maximum<br />
concentration breakthrough is slightly<br />
retarded. The latest breakthrough time is<br />
observed for the chernozem, which is the<br />
soil with the highest Corg (0.88 %) regarded<br />
here. For the strongly sorbing phenanthrene<br />
this behaviour is even more characteristic.<br />
For the chernozem, the maximum concentration<br />
breakthrough is not yet observed<br />
within 275 a <strong>of</strong> simulated contaminant leaching.<br />
In contrast to the tracer, for which<br />
source concentrations are depleted within a<br />
few years, retardation within the DW causes<br />
naphthalene <strong>and</strong> phenanthrene leachate<br />
concentrations to stay on an almost<br />
unreduced level throughout the simulation<br />
period <strong>of</strong> 275 a (Beyer et al., 2007b [EP 6]).<br />
As high contaminant concentrations are<br />
constantly delivered from the source<br />
material, dispersive concentration reductions<br />
remain ineffective. Hence for the<br />
transport path maximum concentration<br />
breakthrough between 70 <strong>and</strong> 90 % <strong>of</strong> C0 is<br />
observed. As for the tracer, these are<br />
reduced by about 30 % when integrated<br />
along the overall lower model boundary.<br />
In comparison to the other type scenarios<br />
(parking lot, noise protection dam) studied<br />
in Beyer et al. (2007b [EP 6]), the contaminant<br />
residence times in the road dam are<br />
rather short, as high amounts <strong>of</strong> run<strong>of</strong>f<br />
water from the road asphalt which infiltrate<br />
along the embankment result in comparably<br />
high <strong>flow</strong> velocities. The short residence<br />
times reduce the effectiveness <strong>of</strong> degradation.<br />
Hence, naphthalene <strong>and</strong> phenanthrene<br />
concentrations are reduced by factors<br />
between 2.5 <strong>and</strong> 5, respectively, while for<br />
the other two type scenarios concentration<br />
reductions by factors between 10 <strong>and</strong> 150<br />
were observed. From this type scenario<br />
<strong>modeling</strong> study the relevant transport <strong>and</strong><br />
attenuation processes for contaminant<br />
leachate from DW used in road constructions<br />
could be identified <strong>and</strong> quantified for<br />
the assumed model structure.
C/C 0 [-]<br />
C/C 0 [-]<br />
1<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0<br />
1<br />
0.1<br />
0.01<br />
0.001<br />
cambisol (dil.)<br />
podzol (dil.)<br />
chernozem (dil.)<br />
cambisol<br />
podzol<br />
chernozem<br />
0 50 100 150 200 250<br />
time [a]<br />
a) naphthalene b) phenanthrene<br />
c) naphthalene<br />
with degradation<br />
d) phenanthrene<br />
with degradation<br />
0 50 100 150 200 250<br />
time [a]<br />
Fig. 17: Concentration breakthrough curves at the groundwater surface below the road dam<br />
for three different subsoil types showing naphthalene without (a) <strong>and</strong> with degradation (c),<br />
<strong>and</strong> phenanthrene without (b) <strong>and</strong> with degradation (d) (Beyer et al., 2007b [EP 6]).<br />
The results allow important conclusions on<br />
mechanisms (e.g. capillary barrier effects)<br />
<strong>and</strong> complementary design criteria for road<br />
constructions, which can be used to reduce<br />
contaminant leaching to groundwater.<br />
4. Conclusions <strong>and</strong> outlook<br />
In this thesis, the utilization <strong>of</strong> process<br />
based <strong>numerical</strong> models <strong>of</strong> <strong>saturated</strong> / un<strong>saturated</strong><br />
<strong>flow</strong> <strong>and</strong> reactive contaminant transport<br />
is demonstrated for two different fields<br />
<strong>of</strong> application.<br />
The Virtual Aquifer (VA) concept is<br />
introduced, which uses <strong>numerical</strong> <strong>modeling</strong><br />
as a tool for the computer based evaluation<br />
<strong>of</strong> investigation <strong>and</strong> remediation strategies<br />
for contaminated soils <strong>and</strong> aquifers. In the<br />
application examples presented, the VA<br />
concept proves its usefulness for assessing<br />
the uncertainty involved in the investigation<br />
<strong>of</strong> heterogeneous sites <strong>and</strong> the parameterization<br />
<strong>of</strong> degradation process models.<br />
Moreover, the VA concept is successfully<br />
applied to test a newly developed approach<br />
for the inference <strong>of</strong> biodegradation parameters<br />
from data typically collected during site<br />
investigation. Both applications exemplify<br />
the importance but also the pitfalls <strong>of</strong> a<br />
careful <strong>and</strong> accurate collection <strong>of</strong> investigation<br />
data, if these are to be used for<br />
prognosis <strong>of</strong> the contaminant behaviour at a<br />
site. The main advantage <strong>of</strong> the VA is, that<br />
individual factors, such as hydraulic heterogeneity<br />
or conceptual model errors can be<br />
studied in detail at low costs <strong>and</strong> without<br />
high effort, either individually or in combination<br />
<strong>and</strong> under otherwise ideal <strong>and</strong><br />
controlled conditions. As the sum <strong>of</strong> these<br />
21
possibilities can not be provided neither by<br />
large scale field experiments nor in the<br />
laboratory, the VA concept can be<br />
considered a valuable contribution complementing<br />
state <strong>of</strong> the art experimental<br />
methods. Future applications <strong>of</strong> the VA<br />
concept will incorporate more realistic degradation<br />
kinetics <strong>and</strong> structures <strong>of</strong> aquifer<br />
heterogeneity, in order to study the<br />
influence <strong>of</strong> <strong>flow</strong> <strong>and</strong> transport channelling<br />
on the effectiveness <strong>of</strong> natural contaminant<br />
attenuation processes.<br />
In the second field <strong>of</strong> interest regarded here,<br />
<strong>numerical</strong> models are applied for an<br />
assessment <strong>of</strong> environmental impact <strong>of</strong><br />
recycling materials used in road constructions.<br />
In this study, the reactive streamtube<br />
model SMART is used for the first time in<br />
combination with the finite element model<br />
GeoSys / Rock<strong>flow</strong> to simulate the extent<br />
<strong>of</strong> contaminant leaching from contaminated<br />
demolition waste within road structures to<br />
the groundwater surface. The coupling<br />
between GeoSys / Rock<strong>flow</strong> <strong>and</strong> SMART<br />
combines the Eulerian <strong>and</strong> Langrangian<br />
frameworks <strong>of</strong> <strong>flow</strong> <strong>and</strong> transport. In so<br />
doing, the complex hydraulic behaviour <strong>of</strong><br />
the two-dimensional model geometries are<br />
successfully modeled by travel time<br />
probability density functions, which allows<br />
to take full advantage <strong>of</strong> the process models<br />
implemented in the SMART code. For the<br />
assessment <strong>of</strong> contaminant leaching consequences<br />
on groundwater recharge quality,<br />
type scenario <strong>modeling</strong> is used. The<br />
relevant transport <strong>and</strong> attenuation processes<br />
in road constructions are identified. The<br />
study allows important conclusions on how<br />
these mechanisms could be used or<br />
enhanced to further reduce contaminant<br />
leaching to groundwater. As an example,<br />
hydraulic processes like capillary barrier<br />
formation from layered composition <strong>of</strong><br />
granular base / subbase materials in a road<br />
dam could be exploited to reduce water<br />
fluxes through contaminant loaded recycling<br />
materials. This study demonstrates<br />
that process based <strong>numerical</strong> models can<br />
provide valuable tools for optimizing<br />
22<br />
design <strong>and</strong> construction with regard to both<br />
functionality <strong>and</strong> environmental impacts.<br />
In as much as the application <strong>of</strong> <strong>numerical</strong><br />
models provides preeminent <strong>and</strong> cost<br />
effective means <strong>of</strong> assessment <strong>and</strong><br />
prognosis, their value <strong>and</strong> credibility relies<br />
on a thorough collection <strong>and</strong> preparation <strong>of</strong><br />
the required parameters <strong>and</strong> input data.<br />
Only if utmost care is taken in the stage <strong>of</strong><br />
setting up the model, reliable results can be<br />
expected. As the data <strong>and</strong> parameter acquisition<br />
is <strong>of</strong> fundamental importance to a<br />
successful <strong>modeling</strong> <strong>of</strong> hydrogeosystems -<br />
be it by direct measurements in laboratory<br />
<strong>and</strong> field experiments, research in literature<br />
<strong>and</strong> databases or utilization <strong>of</strong> expert knowledge<br />
- both processes are closely related<br />
<strong>and</strong> therefore should, if possible, go h<strong>and</strong> in<br />
h<strong>and</strong>.
References<br />
Altfelder, S., Streck, T. (2006): Capability <strong>and</strong> limitations <strong>of</strong> first-order <strong>and</strong> diffusion approaches to<br />
describe long-term sorption <strong>of</strong> chlortoluron in soil. J. Contam. Hydrol., 86, 279-298,<br />
10.1016/j.jconhyd.2006.04.002.<br />
Anderson, M., Woesner, W. (1992): <strong>Applied</strong> groundwater <strong>modeling</strong>. Simulation <strong>of</strong> <strong>flow</strong> <strong>and</strong><br />
advective transport. Academic Press, New York, 381 p.<br />
Baker, A.J. (1983): Finite Element Computational Fluid Mechanics McGraw-Hill Publ., /<br />
Hemisphere, New York, 1983, 544 p.<br />
Ball, W.P., Roberts, P.V. (1991): Long-term sorption <strong>of</strong> halogenated organic chemicals by aquifer<br />
material: 2. Intraparticle diffusion. Environ. Sci. Technol. 25, 1237–1249,<br />
10.1021/es00019a003.<br />
Bauer, S., Kolditz, O. (2006): Assessing contaminant mass <strong>flow</strong> rates by the integral groundwater<br />
investigation method by using the virtual aquifer approach. In: Bierkens, M.F.P., Gehrels, J.C.,<br />
Kovar, K. (eds.), Calibration <strong>and</strong> Reliability in Groundwater Modelling: From Uncertainty to<br />
Decision Making. Proceedings <strong>of</strong> ModelCARE’2005, The Hague, The Netherl<strong>and</strong>s, June 2005.<br />
IAHS Publ. 304, 123–129.<br />
Bauer, S., Beyer, C., Kolditz, O. (2005): Assessing measurements <strong>of</strong> first order degradation rates by<br />
using the virtual aquifer approach. In: Thomson, N.R. (ed.), GQ2004, Bringing Groundwater<br />
Quality Research to the Watershed Scale. Proceedings <strong>of</strong> a Symposium held at Waterloo,<br />
Canada, July 2004. IAHS Publ. 297, 274–281.<br />
Bauer, S., Beyer, C., Kolditz, O. (2006a): Assessing measurement uncertainty <strong>of</strong> first-order<br />
degradation rates in heterogeneous aquifers. Water Resour. Res., 42, W01420,<br />
10.1029/2004WR003878.<br />
Bauer, S., Beyer, C., Kolditz, O. (2007): Einfluss von Heterogenität und Messfehler auf die<br />
Bestimmung von Abbauraten erster Ordnung - eine Virtueller Aquifer Szenarioanalyse.<br />
(Influence <strong>of</strong> heterogeneity <strong>and</strong> measurement error on the determination <strong>of</strong> first order<br />
degradation rates by us1ing the virtual aquifer approach.), Grundwasser, 12, 3–14,<br />
10.1007/s00767-007-0019-8<br />
Bauer, S., Beyer, C., Chen, C., Gronewold, J., Kolditz, O., (2006b): Virtueller Aquifer (VA) -<br />
Computergestützte Bewertung von Erkundungs-, Sanierungs- und Monitoringstrategien im<br />
Hinblick auf das "Natural Attenuation" (NA) und "Enhanced Natural Attenuation" (ENA) -<br />
Potenzial kontaminierter Böden und Grundwässer. Statusseminar des KORA-TV 7, 8.6.2006,<br />
Dresden. Gemeinsame Mitteilungen des DGFZ e.V. und seiner Partner, 3/2006, Dresden, 93-<br />
113.<br />
Baveye, P., Valocchi, A. (1989): An evaluation <strong>of</strong> mathematical models <strong>of</strong> the transport <strong>of</strong><br />
biologically reacting solutes in <strong>saturated</strong> soils <strong>and</strong> aquifers. Water Resour. Res., 25, 1413–1421,<br />
10.1029/89WR00367.<br />
Bear, J. (1972): Dynamics <strong>of</strong> Fluids in Porous Media. American Elsevier Publishing Co., New York,<br />
761 p.<br />
Bear, J. (1979) Hydraulics <strong>of</strong> Groundwater. McGraw-Hill Book Co., London 567 p.<br />
Bear, J., Bachmat, Y. (1990): Introduction to Modeling <strong>of</strong> Transport Phenomena in Porous Media.<br />
Kluwer Academic Publ., 554 p.<br />
BBodSchV (1999): Bundes-Bodenschutz- und Altlastenverordnung vom 16. Juli 1999. Bundesgesetzblatt<br />
Jahrgang 1999, Teil I Nr. 36, 1554-1682.<br />
Bekins, B.A., Warren, E., Godsy, E.M. (1998): A comparison <strong>of</strong> zero-order, first-order, <strong>and</strong> Monod<br />
biotransformation models. Ground Water, 36, 261-268.<br />
Beyer, C., Bauer, S., Kolditz, O. (2006): Uncertainty Assessment <strong>of</strong> Contaminant Plume Length<br />
Estimates in Heterogeneous Aquifers. J. Contam. Hydrol., 87, 73-95,<br />
10.1016/j.jconhyd.2006.04.006.<br />
Beyer, C., Chen, C., Gronewold, J., Kolditz, O., Bauer, S. (2007a): Determination <strong>of</strong> first order<br />
degradation rate constants from monitoring networks. (accepted by Ground Water.).<br />
Beyer, C., Konrad, W., Park, C.H., Bauer, S., Rügner, H., Liedl, R., Grathwohl, P. (2007b):<br />
Modellbasierte Sickerwasserprognose für die Verwertung von Recycling-Baust<strong>of</strong>f in<br />
23
technischen Bauwerken. (Model based prognosis <strong>of</strong> contaminant leaching for reuse <strong>of</strong><br />
demolition waste in construction projects.) (accepted by Grundwasser <strong>and</strong> published online<br />
via SpringerLink), 10.1007/s00767-007-0025-x.<br />
BGR (Bundesanstalt für Geowissenschaften und Rohst<strong>of</strong>fe) (2006): Nutzungsdifferenzierte Bodenübersichtskarte<br />
der Bundesrepublik Deutschl<strong>and</strong> 1:1.000.000 (BÜK 1000 N2.3). Auszugskarten<br />
Acker, Grünl<strong>and</strong>, Wald; Digit. Archiv FISBo BGR; Hannover und Berlin.<br />
Bitterlich, S., Durner, W., Iden, S.C., Knabner, P. (2004): Inverse Estimation <strong>of</strong> the Un<strong>saturated</strong> Soil<br />
Hydraulic Properties from Column Out<strong>flow</strong> Experiments Using Free-Form Parameterizations,<br />
Vadose Zone J., 3, 971-981.<br />
Brooks, R.H., Corey, A.T. (1966): Properties <strong>of</strong> porous media affecting fluid <strong>flow</strong>. Proc. Amer. Soc.<br />
Civ. Eng., 92 [IR2], 61–87.<br />
Brusseau, M.L., Rao, P.S.C. (1989): Sorption nonideality during organic contaminant transport in<br />
porous media. Crit. Rev. Environ. Control, 19, 33–99.<br />
Buscheck, T.E., Alcantar, C.M. (1995): Regression techniques <strong>and</strong> analytical solutions to demonstrate<br />
intrinsic bioremediation. In Hinchee, R.E., Wilson, T.J., Downey, D. (Eds.), Intrinsic<br />
Bioremediation. Battelle Press, Columbus, OH, 109-116.<br />
Carrayrou, J., Mosé, R., Behra, P. (2004): Operator-splitting procedures for reactive transport <strong>and</strong><br />
comparison <strong>of</strong> mass balance errors. J. Contam. Hydrol., 68: 239-268, 10.1016/S0169-<br />
7722(03)00141-4.<br />
Celia, M., Russel, T., Herrera, I. Ewing, R. (1990): An Eulerian-Lagrangian localized adjoint method<br />
for the advection-diffusion equation. Adv. Water Resour., 13, 187-206, 10.1016/0309-<br />
1708(90)90041-2<br />
Cvetkovic, V., Dagan, G. (1996): Reactive transport <strong>and</strong> immiscible <strong>flow</strong> in geological media. 2<br />
Applications.- Proc. Royal Soc. London A, 452, 303-328.<br />
Dagan, G. (1989): Flow <strong>and</strong> Transport in Porous Formations. Springer, Heidelberg. 465 p.<br />
Du, Y., Wang, W. Kolditz, O. (2005): Benchmarking <strong>of</strong> Richards Model. GeoSys – Preprint [2005-<br />
12], GeoSystemsResearch, Center for <strong>Applied</strong> Geosciences, University <strong>of</strong> Tübingen.<br />
Durner, W. (1994): Hydraulic conductivity estimation for soils with heterogeneous pore structure,<br />
Water Resour. Res., 30, 211-223, 10.1029/93WR02676.<br />
Durner, W., Flühler, H. (2005): Chapter 74: Soil Hydraulic Properties, in: Anderson M.G. <strong>and</strong> J. J.<br />
McDonnell (eds.), Encyclopedia <strong>of</strong> Hydrological Sciences, John Wiley & Sons, Ltd., Chapter<br />
74, 1103-1120,<br />
Fetter, C.W. (1993): Contaminant Hydrogeology, Prentice Hall, New Jersey, 500 p.<br />
Finkel, M., Liedl, R., Teutsch, G. (1998): Modelling surfactant-enhanced remediation <strong>of</strong> polycyclic<br />
aromatic hydrocarbons. J. Environ. Modelling & S<strong>of</strong>tware, 14, 203-211. 10.1016/S1364-<br />
8152(98)00071-1.<br />
Freeze, R.A., Cherry, J.A. (1979): Groundwater. Prentice Hall, Englewood Cliffs, New Jersey, 604 p.<br />
Grathwohl, P. (1998): Diffusion in Natural Porous Media: Contaminant Transport,<br />
Sorption/Desorption <strong>and</strong> Dissolution Kinetics. Kluwer Academic Publishers, 224 p.<br />
Haverkamp, R., Vaclin, M., Touma, J., Wierenga, P.J., Vachaud, G. (1977): A comparison <strong>of</strong><br />
<strong>numerical</strong> simulation models for one-dimensional infiltration. Soil Sci. Soc. Am. J., 41, 285-<br />
294.<br />
Helmig, R. (1997): Multiphase Flow <strong>and</strong> Transport Processes in the Subsurface: A Contribution to<br />
the Modeling <strong>of</strong> Hydrosystems. Springer, Berlin, Heidelberg, 255 p.<br />
Huyakorn, P.S., Pinder, G.F. (1983): Computational Methods in Subsurface Flow. Academic Press,<br />
New York, 473 p.<br />
Islam, J., Naresh, S., O’Sullivan, M., (2001): Modeling biogeochemical processes in leachatecontaminated<br />
soils: A review. Transp. Porous Media, 43, 407-440, 10.1023/A:1010737825232.<br />
Jäger, R., Liedl, R. (2000): Prognose der Sorptionskinetik organischer Schadst<strong>of</strong>fe in heterogenem<br />
Aquifermaterial. Grundwasser, 2, 57-66.<br />
Jury, W.A., Gardner, W.R., Gardner, W.H. (1991): Soil Physics. John Wiley & Sons, New York,<br />
328 p.<br />
Kinzelbach, W. (1983): Analytische Lösungen der Schadst<strong>of</strong>ftransportgleichung und ihre Anwendung<br />
auf Schadensfälle mit flüchtigen Chlorkohlenwasserst<strong>of</strong>fen. In: Methoden zur rechnerischen<br />
24
Erfassung und hydraulischen Sanierung von Grundwasserkontaminationen, Mitteilungen des<br />
Instituts für Wasserbau, Universität Stuttgart, Heft 54, 115-199.<br />
Kinzelbach, W., Schäfer, W., Herzer, J. (1991): Numerical <strong>modeling</strong> <strong>of</strong> natural <strong>and</strong> enhanced<br />
denitrification processes in aquifers. Water Resour. Res., 27, 1123-1136, 10.1029/91WR00474.<br />
Knabner, P., Schneid, E. (2002): Adaptive Hybrid Mixed Finite Element Discretization <strong>of</strong><br />
Instationary Variably Saturated Flow in Porous Media. In: Breuer, M.,. et al., (eds.), High<br />
Performance Scientific <strong>and</strong> Engineering Computing, Springer Verlag, Berlin, 37-44.<br />
Korsawe, J., Starke, G., Wang, W., Kolditz, O. (2006): Finite Element Analysis <strong>of</strong> Poro-Elastic<br />
Consolidation in Porous Media: St<strong>and</strong>ard <strong>and</strong> Mixed Approaches. Comput. Methods Appl.<br />
Mech. Engrg., 195, 1096-1115, 10.1016/j.cma.2005.04.011<br />
Kolditz, O. (2002): Computational methods in environmental fluid dynamics. Springer, Heidelberg,<br />
400 p.<br />
Kolditz, O., Bauer, S. (2004): A process-oriented approach to computing multi-field problems in<br />
porous media. J. Hydroinf., 6, 225-244.<br />
Kolditz, O., Xie, M., Kalbacher, T., Bauer, S., Wang, W., McDermott, C., Chen, C., Beyer, C.,<br />
Gronewold, J., Kemmler, D., Walsh, R., Du, Y., Park, C.H., Hess, M., Amanidis, P. (2006):<br />
GeoSys/Rock<strong>flow</strong> version 4.4.03 - Theory <strong>and</strong> users manual, Center for <strong>Applied</strong> Geoscience,<br />
University <strong>of</strong> Tübingen.<br />
Ma, L., Selim, H.M. (1994): Predicting atrazine adsorption–desorption in soils: a modified secondorder<br />
kinetic model. Water Resour. Res., 30, 447–456. 10.1029/93WR02478<br />
Morris, B., Lawrence, A., Chilton, P., Adams, B., Calow, R., Klinck, B. (2003): Groundwater <strong>and</strong> its<br />
susceptibility to degradation: a global assessment <strong>of</strong> the problem <strong>and</strong> options for management.<br />
Nairobi: United Nations Environmental Programme, 140 p.<br />
Mualem, Y. (1976): A New model for predicting the hydraulic conductivity <strong>of</strong> un<strong>saturated</strong> porous<br />
media, Water Resour. Res., 12, 513 - 522.<br />
Neuman, S. (1981): An Eulerian-Lagrangian <strong>numerical</strong> scheme for the dispersion convection<br />
equation using conjugate space time grids, J. Comput. Phys. 41, 270-285.<br />
Newell, C.J., Rifai, H.S., Wilson, J.T., Connor, J.A., Aziz, J.A., Suarez, M.P. (2002): Calculation <strong>and</strong><br />
use <strong>of</strong> first-order rate constants for monitored natural attenuation studies. U.S. EPA Ground<br />
Water Issue, U.S. EPA/540/S-02/500.<br />
Olsen, S.R., Kemper, W.D. (1968): Movement <strong>of</strong> nutrients to plant roots. Adv. Agronomy, 30, 91-<br />
151.<br />
Park, C.H., Beyer, C., Bauer, S., Kolditz, O. (2006): An effcient method <strong>of</strong> r<strong>and</strong>om walk particle<br />
tracking: accuracy <strong>and</strong> resolution. (submitted to Adv. Water Resour.)<br />
Pinder, G.F., Gray, W.G. (1977): Finite element simulation in surface <strong>and</strong> subsurface hydrology.<br />
Academic Press, New York, 295 p.<br />
Rittmann, B., VanBriesen, J.M. (1996): Microbiological processes in reactive <strong>modeling</strong>. In: Lichtner,<br />
P., Steefel, C., Oelkers, E. (eds.), Reactive Transport in Porous Media. Reviews in Mineralogy,<br />
34, Mineralogical Society <strong>of</strong> America, Washington, DC, 311-334.<br />
Rügner, H., Kleineidam, S., Grathwohl, P. (1999): Long term sorption kinetics <strong>of</strong> phenanthrene in<br />
aquifer materials. Environ. Sci. Technol., 33, 1645–1651, 10.1021/es980664x<br />
Rubin, Y. (1983): Transport <strong>of</strong> reacting solutes in porous media: Relationship between mathematical<br />
nature <strong>of</strong> problem formulation <strong>and</strong> chemical nature <strong>of</strong> reactions. Water Res. Resour., 19, 1231-<br />
1252.<br />
Schäfer, D., Schäfer, W., Kinzelbach, W. (1998): Simulation <strong>of</strong> Reactive Processes Related to<br />
Biodegradation in Aquifers. 1. Structure <strong>of</strong> the 3D Reactive Transport Model, J. Contam.<br />
Hydrol., 31, 167-186, 10.1016/S0169-7722(97)00060-0<br />
Schäfer, D., Schlenz, B. Dahmke, A. (2004): Evaluation <strong>of</strong> exploration <strong>and</strong> monitoring methods for<br />
verification <strong>of</strong> natural attenuation using the virtual aquifer approach. Biodegradation, 15, 453-<br />
465, 10.1023/B:BIOD.0000044600.81216.00.<br />
Schäfer, D., Schlenz, B., Dahmke A. (2006a): Virtuelle Aquifere – ein Werkzeug zur Simulation von<br />
Natural Attenuation und zur Bewertung von Monitoringstrategien. Statusseminar des KORA-<br />
TV 7, 8.6.2006, Dresden. Gemeinsame Mitteilungen des DGFZ e.V. und seiner Partner, 3/2006,<br />
Dresden, 115-139.<br />
25
Schäfer, D., Dahmke, A., Kolditz, O., Teutsch, G. (2002): "Virtual Aquifers": A concept for<br />
evaluation <strong>of</strong> exploration, remediation <strong>and</strong> monitoring strategies. In: Kovar, K., Hrkal, Z. (eds.),<br />
Calibration <strong>and</strong> Reliability in Groundwater Modelling: A Few Steps Closer to Reality.<br />
Proceedings <strong>of</strong> the ModelCARE 2002 Conference, Prague, Czech Republic, June 2002. IAHS<br />
Publication 277, 52-59.<br />
Schäfer, D., Hornbruch, G. Schlenz, B., Dahmke A. (2006b): Schadst<strong>of</strong>fausbreitung unter Annahme<br />
verschiedener kinetischer Ansätze zur Modellierung mikrobiellen Abbaus. Grundwasser (in<br />
print).<br />
Scheidleder, A., Grath, J., Winkler, G., Stärk., U, Koreimann., C, Gmeiner, C., Nixon, S., Casillas, J.,<br />
Gravesen, P., Leonard, J., Elvira, M. (1999): Groundwater quality <strong>and</strong> quantity in Europe.<br />
European Environment Agency, Copenhagen, 123 p.<br />
Simkins, S., Alex<strong>and</strong>er, M. (1984): Models for mineralization kinetics with variables <strong>of</strong> substrate<br />
concentration <strong>and</strong> population density. Appl. Envir. Microbiol., 47, 1299-1306.<br />
Starke, G. (2000): Least-squares mixed finite element solution <strong>of</strong> variably <strong>saturated</strong> subsurface <strong>flow</strong><br />
problems. SIAM J. Sci. Comput. 21, 1869-1885, 10.1137/S1064827598339384<br />
Stenback, G.A., Ong, S.K., Rogers, S.W., Kjartanson, B.H. (2004): Impact <strong>of</strong> transverse <strong>and</strong><br />
longitudinal dispersion on first-order degradation rate constant estimation. J. Contam. Hydrol.,<br />
73, 3-14, 10.1016/j.jconhyd.2003.11.004.<br />
Streck, T., Poletika, N.N., Jury, W.A. Farmer, W.J. (1995): Description <strong>of</strong> simazine transport with<br />
rate-limited, two-stage linear <strong>and</strong> nonlinear sorption. Water Resour. Res., 31, 811-822,<br />
10.1029/94WR02822<br />
Thorenz, C. (1999): Model Adaptive Simulation <strong>of</strong> Multiphase <strong>and</strong> Density Driven Flow in Fractured<br />
<strong>and</strong> Porous Media. Dissertation, Institut für Strömungsmechanik, Universität Hannover, Bericht<br />
Nr. 62/2001.<br />
Van Genuchten, M.T. (1980): A closed form equation for predicting the hydraulic conductivity <strong>of</strong><br />
un<strong>saturated</strong> soils. Soil Sci. Soc. Am. J., 44, 892-898.<br />
Van Genuchten, M.T., Alves, W.J. (1982): Analytical solutions <strong>of</strong> the one-dimensional convective<br />
dispersive solute transport equation. USDA ARS Technical Bulletin Number 1661. U.S.<br />
Salinity Laboratory, 4500 Glenwood Drive, Riverside, CA 92501.<br />
Wang, W., Datcheva, M., Schanz, T., Kolditz, O. (2006): A sub-stepping approach for elastoplasticity<br />
with rotational hardening. Computational Mechanics. (in print). 10.1007/s00466-005-<br />
0710-5<br />
Wiedemeier, T.H., Swanson, M.A., Wilson, J.T., Kampbell, D.H., Miller, R.N., Hansen, J.E. (1996):<br />
Approximation <strong>of</strong> biodegradation rate constants for monoaromatic hydrocarbons (BTEX) in<br />
ground water. Ground Water Monit. Remed., 16 (3),186-194.<br />
Wiedemeier, T.H., Rifai, H.S., Newell, C.J., Wilson, J.T. (1999): Natural Attenuation <strong>of</strong> Fuels <strong>and</strong><br />
Chlorinated Solvents in the Subsurface. John Wiley <strong>and</strong> Sons, New York, 632 p.<br />
Wilson, R.D., Thornton, S.F., Mackay, D.M. (2004): Challenges in monitoring the natural attenuation<br />
<strong>of</strong> spatially variable plumes, Biodegradation, 15, 459-469,<br />
10.1023/B:BIOD.0000044591.45542.a9<br />
Xie, M., Bauer, S., Kolditz, O., Nowak, T., Shao, H. (2006): Numerical simulation <strong>of</strong> reactive<br />
processes in an experiment with partially <strong>saturated</strong> bentonite, J. Contam. Hydrol., 83, 122-147.<br />
10.1016/j.jconhyd.2005.11.003<br />
Zhang, Y.-K., Heathcote, R.C. (2003): An improved method for estimation <strong>of</strong> biodegradation rate<br />
with field data. Ground Water Monit. Remed., 23 (3), 112-116.<br />
Zheng, C., Bennett, G.D., (1995): <strong>Applied</strong> Contaminant Transport Modeling: Theory <strong>and</strong> Practice.<br />
Van Nostr<strong>and</strong> Reinhold, New York, 440 p.<br />
Zienkiewicz, O.C., Taylor, R. (1991): The Finite Element Method, Vol. 1, 2 (4th edition). McGraw-<br />
Hill, London, 459 p.<br />
26
Enclosed publications<br />
[EP 1] Bauer, S., Beyer, C., Kolditz, O., (2005): Assessing measurements <strong>of</strong> first order<br />
degradation rates by using the Virtual Aquifer approach. In: Thomson, N.R. (Ed.), GQ2004,<br />
Bringing Groundwater Quality Research to the Watershed Scale. Proceedings <strong>of</strong> a<br />
Symposium held at Waterloo, Canada, July 2004. IAHS Publication 297, IAHS Press,<br />
Wallingford, 274-281.<br />
[EP 2] Bauer, S., Beyer, C., Kolditz, O. (2006a): Assessing measurement uncertainty <strong>of</strong> first-order<br />
degradation rates in heterogeneous aquifers. Water Resour. Res., 42, W01420,<br />
10.1029/2004WR003878.<br />
[EP 3] Beyer, C., Bauer, S., Kolditz, O. (2006): Uncertainty assessment <strong>of</strong> contaminant plume<br />
length estimates in heterogeneous aquifers. J. Contam. Hydrol., 87, 73-95,<br />
[EP 4]<br />
10.1016/j.jconhyd.2006.04.006.<br />
Bauer, S., Beyer, C., Kolditz, O. (2007): Einfluss von Heterogenität und Messfehler auf die<br />
Bestimmung von Abbauraten erster Ordnung - eine Virtueller Aquifer Szenarioanalyse.<br />
(Influence <strong>of</strong> heterogeneity <strong>and</strong> measurement error on the determination <strong>of</strong> first order<br />
degradation rates by us1ing the virtual aquifer approach.), Grundwasser, 12, 3–14,<br />
10.1007/s00767-007-0019-8.<br />
[EP 5] Beyer, C., Chen, C., Gronewold, J., Kolditz, O., Bauer, S. (2007a): Determination <strong>of</strong> first<br />
order degradation rate constants from monitoring networks. (accepted by Ground Water.).<br />
[EP 6] Beyer, C., Konrad, W., Park, C.H., Bauer, S., Rügner, H., Liedl, R., Grathwohl, P. (2007b):<br />
Modellbasierte Sickerwasserprognose für die Verwertung von Recycling-Baust<strong>of</strong>f in<br />
technischen Bauwerken. (Model based prognosis <strong>of</strong> contaminant leaching for reuse <strong>of</strong><br />
demolition waste in construction projects.) (accepted by Grundwasser <strong>and</strong> published online<br />
via SpringerLink), 10.1007/s00767-007-0025-x.
Enclosed Publication 1<br />
Bauer, S., Beyer, C., Kolditz, O. (2005): Assessing measurements <strong>of</strong> first order degradation<br />
rates by using the Virtual Aquifer approach. In: Thomson, N.R. (Ed.), GQ2004, Bringing<br />
Groundwater Quality Research to the Watershed Scale. Waterloo, Canada, July 2004. IAHS<br />
Publication 297, IAHS Press, Wallingford, 274-281.<br />
The enclosed article was reproduced <strong>and</strong> is made available with the permission <strong>of</strong> IAHS<br />
Press.<br />
It can be obtained from IAHS Press at http://www.cig.ensmp.fr/~iahs/redbooks/297.htm.
274<br />
INTRODUCTION<br />
Bringing Groundwater Quality Research to the Watershed Scale (Proceedings <strong>of</strong> GQ2004, the 4th International<br />
Groundwater Quality Conference, held at Waterloo, Canada, July 2004). IAHS Publ. 297, 2005.<br />
Assessing measurements <strong>of</strong> first-order degradation<br />
rates through the virtual aquifer approach<br />
SEBASTIAN BAUER, CHRISTOF BEYER & OLAF KOLDITZ<br />
Center for <strong>Applied</strong> Geoscience, University <strong>of</strong> Tübingen, Sigwartstrasse 10, D 72076 Tübingen,<br />
Germany<br />
sebastian.bauer@uni-tuebingen.de<br />
Abstract The principal idea behind the “virtual aquifer” is to simulate <strong>and</strong><br />
evaluate investigation strategies for contaminated sites by modelling typical<br />
contamination scenarios. In this paper, first-order degradation rates using<br />
various methods were the focus <strong>of</strong> study. A virtual reality <strong>of</strong> a contaminated<br />
aquifer was generated by simulating the spreading <strong>of</strong> a plume, originating<br />
from a defined source zone, subject to first-order degradation. This plume was<br />
investigated through monitoring wells placed along the plume centre-line.<br />
Using information such as head measurements, concentration <strong>and</strong> hydraulic<br />
conductivity, first-order degradation rates were calculated <strong>and</strong> compared to the<br />
true predefined value. This comparison was conducted for varying degrees <strong>of</strong><br />
heterogeneity, represented by ln(KF), r<strong>and</strong>omly distributed conductivity fields.<br />
It was found that when heterogeneity was increased, “measured” degradation<br />
rates overestimated the true degradation rate by several orders <strong>of</strong> magnitude.<br />
The range <strong>of</strong> degradation rates obtained roughly corresponds to the range<br />
stated in literature values.<br />
Key words first-order degradation; modelling; natural attenuation; virtual reality<br />
At a real contaminated site, the true hydrogeological properties (e.g. conductivity,<br />
porosity, recharge rates, source position, degradation rates, etc.) are generally unknown<br />
(spatially). The basic idea behind the “virtual aquifer” is to produce a “virtual” contaminated<br />
site, where the spatial distribution <strong>of</strong> parameters is exactly known. By using a<br />
process-based <strong>flow</strong> <strong>and</strong> transport model, the fate <strong>of</strong> contaminants in the subsurface can<br />
be simulated, including plume development. The second step for creating a virtually<br />
contaminated site is to examine the virtual aquifer properties using st<strong>and</strong>ard investigative<br />
procedures (e.g. interpolating hydraulic head <strong>and</strong> contaminant concentrations<br />
measured at monitoring wells). The result <strong>of</strong> this virtual investigation can be compared<br />
to the true hydrogeological property distribution <strong>of</strong> the virtual aquifer because, contrary<br />
to an actual contaminated site, the exact distribution is known. The investigation<br />
techniques used can thus be tested <strong>and</strong> evaluated with respect to certain influences, i.e.<br />
sensitivity to aquifer heterogeneity or variation <strong>of</strong> other parameters. When the virtual<br />
plume is examined, only the data obtained by these investigation techniques are used,<br />
i.e. only the data that would also be measured at a real field site. Data such as hydraulic<br />
head <strong>and</strong> contaminant concentrations are “measured” in the virtual aquifer by “reading”<br />
the model output. The “virtual aquifer” approach <strong>of</strong>fers the ability to test <strong>and</strong> evaluate<br />
site investigation techniques, which cannot be performed in the real world. In this<br />
paper, four methods for determining first-order degradation rate constants, all based on<br />
the plume centre-line method, are examined by the “Virtual Aquifer” approach.
Assessing measurements <strong>of</strong> first-order degradation rates through the virtual aquifer approach 275<br />
METHODS<br />
Virtual aquifers were produced by generating r<strong>and</strong>om fields <strong>of</strong> hydraulic conductivity<br />
for the model area (dimensions: 184 × 64 m, Fig. 1). Flow direction was from left to<br />
right, with a mean hydraulic gradient <strong>of</strong> 0.003. For this application, a mean hydraulic<br />
conductivity <strong>of</strong> 7.2 × 10 -5 m s -1 was assumed, with ln(KF) variances <strong>of</strong> 0.38, 1.71, 2.7<br />
<strong>and</strong> 4.5; the variances were chosen to simulate different degrees <strong>of</strong> heterogeneity. An<br />
exponential variogram model, with an integral scale <strong>of</strong> 2.33 m, was used to describe<br />
spatial correlation. A virtual contaminant source zone <strong>of</strong> widths 4, 8 <strong>and</strong> 16 m was<br />
introduced into the aquifer, emitting a contaminant which was subject to a first-order<br />
degradation constant, λ, <strong>of</strong> 1 year -1 . A conservative tracer was also released. By using<br />
first order degradation kinetics for the reactive contaminant, the plume evolving<br />
corresponds to the methods used to estimate the first order degradation rate. This is<br />
certainly not true in reality, where the degradation <strong>of</strong> a contaminant follows changing<br />
<strong>and</strong> more complex kinetics. The plume was simulated using a process-based <strong>numerical</strong><br />
<strong>flow</strong> <strong>and</strong> transport model, assuming longitudinal <strong>and</strong> transverse dispersivities <strong>of</strong> 0.25<br />
<strong>and</strong> 0.05, respectively. Thus, the virtual contaminated aquifer was generated. In the<br />
second step, the plume’s properties were examined using the centre-line approach. For<br />
this investigation, not the full data <strong>of</strong> the model is used but only the values which are<br />
obtained by the centreline approach. There were three initial observation wells; one<br />
was directly in the source zone, while the other two were outside <strong>of</strong> the source (Fig. 1).<br />
At these three wells, the hydraulic heads were measured by reading the model output.<br />
A hydrogeological triangle was constructed <strong>and</strong> the direction <strong>of</strong> groundwater <strong>flow</strong> was<br />
thus determined. Along the estimated direction <strong>of</strong> groundwater <strong>flow</strong>, new observation<br />
wells were installed at every 10 m. These wells were then used to measure (arrows in<br />
Fig. 1) hydraulic head, contaminant concentrations, tracer concentrations <strong>and</strong> local<br />
hydraulic conductivity. From the hydraulic head difference, the true porosity <strong>and</strong> the<br />
well positions the respective groundwater <strong>flow</strong> velocities are calculated. Together with<br />
the concentration data, this allows the determination <strong>of</strong> the degradation rate constant<br />
Fig. 1 Method used to obtain plume centreline concentrations. The upper graph<br />
depicts the investigation procedure <strong>and</strong> the lower graph is the virtual site.<br />
virtual reality investigation
276<br />
S. Bauer et al.<br />
Table 1 Methods used for calculating first-order rate constants λ. va is the transport velocity, ∆x is the<br />
observation well distance, C(x) is the downstream concentration <strong>and</strong> C0 the source concentration, while<br />
αL <strong>and</strong> αT are the longitudinal <strong>and</strong> transverse dispersivities, WS is the source width <strong>and</strong> erf is the error<br />
function.<br />
Method Formula for degradation rate Description<br />
1<br />
va � C(<br />
x)<br />
�<br />
λ1<br />
= − ln� �<br />
∆x<br />
� �<br />
� C0<br />
�<br />
Analytical solution to 0-D transport equation<br />
(batch reactor)<br />
2<br />
v � ∗ �<br />
a � C(<br />
x)<br />
C0<br />
λ = −<br />
�<br />
2 ln<br />
∆x<br />
� C ∗ �<br />
� 0 C(<br />
x)<br />
�<br />
Concentration normalized to a non-degrading cocontaminant,<br />
thus accounting for dilution <strong>and</strong><br />
dispersion<br />
3<br />
�<br />
2<br />
v<br />
( ) �<br />
a ��<br />
ln C(<br />
x)<br />
C0<br />
�<br />
λ = � − α<br />
� −1�<br />
3 1 2 L<br />
4α<br />
�<br />
�<br />
L ��<br />
∆x<br />
� �<br />
Analytical solution to the 1-D transport equation.<br />
Accounts for longitudinal dispersion.<br />
4<br />
�<br />
2<br />
v<br />
( ) �<br />
a ��<br />
ln C(<br />
x)<br />
( C0β)<br />
�<br />
λ = � − α<br />
� −1�<br />
3 1 2 L<br />
4α<br />
�<br />
�<br />
L ��<br />
∆x<br />
� �<br />
Analytical solution to the 2-D transport equation.<br />
Accounts for longitudinal as well as transverse<br />
dispersion <strong>and</strong> a finite source width.<br />
� �<br />
with: � WS<br />
β = erf �<br />
� �<br />
� 4 αT<br />
∆x<br />
�<br />
first-order degradation; modelling; natural attenuation; virtual reality. The setup is thus<br />
designed to resemble ideal conditions for the application <strong>of</strong> the four methods for<br />
estimating the degradation rate constant. The only uncertainty <strong>and</strong> variability is<br />
introduced by the aquifer heterogeneity. The data used for the centreline method is<br />
thus only the data which can be measured at the three initial <strong>and</strong> the three downstream<br />
wells. Thus neither the mean hydraulic conductivity nor the variance or correlation<br />
length is known.<br />
The four centre-line methods used are provided in Table 1. Method 1 (Wiedemeier<br />
et al., 1996) is the batch solution to a first-order degradation (i.e. no transport is<br />
included). Method 2 was proposed by Wilson et al. (1994) (see also Wiedemeier et al.,<br />
1996) <strong>and</strong> is similar to Method 1, except amended concentrations are used. The<br />
measured concentrations for the reactive contaminant are corrected by using the ratio<br />
<strong>of</strong> the conservative co-contaminant at the observation well. This method corrects for<br />
dispersion <strong>and</strong> measurements taken outside <strong>of</strong> the plume centre-line at the three new<br />
observation wells. Method 3 was proposed by Buscheck & Alcantar (1995) <strong>and</strong> is<br />
based on the solution <strong>of</strong> a one-dimensional (1-D) transport equation with a first-order<br />
decay constant. This method accounts explicitly for longitudinal dispersion <strong>of</strong> the<br />
plume. To account for transverse dispersion (Method 4), Stenback et al. (2004) suggest<br />
using the analytical solution for a 2-D transport equation with first-order decay. In<br />
order to carry out Methods 3 <strong>and</strong> 4, it is required that the longitudinal <strong>and</strong> the transverse<br />
dispersivities be known. The following dispersivities were used according to<br />
Wiedemeier et al. (1999): 0.1 <strong>of</strong> the plume length for the longitudinal dispersivity, <strong>and</strong><br />
0.33 <strong>of</strong> longitudinal dispersivity for the transverse dispersivity. For each <strong>of</strong> the four<br />
approaches, first-order degradation rates were calculated <strong>and</strong> compared to the value<br />
used in generating the plume. For each degree <strong>of</strong> heterogeneity, 100 realizations were<br />
evaluated to obtain a statistical measure <strong>of</strong> the error introduced by the heterogeneity <strong>of</strong>
Assessing measurements <strong>of</strong> first-order degradation rates through the virtual aquifer approach 277<br />
the hydraulic conductivity. For each realization, the procedures described above were<br />
followed <strong>and</strong> a degradation rate was calculated for each method, at each downstream<br />
well <strong>and</strong> for every source width.<br />
RESULTS AND DISCUSSION<br />
Figure 2 illustrates the results for the calculated first-order rate constants. Calculated rate<br />
constants were reported as normalized rate constants (i.e. the calculated rate constant<br />
was divided by the true rate constant used in the <strong>numerical</strong> simulation). The normalized<br />
rate constant can thus be interpreted as an overestimated factor or an underestimated<br />
factor. Inspection <strong>of</strong> Fig. 2 yields that most calculated rates are higher than one (i.e. the<br />
degradation rate is overestimated). This conclusion is quite concerning for single<br />
realizations, where overestimations can be <strong>of</strong> several orders <strong>of</strong> magnitude. On the lefth<strong>and</strong><br />
side <strong>of</strong> Fig. 2(a), the variation <strong>of</strong> the calculated normalized rate with the source<br />
zone width is illustrated. It is clear that for Method 1, the calculated rates improve when<br />
the source zone width is increased; this is because Method 1 does not account for<br />
dilution, dispersion or measurements outside <strong>of</strong> the plume. These factors become less<br />
relevant with increasing source width since the basic assumptions inherent in Method 1<br />
are better fulfilled, <strong>and</strong> the overestimation factor drops accordingly. On the right-h<strong>and</strong><br />
side <strong>of</strong> Fig. 2(a), the dependence <strong>of</strong> the calculated rate on the degree <strong>of</strong> heterogeneity<br />
(given as variance) is shown. It is obvious that an increase <strong>of</strong> σ²ln(KF) leads to an<br />
overestimation <strong>of</strong> the calculated degradation rate. Furthermore, the st<strong>and</strong>ard deviation <strong>of</strong><br />
the mean calculated degradation rate increases, leading to greater uncertainty in the<br />
calculated rate. For the smallest degree <strong>of</strong> heterogeneity the mean overestimation is a<br />
factor a bit smaller than 2, which increases to values between 3 <strong>and</strong> 5 for medium to<br />
high heterogeneity, <strong>and</strong> 10 for very high heterogeneity.<br />
Figure 2(b) shows the results for Method 2. Degradation rates for this method were<br />
also overestimated. However, when comparing this method to Method 1, the overestimation<br />
factor <strong>and</strong> st<strong>and</strong>ard deviation are generally smaller (i.e. both the error <strong>and</strong><br />
the uncertainty are lower compared to Method 1). When examining the left-h<strong>and</strong> side<br />
<strong>of</strong> Fig. 2(b), the calculated rates show no dependence on source width. This effect is<br />
inherent to the method, since Method 2 accounts for dispersion, dilution <strong>and</strong><br />
measurements taken outside <strong>of</strong> the plume. Method 3 depicts results similar to Method<br />
1 regarding the rate dependence on source width <strong>and</strong> on the degree <strong>of</strong> heterogeneity<br />
(Fig. 2(c)). However, the normalized degradation rates for Method 3 are higher than<br />
for Method 1 (<strong>and</strong> also higher than Method 2); this is due to the dispersivity term. In<br />
comparison to Method 1, a portion <strong>of</strong> the concentration reduction from the source<br />
observation well to the downstream observation well is attributed to dispersion <strong>and</strong><br />
corrected for, <strong>and</strong> thus a higher degradation rate is estimated. Method 4 (Fig. 2(d))<br />
displays behaviour similar to Method 3, except that the rate values are slightly lower.<br />
Lower rate values are attributed to the additional term in the rate equation, which<br />
accounts for transverse dispersion. It should also be noted that at the lowest degree <strong>of</strong><br />
heterogeneity, the normalized degradation rates were actually underestimated; this is<br />
due to an “over correction” <strong>of</strong> the effects for transverse dispersion. To effectively<br />
illustrate the over <strong>and</strong> underestimation <strong>of</strong> the degradation rates for all four methods,<br />
the degradation rates where calculated (for a homogeneous hydraulic conductivity) <strong>and</strong>
278<br />
(a) Method 1<br />
(b) Method 2<br />
(c) Method 3<br />
(d) Method 4<br />
S. Bauer et al.<br />
Fig. 2 “Measured” first-order degradation rate constants normalized to the true<br />
degradation rate constant vs source width (left) <strong>and</strong> degree <strong>of</strong> heterogeneity (right) for<br />
(a) Method 1, (b) Method 2, (c) Method 3 <strong>and</strong> (d) Method 4. All figures show results<br />
for all observations (small symbols) as well as their mean value (large symbols) <strong>and</strong><br />
the corresponding st<strong>and</strong>ard deviations (error bars).
Assessing measurements <strong>of</strong> first-order degradation rates through the virtual aquifer approach 279<br />
plotted as small horizontal bars for σ²ln(KF) <strong>of</strong> 0 on the right-h<strong>and</strong> side graphs <strong>of</strong><br />
Fig. 2. It is obvious that for Method 1 <strong>and</strong> Method 2, the normalized degradation rate<br />
is exactly 1 for the homogeneous case (i.e. these methods yield the correct result).<br />
Method 3 shows a slight overestimation, while Method 4 yields very small, normalized<br />
degradation rates. The smallest rate, from Method 4, was obtained for the smallest<br />
source width, as then the correction is largest (compare Method 4 in Table 1). For large<br />
source widths, the argument <strong>of</strong> the error function <strong>of</strong> method 4 approaches 1.<br />
As mean <strong>and</strong> st<strong>and</strong>ard deviations are true for the ensemble mean, but not for single<br />
observations, an alternative method was chosen for comparing the four methods. The<br />
four methods were contrasted by plotting the probability <strong>of</strong> success against the error<br />
factor; the results are illustrated in Fig. 3. An error factor <strong>of</strong> 10 corresponds to an<br />
interval <strong>of</strong> 0.1 to 10 for normalized degradation rates, i.e. the interval that is obtained<br />
by multiplying 1 with the error factor <strong>and</strong> dividing 1 by the error factor (“within one<br />
order <strong>of</strong> magnitude”). An error factor <strong>of</strong> 5 thus corresponds to an interval <strong>of</strong> 0.2 to 5.<br />
These plots illustrate the probability that the measured degradation rate is within the<br />
interval <strong>of</strong> the corresponding error factor. Figure 3(a) represents the lowest degree <strong>of</strong><br />
heterogeneity <strong>and</strong> shows that the probability <strong>of</strong> calculating the degradation rate with an<br />
error factor <strong>of</strong> less than 2 (i.e. “within a factor <strong>of</strong> 2”) is about 0.7 for Method 1, 0.9 for<br />
Method 2, 0.55 for Method 3 <strong>and</strong> 0.3 for Method 4. When increasing the error factor<br />
(a) (b)<br />
(c) (d)<br />
Fig. 3 Probability plots for all methods for σ²ln(KF) <strong>of</strong>: (a) 0.38, (b) 1.71, (c) 2.7 <strong>and</strong><br />
(d) 4.5, resembling the four degrees <strong>of</strong> heterogeneity, shown for a source width <strong>of</strong><br />
4 m. The probability <strong>of</strong> method success is plotted against the error factor.
280<br />
S. Bauer et al.<br />
to 5, Methods 1, 2 <strong>and</strong> 3 show a success probability <strong>of</strong> about 1, while Method 4 yields<br />
a probability <strong>of</strong> 0.7. Figure 3 also illustrates that all methods exhibit a decreasing<br />
probability <strong>of</strong> success when the degree <strong>of</strong> heterogeneity is increased. An error factor<br />
<strong>of</strong> 5 yields a success probability <strong>of</strong> about 1 for the lowest degree <strong>of</strong> heterogeneity<br />
(0.38) for Method 1; this probability decreases to 0.7, 0.5 <strong>and</strong> 0.35 for a σ²ln(KF) <strong>of</strong><br />
1.71, 2.7 <strong>and</strong> 4.5, respectively. For Method 2 the corresponding success probabilities<br />
are 1.0, 0.9, 0.8 <strong>and</strong> 0.6; Method 3 yields values <strong>of</strong> 1.0, 0.55, 0.35 <strong>and</strong> 0.25; <strong>and</strong><br />
Method 4 0.7, 0.7, 0.6 <strong>and</strong> 0.4.<br />
To achieve the degradation rate within a factor <strong>of</strong> 10 <strong>of</strong> the correct degradation<br />
rate (for high hydraulic heterogeneity, Fig. 3(c)), the success probabilities were found<br />
to be 0.8, 0.95, 0.8 <strong>and</strong> 0.6 for Methods 1 through 4, respectively. For all degrees <strong>of</strong><br />
heterogeneity, Method 2 yielded the highest probability for achieving the correct<br />
degradation rate. For medium to very high heterogeneity, Method 4 was second best,<br />
<strong>and</strong> was the worst for aquifers that were only slightly heterogeneous (Fig. 3(a)). Method<br />
1, although the simplest method, yielded similar probabilities as Method 4, except<br />
Method 1 works well for aquifers <strong>of</strong> low heterogeneity. Method 3 yielded the lowest<br />
success probabilities, with the exception <strong>of</strong> an aquifer <strong>of</strong> low heterogeneity.<br />
CONCLUSIONS<br />
From the study presented, it can be concluded that the four methods for examining<br />
decay coefficients behave differently when heterogeneity is increased. All methods<br />
show a decrease in success probability with increasing heterogeneity. Figure 2<br />
illustrated that the decrease in success probability is attributed to an overestimation <strong>of</strong><br />
the degradation rate constant. The overestimation was largest for Method 3, yielding<br />
the lowest success probability. Method 2 was the least affected by an increase in<br />
heterogeneity <strong>and</strong> was also the method that depicted the lowest overestimation <strong>of</strong><br />
degradation rates. Methods 3 <strong>and</strong> 4, the most realistic since they are based on the 1-D<br />
<strong>and</strong> 2-D transport equations, illustrated low success probabilities <strong>and</strong> high overestimation<br />
<strong>of</strong> degradation rates, Method 3 being the worst. Methods 3 <strong>and</strong> 4 were<br />
prone to errors due to the introduction <strong>of</strong> longitudinal <strong>and</strong> transverse dispersivities,<br />
which were required to calculate the degradation rate constant. Method 1, the simplest<br />
method, yielded results comparable to Method 4.<br />
It can be concluded from this study, that Method 2 (using a non-reactive cocontaminant)<br />
is the preferred method for field investigations, where first-order<br />
degradation rates are to be estimated. If this co-contaminant is not available, then<br />
Method 1 is preferred as then no uncertainty regarding the dispersivities is introduced.<br />
Although widely used <strong>and</strong> published in the literature, the method <strong>of</strong> Buscheck &<br />
Alcantar (1995), Method 3, yielded the worst results in this study.<br />
Acknowledgements This work is funded by the German Ministry <strong>of</strong> Education <strong>and</strong><br />
Research as part <strong>of</strong> the KORA priority programme, sub-project 7.2.
Assessing measurements <strong>of</strong> first-order degradation rates through the virtual aquifer approach 281<br />
REFERENCES<br />
Buscheck, T. E. & Alcantar, C. M. (1995) Regression techniques <strong>and</strong> analytical solutions to demonstrate intrinsic<br />
bioremediation, In: Intrinsic Bioremediation (ed. by R. E. Hinchee, T. J. Wilson, & D. Downey), 109–116. Batelle<br />
Press, Columbus, Ohio, USA.<br />
Stenback, G. A., Ong, S. K., Rogers, S. W. & Kjartonson, B. H. (2004) Impact <strong>of</strong> transverse <strong>and</strong> longitudinal dispersion on<br />
first-order degradation rate constant estimation. J. Contam. Hydrol. 73, 3–14.<br />
Wiedemeier, T. H., Swanson, M. A., Wilson, J. T., Kampbell, D. H., Miller, R. N. & Hansen, J. E. (1996) Approximation<br />
<strong>of</strong> biodegradation rate constants for monoaromatic hydrocarbons (BTEX) in ground water. Ground Water<br />
Monitoring Remed. 16(3), 186–194.<br />
Wiedemeier, T. H., Rifai, H. S., Wilson, J. T. & Newell, C. (1999) Natural Attenuation <strong>of</strong> Fuels <strong>and</strong> Chlorinated Solvents<br />
in the Subsurface. Wiley, New York, USA.<br />
Wilson, J. T., Pfeffer, F. M., Weaver, J. W., Kampbell, D. H., Wiedemeier, T. H., Hansen, J. E., & Miller, R. N. (1994)<br />
Intrinsic bioremediation <strong>of</strong> JP-4 jet fuel. In: Symposium on Intrinsic Bioremediation <strong>of</strong> Ground Water (Denver,<br />
Colorado, USA), 60–72. US-EPA/540 R-94/515, Washington DC, USA.
Enclosed Publication 2<br />
Bauer, S., Beyer, C., Kolditz, O. (2006a): Assessing measurement uncertainty <strong>of</strong> first-order<br />
degradation rates in heterogeneous aquifers, Water Resour. Res., 42, W01420,<br />
doi:10.1029/2004WR003878. Copyright 2006 American Geophysical Union.<br />
Reproduced by permission <strong>of</strong> American Geophysical Union.<br />
The enclosed article can be obtained online from AGU at<br />
http://www.agu.org/pubs/crossref/2006.../2004WR003878.shtml.
WATER RESOURCES RESEARCH, VOL. 42, W01420, doi:10.1029/2004WR003878, 2006<br />
Assessing measurement uncertainty <strong>of</strong> first-order<br />
degradation rates in heterogeneous aquifers<br />
Sebastian Bauer, Christ<strong>of</strong> Beyer, <strong>and</strong> Olaf Kolditz<br />
Center for <strong>Applied</strong> Geoscience, University <strong>of</strong> Tübingen, Tübingen, Germany<br />
Received 7 December 2004; revised 7 October 2005; accepted 18 October 2005; published 31 January 2006.<br />
[1] The principal idea <strong>of</strong> this paper is to simulate <strong>and</strong> evaluate the determination <strong>of</strong><br />
first-order degradation rate constants at heterogeneous contaminated sites under realistic<br />
conditions. First, a set <strong>of</strong> heterogeneous <strong>and</strong> contaminated synthetic aquifers is generated;<br />
second, the spreading <strong>of</strong> a solute plume subject to first-order degradation is simulated.<br />
Third, this plume is investigated using ‘‘monitoring wells’’ placed along the presumed<br />
plume center line. Using only piezometric heads, concentrations <strong>and</strong> hydraulic<br />
conductivities obtained at these monitoring wells, first-order degradation rate constants are<br />
calculated by methods typically used in field applications. The estimated rate constants<br />
are compared to the ‘‘real’’ value known from the simulations. This comparison is<br />
conducted for different degrees <strong>of</strong> heterogeneity, represented by lognormally distributed<br />
r<strong>and</strong>om conductivity fields. The results indicate that, with increasing degree <strong>of</strong><br />
heterogeneity, ‘‘measured’’ degradation rate constants become uncertain with a high<br />
variability around the true constant. Measured rate constants tend to overestimate the true<br />
constant by up to one order <strong>of</strong> magnitude. A sensitivity analysis <strong>of</strong> the influences <strong>of</strong> source<br />
width, transport velocity, <strong>and</strong> dispersivity shows that (1) with increasing source width,<br />
measured rate constants decrease their relative error <strong>and</strong> increase their accuracy; (2) the<br />
choice <strong>of</strong> dispersivity can produce both over- <strong>and</strong> under-estimation <strong>of</strong> the true rate<br />
constant; <strong>and</strong> (3) that large-scale measurements <strong>of</strong> hydraulic conductivity yield better<br />
estimates <strong>of</strong> <strong>flow</strong> velocities as compared to local scale measurements. These results<br />
explain in part the high variability <strong>of</strong> field measured degradation rate constants reported in<br />
the literature.<br />
Citation: Bauer, S., C. Beyer, <strong>and</strong> O. Kolditz (2006), Assessing measurement uncertainty <strong>of</strong> first-order degradation rates in<br />
heterogeneous aquifers, Water Resour. Res., 42, W01420, doi:10.1029/2004WR003878.<br />
1. Introduction<br />
[2] This work studies the uncertainty involved in estimating<br />
first order degradation rate constants by the plume<br />
center line method for the assessment <strong>of</strong> natural attenuation<br />
at contaminated groundwater sites. Natural attenuation, also<br />
known as intrinsic bioremediation, refers to the observed<br />
reduction in contaminant concentration via natural processes<br />
as contaminants migrate from the source into environmental<br />
media [U.S. Environmental Protection Agency (EPA), 1999;<br />
Wiedemeier et al., 1999]. The processes contributing to<br />
natural attenuation include dilution, dispersion, sorption,<br />
volatilization <strong>and</strong> biodegradation, where biodegradation is<br />
the only process that decreases the total contaminant mass.<br />
The relative efficiencies <strong>of</strong> the attenuation processes active<br />
at a contaminated site must be carefully assessed before<br />
natural attenuation can be adopted as a cleanup remedy or<br />
risk reduction strategy. Thus degradation rates <strong>of</strong> the contaminants<br />
under consideration may play an important role in<br />
decision making <strong>and</strong> site management, when natural attenuation<br />
is considered as a remedial alternative or a remedial<br />
step in contaminated site management. Degradation rate<br />
constants can be used to estimate (1) the total overall natural<br />
Copyright 2006 by the American Geophysical Union.<br />
0043-1397/06/2004WR003878$09.00<br />
W01420<br />
attenuation potential <strong>of</strong> an aquifer, (2) contaminant plume<br />
lengths <strong>and</strong> (3) downstream concentrations. They can also be<br />
used for identifying potential receptors <strong>and</strong> exposure levels<br />
in case <strong>of</strong> a risk analysis.<br />
[3] Several approaches for estimating biodegradation<br />
rates in ground water in the field are commonly used,<br />
including mass balances, in situ microcosm studies <strong>and</strong><br />
the use <strong>of</strong> concentration-distance relations obtained along<br />
the plume center line [Chapelle et al., 1996; Wiedemeier et<br />
al., 1999]. The latter include a batch-reaction solution<br />
[Wiedemeier et al., 1996], normalization to a recalcitrant<br />
co-contaminant [Wiedemeier et al., 1996, 1999] <strong>and</strong> the<br />
method <strong>of</strong> Buscheck <strong>and</strong> Alcantar [1995]. The method <strong>of</strong><br />
Buscheck <strong>and</strong> Alcantar [1995] utilizes contaminant concentrations<br />
measured along the plume center line, which are<br />
evaluated by an analytical solution to the one-dimensional<br />
transport equation with first-order degradation. The firstorder<br />
degradation rate is calculated from the concentrations<br />
<strong>and</strong> an assumed longitudinal dispersivity. An<br />
additional requirement is, that the plume has reached<br />
steady state. This approach has been used by a number<br />
<strong>of</strong> authors, e.g., Chapelle et al. [1996], Wiedemeier et al.<br />
[1996], Zamfirescu <strong>and</strong> Grathwohl [2001], Suarez <strong>and</strong> Rifai<br />
[2002] or Bockelmann et al. [2003]. Recently, two- <strong>and</strong><br />
three-dimensional approaches were suggested [Zhang <strong>and</strong><br />
Heathcote, 2003; Stenback et al., 2004] as extensions to the<br />
1<strong>of</strong>14
W01420 BAUER ET AL.: ASSESSING FIRST-ORDER RATES W01420<br />
method by Buscheck <strong>and</strong> Alcantar [1995], which are based<br />
on analytical solutions for transport in two <strong>and</strong> three dimensions<br />
<strong>and</strong> account for finite source widths as well as<br />
transverse dispersion. By a method comparison with the<br />
original data Zhang <strong>and</strong> Heathcote [2003] showed that the<br />
method <strong>of</strong> Buscheck <strong>and</strong> Alcantar [1995] overestimates<br />
the degradation rate by 21% <strong>and</strong> 65% in case <strong>of</strong> a two<strong>and</strong><br />
three-dimensional plume, respectively. McNab <strong>and</strong><br />
Dooher [1998] reported that the method by Buscheck <strong>and</strong><br />
Alcantar [1995] is easily subject to misinterpretation, as<br />
transverse dispersivities <strong>and</strong> temporal effects can produce<br />
center line concentration pr<strong>of</strong>iles which resemble a degrading<br />
contaminant, even in the absence <strong>of</strong> degradation.<br />
[4] The spatial variability <strong>of</strong> aquifer properties has a<br />
significant influence on the distribution <strong>of</strong> contaminants<br />
<strong>and</strong> plume development. As a consequence, the methods for<br />
the estimation <strong>of</strong> degradation rates presented above are<br />
prone to effects <strong>of</strong> hydraulic heterogeneity, as they rely on<br />
concentration samples along the (presumed) plume center<br />
line as well as on estimations <strong>of</strong> site specific dispersivity. As<br />
Wilson et al. [2004] point out, the center line <strong>of</strong> a plume can<br />
easily be missed by monitoring wells installed based on<br />
assumed, but incorrect, groundwater <strong>flow</strong> directions. Moreover,<br />
contaminant plumes may w<strong>and</strong>er in all three dimensions<br />
due to macroscale heterogeneities [Wilson et al.,<br />
2004]. However, so far no study has been reported in<br />
literature which investigates these effects. Aim <strong>of</strong> this work<br />
is therefore to assess the influence <strong>of</strong> spatially heterogeneous<br />
hydraulic conductivities on the determination <strong>of</strong> firstorder<br />
degradation rates using sets <strong>of</strong> synthetic aquifer<br />
models.<br />
[5] Owing to the limited accessibility <strong>of</strong> the subsurface,<br />
measurements <strong>of</strong> piezometric heads <strong>and</strong> contaminant concentrations<br />
at contaminated sites are sparse <strong>and</strong> may not be<br />
representative <strong>of</strong> the heterogeneous hydrogeologic conditions.<br />
Therefore site investigation is subject to uncertainty,<br />
reflecting the limited knowledge on the aquifer properties<br />
<strong>and</strong> the extent <strong>of</strong> the contamination. Owing to this uncertainty,<br />
field investigation methods for plume screening or<br />
measuring hydraulic conductivity or degradation rates can<br />
neither be tested nor verified in the field. The only way <strong>of</strong><br />
assessing the performance <strong>and</strong> reliability <strong>of</strong> field investigation<br />
methods is by studying them in synthetic aquifers<br />
within a Monte Carlo framework. By applying the investigation<br />
method under consideration in the synthetic contaminated<br />
<strong>and</strong> heterogeneous aquifer, the method results can be<br />
compared to the true values. These are known from the<br />
synthetic aquifer, unlike in reality, where the true values are<br />
unknown.<br />
[6] This approach uses synthetic aquifer models, which<br />
are generated as the first step based on statistical properties<br />
<strong>of</strong> real aquifers <strong>and</strong> have a defined source <strong>of</strong> contamination.<br />
A reactive transport model is then used to simulate the<br />
spreading <strong>of</strong> the plume, resulting in realistic concentration<br />
distributions in the synthetic aquifer. In comparison to the<br />
‘‘real world,’’ the unique advantage <strong>of</strong> the synthetic aquifer<br />
is that the spatial distribution <strong>of</strong> all physical <strong>and</strong> geochemical<br />
properties <strong>and</strong> parameters as well as the contaminant<br />
concentrations are exactly known. In the second step, the<br />
synthetic aquifer is investigated by st<strong>and</strong>ard monitoring <strong>and</strong><br />
investigation techniques. In this step, only the data obtained<br />
by the investigation methods, i.e., heads <strong>and</strong> concentrations<br />
2<strong>of</strong>14<br />
at the observation wells, is used, because in case <strong>of</strong> a real<br />
site investigation the true parameter distribution is unknown.<br />
In the third step, the results from the investigation<br />
are compared to the true values, which allows to test <strong>and</strong><br />
evaluate the investigation method used. Using synthetic<br />
aquifers <strong>of</strong>fers furthermore the possibility to single out the<br />
influence <strong>of</strong> different parameters, such that sources <strong>of</strong><br />
uncertainty <strong>and</strong> error for the investigation method can be<br />
studied individually. Owing to this possibility <strong>of</strong> extensive<br />
<strong>and</strong> detailed scenario analysis <strong>and</strong> visualization, this approach<br />
is well suited to explore the uncertainty involved in<br />
hydrogeologic investigation <strong>and</strong> management. It has been<br />
applied under the term ‘‘virtual aquifer’’ by Schäfer et al.<br />
[2002, 2004], Bauer et al. [2005] <strong>and</strong> Bauer <strong>and</strong> Kolditz<br />
[2006].<br />
[7] This paper uses synthetic heterogeneous <strong>and</strong> contaminated<br />
aquifers in a Monte Carlo approach to assess for the<br />
first time the influence <strong>of</strong> spatially heterogeneous hydraulic<br />
conductivities on the determination <strong>of</strong> first-order degradation<br />
rates. To this end, plumes formed by contaminants<br />
degrading according to a first-order degradation rate in<br />
aquifers <strong>of</strong> different degrees <strong>of</strong> heterogeneity are investigated<br />
by the center line approach. By comparison <strong>of</strong> the<br />
estimated degradation rate constant with the true degradation<br />
rate constant the methods are tested <strong>and</strong> evaluated. This<br />
is performed by individually studying the influence <strong>of</strong><br />
aquifer heterogeneity, source width, <strong>flow</strong> velocity <strong>and</strong><br />
dispersivity on the estimated rate constant.<br />
2. Methods<br />
2.1. Model Domain<br />
[8] The model domain used for the <strong>numerical</strong> investigation<br />
is a two-dimensional aquifer with 184 m length <strong>and</strong><br />
64 m width (Figure 1). Flow is from left to right, with a<br />
mean hydraulic gradient <strong>of</strong> 0.003, which is induced by<br />
fixed head boundary conditions on the left <strong>and</strong> the right<br />
h<strong>and</strong> side <strong>of</strong> the model domain. No <strong>flow</strong> boundary<br />
conditions are assigned to all other sides <strong>of</strong> the model<br />
domain. The model domain is discretized with a grid<br />
density <strong>of</strong> 0.5 m in both directions. A contaminant source<br />
is emplaced 11.5 m downstream <strong>of</strong> the in<strong>flow</strong> boundary in<br />
the center <strong>of</strong> the aquifer, emitting a contaminant subject to<br />
first-order degradation with a degradation rate constant l<br />
<strong>of</strong> 1 a 1 (one per year). The contaminant source is<br />
represented by a fixed concentration boundary condition<br />
at the source position. Neither sorption, i.e., retardation,<br />
nor volatilization or dilution by recharge are accounted for.<br />
Additionally, a conservative compound is emitted from the<br />
source. The model setup is thus designed to provide ideal<br />
conditions for the application <strong>of</strong> the four center line<br />
methods to be studied. This is certainly not the case in<br />
nature, where the reaction kinetics will follow more<br />
complicated laws <strong>and</strong> may be spatially dependent, or<br />
influences from sorption <strong>and</strong> dilution have to be accounted<br />
for. However, these assumptions are used here to be able<br />
to study the st<strong>and</strong>ard methods closely <strong>and</strong> evaluate individually<br />
the influence <strong>of</strong> heterogeneity <strong>of</strong> the hydraulic<br />
conductivity. Further studies will use model setups which<br />
incorporate, e.g., different degradation kinetics.<br />
[9] A plume is generated using a process based <strong>numerical</strong><br />
<strong>flow</strong> <strong>and</strong> reactive transport model. The simulation code
W01420 BAUER ET AL.: ASSESSING FIRST-ORDER RATES<br />
Figure 1. Model area <strong>of</strong> the synthetic aquifer <strong>and</strong><br />
boundary conditions applied.<br />
GeoSys/RockFlow [Kolditz, 2002; Kolditz et al., 2004] is<br />
used here, which solves the <strong>flow</strong> <strong>and</strong> transport equations by<br />
st<strong>and</strong>ard Galerkin finite element methods [e.g., Huyakorn<br />
<strong>and</strong> Pinder, 1983] <strong>and</strong> using implicit Euler time stepping.<br />
The governing equations are given as [e.g., Bear, 1972]:<br />
<strong>and</strong><br />
S @h<br />
@t<br />
¼rðKrhÞþq ð1Þ<br />
@C<br />
@t ¼ varC þrðDrCÞ lC ð2Þ<br />
where S is the storage coefficient, h is the piezometric head,<br />
K is the tensor <strong>of</strong> hydraulic conductivity, q are sources <strong>and</strong><br />
sinks <strong>of</strong> water, C is concentration, v a is the transport<br />
velocity, D is the dispersion tensor, l is the first order<br />
degradation rate constant <strong>and</strong> t is time. The model<br />
parameters used in this study are given in Table 1. Details<br />
on <strong>numerical</strong> <strong>and</strong> s<strong>of</strong>tware issues can be found in the work<br />
<strong>of</strong> Kolditz [2002] <strong>and</strong> Kolditz <strong>and</strong> Bauer [2004]. The<br />
simulation code has been used for ground water <strong>flow</strong> <strong>and</strong><br />
transport simulations by Kolditz et al. [1998], Diersch <strong>and</strong><br />
Kolditz [1998, 2002], Thorenz et al. [2002] <strong>and</strong> Beinhorn et<br />
al. [2005].<br />
[10] To study the effects <strong>of</strong> spatially variable hydraulic<br />
conductivity, K is regarded as a r<strong>and</strong>om variable following a<br />
lognormal distribution with an expected value <strong>of</strong> E[Y =<br />
ln(K)] = 9.54. This corresponds to an effective hydraulic<br />
conductivity K ef <strong>of</strong> 7.2 10 5 ms 1 using the geometric<br />
mean [Rubin, 2003]. Using a porosity n <strong>of</strong> 0.33, the mean<br />
transport velocity is given by 6.5 10 7 ms 1 . The spatial<br />
correlation structure is characterized by an isotropic exponential<br />
covariance function C Y = s Y 2 exp( Dh/lY), with an<br />
integral scale <strong>of</strong> l Y = 2.67 m <strong>and</strong> the variance s Y 2 . Four<br />
different cases <strong>of</strong> increasing heterogeneity with ln(K) variances<br />
s Y 2 <strong>of</strong> 0.38, 1.71, 2.70 <strong>and</strong> 4.50 are considered,<br />
representing mildly to highly heterogeneous conductivity<br />
fields. The value <strong>of</strong> s Y 2 = 0.38 as well as the integral scale lY<br />
is taken from the Borden field site [Sudicky, 1986]. The<br />
value <strong>of</strong> 1.71 stems from an alluvial valley aquifer in<br />
southern Germany [Herfort, 2000]. The values <strong>of</strong> 2.70<br />
<strong>and</strong> 4.50 were reported for the Columbus Air Force Base<br />
site [Rehfeldt et al., 1992]. The geostatistical s<strong>of</strong>tware tool<br />
gstat2.4 [Pebesma <strong>and</strong> Wesseling, 1998] is used to generate<br />
100 realizations <strong>of</strong> the r<strong>and</strong>om field for each value <strong>of</strong> s Y 2 by<br />
unconditional sequential Gaussian simulation. The r<strong>and</strong>om<br />
3<strong>of</strong>14<br />
K values are generated over a two-dimensional grid <strong>of</strong><br />
density 0.5 m, exactly matching the <strong>numerical</strong> grid. Thus,<br />
following a rule <strong>of</strong> thumb <strong>of</strong> Ababou et al. [1989], a<br />
sufficient resolution <strong>of</strong> 5.33 > 1 + sY 2 grid nodes per integral<br />
scale is ensured.<br />
[11] To generate steady state plumes, as required by the<br />
methods under consideration, a stationary <strong>flow</strong> field is<br />
assumed. The time development <strong>of</strong> the plume is calculated,<br />
until the plume has reached steady state. A local longitudinal<br />
dispersivity aL = 0.25 m <strong>and</strong> a local transversal<br />
dispersivity <strong>of</strong> aT = 0.05 m are used for the <strong>numerical</strong><br />
simulations (compare Table 1).<br />
2.2. Center Line Method<br />
[12] Four methods for the determination <strong>of</strong> first-order<br />
degradation rate constants are investigated here, which are<br />
all based on the plume center line method. Method 1 is<br />
based on the one-dimensional transport equation, considering<br />
advection <strong>and</strong> first-order degradation only. The steady<br />
state solution for the concentration pr<strong>of</strong>ile can be rearranged<br />
to yield the first-order degradation rate constant for method<br />
1, i.e., l 1 [T 1 ]as:<br />
l1 ¼ va<br />
Dx<br />
Cx<br />
ln ðÞ<br />
C0<br />
where va [L T 1 ] is the transport velocity, Dx [L] is the<br />
distance between the observation wells, <strong>and</strong> C0 <strong>and</strong> C(x)<br />
[M L 3 ] are the upstream <strong>and</strong> downstream contaminant<br />
concentrations at the observation wells. In this formulation,<br />
all concentration changes resulting from processes other<br />
than degradation, i.e., diffusion, dispersion <strong>and</strong> dilution,<br />
are attributed to degradation. Therefore the rate constant<br />
l1 determined with method 1 can be considered rather<br />
an overall (or bulk) attenuation rate than a degradation<br />
rate constant [Newell et al., 2002]. Also, if the<br />
downstream observation well is not placed on the plume<br />
center line, the measured concentration is smaller than on<br />
the plume center line <strong>and</strong> the degradation rate constant is<br />
overestimated.<br />
[13] Method 2 was proposed by Wiedemeier et al. [1996]<br />
<strong>and</strong> is based on the same transport equation as method 1.<br />
However, to overcome the above mentioned drawbacks,<br />
amended concentrations are used: The measured concentrations<br />
<strong>of</strong> the reactive contaminant are corrected by the ratio<br />
<strong>of</strong> upgradient concentration C* 0 to downgradient concentration<br />
C(x)* [M L 3 ] <strong>of</strong> a nondegrading co-contaminant at the<br />
same observation wells. Thus the method corrects for<br />
dispersion <strong>of</strong> the plume or for the effects <strong>of</strong> unintended<br />
measurements <strong>of</strong>f the plume center line. The degradation<br />
Table 1. Model Parameters Used in the Simulations<br />
Parameter Value<br />
Kef 7.2 10 5 ms 1<br />
lY sy<br />
2.67 m<br />
2<br />
n<br />
0, 0.38, 1.71, 2.7, 4.5<br />
0.33<br />
l 1a 1<br />
S, q 0<br />
aL 0.25 m<br />
0.05 m<br />
a T<br />
W01420<br />
ð3Þ
W01420 BAUER ET AL.: ASSESSING FIRST-ORDER RATES W01420<br />
rate constant for method 2 is then calculated as [Wiedemeier<br />
et al., 1996]:<br />
l2 ¼ va<br />
Dx<br />
Cx<br />
ln ðÞ<br />
C0<br />
C* 0<br />
C* ðÞ x<br />
A prerequisite for the application <strong>of</strong> method 2 is that<br />
physicochemical properties <strong>of</strong> degradable <strong>and</strong> recalcitrant<br />
compounds like Henry’s Law constants <strong>and</strong> sorption<br />
coefficients must be comparable. A group <strong>of</strong> substances<br />
that has been proven to be well suited for the normalization<br />
<strong>of</strong> downgradient concentrations in BTEX plumes under<br />
anaerobic conditions are several trimethylbenzene (TMB)<br />
isomers [EPA, 1998; Wiedemeier et al., 1996, 1999]. When<br />
TMB is subject to biodegradation the estimated rate<br />
constant will be less than the actual value. For chlorinated<br />
solvent plumes, inorganic compounds like chloride may be<br />
appropriate substances for the normalization. Reductive<br />
dechlorination results in the production <strong>of</strong> chloride along<br />
the <strong>flow</strong> path, which by means <strong>of</strong> a mass balance can be<br />
used to derive the correction factor [EPA, 1998].<br />
[14] The third method investigated here was proposed by<br />
Buscheck <strong>and</strong> Alcantar [1995]. It is based on the steady<br />
state solution to the one-dimensional transport equation,<br />
accounting for advection, dispersion <strong>and</strong> first-order degradation.<br />
In comparison to method 1, method 3 accounts<br />
additionally for effects <strong>of</strong> longitudinal dispersion <strong>and</strong> thus<br />
requires an estimate <strong>of</strong> the longitudinal dispersivity aL [m].<br />
The degradation rate constant for method 3 is given by<br />
Buscheck <strong>and</strong> Alcantar [1995]:<br />
l3 ¼<br />
va<br />
ln<br />
1 2aL<br />
4aL<br />
Cx ðÞ<br />
00<br />
B<br />
C0<br />
@@<br />
Dx<br />
1<br />
A<br />
2<br />
1<br />
ð4Þ<br />
C<br />
1A<br />
ð5Þ<br />
Method 4 used here is the modified method <strong>of</strong> Buscheck <strong>and</strong><br />
Alcantar [1995], as proposed by Zhang <strong>and</strong> Heathcote<br />
[2003]. Since in this study a two-dimensional synthetic<br />
aquifer is used to assess the different approaches, for method<br />
4 the analytical solution to the two-dimensional transport<br />
equation including first order decay [Domenico, 1987] is<br />
adopted. Method 4 accounts for a finite source width as well<br />
as longitudinal <strong>and</strong> transverse dispersion. Therefore longitudinal<br />
<strong>and</strong> transverse dispersivities aL [m] <strong>and</strong> aT [m] as<br />
well as the source width WS [m] perpendicular to the average<br />
<strong>flow</strong> direction have to be known or estimated. Given these<br />
prerequisites, the degradation rate constant is given as<br />
[Zhang <strong>and</strong> Heathcote, 2003]:<br />
l4 ¼<br />
va<br />
1<br />
4aL<br />
ln<br />
2aL<br />
Cx ðÞ<br />
00<br />
B<br />
@@<br />
C0b<br />
Dx<br />
b ¼ erf<br />
WS<br />
1<br />
A<br />
2<br />
1<br />
C<br />
1A<br />
4 ffiffiffiffiffiffiffiffiffiffiffi p ð6Þ<br />
aT Dx<br />
2.3. Investigation Scenario<br />
[15] The site investigation mimicked in this study is<br />
depicted in Figure 2, where a steady state plume has<br />
evolved from a contaminant source. This plume is investigated<br />
by the center line approach. Figure 2a shows the<br />
initial situation, where three observation wells are present in<br />
4<strong>of</strong>14<br />
Figure 2. Representation <strong>of</strong> the investigation method to<br />
obtain plume center line concentrations: (a) initial<br />
situation, (b) estimation <strong>of</strong> <strong>flow</strong> direction by application<br />
<strong>of</strong> a hydrogeologic triangle, (c) measured concentrations on<br />
the inferred center line, <strong>and</strong> (d) comparison to true<br />
concentrations <strong>and</strong> heads.<br />
the aquifer. One <strong>of</strong> these wells (solid circle) is in the source<br />
<strong>and</strong> concentrations are high, while the other two show no<br />
concentration. This situation is the starting point for the<br />
investigation scenario for all realizations <strong>and</strong> represents the<br />
initial knowledge on the site. In the first investigation step,<br />
the <strong>flow</strong> direction is determined (Figure 2b). Hydraulic<br />
heads are measured in the three wells (by reading the model<br />
output at the well positions), a hydrogeologic triangle is<br />
constructed <strong>and</strong> the hydraulic gradient is calculated. In the<br />
second investigation step, three new observation wells are<br />
installed every 10 m along the estimated direction <strong>of</strong> <strong>flow</strong><br />
(Figure 2c). These wells are then used to obtain hydraulic<br />
heads <strong>and</strong> concentrations <strong>of</strong> the contaminant <strong>and</strong> the nonreactive<br />
co-contaminant as well as local hydraulic conductivities<br />
at the observation wells by using the model input for<br />
hydraulic conductivity at the corresponding location. From<br />
the head difference, the true porosity <strong>and</strong> the well positions<br />
the respective groundwater <strong>flow</strong> velocities are calculated.<br />
An effective conductivity Kef between each upstream <strong>and</strong><br />
downstream well is estimated by calculating the geometric<br />
mean <strong>of</strong> the local conductivities. Together with the concentration<br />
data, this information allows for the determination <strong>of</strong>
W01420 BAUER ET AL.: ASSESSING FIRST-ORDER RATES<br />
the first-order degradation rate constants by the four methods<br />
presented above. The investigation setup is designed to<br />
resemble ideal conditions for the application <strong>of</strong> the four<br />
methods for estimating the degradation rate constant. All<br />
measurements are assumed to be exact, which means that<br />
there is no measurement error involved. The only uncertainty<br />
<strong>and</strong> variability is introduced by the heterogeneity <strong>of</strong><br />
hydraulic conductivity. For methods 3 <strong>and</strong> 4, additionally<br />
aL <strong>and</strong> aT have to be known. These are estimated following<br />
Wiedemeier et al. [1999] as 0.1 <strong>of</strong> the plume length for aL,<br />
with aT being about 0.33 <strong>of</strong> the longitudinal dispersivity. As<br />
plume length the maximum distance covered by the observation<br />
wells, i.e., 30 m, is used. As the plumes are generally<br />
longer, this assumption yields rather low dispersivities.<br />
Thus aL <strong>and</strong> aT are estimated to be 3.0 <strong>and</strong> 1.0 m,<br />
respectively. These estimates <strong>of</strong> dispersivities are not optimal,<br />
as they are not based on the heterogeneity <strong>of</strong> the<br />
hydraulic conductivity. However, dispersivities based on<br />
results from stochastic hydrogeology are difficult to obtain,<br />
as for most field sites structure <strong>and</strong> degree <strong>of</strong> heterogeneity<br />
are not well known. Also, the four samples taken in this<br />
investigation scenario do not allow for an estimation <strong>of</strong> the<br />
correlation length or the ln(K) variances. Both the correct<br />
source width <strong>and</strong> the correct porosity are used. In the last<br />
step, by each <strong>of</strong> the four approaches the corresponding<br />
first-order degradation rate constants l 1 through l 4 are<br />
calculated. These values can be compared to the value used<br />
to generate the plume (l =1a 1 ). For each realization, the<br />
investigation procedure described above is followed <strong>and</strong> a<br />
degradation rate is calculated for each method, each downstream<br />
well <strong>and</strong> for each source width. For each <strong>of</strong> the four<br />
classes <strong>of</strong> heterogeneity used in this study (ln(K) variances<br />
s Y 2 <strong>of</strong> 0.38, 1.71, 2.70 <strong>and</strong> 4.50) a minimum <strong>of</strong> 100 realizations<br />
is evaluated. Thus statistical measures <strong>of</strong> the errors<br />
<strong>and</strong> uncertainties introduced by the heterogeneity <strong>of</strong> the<br />
hydraulic conductivity are obtained. Additionally, also the<br />
impact <strong>of</strong> the width <strong>of</strong> the source zone is studied. Here it<br />
is expected, that for increasing source width the onedimensional<br />
methods yield better results, as then the<br />
investigated situation corresponds better to the assumptions<br />
<strong>of</strong> the method. Source widths W S <strong>of</strong> 4 m, 8 m <strong>and</strong> 16 m are<br />
used, corresponding to 1.5, 3 <strong>and</strong> 6 integral scales l Y. Then<br />
methods for estimating the <strong>flow</strong> velocity are elucidated for<br />
the different degrees <strong>of</strong> heterogeneity. This is because the<br />
goodness <strong>of</strong> the calculated value for lambda is directly<br />
related to estimated transport velocity accuracy. Finally the<br />
influence <strong>of</strong> estimated longitudinal <strong>and</strong> transversal dispersivities<br />
on results by methods 3 <strong>and</strong> 4 is studied in a<br />
sensitivity analysis.<br />
2.4. Numerical Tests<br />
[16] Convergence <strong>of</strong> the Monte Carlo simulation with<br />
regard to the sample size N <strong>of</strong> estimated degradation rate<br />
constants was tested by a procedure following Goovaerts<br />
[1999]. The test is only conducted for the highest degree <strong>of</strong><br />
heterogeneity used in this study (sY 2 = 4.5) <strong>and</strong> the smallest<br />
source width <strong>of</strong> 4 m, as this is the case <strong>of</strong> highest variability.<br />
A total <strong>of</strong> 1000 realizations <strong>of</strong> the r<strong>and</strong>om conductivity field<br />
was generated. For each realization, plume development<br />
was simulated <strong>and</strong> the degradation rate constant l1 was<br />
calculated using method 1. The resulting set <strong>of</strong> 1000<br />
degradation rates is assumed to be sufficiently large to<br />
Figure 3. Influence <strong>of</strong> Monte Carlo sample size N on the<br />
average subset mean (l1), the st<strong>and</strong>ard deviation <strong>of</strong> subset<br />
means (sl1 ), <strong>and</strong> the average st<strong>and</strong>ard deviation <strong>of</strong> the<br />
subset population (sl1 ) for the highest degree <strong>of</strong> heterogeneity<br />
(sY 2 ).<br />
represent the global population. The global population <strong>of</strong><br />
l1 was r<strong>and</strong>omly sampled with a sample size <strong>of</strong> N =2,<br />
yielding a subset <strong>of</strong> two l1. For this subset the mean<br />
degradation rate as well as the st<strong>and</strong>ard deviation were<br />
calculated. R<strong>and</strong>om sampling was repeated 999 times, resulting<br />
in 1000 subsets <strong>of</strong> N = 2. From these subsets, the average<br />
subset mean l1, the st<strong>and</strong>ard deviation <strong>of</strong> subset means sl1 <strong>and</strong> the average st<strong>and</strong>ard deviation <strong>of</strong> the subset population<br />
sl1 are calculated. R<strong>and</strong>om sampling was repeated with<br />
increasing subset sizes N =3,4,..., 1000, resulting in 999<br />
triplets <strong>of</strong> the statistics, one for each subset size. Dependence<br />
<strong>of</strong> the three statistics on N is shown in Figure 3.<br />
[17] The middle curve in Figure 3 displays the average<br />
subset mean l1, which shows almost no dependence on N<br />
<strong>and</strong> yields values very close to the global mean <strong>of</strong> 8.9. The<br />
st<strong>and</strong>ard deviation <strong>of</strong> the subset means (s , lower curve)<br />
l1<br />
shows a strong decrease from 12.3 (N = 2) to 2.2 (N = 50)<br />
<strong>and</strong> 1.5 (N = 100), with a significantly reduced decrease for<br />
larger subset sizes. In relation to the global mean <strong>of</strong> 8.9, the<br />
variation among the subsets is therefore small for N 100.<br />
The upper curve in Figure 3 shows the average st<strong>and</strong>ard<br />
deviation <strong>of</strong> the r<strong>and</strong>om sample subsets sl1 , which strongly<br />
increases with subset size for small N, but with a much<br />
smaller increase for N > 50. For N = 100 a value <strong>of</strong> 15.5 is<br />
found, which is 95% <strong>of</strong> the st<strong>and</strong>ard deviation <strong>of</strong> the global<br />
population, as obtained for N = 1000. The observed reduction<br />
in increase <strong>of</strong> sl1 with N indicates the redundancy <strong>of</strong><br />
additional realizations with regard to the subset variability.<br />
As the rate <strong>of</strong> decrease <strong>of</strong> s as well as the rate <strong>of</strong> increase<br />
l1<br />
<strong>of</strong> sl1 becomes small for more than 100 realizations, we feel<br />
confident that a sample size <strong>of</strong> N = 100 is sufficient to yield<br />
stable ensemble averaged rate coefficients. Since the analysis<br />
was conducted for the largest degree <strong>of</strong> heterogeneity,<br />
5<strong>of</strong>14<br />
W01420
W01420 BAUER ET AL.: ASSESSING FIRST-ORDER RATES W01420<br />
Figure 4. Estimated first-order degradation rate constants Li (normalized to the true rate constant l)<br />
versus degree <strong>of</strong> heterogeneity sY 2 for (a) method 1, (b) method 2, (c) method 3, (d) <strong>and</strong> method 4. All<br />
figures show results for all single realizations (small symbols) as well as their ensemble means (large<br />
crosses) with their corresponding st<strong>and</strong>ard deviations (error bars) <strong>and</strong> ensemble medians (large diamonds).<br />
The reference rate constant used in the <strong>numerical</strong> simulations is indicated by the horizontal line.<br />
for lower values <strong>of</strong> s Y 2 convergence can be expected already<br />
at lower sample sizes.<br />
[18] Another check was performed concerning the mean<br />
flux over the <strong>flow</strong> domain. Although deviations between<br />
single realizations are quite distinct <strong>and</strong> increase with s Y 2 ,<br />
the ensemble averages for each s Y 2 match the theoretical<br />
value with less than a 1% error. Furthermore, the correct<br />
operation <strong>of</strong> the investigation methods was verified by<br />
applying the investigation procedure for a source <strong>of</strong> infinite<br />
width, i.e., a width equal to the model area, <strong>and</strong> by<br />
assuming the aquifer is homogeneous. Then all methods<br />
reduce to the one-dimensional method <strong>and</strong> yield the correct<br />
degradation rate constant. This result was obtained <strong>and</strong> thus<br />
the correct operation verified.<br />
3. Results <strong>and</strong> Discussion<br />
3.1. Influence <strong>of</strong> Heterogeneous Conductivity<br />
[19] To examine the influence <strong>of</strong> heterogeneity on the<br />
estimation <strong>of</strong> the rate constants, a contaminant source <strong>of</strong><br />
6<strong>of</strong>14<br />
width W S = 4 m perpendicular to the average <strong>flow</strong> direction<br />
is emplaced in the synthetic aquifer <strong>and</strong> <strong>flow</strong> as well as<br />
reactive transport are simulated. With the investigation<br />
scenario described in section 2.3 applied to a single realization,<br />
each <strong>of</strong> the four methods yields differing rate<br />
constants, l i, for each <strong>of</strong> the three center line observation<br />
wells (10 m, 20 m, <strong>and</strong> 30 m distance from the source). For<br />
the assessment <strong>of</strong> the four methods, these rate constants<br />
l i,10, l i,20 <strong>and</strong> l i,30 are averaged to yield one single<br />
estimated l i for each method. This procedure is repeated<br />
for all realizations. Figure 4 presents results <strong>of</strong> the calculated<br />
rate constants l i versus the degree <strong>of</strong> heterogeneity<br />
(given as ln conductivity variance s Y 2 ). Calculated rates li<br />
are reported as normalized rates L i, i.e., the calculated rate<br />
l i is divided by the true rate l used in the <strong>numerical</strong><br />
simulation. The normalized rate constants L i thus can be<br />
interpreted as over- or under-estimation factors. The homogeneous<br />
case (s Y 2 = 0) is included for reference.<br />
[20] Figure 4a presents the results for method 1, i.e., based<br />
on the one-dimensional advection-degradation solution
W01420 BAUER ET AL.: ASSESSING FIRST-ORDER RATES<br />
(equation (3)). It can be clearly seen, that an increase <strong>of</strong><br />
sY 2 leads to an increase in spread <strong>of</strong> the single realizations.<br />
Quite a number <strong>of</strong> realizations exhibit normalized rate<br />
constants L1 <strong>of</strong> more than 10 up to about 100, i.e., the<br />
true rate constant is severely overestimated by a factor <strong>of</strong><br />
10 to 100. Also, L1 increases with increasing sY 2 as well<br />
as the st<strong>and</strong>ard deviations <strong>of</strong> the ensemble means. This<br />
points to an increase in uncertainty <strong>and</strong> reflects the spread<br />
<strong>of</strong> the single realizations. Mean overestimation increases<br />
from a factor <strong>of</strong> about 1.6 for sY 2 = 0.38 to about 10.2 in<br />
the case <strong>of</strong> sY 2 = 4.50. In the homogeneous case (sY 2 =0),<br />
method 1 yields the correct result. Although on average l<br />
is overestimated, even for high values <strong>of</strong> sY 2 in single<br />
realizations l may actually be underestimated by L1. For<br />
the highest degree <strong>of</strong> heterogeneity, the L1 <strong>of</strong> the single<br />
realizations span about three orders <strong>of</strong> magnitude. Comparison<br />
<strong>of</strong> ensemble means with the corresponding<br />
medians shows that in all cases the medians are significantly<br />
lower. The populations <strong>of</strong> estimated l1 are positively<br />
skewed as some exceedingly large values <strong>of</strong> l1<br />
shift the means to high values. However, the general trend<br />
<strong>of</strong> increasing overestimation with heterogeneity is also<br />
distinct for the medians.<br />
[21] Figure 4b shows the corresponding results obtained<br />
with method 2, i.e., using the one dimensional advectiondegradation<br />
equation with normalization to a recalcitrant cocontaminant<br />
(equation (4)) [Wiedemeier et al., 1996]. As for<br />
method 1, the spread <strong>of</strong> the single realizations increases<br />
with increasing s Y 2 . Compared to method 1, however, the<br />
spread is smaller <strong>and</strong> more equally distributed about L 2 =1.<br />
Therefore the average overestimation factors as well as the<br />
st<strong>and</strong>ard deviations are much smaller than for method 1 <strong>and</strong><br />
both the error <strong>and</strong> the uncertainty are lower. Average rate<br />
constants L 2 are 1.1 for s Y 2 = 0.38, increasing to 3.3 for sY 2 =<br />
4.50. For homogeneous conditions, also method 2 yields the<br />
correct result, i.e., L 2 = 1 for s Y 2 = 0. Ensemble medians are<br />
just slightly above the true l, i.e., deviating less than a<br />
factor <strong>of</strong> 2.<br />
[22] Degradation rate constants calculated with method<br />
3, i.e., the one-dimensional method introduced by<br />
Buscheck <strong>and</strong> Alcantar [1995] (equation (5)), are displayed<br />
in Figure 4c. They exhibit a similar general<br />
behavior as found with method 1, i.e., increasing spread<br />
<strong>and</strong> increasing overestimation <strong>of</strong> the true rate constant for<br />
higher s Y 2 . However, mean L3 are significantly higher<br />
than the corresponding L 1, with ensemble means <strong>of</strong> 2.0<br />
for s Y 2 = 0.38 increasing to 29.4 for sY 2 = 4.50. In the<br />
homogeneous case, the true rate constant is slightly<br />
overestimated, i.e., L 3 =1.25fors Y 2 = 0. Spread in the<br />
single realizations is higher compared to method 1, now<br />
spanning nearly four orders <strong>of</strong> magnitude for the largest<br />
variance value <strong>of</strong> s Y 2 .<br />
[23] Results for method 4, i.e., the two-dimensional<br />
solution (equation (6)) suggested by Zhang <strong>and</strong> Heathcote<br />
[2003], are depicted in Figure 4d. Compared to the other<br />
methods, method 4 displays the largest spread <strong>of</strong> calculated<br />
L 4 around the mean values. For the highest degree <strong>of</strong><br />
heterogeneity, the spread <strong>of</strong> the single realizations covers<br />
nearly five orders <strong>of</strong> magnitude. Ensemble means <strong>of</strong> L 4<br />
increase from 0.6 for s Y 2 = 0.38 to about 23.0 for sY 2 = 4.5,<br />
i.e., for low s Y 2 the normalized rate constant L4 is actually<br />
underestimated in most realizations, while for larger varian-<br />
7<strong>of</strong>14<br />
W01420<br />
ces the ensemble averages approach the results <strong>of</strong> method 3.<br />
In the homogeneous case (sY 2 = 0), the estimated rate<br />
constant l4 is about two orders <strong>of</strong> magnitude lower than<br />
the true rate constant l. Ensemble medians for method 4<br />
show a similar behavior as the ensemble means, with their<br />
values closer to the true rate constant than for methods 1<br />
<strong>and</strong> 3 for large heterogeneities (sY 2 = 1.71). This reflects the<br />
fact, that the spread <strong>of</strong> the single realizations is distributed<br />
symmetrically around L4 = 1, however, the spread <strong>of</strong> the<br />
populations <strong>and</strong> thus the uncertainty is significantly larger<br />
than for methods 1 <strong>and</strong> 3.<br />
[24] Method 1 is based on the one-dimensional solution<br />
to first-order biodegradation <strong>and</strong> advection. Therefore it is<br />
expected that rate constants estimated with method 1<br />
overestimate the true rate constant due to two effects.<br />
Firstly, method 1 does not account for measuring <strong>of</strong>f the<br />
center line. So if an observation well is placed <strong>of</strong>f the plume<br />
center line, concentrations sampled there will be smaller<br />
than on the center line <strong>and</strong> therefore the degradation rate<br />
constant will be estimated too high. Secondly, as method 1<br />
is based on a one-dimensional solution <strong>of</strong> the transport<br />
equation, it does not account for transverse dispersion,<br />
which lowers concentrations on the plume center line. This<br />
second effect also causes an overestimation <strong>of</strong> the rate<br />
constant. Both effects together cause the overestimation<br />
<strong>of</strong> the rate constant as shown in Figure 4a. Method 2<br />
tries to overcome these two problems by normalization<br />
to a conservative tracer. Both above effects also determine<br />
the concentration <strong>of</strong> the nonreactive component, <strong>and</strong> are<br />
thus corrected for by the normalization. As is shown in<br />
Figure 4b, results <strong>of</strong> method 2 are considerably better than<br />
<strong>of</strong> method 1, both considering spread <strong>and</strong> ensemble averages.<br />
However, as can be seen from Figure 4b, method 2<br />
does not correct for all effects, as overestimation is observed<br />
with increasing s Y 2 . Effects <strong>of</strong> dispersion <strong>and</strong> measuring <strong>of</strong>f<br />
the center line are accounted for by method 2, so the<br />
deviation seen for method 2 has to have a hydraulic cause.<br />
This deviation is introduced by the determination <strong>of</strong> the<br />
average <strong>flow</strong> velocity between the observation wells, which<br />
is calculated using an averaged value <strong>of</strong> the hydraulic<br />
conductivity at the two observation wells. This averaged<br />
value may not be representative <strong>of</strong> the <strong>flow</strong> path between the<br />
two wells <strong>and</strong> bias may be introduced into the calculation <strong>of</strong><br />
degradation rate constants. This effects is studied closely<br />
below in section 3.3.<br />
[25] Method 3 is based on the one-dimensional transport<br />
equation including advection, degradation <strong>and</strong> longitudinal<br />
dispersion. Results from method 3, as shown in Figure 4c,<br />
display a higher spread <strong>and</strong> higher ensemble averages<br />
compared to method 1. Because method 3 includes longitudinal<br />
dispersion, it should be closer to reality <strong>and</strong> advantageous<br />
over method 1. The differences in estimated rate<br />
constants between method 1 <strong>and</strong> 3 are therefore due to the<br />
longitudinal dispersivity a L in method 3. With the onedimensional<br />
transport model used, pronounced longitudinal<br />
dispersion <strong>of</strong> a degrading contaminant results in a stronger<br />
spreading <strong>of</strong> the solute downstream <strong>and</strong> thus in higher<br />
concentrations along the plume center line compared to an<br />
advection only case. Therefore a larger rate constant is<br />
calculated to accomplish a given concentration decrease<br />
between the upgradient <strong>and</strong> the downgradient observation<br />
well. l 3 grows linearly with a L <strong>and</strong> is always larger than l 1,
W01420 BAUER ET AL.: ASSESSING FIRST-ORDER RATES W01420<br />
as can be seen by exp<strong>and</strong>ing the squared brackets in<br />
equation (3) <strong>and</strong> using C(x) C0:<br />
l3 ¼ va aL<br />
va<br />
lnð Cx ðÞ=C0<br />
Dx<br />
lnð Cx ðÞ=C0<br />
Dx<br />
2<br />
¼ l1<br />
!<br />
lnðCx ðÞ=C0Þ<br />
Dx<br />
[26] If a L = 0, method 3 reduces to the advection only<br />
case, i.e., method 1. Method 3 still does not account for<br />
transverse dispersion, which is the process causing smaller<br />
concentrations on the plume center line. Method 4 is based<br />
on a two-dimensional solution to the transport equation<br />
including advection, longitudinal <strong>and</strong> transverse dispersion<br />
<strong>and</strong> first-order degradation. Results from method 4<br />
(Figure 4d) show an underestimation <strong>of</strong> the true rate<br />
constant for homogeneous or slightly heterogeneous conditions<br />
(sY 2<br />
1.71), while for high degrees <strong>of</strong> heterogeneity<br />
(sY 2<br />
2.7), the ensemble averages <strong>of</strong> the estimated rate<br />
constants approach the respective values obtained with<br />
method 3. The underestimation for low heterogeneities is<br />
a consequence <strong>of</strong> the correction for transverse dispersion,<br />
represented by the error function b in equation (4). The<br />
effect <strong>of</strong> lower concentrations along the plume center line<br />
due to transverse dispersion is strong for small source<br />
widths, large transversal dispersivities <strong>and</strong> large well spacings.<br />
b is always less than 1 <strong>and</strong> asymptotically approaches<br />
unity for arguments <strong>of</strong> the error function larger than 2, i.e.,<br />
l4 converges toward l3 for small aT or large WS. For<br />
arguments
W01420 BAUER ET AL.: ASSESSING FIRST-ORDER RATES<br />
Figure 5. Estimated first-order degradation rate constants versus degree <strong>of</strong> heterogeneity sY 2 for<br />
(a) method 1, (b) method 2, (c) method 3, <strong>and</strong> (d) method 4 <strong>and</strong> source widths WS <strong>of</strong> 4, 8, <strong>and</strong> 16 m,<br />
respectively. Solid lines show ensemble means normalized to the true degradation rate constant (left<br />
axis), dashed lines show the corresponding st<strong>and</strong>ard deviations (right axis).<br />
[29] In the following, the influence <strong>of</strong> different approaches<br />
for estimating K (<strong>and</strong> thus va) on the determination <strong>of</strong> li<br />
is analyzed. As the plume samples only part <strong>of</strong> the full<br />
domain, the hydraulic conductivity needed is not the effective<br />
conductivity for the complete domain, but an equivalent<br />
hydraulic conductivity valid for the <strong>flow</strong> path along the<br />
center line wells. This equivalent hydraulic conductivity<br />
may differ from the global effective value. During the field<br />
investigation by the center line method, only the local<br />
hydraulic conductivities measured at the observation wells<br />
placed along the inferred plume center line or only the<br />
global effective hydraulic conductivity are known. Neither<br />
the underlying statistical parameters nor the complete conductivity<br />
field are known. The first approach represents the<br />
situation, where local hydraulic conductivities have been<br />
obtained in the observation wells, e.g., by slug tests or sieve<br />
analysis, <strong>and</strong> are averaged to obtain an estimate <strong>of</strong> the<br />
equivalent hydraulic conductivity between the two observation<br />
wells used for rate determination. In addition to the<br />
geometric mean Kg, as used so far, also the arithmetic mean<br />
Ka <strong>and</strong> the harmonic mean Kh are used, as they present<br />
9<strong>of</strong>14<br />
W01420<br />
upper <strong>and</strong> lower limits for Kef [e.g., Renard <strong>and</strong> de Marsily,<br />
1997]. As hydraulic conductivity data is available only at<br />
four wells, a full characterization <strong>of</strong> the conductivity field is<br />
not possible. Often at a real site, values <strong>of</strong> hydraulic<br />
conductivity are not available at the locations <strong>of</strong> the<br />
observation wells at the plume center line, but only from<br />
a single well not on the plume center line <strong>and</strong> thus not used<br />
for rate determination. This case is simulated by using a<br />
value KS <strong>of</strong> hydraulic conductivity from a well at point<br />
[55.10 m, 55.20 m] (see Figure 1). In the last approach, the<br />
true global effective conductivity KG <strong>of</strong> the synthetic<br />
aquifer is known, i.e., from a large-scale pumping test,<br />
<strong>and</strong> KG is used as an estimator for the local equivalent<br />
hydraulic conductivity at the observation wells.<br />
[30] Figure 6 shows the ensemble averages (Figure 6a)<br />
<strong>and</strong> medians (Figure 6b) <strong>of</strong> the degradation rate constants<br />
estimated using the above five approaches. Only results for<br />
method 2 <strong>and</strong> WS = 16 m are presented here, because<br />
method 2 is only affected by the hydraulic error due to a<br />
wrong estimation <strong>of</strong> <strong>flow</strong> velocity. Inspection <strong>of</strong> Figure 6a<br />
shows that all approaches lead to an overestimation <strong>of</strong> the
W01420 BAUER ET AL.: ASSESSING FIRST-ORDER RATES W01420<br />
Figure 6. Ensemble (a) means <strong>and</strong> (b) medians <strong>of</strong><br />
calculated L 2 for the different approaches <strong>of</strong> average <strong>flow</strong><br />
velocity estimation <strong>and</strong> (c) the corresponding coefficients <strong>of</strong><br />
variation.<br />
ensemble averaged L2, which increases with sY 2 . This<br />
increase is most pronounced for Ka <strong>and</strong> KS. For sY 2 = 4.50<br />
the average L2,K a is about 4.3 times larger than L2,K g ,<br />
whereas for single realizations the difference may reach a<br />
10 <strong>of</strong> 14<br />
factor <strong>of</strong> 70 (not shown here). If only the single KS value is<br />
used, overestimation is even larger for all degrees <strong>of</strong><br />
heterogeneity. In comparison to the geometric average,<br />
using the harmonic average Kh or the true global effective<br />
conductivity KG reduces the overestimation <strong>of</strong> L2 considerably.<br />
Both approaches result in nearly identical ensemble<br />
averages. However, inspecting ensemble medians as shown<br />
in Figure 6b, harmonic averaging <strong>of</strong> K measurements results<br />
in an underestimation <strong>of</strong> L2 in more than 50% <strong>of</strong> all<br />
realizations. Taking this into account, usage <strong>of</strong> the global<br />
KG seems to be the best approach for the estimation <strong>of</strong> the<br />
local average <strong>flow</strong> velocity between two observation wells<br />
along the plume center line. This finding is supported by<br />
Figure 6c, which shows coefficients <strong>of</strong> variation (CV) <strong>of</strong>L2<br />
as a measure <strong>of</strong> uncertainty. Using KG yields the smallest<br />
spread. Comparison <strong>of</strong> all five approaches yields: CVK G <<br />
CVK g CVK h < CVK a < CVK S for all sY 2 . These findings<br />
indicate that for the conditions given in this work the best<br />
estimate <strong>of</strong> local <strong>flow</strong> velocities <strong>and</strong> <strong>of</strong> the degradation rate<br />
constant is given by the global geometric average <strong>of</strong> the<br />
hydraulic conductivity. This result is expected for large well<br />
distances, as then ergodic conditions have been reached.<br />
Our results show that also for nonergodic conditions due to<br />
small well distances <strong>of</strong> just a few integral scales the global<br />
geometric average yields better estimates <strong>of</strong> local <strong>flow</strong><br />
velocity compared to using locally measured hydraulic<br />
conductivities. The spread observed in Figure 6 is thus<br />
due to the nonergodic conditions <strong>of</strong> single realizations, i.e.,<br />
the fact that the plume has sampled only a part <strong>of</strong> the<br />
domain. Thus for a field case, a single large-scale pumping<br />
test would be preferable to small-scale local information<br />
obtained directly at the sampling wells.<br />
3.4. Influence <strong>of</strong> Dispersivity Parameterization<br />
[31] As demonstrated in the previous sections, the parameterization<br />
<strong>of</strong> macrodispersivities a L <strong>and</strong> a T solely<br />
based on the scale <strong>of</strong> the contaminant plume according to<br />
Wiedemeier et al. [1999] using 0.1 <strong>of</strong> the assumed plume<br />
length for a L <strong>and</strong> a T = a L/3 introduces a significant<br />
additional error when rate constants are estimated with<br />
methods 3 or 4. For method 3, this error is always toward<br />
higher rate constants, while for method 4 errors in both<br />
directions may occur. It is known that aL <strong>and</strong> aT do not only<br />
depend on plume scale, because the variability <strong>of</strong> <strong>flow</strong><br />
velocity resulting from the heterogeneity <strong>of</strong> hydraulic conductivity<br />
is also important for the spreading during solute<br />
transport. If additional information on the spatial distribution<br />
<strong>and</strong> variability <strong>of</strong> hydraulic conductivity is available,<br />
i.e., from an geologically analogous aquifer, where these<br />
parameters have been determined, this information can be<br />
used to obtain estimates <strong>of</strong> the dispersivity values based on<br />
aquifer heterogeneity. So far in the manuscript, only estimates<br />
based on plume length have been used.<br />
[32] In a two-dimensional isotropic domain with zero<br />
local dispersivity <strong>and</strong> for ergodic conditions, the asymptotic<br />
large time longitudinal dispersivity is given by aL = sY 2 lY,<br />
whereas the asymptotic limit <strong>of</strong> aT is 0 [Dagan, 1989;<br />
Rubin, 2003]. For the two-dimensional model setup investigated<br />
in this manuscript <strong>and</strong> the values <strong>of</strong> sY 2 (0.38, 1,71,<br />
2.7, 4.5) <strong>and</strong> lY (2.67) m, the corresponding values for the<br />
asymptotic limit <strong>of</strong> aL are 1.02, 4.57, 7.21 <strong>and</strong> 12.02 m,<br />
respectively. Close to the source, i.e., in the preasymptotic
W01420 BAUER ET AL.: ASSESSING FIRST-ORDER RATES<br />
Figure 7. Ensemble medians <strong>of</strong> estimated first-order rate constants L 4 for (a) s Y 2 = 0.38, (b) 1.71,<br />
(c) 2.7, <strong>and</strong> (d) 4.5 for different values <strong>of</strong> a L versus a T. Values for method 3 are obtained using<br />
L 3(a L)=L 4(a L, a T = 0). Medians for L 1 <strong>and</strong> L 2 are depicted in each diagram by dashed <strong>and</strong> dash-dotted<br />
horizontal lines for reference.<br />
region, aL will show lower values. aT grows with travel<br />
distance until aT reaches peak values at about 2.5 lY <strong>of</strong> 0.1,<br />
0.44, 0.69 <strong>and</strong> 1.15 m, respectively. The asymptotic limit<br />
<strong>of</strong> aT is 0. For nonzero but small local dispersivities (aL,T<br />
l 1<br />
Y 1), aL is unaffected or slightly reduced, while aT<br />
increases <strong>and</strong> a non zero long time limit will establish<br />
[Zhang <strong>and</strong> Neuman, 1996]. As the observation wells used<br />
for the investigation <strong>of</strong> the plumes are located within 3.8 to<br />
11.2 lY from the contaminant source, it is expected, that the<br />
asymptotic limits <strong>of</strong> aL <strong>and</strong> aT are not yet reached. Thus<br />
for methods 3 <strong>and</strong> 4, aL values below those listed above<br />
should be used. Similarly, aT values in the range between<br />
the peak values <strong>and</strong> the asymptotic limit should be used<br />
for method 4. As this rough estimation already shows,<br />
applying the stochastically derived values in a real field<br />
case may introduce uncertainty, as the local conditions are<br />
generally not known exactly.<br />
[33] To study the influence <strong>of</strong> dispersivity parameterization<br />
on rate constants estimated with methods 3 <strong>and</strong> 4, the<br />
sensitivity <strong>of</strong> L3 <strong>and</strong> L4 on different values <strong>of</strong> aL <strong>and</strong> aT is<br />
investigated. Because in a real field application these values<br />
11 <strong>of</strong> 14<br />
W01420<br />
will be always uncertain, this investigation is performed as a<br />
sensitivity analysis, which allows to cover a wider range <strong>of</strong><br />
values for aL <strong>and</strong> aT. For aL values <strong>of</strong> 12, 9, 6, 3, 0.8 <strong>and</strong><br />
0.25 m are used, while for aT values <strong>of</strong> 2, 1.15, 1, 0.75, 0.5,<br />
0.3, 0.07 <strong>and</strong> 0 m are used. Thus the range <strong>of</strong> reasonable<br />
values <strong>of</strong> aL <strong>and</strong> aT up to the maximum values for each sY 2<br />
is well represented. Figure 7 presents for each sY 2 the<br />
medians <strong>of</strong> estimated rate constants L3, using L3(aL) =<br />
L4(aL, aT = 0), <strong>and</strong> L4(aL, aT) in dependence on aT for the<br />
different values <strong>of</strong> aL. For reference, also medians for L1<br />
<strong>and</strong> L2 (compare Figure 4) are included. Because method 2<br />
corrects for dispersion as well as for measuring <strong>of</strong>f the<br />
plume center line, but not for errors in the estimated<br />
transport velocity, <strong>and</strong> the same value <strong>of</strong> the transport<br />
velocity is used for all four methods, L2 can be seen as<br />
the optimal lower limit for methods 3 <strong>and</strong> 4. Method 1<br />
involves no correction for longitudinal <strong>and</strong> transverse dispersion<br />
or measuring <strong>of</strong>f the center line. Thus a good<br />
parameterization <strong>of</strong> aL <strong>and</strong> aT for method 4 should result<br />
in rate constant estimates L4 < L1. However, L4 < L2 would<br />
indicate over correction for transverse dispersion.
W01420 BAUER ET AL.: ASSESSING FIRST-ORDER RATES W01420<br />
[34] Results <strong>of</strong> the sensitivity study are presented in<br />
Figure 7. For all degrees <strong>of</strong> heterogeneity, decreasing<br />
median degradation rates are found with increasing aT.<br />
Differences between the different aL are most distinct for<br />
low values <strong>of</strong> aT. Considering method 3, it is clearly shown<br />
that L3 = L1 only for aL = 0, <strong>and</strong> L3 > L1 for aL >0.<br />
This demonstrates again that method 3 always yields<br />
higher estimated degradation rate constants as compared<br />
to method 1 (compare equation (7)). Considering that<br />
generally l is overestimated in our study, accounting only<br />
for aL by using method 3 aggravates this problem.<br />
[35] For method 4 it is found that for low <strong>and</strong> medium<br />
heterogeneity values for aL <strong>and</strong> aT exist, which allow for<br />
an optimal estimation <strong>of</strong> the degradation rate constant, i.e.,<br />
L4 = 1, <strong>and</strong> thus L4 L2. For sY 2 = 2.70, L4 > 1.0<br />
always, but L4 < L2 can be achieved for large values <strong>of</strong><br />
aT <strong>and</strong> small values <strong>of</strong> aL. However, for the highest<br />
degree <strong>of</strong> heterogeneity, L4 > L2 > 1 always. For small<br />
values <strong>of</strong> aT, the degradation rate is overestimated for all<br />
degrees <strong>of</strong> heterogeneity, while for low <strong>and</strong> medium<br />
heterogeneity L4 < 1 is possible for large values <strong>of</strong> aT.<br />
[36] For sY 2 = 0.38 <strong>and</strong> aT = 0.07 m, all medians are larger<br />
than L 1 (when a L > 0.8), at a T = 0.3 all medians are 0.5 m (Figure 7c). However, even<br />
for an unrealistically large a T <strong>of</strong> 2 m, L 4 is still significantly<br />
larger than L 2. A similar behavior is found for s Y 2 =4.5<br />
(Figure 7d), where only using a L = 9 m <strong>and</strong> a T >1.15m<br />
will result in L 4 being closer to the true rate constant than<br />
the corresponding L 1.<br />
[37] These results show, that a wide range <strong>of</strong> dispersivities<br />
can <strong>and</strong> must be used to obtain better estimates using<br />
method 4 than using method 1 or method 2. It is also found<br />
that the theoretical values (as given above) lead only for the<br />
case <strong>of</strong> low heterogeneity to considerably better estimates<br />
than using method 1. Therefore a large fraction <strong>of</strong> the<br />
observed overestimation must result from <strong>of</strong>f center line<br />
measurements, as it cannot be corrected for by reasonable<br />
values <strong>of</strong> a T. Especially for the cases <strong>of</strong> high heterogeneity,<br />
the effects <strong>of</strong> dispersion seem to be minor in comparison to<br />
the effect <strong>of</strong> missing the center line.<br />
[38] As shown above, method 4 could yield estimated<br />
rate coefficients that are as close or even closer to the true<br />
rate constant than rate coefficients estimated with method 2,<br />
regardless <strong>of</strong> the degree <strong>of</strong> heterogeneity. However, this<br />
requires unreasonably low values for a L <strong>and</strong> very high<br />
values for a T, as these parameters would have to correct for<br />
the <strong>of</strong>f center line measurement errors. In this case, a L<br />
<strong>and</strong> a T would no longer represent the actual dispersivities,<br />
but are lumped fitting parameters. As the magnitude <strong>of</strong><br />
bias introduced by missing the center line as well as the<br />
exact values <strong>of</strong> s Y 2 or lY are usually not known at a real<br />
field site, choosing dispersivities is highly uncertain <strong>and</strong><br />
may cause over- as well as under-estimation <strong>of</strong> the<br />
degradation rate constant. Thus estimation <strong>of</strong> dispersivity<br />
12 <strong>of</strong> 14<br />
introduces an additional error into the estimation <strong>of</strong> degradation<br />
rate constants using methods 3 or 4. Only for aquifers<br />
<strong>of</strong> low heterogeneity method 4 yields better estimates than<br />
method 1. Better estimates than using method 2 are only<br />
possible by assuming unphysical dispersivity values.<br />
4. Summary <strong>and</strong> Conclusions<br />
[39] In this paper the performance <strong>of</strong> four different<br />
methods for the estimation <strong>of</strong> degradation rate constants<br />
in an aquifer with a heterogeneous distribution <strong>of</strong> the<br />
hydraulic conductivity is studied. All four methods are<br />
based on the center line approach. The results demonstrate<br />
that a heterogeneous distribution <strong>of</strong> the hydraulic conductivity<br />
may lead to severe overestimation <strong>of</strong> the ensemble<br />
averaged degradation rate constant. Furthermore, the single<br />
realizations show a large spread <strong>and</strong> a large st<strong>and</strong>ard<br />
deviation, indicating that results obtained from any one<br />
estimation are highly uncertain. Mean overestimation as<br />
well as spread increase with degree <strong>of</strong> aquifer heterogeneity.<br />
By method comparison, the main reasons are identified as<br />
‘‘measuring <strong>of</strong>f the plume center line’’ <strong>and</strong> effects <strong>of</strong><br />
transverse dispersion. Best method performance is observed<br />
for method 2, which is based on the one-dimensional<br />
transport equation including advection <strong>and</strong> first order degradation.<br />
By normalizing the measured concentrations <strong>of</strong> the<br />
degrading contaminant to a nonreactive co-contaminant<br />
emitted by the same source, the above mentioned effects<br />
are corrected for. Method 2 thus shows the lowest spread<br />
<strong>and</strong> the lowest overestimation <strong>of</strong> estimated degradation rate<br />
constants <strong>of</strong> all four methods. However, the presence <strong>of</strong> the<br />
recalcitrant co-contaminant needed for the normalization<br />
approach may not always be given at a site. Second best<br />
performance is observed for method 1, which yields consistently<br />
higher spread <strong>and</strong> degradation rate constant overestimation<br />
as compared to method 2. Methods 3 <strong>and</strong> 4,<br />
although more realistic in the sense that they base on the<br />
one-dimensional <strong>and</strong> two-dimensional transport equation,<br />
respectively, show higher spread <strong>and</strong> larger overestimation.<br />
Both methods are prone to errors introduced by estimating<br />
longitudinal <strong>and</strong> transverse dispersivities. The choice <strong>of</strong><br />
these values introduces additional uncertainty without yielding<br />
substantially better results than methods 1 or 2. The<br />
ensemble averaged degradation rate constant is highest for<br />
method 3, due to the longitudinal dispersivity term in<br />
equation (3), while performance <strong>of</strong> method 4 crucially<br />
depends on the choice <strong>of</strong> an appropriate transverse dispersivity<br />
value a T.Ifa T is chosen too small with respect to the<br />
real macrodispersivity at the field site under consideration,<br />
the degradation rate constant may be overestimated. If a T is<br />
chosen too large, the degradation rate may be underestimated.<br />
For all methods, a high spread <strong>of</strong> the results from the<br />
single realizations is found, causing a high uncertainty <strong>of</strong><br />
the estimated degradation rate constant for all methods. For<br />
a single realization, the estimated degradation rate may<br />
deviate by one or even two orders <strong>of</strong> magnitude from the<br />
correct value. This deviation is caused only by the heterogeneity<br />
<strong>of</strong> the hydraulic conductivity. The spread observed<br />
here may contribute to the spread observed in degradation<br />
rate constants observed in the field. Wiedemeier et al. [1999]<br />
<strong>and</strong> Aronson <strong>and</strong> Howard [1997] report measured degradation<br />
rate constants for PCE ranging over two orders <strong>of</strong>
W01420 BAUER ET AL.: ASSESSING FIRST-ORDER RATES<br />
magnitude from 0.07 a 1 to 1.2 a 1 <strong>and</strong> for TCE ranging<br />
from 0.05 a 1 to 4.75 a 1 . For benzene, toluene <strong>and</strong> xylene<br />
rate constants <strong>of</strong> 0.07–3.0 a 1 , 0.36–21.0 a 1 <strong>and</strong> 0.32–<br />
76.0 a 1 , respectively, are reported [Wiedemeier et al.,<br />
1999]. Method performance increases with increasing<br />
source width for all methods. For sources very wide with<br />
respect to the integral scale <strong>of</strong> the hydraulic conductivity<br />
field, all methods yield reasonable results. In reality, however,<br />
when sources are heterogeneous or formed by a<br />
complex combination <strong>of</strong> a number <strong>of</strong> zones, the total source<br />
width may be difficult to estimate. If degradation rates are<br />
used for assessing the NA potential at a contaminated site,<br />
overestimation <strong>of</strong> the degradation rates is a critical point.<br />
Overestimation <strong>of</strong> the degradation rate constant leads to an<br />
overestimation <strong>of</strong> the overall natural attenuation potential. If<br />
plume lengths are calculated with too high degradation rate<br />
constants, then estimated plume lengths are too short.<br />
Remediation times as well as downgradient concentrations<br />
may be underestimated. The results presented show that<br />
determination <strong>of</strong> degradation rate constants suffers from two<br />
main sources <strong>of</strong> error, i.e., sampling <strong>of</strong>f the plume center<br />
line <strong>and</strong> an incorrect estimate <strong>of</strong> the average transport<br />
velocity. The first can be overcome by using method 2,<br />
the second can be resolved by conducting tracer tests or<br />
additional measurements <strong>of</strong> the hydraulic conductivity. A<br />
tracer test would furthermore prove, that the observation<br />
wells under consideration are sampling the same <strong>flow</strong> path.<br />
Further work on this subject will include the effects <strong>of</strong><br />
measurement error on the estimated degradation rates, both<br />
in measuring hydraulic head <strong>and</strong> contaminant concentration.<br />
Also effects <strong>of</strong> different formulations <strong>of</strong> the kinetic reactions<br />
used to simulate the plume will be investigated.<br />
[40] Acknowledgments. This work is funded by the German Ministry<br />
<strong>of</strong> Education <strong>and</strong> Research as part <strong>of</strong> the KORA priority program,<br />
subproject 7.1 Virtual Aquifer. We would like to acknowledge the thoughtful<br />
reviews <strong>of</strong> three anonymous reviewers. Their comments have greatly<br />
improved this manuscript.<br />
References<br />
Ababou, R., D. McLaughlin, A. L. Gutjahr, <strong>and</strong> A. F. B. Tompson (1989),<br />
Numerical simulation <strong>of</strong> three-dimensional <strong>saturated</strong> <strong>flow</strong> in r<strong>and</strong>omly<br />
heterogeneous porous media, Transp. Porous Media, 4, 549–565.<br />
Aronson, D., <strong>and</strong> P. H. Howard (1997), Anaerobic biodegradation <strong>of</strong> organic<br />
chemicals in groundwater: A summary <strong>of</strong> field <strong>and</strong> laboratory<br />
studies, SRC TR-97-0223F, Sci. Cent. Rep., Syracuse Res. Corp., New<br />
York.<br />
Bauer, S., <strong>and</strong> O. Kolditz (2006), Assessing contaminant mass <strong>flow</strong> rates<br />
obtained by the integral groundwater investigation method by using the<br />
virtual aquifer approach, IAHS Publ., in press.<br />
Bauer, S., C. Beyer, <strong>and</strong> O. Kolditz (2005), Assessing measurements <strong>of</strong> first<br />
order degradation rates by using the Virtual Aquifer approach, IAHS<br />
Publ., 297, 274–281.<br />
Bear, J. (1972), Dynamics <strong>of</strong> fluids in Porous Media, Elsevier, New York.<br />
Beinhorn, M., O. Kolditz, <strong>and</strong> P. Dietrich (2005), 3-D <strong>numerical</strong> evaluation<br />
<strong>of</strong> density effects on tracer tests, J. Contam. Hydrol., 81, 89–105,<br />
doi:10.1016/j.jconhyd.2005.08.001.<br />
Bockelmann, A., D. Zamfirescu, T. Ptak, P. Grathwohl, <strong>and</strong> G. Teutsch<br />
(2003), Quantification <strong>of</strong> mass fluxes <strong>and</strong> natural attenuation rates<br />
at an industrial site with a limited monitoring network: A case study,<br />
J. Contam. Hydrol., 60, 97–121, doi:10.1016/S0169-7722(02)00060-8.<br />
Buscheck, T. E., <strong>and</strong> C. M. Alcantar (1995), Regression techniques <strong>and</strong><br />
analytical solutions to demonstrate intrinsic bioremediation, in Intrinsic<br />
Bioremediation, edited by R. E. Hinchee, T. J. Wilson, <strong>and</strong> D. Downey,<br />
pp. 109–116, Battelle Press, Columbus, Ohio.<br />
Chapelle, F. H., P. M. Bradley, D. R. Lovley, <strong>and</strong> D. A. Vroblesky (1996),<br />
Measuring rates <strong>of</strong> biodegradation in a contaminated aquifer using field<br />
<strong>and</strong> laboratory methods, Ground Water, 34(4), 691–698.<br />
13 <strong>of</strong> 14<br />
W01420<br />
Dagan, G. (1989), Flow <strong>and</strong> Transport in Porous Formations, Springer,<br />
New York.<br />
Diersch, H.-J., <strong>and</strong> O. Kolditz (1998), Coupled groundwater <strong>flow</strong> <strong>and</strong><br />
transport: 2. Thermohaline <strong>and</strong> 3-D convection processes, Adv. Water<br />
Resour., 21(5), 401–425.<br />
Diersch, H.-J., <strong>and</strong> O. Kolditz (2002), Variable-density <strong>flow</strong> <strong>and</strong> transport<br />
in porous media: Approaches <strong>and</strong> challenges, Adv. Water Resour., 25(8–<br />
12), 899–944.<br />
Domenico, P. A. (1987), An analytical model for multidimensional transport<br />
<strong>of</strong> a decaying contaminant species, J. Hydrol., 91, 49–59.<br />
Dykaar, B. B., <strong>and</strong> P. K. Kitanidis (1992), Determination <strong>of</strong> the effective<br />
hydraulic conductivity for heterogeneous porous media using a <strong>numerical</strong><br />
spectral approach: 2. Results, Water Resour. Res., 28(4), 1167–<br />
1178.<br />
Goovaerts, P. (1999), Impact <strong>of</strong> the simulation algorithm, magnitude <strong>of</strong><br />
ergodic fluctuations <strong>and</strong> number <strong>of</strong> realizations on the spaces <strong>of</strong> uncertainty<br />
<strong>of</strong> <strong>flow</strong> properties, Stochastic Environ. Res. Risk Assess., 13(3),<br />
161–182, doi:10.1007/s004770050037.<br />
Herfort, M. (2000), Reactive transport <strong>of</strong> organic compounds within a<br />
heterogeneous porous aquifer, Tübinger Geowiss. Arbeiten, Univ. <strong>of</strong><br />
Tübingen, Tübingen.<br />
Huyakorn, P. S., <strong>and</strong> G. F. Pinder (1983), Computational Methods in Subsurface<br />
Flow, Academic, San Diego, Calif.<br />
Kolditz, O. (2002), Computational Methods in Environmental Fluid Dynamics,<br />
Springer, New York.<br />
Kolditz, O., <strong>and</strong> S. Bauer (2004), A process-oriented approach to computing<br />
multi-field problems in porous media, J. Hydroinf., 6(3), 225–<br />
244.<br />
Kolditz, O., R. Ratke, H.-J. Diersch, <strong>and</strong> W. Zielke (1998), Coupled<br />
groundwater <strong>flow</strong> <strong>and</strong> transport: 1. Verification <strong>of</strong> variable density <strong>flow</strong><br />
<strong>and</strong> transport models, Adv. Water Resour., 21(1), 27–46.<br />
Kolditz, O., et al. (2004), GeoSys—Theory <strong>and</strong> users Manual, Release 4.1.<br />
GeoSystems Res., Cent. for Appl. Geosci., Univ. <strong>of</strong> Tübingen, Tübingen.<br />
(Available at http://www.uni-tuebingen.de/zag/geohydrology.)<br />
McNab, W. W., <strong>and</strong> B. P. Dooher (1998), A critique <strong>of</strong> a steady-state<br />
analytical method for estimating contaminant degradation rates, Ground<br />
Water, 36(6), 983–987.<br />
Newell, C. J., H. S. Rifai, J. T. Wilson, J. A. Connor, J. A. Aziz, <strong>and</strong> M. P.<br />
Suarez (2002), Calculation <strong>and</strong> use <strong>of</strong> first-order rate constants for monitored<br />
natural attenuation studies, U.S. EPA Ground Water Issue, EPA/<br />
540/S-02/500, U.S. Environ. Protect. Ag., Washington, D. C.<br />
Pebesma, E. J., <strong>and</strong> C. G. Wesseling (1998), Gstat: A program for geostatistical<br />
<strong>modeling</strong>, prediction <strong>and</strong> simulation, Comput. Geosci., 24(1),<br />
17–31, doi:10.1016/S0098-3004(97)00082-4.<br />
Rehfeldt, K. R., J. M. Boggs, <strong>and</strong> L. W. Gelhar (1992), Field study <strong>of</strong><br />
dispersion in a heterogeneous aquifer: 3. Geostatistical analysis <strong>of</strong> hydraulic<br />
conductivity, Water Resour. Res., 28(12), 3309–3324.<br />
Renard, P., <strong>and</strong> G. de Marsily (1997), Calculating equivalent permeability:<br />
A review, Adv. Water Resour., 20(5–6), 253–278.<br />
Rubin, Y. (2003), <strong>Applied</strong> Stochastic Hydrogeology, Oxford Univ. Press,<br />
New York.<br />
Schäfer, D., A. Dahmke, O. Kolditz, <strong>and</strong> G. Teutsch (2002), ‘‘Virtual<br />
Aquifers’’: A concept for evaluation <strong>of</strong> exploration, remediation <strong>and</strong><br />
monitoring strategies, IAHS Publ., 277, 52–59.<br />
Schäfer, D., B. Schlenz, <strong>and</strong> A. Dahmke (2004), Evaluation <strong>of</strong> exploration<br />
<strong>and</strong> monitoring methods for verification <strong>of</strong> natural attenuation using the<br />
virtual aquifer approach, Biodegradation J., 15(6), 453–465,<br />
doi:10.1023/b:biod.0000044600.81216.00.<br />
Stenback, G. A., S. K. Ong, S. W. Rogers, <strong>and</strong> B. H. Kjartonson (2004),<br />
Impact <strong>of</strong> transverse <strong>and</strong> longitudinal dispersion on first-order degradation<br />
rate constant estimation, J. Contam. Hydrol., 73, 3 – 14, doi:10.1016/<br />
j.jconhyd.2003.11.004.<br />
Suarez, M. P., <strong>and</strong> H. S. Rifai (2002), Evaluation <strong>of</strong> BTEX remediation by<br />
natural attenuation at a coastal facility, Ground Water Monit. Remed.,<br />
22(1), 62–77.<br />
Sudicky, E. A. (1986), A natural gradient experiment on solute transport<br />
in a s<strong>and</strong> aquifer: Spatial variability <strong>of</strong> hydraulic conductivity <strong>and</strong> its<br />
role in the dispersion process, Water Resour. Res., 22(13), 2069 –<br />
2082.<br />
Thorenz, C., G. Kosakowski, O. Kolditz, <strong>and</strong> B. Berkowitz (2002), An<br />
experimental <strong>and</strong> <strong>numerical</strong> investigation <strong>of</strong> saltwater movement in<br />
coupled <strong>saturated</strong>-partially <strong>saturated</strong> systems, Water Resour. Res.,<br />
38(6), 1069, doi:10.1029/2001WR000364.<br />
U.S. Environmental Protection Agency (EPA) (1998), Technical protocol<br />
for evaluating natural attenuation <strong>of</strong> chlorinated solvents in groundwater,<br />
Rep. EPA/600/R/128, Washington, D. C.
W01420 BAUER ET AL.: ASSESSING FIRST-ORDER RATES W01420<br />
U.S. Environmental Protection Agency (EPA) (1999), Use <strong>of</strong> monitoring<br />
natural attenuation at Superfund, RCRA Corrective Action, <strong>and</strong> Underground<br />
Storage Tank Sites, Office <strong>of</strong> Solid Waste <strong>and</strong> Emergency Response<br />
Directive 9200.4-17, Washington, D. C.<br />
Wiedemeier, T. H., M. A. Swanson, J. T. Wilson, D. H. Kampbell, R. N.<br />
Miller, <strong>and</strong> J. E. Hansen (1996), Approximation <strong>of</strong> biodegradation rate<br />
constants for monoaromatic hydrocarbons (BTEX) in ground water,<br />
Ground Water Monit. Remed., 16(3), 186–194.<br />
Wiedemeier, T. H., H. S. Rifai, T. J. Wilson, <strong>and</strong> C. Newell (1999), Natural<br />
Attenuation <strong>of</strong> Fuels <strong>and</strong> Chlorinated Solvents in the Subsurface, John<br />
Wiley, Hoboken, N. J.<br />
Wilson, R. D., S. F. Thornton, <strong>and</strong> D. M. Mackay (2004), Challenges<br />
in monitoring the natural attenuation <strong>of</strong> spatially variable plumes, Biodegradation<br />
J., 15(6), 459–469, doi:10.1023/b:biod.0000044591. 45542.a9.<br />
Zamfirescu, D., <strong>and</strong> P. Grathwohl (2001), Occurrence <strong>and</strong> attenuation <strong>of</strong><br />
specific organic compounds in the groundwater plume at a former<br />
gasworks site, J. Contam. Hydrol., 53, 407–427, doi:10.1016/S0169-<br />
7722(01)00176-0.<br />
14 <strong>of</strong> 14<br />
Zhang, D., <strong>and</strong> S. P. Neuman (1996), Effect <strong>of</strong> local dispersion on solute<br />
transport in r<strong>and</strong>omly heterogeneous media, Water Resour. Res., 32,<br />
2715–2723.<br />
Zhang, Y.-K. (2003), Non-ergodic solute transport in physically <strong>and</strong><br />
chemically heterogeneous porous media, Water Resour. Res., 39(7),<br />
1197, doi:10.1029/2003WR002116.<br />
Zhang, Y.-K., <strong>and</strong> R. C. Heathcote (2003), An improved method for estimation<br />
<strong>of</strong> biodegradation rate with field data, Ground Water Monit.<br />
Remed., 23(3), 112–116.<br />
S. Bauer, C. Beyer, <strong>and</strong> O. Kolditz, Center for <strong>Applied</strong> Geoscience,<br />
University <strong>of</strong> Tübingen, Sigwartstr. 10, D-72076 Tübingen, Germany.<br />
(sebastian.bauer@uni-tuebingen.de; christ<strong>of</strong>.beyer@uni-tuebingen.de;<br />
kolditz@uni-tuebingen.de)
Enclosed Publication 3<br />
Beyer, C., Bauer, S., Kolditz, O. (2006): Uncertainty Assessment <strong>of</strong> Contaminant Plume<br />
Length Estimates in Heterogeneous Aquifers. J. Contam. Hydrol., 87, 73-95, doi:<br />
10.1016/j.jconhyd.2006.04.006.<br />
Reprinted from Journal <strong>of</strong> Contaminant Hydrology, 87, Beyer, Christ<strong>of</strong>, Bauer, Sebastian <strong>and</strong><br />
Olaf Kolditz, Uncertainty Assessment <strong>of</strong> Contaminant Plume Length Estimates in<br />
Heterogeneous Aquifers, 73-95, Copyright 2006, with permission from Elsevier.<br />
The enclosed article can be obtained online via ScienceDirect at<br />
http://www.sciencedirect.com/science/journal/01697722.
Uncertainty assessment <strong>of</strong> contaminant plume length<br />
estimates in heterogeneous aquifers<br />
Abstract<br />
Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
Christ<strong>of</strong> Beyer ⁎ , Sebastian Bauer, Olaf Kolditz<br />
www.elsevier.com/locate/jconhyd<br />
Center for <strong>Applied</strong> Geoscience, University <strong>of</strong> Tübingen, Sigwartstraße 10, D 72076 Tübingen, Germany<br />
Received 4 November 2005; received in revised form 14 April 2006; accepted 25 April 2006<br />
Available online 16 June 2006<br />
The Virtual Aquifer approach is used in this study to assess the uncertainty involved in the estimation <strong>of</strong><br />
contaminant plume lengths in heterogeneous aquifers. Contaminant plumes in heterogeneous twodimensional<br />
conductivity fields <strong>and</strong> subject to first order <strong>and</strong> Michaelis–Menten (MM) degradation kinetics<br />
are investigated by the center line method. First order degradation rates <strong>and</strong> plume lengths are estimated<br />
from point information obtained along the plume center line. Results from a Monte-Carlo investigation<br />
show that the estimated rate constant is highly uncertain <strong>and</strong> biased towards overly high values. Uncertainty<br />
<strong>and</strong> bias amplify with increasing heterogeneity up to maximum values <strong>of</strong> one order <strong>of</strong> magnitude.<br />
Calculated plume lengths reflect this uncertainty <strong>and</strong> bias. On average, plume lengths are estimated to about<br />
50% <strong>of</strong> the true plume length. When plumes subject to MM degradation kinetics are investigated by using a<br />
first order rate law, an additional error is introduced <strong>and</strong> uncertainty as well as bias increase, causing plume<br />
length estimates to be less than 40% <strong>of</strong> the true length. For plumes with MM degradation kinetics,<br />
therefore, a regression approach is used which allows the determination <strong>of</strong> the MM parameters from center<br />
line data. Rate parameters are overestimated by a factor <strong>of</strong> two on average, while plume length estimates are<br />
about 80% <strong>of</strong> the true length. Plume lengths calculated using the MM parameters are thus closer to the<br />
correct length, as compared to the first order approximation. This approach is therefore recommended if<br />
field data collected along the center line <strong>of</strong> a plume give evidence <strong>of</strong> MM kinetics.<br />
© 2006 Elsevier B.V. All rights reserved.<br />
Keywords: Natural Attenuation; Heterogeneity; Plume length; Center line method; Uncertainty analysis; Monte-Carlo;<br />
Numerical modelling<br />
⁎ Corresponding author. Tel.: +49 7071 29 73176; fax: +49 7071 5059.<br />
E-mail address: christ<strong>of</strong>.beyer@uni-tuebingen.de (C. Beyer).<br />
0169-7722/$ - see front matter © 2006 Elsevier B.V. All rights reserved.<br />
doi:10.1016/j.jconhyd.2006.04.006
74 C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
1. Introduction<br />
One major requirement for the implementation <strong>of</strong> natural attenuation (NA) as a remedial <strong>and</strong> risk<br />
reduction strategy for contaminated aquifers is an assessment <strong>of</strong> the dimensions <strong>of</strong> contaminant<br />
plumes <strong>and</strong> to predict their fate. Down gradient contaminant concentrations, i.e. plume lengths, must<br />
be calculated or estimated to identify potential receptors <strong>and</strong> predict exposure levels. For this purpose<br />
analytical <strong>and</strong> <strong>numerical</strong> solute transport models (e.g. Bioscreen (Newell et al., 1996), Biochlor<br />
(Aziz et al., 2000), Bioplume III (Rifai et al., 1998)) are routinely used. The rate <strong>of</strong> contaminant<br />
removal through biodegradation is a key parameter, as concentrations <strong>and</strong> the modeled plume<br />
lengths are highly sensitive to the degradation rate (McNab, 2001; Suarez <strong>and</strong> Rifai, 2004). Although<br />
very detailed mathematical descriptions <strong>of</strong> contaminant degradation in the subsurface are available<br />
(Baveye <strong>and</strong> Valocchi, 1989; Rittmann <strong>and</strong> VanBriesen, 1996; Wiedemeier et al., 1999; Islam et al.,<br />
2001), for applications in the field, usually simplified approaches are used because the identification<br />
<strong>of</strong> a large number <strong>of</strong> parameters <strong>and</strong> processes from field data is <strong>of</strong>ten impossible. Due to its<br />
mathematical simplicity, its ease <strong>of</strong> implementation into transport models <strong>and</strong> the necessity <strong>of</strong><br />
determining only a single parameter, the most frequently used degradation model is first order<br />
kinetics (Wiedemeier et al., 1999). Field methods for the determination <strong>of</strong> biodegradation rates in<br />
ground water include mass balance calculations, in-situ microcosm studies <strong>and</strong> the center line<br />
method (Chapelle et al., 1996; Wiedemeier et al., 1999). The latter is frequently used for plume<br />
monitoring <strong>and</strong> degradation rate evaluation (e.g. Chapelle et al., 1996; Wiedemeier et al., 1996;<br />
Zamfirescu <strong>and</strong> Grathwohl, 2001; Suarez <strong>and</strong> Rifai, 2002; Wilson <strong>and</strong> Kolhatkar, 2002; Bockelmann<br />
et al., 2003), <strong>and</strong> is based on contaminant concentrations measured in observation wells<br />
installed along the presumed center line <strong>of</strong> a plume. This approach, however, is only applicable for<br />
plumes that have reached a (quasi-) steady state, i.e. the plume is neither shrinking nor exp<strong>and</strong>ing <strong>and</strong><br />
the measured concentrations do not change with time. The concentration-distance relations thus<br />
obtained for a steady state plume can be used to estimate the first order rate constant λ. This<br />
parameter can then be used with an appropriate transport model to estimate the contaminant<br />
distribution in the aquifer. However, as the spatial variability <strong>of</strong> aquifer properties has a substantial<br />
influence on the distribution <strong>of</strong> contaminants <strong>and</strong> plume development, also the results <strong>of</strong> such an<br />
assessment are affected. Wilson et al. (2004) point out that the approach is prone to errors because the<br />
center line <strong>of</strong> a plume may be missed by monitoring wells if the inferred ground water <strong>flow</strong> direction<br />
is incorrect or the contaminant plume me<strong>and</strong>ers in all three dimensions due to macro-scale<br />
heterogeneities. McNab <strong>and</strong> Dooher (1998) demonstrated that, even in a homogeneous aquifer,<br />
transverse dispersion can produce center line concentration pr<strong>of</strong>iles <strong>of</strong> recalcitrant compounds that<br />
exhibit characteristics consistent with first order degradation; this can easily lead to misinterpretation<br />
<strong>of</strong> the monitoring data. The result <strong>of</strong> these complicating factors is that the degradation potential may<br />
be severely overestimated, causing underestimation <strong>of</strong> plume length or contaminant mass <strong>and</strong> an<br />
over optimistic prognosis <strong>of</strong> down gradient concentrations <strong>and</strong> exposure levels. Moreover, it is well<br />
known that the use <strong>of</strong> first order kinetics may be problematic in some situations, as it is a poor<br />
representation <strong>of</strong> the processes occurring in contaminated aquifers. Usage <strong>of</strong> a first order model<br />
outside its range <strong>of</strong> validity may result either in significant under- or overestimation <strong>of</strong> the<br />
attenuation potential at a site (Bekins et al., 1998). In a <strong>numerical</strong> experiment, Schäfer et al. (2004a)<br />
demonstrated that for specific points in time, first order kinetics may be able to approximately<br />
reproduce mass <strong>and</strong> dimensions <strong>of</strong> contaminant plumes that follow from a far more complex<br />
degradation model. For a long term prognosis, however, the first order approximation proved<br />
inappropriate, resulting in an underestimation <strong>of</strong> plume length <strong>and</strong> contaminant mass. Recently,<br />
Bauer et al. (2006) performed a sensitivity study on the influences <strong>of</strong> aquifer heterogeneity on first
order degradation rate constants estimated from using the center line method. This study demonstrated<br />
that aquifer heterogeneity introduces significant uncertainty in the estimated rate constants<br />
<strong>and</strong> may cause a severe overestimation <strong>of</strong> the degradation potential.<br />
Since the determination <strong>of</strong> degradation rates is usually only an intermediate step for the<br />
characterization <strong>of</strong> contaminated sites, the present paper takes the approach <strong>of</strong> Bauer et al. (2006) one<br />
step further. Here, the uncertainty involved in the estimation <strong>of</strong> contaminant plume lengths in<br />
heterogeneous aquifers is evaluated using the Virtual Aquifer concept. Three different scenarios are<br />
studied in detail. In case A, synthetic contaminant plumes following first order degradation kinetics<br />
are investigated. First order rate constants are estimated by methods typically used in field<br />
applications. The rate constants are then used to calculate the corresponding contaminant plume<br />
lengths with analytical transport models. As the first order degradation model results in theoretically<br />
infinite plumes, a relative concentration contour line is defined as the plume length here. Results are<br />
analysed with regard to errors <strong>and</strong> uncertainty in the rate constants <strong>and</strong> their propagation to the plume<br />
length estimates. In case B, the additional error is studied that arises when a first order approximation<br />
is used although degradation kinetics deviate from a first order rate law. Here the attenuation<br />
potential for plumes following Michaelis–Menten (MM) degradation kinetics is assessed using the<br />
same first order methods employed in case A. In case C, a regression approach is used to estimate the<br />
MM parameters <strong>and</strong> plume lengths for the plumes with MM degradation kinetics. Results are<br />
compared to cases A <strong>and</strong> B to allow conclusions about potentials <strong>and</strong> limitations <strong>of</strong> this approach.<br />
2. Virtual Aquifer concept<br />
C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
Due to the limited accessibility <strong>of</strong> the subsurface, measurements <strong>of</strong> piezometric heads <strong>and</strong><br />
pollutant concentrations at contaminated sites are sparse <strong>and</strong> may not be representative <strong>of</strong> the<br />
heterogeneous hydrogeologic conditions. Any site investigation is thus subject to uncertainty,<br />
reflecting the limited knowledge <strong>of</strong> the aquifer properties <strong>and</strong> the extent <strong>of</strong> the contamination. Due<br />
to this uncertainty, field investigation methods for plume screening <strong>and</strong> measuring <strong>of</strong> hydraulic<br />
conductivity or degradation rates can neither be tested nor verified in the field. One appropriate<br />
method <strong>of</strong> assessing the performance <strong>and</strong> reliability <strong>of</strong> field investigation methods is by studying<br />
them in heterogeneous synthetic (virtual) aquifers. With this approach the results <strong>of</strong> a particular<br />
method can be compared to the “true” values, as these values are known from the synthetic aquifer.<br />
The “Virtual Aquifer” concept is a combination <strong>of</strong> different methodologies, tools <strong>and</strong> techniques,<br />
particularly aimed at this type <strong>of</strong> problem. Its two key components are (1) a flexible <strong>and</strong> efficient<br />
modelling system, allowing the <strong>numerical</strong> simulation <strong>of</strong> reactive multi-component transport in the<br />
subsurface, <strong>and</strong> (2) an extensive database, containing statistical information, physical <strong>and</strong> (bio-)<br />
geochemical data from a large number <strong>of</strong> well investigated sites <strong>and</strong> aquifers. Moreover, the concept<br />
comprises a collection <strong>of</strong> analytical <strong>and</strong> <strong>numerical</strong> methods, that are commonly used for the<br />
investigation <strong>of</strong> contaminated sites <strong>and</strong> aquifers or the interpretation <strong>of</strong> measured data. The synthesis<br />
<strong>of</strong> both, the database <strong>and</strong> the simulation system allows a proper definition <strong>and</strong> computer based<br />
evaluation <strong>of</strong> scenarios <strong>and</strong> case studies, focussed on investigation strategies, redevelopment <strong>and</strong><br />
monitoring at contaminated sites.<br />
As a first step for such an analysis, synthetic aquifer models are generated based on the statistical<br />
properties <strong>of</strong> real aquifers. Thus, to a certain degree, these aquifer models represent realistic analogues<br />
<strong>of</strong> existing sites. A defined source <strong>of</strong> contamination is then introduced <strong>and</strong> a reactive transport model<br />
is used to simulate the evolution <strong>of</strong> the plume, resulting in realistic concentration distributions in the<br />
synthetic aquifer. In comparison to the “real world”, the unique advantage <strong>of</strong> the synthetic aquifer is<br />
that the spatial distribution <strong>of</strong> all physical <strong>and</strong> geochemical properties <strong>and</strong> parameters as well as the<br />
75
76 C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
contaminant concentrations are exactly known. In the second step, the synthetic aquifer is investigated<br />
by st<strong>and</strong>ard monitoring <strong>and</strong> investigation techniques. Although the parameter distribution <strong>of</strong> the<br />
synthetic aquifer is known a priori, only the data “measured” at the observation wells (i.e. hydraulic<br />
heads <strong>and</strong> concentrations) are used <strong>and</strong> interpreted. This is done because in a real site investigation<br />
also only a limited amount <strong>of</strong> measured data would be available, with the amount <strong>of</strong> information<br />
depending on investigation intensity <strong>and</strong> project finances. In the third step, the investigation results are<br />
compared to the “true” values, allowing an evaluation <strong>of</strong> the accuracy <strong>of</strong> the investigation method. In<br />
addition, the use <strong>of</strong> synthetic aquifers <strong>of</strong>fers the possibility to assess the influence <strong>of</strong> different<br />
parameters, such that sources <strong>of</strong> uncertainty <strong>and</strong> error for the investigation method can be considered<br />
individually <strong>and</strong> the sensitivity <strong>of</strong> investigation results on these can be studied. Stochastic simulation<br />
techniques like the Monte-Carlo method are applied to study the propagation <strong>of</strong> parameter variability<br />
<strong>and</strong> uncertainty into the investigation results. Due to the ability to perform extensive <strong>and</strong> detailed<br />
scenario analysis <strong>and</strong> visualization, this approach is well suited to the exploration <strong>of</strong> the uncertainty<br />
involved in hydrogeologic investigation <strong>and</strong> management. The methodology has been applied under<br />
the term “Virtual Aquifer” by Schäfer et al. (2002, 2004b), Bauer <strong>and</strong> Kolditz (2005) <strong>and</strong> Bauer et al.<br />
(2005, 2006).<br />
3. Scenario definition<br />
In this study, the Virtual Aquifer concept is used in a Monte-Carlo framework to assess the<br />
influence <strong>of</strong> spatially heterogeneous hydraulic conductivities on the estimation <strong>of</strong> degradation<br />
rates <strong>and</strong> contaminant plume lengths. Multiple plume realizations <strong>of</strong> contaminants degrading<br />
according to a first order degradation kinetics or Michaelis–Menten kinetics in aquifers with<br />
different degrees <strong>of</strong> heterogeneity are investigated using the center line approach. By comparison<br />
<strong>of</strong> the estimated degradation rates <strong>and</strong> plume lengths with the respective virtual reality data the<br />
investigation methods are tested <strong>and</strong> evaluated. Three different cases are studied in detail:<br />
In case A, four different st<strong>and</strong>ard methods for the determination <strong>of</strong> the first order rate constant λ<br />
are applied to concentration vs. distance data obtained from investigation <strong>of</strong> synthetic contaminant<br />
plumes following first order degradation kinetics. Accordingly, four different rate constants are<br />
estimated for each plume realization. For each λ the length <strong>of</strong> the contaminant plume is estimated<br />
using an analytical transport model. The four methods <strong>and</strong> the corresponding analytical transport<br />
models are introduced in Sections 4.1 <strong>and</strong> 4.2. The main objectives <strong>of</strong> case A are to test the<br />
applicability <strong>and</strong> performance <strong>of</strong> the four different methods <strong>of</strong> determining the first order degradation<br />
rate constant in heterogeneous aquifers <strong>and</strong> to analyse the propagation <strong>of</strong> errors <strong>and</strong><br />
uncertainty from the rate constant to the plume length estimate.<br />
In case B, the same four methods are evaluated with regard to their ability to approximate the<br />
degradation potential <strong>and</strong> estimate the plume length, when the true degradation kinetics deviate<br />
from first order. Here plumes following MM degradation kinetics are investigated in an analogous<br />
manner to case A. The additional error that arises from the first order approximation is studied. The<br />
motivation behind this scenario is that although it is well known that contaminant degradation in<br />
natural aquifers may follow far more complicated processes <strong>and</strong> kinetic laws than a simple first<br />
order model, the latter is routinely used at many field sites. Therefore, this scenario highlights some<br />
<strong>of</strong> the problems that result from this discrepance.<br />
In case C, a regression approach is studied which allows the estimation <strong>of</strong> MM kinetic parameters<br />
from plume center line data. This method is developed in Section 4.1 <strong>and</strong> tested under the influence<br />
<strong>of</strong> aquifer heterogeneity in case C. Here the plumes following MM degradation kinetics are<br />
investigated. As for cases A <strong>and</strong> B the propagation <strong>of</strong> errors <strong>and</strong> uncertainty from the estimated MM
Table 1<br />
Overview <strong>of</strong> scenarios studied<br />
Case Contaminant plume<br />
following<br />
parameters to the plume lengths is analysed. Results are set in relation to cases A <strong>and</strong> B to discuss<br />
advantages <strong>and</strong> disadvantages, potentials <strong>and</strong> limitations <strong>of</strong> the MM parameter estimation approach.<br />
An overview <strong>of</strong> the three different cases is given in Table 1.<br />
4. Methods<br />
C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
Degradation rate<br />
determined with<br />
4.1. Estimation <strong>of</strong> degradation rate constants<br />
Plume length<br />
determined with<br />
A First order kinetics First order kinetics (Eqs. (1)–(4)) First order kinetics (Eqs. (7)–(9)) 6.1<br />
B Michaelis–Menten<br />
kinetics<br />
First order kinetics (Eqs. (1)–(4)) First order kinetics (Eqs. (7)–(9)) 6.2<br />
C Michaelis–Menten<br />
kinetics<br />
Michaelis–Menten kinetics (Eq. (5)) Michaelis–Menten kinetics (Eq. (10)) 6.3<br />
Results in<br />
section<br />
Four different st<strong>and</strong>ard methods for the determination <strong>of</strong> the first order degradation rate<br />
constant λ (Table 2, Eqs. (1)–(4)) are compared in this study. Each <strong>of</strong> the four methods requires<br />
concentration-distance relations obtained by measuring contaminant concentrations in several<br />
observation wells along the center line <strong>of</strong> a steady state plume. The degradation rate constant λ is<br />
estimated by fitting a linear function to the logarithms <strong>of</strong> concentration vs. distance from the source<br />
by linear regression. Methods 1–4 are introduced only briefly here, more details are given in Bauer<br />
et al. (2006). Furthermore, a regression approach that allows the estimation <strong>of</strong> MM kinetics<br />
parameters from center line data is developed <strong>and</strong> tested as method 5 (Table 2, equation (5)).<br />
Method 1 (Table 2, equation (1)) is based on the one dimensional transport equation, considering<br />
advection <strong>and</strong> first order degradation only. Rate constants determined with method 1 can<br />
be considered rather an overall or bulk attenuation rate than a degradation rate constant (Newell et<br />
al., 2002) as all concentration changes that result from processes other than degradation, such as<br />
diffusion, dispersion, volatilization <strong>and</strong> dilution, are attributed to the degradation process. Method<br />
2(Table 2, equation (2); Buscheck <strong>and</strong> Alcantar, 1995) is based on the steady state solution to the<br />
one dimensional transport equation considering advection, longitudinal dispersion <strong>and</strong> first order<br />
degradation. Method 2 thus requires an estimate <strong>of</strong> longitudinal dispersivity αL [m]. Method 3<br />
(Table 2, equation (3)) was proposed by Zhang <strong>and</strong> Heathcote (2003) <strong>and</strong> represents a twodimensional<br />
modification <strong>of</strong> method 2. A correction term derived from the analytical solution <strong>of</strong><br />
the two-dimensional transport equation including first order decay (Domenico, 1987) is used to<br />
account for lateral spreading <strong>and</strong> the width <strong>of</strong> the source zone. A similar approach, which also<br />
allows the inclusion <strong>of</strong> measurements <strong>of</strong>f the plume center line was presented by Stenback et al.<br />
(2004). Method 4 (Table 2, equation (4); Wilson et al., 1994; Wiedemeier et al., 1996) is based on<br />
the same transport equation as method 1. However, measured concentrations <strong>of</strong> the reactive<br />
contaminant are scaled by up <strong>and</strong> down gradient concentration C0 ⁎ <strong>and</strong> C(x)⁎ [M L − 3 ] <strong>of</strong> a nondegrading<br />
conservative solute spreading from the same source. Since dispersion <strong>and</strong> measuring <strong>of</strong>f<br />
the center line also affect the measured concentrations <strong>of</strong> non-reactive solutes, this procedure<br />
allows a correction for both effects. Method 5 (Table 2, equation (5)) is valid for contaminant<br />
plumes following Michaelis–Menten (MM) degradation kinetics (Beyer et al., 2005). When the<br />
77
78 C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
Table 2<br />
Estimation <strong>of</strong> first order degradation rate constants (methods 1–4) <strong>and</strong> Michaelis–Menten degradation kinetics parameters (method 5) from center line data<br />
Formula Description Parameter<br />
estimated<br />
Method<br />
(equation)<br />
λ<br />
1D transport equation with advection <strong>and</strong><br />
first order degradation; advective velocity,<br />
va, source concentration C0, down gradient<br />
concentration, C(x), first order degradation<br />
rate constant, λ, distance, x<br />
(1) k1 ¼ − va CðxÞ<br />
ln<br />
Dx C0<br />
!<br />
2<br />
λ<br />
1D transport equation with advection,<br />
longitudinal dispersion <strong>and</strong> first order<br />
degradation; longitudinal dispersivity, αL<br />
−1<br />
lnðCðxÞ=C0Þ<br />
Dx<br />
1−2aL<br />
(2) k2 ¼ va<br />
4aL<br />
λ<br />
!<br />
2D transport equation with advection,<br />
longitudinal <strong>and</strong> transverse dispersion,<br />
source width <strong>and</strong> first order degradation;<br />
transverse dispersivity, αT, source area<br />
width, WS 2<br />
−1<br />
lnðCðxÞ=ðC0bÞÞ Dx<br />
1−2aL<br />
k3 ¼ va<br />
4aL<br />
(3)<br />
4 ffiffiffiffiffiffiffiffiffiffi p<br />
aTDx<br />
C * 0<br />
CðxÞ *<br />
!<br />
WS<br />
with b ¼ erf<br />
λ<br />
Same as method 1; contaminant<br />
concentrations normalized with<br />
regard to conservative solute<br />
concentrations, C0⁎, C(x)⁎<br />
CðxÞ<br />
ln<br />
C0<br />
(4) k4 ¼ − va<br />
Dx<br />
k max, M C<br />
1D transport equation with advection <strong>and</strong><br />
Michaelis–Menten degradation kinetics;<br />
maximum degradation rate, kmax,<br />
half saturation concentration, MC<br />
1<br />
þ<br />
kmax<br />
lnðC0=CðxÞÞ C0−CðxÞ<br />
MC<br />
¼<br />
kmax<br />
Dx<br />
ð Þ<br />
(5)<br />
va C0−CðxÞ
Table 3<br />
Calculation <strong>of</strong> contaminant plume lengths for first order <strong>and</strong> Michaelis–Menten degradation kinetics<br />
Method Formula Equation<br />
1 L1 ¼ Dx ¼ − va<br />
k lnðCðxÞ=C0Þ (7)<br />
2<br />
lnðCðxÞ=C0Þ<br />
L2 ¼ Dx ¼ 2aL<br />
1−ð1 þ 4kaL=vaÞ 0:5<br />
( " sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi#<br />
)<br />
(8)<br />
3, 4 L3;4 ¼ Dxi; for CðxÞ<br />
5 LMM ¼ Dx ¼ va<br />
degradation <strong>of</strong> a contaminant is not limited by electron acceptor availability <strong>and</strong> the microbial<br />
density is assumed to be constant with time, Eq. (6) applies (Simkins <strong>and</strong> Alex<strong>and</strong>er, 1984):<br />
dC<br />
dt<br />
C<br />
¼ −kmax<br />
C þ MC<br />
C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
where kmax is the maximum degradation rate [M L − 3 T − 1 ] <strong>and</strong> MC is the MM half-saturation<br />
concentration [ML − 3 ]. This approximation may be applicable when aquifer sediments have been<br />
exposed to contaminants for several years (Bekins et al., 1998). The integral form <strong>of</strong> Eq. (6) can be<br />
rearranged to yield equation (5) <strong>of</strong> Table 2. According to Robinson (1985), equation (5) is the most<br />
reliable <strong>of</strong> several different formulations <strong>of</strong> the integrated MM model for estimation <strong>of</strong> kmax <strong>and</strong><br />
MC. Both parameters are estimated by a linear least squares fit <strong>of</strong> Δx/[va(C0−C(x))] vs. ln(C0/C<br />
(x))/(C0−C(x)). Thus kmax is obtained as the reciprocal <strong>of</strong> the intercept <strong>of</strong> the linear function <strong>and</strong><br />
MC as its slope multiplied by kmax. Application <strong>of</strong> method 5 assumes advective transport only (see<br />
also Parlange et al., 1984). As Robinson (1985) points out, application <strong>of</strong> a linearized integrated<br />
MM model for least squares estimation <strong>of</strong> its parameters may be problematic, because measured<br />
concentrations C(x) appear in the dependent as well as in the independent variable. A preliminary<br />
examination <strong>of</strong> several nonlinear least squares approaches for fitting the MM parameters to the<br />
concentration vs. distance data showed that the parameters determined by method 5 were on<br />
average more accurate than those obtained by other methods.<br />
4.2. Estimation <strong>of</strong> contaminant plume lengths<br />
C0<br />
kmax<br />
−exp Dxi<br />
2aL<br />
1−<br />
1 þ 4kaL<br />
va<br />
Given a first order degradation rate constant λ (by estimation with one <strong>of</strong> the four center line<br />
methods presented above), equations (1)–(3) (Table 2) can be rearranged to calculate the length <strong>of</strong><br />
the steady state contaminant plume (see Table 3). The plume length here is defined as the largest<br />
distance between the source <strong>and</strong> the concentration isoline for concentration CPL [ML − 3 ]. Equation<br />
(7) gives this distance for the purely advective case <strong>of</strong> method 1, equations (8) <strong>and</strong> (9) correspond<br />
with methods 2 <strong>and</strong> 3, respectively. As the rate constant estimated with method 4 is corrected for<br />
dispersion, this process has to be accounted for when the plume length is calculated. Therefore<br />
equation (9) is also used for calculation <strong>of</strong> plume lengths based on λ4. A steady state plume length<br />
can also be calculated using the MM parameters estimated using method 5. Rearrangement <strong>of</strong><br />
equation (5) <strong>of</strong> Table 2 yields equation (10) <strong>and</strong> gives the distance L MM at which concentrations fall<br />
erf<br />
WS<br />
4 ffiffiffiffiffiffiffiffiffiffiffi p ¼ 0 (9)<br />
aTDxi<br />
MCln C0<br />
CðxÞ þ C0−CðxÞ (10)<br />
79<br />
ð6Þ
80 C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
below CPL. To define the plume length for this study, the 1% relative concentration contour line <strong>of</strong><br />
the contaminants is used, i.e. CPL=C(x)/C0=0.01.<br />
4.3. Numerical Monte-Carlo simulations<br />
To study the influence <strong>of</strong> heterogeneous hydraulic conductivity on the investigation results<br />
two-dimensional virtual aquifers are used. The model domain has dimensions <strong>of</strong> 184 m length <strong>and</strong><br />
64 m width (Fig. 1). A mean hydraulic gradient I <strong>of</strong> 0.053 is induced by fixed head boundary<br />
conditions on the left <strong>and</strong> the right h<strong>and</strong> side <strong>of</strong> the model domain. No <strong>flow</strong> boundary conditions<br />
are assigned to all other sides. Flow conditions are at steady state.<br />
The model domain is discretized with a grid density <strong>of</strong> 0.5 m in both directions. A contaminant<br />
source <strong>of</strong> 3 m×8 m, represented by a fixed concentration boundary condition, is centered at [11.5 m;<br />
32.0 m] down stream <strong>of</strong> the in<strong>flow</strong> boundary. The source emits two reactive contaminants <strong>and</strong> a<br />
conservative compound, each with a unit concentration <strong>of</strong> 1. The first reactive contaminant is<br />
degraded by first order kinetics with a rate constant λ=5.87·10 −7 s −1 . Degradation <strong>of</strong> the second<br />
reactive contaminant follows MM kinetics. MM parameters are taken from Bekins et al. (1998)<br />
(kmax=3.9·10 −9 g L −1 s −1 <strong>and</strong> MC=1.33·10 −3 g L). Using the source concentration <strong>of</strong> 2.68·10 −2 g<br />
L −1 given in Bekins et al. (1998) these parameters were scaled to a dimensionless source<br />
concentration <strong>of</strong> 1.0, as used here, yielding relative values (in normalized units) <strong>of</strong><br />
kmax=1.45·10 −7 s −1 <strong>and</strong> MC=4.97·10 −2 . Thus the first order <strong>and</strong> MM plume lengths for both<br />
compounds are equal in a two-dimensional homogeneous aquifer for CPL=C(x)/C0=0.01. Neither<br />
growth <strong>of</strong> microorganisms nor limitation or inhibition <strong>of</strong> degradation by other substances is<br />
considered here. All compounds are not retarded <strong>and</strong> show no volatilization. The conceptual model<br />
used in this study is a rigorous simplification <strong>of</strong> the processes observed in natural aquifer systems,<br />
where degradation follows more complicated laws <strong>and</strong> is spatially dependent. The model setup is thus<br />
designed to provide ideal conditions for the application <strong>of</strong> the center line methods to be studied. This<br />
is certainly not the case in nature, where the reaction kinetics will follow more complicated laws, may<br />
be spatially dependent, be steered by the availability <strong>of</strong> electron donors <strong>and</strong> acceptors, or additional<br />
influences from transient effects <strong>and</strong> dilution have to be accounted for. However, these simplifying<br />
assumptions are used here to be able to study the st<strong>and</strong>ard methods closely <strong>and</strong> evaluate individually<br />
the influence <strong>of</strong> heterogeneity <strong>of</strong> the hydraulic conductivity <strong>and</strong> the influence <strong>of</strong> degradation kinetics<br />
on the performance <strong>of</strong> the methods under otherwise ideal conditions. Case B (Table 1) is the case were<br />
we study the combination <strong>of</strong> errors stemming from hydraulics <strong>and</strong> from reaction kinetics.<br />
The hydraulic conductivity K <strong>of</strong> the virtual aquifers is regarded as a spatial r<strong>and</strong>om variable,<br />
following a lognormal distribution with an expected value <strong>of</strong> E[Y=ln(K)]=−9.54, which corresponds<br />
to an effective conductivity Kef <strong>of</strong> 7.19·10 −5 ms −1 using the geometric mean. An isotropic<br />
exponential covariance function with an integral scale lY<strong>of</strong> 2.67 m is used for the spatial correlation<br />
Fig. 1. Virtual Aquifer model domain <strong>and</strong> boundary conditions.
structure. Four different cases with increasing heterogeneity, i.e. ln(K) variances σY 2 <strong>of</strong> 0.38, 1.71, 2.7<br />
<strong>and</strong> 4.5, respectively, are considered in this study, representing mildly to highly heterogeneous<br />
conductivity fields. The parameters lY <strong>and</strong> σY 2 =0.38 are taken from the Borden field site (Sudicky,<br />
1986); the value <strong>of</strong> 1.71 was found at the Testfeld Süd in southern Germany (Herfort, 2000); the<br />
values <strong>of</strong> 2.7 <strong>and</strong> 4.5 were reported for the Columbus Air Force Base site (Rehfeldt et al., 1992). A<br />
constant porosity n <strong>of</strong> 0.33 is used, resulting in a mean <strong>flow</strong> velocity va <strong>of</strong> 1.16·10 −5 ms −1 . For each<br />
<strong>of</strong> the four degrees <strong>of</strong> heterogeneity, ensembles <strong>of</strong> 100 realizations <strong>of</strong> the r<strong>and</strong>om field are generated<br />
by unconditional Gaussian simulation. The Monte-Carlo strategy is chosen in order to obtain<br />
statistical measures <strong>of</strong> the errors <strong>and</strong> uncertainties introduced by the heterogeneity <strong>of</strong> K. Plumes <strong>of</strong><br />
the three compounds are generated in each virtual aquifer using a process based <strong>numerical</strong> <strong>flow</strong> <strong>and</strong><br />
reactive transport model. The GeoSys/RockFlow simulation code (Kolditz et al., 2004)isusedhere,<br />
which solves the <strong>flow</strong> <strong>and</strong> transport equations by finite element methods. The governing equations<br />
are given as (e.g. Bear, 1972; Kolditz, 2002):<br />
jðKjhÞ ¼0 ð11Þ<br />
∂C<br />
∂t ¼ −vajC þ jðDjCÞ−C ð12Þ<br />
with h as the piezometric height, K the tensor <strong>of</strong> hydraulic conductivity, C concentration, D the<br />
dispersion tensor, t time <strong>and</strong> Γ a sink term representing first order or MM degradation kinetics. For<br />
local dispersivities αL <strong>and</strong> αT values <strong>of</strong> 0.25 m <strong>and</strong> 0.05 m were used. Details <strong>of</strong> <strong>numerical</strong> <strong>and</strong><br />
s<strong>of</strong>tware issues can be found in Kolditz (2002) <strong>and</strong> Kolditz <strong>and</strong> Bauer (2004). All model parameters<br />
are summarized in Table 4.<br />
5. Investigation <strong>of</strong> the synthetic plumes<br />
The steady state contaminant plumes in each virtual aquifer are investigated by the center line<br />
method (see Fig. 2). Initially three observation wells are present in the aquifer (Fig. 2 (a)), one being<br />
located directly in the source in the center <strong>of</strong> the aquifer (full circle) at [13.0 m; 32.0 m], showing<br />
high concentrations. This setup is the starting point for the investigation <strong>of</strong> all realizations. The initial<br />
knowledge on the site comprises only the hydraulic heads at the three wells. The full concentration,<br />
Table 4<br />
Model parameters used in the <strong>numerical</strong> simulations<br />
Parameter Value<br />
Kef Effective conductivity 7.19·10 − 5 − 1<br />
ms<br />
σY 2<br />
ln(K)-variance 0, 0.38, 1.71, 2.7, 4.5<br />
lY Integral scale 2.67 m<br />
n Porosity 0.33<br />
I Hydraulic gradient 0.053<br />
αL Longitudinal dispersivity 0.25 m<br />
α T Transverse dispersivity 0.05 m<br />
λ First order degradation rate constant 5.87·10 − 7 − 1<br />
s<br />
kmax<br />
Maximum degradation velocity a<br />
1.45·10 − 7 − 1<br />
s<br />
MC<br />
C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
Half saturation concentration a<br />
a In normalized units, see explanation in the text.<br />
0.0497<br />
81
82 C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
Fig. 2. Investigation <strong>of</strong> a virtual plume by the center line approach: (a) initial situation, (b) estimation <strong>of</strong> <strong>flow</strong> direction by<br />
application <strong>of</strong> a hydrogeologic triangle, (c) observation wells on the inferred center line, (d) comparison to virtual reality<br />
(concentration <strong>and</strong> heads).<br />
head <strong>and</strong> conductivity distributions <strong>of</strong> the virtual sites are assumed unknown. In the first investigation<br />
step, the <strong>flow</strong> direction is determined (Fig. 2 (b)). A hydrogeologic triangle is constructed<br />
<strong>and</strong> the hydraulic gradient is calculated using the heads measured at the three wells. After this, five<br />
new observation wells are installed along the estimated <strong>flow</strong> direction with distances <strong>of</strong> 15, 50, 75,<br />
100 <strong>and</strong> 125 m from the source (Fig. 2 (c)). At these <strong>and</strong> at the well at the source hydraulic heads <strong>and</strong><br />
concentrations <strong>of</strong> the three compounds are measured. Additionally, local hydraulic conductivities are<br />
determined (e.g. by a slug test). The geometric mean <strong>of</strong> these six K values is used as an estimator for<br />
the effective conductivity Kef along the <strong>flow</strong> path. From the head difference, the true porosity <strong>and</strong> Kef<br />
an average va is calculated. For methods 3 <strong>and</strong> 4, estimates <strong>of</strong> dispersivities αL <strong>and</strong> αT are required.<br />
Following Wiedemeier et al. (1999) αL is taken as 0.1 <strong>of</strong> the plume length <strong>and</strong> αT is assumed to be 0.1<br />
<strong>of</strong> αL. As the true plume length is unknown at this stage <strong>of</strong> the site investigation, the maximum<br />
distance covered by the observation wells, i.e. 125 m, is used. Consequently, α L <strong>and</strong> α Tare estimated<br />
to be 12.5 <strong>and</strong> 1.25 m, respectively. Of course, such rather rough estimates <strong>of</strong> α L <strong>and</strong> α T are not<br />
optimal, as they are not based on the heterogeneity structure <strong>of</strong> the aquifer. In practice, however,<br />
dispersivities based on results from stochastic hydrogeology are difficult to obtain, as for most field<br />
sites structure <strong>and</strong> degree <strong>of</strong> heterogeneity are not well characterized. Also with this scenario, the
samples taken at all eight wells do not allow for an estimation <strong>of</strong> lY<strong>and</strong> σY 2 , which would be required<br />
to derive αL <strong>and</strong> αT. A detailed study on the effects <strong>of</strong> dispersivity parameterization on the<br />
performance <strong>of</strong> methods 2 <strong>and</strong> 3 is presented in Bauer et al. (2006).<br />
The information obtained by the site investigation allows the application <strong>of</strong> the five methods for<br />
estimation <strong>of</strong> the degradation kinetic parameters <strong>and</strong> the subsequent calculation <strong>of</strong> contaminant<br />
plume lengths as presented in Sections 4.1 <strong>and</strong> 4.2. The investigation setup is designed to resemble<br />
ideal conditions for this purpose. All measurements are assumed to be exact, i.e. without measurement<br />
error, <strong>and</strong> are obtained by reading the model output at the respective well positions. The<br />
only uncertainty <strong>and</strong> variability is introduced by the heterogeneity <strong>of</strong> the hydraulic conductivity.<br />
6. Results <strong>and</strong> discussion<br />
C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
As outlined in Section 3, three different cases A, B <strong>and</strong> C are investigated here (see Table 1). The<br />
next three Sections 6.1, 6.2 <strong>and</strong> 6.3 present detailed results <strong>and</strong> discussions for each case studied. A<br />
detailed comparison <strong>and</strong> discussion <strong>of</strong> the performance <strong>of</strong> the different approaches is given in<br />
Section 6.4.<br />
6.1. Case A: estimation <strong>of</strong> first order degradation rate constants <strong>and</strong> plume lengths for plumes<br />
following a first order degradation kinetics<br />
6.1.1. Estimation <strong>of</strong> rate constants<br />
In case A methods 1–4 are tested <strong>and</strong> compared based on their ability to estimate the first order<br />
degradation rate constant λ <strong>and</strong> the contaminant plume length in heterogeneous aquifers.<br />
Therefore, plumes following first order degradation kinetics are investigated here. The four<br />
estimated rate constants λi are divided by the true rate constant to yield corresponding normalized<br />
Λi, which can be interpreted as an over- or underestimation factor. Fig. 3 presents ensemble means<br />
with corresponding st<strong>and</strong>ard deviations as error bars, medians, coefficients <strong>of</strong> variation <strong>and</strong> the<br />
single realization results for methods 1–4 against the aquifer heterogeneity σY 2 .<br />
Results for method 1 (Fig. 3 (a)) show that Λ1=1 in the homogeneous case (σY 2 =0), thus in this case<br />
the true rate is obtained. However, already for the lowest degree <strong>of</strong> heterogeneity (σY 2 =0.38), about<br />
77% <strong>of</strong> the rate constants estimated fall above the reference line which indicates the true rate constant,<br />
overestimating λ up to factors <strong>of</strong> 4.79 in the worst case. In the remaining realizations λ is<br />
underestimated. The spread <strong>of</strong> the ensemble <strong>of</strong> 100 realizations covers about one order <strong>of</strong> magnitude.<br />
When σY 2 is further increased, spread <strong>and</strong> st<strong>and</strong>ard deviation also increase. For σY 2 =4.5 the Λ1 cover<br />
almost two orders <strong>of</strong> magnitude with single realizations showing overestimation factors N10. Mean Λ1<br />
is 3.3 with st<strong>and</strong>ard deviation <strong>and</strong> coefficient <strong>of</strong> variation <strong>of</strong> 3.65 <strong>and</strong> 1.11, respectively. Increasing the<br />
heterogeneity <strong>of</strong> the aquifer thus results in a higher uncertainty <strong>of</strong> λ <strong>and</strong> increases the probability <strong>of</strong> a<br />
significant overestimation. For method 2 normalized rate constants Λ2 are shown in Fig. 3 (b). As seen<br />
for method 1 a systematic overestimation <strong>of</strong> λ can be observed, increasing with σ Y 2 . However, the Λ2<br />
are significantly larger than the corresponding Λ 1. In the homogeneous case method 2 yields Λ 2=1.7,<br />
increasing to 9.38 for σY 2 =4.50. In extreme cases, overestimation <strong>of</strong> λ is larger than a factor <strong>of</strong> 50. Also<br />
spread <strong>and</strong> st<strong>and</strong>ard deviations are larger than for Λ1. Estimated rate constants Λ3 obtained by method<br />
3(Fig. 3 (c)) are only slightly lower than those <strong>of</strong> method 2, ranging from 1.17 in the homogeneous<br />
case to 7.39 for σY 2 =4.5. Method 4 (Fig. 3(d)) yields the true rate for homogeneous conditions, i.e.<br />
Λ4=1. As for the other methods, spread <strong>and</strong> uncertainty increase with σY 2 . Compared to methods 1–3<br />
the spread is smaller <strong>and</strong> balanced around the true rate constant. For σY 2 ≤2.7 the ensemble averages<br />
<strong>and</strong> medians match the true rate constant well. For σ Y 2 =4.50 the average Λ4 reaches 1.73.<br />
83
84 C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
The overestimation observed for the different Λi results from a combination <strong>of</strong> several effects.<br />
Method 1, which is the simplest approach used in this study, is based on the one dimensional transport<br />
equation only accounting for advection <strong>and</strong> degradation. Concentration reductions on the plume<br />
center line caused by transverse dispersion therefore are incorrectly attributed to the degradation<br />
process <strong>and</strong> Λ1 is overestimated. Moreover, when the inferred center line deviates from the true center<br />
line, concentrations measured are too low, which further increases rate overestimation. A third source<br />
<strong>of</strong> error is the estimate <strong>of</strong> v a (see Section 5), which may not be representative for the <strong>flow</strong> path.<br />
Overestimation is caused if v a is estimated too high, while a too low value <strong>of</strong> v a causes underestimation<br />
<strong>of</strong> the rate constant. All three error types are also relevant for method 2. The additional bias <strong>of</strong> method<br />
2 towards too large rate constants in comparison with method 1 is a consequence <strong>of</strong> accounting for αL<br />
in equation (2) <strong>of</strong> Table 2. Longitudinal dispersion <strong>of</strong> a degrading contaminant results in a stronger<br />
spreading <strong>of</strong> the solute down stream <strong>and</strong> consequently in higher concentrations along the center line <strong>of</strong><br />
a steady state plume. Equation (2) can be rearranged to show that λ2=λ1+αLva(ln(C(x)/C0)/Δx) 2 , i.e.<br />
λ2 grows linearly with αL <strong>and</strong> is always larger than λ1 for αLN0.Method3isaffectedbyerrorsinva<br />
<strong>and</strong> <strong>of</strong>f center line measurements. By accounting for transverse dispersion, rate constant estimates are<br />
improved compared to method 2, because βb1.0 in equation (4) <strong>and</strong> thus Λ 3bΛ 2 always. Because β<br />
approaches unity for arguments N2, Λ3 converges with Λ2 for small αT, short transport distances Δx<br />
Fig. 3. Estimated first order degradation rate constants Λi (normalized to the true rate constant λ, indicated by the<br />
horizontal line) vs. heterogeneity <strong>of</strong> the aquifer σ Y 2 for methods 1 (a), 2 (b), 3 (c) <strong>and</strong> 4 (d) for case A.
C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
<strong>and</strong> large source widths WS. A surprising result is that method 1, despite its simplicity, yields closer<br />
estimates <strong>of</strong> the true rate constant than the more comprehensive description by method 3. Since<br />
method 3 depends on longitudinal <strong>and</strong> transverse dispersivities, an adequate parameterization is<br />
crucial for its success. From stochastic hydrogeology it is known that αL as well as αT strongly<br />
depends on travel time <strong>and</strong> distance as well as on the correlation structure <strong>of</strong> hydraulic conductivity<br />
<strong>and</strong> <strong>flow</strong> velocity (e.g. Dagan, 1989). Consequently, a uniform parameterization solely based on the<br />
scale <strong>of</strong> the contaminant problem as used in this study (<strong>and</strong> with many field applications) is not<br />
adequate. A detailed sensitivity study on the influence <strong>of</strong> dispersivity parameterization on the<br />
performance <strong>of</strong> methods 2 <strong>and</strong> 3 is presented in Bauer et al. (2006). It is found that for method 2 no<br />
value <strong>of</strong> αL <strong>and</strong> for method 3 only very high <strong>and</strong> thus unphysical values <strong>of</strong> αT yield the correct<br />
degradation rate constant. The required values, however, cannot be deduced from aquifer heterogeneity<br />
σY 2 alone, as the other errors also influence the estimated degradation rate constant.<br />
Method 4 circumvents this problem <strong>and</strong> corrects for transverse dispersion as well as for measuring<br />
<strong>of</strong>f the center line by normalizing concentrations to a conservative tracer. The bias towards too large<br />
degradation rate constants observed for the other methods is significantly reduced yielding the closest<br />
estimates <strong>of</strong> λ <strong>of</strong> the four methods. The remaining deviation from the true rate constant is due to the<br />
hydraulic error introduced by the approximation <strong>of</strong> va. For low heterogeneities there is no evidence for<br />
a systematic bias towards either too high or too low rate constants. A prerequisite which may limit the<br />
applicability <strong>of</strong> method 4 is the presence <strong>of</strong> a suited normalization compound. A discussion <strong>of</strong> potential<br />
normalization compounds is provided in U.S. EPA (1998) <strong>and</strong> Wiedemeier et al. (1996, 1999).<br />
6.1.2. Estimation <strong>of</strong> plume lengths<br />
During site characterization, estimating degradation rate constants rarely is a goal per se. Here, the<br />
kinetics <strong>of</strong> contaminant degradation are quantified to be used for prediction <strong>of</strong> the steady state length<br />
<strong>of</strong> the plumes. In site assessment, such information could be used to identify potential receptors <strong>and</strong><br />
exposure levels. Rate constants λ1, λ2 <strong>and</strong> λ3 are evaluated using the respective corresponding<br />
equations (7), (8) <strong>and</strong> (9) <strong>of</strong> Table 3 yielding plume length estimates L1, L2 <strong>and</strong> L3. Forλ4 also<br />
equation (9) is used yielding L4. To be able to compare the results <strong>of</strong> all realizations, the Li are<br />
normalized by the respective true length L read from the model output. Resulting over- respectively<br />
underestimation factors against aquifer heterogeneity σY 2 are presented in Fig. 4 (single realization<br />
results, ensemble means with st<strong>and</strong>ard deviations as error bars, medians, coefficients <strong>of</strong> variation).<br />
Plume lengths L 1 <strong>and</strong> L 2, calculated from λ 1 <strong>and</strong> λ 2, show exactly identical results, although<br />
the λ2 show a stronger overestimation than the corresponding λ1 for all realizations. The reason<br />
for the equivalence <strong>of</strong> L1 <strong>and</strong> L2 is that the bias introduced by estimating λ2 with a one<br />
dimensional model accounting for longitudinal dispersion only is reversed by using the same<br />
transport equation to calculate the plume length. While for homogeneous conditions the true<br />
plume length is obtained, L is underestimated in most realizations for all degrees <strong>of</strong> heterogeneity.<br />
Mean L1 <strong>and</strong> L2 decrease to 0.59 for σY 2 =4.5. As for λ, spread <strong>and</strong> uncertainty <strong>of</strong> L increase with<br />
σ Y 2 . The overestimation <strong>of</strong> the degradation rate constant is thus reflected in an underestimation <strong>of</strong><br />
the plume length.<br />
Plume lengths L3 calculated with the two-dimensional transport equation on average are lower<br />
than the corresponding L1 <strong>and</strong> L2. This is a consequence <strong>of</strong> the β term in equation (5) (Table 2),<br />
which is used to correct down gradient concentrations for transverse dispersion. When estimating<br />
the rate constant with method 3, each down gradient concentration C(x) is scaled by a different<br />
value β, as the correction factor is dependent on the distance from the source. This scaling is not<br />
fully reversed when the plume length is calculated using equation (9) (Table 3), as then only one<br />
single Δx is used.<br />
85
86 C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
As shown before, degradation rate constants λ4 in general are significantly better estimated<br />
than those obtained by the other three approaches. This is also reflected in plume lengths L4 (Fig.<br />
4 (c)). For all σY 2 the ensemble means almost perfectly match the true L. Of all four approaches the<br />
L4 consistently show the lowest coefficients <strong>of</strong> variation, indicating the lowest spread <strong>and</strong> thus the<br />
lowest uncertainty.<br />
The plume length results presented for case A demonstrate that the determination <strong>of</strong><br />
degradation rate constants using the center line method as well as the subsequent calculation <strong>of</strong><br />
contaminant plume lengths both are subject to uncertainty induced by the heterogeneity <strong>of</strong> the<br />
medium. However, the uncertainty observed for the rate constants is only partially transferred to<br />
the calculated plume lengths. For all approaches the observed spread <strong>of</strong> the ensembles <strong>of</strong> estimated<br />
plume lengths is smaller than for the corresponding rate constants. This is most obvious for the L4,<br />
which also show the best agreement with the true plume lengths. Moreover, the L4 are unbiased,<br />
showing equal amounts <strong>of</strong> under- as well as overestimation. However, both types <strong>of</strong> error are<br />
undesirable. Underestimation <strong>of</strong> the contaminant plume length, i.e. a “non-conservative” result,<br />
may pose a threat to down gradient receptors. On the other h<strong>and</strong>, a “conservative” result, i.e. an<br />
overestimation <strong>of</strong> plume dimensions, might result in wrong decisions regarding the necessity <strong>and</strong><br />
dimensioning <strong>of</strong> engineered remediation measures with unnecessary financial expenses. In<br />
Fig. 4. Plume lengths Li calculated with degradation rate constants λ1 through λ4. The Li are normalized by the true plume<br />
length L (indicated by the horizontal line) for case A.
C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
contrast to L4, plume lengths L1, L2 <strong>and</strong> L3 show a clear tendency <strong>of</strong> underestimation, demonstrating<br />
the non-conservativeness <strong>of</strong> too large rate constants. Comparing methods 1, 2 <strong>and</strong> 3,<br />
differences in calculated plume lengths are not as pronounced as for the rate constants. Thus it can<br />
be concluded, that the same underlying equation should be used for rate constant <strong>and</strong> plume length<br />
estimation. However, if e.g. a rate constant estimated with method 2 is used in a two- or threedimensional<br />
(<strong>numerical</strong>) transport model, results are susceptible to the bias in method 2, resulting<br />
in a significantly stronger underestimation <strong>of</strong> the plume length.<br />
6.2. Case B: estimation <strong>of</strong> first order degradation rate constants <strong>and</strong> plume lengths for plumes<br />
following Michaelis–Menten degradation kinetics<br />
In case B the additional error is studied, that arises when the four methods for the estimation <strong>of</strong><br />
first order degradation rate constants (Table 2, equations (1)–(4)) <strong>and</strong> the respective equations for<br />
plume lengths (Table 3, equations (7)–(9)) are used for plumes with a degradation kinetics<br />
deviating from first order but following MM degradation kinetics instead (see Table 1).<br />
Since a direct comparison <strong>of</strong> estimated rate constants λ 1 through λ 4 with the MM parameters k max<br />
<strong>and</strong> MC used in the <strong>numerical</strong> simulations is not possible, the evaluation here is based on calculated<br />
Fig. 5. Normalized plume lengths Li calculated with degradation rate constants λ1 through λ4 for the contaminant plumes<br />
following Michaelis–Menten degradation kinetics.<br />
87
88 C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
plume lengths only. The estimated contaminant plume lengths Li are displayed in Fig. 5 (single<br />
realization results, ensemble means with st<strong>and</strong>ard deviations as error bars, medians, coefficients <strong>of</strong><br />
variation). As for case A, L1 <strong>and</strong> L2 (Fig. 5 (a)) yield very similar results. However, in comparison to<br />
case A, underestimation is clearly increased. This is most obvious for homogeneous conditions, where<br />
L1 <strong>and</strong> L2 are only about 50% <strong>of</strong> the true length <strong>and</strong> thus no longer yield the correct result.<br />
Underestimation <strong>of</strong> L increases with heterogeneity, yielding mean values <strong>of</strong> L1 <strong>and</strong> L2 <strong>of</strong> 0.45 for<br />
σ Y 2 =4.5. Spread is about 0.75 orders <strong>of</strong> magnitude for low heterogeneity <strong>and</strong> one order <strong>of</strong> magnitude<br />
for high heterogeneities, which is similar to case A. Plume lengths L 3 (Fig. 5 (b)) show the same<br />
general behaviour as the L1 <strong>and</strong> L2, with mean values being slightly lower. In contrast to this, L is<br />
overestimated for most realizations by L4 (Fig. 5 (c)). While for homogeneous conditions a too short<br />
L4 <strong>of</strong> 0.65 is obtained, L4 increases to values larger than one for higher degrees <strong>of</strong> heterogeneity.<br />
Spread <strong>of</strong> single realizations is significantly increased in comparison to case A (compare Figs. 4 (c)<br />
<strong>and</strong> 5(c)), as a spread <strong>of</strong> about one order <strong>of</strong> magnitude can be observed for all degrees <strong>of</strong> heterogeneity.<br />
The additional error introduced by using methods 1 through 4 for plumes following Michaelis–<br />
Menten kinetics degradation is most pronounced for low heterogeneities. Compared to case A,<br />
plume lengths are underestimated to a larger extent, as can be seen by the lower mean values <strong>and</strong><br />
Fig. 6. Normalized Michaelis–Menten kinetics parameters kmax vs. MC estimated for σY 2 =0.38 (a), 1.71 (b), 2.7 (c) <strong>and</strong> 4.5<br />
(d), respectively.
C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
medians. Only for L4 plume length overestimation is observed <strong>and</strong> uncertainty is increased in<br />
comparison to case A.<br />
6.3. Case C: estimation <strong>of</strong> Michaelis–Menten kinetics parameters <strong>and</strong> plume lengths for plumes<br />
following Michaelis–Menten degradation kinetics<br />
6.3.1. Estimation <strong>of</strong> rate parameters<br />
As demonstrated in case B, using a first order kinetics approximation for plumes following MM<br />
degradation kinetics introduces an additional error, yielding less conservative estimates <strong>of</strong> plume<br />
lengths (L1−L3) as well as higher uncertainty (L4). Therefore, method 5 for the estimation <strong>of</strong> the<br />
MM parameters is tested by application to the contaminant plumes following MM degradation<br />
kinetics (see Table 1). Using the same concentration vs. distance data as in Section 6.2, MM<br />
parameters kmax <strong>and</strong> MC are estimated by a linear fit as explained in Section 4.1. These parameters<br />
then are used to calculate the length <strong>of</strong> the contaminant plumes LMM by equation (10) (Table 3).<br />
Fig. 6 presents the results <strong>of</strong> the parameter estimation in single diagrams for each σ Y 2 (results<br />
for single realizations, ensemble means with corresponding st<strong>and</strong>ard deviations as error bars,<br />
medians). Since for each realization the maximum degradation rate kmax <strong>and</strong> half-saturation<br />
concentration MC are estimated simultaneously, the parameters are shown in scatter plots. For the<br />
homogeneous case (not shown), kmax is slightly overestimated with a value <strong>of</strong> 1.06, while for MC<br />
an underestimation is observed with MC=0.91. This error results from neglecting the dispersion<br />
process. Fig. 6 shows that normalized kmax <strong>and</strong> MC are increasingly overestimated with increasing<br />
σY 2 . Also uncertainty increases with σY 2 . This is a similar behaviour as found for the first order rate<br />
constants in case A. Approximate orientation <strong>of</strong> the data points along a diagonal axis with positive<br />
slope gives evidence <strong>of</strong> a weak positive correlation between k max <strong>and</strong> M C. An overestimation <strong>of</strong><br />
kmax increases the degradation rate as long as concentrations are higher than MC. However,<br />
overestimation <strong>of</strong> MC raises the concentration threshold at which kinetic degradation transits from<br />
zeroth to slower first order, which counter-balances the effects <strong>of</strong> the kmax overestimation.<br />
Fig. 7. Normalized plume lengths LMM calculated with Eq. (10) based on estimated Michaelis–Menten kinetics parameters<br />
k max <strong>and</strong> M C.<br />
89
90 C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
6.3.2. Estimation <strong>of</strong> plume lengths<br />
For the determination <strong>of</strong> contaminant plume lengths, the estimated parameters kmax <strong>and</strong> MC are<br />
evaluated with equation (10) (Table 3). Normalized plume lengths LMM are presented in Fig. 7<br />
(single realization results, ensemble means with corresponding st<strong>and</strong>ard deviations as error bars,<br />
medians, coefficients <strong>of</strong> variation). For the homogeneous case the estimated <strong>and</strong> true plume<br />
lengths agree well with LMM=0.97. Also for σY 2 =0.38 the mean yields the correct result with<br />
L MM=1.0 <strong>and</strong> spread is about 0.5 orders <strong>of</strong> magnitude. For higher heterogeneities, however,<br />
plume lengths again tend to be underestimated with L MM decreasing to 0.80 for σ Y 2 =4.5. Spread<br />
for all degrees <strong>of</strong> heterogeneity is smaller than one order <strong>of</strong> magnitude. When the spread <strong>of</strong><br />
estimated kmax <strong>and</strong> MC values is compared to the spread <strong>of</strong> calculated MM plume lengths, it is<br />
found – as for case A – that the uncertainty in MM parameters is not fully propagated to the<br />
plume lengths. Comparing MM plume lengths with those determined for the plumes subject to<br />
first order degradation in case A (see Fig. 4), the LMM outperform L1, L2 or L3 with regard to<br />
accuracy as well as uncertainty. Only the L4 are closer to the true lengths on average, while the<br />
spread <strong>of</strong> single realizations is comparable. It is thus found in this study that the proposed method<br />
<strong>of</strong> estimating MM kinetics parameters is adequate <strong>and</strong> performs better than the methods for<br />
estimating first order rate constants.<br />
6.4. Comparison <strong>and</strong> discussion <strong>of</strong> cases B <strong>and</strong> C<br />
In this section the plume lengths LMM calculated in case C using the MM model (Table 3,<br />
equation (10)) are compared to the L1 −L4 <strong>of</strong> case B, where plume lengths were calculated using<br />
the first order kinetics approximations <strong>of</strong> equations (7) through (9) <strong>of</strong> Table 3. Table 5 compares<br />
the accuracy <strong>of</strong> calculated plume lengths using MM parameters estimated by method 5 with the<br />
four different first order approaches. Given are the percentages <strong>of</strong> realizations for which LMM<br />
constitutes a closer estimate <strong>of</strong> L than L1, L2, L3, orL4. In comparison to L1, L2 <strong>and</strong> L3 it is found<br />
that LMM is the closer estimate for 93 up to 99% <strong>of</strong> all realizations, depending on the degree <strong>of</strong><br />
heterogeneity <strong>and</strong> the first order method used. In comparison with L4 LMM constitutes the closer<br />
estimate still for 53 up to 72% <strong>of</strong> all realizations.<br />
As a more quantitative criterion, plume length error factors EF are calculated for each<br />
realization <strong>and</strong> all five estimation methods:<br />
EF ¼ L *<br />
L<br />
a<br />
with<br />
a ¼ −1jL * bL<br />
a ¼þ1jL * zL<br />
where L ⁎ <strong>and</strong> L are the estimated <strong>and</strong> true plume lengths, respectively, <strong>and</strong> the exponent a is used to<br />
obtain a st<strong>and</strong>ardized measure for over- as well as for underestimation. Hence, EF gives the degree <strong>of</strong><br />
accuracy <strong>of</strong> the plume length estimate L ⁎. For each ensemble <strong>of</strong> Li <strong>and</strong> LMM cumulative empirical<br />
Table 5<br />
Percentage <strong>of</strong> realizations for which LMM is a closer estimate <strong>of</strong> L than L1−L4<br />
LMM vs. L1, L2 % LMM vs. L3 % LMM vs. L4 %<br />
0.38 97 99 72<br />
1.71 99 99 57<br />
2.7 98 99 59<br />
4.5 94 93 53<br />
σ Y 2<br />
ð13Þ
C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
Fig. 8. Cumulative empirical distribution functions <strong>of</strong> plume length error factors for estimated plume lengths L1 through L4<br />
<strong>and</strong> LMM <strong>and</strong> aquifer heterogeneities σY 2 <strong>of</strong> 0.38 (a), 1.71 (b), 2.7 (c) <strong>and</strong> 4.5 (d). The diagrams yield the probability <strong>of</strong><br />
obtaining an estimate <strong>of</strong> the plume length L with an accuracy (EF) as given on the abscissa.<br />
distribution functions (edf) <strong>of</strong> the EF were calculated. These are presented in Fig. 8 (a) through (d) for<br />
each degree <strong>of</strong> heterogeneity. The edf give the probability <strong>of</strong> obtaining an estimate <strong>of</strong> the plume<br />
length with an EF less than or equal to the associated quantile on the abscissa. In Fig. 8 (a) with<br />
σY 2 =0.38, for example, the probability <strong>of</strong> estimating L by LMM, given an accuracy <strong>of</strong> factor 2, i.e.<br />
allowing a maximum underestimation by 50% or overestimation by 100%, is about 0.98. In contrast<br />
to this, the probability <strong>of</strong> obtaining L1 (=L2)orL3 as accurate as a factor <strong>of</strong> 2 is only 0.68 <strong>and</strong> 0.55,<br />
respectively, while for L4 the probability is approximately 0.89. When aquifer heterogeneity<br />
increases, the probabilities are reduced, as can be seen by the shift towards higher EF <strong>and</strong> the<br />
flattening <strong>of</strong> the slopes (note the logarithmic scale <strong>of</strong> the abscissa). Moreover, differences between<br />
LMM <strong>and</strong> L4 diminish. Thus for σY 2 =4.5 (Fig. 8 (d)) edf for both approaches almost coincide <strong>and</strong><br />
show the same degree <strong>of</strong> accuracy. However, LMM still yields significantly higher probabilities for a<br />
given EF than L1, L2 or L3.<br />
The results obtained in this comparison clearly support that for plumes following MM degradation<br />
kinetics usage <strong>of</strong> the MM parameter estimation approach allows a distinct improvement <strong>of</strong><br />
plume length estimates over those obtained using a first order approximation, especially when<br />
aquifer heterogeneity is not too high. For the majority <strong>of</strong> realizations investigated in this study, L MM<br />
91
92 C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
constitutes a more accurate estimate <strong>of</strong> the true plume length than any <strong>of</strong> the first order methods used.<br />
Only when aquifer heterogeneity is very high, L4 is able to yield plume length estimates <strong>of</strong> almost<br />
comparable accuracy. This is mainly, because increased transverse dispersion produces concentration<br />
pr<strong>of</strong>iles that allow a log-linear fit <strong>of</strong> a first order rate constant.<br />
7. Summary <strong>and</strong> conclusions<br />
In this study the Virtual Aquifer concept is used to assess the uncertainty involved in estimating<br />
down stream contaminant concentrations <strong>and</strong> plume lengths in heterogeneous aquifers. For such<br />
an analysis a key element is the quantification <strong>of</strong> the degradation rate at the site under study.<br />
Therefore, the main focus <strong>of</strong> this work is on the influence <strong>of</strong> this parameter on the estimated plume<br />
lengths. Three different scenarios are analysed.<br />
In case A, four different st<strong>and</strong>ard field methods based on the center line investigation strategy<br />
are tested <strong>and</strong> compared with regard to their capability <strong>of</strong> estimating first order degradation rate<br />
constants <strong>and</strong> the contaminant plume length in heterogeneous aquifers. The four methods are<br />
applied to plumes following first order degradation kinetics. It is found that both, the estimated<br />
degradation rate constants <strong>and</strong> the calculated plume lengths, are subject to high uncertainty. On<br />
average, estimated rate constants exceed the true degradation rate constant, causing calculated<br />
plume lengths that are too short. Both bias <strong>and</strong> uncertainty <strong>of</strong> estimated degradation rate constants<br />
increase with the degree <strong>of</strong> heterogeneity to about a factor in the order <strong>of</strong> a magnitude, respectively.<br />
However, the uncertainty observed in the rate constants does not fully propagate to the plume length<br />
estimates. On average plume lengths are underestimated by about 50% <strong>of</strong> the true plume length, <strong>and</strong><br />
up to a factor <strong>of</strong> ten in the worst cases. Of the four different methods, the approach <strong>of</strong> Wiedemeier et<br />
al. (1996) using concentrations normalized to a conservative tracer yields the best results for the rate<br />
constants <strong>and</strong> for the plume lengths. Consequently, this method should preferably be used.<br />
However, the presence <strong>of</strong> a suitable recalcitrant compound may not always be given. In this case,<br />
the most simple <strong>of</strong> the four approaches, which is based on the advection equation <strong>and</strong> neglects<br />
dispersive processes should be used to determine the degradation potential, as this method yields<br />
closer estimates <strong>of</strong> the first order rate constant than the approaches using the one- (Buscheck <strong>and</strong><br />
Alcantar, 1995) <strong>and</strong> two-dimensional advection dispersion equations (Zhang <strong>and</strong> Heathcote, 2003).<br />
The non-conservative plume length estimates might cause threats to down stream receptors, as the<br />
risk <strong>of</strong> contamination could be underrated. The high uncertainty when estimating plume dimensions<br />
could result in incorrect decisions regarding the necessity <strong>and</strong> dimensioning <strong>of</strong> engineered remediation<br />
measures or when considering the applicability <strong>of</strong> natural attenuation. All four methods are only<br />
applicable to steady state plumes. In reality, however, contaminant plumes <strong>of</strong>ten show temporal<br />
variations in extent <strong>and</strong> orientation as the plume exp<strong>and</strong>s, the source slowly depletes, or the <strong>flow</strong><br />
regime changes over time. For exp<strong>and</strong>ing plumes, application <strong>of</strong> methods 1–4 wouldresultinan<br />
overestimation <strong>of</strong> the rate constant, <strong>and</strong> thus in underestimation <strong>of</strong> the plume length at steady state. This<br />
is because the down stream transient contaminant concentrations are lower than those at steady state,<br />
causing a higher concentration decrease along the center line, which would be falsely attributed to the<br />
degradation process. For the same reason rate constants would also be overestimated for shrinking<br />
plumes. An examination <strong>of</strong> the current state <strong>of</strong> the plume is necessary, when one <strong>of</strong> the four methods is<br />
to be applied at a site. An overview <strong>of</strong> techniques for this purpose is given by Newell et al. (2002).<br />
Case B analyses the additional error that results from application <strong>of</strong> the four methods, although<br />
the true degradation kinetics deviate from first order. Here plumes following Michaelis–Menten<br />
degradation kinetics are investigated. Results show that for the three approaches without<br />
correction <strong>of</strong> concentrations to a conservative tracer, an additional underestimation occurs which
is largest for low heterogeneities. For the approach using the tracer correction, the estimated<br />
plume lengths still match the true lengths on average. However, the uncertainty <strong>of</strong> the plume<br />
lengths is significantly increased as compared to case A.<br />
In case C, a regression approach is introduced to estimate the parameters <strong>of</strong> the MM kinetics <strong>and</strong> to<br />
calculate the plume lengths for the same plumes as investigated in case B. Since longitudinal <strong>and</strong><br />
transverse dispersion are neglected in this method, the MM parameters are overestimated on average.<br />
Overestimation increases from less than a factor <strong>of</strong> two for low degrees <strong>of</strong> heterogeneity to a factor <strong>of</strong><br />
roughly four for high heterogeneity. Consequently an underestimation <strong>of</strong> the corresponding plume<br />
lengths is observed for most realizations. For low heterogeneity, the plume length is estimated<br />
precisely with low uncertainty, for high heterogeneity the average estimated plume length is about<br />
80% <strong>of</strong> the true plume length, with estimates as low as 40–35% in the worst cases. However, in<br />
comparison with the first order approximation investigated in case B the error resulting from this<br />
approach is significantly reduced, with the uncertainty <strong>of</strong> the plume length estimates being reduced by<br />
a factor <strong>of</strong> three. Therefore, if field data collected along the center line <strong>of</strong> a plume gives evidence <strong>of</strong><br />
MM kinetics (linear behaviour <strong>of</strong> concentrations vs. distance in a linear plot, concave down behaviour<br />
in a semi-logarithmic plot (Bekins et al., 1998)), this approach is recommended.<br />
Generally, the Virtual Aquifer concept has proven useful in assessing the performance <strong>of</strong><br />
methods for investigating first order rate constants <strong>and</strong> plume lengths from plume center line<br />
measurements. The main advantage is, that individual factors, as here the hydraulic heterogeneity<br />
<strong>and</strong> the assumption <strong>of</strong> a wrong plume kinetics, can be studied either individually in detail or in<br />
combination <strong>and</strong> under otherwise ideal conditions. However, the restrictions from the simplified<br />
setup <strong>of</strong> the model scenario have to be kept in mind when drawing conclusions for field<br />
applications. In reality, contaminant degradation follows more complicated rate laws, depending<br />
e.g. on electron acceptor <strong>and</strong> donor availability, may include transient effects, or dilution <strong>and</strong><br />
phase changes to the un<strong>saturated</strong> zone. Further studies will thus incorporate more realistic<br />
degradation kinetics as well as influences from e.g. measurement errors.<br />
Acknowledgements<br />
This work is funded by the German Ministry <strong>of</strong> Education <strong>and</strong> Research (BMBF) under grant<br />
033 05 12/033 05 13 as part <strong>of</strong> the KORA priority program, sub-project 7.2. We wish to thank<br />
Robert Walsh for his helpful comments on the manuscript. We also wish to thank our project<br />
partners at the Christian-Albrechts-University Kiel Andreas Dahmke <strong>and</strong> Dirk Schäfer for their<br />
support in our research. We acknowledge Uwe Wittmann, Iris Bernhardt <strong>and</strong> Ludwig Luckner for<br />
coordination <strong>of</strong> the project work. Furthermore we would like to acknowledge the thoughtful<br />
reviews <strong>of</strong> the anonymous reviewers <strong>and</strong> the Editor. Their comments have greatly improved the<br />
manuscript.<br />
References<br />
C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
Aziz, C.E., Newell, C.J., Gonzales, A.R., Haas, P., Clement, T.P., Sun, Y., 2000. BIOCHLOR Natural Attenuation<br />
Decision Support System User's Manual. U.S. EPA, Office <strong>of</strong> Research <strong>and</strong> Development, EPA/600/R-00/008,<br />
Washington D.C.<br />
Bauer, S., Kolditz, O., 2005. Assessing contaminant mass <strong>flow</strong> rates by the integral groundwater investigation method by<br />
using the virtual aquifer approach. In: ModelCARE 2005, Fifth International Conference on Calibration <strong>and</strong> Reliability<br />
in Groundwater Modelling. From Uncertainty to Decision Making. Pre-published Proceedings, The Hague<br />
(Scheveningen), The Netherl<strong>and</strong>s, June 2005, pp.301–307.<br />
93
94 C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
Bauer, S., Beyer, C., Kolditz, O., 2005. Assessing measurements <strong>of</strong> first order degradation rates by using the virtual aquifer<br />
approach. In: Thomson, N.R. (Ed.), GQ2004, Bringing Groundwater Quality Research to the Watershed Scale.<br />
Proceedings <strong>of</strong> a Symposium held at Waterloo, Canada, July 2004. IAHS Publication, vol. 297. IAHS Press,<br />
Wallingford, pp. 274–281.<br />
Bauer, S., Beyer, C., Kolditz, O., 2006. Assessing measurement uncertainty <strong>of</strong> first-order degradation rates in<br />
heterogeneous aquifers. Water Resour. Res. 42, W01420. doi:10.1029/2004WR003878.<br />
Baveye, P., Valocchi, A., 1989. An evaluation <strong>of</strong> mathematical models <strong>of</strong> the transport <strong>of</strong> biologically reacting solutes in<br />
<strong>saturated</strong> soils <strong>and</strong> aquifers. Water Resour. Res. 25 (6), 1413–1421.<br />
Bear, J., 1972. Dynamics <strong>of</strong> Fluids in Porous Media. Elsevier, Amsterdam. 784 pp.<br />
Bekins, B.A., Warren, E., Godsy, E.M., 1998. A comparison <strong>of</strong> zero-order, first-order, <strong>and</strong> Monod biotransformation<br />
models. Ground Water 36 (2), 261–268.<br />
Beyer, C., Bauer, S., Kolditz, O., 2005. Uncertainty assessment <strong>of</strong> degradation rate measurements in heterogeneous media<br />
using the virtual aquifer approach. In: Kolditz, O., Bauer, S., Gronewold, J. (Eds.), Proceedings <strong>of</strong> the 5th Workshop<br />
“Porous Media”, Blaubeuren, Germany, December 2004. ZAG Publisher, Tübingen.<br />
Bockelmann, A., Zamfirescu, D., Ptak, T., Grathwohl, P., Teutsch, G., 2003. Quantification <strong>of</strong> mass fluxes <strong>and</strong> natural<br />
attenuation rates at an industrial site with a limited monitoring network: a case study. J. Contam. Hydrol. 60 (1–2), 97–121.<br />
Buscheck, T.E., Alcantar, C.M., 1995. Regression techniques <strong>and</strong> analytical solutions to demonstrate intrinsic<br />
bioremediation. In: Hinchee, R.E., Wilson, T.J., Downey, D. (Eds.), Intrinsic Bioremediation. Battelle Press,<br />
Columbus, OH, pp. 109–116.<br />
Chapelle, F.H., Bradley, P.M., Lovley, D.R., Vroblesky, D.A., 1996. Measuring rates <strong>of</strong> biodegradation in a contaminated<br />
aquifer using field <strong>and</strong> laboratory methods. Ground Water 34 (4), 691–698.<br />
Dagan, G., 1989. Flow <strong>and</strong> Transport in Porous Formations. Springer, Heidelberg. 465 pp.<br />
Domenico, P.A., 1987. An analytical model for multidimensional transport <strong>of</strong> a decaying contaminant species. J. Hydrol.<br />
91 (1–2), 49–59.<br />
Islam, J., Naresh, S., O'Sullivan, M., 2001. Modeling biogeochemical processes in leachate-contaminated soils: a review.<br />
Transp. Porous Media 43 (3), 407–440.<br />
Herfort, M., 2000. Reactive Transport <strong>of</strong> Organic Compounds Within a Heterogeneous Porous Aquifer, PhD Thesis,<br />
University <strong>of</strong> Tübingen, Tübinger Geowissenschaftliche Arbeiten (TGA), 54, Tübingen. 76 pp.<br />
Kolditz, O., 2002. Computational Methods in Environmental Fluid Dynamics. Springer, Heidelberg. 400 pp.<br />
Kolditz, O., Bauer, S., 2004. A process-oriented approach to computing multifield problems in porous media. J. Hydroinform.<br />
6(3),225–244.<br />
Kolditz, O., de Jonge, J., Beinhorn, M., Xie, M., Kalbacher, T., Wang, W., Bauer, S., McDermott, C., Chen, C., Beyer, C.,<br />
Gronewold, J., Kemmler, D., Manabe, T., Legeida, D., Adamidis, P., 2004. GeoSys- Theory <strong>and</strong> users manual. release<br />
4.1. GeoSystems Research, Center for <strong>Applied</strong> Geoscience, University <strong>of</strong> Tübingen, http://www.uni-tuebingen.de/zag/<br />
geohydrology, Tübingen.<br />
McNab Jr., W.W., 2001. A Monte Carlo simulation method for assessing biotransformation effects on groundwater fuel<br />
hydrocarbon plume lengths. Comput. Geosci. 27 (1), 31–42.<br />
McNab Jr., W.W., Dooher, B.P., 1998. A critique <strong>of</strong> a steady-state analytical method for estimating contaminant<br />
degradation rates. Ground Water 36 (6), 983–987.<br />
Newell, C.J., McLeod, R.K., Gonzales, J.R., 1996. BIOSCREEN Natural Attenuation Decision Support System User's<br />
Manual. Version 1.3. U.S. EPA, Office <strong>of</strong> Research <strong>and</strong> Development, EPA/600/R-96/087, Washington D.C.<br />
Newell, C.J., Rifai, H.S., Wilson, J.T., Connor, J.A., Aziz, J.A., Suarez, M.P., 2002. Calculation <strong>and</strong> use <strong>of</strong> first-order rate<br />
constants for monitored natural attenuation studies. U.S. EPA Ground Water Issue, U.S. EPA, Office <strong>of</strong> Research <strong>and</strong><br />
Development, EPA/540/S-02/500, Washington D.C.<br />
Parlange, J.-Y., Starr, J.L., Barry, D.A., Braddock, R.D., 1984. Some approximate solutions <strong>of</strong> the transport equation with<br />
irreversible reactions. Soil Sci. 137, 434–442.<br />
Rehfeldt, K.R., Boggs, J.M., Gelhar, L.W., 1992. Field study <strong>of</strong> dispersion in a heterogeneous aquifer, 3, geostatistical<br />
analysis <strong>of</strong> hydraulic conductivity. Water Resour. Res. 28 (12), 3309–3324.<br />
Rifai, H.S., Newell, C.J., Gonzales, J.R., Dendrou, S., Kennedy, L., Wilson, J.T., 1998. BIOPLUME III Natural<br />
Attenuation Decision Support System Version 1.0 User's Manual. U.S. EPA, Office <strong>of</strong> Research <strong>and</strong> Development,<br />
EPA/600/R-98/010, Washington D.C.<br />
Rittmann, B., VanBriesen, J.M., 1996. Microbiological processes in reactive <strong>modeling</strong>. In: Lichtner, P., Steefel, C.,<br />
Oelkers, E. (Eds.), Reactive Transport in Porous Media. Reviews in Mineralogy, vol. 34. Mineralogical Society <strong>of</strong><br />
America, Washington, DC, pp. 311–334.<br />
Robinson, J.A., 1985. Determining microbial kinetic parameters using nonlinear regression analysis. Adv. Microb. Ecol. 8,<br />
61–114.
C. Beyer et al. / Journal <strong>of</strong> Contaminant Hydrology 87 (2006) 73–95<br />
Schäfer, D., Dahmke, A., Kolditz, O., Teutsch, G., 2002. “Virtual Aquifers”: a concept for evaluation <strong>of</strong> exploration,<br />
remediation <strong>and</strong> monitoring strategies. In: Kovar, K., Hrkal, Z. (Eds.), Calibration <strong>and</strong> Reliability in Groundwater<br />
Modelling: A Few Steps Closer to Reality. Proceedings <strong>of</strong> the ModelCARE 2002 Conference, Prague, Czech<br />
Republic, June 2002. IAHS Publication, vol. 277. IAHS Press, Oxfordshire, pp. 52–59.<br />
Schäfer, D., Manconi, A., Dahmke, A., 2004a. Identification <strong>and</strong> consequences <strong>of</strong> different kinetic approaches for simulation<br />
<strong>of</strong> microbial degradation. In: Aagaard, P., Bedbur, E., Bidoglio, G., C<strong>and</strong>ela, L., Nuetzmann, G., Trevisan, M., Vanclooster,<br />
M., Viotti, P. (Eds.), Proceedings <strong>of</strong> the International Workshop: Saturated <strong>and</strong> Un<strong>saturated</strong> Zone, Integration <strong>of</strong> Process<br />
Knowledge into Effective Models. University <strong>of</strong> Rome La Sapienza“, Rome, Italy, May 2004, pp. 291–296.<br />
Schäfer, D., Schlenz, B., Dahmke, A., 2004b. Evaluation <strong>of</strong> exploration <strong>and</strong> monitoring methods for verification <strong>of</strong> natural<br />
attenuation using the virtual aquifer approach. Biodegradation 15 (6), 453–465.<br />
Simkins, S., Alex<strong>and</strong>er, M., 1984. Models for mineralization kinetics with variables <strong>of</strong> substrate concentration <strong>and</strong><br />
population density. Appl. Environ. Microbiol. 47 (6), 1299–1306.<br />
Stenback, G.A., Ong, S.K., Rogers, S.W., Kjartanson, B.H., 2004. Impact <strong>of</strong> transverse <strong>and</strong> longitudinal dispersion on<br />
first-order degradation rate constant estimation. J. Contam. Hydrol. 73, 3–14.<br />
Suarez, M.P., Rifai, H.S., 2002. Evaluation <strong>of</strong> BTEX remediation by natural attenuation at a coastal facility. Ground Water<br />
Monit. Remediat. 22 (1), 62–77.<br />
Suarez, M.P., Rifai, H.S., 2004. Modeling natural attenuation <strong>of</strong> total BTEX <strong>and</strong> benzene plumes with different kinetics.<br />
Ground Water Monit. Remediat. 24 (3), 53–68.<br />
Sudicky, E.A., 1986. A natural gradient experiment on solute transport in a s<strong>and</strong> aquifer: spatial variability <strong>of</strong> hydraulic<br />
conductivity <strong>and</strong> its role in the dispersion process. Water Resour. Res. 22 (13), 2069–2082.<br />
U.S. EPA, 1998. Technical protocol for evaluating natural attenuation <strong>of</strong> chlorinated solvents in groundwater, U.S. EPA,<br />
Office <strong>of</strong> Research <strong>and</strong> Development, EPA/600/R/128, Washington, D.C.<br />
Wiedemeier, T.H., Swanson, M.A., Wilson, J.T., Kampbell, D.H., Miller, R.N., Hansen, J.E., 1996. Approximation <strong>of</strong><br />
biodegradation rate constants for monoaromatic hydrocarbons (BTEX) in ground water. Ground Water Monit.<br />
Remediat. 16 (3), 186–194.<br />
Wiedemeier, T.H., Rifai, H.S., Wilson, T.J., Newell, C., 1999. Natural Attenuation <strong>of</strong> Fuels <strong>and</strong> Chlorinated Solvents in the<br />
Subsurface. Wiley, New York. 632 pp.<br />
Wilson, J.T., Kolhatkar, R., 2002. Role <strong>of</strong> natural attenuation in life cycle <strong>of</strong> MTBE plumes. J. Environ. Eng. 128 (9),<br />
876–882.<br />
Wilson, J.T., Pfeffer, F.M., Weaver, J.W., Kampbell, D.H., Wiedemeier, T.H., Hansen, J.E., Miller, R.N., 1994. Intrinsic<br />
bioremediation <strong>of</strong> JP-4 jet fuel. Symposium on Intrinsic Bioremediation <strong>of</strong> Ground Water, Denver, Colorado, EPA/<br />
540/R-94/515, pp. 60–72.<br />
Wilson, R.D., Thornton, S.F., Mackay, D.M., 2004. Challenges in monitoring the natural attenuation <strong>of</strong> spatially variable<br />
plumes. Biodegradation 15 (6), 459–469.<br />
Zamfirescu, D., Grathwohl, P., 2001. Occurrence <strong>and</strong> attenuation <strong>of</strong> specific organic compounds in the groundwater plume<br />
at a former gasworks site. J. Contam. Hydrol. 53 (1–2), 407–427.<br />
Zhang, Y.-K., Heathcote, R.C., 2003. An improved method for estimation <strong>of</strong> biodegradation rate with field data. Ground<br />
Water Monit. Remediat. 23 (3), 112–116.<br />
95
Enclosed Publication 4<br />
Bauer, S., Beyer, C., Kolditz, O. (2007): Einfluss von Heterogenität und Messfehler auf die<br />
Bestimmung von Abbauraten erster Ordnung - eine Virtueller Aquifer Szenarioanalyse.<br />
(Influence <strong>of</strong> heterogeneity <strong>and</strong> measurement error on the determination <strong>of</strong> first order<br />
degradation rates by us1ing the virtual aquifer approach.), Grundwasser, 12, 3–14, doi:<br />
10.1007/s00767-007-0019-8<br />
The enclosed article is made available with the permission <strong>of</strong> Springer <strong>and</strong> was published in<br />
the journal Grundwasser, 12, 3-14 (2007). Copyright © 2007 Springer.<br />
The article can be obtained online via SpringerLink at<br />
http://www.springerlink.com/openurl.asp?genre=journal&eissn=1432-1165.
Einfluss von Heterogenität und Messfehler auf die<br />
Bestimmung von Abbauraten erster Ordnung – eine<br />
Virtueller Aquifer Szenarioanalyse ∗<br />
Influence <strong>of</strong> heterogeneity <strong>and</strong> measurement error on the determination<br />
<strong>of</strong> first order degradation rates - a virtual aquifer scenario analysis<br />
Sebastian Bauer, Christ<strong>of</strong> Beyer, Olaf Kolditz<br />
Zentrum für Angew<strong>and</strong>te Geowissenschaften, Universität Tübingen, Sigwartstraße 10,<br />
D 72076 Tübingen. Tel: 07071-2973171. Fax: 07071 5059<br />
e-mail: sebastian.bauer@uni-tuebingen.de, christ<strong>of</strong>.beyer@uni-tuebingen.de, kolditz@uni-tuebingen.de<br />
Header: Heterogenität und Messfehler bei Abbauratenbestimmung<br />
Kurzfassung<br />
Die grundlegende Idee des Virtuellen Aquifers ist, durch numerische Modellierung von typischen<br />
Schadensfällen Erkundungsstrategien zu simulieren und zu bewerten. In diesem Beitrag wird die<br />
Bestimmung von Abbauraten erster Ordnung untersucht. Eine Schadensquelle wird dabei in einen<br />
virtuellen Aquifer eingebracht und die stationäre Schadst<strong>of</strong>ffahne unter der Annahme einer<br />
Abbaukinetik erster Ordnung simuliert. Diese Fahne wird dann durch Beobachtungspegel entlang der<br />
Zentrallinie der Fahne untersucht. Anh<strong>and</strong> von vier typischen Methoden werden Abbauraten erster<br />
Ordnung berechnet und mit dem vorgegebenen Wert verglichen. Dieser Vergleich wird für<br />
unterschiedlich stark ausgeprägte hydraulische Heterogenitäten durchgeführt. Dabei zeigt sich, dass<br />
mit zunehmender Heterogenität die ermittelten Abbauraten die tatsächliche Abbaurate um Größenordnungen<br />
überschätzen können und sie somit sehr unsicher sind. Bei der Untersuchung der<br />
Messfehler wurden Abweichungen bei der Bestimmung der Piezometerhöhe und der Konzentration<br />
angenommen. Hierbei ergibt sich, dass Messfehler ebenfalls zu einer hohen Unsicherheit der<br />
Ratenkonstante führen können, wobei Messfehler der Piezometerhöhe einen stärkeren Einfluss<br />
haben.<br />
Abstract<br />
The principal idea <strong>of</strong> the Virtual Aquifer is to simulate <strong>and</strong> evaluate monitoring strategies <strong>and</strong><br />
remediation options for contaminated sites by modelling <strong>of</strong> typical contamination scenarios. Here the<br />
determination <strong>of</strong> first order degradation rates is studied. A virtual reality is generated by simulating the<br />
spreading <strong>of</strong> a plume, originating from a defined source <strong>and</strong> subject to first order degradation. This<br />
plume is investigated using monitoring wells placed along the plume center line. From the information<br />
thus obtained first order degradation rates are calculated by methods typically used <strong>and</strong> are then<br />
compared to the predefined value. This comparison is conducted for different degrees <strong>of</strong><br />
heterogeneity. It is found that with increasing heterogeneity degradation rates overestimate the real<br />
degradation rate by up to orders <strong>of</strong> magnitude <strong>and</strong> show a high uncertainty. Then measurement errors<br />
are introduced for piezometric head <strong>and</strong> concentration measurements. It is found that deviations <strong>of</strong> the<br />
estimated first order rate constant from the true one <strong>of</strong> up to orders <strong>of</strong> magnitude can occur, with<br />
errors <strong>of</strong> the piezometric head measurement causing the dominant uncertainty.<br />
Keywords: natural attenuation, heterogeneity, first-order degradation rate, virtual aquifer, scenario<br />
analysis<br />
∗ Bauer, S., Beyer, C., Kolditz, O. (2007): Einfluss von Heterogenität und Messfehler auf die Bestimmung von Abbauraten erster<br />
Ordnung - eine Virtueller Aquifer Szenarioanalyse. (Influence <strong>of</strong> heterogeneity <strong>and</strong> measurement error on the determination <strong>of</strong><br />
first order degradation rates by us1ing the virtual aquifer approach.), Grundwasser, 12, 3–14, doi: 10.1007/s00767-007-0019-8.<br />
Der Artikel wurde in der Zeitschrift Grundwasser publiziert und mit Erlaubnis von Springer reproduziert. Copyright © 2007 Springer.<br />
Der Artikel ist online abrufbar via SpringerLink: http://www.springerlink.com/openurl.asp?genre=journal&eissn=1432-1165.<br />
1
Einführung<br />
In dieser Arbeit wird eine „Virtueller Aquifer“ Szenarioanalyse verwendet, um die Unsicherheit bei der<br />
Bestimmung von Abbauratenkonstanten erster Ordnung anh<strong>and</strong> von Messstellen auf der Zentrallinie<br />
der Schadst<strong>of</strong>ffahne („Center line method“) im Kontext von Natural Attenuation zu untersuchen und zu<br />
quantifizieren. Natural Attenuation (NA), auch als Bioremediation bekannt, bezieht sich auf die Abnahme<br />
von Schadst<strong>of</strong>fkonzentrationen durch natürliche Abbauprozesse mit zunehmendem Abst<strong>and</strong> von<br />
der Quelle (US-EPA, 1999; WIEDEMEIER et al., 1999). Dabei werden Dispersion, Verdünnung,<br />
Sorption, Ausgasung und Bioabbau betrachtet, wobei der Bioabbau der einzige Prozess ist, der zu<br />
einer Verringerung der Schadst<strong>of</strong>fmasse führt. Die an einem St<strong>and</strong>ort ablaufenden Prozesse müssen<br />
sorgfältig charakterisiert werden, um Aussagen über NA treffen zu können. Hierbei können<br />
insbesondere die Abbauraten der betrachteten Schadst<strong>of</strong>fe für mögliche Sanierungen und das<br />
St<strong>and</strong>ortmanagement eine Rolle spielen. Die Abbauraten werden verwendet, um das gesamte NA-<br />
Potential am St<strong>and</strong>ort zu charakterisieren, um die Länge von Schadst<strong>of</strong>ffahnen in der Zukunft zu<br />
prognostizieren und um unterstromige Konzentrationen zu berechnen, die für eine Auswirkungsprognose<br />
benötigt werden (WIEDEMEIER et al., 1999).<br />
Zur Bestimmung von Abbauraten im Feld stehen derzeit mehrere Methoden zur Verfügung, wie z.B.<br />
Massenbilanzen, in-situ Mikrokosmosstudien oder die Verwendung von Konzentrations-Abst<strong>and</strong>s-<br />
Beziehungen, die auf der Fahnenzentrallinie einer stationären Schadst<strong>of</strong>ffahne ermittelt wurden<br />
(CHAPELLE, 1996). Für letztere sind in der Literatur vier unterschiedliche Methoden beschrieben. Die<br />
erste beruht auf der eindimensionalen Transportgleichung mit Advektion und Abbau erster Ordnung<br />
(WIEDEMEIER et al., 1996). Die zweite Methode ist eine Erweiterung der ersten, indem die<br />
Konzentrationen des betrachteten Schadst<strong>of</strong>fes auf die Konzentrationen eines nichtreaktiven Mitkontamin<strong>and</strong>en<br />
normiert werden (WIEDEMEIER et al., 1996, 1999; WILSON et. al., 1994). Die dritte<br />
Methode, von BUSCHECK & ALCANTAR (1995) vorgeschlagen, beruht auf der eindimensionalen<br />
Transportgleichung mit Advektion, Dispersion und Abbau erster Ordnung und wurde bereits an einigen<br />
St<strong>and</strong>orten eingesetzt (CHAPELLE et al., 1996; WIEDEMEIER et al., 1996; ZAMFIRESCU & GRATHWOHL,<br />
2001; SUAREZ & RIFAI, 2002; BOCKELMANN et al., 2003). In letzter Zeit wurden als Erweiterung zur<br />
Methode von BUSCHECK & ALCANTAR (1995) zwei- und dreidimensionale Methoden entwickelt (ZHANG<br />
& HEATHCOTE, 2003; STENBACK et al., 2004). ZHANG & HEATHCOTE (2003) konnten zeigen, dass die<br />
eindimensionale Methode durch Vernachlässigung der Querdispersion die Abbauraten um 21% im<br />
Vergleich zum zweidimensionalen und um 65% im Vergleich zum dreidimensionalen Fall überschätzt.<br />
MCNAB Jr. & DOOHER (1998) beschrieben, wie transversale Dispersion und instationäre Strömungszustände<br />
Konzentrationsverteilungen erzeugen können, die durch die Methode von BUSCHECK &<br />
ALCANTAR (1995) dann fälschlich als Bioabbau klassifiziert werden können, obwohl am St<strong>and</strong>ort kein<br />
Abbau stattfindet.<br />
Aufgrund der eingeschränkten Zugänglichkeit des Untergrundes sind Beobachtungen an<br />
kontaminierten St<strong>and</strong>orten nur an einzelnen räumlichen Punkten möglich. Die aus einzelnen<br />
Messpunkten abgeleiteten Ergebnisse und Aussagen sind daher immer mit Unsicherheit behaftet, die<br />
das eingeschränkte und punktuelle Wissen über den St<strong>and</strong>ort widerspiegelt. Eine „Virtuelle Aquifer“<br />
Szenarioanalyse kann verwendet werden, um diese Unsicherheit näher zu betrachten und zu quantifizieren<br />
(SCHÄFER et al., 2002; SCHÄFER et al., 2004; BAUER et al., 2006; BEYER et al., 2005a; BAUER &<br />
KOLDITZ, 2005). Eine Untersuchung des Einflusses der Parametrisierung der Abbaukinetik auf die<br />
prognostizierte Fahnenlänge wird in SCHÄFER et al. (2005) durchgeführt. Dabei werden numerische<br />
synthetische Modelle typischer Aquifere generiert. Mithilfe eines reaktiven Transportmodells können<br />
dann typische Schadensszenarien, wie beispielsweise die Ausbreitung einer Schadst<strong>of</strong>ffahne ausgehend<br />
von einer Schadst<strong>of</strong>fquelle, simuliert werden. Der große Vorteil der Virtuellen Aquifere ist, dass<br />
die so erhaltene realistische räumliche Verteilung aller Parameter, wie z.B. Piezometerhöhe oder<br />
Konzentration, exakt bekannt ist. Die virtuelle Schadst<strong>of</strong>fahne wird dann in einem zweiten Schritt<br />
durch typische Erkundungsstrategien (hier: Zentrallinienmethode) untersucht. Eine Messung im<br />
virtuellen Aquifer bedeutet, die entsprechenden Werte aus der Modellausgabedatei zu lesen. Bei<br />
dieser Erkundung der virtuellen Schadst<strong>of</strong>ffahne werden nur die Messwerte (hier: Piezometerhöhe,<br />
Konzentration) berücksichtigt, die auch an einem echten St<strong>and</strong>ort erhalten werden können.<br />
Ausgehend von diesen (virtuellen) Messwerten werden dann weitere Parameter ermittelt (hier:<br />
hydraulische Durchlässigkeiten, Abbauraten). Bei diesem zweiten Schritt ist also die richtige<br />
Parameterverteilung nicht bekannt. Indem das Ergebnis der virtuellen Erkundung im dritten Schritt mit<br />
der wahren Parameterverteilung (hier: Abbauratenkoeffizient) verglichen wird, können die<br />
verwendeten Untersuchungsmethoden getestet und bewertet werden.<br />
Die räumliche Heterogenität von Aquiferparametern hat einen erheblichen Einfluss auf die<br />
Fahnenentwicklung und die resultierende Konzentrationsverteilung eines Schadst<strong>of</strong>fes im Aquifer. Die<br />
2
oben erwähnten Methoden zur Bestimmung von Abbauratenkonstanten erster Ordnung unterliegen<br />
daher dem Einfluss der Heterogenität, da sie auf Messdaten entlang der vermuteten Fahnenzentrallinie<br />
und auf Abschätzungen der Dispersivität beruhen. Wird die Grundwasserfließrichtung<br />
falsch abgeschätzt, kann die Fahnenzentrallinie leicht verfehlt werden (WILSON et al., 2004). Bisher<br />
existiert in der Literatur keine Studie, die den Einfluss der Heterogenität auf die Ermittlung von<br />
Abbauratenkonstanten erster Ordnung untersucht. Bei der Probenahme im realen Fall können darüber<br />
hinaus noch Messfehler auftreten, die ebenfalls die Ratenkonstanten beeinflussen. Diese können<br />
sowohl bei der Pegeleinmessung, bei der Bestimmung der hydraulischen Leitfähigkeit, bei Wasserst<strong>and</strong>smessungen<br />
und insbesondere bei Konzentrationsmessungen auftreten. Auch hierzu existieren<br />
noch keine Untersuchungen in der Literatur. Ziel dieser Arbeit ist daher a) die Genauigkeit von<br />
Abbauraten, wie sie aus Feldmessungen gewonnen werden, zu ermitteln, und b) die verschiedenen<br />
Methoden zu Bestimmung von Abbauraten zu bewerten. Dazu wird der Einfluss von hydraulischer<br />
Heterogenität und Messfehlern anh<strong>and</strong> einer „Virtueller Aquifer“ Szenarioanalyse untersucht. Hierzu<br />
werden in einem synthetischen Aquifer mit unterschiedlich stark ausgeprägter Heterogenität virtuelle<br />
stationäre Fahnen simuliert, die einer Abbaukinetik erster Ordnung unterliegen. Die anh<strong>and</strong> der oben<br />
genannten vier Methoden bestimmten Abbauratenkonstanten erster Ordnung werden mit dem wahren<br />
Wert verglichen und so Aussagen über die Zuverlässigkeit und Unsicherheit der Methoden und der<br />
resultierenden Ratenkonstanten abgeleitet. Bei der Untersuchung des Einflusses des Messfehlers auf<br />
die Ratenkonstante wird anschließend von einem homogenen Aquifer ausgegangen. Der Einfluss von<br />
Messfehlern bei der Bestimmung der Piezometerhöhe und bei der Konzentrationsmessung wird<br />
separat untersucht und quantifiziert.<br />
Methodik<br />
Das verwendete Modellgebiet ist zweidimensional und misst 184 m auf 64 m (vergl. Abb. 1). Das<br />
Grundwasser strömt von links nach rechts, der mittlere hydraulische Gradient beträgt 0.003 und wird<br />
durch Festpotentialr<strong>and</strong>bedingungen am linken und rechten Modellr<strong>and</strong> erzeugt. Alle <strong>and</strong>eren Ränder<br />
sind undurchlässig. Das Modellgebiet ist mit einer Gitterweite von 0.5 m regelmäßig diskretisiert. 11.5<br />
m vom oberstromigen R<strong>and</strong> ist eine Schadst<strong>of</strong>fquelle, durch eine Festkonzentrationsr<strong>and</strong>bedingung<br />
dargestellt, eingebracht, die einen Schadst<strong>of</strong>f emittiert, der einer Abbaukinetik erster Ordnung folgend<br />
mit einer Ratenkonstanten λ abgebaut wird. Zusätzlich wird ein nichtreaktiver St<strong>of</strong>f aus der Quelle<br />
freigesetzt. Es werden Quellbreiten von 4, 8 und 16 m betrachtet. Weder Sorption noch Ausgasung<br />
werden berücksichtigt. Der Schadst<strong>of</strong>f verhält sich also genau so, wie es von den Methoden zur<br />
Bestimmung der Abbaurate angenommen wird – eine Annahme, die in der Realität nur annähernd<br />
erfüllt ist, da die Abbauraten komplexeren Gesetzmäßigkeiten folgen. Diese Abstraktion ist hier<br />
notwendig, um die vier Methoden möglichst genau untersuchen zu können. Als longitudinale und<br />
transversale Dispersionslängen werden im Modell 0.25 m und 0.05 m angenommen. Diese stellen die<br />
lokale Dispersion dar, in die effektive Dispersion geht noch die Heterogenität der hydraulischen<br />
Durchlässigkeit ein, die hier im Modell explizit dargestellt wird. Mit Hilfe des Programms GeoSys<br />
(KOLDITZ et al., 2005) wird dann eine stationäre Schadst<strong>of</strong>ffahne erzeugt. Um den Einfluss räumlicher<br />
Heterogenität untersuchen zu können, wird die hydraulische Durchlässigkeit K als eine ln –<br />
normalverteilte Zufallsvariable aufgefasst mit einem Erwartungswert von ln(K) = -9.54, was einer<br />
mittleren hydraulischen Durchlässigkeit von 7.2 10 -5 m s -1 entspricht. Mit der verwendeten Porosität<br />
von 0.33 ergibt sich so eine mittlere Transportgeschwindigkeit von 6.5·10 -7 m s -1 . Die räumliche<br />
Struktur wird durch ein isotropes exponentielles Covarianzmodell Cln(K) = σ 2 exp (-∆h/l) abgebildet,<br />
wobei eine integrale Länge von l = 2.67 m verwendet wird, was einer Korrelationslänge von 8.0 m<br />
entspricht (RUBIN, 2003). Vier Klassen von hydraulischer Heterogenität werden betrachtet, die durch<br />
ln(K)-Varianzen σ 2 von 0.38, 1.71, 2.70 und 4.50 gegeben sind und das Spektrum von wenig bis stark<br />
heterogenen Bedingungen abdecken. Der Wert σ 2 = 0.38, das Covarianzmodell, die mittlere<br />
hydraulische Durchlässigkeit und Porosität als auch die integrale Länge l entsprechen dem St<strong>and</strong>ort<br />
Borden (SUDICKY, 1986). Der Wert 1.71 wurde in einem alluvialen Aquifer in Süddeutschl<strong>and</strong> bestimmt<br />
(HERFORT, 2000), während die Werte 2.70 und 4.50 von Untersuchungen der Columbus Air Force<br />
Base stammen (REHFELDT et al., 1992). Die geostatistische S<strong>of</strong>tware gstats2.4 (PEBESMA &<br />
WESSELING, 1998) wurde zur Generierung der Zufallsverteilungen der hydraulischen Durchlässigkeit<br />
durch unkonditionierte Gauss’sche Simulation verwendet.<br />
Die so erzeugten Fahnen werden nun anh<strong>and</strong> der Zentrallinien-Methode untersucht. Dabei sind<br />
anfänglich drei Beobachtungsbrunnen vorgegeben, von denen einer direkt in der Schadst<strong>of</strong>fquelle<br />
liegt, während die <strong>and</strong>eren keine Kontamination zeigen (Abb.1 a). An diesen drei Brunnen werden nun<br />
die Piezometerhöhen „gemessen“, indem die Modellausgabe gelesen wird. Durch Konstruktion eines<br />
hydrogeologischen Dreiecks wird die lokale Fließrichtung an der Schadensquelle bestimmt (Abb.1b).<br />
3
Entlang der so bestimmten Grundwasserfließrichtung werden nun drei neue Messstellen im Abst<strong>and</strong><br />
von je 10 m installiert. An diesen Messstellen werden dann Piezometerhöhen, Konzentrationen des<br />
reaktiven und des nichtreaktiven Schadst<strong>of</strong>fes und die lokale hydraulische Durchlässigkeit „gemessen“<br />
(Abb. 1c). Aus der Differenz der Piezometerhöhen, den hydraulischen Durchlässigkeiten und der<br />
wahren, d.h. der auch im Modell verwendeten, Porosität wird die Grundwasserfließgeschwindigkeit<br />
bestimmt. Zusammen mit den gemessenen Schadst<strong>of</strong>fkonzentrationen können dann Abbauraten<br />
erster Ordnung anh<strong>and</strong> von vier Methoden berechnet werden. Der Virtuelle Aquifer und das Messszenario<br />
sind so ausgelegt, dass optimale Bedingungen für eine Bestimmung der Abbauraten<br />
vorliegen. Die einzige Unsicherheit wird durch die heterogene Verteilung der hydraulischen<br />
Durchlässigkeit erzeugt.<br />
4<br />
Anfangssituation<br />
Erkundungsschritt 1<br />
Erkundungsschritt 2<br />
Virtuelle Realität<br />
Abb. 1 Verwendete Methodik, um Konzentrationen von der Zentrallinie der Fahne zu erhalten: a)<br />
Anfangszust<strong>and</strong>, b) Abschätzung der Fließrichtung anh<strong>and</strong> des hydrogeologischen Dreiecks, c)<br />
„gemessene“ Konzentrationen auf der ermittelten Zentrallinie, d) Vergleich mit der Virtuellen Realität<br />
(Piezometerhöhen und Fahne).<br />
Für den zweiten Teil dieses Beitrags, in dem der Einfluss von Messfehlern untersucht wird, wird der<br />
Aquifer als homogen angenommen. Die Unsicherheit wird nun also nicht durch das heterogene<br />
Fließfeld erzeugt, sondern durch fehlerhaftes „Messen“ von Piezometerhöhe und Schadst<strong>of</strong>fkonzentration.<br />
Im Falle der Piezometerhöhe lautet das Fehlermodell:<br />
[1] h′ = h + z∆hmax<br />
wobei h’ die fehlerbehaftete, hh die wahre Piezometerhöhe und ∆hmax der maximale Messfehler für die<br />
Piezometerhöhe ist. z ist eine gleichverteilte Zufallszahl aus dem Intervall [-1, 1], mit der ∆hmax
multipliziert wird, um einen fehlerbehafteten Messwert für die Piezometerhöhe aus dem Intervall [h-<br />
∆hmax, h+∆hmax ] zu erhalten. Die fehlerbehafteten Messwerte streuen also symmetrisch um den<br />
wahren Wert. ∆hmax wird zwischen 0 und 5 cm variiert. Bei der Ermittlung der Ratenkonstante wird nun<br />
bei der Probenahme für die Piezometerhöhe ein Messfehler gemäß des vorgestellten Fehlermodells<br />
berücksichtigt. Die gesamte Auswertung wird einhundert mal durchgeführt, um statistisch<br />
repräsentative Aussagen über den Einfluss des Messfehlers auf die Ratenkonstante zu erlangen.<br />
Da für die Messung von Konzentrationen ein höherer Messfehlerbereich zu erwarten ist, wurde das<br />
Fehlermodell angepasst:<br />
a<br />
[2] c ′ = c z∆c<br />
)<br />
5<br />
( max<br />
wobei c’ die fehlerbehaftete, ch die wahre Konzentration und ∆cmax der maximale Messfehlerfaktor für<br />
die Konzentration ist. Der Exponent a ist –1 oder 1 und wird zufällig bestimmt, z ist eine gleichverteilte<br />
Zufallszahl aus dem Intervall [0, 1], mit der ∆cmax multipliziert wird. Man erhält so einen fehlerbehaftet<br />
Messwert für die Konzentration aus dem Intervall [c/∆cmax, c∆cmax ]. ∆cmax wird zwischen 1 und 100<br />
variiert. Dieses Verfahren ist ähnlich zu dem für die Piezometerhöhe angewendeten Verfahren und ein<br />
Messfehlerfaktor von 2 entspricht dem umgangsprachlichen „auf einen Faktor 2 genau“. Für die<br />
Bestimmung des Einflusses von Konzentrationsmessfehlern wurden ein homogenes Strömungsfeld<br />
und keine Messfehler für die Piezometerhöhe angenommen. Für den konservativen und den reaktiven<br />
St<strong>of</strong>f wird dasselbe Messfehlermodell angenommen.<br />
Die vier Methoden, anh<strong>and</strong> derer die Abbauraten bestimmt werden, sind in Tabelle 1 aufgeführt.<br />
Methode 1 stellt die Lösung zur eindimensionalen Advektionsgleichung mit Abbau erster Ordnung dar.<br />
Methode 2 wurde von WIEDEMEIER et al. (1996) vorgeschlagen, und baut auf Methode 1 auf.<br />
Allerdings werden die Konzentrationen des reaktiven St<strong>of</strong>fes auf die Konzentrationen des nichtreaktiven<br />
St<strong>of</strong>fes bezogen. Dadurch berücksichtigt Methode 2 Dispersion, Verdünnung und „Aus der<br />
Fahne Messen“ an Beobachtungspegeln, die nicht genau auf der Zentrallinie der Fahne liegen. Die<br />
dritte Methode wurde von BUSCHECK & ALCANTAR (1995) vorgestellt und basiert auf der eindimensionalen<br />
Transportgleichung mit Advektion, Dispersion und Abbau erster Ordnung. Diese Methode<br />
beinhaltet somit explizit die longitudinale Dispersion. ZHANG & HEATHCOTE (2003) beschrieben die<br />
vierte Methode, die auf der analytischen Lösung zur zweidimensionalen Transportgleichung beruht<br />
und zusätzlich zur longitudinalen auch die transversale Dispersion berücksichtigt.<br />
Tab. 1 Methoden zur Bestimmung von Abbauratenkonstanten erster Ordnung λ. va ist die<br />
Transportgeschwindigkeit, ∆x ist der Abst<strong>and</strong> der verwendeten Beobachtungspegel, C(x) ist die unterstromige<br />
und C0 die Quellkonzentration. αL und αT sind die longitudinale und transversale Dispersivität, WS ist die<br />
Quellbreite und erf ist die Fehlerfunktion.<br />
Methode Formel für Abbauratenkonstante Beschreibung<br />
1 ⎟ v ⎛ ⎞<br />
a C(<br />
x)<br />
λ = −<br />
⎜<br />
1 ln<br />
∆x<br />
⎝ C0<br />
⎠<br />
2<br />
3<br />
4<br />
v ⎛ ∗ ⎞<br />
a ⎜ C(<br />
x)<br />
C0<br />
λ = −<br />
⎟<br />
2 ln<br />
∆x<br />
⎜ C ∗ ⎟<br />
⎝ 0 C(<br />
x)<br />
⎠<br />
( C(<br />
x)<br />
C )<br />
2<br />
v ⎛<br />
⎞<br />
a ⎜⎛<br />
ln<br />
0 ⎞<br />
λ<br />
− ⎟<br />
3 =<br />
⎜<br />
⎜1−<br />
2α<br />
L<br />
⎟ 1<br />
4α<br />
⎟<br />
L ⎝⎝<br />
∆x<br />
⎠ ⎠<br />
( C(<br />
x)<br />
( C β ) )<br />
2<br />
v ⎛<br />
⎞<br />
a ⎜⎛<br />
ln<br />
0 ⎞<br />
λ<br />
− ⎟<br />
3 =<br />
⎜<br />
⎜1−<br />
2α<br />
L<br />
⎟ 1<br />
4α<br />
⎟<br />
L ⎝⎝<br />
∆x<br />
⎠ ⎠<br />
mit:<br />
⎛ ⎞<br />
⎜<br />
WS<br />
β = erf ⎟<br />
⎜ ⎟<br />
⎝ 4 αT<br />
∆x<br />
⎠<br />
Analytische Lösung der 1D<br />
Advektionsgleichung mit Abbau erster<br />
Ordnung<br />
Wie Methode 1, aber Konzentration normiert<br />
auf einen nichtreaktiven Mitkontamin<strong>and</strong>,<br />
berücksichtigt Verdünnung und Dispersion<br />
Analytische Lösung der 1D Transportgleichung<br />
mit longitudinaler Dispersion<br />
Analytische Lösung der 2D Transport-<br />
gleichung. Berücksichtigt sind longitudinale<br />
und transversale Dispersion und die<br />
Quellbreite
Für die letztgenannten Methoden müssen die longitudinale und die transversale Dispersivität bekannt<br />
sein. Gemäß einem Ansatz in WIEDEMEIER et al. (1999) wurde die longitudinale Dispersivität als 10%<br />
der Fahnenlänge angenommen, die transversale Dispersivität beträgt 33% der longitudinalen<br />
Dispersivität. Als Fahnenlänge wurde der Abst<strong>and</strong> von der Quelle zur entferntesten Messstelle<br />
angenommen. Dieser Abst<strong>and</strong> beträgt 30 m, die longitudinale Dispersivität somit 3 m und die<br />
Transversale Dispersivität 1 m. Die Dispersivitäten sind damit recht gering geschätzt, eine detaillierte<br />
Untersuchung des Einflusses der angenommenen Dispersivitäten ist in BAUER et al. (2005) zu finden.<br />
Mithilfe der vier vorgestellten Ansätze werden Abbauraten erster Ordnung ermittelt und mit dem<br />
Modelleingabewert verglichen. Für jede Klasse der hydraulischen Heterogenität werden je 100<br />
Realisierungen betrachtet, um ein statistisches Maß für die Unsicherheit zu erhalten, die durch die<br />
hydraulische Heterogenität erzeugt wird. Für jede Realisierung wurde die oben beschriebene<br />
Auswertung für jede Methode und jede Quellbreite durchgeführt. Dabei wurden jeweils die Abbauraten<br />
erster Ordnung ausgehend von der Quelle für die unterstromigen Brunnen im Abst<strong>and</strong> von 10 m, 20 m<br />
und 30 m bestimmt und anschließend arithmetisch gemittelt. Für Methode 4 wird für die Ermittlung der<br />
Abbaurate die wahre Quellbreite angenommen.<br />
Ergebnisse und Diskussion<br />
Einfluss der Heterogenität<br />
Abb. 2 zeigt die Ergebnisse der Berechnung der Abbauratenkonstante erster Ordnung. Die Raten<br />
werden normalisiert dargestellt, d.h. die berechnete Ratenkonstante wird durch die wahre Ratenkonstante<br />
geteilt. Die normalisierte Rate kann so als Überschätzungs- bzw. Unterschätzungsfaktor<br />
interpretiert werden. Abb. 2 zeigt, dass die meisten berechneten Ratenkonstanten größer als 1.0 sind,<br />
d.h. dass die Ratenkonstante überschätzt wird. Dieser Effekt ist in einzelnen Realisierungen sehr<br />
stark, wo Überschätzungen der Ratenkonstante von einigen Größenordnungen auftreten können.<br />
Abb. 2a zeigt links die Variation der berechneten normalisierten Ratenkonstante mit der Quellbreite<br />
der emittierenden Schadst<strong>of</strong>fquelle. Es ist deutlich, dass mit zunehmender Quellbreite die berechnete<br />
Ratenkonstante sich der wahren Ratenkonstante annähert, d.h. sich dem Wert 1 annähert, was am<br />
deutlichsten für große Heterogenitäten sichtbar ist. Dies ist in Methode 1 begründet, die weder<br />
Verdünnung noch Dispersion oder „Aus der Fahne messen“ berücksichtigt. Diese Effekte werden mit<br />
zunehmender Quellbreite weniger signifikant, da dann die Annahmen zur Anwendung von Methode 1<br />
besser erfüllt sind. Auf der rechten Seite der Abb. 2 ist die Abhängigkeit der berechneten Ratenkonstante<br />
von der verwendeten Heterogenitätsklasse, durch die zugehörige Varianz σ²ln(K) bezeichnet,<br />
dargestellt. Es ist zu erkennen, dass eine Erhöhung von σ²ln(K) im Mittel zu einer Überschätzung der<br />
Ratenkonstante führt. Zusätzlich zu diesem Trend nimmt auch die St<strong>and</strong>ardabweichung der<br />
berechneten Ratenkonstanten zu, was die Zunahme der Spanne der ermittelten Ratenkonstanten<br />
wiederspiegelt. Diese Zunahme der Spanne kann als Zunahme der Unsicherheit der berechneten<br />
Ratenkonstante verst<strong>and</strong>en werden. Die Überschätzung beträgt im Mittel ca. 2 für geringe Heterogenität,<br />
steigt jedoch auf Werte von 4 bis 10 für mittlere und hohe Heterogenität und erreicht einen<br />
Wert von ca. 100 im Falle der sehr hohen hydraulischen Heterogenität.<br />
Abb. 2b stellt die Ergebnisse für Methode 2 dar. Auch anh<strong>and</strong> von Methode 2 werden die<br />
Ratenkonstanten überschätzt, jedoch sind, verglichen mit Methode 1, die Überschätzungen als auch<br />
die St<strong>and</strong>ardabweichung generell geringer. Methode 2 liefert also bessere und mit weniger<br />
Unsicherheit behaftete Ergebnisse. Wie man an der linken Grafik in Abb. 2b erkennen kann, ergibt<br />
sich für Methode 2 keine Abhängigkeit von der Quellbreite. Dies ist in der Methode begründet, da sie<br />
Effekte von Dispersion, Verdünnung und „Aus der Fahne messen“ durch die Normierung explizit<br />
berücksichtigt. Methode 3 zeigt ein ähnliches Verhalten wie Methode 1, sowohl in Abhängigkeit von<br />
der Quellbreite als auch in Abhängigkeit von der Heterogenität (Abb. 2c). Jedoch sind die<br />
normalisierten Abbauratenkonstanten höher als für Methode 1, was an der Berücksichtigung der<br />
Dispersion in Methode 3 liegt.<br />
Im eindimensionalen Aquifer mit einer Festkonzentration als R<strong>and</strong>bedingung führt die<br />
Berücksichtigung der longitudinalen Dispersion zu höheren Konzentrationen entlang der stationären<br />
Fahne, verglichen mit dem Fall ohne Dispersion. Dies wird durch den zusätzlichen dispersiven<br />
Massenaustrag aus der Schadst<strong>of</strong>fquelle verursacht. Um eine gemessene Konzentrationsabnahme<br />
zwischen Quelle und unterstromigen Brunnen mit Methode 3 erklären zu können ist also eine höhere<br />
Abbauratenkonstante notwendig als mit Methode 1. Methode 4 (Abb. 2d) schließlich zeigt ein sehr<br />
ähnliches Bild wie Methode 3. Aufgrund der in Methode 4 berücksichtigten Querdispersivität sind die<br />
bestimmten Abbauratenkonstanten etwas geringer als bei Methode 3. Für geringe Heterogenität wird<br />
die Ratenkonstante durch Methode 4 sogar unterschätzt. Dies liegt an einer zu hohen Korrektur durch<br />
6
den Querdispersionsterm, der die Effekte der Querdispersion für geringe Heterogenität überschätzt.<br />
Für den Fall geringer oder keiner Heterogenität sind die hier gewählten und in Methode 4<br />
verwendeten Dispersivitäten zu groß.<br />
a)<br />
b)<br />
c)<br />
d)<br />
normalisierte Ratenkonstante [-]<br />
normalisierte Ratenkonstante [-]<br />
normalisierte Ratenkonstante [-]<br />
normalisierte Ratenkonstante [-]<br />
1000<br />
100<br />
10<br />
1<br />
0.1<br />
0.01<br />
0 4 8 12 16 20<br />
1000<br />
100<br />
10<br />
1<br />
0.1<br />
0.01<br />
0 4 8 12 16 20<br />
1000<br />
100<br />
10<br />
1<br />
0.1<br />
0.01<br />
0 4 8 12 16 20<br />
1000<br />
100<br />
10<br />
1<br />
0.1<br />
0.01<br />
0 4 8 12 16 20<br />
Quellbreite [m]<br />
7<br />
0 1 2 3 4 5<br />
0 1 2 3 4 5<br />
0 1 2 3 4 5<br />
0 1 2 3 4 5<br />
σ 2 ln(K)<br />
Abb. 2 “Gemessene” Abbauratenkonstanten erster Ordnung, normalisiert auf die wahre<br />
Abbauratenkonstante, aufgetragen gegen Quellbreite (links) und Heterogenität (rechts). a) Methode 1,<br />
b) Methode 2, c) Methode 3 und d) Methode 4. Alle Abbildungen zeigen die Einzelergebnisse aller<br />
Realisierungen (kleine Symbole), ihren Mittelwert (große Symbole) und die zugehörige<br />
St<strong>and</strong>ardabweichung (Fehlerbalken). Unterschiedliche Grautöne geben die Quellbreite,<br />
unterschiedliche Symbole die verwendeten σ²ln(K) an.
Um diesen Effekt zu verdeutlichen sind in Abb. 2 für alle Methoden noch die für den homogenen Fall<br />
bestimmten Ratenkonstanten eingetragen. Die Ergebnisse sind für σ²ln(K) = 0 in den Abbildungen<br />
rechter H<strong>and</strong> als kleine horizontale Balken dargestellt. Für Methode 1 und Methode 2 beträgt die<br />
normalisierte Ratenkonstante im homogenen Fall genau 1.0, d.h. diese Methoden liefern exakt die<br />
wahre Ratenkonstante. Mit Methode 3 ergibt sich eine geringe Überschätzung, mit Methode 4 eine<br />
deutliche Unterschätzung der wahren Ratenkonstante. Die geringste Ratenkonstante wird für die<br />
geringste Quellbreite ermittelt, da hier die Korrektur durch β am größten ist (vergl. Tab 1).<br />
Da die bisher gezeigten Mittelwerte und St<strong>and</strong>ardabweichungen repräsentativ für das Ensemblemittel<br />
sind, nicht jedoch für die einzelnen Realisierungen, wurde ein weiteres Maß zum Vergleich der<br />
anh<strong>and</strong> der vier Methoden bestimmten Ratenkonstanten entwickelt. Abb. 3 zeigt die Wahrscheinlichkeit,<br />
mit der eine Methode zum Erfolg führen kann, worunter hier verst<strong>and</strong>en wird, dass die Abbaurate<br />
anh<strong>and</strong> der Methode mit einer gewünschten Genauigkeit ermittelt wird. Die gewünschte Genauigkeit<br />
wird als sogenannten Fehlerfaktor angegeben. Ein Fehlerfaktor von 10 entspricht dem Intervall 0.1 bis<br />
10 der normierten Ratenkonstanten, wobei dieses Intervall durch Division und Multiplikation von 1.0<br />
mit dem Fehlerfaktor (10) ermittelt wird. Dies entspricht somit der umgangssprachlichen Formulierung<br />
„... innerhalb einer Größenordnung ...“. Ein Fehlerfaktor von 5 entspricht somit dem Intervall 0.2 bis 5<br />
der normierten Ratenkonstanten. Abb. 3 gibt daher die Wahrscheinlichkeit dafür an, dass eine<br />
„gemessene“ Ratenkonstante innerhalb des durch den Fehlerfaktor aufgespannten Intervalls liegt.<br />
Abb. 3a zeigt für die geringste Heterogenität, dass die Wahrscheinlichkeit, die Abbauratenkonstante<br />
mit einem Fehlerfaktor kleiner als 2.0 zu bestimmen („ ... bis auf einen Faktor zwei ...“), für Methode 1<br />
ca. 70 %, für Methode 2 ca. 90%, für Methode 3 ca. 55% und für Methode 4 ca. 30% beträgt. Wird der<br />
Fehlerfaktor auf 5 erhöht, dann erhält man mit Methoden 1 bis 3 eine Erfolgswahrscheinlichkeit von<br />
100%, nur mit Methode 4 beträgt die Erfolgswahrscheinlichkeit ca. 70%. Je geringer die Werte einer<br />
Methode in Abb. 3 sind, desto geringer ist die Wahrscheinlichkeit, die Abbauratenkonstante mit der<br />
gewünschten Genauigkeit bestimmen zu können. Wie Abb. 3 zeigt, sinkt für alle Methoden die Erfolgswahrscheinlichkeit<br />
mit zunehmender Heterogenität. Beträgt die Erfolgswahrscheinlichkeit von Methode<br />
1 für einen Fehlerfaktor von 5 noch ca. 100% für die geringste Heterogenität, so sinkt diese<br />
Wahrscheinlichkeit auf 70%, 50% und schließlich 35% für die höheren Heterogenitätsklassen. Für<br />
Methoden 2 bis 4 sind die Werte analog Abbildung 3 zu entnehmen.<br />
Abb. 3 erlaubt daher einen direkten Vergleich der in dieser Arbeit untersuchten vier Methoden. Wird<br />
die Ratenkonstante in einem stark heterogenen Aquifer beispielsweise auf einen Faktor zehn genau<br />
benötigt, betragen die Erfolgswahrscheinlichkeiten ca. 80%, 95%, 80% und 60% für Methoden 1 bis 4<br />
(Abb. 3c). Für alle Heterogenitätsklassen liefert Methode 2 die höchste Wahrscheinlichkeit, das<br />
richtige Ergebnis zu bekommen. Für mittlere bis hohe Heterogenitäten folgt Methode 4 an zweiter<br />
Stelle, während diese Methode für geringe Heterogenität die schlechtesten Erfolgswahrscheinlichkeiten<br />
aufweist. Obwohl Methode 1 die einfachste Methode ist, liefert sie sehr ähnliche Ergebnisse wie<br />
Methode 4. Für geringe Heterogenitäten ergeben sich mit Methode 1 sogar die besseren<br />
Abschätzungen der Ratenkonstante. Methode 3 zeigt – bis auf geringe Heterogenität – immer die<br />
geringste Erfolgswahrscheinlichkeit.<br />
Für größere Quellbreiten ergeben sich qualitativ dieselben Ergebnisse. Methode 2 ist unabhängig von<br />
der Quellbreite, daher ist auch für Quellbreiten von 8 m oder 16 m die Erfolgswahrscheinlichkeit von<br />
Methode 2 dieselbe wie im Falle von 4 m. Für die <strong>and</strong>eren Methoden ergeben sich mit zunehmender<br />
Quellbreite höhere Erfolgswahrscheinlichkeiten. So steigt z.B. die Erfolgswahrscheinlichkeit von<br />
Methode 3 für σ²ln(K) = 2.7 von 35% für eine Quellbreite von 4 m (Abb. 3c) auf 40% für eine<br />
Quellbreite von 8 m und ca. 50% für eine Quellbreite von 16 m. Somit steigt die Erfolgswahrscheinlichkeit<br />
für Methoden 1, 3 und 4 mit der Quellbreite an, erreichen aber dennoch nicht die<br />
Erfolgswahrscheinlichkeit von Methode 2.<br />
8
1 a) b) 1<br />
c)<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
Methode 1<br />
Methode 2<br />
Methode 3<br />
Methode 4<br />
0<br />
1 10 100 1000<br />
1<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0<br />
Methode 1<br />
Methode 2<br />
Methode 3<br />
Methode 4<br />
1 10 100 1000<br />
Fehlerfaktor<br />
9<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
d)<br />
Methode 1<br />
Methode 2<br />
Methode 3<br />
Methode 4<br />
0<br />
1 10 100 1000<br />
1<br />
0.8<br />
0.6<br />
0.4<br />
0.2<br />
0<br />
Methode 1<br />
Methode 2<br />
Methode 3<br />
Methode 4<br />
1 10 100 1000<br />
Fehlerfaktor<br />
Abb. 3 Erfolgswahrscheinlichkeit für alle vier Methoden gegen den Fehlerfaktor für σ²ln(K) a) 0.38, b)<br />
1.71, c) 2.7 <strong>and</strong> d) 4.5. Die Quellbreite beträgt 4 m.<br />
Einfluss des Messfehlers<br />
Bei der Untersuchung des Einflusses des Messfehlers wird von einem homogenen Aquifer ausgegangen.<br />
Die Unsicherheit wird nun nicht durch das heterogene Fließfeld erzeugt, sondern durch<br />
falsches „Messen“ von Piezometerhöhe und Schadst<strong>of</strong>fkonzentration.<br />
Da in der vorigen Untersuchungen gezeigt wurde, dass Methode 4 mit der gewählten transversalen<br />
Dispersivität im homogenen Fall zu kleine Abbauratenkonstanten liefert, wurde die transversale<br />
Dispersivität, die für die Auswertung mit Methode 4 (Tabelle 1) verwendet wird, auf 0.15 m verringert.<br />
Damit ergibt sich im homogenen Fall die richtige Abbauratenkonstante.<br />
In Abb. 4 ist die auf die wahre Ratenkonstante normierte fehlerbehaftete Ratenkonstante gegen den<br />
maximal möglichen Fehler bei der Bestimmung der Piezometerhöhe, ∆hmax, gesondert für die vier<br />
Auswertemethoden aufgetragen. In Abb. 4a erkennt man ein deutliches Ansteigen der normierten<br />
Ratenkonstanten mit zunehmendem ∆hmax für die Auswertung mit Methode 1. Bereits für einen<br />
maximalen Messfehler von 1 cm ergibt sich eine mittlere Überschätzung von ca. 2, der für ein ∆hmax<br />
von 5 cm auf ca. 5 steigt. Außerdem ist zu erkennen, dass für einzelne Messungen Überschätzungen<br />
der Ratenkonstante von mehr als 10 möglich sind. Generell liegt eine Überschätzung vor, da eine<br />
fehlerhaft bestimmte Piezometerhöhe zu einer falschen Abschätzung der Fließrichtung führt. Die neu<br />
platzierten unterstromigen Brunnen liegen dann nicht mehr auf der Zentrallinie, d.h. die gemessenen<br />
Konzentrationen sind kleiner als auf der Zentrallinie und dies führt zu einer Überschätzung der<br />
Ratenkonstanten. Es wird jedoch nicht nur die Fließrichtung falsch ermittelt, sondern auch die<br />
Fließgeschwindigkeit entlang der (angenommenen) Zentrallinie, da diese aus den gemessenen<br />
Piezometerhöhen, Brunnenabständen und hydraulischen Durchlässigkeiten berechnet wird. Der<br />
hieraus resultierende Fehler kann sowohl zu große als auch zu kleine Ratenkonstanten produzieren.<br />
Für Methode 2 (Abb. 4b) ergibt sich ein <strong>and</strong>eres Bild. Die mittlere Ratenkonstante ist auch für große<br />
∆hmax sehr nahe bei 1, d.h. es liegt kein Trend zu Über- oder Unterschätzung vor. Die Unsicherheit der<br />
bestimmten Abbaurate ist kleiner als bei Methode 1, wie an den Fehlerbalken zu erkennen ist. Da<br />
Methode 2 das „aus der Fahne messen“ korrigiert, lässt sich dieser Fehler für Methode 2 nicht<br />
beobachten (Abb. 4b). Es kommt sowohl zu einer Über- als auch Unterschätzung der Ratenkonstante,<br />
also einem eher symmetrischen Fehler ohne generelle Tendenz, der auf die falsche Abschätzung der<br />
Fließgeschwindigkeit zurückzuführen ist. Mit zunehmendem ∆hmax nimmt auch für Methode 2 die
Unsicherheit zu. Für Methode 3 ( Abb. 4c) und Methode 4 (Abb. 4d) ergibt sich – wie schon für die<br />
hydraulische Heterogenität – ein sehr ähnliches Bild wie für Methode 1, jedoch mit höheren<br />
Überschätzungen und Unsicherheiten als für Methode 1. Für Methode 3 und 4 beträgt die mittlere<br />
Überschätzung der Ratenkonstanten für ∆hmax = 1 cm bereits ca. 5, jeweils verbunden mit einer hohen<br />
Unsicherheit.<br />
normalisierte Abbauratenkonstante [-]<br />
normalisierte Abbauratenkonstante [-]<br />
100<br />
10<br />
1<br />
0.1<br />
100<br />
10<br />
1<br />
0.1<br />
0 1 2 3 4 5 6<br />
10<br />
100<br />
a) b)<br />
0 1 2 3 4 5 6<br />
maximaler Messfehler Piezometerhöhe [cm]<br />
10<br />
1<br />
0.1<br />
100<br />
c) d)<br />
10<br />
1<br />
0.1<br />
0 1 2 3 4 5 6<br />
0 1 2 3 4 5 6<br />
maximaler Messfehler Piezometerhöhe [cm]<br />
Abb. 4 Normalisierte Abbauratenkonstante, aufgetragen gegen den maximalen Messfehler bei der<br />
Bestimmung der Piezometerhöhe für a) Methode 1, b) Methode 2, c) Methode 3 und d) Methode 4.<br />
Dargestellt sind die Einzelergebnisse (kleine Symbole), die Mittelwerte (durch Linie verbundene<br />
Punkte) und die zugehörige St<strong>and</strong>ardabweichung (Fehlerbalken).<br />
Für Konzentrationsmessungen wurde ein höherer Messfehlerfaktor von 100 und eine exakte Messung<br />
der Piezometerhöhe angenommen (siehe Kapitel Methodik). Ein Messfehlerfaktor von 100 ist<br />
sicherlich sehr hoch gegriffen und wird hier nur zu Demonstrationszwecken eingesetzt. Das Ergebnis<br />
für Methode 1 zeigt Abb. 5a, in der die normierte Ratenkonstanten gegen den Fehlerfaktor<br />
aufgetragen sind. Man erkennt, dass für geringe Messfehlerfaktoren die Abbaukonstante sehr gut<br />
bestimmt werden kann, für einen Messfehlerfaktor von 2 erhält man im Mittel noch das richtige<br />
Ergebnis, für einen Messfehlerfaktor von 5 erhält man eine mittlere Überschätzung der Ratenkonstante<br />
von ca. 3. Die maximale mittlere Überschätzung der Ratenkonstante von ca. 5 wird bei einem<br />
Messfehlerfaktor von ca. 10 erreicht und steigt auch für größere Fehlerfaktoren nicht weiter an.<br />
Allerdings nimmt die Unsicherheit, dargestellt als St<strong>and</strong>ardabweichung, mit dem Messfehlerfaktor<br />
stark zu und man kann sowohl zu große als auch zu kleine Ratenkonstanten erhalten. Für Methode 2<br />
(Abb. 5b) erhält man generell eine geringere Überschätzung der Ratenkonstante von maximal ca. drei,<br />
jedoch eine ähnlich hohe Unsicherheit wie mit Methode 1. Generell werden bei Methode 2 eher<br />
kleinere Ratenkonstanten erzeugt als bei Methode 1. Methoden 3 und 4 (Abb 5.c und Abb. 5d) zeigen<br />
einen deutlichen Anstieg der ermittelten Überschätzung mit dem Messfehlerfaktor, maximale Werte<br />
liegen hier bei etwa 10. Ebenso ist die Unsicherheit dieser Werte sehr hoch.
normalisierte Abbauratenkonstante [-]<br />
normalisierte Abbauratenkonstante [-]<br />
1000<br />
100<br />
10<br />
1<br />
0.1<br />
0.01<br />
1000<br />
100<br />
10<br />
1<br />
0.1<br />
0.01<br />
a)<br />
1 10 100<br />
c)<br />
1 10 100<br />
maximaler Messfehlerfaktor Konzentration [-]<br />
11<br />
1000<br />
100<br />
10<br />
1<br />
0.1<br />
0.01<br />
1000<br />
100<br />
10<br />
1<br />
0.1<br />
0.01<br />
b)<br />
1 10 100<br />
d)<br />
1 10 100<br />
maximaler Messfehlerfaktor Konzentration [-]<br />
Abb. 5 Normalisierte Abbauratenkonstante, aufgetragen gegen den maximalen Messfehlerfaktor bei<br />
der Bestimmung der Konzentration für a) Methode 1, b) Methode 2, c) Methode 3 und d) Methode 4.<br />
Dargestellt sind die Einzelergebnisse (kleine Symbole), die Mittelwerte (durch Linie verbundene<br />
Punkte) und die zugehörige St<strong>and</strong>ardabweichung (Fehlerbalken).<br />
Schlussfolgerungen<br />
Die vorgestellten Ergebnisse zeigen, dass sich die anh<strong>and</strong> der vier untersuchten Methoden ermittelten<br />
Abbauratenkonstanten erster Ordnung sowohl für zunehmende hydraulische Heterogenität als auch<br />
für zunehmende Messfehler bezüglich der Piezometerhöhe und der Konzentration unterscheiden. Alle<br />
Methoden zeigen geringere Erfolgswahrscheinlichkeiten mit zunehmender Heterogenität (Abb.3)<br />
sowie eine erhöhte Unsicherheit (Abb. 2). Aus Abb. 2 ist ersichtlich, dass diese Abnahme der<br />
Erfolgswahrscheinlichkeit durch eine generelle Überschätzung der Ratenkonstante verursacht wird.<br />
Diese Überschätzung ist am größten für Methode 3. Methode 2 wird dagegen am wenigsten von der<br />
hydraulischen Heterogenität beeinflusst und zeigt die geringsten Überschätzungen (Abb. 2) und<br />
Unsicherheiten. Obwohl Methoden 3 und 4 realitätsnäher sind, da sie auf der eindimensionalen bzw.<br />
zweidimensionalen Transportgleichung beruhen, zeigen sie eine geringere Erfolgswahrscheinlichkeit<br />
und eine größere Überschätzung der Ratenkonstanten als Methoden 1 und 2. Beide Methoden sind<br />
anfällig für Fehler, die durch die Abschätzung der longitudinalen und transversalen Dispersivitäten<br />
erzeugt werden können. Methode 1 als die einfachste Methode, da sie weder die Abschätzung der<br />
Dispersivität noch einen nichtreaktiven Mitkontamin<strong>and</strong>en benötigt, zeigt bessere oder ähnlich gute<br />
Ergebnisse wie Methode 4. Sowohl mittlere Überschätzung als auch Unsicherheit steigen mit<br />
zunehmender Größe des Messfehlers. Die Untersuchung des Messfehlers der Piezometerhöhe zeigt,<br />
dass bereits geringe Messfehler von 1 cm zu deutlicher Überschätzung der Ratenkonstanten und<br />
einer erhöhten Unsicherheit führen können. Der Einfluss des Messfehlers der Konzentration auf die<br />
Ratenkonstante ist generell geringer. Auch bei der Untersuchung der Messfehler erhält man, wie<br />
schon im Falle von hydraulischer Heterogenität, die besten Ergebnisse mit Methode 2. Methoden 3<br />
und 4 zeigen sehr ähnliche Ergebnisse und weisen größere Überschätzungen auf als Methode 2 oder<br />
Methode 1.<br />
Insgesamt ergibt sich, dass die Verwendung von Methode 2, die auf der Normierung der<br />
Schadst<strong>of</strong>fkonzentration mit einem nichtreaktiven Mitkontamin<strong>and</strong>en beruht, zu den besten<br />
Ergebnissen bei der Bestimmung von Abbauratenkonstanten erster Ordnung führt. Daher sollte diese<br />
Methode, wenn möglich, angewendet werden. Ist kein nichtreaktiver Mitkontamin<strong>and</strong> vorh<strong>and</strong>en, sollte<br />
Methode 1 verwendet werden, da sie die Abbauratenkonstante besser als Methoden 3 und 4 vorher-
sagt sowie die zusätzliche Problematik des Abschätzens der Dispersivitäten umgeht. In BAUER et al.<br />
(2005) wird der Einfluss der geschätzten Dispersivitäten auf die Abbauratenkonstante sowohl für die<br />
longitudinale als auch transversale Dispersivität anh<strong>and</strong> einer Sensitivitätsstudie eingehend untersucht.<br />
Es zeigt sich dabei, dass mit Methode 3 nie, mit Methode 4 nur unter Annahme sehr hoher<br />
transversaler Dispersivitäten, die nicht mit der Heterogenität (σ²ln(K)) des Aquifers begründet werden<br />
können, die korrekten Abbauraten ermittelt werden. In diesem Aufsatz werden weiterhin der Einfluss<br />
der Quellbreite sowie verschiedene Methoden zur Berechnung der Fliessgeschwindigkeit untersucht.<br />
Es zeigt sich, dass die Anordnung der Messpegel, abgeleitet aus dem hydraulischen Dreieck, großen<br />
Einfluss auf die geschätzte Abbaurate hat, wie er sich auch in Abbildung 1 widerspiegelt. Ebenso ist<br />
die Mittelung der an den einzelnen Messpegeln ermittelten hydraulischen Durchlässigkeiten von<br />
Bedeutung. Hier ergibt sich, dass eine geometrische Mittelung zu bevorzugen ist (BAUER et al., 2006).<br />
Die gefundene generelle Überschätzung der Ratenkonstante ist im Hinblick auf eine Prognose der<br />
Fahnenlänge problematisch, da eine zu hohe Ratenkonstante den Abbau überschätzt und daher zu<br />
geringe prognostizierte Fahnenlängen verursacht. Dieser Fehler ist daher nicht konservativ und kann<br />
zu einer zu optimistischen Einschätzung der St<strong>and</strong>ortverhältnisse führen. Im Folgenden ist die<br />
Fahnenlänge definiert als die Strecke zwischen Schadensquelle und dem Ort auf der<br />
Fahnenzentralachse, an dem das Verhältnis C/C0 einen vorgegebenen Wert erreicht hat<br />
(beispielsweise 10 -3 ), wobei C die Konzentration in der Fahne und C0 die Konzentration in der Quelle<br />
ist. Für eindimensionale Lösungen ohne Dispersion ergibt sich, dass Abbaurate und Fahnenlänge<br />
umgekehrt proportional zuein<strong>and</strong>er sind und die Fahnenlänge linear mit dem Kehrwert der Abbaurate<br />
wächst (Vergleiche Methoden 1 und 2 in Tabelle 1). Eine Überschätzung der Abbaurate um den<br />
Faktor 5 verursacht also eine Unterschätzung der Fahnenlänge ebenfalls um den Faktor 5. Im<br />
eindimensionalen Fall mit Dispersion sowie im zweidimensionalen Fall ist die Fahnenlänge annähernd<br />
umgekehrt proportional zum Kehrwert der Quadratwurzel der Abbaurate, die genauen Werte hängen<br />
von den gegebenen Dispersivitäten, der Quellbreite und dem Verhältnis von Abst<strong>and</strong>sgeschwindigkeit<br />
zu Abbaurate ab (vergleiche Methoden 3 und 4 in Tabelle 1). Eine Überschätzung der Fahnenlänge<br />
um den Faktor 5 bzw. 10 kann beispielsweise eine Unterschätzung der Fahnenlänge um den Faktor 3<br />
bzw. 5 verursachen. Der Einfluss der Dispersion verringert die Sensitivität der Fahnenlänge auf die<br />
Abbaurate. Eine genauere Untersuchung des Einflusses der ermittelten Abbaurate auf die<br />
prognostizierte Fahnenlänge wurde von BEYER et al., (2005b) durchgeführt. Dabei zeigte sich eine<br />
durchschnittliche Unterschätzung der Fahnenlängen um 50 %. Für einzelne Realisierungen ergaben<br />
sich zum Teil jedoch erheblich größere Fehler, sodass in ungünstigen Fällen die geschätzte<br />
Fahnenlänge lediglich 10% des tatsächlichen Wertes betrug. Dabei zeigt sich, dass die mit Methoden<br />
1, 3 und 4 berechneten Abbauraten prinzipiell zu ähnlichen Abschätzungen der Fahnenlänge führen,<br />
wenn diese mit einem mit der Abbaurate korrespondierenden Transportmodell berechnet wird. Die<br />
durch Vernachlässigung bzw. falsche Parametrisierung der Längs- und Querdispersion verursachten<br />
Fehler in der Abbaurate werden bei der Fahnenlängenprognose teilweise wieder aufgehoben. Als<br />
problematisch stellte es sich jedoch heraus, Abbauraten, die mit der eindimensionalen Lösung<br />
berechnet wurden, in einem zwei- oder dreidimensionalen Transportmodell zur Prognose der<br />
Fahnenlänge zu verwenden (BEYER et al., 2005b).<br />
Danksagung: Der Beitrag entst<strong>and</strong> im Rahmen des Projekts TV 7.1 „Modellierung und<br />
Prognose“ des BMBF- Förderschwerpunktes “Kontrollierter natürlicher Rückhalt und Abbau<br />
von Schadst<strong>of</strong>fen bei der Sanierung kontaminierter Grundwässer und Böden (KORA)“ an den<br />
Universitäten Kiel und Tübingen. Den Projektträgern und dem BMBF sei für die hierbei<br />
gewährte Unterstützung gedankt.<br />
Literaturangaben<br />
Bauer, S., Beyer, C. , Kolditz, O. (2006): Assessing measurement uncertainty <strong>of</strong> first order degradation<br />
rates in heterogeneous aquifers. - Water Resour. Res., im Druck.<br />
Bauer, S., Beyer, C., Kolditz, O. (2005): Assessing measurements <strong>of</strong> first order degradation rates by<br />
using the Virtual Aquifer approach. - IAHS Publication, 297: 274-281.<br />
Bauer, S., Kolditz, O. (2005): Uncertainty assessment <strong>of</strong> integral pumping tests in heterogeneous<br />
aquifers. - Angenommener Beitrag für eine IAHS Publikation.<br />
Beyer, C., Bauer S., Kolditz, O. (2005a): Uncertainty assessment <strong>of</strong> degradation rate measurements in<br />
heterogeneous media using the virtual aquifer approach. - In: Kolditz, O., Bauer, S., <strong>and</strong><br />
Gronewold, J. (Eds.): Proceedings <strong>of</strong> the 5th Workshop “Porous Media”, Blaubeuren, Germany,<br />
December 2004,. ZAG Publisher, Tübingen.<br />
12
Beyer, C., Bauer S., Kolditz, O. (2005b): Using the Virtual Aquifer approach for uncertainty<br />
assessment <strong>of</strong> contaminant plume length estimates. GeoSys Preprint, GeoSystemForschung,<br />
Tübingen. http://www.uni-tuebingen.de/zag/geohydrology/index.html<br />
Bockelmann, A., Zamfirescu, D., Ptak, T., Grathwohl, P., Teutsch, G. (2003): Quantification <strong>of</strong> mass<br />
fluxes <strong>and</strong> natural attenuation rates at an industrial site with a limited monitoring network: A case<br />
study. - J. Contam. Hydrol., 60: 97-121.<br />
Buscheck, T.E., Alcantar, C.M. (1995): Regression techniques <strong>and</strong> analytical solutions to demonstrate<br />
intrinsic bioremediation. - In: Proceedings <strong>of</strong> the 1995 Battelle International Conference on In-Situ<br />
<strong>and</strong> On Site Bioreclamation, Batelle, CA, April 1995. 109-116.<br />
Chapelle, F.H., Bradley, P.M., Lovley, D.R., Vroblesky, D.A. (1996): Measuring rates <strong>of</strong> biodegradation<br />
in a contaminated aquifer using field <strong>and</strong> laboratory methods. -Ground Water, 34(4): 691-698.<br />
Herfort, M. (2000): Reactive Transport <strong>of</strong> Organic Compounds Within a Heterogeneous Porous<br />
Aquifer. - Tübinger Geowissenschaftliche Arbeiten, 54, Universität Tübingen, Tübingen.<br />
Kolditz, O., de Jonge, J., Beinhorn, M., Xie, M., Kalbacher, M., Wang, W., Bauer, S., McDermott, C.,<br />
Chen, C., Beyer, C., Gronewold, J., Kemmler, D., Manabe, T., Legeida D., Adamidis, P. (2005):<br />
GeoSys - Theory <strong>and</strong> users manual, release 4.2. GeoHydrology / HydroInformatics, Center for<br />
<strong>Applied</strong> Geoscience, Universität Tübingen, Tübingen.<br />
McNab Jr, W. W., Dooher, B.P. (1998): A critique <strong>of</strong> a steady-state analytical method for estimating<br />
contaminant degradation rates. - Ground Water 36(6): 983–987.<br />
Pebesma, E. J., Wesseling, C. G. (1998): Gstat: A program for geostatistical <strong>modeling</strong>, prediction <strong>and</strong><br />
simulation. - Computers & Geosciences, 24(1): 17-31, doi:10.1016/S0098- 3004(97)00082-4.<br />
Rehfeldt, K.R., Boggs, J.M., Gelhar, L.W. (1992): Field study <strong>of</strong> dispersion in a heterogeneous aquifer,<br />
3, geostatistical analysis <strong>of</strong> hydraulic conductivity. - Water Resour. Res., 28(12): 3309–3324.<br />
Rubin,Y. (2003): <strong>Applied</strong> Stochastic Hydrogeology. - 416 S.; Oxford University Press, New York, NY.<br />
Schäfer, D., Dahmke, A., Kolditz, O., Teutsch, G. (2002): "Virtual Aquifers": A concept for evaluation <strong>of</strong><br />
exploration, remediation <strong>and</strong> monitoring strategies. - In: Calibration <strong>and</strong> Reliability in Groundwater<br />
Modelling: A Few Steps Closer to Reality (Proceedings <strong>of</strong> the ModelCARE 2002 Conference held<br />
in Prague, Czech Republic, June 2002), edited by K. Kovar & Z. Hrkal, IAHS Publication, 277: 52-<br />
59.<br />
Schäfer, D., Schlenz B., Dahmke, A. (2004): Evaluation <strong>of</strong> exploration <strong>and</strong> monitoring methods for<br />
verification <strong>of</strong> natural attenuation using the virtual aquifer approach. - Biodegradation Journal<br />
15(6): 453-465.<br />
Schäfer, D., Hornbruch, G., Schlenz B., Dahmke, A. (2005): Untersuchung kinetischer Ansätze zur<br />
Modellierung mikrobieller Abbauprozesse mit Hilfe des Virtuelle Aquifere Ansatzes. - Eingereicht<br />
bei Grundwasser.<br />
Suarez, M. P., Rifai, H.S. (2002): Evaluation <strong>of</strong> BTEX remediation by natural attenuation at a coastal<br />
facility. - Ground Water Monit. Remed. 22(1): 62-77.<br />
Stenback, G.A., Ong, S.K., Rogers, S.W., Kjartanson, B.H. (2004): Impact <strong>of</strong> transverse <strong>and</strong> longitudinal<br />
dispersion on first-order degradation rate constant estimation. - J. Cont. Hydrol. 73: 3-14.<br />
Sudicky, E. A. (1986): A natural gradient experiment on solute transport in a s<strong>and</strong> aquifer: Spatial<br />
variability <strong>of</strong> hydraulic conductivity <strong>and</strong> its role in the dispersion process. - Water Resour. Res.<br />
22(13): 2069-2082.<br />
U.S. Environmental Protection Agency (1999): Use <strong>of</strong> monitoring natural attenuation at Superfund,<br />
RCRA Corrective Action, <strong>and</strong> Underground Storage Tank Sites, - Office <strong>of</strong> Solid Waste <strong>and</strong><br />
Emergency Response Directive 9200.4-17, Washington, D.C.<br />
Wiedemeier, T.H., Swanson, M.A., Wilson, J.T., Kampbell, D.H., Miller, R.N., Hansen,J.E. (1996):<br />
Approximation <strong>of</strong> biodegradation rate constants for monoaromatic hydrocarbons (BTEX) in ground<br />
water. - Ground Water Monit. Remed., 16(3): 186-194.<br />
Wiedemeier, T.H., Rifai, H.S., Wilson, J.T., Newell, C. (1999): Natural Attenuation <strong>of</strong> Fuels <strong>and</strong><br />
Chlorinated Solvents in the Subsurface. - 617 S.; Wiley, New York, NY.<br />
Wilson, J. T., Pfeffer, F. M., Weaver, J. W., Kampbell, D. H., Wiedemeier, T. H., Hansen, J. E., Miller,<br />
R. N. (1994): Intrinsic Bioremediation <strong>of</strong> JP-4 Jet Fuel. - In: Symposium on Intrinsic<br />
Bioremediation <strong>of</strong> Ground Water, Denver, Colorado, US-EPA: 60-72.<br />
Wilson, R. D., Thornton, S.F. Mackay, D.M. (2004): Challenges in monitoring the natural attenuation <strong>of</strong><br />
spatially variable plumes. - Biodegradation Journal 15(6): 459-469.<br />
Zamfirescu, D., Grathwohl, P.(2001): Occurrence <strong>and</strong> attenuation <strong>of</strong> specific organic compounds in the<br />
groundwater plume at a former gasworks site. - J. Contam. Hydrol. 53: 407-427.<br />
Zhang, Y.-K. , Heathcote, R.C. (2003): An improved method for estimation <strong>of</strong> biodegradation rate with<br />
field data. - Ground Water Monit. Remed., 23(3): 112-116.<br />
13
Enclosed Publication 5<br />
Beyer, C., Chen, C., Gronewold, J., Kolditz, O., Bauer, S. (2007a): Determination <strong>of</strong> first<br />
order degradation rate constants from monitoring networks. (accepted by Ground Water.).<br />
The enclosed article was accepted for publication in the journal Ground Water. Copyright ©<br />
2007 Blackwell Publishing.<br />
The definite version will be available at www.blackwell-synergy.com.
Determination <strong>of</strong> first order degradation rate constants from<br />
monitoring networks ∗<br />
Christ<strong>of</strong> Beyer, Cui Chen, Jan Gronewold, Olaf Kolditz, Sebastian Bauer<br />
Center for <strong>Applied</strong> Geoscience, Eberhard-Karls-University <strong>of</strong> Tübingen,<br />
Sigwartstraße 10, D 72076 Tübingen, Germany<br />
Tel.: +49 7071 29 73176; Fax: +49 7071 5059. E-mail: christ<strong>of</strong>.beyer@uni-tuebingen.de<br />
Abstract<br />
In this paper different strategies for estimating first order degradation rate constants from measured field data are<br />
compared by application to multiple synthetic contaminant plumes. The plumes were generated by <strong>numerical</strong> simulation<br />
<strong>of</strong> contaminant transport <strong>and</strong> degradation in virtual heterogeneous aquifers. These sites then were individually <strong>and</strong><br />
independently investigated on the computer by installation <strong>of</strong> extensive networks <strong>of</strong> observation wells. From the data<br />
measured at the wells, i.e. contaminant concentrations, hydraulic conductivities <strong>and</strong> heads, first order degradation rates<br />
were estimated by three one-dimensional center line methods, which use only measurements located on the plume axis,<br />
<strong>and</strong> a two-dimensional method, which uses all concentration measurements available downgradient from the<br />
contaminant source. Results for both strategies show that the true rate constant used for the <strong>numerical</strong> simulation <strong>of</strong> the<br />
plumes in general tends to be overestimated. Overestimation is stronger for narrow plumes from small source zones with<br />
an average overestimation factor <strong>of</strong> about 5 <strong>and</strong> single values ranging from 0.5 to 20, decreasing for wider plumes, with<br />
an average overestimation factor <strong>of</strong> about 2 <strong>and</strong> similar spread. Reasons for this overestimation are identified in the<br />
velocity calculation, the dispersivity parameterization <strong>and</strong> <strong>of</strong>f-center-line measurements. For narrow plumes the onedimensional<br />
<strong>and</strong> the two-dimensional strategies show approximately the same amount <strong>of</strong> overestimation. For wide<br />
plumes, however, incorporation <strong>of</strong> all measurements in the two-dimensional approach reduces the estimation error. No<br />
significant relation between the number <strong>of</strong> observation wells in the monitoring network <strong>and</strong> the quality <strong>of</strong> the estimated<br />
rate constant is found for the two-dimensional approach.<br />
Introduction<br />
Although detailed mathematical descriptions <strong>of</strong> contaminant degradation in the subsurface are available (e.g. Rittmann<br />
<strong>and</strong> VanBriesen 1996; Wiedemeier et al. 1999; Islam et al. 2001), in field studies <strong>of</strong>ten simplified approaches are used for<br />
contaminant transport <strong>modeling</strong>. This is mainly because the identification <strong>of</strong> a large number <strong>of</strong> parameters <strong>and</strong> processes<br />
from field data <strong>of</strong>ten is impossible. Due to its mathematical simplicity, its easy implementation into transport models <strong>and</strong><br />
the necessity <strong>of</strong> determining only a single parameter, the biodegradation model most frequently used is first order kinetics<br />
(Wiedemeier et al. 1999). Field methods for the determination <strong>of</strong> biodegradation rates in ground water comprise a variety<br />
<strong>of</strong> approaches, including mass balance calculations, in-situ microcosm studies <strong>and</strong> the so called center line method<br />
(Chapelle et al. 1996). With the center line method, only concentration measurements in observation wells located on the<br />
plume axis are evaluated <strong>and</strong> additional information possibly at h<strong>and</strong> (e.g. well data downgradient from the source but<br />
not on the center line) is not explicitly accounted for in the rate estimation. Thus, the center line method is an essentially<br />
one-dimensional approach to rate constant estimation, as <strong>flow</strong>, transport <strong>and</strong> degradation are only evaluated for a single<br />
streamline. Assuming that contaminant biodegradation can be approximated by a first order kinetics, the degradation rate<br />
constant can be calculated with analytical transport models to yield the sought for first order rate constant. An overview<br />
<strong>of</strong> common methods for estimating degradation rate constants from field data within the context <strong>of</strong> monitored natural<br />
attenuation is given by Newel et al. (2002). The authors discuss approaches based on (point-) concentration over time<br />
data <strong>and</strong> different concentration vs. distance relations, where dispersion is completely neglected as well as accounted for.<br />
Important to note is that the different approaches yield different types <strong>of</strong> degradation rates that should be used for<br />
destined purposes only. Rate constants estimated from concentration vs. time data for single monitoring wells for<br />
example represent point decay rates that are typically used to assess source decay or the time required to reach defined<br />
remediation goals at a particular location (Newell et al. 2002). Complete neglect <strong>of</strong> dispersion in the rate estimation<br />
procedure for concentration vs. distance data along the center line <strong>of</strong> the plume yields bulk attenuation rates, which<br />
quantify the reduction <strong>of</strong> contaminant concentrations with distance from the source due to the combined effects <strong>of</strong><br />
dispersion, dilution (e.g. by recharge) <strong>and</strong> degradation processes. The frequently used method <strong>of</strong> Buscheck <strong>and</strong> Alcantar<br />
∗<br />
Beyer, C., Chen, C., Gronewold, J., Kolditz, O., Bauer, S. (2007): Determination <strong>of</strong> first order degradation rate constants from monitoring networks.<br />
(accepted by Ground Water.).<br />
The manuscript was accepted for publication in the Journal Ground Water. Copyright © 2007 Blackwell Publishing. The definite version will be<br />
available at www.blackwell-synergy.com.<br />
1
(1995) yields a “hybrid” rate constant between a bulk attenuation <strong>and</strong> a pure biodegradation rate constant, as it accounts<br />
for longitudinal dispersion whereas the effects <strong>of</strong> transverse dispersion are still reflected in the rate constant. A pure<br />
biodegradation rate constant can be estimated analytically by normalizing contaminant concentrations to concentrations<br />
<strong>of</strong> a recalcitrant tracer, if such a compound is emitted from the same source zone as the contaminant <strong>of</strong> concern<br />
(Wiedemeier et al. 1996). If this is not the case, an approach <strong>of</strong> Zhang <strong>and</strong> Heathcote (2003) may be used, in which the<br />
Buscheck <strong>and</strong> Alcantar (1995) method is extended to account for dispersion in two or three dimensions. Pure<br />
biodegradation rate constants exclusive <strong>of</strong> dispersion or other attenuation processes (<strong>and</strong> only those) may be used in<br />
contaminant transport models for prognoses <strong>of</strong> plume trends. It is important to note that these center line based rate<br />
constant estimation approaches are only applicable to contaminant plumes that have reached steady state conditions, as<br />
for still exp<strong>and</strong>ing plumes the rate constant would be overestimated.<br />
Although it is known, that biodegradation rate estimates obtained from an investigation <strong>of</strong> the plume center line are<br />
subject to substantial uncertainty, this strategy is frequently used in practice. Even in homogeneous aquifers vertical <strong>and</strong><br />
horizontal transverse dispersion can produce center line concentration pr<strong>of</strong>iles <strong>of</strong> recalcitrant compounds that could be<br />
mistaken as following from first order degradation (McNab Jr. <strong>and</strong> Dooher 1998). Moreover, the plume axis may easily<br />
be missed by monitoring wells when the inferred ground water <strong>flow</strong> direction is incorrect, changes over time due to<br />
transient <strong>flow</strong> behavior or when the contaminant plume shows large scale me<strong>and</strong>ering due to aquifer heterogeneity<br />
(Newell et al. 2002; Wilson et al. 2004). Therefore, biodegradation rate constants obtained from such field data should be<br />
taken as rough estimates only (Chapelle et al. 2003). As first order rates calculated from center line data include all<br />
effects <strong>and</strong> processes that lower local contaminant concentrations, the precarious result <strong>of</strong> the different sources <strong>of</strong><br />
uncertainty is that the degradation potential may be severely overestimated (Rittmann 2004), causing underestimation <strong>of</strong><br />
plume length <strong>and</strong> contaminant mass as well as a too optimistic prognosis <strong>of</strong> down gradient concentrations <strong>and</strong> exposure<br />
levels. These aspects were recently studied in-depth by Bauer et al. (2006a) <strong>and</strong> Beyer et al. (2006) by means <strong>of</strong><br />
<strong>numerical</strong> experiments in two-dimensional synthetic heterogeneous contaminated aquifers. The <strong>numerical</strong> experiments<br />
were based on coarse monitoring networks with 6 to 8 wells, which were all placed along an inferred plume center line.<br />
In reality, however, monitoring networks typically are designed to suit multiple <strong>and</strong> sometimes conflicting requirements<br />
<strong>and</strong> objectives, which may also change with time, depending on the stage <strong>of</strong> the site investigation. Objectives in this<br />
context are the detection <strong>of</strong> ground water contamination (e.g. Storck et al. 1997), site characterization <strong>and</strong> spatial<br />
delineation <strong>of</strong> the contamination (e.g. McGrath <strong>and</strong> Pinder 2003) or the long term monitoring <strong>of</strong> the plume behavior (e.g.<br />
Wu et al. 2006). These aims require spatially more extensive monitoring networks with larger numbers <strong>of</strong> wells<br />
compared to the relatively simple center line well configurations necessary to estimate a degradation rate constant.<br />
Recently, Stenback et al. (2004) demonstrated that additional <strong>of</strong>f center line measurements can be incorporated in the<br />
estimation <strong>of</strong> the degradation rate, when a two-dimensional analytical transport model is fitted to contaminant<br />
concentrations <strong>of</strong> all monitoring wells <strong>of</strong> an extensive monitoring network downgradient from the source. In a field<br />
application example Stenback et al. (2004) showed, that accounting for the additional information on contaminant<br />
concentrations <strong>and</strong> distribution significantly reduced rate constant estimates obtained from the conventional center line<br />
approach to about 50 %, pointing out the well known problem <strong>of</strong> rate constant overestimation with the 1D center line<br />
method.<br />
This paper therefore studies the performance <strong>of</strong> several rate constant estimation approaches based on center line<br />
investigation data for sites with extensive monitoring networks <strong>and</strong> compares the results to the two-dimensional approach<br />
<strong>of</strong> Stenback et al. (2004). Adopting the terminology <strong>of</strong> Newell et al. (2002) the types <strong>of</strong> degradation rate constants<br />
regarded here comprise bulk attenuation, biodegradation <strong>and</strong> “hybrid” rate constants. Point decay rates are not addressed<br />
in this paper. Both strategies for rate constant estimation, i.e. the investigation <strong>of</strong> the plume center line <strong>and</strong> the approach<br />
<strong>of</strong> Stenback et al. (2004), are applied to a set <strong>of</strong> synthetic sites with extensive monitoring networks, which were<br />
independently designed <strong>and</strong> installed by individual test persons engaged in hydrogeological research <strong>and</strong> consulting. The<br />
networks were installed for a general characterization <strong>and</strong> quantification <strong>of</strong> the contaminant plume (Chen et al. 2005;<br />
Bauer et al. 2006b), <strong>and</strong> are used here as a basis for degradation rate estimation. Estimated rate constants for both<br />
strategies are compared with regard to magnitude <strong>of</strong> errors <strong>and</strong> variability to draw conclusions on their limitations in<br />
view <strong>of</strong> the monitoring network used. Thereby the studies <strong>of</strong> Bauer et al. (2006a) <strong>and</strong> Beyer et al. (2006) are considerably<br />
extended, as the monitoring networks used in this paper are representative <strong>of</strong> real field situations by using all installed<br />
observation wells, <strong>and</strong> are not restricted to an one-dimensional plume center line. Moreover, this analysis allows an<br />
evaluation <strong>of</strong> the “human factor” on estimated rate constants resulting from individual notions <strong>of</strong> “sufficient accuracy” in<br />
plume investigation.<br />
Background <strong>and</strong> Scope<br />
This study is based on a set <strong>of</strong> synthetic contaminated two-dimensional aquifers, generated by multiple stochastic<br />
simulations <strong>of</strong> heterogeneous hydraulic conductivity fields <strong>and</strong> subsequent <strong>numerical</strong> simulation <strong>of</strong> contaminant<br />
spreading from a defined source zone in the synthetic aquifers. The evolved virtual plumes were independently<br />
investigated by a number <strong>of</strong> German scientists engaged in hydrogeological <strong>and</strong> environmental research using an<br />
2
interactive plume investigation <strong>and</strong> mapping s<strong>of</strong>tware (Chen et al. 2005) with any number <strong>of</strong> wells considered necessary<br />
for a characterization <strong>of</strong> the synthetic sites <strong>and</strong> the delineation <strong>of</strong> the contaminant plume. The plume investigations were<br />
performed within a related project (Bauer et al. 2006b) <strong>and</strong> are described in detail below. The results <strong>of</strong> these virtual<br />
plume investigations yield individually <strong>and</strong> realistically investigated plumes. The “measured” concentrations, hydraulic<br />
heads <strong>and</strong> conductivities at the observation well networks <strong>of</strong> the different plumes investigated are used as a basis for the<br />
determination <strong>of</strong> the degradation rate constants. Two different strategies are compared here:<br />
• Strategy A - One-dimensional center line approach: From all observation wells installed the concentration<br />
distribution is analyzed <strong>and</strong> wells along the approximate center line <strong>of</strong> the contaminant plume are identified (see<br />
Figure 1(a)). Concentrations measured in these wells are evaluated using three different methods for the<br />
inference <strong>of</strong> a degradation rate constant (Newell et al. 2002; Buscheck <strong>and</strong> Alcantar 1995; Zhang <strong>and</strong> Heathcote<br />
2003). Thus, here only a subset <strong>of</strong> concentrations measured in the original monitoring network is used for rate<br />
constant estimation.<br />
• Strategy B - Two-dimensional evaluation: An approximate solution for the two-dimensional steady state<br />
concentration distribution is fitted to concentrations measured in all observation wells <strong>of</strong> the monitoring network<br />
downgradient <strong>of</strong> the source zone using a residual least squares criterion (Stenback et al. 2004) (Figure 1(b)).<br />
Fitting parameters are the first order rate constant <strong>and</strong> the source concentration.<br />
Figure 1: a) Strategy A <strong>and</strong> b) strategy B for the inference <strong>of</strong> the degradation rate constant using an existing monitoring<br />
network. Shown is the true virtual plume (unknown to the investigator), the monitoring network (small squares) <strong>and</strong> the<br />
identified center line wells (larger squares) (strategy A) as well as an example <strong>of</strong> a fitted 2D analytical plume (contour<br />
lines) (strategy B).<br />
Generation <strong>of</strong> synthetic sites<br />
The data basis <strong>of</strong> this study consists <strong>of</strong> 20 different realizations <strong>of</strong> heterogeneous steady state contaminant plumes. The<br />
plumes were generated by <strong>numerical</strong> simulation <strong>of</strong> ground water <strong>flow</strong> <strong>and</strong> contaminant transport in 20 different twodimensional<br />
horizontal aquifer realizations with heterogeneous hydraulic conductivity distributions. Into ten <strong>of</strong> these<br />
conductivity fields a source with a width <strong>of</strong> 4 m perpendicular to the mean <strong>flow</strong> direction was introduced, for the other 10<br />
aquifer realizations, a source width <strong>of</strong> 16 m was used. For the <strong>numerical</strong> simulations the GeoSys/Rock<strong>flow</strong> code (Kolditz<br />
et al. 2006; Kolditz <strong>and</strong> Bauer 2004) was used, which solves the <strong>flow</strong> <strong>and</strong> transport equations by finite element methods.<br />
The governing equations for steady state <strong>flow</strong> conditions are given as (e.g. Bear 1972):<br />
∇( K ∇h)<br />
= 0<br />
(1)<br />
∂C<br />
= −va∇C<br />
+ ∇(<br />
D∇C)<br />
− λC<br />
(2)<br />
∂t<br />
with h [L] the hydraulic head, K [L T -1 ] the tensor <strong>of</strong> hydraulic conductivity, C [M L -3 ] concentration, D [L 2 T -1 ] the<br />
dispersion tensor which is calculated acording to Bear (1972), va [L T -1 ] the transport velocity, t [T] time <strong>and</strong> λ [T -1 ]<br />
representing the first order degradation rate constant. Equation (2) was solved over a two-dimensional <strong>numerical</strong> grid <strong>of</strong><br />
184 m length by 64 m width with node spacing <strong>of</strong> 0.5 m in both directions <strong>and</strong> setting the left h<strong>and</strong> side <strong>of</strong> equation (2) to<br />
0. This guarantees that the simulated plumes are at steady state. For local dispersivities αL <strong>and</strong> αT values <strong>of</strong> 0.25 m <strong>and</strong><br />
0.05 m are used. A mean hydraulic gradient I <strong>of</strong> 0.003 is induced by fixed head boundary conditions on the left <strong>and</strong> the<br />
right h<strong>and</strong> side <strong>of</strong> the model domain (Figure 2). Flow conditions are at steady state. Hydraulic conductivity K <strong>of</strong> the<br />
virtual aquifers is regarded as a spatial r<strong>and</strong>om variable (Figure 2), following a lognormal distribution with an expected<br />
value <strong>of</strong> E[Y = ln(K)] = -9.54, which corresponds to an effective conductivity Kef <strong>of</strong> 7.19·10 -5 m s -1 using the geometric<br />
mean. As the aquifer models are two-dimensional <strong>and</strong> horizontal, an isotropic exponential covariance function with an<br />
3
2<br />
integral scale lY <strong>of</strong> 2.67 m <strong>and</strong> ln-conductivity variance σ Y = 2.7 is used for the spatial correlation structure. Kef <strong>and</strong> lY<br />
2<br />
are taken from the Borden field site (Sudicky 1986), whereas the degree <strong>of</strong> heterogeneity σ Y was reported for the<br />
Columbus Air Force Base site (Rehfeldt et al. 1992). Porosity n is set to 0.33, resulting in va = 6.54 10 -7 m s -1 . For local<br />
dispersivities αL <strong>and</strong> αT values <strong>of</strong> 0.25 m <strong>and</strong> 0.05 m are used. The contaminant source introduced in the aquifers is <strong>of</strong><br />
rectangular shape centered at [11.5 m; 32.0 m] downstream <strong>of</strong> the in<strong>flow</strong> boundary <strong>and</strong> has an area <strong>of</strong> either 3 * 4 m or 3<br />
* 16 m, corresponding to a source width WS <strong>of</strong> 1.5 or 6 correlation lengths transverse to the mean <strong>flow</strong> direction. It emits<br />
a contaminant with the contaminant concentration fixed in the source area at C / C0 = 1.0. The dissolved contaminant is<br />
subject to a first order kinetics degradation reaction with a rate constant λ <strong>of</strong> 1.59·10 -8 s -1 (0.5 a -1 = 0.5 yr -1 ). This value is<br />
well within the range <strong>of</strong> reported / recommended first order degradation rate constants for chlorinated solvents as well as<br />
petroleum hydrocarbons under anaerobic conditions listed in Aronson <strong>and</strong> Howard (1997). The model parameters used<br />
are summarized in Table 1.<br />
contaminant source: C = 1.0<br />
0<br />
fixed head<br />
h = 0.552 m<br />
4<br />
no <strong>flow</strong><br />
no <strong>flow</strong><br />
fixed head<br />
h = 0.0 m<br />
-1<br />
K [m s ]<br />
Figure 2: A single realization <strong>of</strong> the spatially correlated r<strong>and</strong>om hydraulic conductivity field <strong>and</strong> model boundary<br />
conditions.<br />
The conceptual model used here is a rigorous simplification <strong>of</strong> contaminant transport <strong>and</strong> degradation observed in natural<br />
aquifer systems, where biodegradation is a function <strong>of</strong> electron acceptor <strong>and</strong> donor availability <strong>and</strong> the aquifer structure<br />
<strong>and</strong> thus reaction kinetics follow more complicated laws, show spatial dependence <strong>and</strong> may include transient effects,<br />
dilution or phase changes to the un<strong>saturated</strong> zone. Furthermore, the contaminant is not retarded <strong>and</strong> shows no<br />
volatilization. However, as the methods assume a uniform rate constant to be valid, only using such simplified<br />
representations <strong>of</strong> reality as virtual test sites allows for a detailed analysis <strong>of</strong> the influence <strong>of</strong> heterogeneous hydraulic<br />
conductivity on the different methods under otherwise ideal conditions.<br />
1.0*10 -2<br />
1.0*10 -3<br />
1.0*10 -4<br />
1.0*10 -5<br />
1.0*10 -6<br />
1.0*10 -7<br />
1.0*10 -8<br />
1.0*10 -9<br />
Table 1: Model parameters used in the <strong>numerical</strong> simulations.<br />
parameter description value<br />
Kef effective conductivity 7.19·10 -5 m s -1<br />
2<br />
σ Y<br />
ln(K)-variance 2.7<br />
lY integral scale 2.67 m<br />
n porosity 0.33<br />
αL longitudinal dispersivity 0.25 m<br />
αT transverse dispersivity 0.05 m<br />
I hydraulic gradient 0.003<br />
λ first order degradation rate constant 1.59·10 -8 s -1<br />
Investigation <strong>of</strong> synthetic sites<br />
The plumes were independently investigated by 85 different test persons, engaged in hydrogeological <strong>and</strong> environmental<br />
research, consulting or administration. These investigators were confronted with the scenario <strong>of</strong> contaminant migration<br />
downstream from a source zone in a virtual aquifer. A two-dimensional top view <strong>of</strong> the site, the mean ground water <strong>flow</strong><br />
direction <strong>and</strong> the approximate location <strong>of</strong> the contaminant source were given as the only prior information for the site<br />
investigation. Neither the plume nor the hydraulic heads calculated are known at this stage. The task <strong>of</strong> the investigators<br />
was an as exact as possible characterization <strong>of</strong> the contaminated aquifer by the following procedure:<br />
• Step 1) Emplacing observation wells into the virtual aquifer: Using an interactive graphical user interface, the<br />
investigator positions observation wells on the virtual site. At the wells local contaminant concentrations <strong>and</strong><br />
hydraulic heads are measured.<br />
• Step 2) Regionalization <strong>of</strong> local measurements: Using different interpolation schemes (e.g. Kriging or Inverse<br />
Distance Weighting) the investigator interpolates contaminant concentrations <strong>and</strong> hydraulic heads measured at<br />
the observation wells to the virtual aquifer.
Steps 1 <strong>and</strong> 2 could be repeated as <strong>of</strong>ten as desired by the investigator, until the interpolated contaminant plume was<br />
deemed to be investigated accurately enough to properly characterize the contaminant distribution. Thus the interactive<br />
site investigation is an iterative procedure <strong>and</strong> is evaluated by a comparison <strong>of</strong> the “true” plume with the investigation<br />
result. More details on this procedure can be found in Chen et al. (2005).<br />
Using this methodology, a total number <strong>of</strong> 85 individual investigation results were obtained. 47 <strong>of</strong> these investigations<br />
were conducted for plumes originating from a source <strong>of</strong> width 4 m, while the plumes with a source width <strong>of</strong> 16 m were<br />
investigated 38 times. Accordingly, the majority <strong>of</strong> the 20 different plumes were investigated three or four times by<br />
different investigators.<br />
The configuration <strong>and</strong> the number <strong>of</strong> monitoring wells varies substantially between the different realizations <strong>and</strong> even for<br />
the same realization but different investigators (12 – 93 wells), as the decision about how many wells were needed for an<br />
accurate characterization <strong>of</strong> the site was left to the individual investigators. For two contaminant plumes (one for each<br />
source width) the investigation was repeated 13 times. Each investigation was performed by a different investigator. In<br />
addition to the general comparison <strong>of</strong> strategies A <strong>and</strong> B for the inference <strong>of</strong> the degradation rate constant, this subset <strong>of</strong><br />
the whole data set allows for an analysis <strong>of</strong> estimated rate constant variability for a single site, resulting from different<br />
notions <strong>of</strong> “sufficient accuracy” <strong>of</strong> the virtual plume investigation.<br />
Rate constant estimation<br />
Using the monitoring networks installed by the investigators <strong>of</strong> the virtual plumes, first order rate constants were<br />
estimated following strategies A <strong>and</strong> B. Methods A1, A2 <strong>and</strong> A3 are used in strategy A <strong>and</strong> are described in detail in<br />
Bauer et al. (2006a). Therefore, only a brief description is given here:<br />
Method A1 (equation (3)) is based on the one-dimensional transport equation, considering advection <strong>and</strong> first order<br />
degradation only. The steady state solution for the concentration pr<strong>of</strong>ile is rearranged to yield λA1, i.e. the first order<br />
degradation rate constant for method A1:<br />
va ⎛ C(<br />
x)<br />
⎞<br />
λ A1<br />
= − ln⎜ ⎟<br />
∆x<br />
⎜ ⎟<br />
(3)<br />
⎝ C0<br />
⎠<br />
va [L T -1 ] is the transport velocity, ∆x [L] the distance between the observation wells <strong>and</strong> C0 <strong>and</strong> C(x) [M L -3 ] are<br />
upstream <strong>and</strong> downstream contaminant concentrations at the observation wells, respectively. λA1 can be considered rather<br />
an overall or bulk attenuation rate than a degradation rate constant (Newell et al. 2002), as all concentration changes due<br />
to diffusion, dispersion, volatilization or dilution are attributed to the degradation process.<br />
The method introduced by Buscheck <strong>and</strong> Alcantar (1995) is the second approach applied in this study (equation (4)). It is<br />
based on the steady state solution <strong>of</strong> the one-dimensional transport equation considering advection, longitudinal<br />
dispersion <strong>and</strong> first order degradation. Method A2 requires an estimate <strong>of</strong> longitudinal dispersivity αL [L] <strong>and</strong> yields a<br />
“hybrid” rate constant between a bulk attenuation <strong>and</strong> a pure biodegradation rate constant, as the effects <strong>of</strong> transverse<br />
dispersion are still reflected in the rate constant estimate λA2.<br />
⎛<br />
2<br />
v<br />
( ) ⎞<br />
a ⎜⎛<br />
ln C(<br />
x)<br />
C0<br />
⎞<br />
λ = ⎜1<br />
− 2<br />
⎟ −1⎟<br />
A2<br />
α L<br />
4α<br />
⎜<br />
⎟<br />
L ⎝⎝<br />
∆x<br />
⎠ ⎠<br />
(4)<br />
Zhang <strong>and</strong> Heathcote (2003) proposed modifications to the method <strong>of</strong> Buscheck <strong>and</strong> Alcantar (1995) to improve the<br />
estimation <strong>of</strong> λ with regard to transverse dispersion. Correction terms derived from analytical solutions to the two- <strong>and</strong><br />
three-dimensional advection dispersion equations including first order decay are used to account for lateral spreading <strong>and</strong><br />
the width <strong>of</strong> the source zone WS [L] in two <strong>and</strong> three dimensions, respectively. Therefore information about WS, αL <strong>and</strong><br />
αT are required for this approach. Method A3 (equation (5)) is the two-dimensional form <strong>of</strong> the method by Zhang <strong>and</strong><br />
Heathcote (2003) <strong>and</strong> yields the biodegradation rate constant estimate λA3:<br />
2<br />
v ⎛ ( ) ⎞<br />
a ⎜⎛<br />
ln C(<br />
x)<br />
( C0β<br />
) ⎞<br />
⎛ ⎞<br />
λ = − ⎟<br />
A3<br />
⎜<br />
⎜1−<br />
2α<br />
L<br />
⎟ 1 with<br />
⎜ WS<br />
β = erf ⎟ (5)<br />
4α<br />
⎟<br />
L ⎝⎝<br />
∆x<br />
⎠<br />
⎜ ⎟<br />
⎠<br />
⎝ 4 αT<br />
∆x<br />
⎠<br />
For evaluation strategy B, method B (equation (6)) is used, which corresponds to the approach <strong>of</strong> Stenback et al. (2004).<br />
An approximate solution for the steady state concentration distribution derived from the two-dimensional advectiondispersion<br />
equation with first order degradation (Domenico <strong>and</strong> Schwartz 1990) is fitted to measured concentrations:<br />
C ⎪⎧<br />
⎛<br />
⎞⎪⎫<br />
⎪<br />
⎧ ⎛ ⎞ ⎛ ⎞⎪<br />
⎫<br />
0 ⎛ x ⎞<br />
⎨<br />
⎜ + ⎟ ⎜ −<br />
⎨ ⎜<br />
4λ<br />
⎟⎬<br />
−<br />
⎟<br />
⎜<br />
⎟<br />
Bα<br />
L 2y<br />
WS<br />
2y<br />
WS<br />
C(<br />
x,<br />
y)<br />
= exp 1−<br />
1+<br />
erf<br />
erf ⎬ (6)<br />
2 ⎪⎩ ⎝ 2α<br />
⎠<br />
⎜<br />
⎟<br />
⎪⎭ ⎪⎩<br />
⎜ ⎟ ⎜ ⎟<br />
L ⎝<br />
va<br />
⎠ ⎝ 4 αT<br />
x ⎠ ⎝ 4 αT<br />
x ⎠⎪⎭<br />
5
The approximate solution employed has the same basis as method 3. Here, however, a real two-dimensional approach is<br />
used, as equation (6) is fitted to the concentrations <strong>of</strong> all observation wells <strong>and</strong> not only to those measured along the<br />
center line.<br />
In strategy A the first step is an analysis <strong>of</strong> the concentration distribution based on all measurements made at the<br />
observation wells <strong>of</strong> the monitoring network. Starting at the well with the highest measured concentration, those<br />
downgradient wells are identified, that best represent the center line <strong>of</strong> the plume. These then are used to estimate the rate<br />
constant. Only locally measured quantities are used here, i.e. local hydraulic conductivities, hydraulic heads <strong>and</strong><br />
contaminant concentrations are “measured” at these wells by reading the model data at the respective nodes <strong>of</strong> the<br />
<strong>numerical</strong> grid. The transport velocity va between each pair <strong>of</strong> center line wells is approximated by:<br />
∆h<br />
va = K ef<br />
(7)<br />
n∆x<br />
with Kef the effective conductivity <strong>of</strong> local hydraulic conductivities at up <strong>and</strong> down gradient wells, n the porosity, ∆h the<br />
head difference <strong>and</strong> ∆x the distance between the wells. According to Rubin (2003) in stationary isotropic twodimensional<br />
domains with gaussian probability density functions <strong>of</strong> Y = ln(K) the effective conductivity Kef can be<br />
calculated as the geometric mean (cf. Bauer et al. 2006a). The porosity is assumed to be known correctly. With va, λ can<br />
then be calculated for each pair <strong>of</strong> center line wells using methods A1 – A3. For a set <strong>of</strong> k center line wells thus k-1 rate<br />
constants are calculated for one method. These are averaged to yield an estimate <strong>of</strong> the mean degradation rate constant λ.<br />
As an alternative to using only the locally measured conductivities, rate constant estimation is also performed using a<br />
global estimate <strong>of</strong> Kef, which is obtained from the geometric mean value <strong>of</strong> all hydraulic conductivities measured at all<br />
observation wells.<br />
In strategy B, method B (equation (6)) is fitted to measured concentrations <strong>of</strong> all observation wells <strong>of</strong> the monitoring<br />
network. Both the biodegradation rate constant λ <strong>and</strong> the source concentration C0 are varied simultaneously to achieve<br />
correspondence <strong>of</strong> measured <strong>and</strong> calculated concentrations, as a preliminary analysis (in agreement with results <strong>of</strong><br />
Stenback et al. (2004)) showed that this procedure on average yields closer estimates <strong>of</strong> the true rate constant than fitting<br />
only λ with a single fixed estimate <strong>of</strong> the source concentration. A least squares criterion for the concentration residuals is<br />
used in the fitting procedure. As in strategy A the <strong>flow</strong> velocity va is approximated using equation (7). Kef is calculated as<br />
the geometric mean <strong>of</strong> hydraulic conductivities measured at all wells <strong>of</strong> the network. The average hydraulic gradient over<br />
the entire site is approximated by fitting a linear trend surface to all head measurements by ordinary least squares<br />
regression.<br />
For methods A2, A3 <strong>and</strong> B estimates <strong>of</strong> longitudinal <strong>and</strong> transverse dispersivities are required. Practical guidance on<br />
estimating these parameters at the field scale is given e.g. by Wiedemeier et al. (1999), where one suggestion is to use 0.1<br />
times the plume length for αL <strong>and</strong> αT as 0.1 αL. Here, however, an alternative strategy is employed: Macrodispersivities<br />
2<br />
αL <strong>and</strong> αT are derived from correlation scale, aquifer heterogeneity σ Y <strong>and</strong> travel distance (Dagan 1984; Hsu 2003;<br />
Rubin et al. 2003). Thus, αL is taken as 7 m, which roughly corresponds to the large time asymptotic limit for the given<br />
conductivity distribution, while αT is taken as the approximate peak value <strong>of</strong> transverse macrodispersivity, calculated as<br />
0.7 m (which thus also corresponds to the frequently used relationship as αT ≈ 0.1 αL). These values are well within the<br />
ranges <strong>of</strong> dispersivities commonly used for the field scale <strong>modeling</strong> <strong>of</strong> contaminant transport. The true value <strong>of</strong> the<br />
source width WS is assumed to be known from the site investigation. These approximations <strong>and</strong> assumptions were made<br />
to ensure that the error introduced in estimated rate constants due to the parameterization <strong>of</strong> methods A2, A3 <strong>and</strong> B is as<br />
small as possible.<br />
Results <strong>and</strong> Discussion<br />
Strategy A - One-dimensional center line approach<br />
The rate constants λA1 - λA3 estimated with equations (3) – (5) are divided by the “true” value used in the <strong>numerical</strong><br />
simulations to yield normalized rate constants ΛA1 - ΛA3, which can directly be interpreted as overestimation or<br />
underestimation factors. Results in terms <strong>of</strong> mean values, medians, st<strong>and</strong>ard deviations <strong>and</strong> coefficients <strong>of</strong> variation (cv)<br />
as well as the number <strong>of</strong> realizations (N) used for strategy A are presented in Table 2 <strong>and</strong> Figure 3. Comparing methods<br />
A1 - A3 for the small source width WS = 4 m yields that all approaches on average result in a distinct overestimation <strong>of</strong> λ.<br />
For method A1 λ is overestimated on average by a factor <strong>of</strong> 6.88, while for A2 an mean ΛA2 = 8.24 is observed. Hence,<br />
method A1 performs better than A2. Method A3 yields a slightly lower mean <strong>of</strong> ΛA3 = 6.82, while the spread <strong>of</strong> results<br />
for A3 is noticeably larger, as can also be seen by the higher cv. Three main error sources for the observed<br />
overestimation exist:<br />
• neglect <strong>of</strong> dispersion by method A1, which thus is attributed to the degradation process (λA1 represents a bulk<br />
attenuation rate constant)<br />
6
• deviation <strong>of</strong> sampling well locations from the true center line position, which causes the sampling <strong>of</strong> too low<br />
concentrations<br />
• unrepresentative estimates <strong>of</strong> the local va along the <strong>flow</strong> path, as the estimated λ increases <strong>and</strong> decreases with va<br />
(cf. equations (3) – (5)).<br />
These effects are discussed in detail in Bauer et al. (2006a). For method A2 the additional bias towards too large rate<br />
constants is a consequence <strong>of</strong> accounting only for αL in equation (4), as with a one-dimensional transport model<br />
longitudinal dispersion <strong>of</strong> a degrading contaminant results in a stronger spreading <strong>of</strong> the solute downstream <strong>and</strong><br />
consequently in higher concentrations along the center line <strong>of</strong> a steady state plume compared to an advection only case.<br />
Therefore a larger rate constant is needed to fit a given concentration decrease <strong>and</strong> the “hybrid” rate constant estimate ΛA2<br />
is always larger than the bulk attenuation rate constant ΛA1 (cf. Bauer et al. 2006a), which appears contradictory.<br />
Table 2: Normalized degradation rate constants estimated<br />
with strategy A.<br />
WS = 4 m<br />
method A1 method A2 method A3<br />
mean 6.88 8.24 6.82<br />
median 4.43 5.50 2.65<br />
stdv. 6.36 7.41 10.09<br />
cv 0.92 0.90 1.48<br />
N 47 47 47<br />
WS = 16 m<br />
method A1 method A2 method A3<br />
mean 4.46 5.33 4.99<br />
median 2.66 3.23 2.45<br />
stdv. 4.81 6.29 9.59<br />
cv 1.08 1.18 1.92<br />
N 38 38 36<br />
For the source width WS = 16 m all methods improve (Table 2, Figure 3). Still, method A1 shows the closest estimates <strong>of</strong><br />
the rate constant <strong>and</strong> the lowest variability <strong>of</strong> single realization results in comparison to methods A2 <strong>and</strong> A3. With<br />
method A3 for two out <strong>of</strong> 38 plumes no reasonable rate constant estimate is obtained. This effect is due to an overcorrection<br />
for transverse dispersion in the β term <strong>of</strong> equation (5), which can cause the corrected C(x) to be very close to<br />
or even larger than the respective upgradient concentration C0, yielding very small or even negative λA3.<br />
norm. degradation rate constant [-]<br />
100<br />
10<br />
1<br />
W S = 4 m W S = 16 m<br />
single realization result<br />
ensemble mean<br />
0.1<br />
A1 1 A2 2 A3 3<br />
A1 1 A2 2 A3 3<br />
method<br />
method<br />
Figure 3: Normalized degradation rate constants estimated with strategy A <strong>and</strong> methods A1, A2 <strong>and</strong> A3 for source widths<br />
<strong>of</strong> 4 m (left diagram) <strong>and</strong> 16 m (right diagram). Small symbols represent results <strong>of</strong> individual realizations, large symbols<br />
the mean <strong>of</strong> all realizations. Kef used for rate estimation is calculated from measurements at center line wells only.<br />
7
As Bauer et al. (2006a) demonstrated, the local <strong>flow</strong> velocity estimate is a crucial parameter for the estimation <strong>of</strong> the<br />
degradation rate constant. In a sensitivity study the authors found that inclusion <strong>of</strong> field scale information on hydraulic<br />
conductivity, which is more representative for the entire site, can improve the accuracy <strong>of</strong> estimated rate constants over<br />
usage <strong>of</strong> local information only. Therefore the center line data obtained in strategy A are re-evaluated using a field scale<br />
estimate <strong>of</strong> the effective conductivity at the investigated sites: Kef is calculated as the geometric mean <strong>of</strong> local<br />
conductivity values measured at all observation wells <strong>of</strong> the site <strong>and</strong> not only using those measured on the center line.<br />
Results for this approximation are presented in Table 3 <strong>and</strong> Figure 4. It is found that consideration <strong>of</strong> all conductivity<br />
measurements available improves the rate constants estimated for all three methods <strong>and</strong> both source widths. On average<br />
all estimates are closer to the true rate than those estimated using only local conductivity information. The average<br />
improvement is between a factor <strong>of</strong> 1.5 for WS = 4 m <strong>and</strong> a factor <strong>of</strong> two for WS = 16 m. Also the st<strong>and</strong>ard deviations<br />
from the mean values <strong>and</strong> the cv are lower. This finding is true for the settings investigated here, i.e. a correlation length<br />
smaller than the average distance between observation wells <strong>and</strong> a stationary K distribution. If the correlation length is on<br />
the order or longer than the average observation well distance or K is not stationary, this finding is probably not<br />
transferable.<br />
norm. degradation rate constant [-]<br />
100<br />
10<br />
1<br />
0.1<br />
Table 3: Normalized degradation rate constants estimated<br />
with strategy A, re-evaluated using a field scale estimate <strong>of</strong><br />
Kef.<br />
WS = 4 m<br />
method A1 method A2 method A3<br />
mean 4.75 5.73 4.14<br />
median 4.04 4.67 2.66<br />
stdv. 3.75 4.33 4.06<br />
cv 0.79 0.76 0.98<br />
N 47 47 47<br />
WS = 16 m<br />
method A1 method A2 method A3<br />
mean 2.16 2.53 2.42<br />
median 1.14 1.50 0.90<br />
stdv. 2.08 2.45 3.98<br />
cv 0.96 0.97 1.64<br />
N 38 38 36<br />
W S= 4 m W S = 16 m<br />
single realization result<br />
ensemble mean<br />
A1 1 A2 2 A3 3<br />
A1 1 A2 2 A3 3<br />
method<br />
method<br />
Figure 4: Normalized degradation rate constants estimated with strategy A <strong>and</strong> methods A1, A2 <strong>and</strong> A3 for source widths<br />
<strong>of</strong> 4 m (left diagram) <strong>and</strong> 16 m (right diagram), respectively. Kef used for rate estimation is calculated as a field scale<br />
estimate from all monitoring wells available for each investigated plume.<br />
8
An interesting observation is that the differences between the three methods <strong>of</strong> strategy A are not as distinct as observed<br />
in Bauer et al. (2006a). One explanation for this finding is that due to the larger number <strong>of</strong> observation wells used for the<br />
site investigation in this study, the plume center line positions are better identified on average <strong>and</strong> concentration samples<br />
are taken closer to the true plume axis. Methods A1, A2 <strong>and</strong> A3 show a different sensitivity on deviations <strong>of</strong><br />
measurement locations from the plume center line. This is because in method A1, the rate constant estimate is linearly<br />
related to ln( C( x)<br />
/ C0<br />
) / ∆x<br />
, while in method A2 this term appears in linear as well as squared form after rearrangement<br />
<strong>of</strong> equation (4). As a consequence, increasing deviations <strong>of</strong> observation well locations from the plume center line <strong>and</strong><br />
thus lower measured contaminant concentrations will result in increasingly stronger overestimation <strong>of</strong> λ by A2 relative to<br />
A1. For method A3, the correction factor β has to be taken into account additionally (cf. equation (5)). For significant<br />
deviations from the center line, however, the same effect as for A2 can be shown.<br />
In Figure 5 the degradation rate constants ΛA1 estimated with method A1 were plotted against the number <strong>of</strong> observation<br />
wells with relative concentrations C/C0 > 0.001 in the respective monitoring networks. A clear relationship between the<br />
number <strong>of</strong> wells <strong>and</strong> the accuracy <strong>of</strong> ΛA1 is neither observed for WS = 4 m nor for WS = 16 m. It seems, however, that the<br />
spread <strong>of</strong> estimated rate constants decreases slightly with increasing numbers <strong>of</strong> wells, but a larger number <strong>of</strong> samples,<br />
especially for numbers <strong>of</strong> monitoring wells > 30 would be required to derive more meaningful results. For the other<br />
methods A2 <strong>and</strong> A3, the similar observations are made (not shown here).<br />
norm. deg. rate constant Λ A1 [-]<br />
50<br />
10<br />
1<br />
0.1<br />
Y = -0.074 * X + 6.531; R<br />
0 20 40 60 80<br />
no. <strong>of</strong> wells with C/C0 > 0.001<br />
2 = 0.063<br />
Y = -0.043 * X + 3.174; R2 linear fits<br />
= 0.072<br />
Figure 5: Degradation rate constants ΛA1 for source widths <strong>of</strong> 4 m (grey diamonds) <strong>and</strong> 16 m (black crosses) versus the<br />
number <strong>of</strong> wells showing relative concentrations C/C0 > 0.001.<br />
9<br />
W S = 4 m<br />
W S = 16 m<br />
Strategy B - Two-dimensional evaluation<br />
Applying method B it is not always possible to obtain a rate constant λB > 0 when minimizing the sum <strong>of</strong> squared<br />
residuals <strong>of</strong> concentration. The concentration distribution calculated with the analytical model indicates absence <strong>of</strong><br />
contaminant degradation for the closest fit to the measured data for six out <strong>of</strong> 47 plumes when WS = 4 m. For the<br />
remaining 41 plumes, usage <strong>of</strong> method B on average yields ΛB = 4.51 (Table 4, Figure 6), which is considerably closer to<br />
the true rate constant than for methods A1 - A3 with using only local measurements <strong>of</strong> hydraulic conductivity along the<br />
center line (cf. Table 2). The st<strong>and</strong>ard deviation for method B, however, is slightly larger than for A1 <strong>and</strong> A2. With WS =<br />
16 m for five out <strong>of</strong> 38 plumes no ΛB > 0 is found. Here, the improvement over methods A1 – A3 is even more distinct<br />
with the mean ΛB = 1.92. Also the spread <strong>of</strong> single realizations is reduced for the larger source width. However, no<br />
definite improvement <strong>of</strong> ΛB over those obtained by strategy A using the field scale estimate <strong>of</strong> Kef (cf. Table 3) is<br />
observed. For WS = 4 m the mean ΛB is slightly lower than the mean ΛA1, but slightly larger than ΛA3. Also the variability<br />
<strong>of</strong> estimated ΛB is larger than for methods A1 – A3. For WS = 16 m the mean ΛB is slightly lower than for all methods <strong>of</strong><br />
strategy A, while the observed spread is only lower in comparison to A3.<br />
In the estimation <strong>of</strong> λ with method B the source concentration is included as a fitting parameter. On average the true<br />
source concentration is overestimated by a factor <strong>of</strong> three for WS = 4m, while for WS = 16m deviations are lower than 5<br />
%. As for strategy A a distinct relationship between the number <strong>of</strong> observation wells in the monitoring network <strong>and</strong> the<br />
quality <strong>of</strong> the degradation rate estimates is not observed.
Table 4: Normalized degradation rate<br />
constants estimated with method B.<br />
WS = 4 m WS = 16 m<br />
mean 4.51 1.92<br />
median 1.96 1.14<br />
stdv. 8.34 2.97<br />
cv 1.85 1.55<br />
N 41 33<br />
norm. degradation rate constant [-]<br />
100<br />
10<br />
1<br />
0.1<br />
WS = 4 m WS = 16 m<br />
Figure 6: Normalized degradation rate constants estimated with method B for source widths <strong>of</strong> 4 m <strong>and</strong> 16 m.<br />
10<br />
single realization result<br />
ensemble mean<br />
Variability <strong>of</strong> estimated rate constants for a single site<br />
For each source width, one single plume was repeatedly investigated 13 times by different investigators. For this subset<br />
<strong>of</strong> plumes the variability <strong>of</strong> estimated rate constants for a single site, resulting from different notions on “sufficient<br />
accuracy” <strong>of</strong> the plume investigation, is studied. As the incorporation <strong>of</strong> conductivity measurements from all observation<br />
wells <strong>of</strong> the monitoring network significantly improves rate constant estimation with methods A1, A2 <strong>and</strong> A3, this<br />
approach is used in this comparison. Results for strategies A <strong>and</strong> B are summarized in Table 5 <strong>and</strong> Figure 7.<br />
For source width WS = 4 m a slightly stronger overestimation <strong>of</strong> λ can be observed for strategy A (methods A1 - A3) in<br />
comparison to the results for the complete data set. Here, the mean ΛA1 resulting from 13 investigations <strong>of</strong> the single site<br />
is 5.26, while ΛA1 for the complete data set is 4.75. For methods A2 <strong>and</strong> A3 the same tendency <strong>of</strong> increased<br />
overestimation is found. The variability among the sets <strong>of</strong> 13 estimated ΛA1, ΛA2 <strong>and</strong> ΛA3, however, is much smaller. The<br />
cv here are only 0.3, 0.34 <strong>and</strong> 0.68, respectively. With strategy B, i.e. method B, a slightly smaller mean ΛB = 4.26 is<br />
found than for the complete data set <strong>and</strong> the cv <strong>of</strong> 0.78 here also is reduced. For a source width <strong>of</strong> 16 m, all four methods<br />
show a much lower mean value <strong>of</strong> normalized rate constants than for the complete data set. On average, method A1 <strong>and</strong><br />
A3 yield the correct result for the single investigated site, while method A2 <strong>and</strong> B show a slight overestimation.<br />
Variability <strong>of</strong> the estimated rate constants in terms <strong>of</strong> cv is lower than for the complete data set <strong>of</strong> WS = 16 m. For<br />
strategy B no clear dependence between the precision <strong>of</strong> the rate constant estimate <strong>and</strong> the absolute number <strong>of</strong> wells in<br />
the monitoring network nor the number <strong>of</strong> respective wells showing considerable contaminant concentrations (e.g.<br />
C/C0 > 0.001) could be observed. The absence <strong>of</strong> such an obvious relationship may be due to the low sample size <strong>of</strong> only<br />
13 investigations <strong>of</strong> both plume realizations. As described before, it seems, however, that the spread <strong>of</strong> estimated rate<br />
constants decreases slightly with increasing numbers <strong>of</strong> wells (not shown here).<br />
The stronger overestimation observed for WS = 4 m as well as the more precise estimation <strong>of</strong> λ for WS = 16 m do not<br />
allow for any general conclusions regarding the absolute magnitude <strong>of</strong> errors, as this might probably be an effect <strong>of</strong> the<br />
two individual realizations investigated <strong>and</strong> their respective plume geometry. The variability observed among the 13<br />
results for each <strong>of</strong> the two single plumes however shows, that for a single realization <strong>of</strong> a contaminant plume,<br />
substantially different investigation results, in this case degradation rate constants, may be obtained, when investigated<br />
by individual persons. For both, the narrow plume with a comparatively small source zone <strong>of</strong> 4 m width as well as for the<br />
wider plume with WS = 16 m, the variability <strong>of</strong> investigation results here has a magnitude <strong>of</strong> 25 % up to 69 % <strong>of</strong> the<br />
variability observed among different realizations <strong>of</strong> the plume, depending on the method used for rate constant<br />
estimation. This clearly shows, that a substantial part <strong>of</strong> rate constant estimation uncertainty is due to the individual<br />
configuration <strong>of</strong> the monitoring network.
norm. degradation rate constant [-]<br />
100<br />
10<br />
1<br />
0.1<br />
W S = 4 m W S = 16 m<br />
single realization result<br />
ensemble mean<br />
A1 1 A2 2 A3 3 4B<br />
A1 1 A2 2 A3 3 4B<br />
method<br />
method<br />
Figure 7: Comparison <strong>of</strong> normalized degradation rate constants estimated with methods A1 – A3 <strong>and</strong> B for two single<br />
plume realizations with different source widths (Ws = 4 m: left diagram; Ws = 16 m: right diagram). Both plume<br />
realizations were repeatedly <strong>and</strong> independently investigated by 13 individual persons each.<br />
Summary <strong>and</strong> Conclusions<br />
Table 5: Normalized degradation rate constants estimated with methods<br />
A1 – A3 <strong>and</strong> B for two individual plume realizations with different source<br />
widths (Ws = 4 m <strong>and</strong> 16 m). Both plume realizations were independently<br />
investigated by 13 individual persons each.<br />
WS = 4 m<br />
method A1 method A2 method A3 method B<br />
mean 5.26 6.25 4.56 4.26<br />
median 5.02 5.88 2.94 3.47<br />
stdv. 1.59 2.11 3.10 3.33<br />
cv 0.30 0.34 0.68 0.78<br />
N 13 13 13 13<br />
WS = 16 m<br />
method A1 method A2 method A3 method B<br />
mean 1.01 1.28 1.00 1.21<br />
median 0.97 1.15 0.87 1.20<br />
stdv. 0.31 0.51 0.48 0.51<br />
cv 0.31 0.40 0.48 0.42<br />
N 13 13 13 13<br />
In this study, the frequently used center line approach for estimation <strong>of</strong> degradation rate constants is compared to a<br />
methodology suggested by Stenback et al. (2004), which uses all concentration measurements downstream <strong>of</strong> the<br />
contaminant source zone to estimate the degradation rate. In a <strong>numerical</strong> experiment, both strategies are applied to a set<br />
<strong>of</strong> 85 synthetic contaminant plumes, subject to first order degradation, <strong>and</strong> investigated using extensive monitoring<br />
networks. Rate constants are estimated using concentrations, hydraulic heads <strong>and</strong> conductivities locally measured at the<br />
observation wells <strong>of</strong> the monitoring network.<br />
Results show that rate constants tend to be overestimated in general. Using the center line approach (strategy A) <strong>and</strong> a<br />
simplified observation well set up causes a general overestimation due to three factors:<br />
• biased estimates <strong>of</strong> the mean <strong>flow</strong> velocity,<br />
• inadequate parameterizations <strong>of</strong> longitudinal <strong>and</strong> transverse dispersivities <strong>and</strong><br />
• <strong>of</strong>f center line measurements.<br />
11
Comparing results for source widths <strong>of</strong> 4 <strong>and</strong> 16 m yields, that rate constant estimation improves significantly for wider<br />
plumes, as it is less likely, that observation wells are placed <strong>of</strong>f the center line. Estimated rate constants also become<br />
more accurate when more than only local information on hydraulic conductivity is used to approximate the effective<br />
conductivity along the <strong>flow</strong> path: Taking Kef as the geometric mean <strong>of</strong> all measured K values from the monitoring<br />
network reduces rate constant overestimation, although wells are not necessarily located on the <strong>flow</strong> path or even within<br />
the contaminant plume.<br />
For the two-dimensional approach (strategy B), sources <strong>of</strong> error are inadequate parameterizations <strong>of</strong> longitudinal <strong>and</strong><br />
transverse dispersivities <strong>and</strong> an erroneous approximation <strong>of</strong> the mean <strong>flow</strong> velocity. As all observation wells<br />
downgradient <strong>of</strong> the source zone are incorporated in the estimation procedure, <strong>of</strong>f center line measurements are not<br />
relevant for this strategy. However, when the contaminant plume shows significant me<strong>and</strong>ering, the applicability <strong>of</strong><br />
strategy B is hampered, as the analytical model which is fitted to the concentration measurements presumes a linear<br />
plume axis. This problem is observed for the small source width <strong>of</strong> 4 m, where overestimation by strategy B is<br />
comparable to the one-dimensional approaches using the global estimate <strong>of</strong> Kef. For larger source widths like the 16 m<br />
used in this study, however, strategy B yields closer estimates <strong>of</strong> the degradation rate constant than strategy A on<br />
average.<br />
One drawback <strong>of</strong> the two-dimensional approach is that source concentration <strong>and</strong> degradation rate constant have to be<br />
fitted simultaneously to obtain accurate estimates <strong>of</strong> the degradation rate constant. This proves problematic especially for<br />
the small source width <strong>of</strong> 4 m where for some realizations the source concentration is overestimated by up to 300 %. For<br />
the source width <strong>of</strong> 16 m, however, the source concentration is fitted precisely within 5% <strong>of</strong> the “true” source<br />
concentration. These results suggest that incorporation <strong>of</strong> <strong>of</strong>f center line information in the estimation <strong>of</strong> the degradation<br />
rate can improve results <strong>of</strong> the plume investigation considerably. Especially for wide plumes, the two-dimensional<br />
approach <strong>of</strong> Stenback et al. (2004) proves superior to the one-dimensional center-line investigation strategy, resulting in<br />
more precise estimates <strong>of</strong> the degradation rate constant.<br />
Studying the variability <strong>of</strong> estimated degradation rates due to different monitoring network configurations, i.e. the<br />
personal factor due to different “site investigators”, it is found that for both source widths the variability <strong>of</strong> investigation<br />
results <strong>of</strong> one realization ranges from 25 % up to 69 % <strong>of</strong> the variability observed among different realizations. This<br />
demonstrates that the configuration <strong>of</strong> the monitoring network following from the investigators individual site<br />
investigation approach represents a substantial part <strong>of</strong> the uncertainty <strong>of</strong> the estimated rate constant.<br />
Overestimation <strong>of</strong> the degradation rate at a contaminated site is a critical point if monitored natural attenuation is<br />
considered as an alternative to conventional engineered remediation measures because the overall NA potential is<br />
assessed too positive. As demonstrated by Beyer et al. (2006) plume lengths calculated with biased degradation rates will<br />
result in an underestimation <strong>of</strong> the contaminated regions <strong>of</strong> the aquifer. Such an application, however, might not be <strong>of</strong><br />
primary concern, when monitoring networks <strong>of</strong> sufficient density have already been established over a site. Nonetheless,<br />
a too high rate constant could falsely lead to the conclusion, that a plume is at steady state, when the present day length<br />
observed fits the calculated steady state plume length. In the assessment <strong>of</strong> contaminated sites, indication for a steady<br />
state plume is an important result, e.g. for the acceptance <strong>of</strong> NA as a remediation scheme. Also for well investigated sites,<br />
a reliable determination <strong>of</strong> contaminant degradation rates is important, when the rates are to be used as <strong>modeling</strong><br />
parameters, for comparison with other sites <strong>and</strong> for comparing <strong>and</strong> optimizing alternatives <strong>of</strong> remediation measures.<br />
Acknowledgements: This work is funded by the German Ministry <strong>of</strong> Education <strong>and</strong> Research (BMBF) under grant<br />
033 05 12 / 033 05 13 as part <strong>of</strong> the KORA priority program, sub-project 7.2. We wish to thank our project partners at the<br />
Christian-Albrechts-University Kiel Andreas Dahmke <strong>and</strong> Dirk Schäfer for their support in our research. We<br />
acknowledge the support <strong>of</strong> Uwe Wittmann, Iris Bernhardt <strong>and</strong> Ludwig Luckner in coordination <strong>of</strong> the project work. Last<br />
but not least we are grateful to Bernie Kueper <strong>and</strong> two anonymous reviewers for their thoughtful comments <strong>and</strong><br />
suggestions which considerably improved this paper.<br />
References:<br />
Aronson, D., <strong>and</strong> P.H. Howard. 1997. Anaerobic biodegradation <strong>of</strong> organic chemicals in groundwater: A summary <strong>of</strong><br />
field <strong>and</strong> laboratory studies, SRC TR-97-0223F, Science Center Report, Syracuse Research Corporation, New York.<br />
Bauer, S., C. Beyer, <strong>and</strong> O. Kolditz. 2006a. Assessing measurement uncertainty <strong>of</strong> first-order degradation rates in<br />
heterogeneous aquifers. Water Resources Research 42, no. 1: W01420. 10.1029/2004WR003878.<br />
Bauer, S., C. Beyer, C. Chen, J. Gronewold, <strong>and</strong> O. Kolditz. 2006b. Virtueller Aquifer (VA) - Computergestützte<br />
Bewertung von Erkundungs-, Sanierungs- und Monitoringstrategien im Hinblick auf das "Natural Attenuation" (NA)<br />
und "Enhanced Natural Attenuation" (ENA) -Potenzial kontaminierter Böden und Grundwässer. In Statusseminar<br />
des KORA-TV 7, 8.6.2006, Dresden. Gemeinsame Mitteilungen des DGFZ e.V. und seiner Partner, vol. 3/2006, 93-<br />
113. Dresden.<br />
Bear, J. 1972. Dynamics <strong>of</strong> Fluids in porous Media. Amsterdam: Elsevier.<br />
Beyer, C., S. Bauer, <strong>and</strong> O. Kolditz. 2006. Uncertainty Assessment <strong>of</strong> Contaminant Plume Length Estimates in<br />
12
Heterogeneous Aquifers. Journal <strong>of</strong> Contaminant Hydrology 87, no. 1-2: 73-95. 10.1016/j.jconhyd.2006.04.006.<br />
Buscheck, T.E., <strong>and</strong> C.M. Alcantar. 1995. Regression techniques <strong>and</strong> analytical solutions to demonstrate intrinsic<br />
bioremediation. In Intrinsic Bioremediation, ed. R.E. Hinchee, T.J. Wilson, <strong>and</strong> D. Downey, 109-116. Columbus<br />
OH: Battelle Press.<br />
Chapelle, F.H., P.M. Bradley, D.R. Lovley, <strong>and</strong> D.A. Vroblesky. 1996. Measuring rates <strong>of</strong> biodegradation in a<br />
contaminated aquifer using field <strong>and</strong> laboratory methods. Ground Water 34, no. 4: 691-698.<br />
Chapelle, F.H., Widdowson, M.A., Brauner, J.S., Mendez, E. <strong>and</strong> C.C. Casey. 2003. Methodology for estimating times <strong>of</strong><br />
remediation associated with monitored natural attenuation. USGS Water-Resources Investigations Report 03-4057.<br />
Chen, C., C. Beyer, S. Bauer, <strong>and</strong> O. Kolditz. 2005. Interactive visual framework to demonstrate the uncertainty <strong>of</strong><br />
contaminant plume investigation. In 5. Workshop Porous Media, 02.-03.12.2004 Blaubeuren, Workshop<br />
Proceedings, Center for <strong>Applied</strong> Geosciences, University <strong>of</strong> Tübingen. http://www.virtual-aquifer.unituebingen.de/pdf/2005_29_chen.pdf<br />
Dagan, G. 1984. Solute transport in heterogeneous porous formations. Journal <strong>of</strong> Fluid Mechanics 145: 151–177.<br />
Domenico, P.A., <strong>and</strong> F.W. Schwartz. 1990. Physical <strong>and</strong> Chemical Hydrogeology, 2nd ed. New York: John Wiley <strong>and</strong><br />
Sons.<br />
Hsu, K.-C. 2003. The influence <strong>of</strong> the log-conductivity autocovariance structure on macrodispersion coefficients. Journal<br />
<strong>of</strong> Contaminant Hydrology 65, no. 1-2: 65-77. 10.1016/S0169-7722(02)00231-0.<br />
Islam J., S. Naresh, <strong>and</strong> M. O’Sullivan. 2001. Modeling biogeochemical processes in leachate-contaminated soils: A<br />
review. Transport in Porous Media 43, no. 3: 407 – 440. 10.1023/A:1010737825232.<br />
Kolditz, O., <strong>and</strong> S. Bauer, 2004. A process-oriented approach to computing multifield problems in porous media. Journal<br />
<strong>of</strong> Hydroinformatics 6, no. 3: 225-244.<br />
Kolditz, O., M. Beinhorn, M. Xie, T. Kalbacher, S. Bauer, W. Wang, C. McDermott, C. Chen, C. Beyer, J. Gronewold,<br />
D. Kemmler, R. Walsh, Y. Du, C.H. Park, M. Hess, <strong>and</strong> C. Büurger <strong>and</strong> J.O. Delfs. 2006. GeoSys/Rock<strong>flow</strong><br />
version 4.4.03 - Theory <strong>and</strong> users manual. Center for <strong>Applied</strong> Geoscience, University <strong>of</strong> Tübingen.<br />
McGrath, W.A., <strong>and</strong> G.F. Pinder. 2003. Search strategy for groundwater contaminant plume delineation. Water<br />
Resources Research 39, no. 10: 1298. 10.1029/2002WR001636.<br />
McNab Jr., W.W., <strong>and</strong> B.P. Dooher. 1998. A critique <strong>of</strong> a steady-state analytical method for estimating contaminant<br />
degradation rates. Ground Water 36, no. 6: 983–987.<br />
Newell, C.J., H.S. Rifai, J.T. Wilson, J.A. Connor, J.A. Aziz, <strong>and</strong> M.P. Suarez. 2002. Calculation <strong>and</strong> use <strong>of</strong> first-order<br />
rate constants for monitored natural attenuation studies. U.S. EPA Ground Water Issue, U.S. EPA/540/S-02/500.<br />
Rehfeldt, K.R., J.M. Boggs, <strong>and</strong> L.W. Gelhar. 1992. Field study <strong>of</strong> dispersion in a heterogeneous aquifer, 3, geostatistical<br />
analysis <strong>of</strong> hydraulic conductivity. Water Resources Research 28, no. 12: 3309–3324. 10.1029/92WR01758.<br />
Rittmann, B.E. 2004. Definition, objectives, <strong>and</strong> evaluation <strong>of</strong> natural attenuation. Biodegradation 15, no. 6: 349-357.<br />
10.1023/B:BIOD.0000044587.05189.99.<br />
Rittmann, B.E., <strong>and</strong> J.M. VanBriesen. 1996. Microbiological processes in reactive <strong>modeling</strong>. In Reactive Transport in<br />
Porous Media. Reviews in Mineralogy, vol. 34, ed. P. Lichtner, C. Steefel, <strong>and</strong> E. Oelkers, 311-334. Mineralogical<br />
Society <strong>of</strong> America: Washington DC.<br />
Rubin, Y., A. Bellin, <strong>and</strong> A.E. Lawrence. 2003. On the use <strong>of</strong> block-effective macrodispersion for <strong>numerical</strong> simulations<br />
<strong>of</strong> transport in heterogeneous formations. Water Resources Research 39, no. 9: 1242, 10.1029/2002WR001727.<br />
Stenback, G.A., S.K. Ong, S.W. Rogers, <strong>and</strong> B.H. Kjartanson. 2004. Impact <strong>of</strong> transverse <strong>and</strong> longitudinal dispersion on<br />
first-order degradation rate constant estimation. Journal <strong>of</strong> Contaminant Hydrology 73, no. 1-4: 3-14.<br />
10.1016/j.jconhyd.2003.11.004<br />
Storck, P., J.W. Eheart, <strong>and</strong> A.J. Valocchi. 1997. A method for the optimal location <strong>of</strong> monitoring wells for detection <strong>of</strong><br />
groundwater contamination in three-dimensional aquifers. Water Resources Research 33, no. 9: 2081–2088.<br />
10.1029/97WR01704.<br />
Sudicky, E.A.. 1986. A natural gradient experiment on solute transport in a s<strong>and</strong> aquifer: Spatial variability <strong>of</strong> hydraulic<br />
conductivity <strong>and</strong> its role in the dispersion process. Water Resources Research 22, no. 13: 2069-2082.<br />
Wiedemeier, T.H., M.A. Swanson, J.T. Wilson, D.H. Kampbell, R.N. Miller, <strong>and</strong> J.E. Hansen. 1996. Approximation <strong>of</strong><br />
biodegradation rate constants for monoaromatic hydrocarbons (BTEX) in ground water. Ground Water Monitoring<br />
& Remediation 16, no. 3: 186-194.<br />
Wiedemeier, T.H., H.S. Rifai, T.J. Wilson, <strong>and</strong> C. Newell. 1999. Natural Attenuation <strong>of</strong> Fuels <strong>and</strong> Chlorinated Solvents<br />
in the Subsurface. New York: Wiley.<br />
Wilson, R.D., S.F. Thornton, <strong>and</strong> D.M. Mackay. 2004. Challenges in monitoring the natural attenuation <strong>of</strong> spatially<br />
variable plumes. Biodegradation 15, no. 6: 459-469. 10.1023/B:BIOD.0000044591.45542.a9.<br />
Wu, J., C. Zheng, C.C. Chien, <strong>and</strong> L. Zheng. 2006. A comparative study <strong>of</strong> Monte Carlo simple genetic algorithm <strong>and</strong><br />
noisy genetic algorithm for cost-effective sampling network design under uncertainty. Advances in Water Resources<br />
29, no. 6: 899–911. 10.1016/j.advwatres.2005.08.005<br />
Zhang, Y.-K., <strong>and</strong> R.C. Heathcote. 2003. An improved method for estimation <strong>of</strong> biodegradation rate with field data.<br />
Ground Water Monitoring & Remediation 23, no. 3: 112-116.<br />
13
Enclosed Publication 6<br />
Beyer, C., Konrad, W., Park, C.H., Bauer, S., Rügner, H., Liedl, R., Grathwohl, P. (2007b):<br />
Modellbasierte Sickerwasserprognose für die Verwertung von Recycling-Baust<strong>of</strong>f in<br />
technischen Bauwerken. (Model based prognosis <strong>of</strong> contaminant leaching for reuse <strong>of</strong><br />
demolition waste in construction projects.) (accepted by Grundwasser <strong>and</strong> published online<br />
via SpringerLink), doi:10.1007/s00767-007-0025-x<br />
The enclosed article is made available with the permission <strong>of</strong> Springer <strong>and</strong> was published in<br />
the journal Grundwasser, online. Copyright © 2007 Springer.<br />
The article can be obtained online via SpringerLink at<br />
http://www.springerlink.com/openurl.asp?genre=journal&eissn=1432-1165.
Modellbasierte Sickerwasserprognose für die Verwertung von<br />
Recycling-Baust<strong>of</strong>f in technischen Bauwerken *<br />
Model based prognosis <strong>of</strong> contaminant leaching for reuse <strong>of</strong> demolition waste<br />
in construction projects<br />
Christ<strong>of</strong> Beyer 1 , Wilfried Konrad 1 , Hermann Rügner 2 , Sebastian Bauer 1 , Park Chan Hee 1 , Rudolf<br />
Liedl 3 , Peter Grathwohl 1<br />
1 Eberhard-Karls-Universität Tübingen, Zentrum für Angew<strong>and</strong>te Geowissenschaften (ZAG), Sigwartstraße 10, 72076<br />
Tübingen; Telefon: 07071-29 73176, Telefax: 07071-5059, E-mail: christ<strong>of</strong>.beyer@uni-tuebingen.de<br />
2<br />
Umweltforschungszentrum Leipzig-Halle (UFZ), Permoserstraße 15,<br />
04318 Leipzig<br />
3<br />
TU Dresden, Institut für Grundwasserwirtschaft, 01062 Dresden<br />
Header: Sickerwasserprognose für Recycling-Baust<strong>of</strong>f<br />
Kurzfassung:<br />
In dieser Studie wird die in der BBodSchV rechtlich etablierte Sickerwasserprognose in Bezug auf die<br />
Beurteilung der von Recycling-Baust<strong>of</strong>f-Verwertungen im Straßenbau ausgehenden Schadst<strong>of</strong>feinträge ins<br />
Grundwasser weiterentwickelt. Anh<strong>and</strong> numerischer reaktiver St<strong>of</strong>ftransportsimulationen für drei<br />
praxisrelevante Verwertungsszenarien (Parkplatz, Lärmschutzwall, Straßendamm) sowie eine Auswahl<br />
regionaltypischer Unterbodeneinheiten Deutschl<strong>and</strong>s werden zeitliche Konzentrationsverläufe verschiedener<br />
St<strong>of</strong>fklassen an der Grundwasseroberfläche berechnet. Der Durchbruchszeitpunkt konservativer Tracer wird<br />
allein von den hydraulischen Eigenschaften der Unterböden gesteuert, für organische Schadst<strong>of</strong>fe sind vor<br />
allem deren KOC-Werte und die Corg-Gehalte der Unterböden ausschlaggebend. Signifikante dispersive<br />
Konzentrationsverminderungen ergeben sich nur bei deutlicher Abnahme der Quellstärke vor dem<br />
Durchbruch der Konzentrationspeaks. Bei lang anhaltend hohen Quellkonzentrationen relativ zur<br />
Transportzeit bleiben die Konzentrationsdurchbrüche unvermindert. Biologischer Schadst<strong>of</strong>fabbau führt zu<br />
deutlich reduzierten Durchbruchskonzentrationen. Für die Szenarien Lärmschutzwall und Straßendamm<br />
werden Kapillarsperreneffekte beobachtet, die zu einem teilweisen Umfließen der Schadst<strong>of</strong>fquelle führen.<br />
Bei Berücksichtigung des am Recyclingmaterial vorbeiströmenden Sickerwassers durch Konzentrationsmittelung<br />
über die gesamte Bauwerksbreite ergeben sich Konzentrationsminderungen um 30-40%.<br />
Abstract:<br />
In this study contaminant leaching from recycling materials in road constructions to groundwater is assessed<br />
by the “Sickerwasserprognose”. Numerical transport simulations for three scenarios (parking lot, noise<br />
protection dam, road) <strong>and</strong> a number <strong>of</strong> characteristic subsoils <strong>of</strong> Germany are performed to estimate the<br />
breakthrough <strong>of</strong> different contaminant classes at the groundwater table. Conservative tracer breakthrough<br />
times (BTT) primarily depend on subsoil hydraulic properties, for organic pollutants Koc <strong>and</strong> subsoil OC are<br />
the controlling parameters. Significant concentration reductions from dispersion only occur when source<br />
concentrations decrease prior to contaminant breakthrough. If source concentrations remain high for long<br />
periods relative to peak BTT, concentration breakthrough is undamped. Accounting for biodegradation<br />
reduces breakthrough concentrations significantly. For the scenarios "noise protection dam" <strong>and</strong> "road" capillary<br />
barrier effects cause the seepage water to partially bypass the recycling material. Accounting for this<br />
bypass <strong>flow</strong> <strong>and</strong> averaging spatially across the constructions reduces concentrations by about 30-40%.<br />
Keywords: ground water risk assessment; reuse; demolition waste; type-scenarios; road construction;<br />
modelling<br />
* Beyer, C., Konrad, W., Park, C.H., Bauer, S., Rügner, H., Liedl, R., Grathwohl, P. (2007b): Modellbasierte Sickerwasserprognose für<br />
die Verwertung von Recycling-Baust<strong>of</strong>f in technischen Bauwerken. (Model based prognosis <strong>of</strong> contaminant leaching for reuse <strong>of</strong> demolition<br />
waste in construction projects.) (accepted by Grundwasser), doi:10.1007/s00767-007-0025-x<br />
Der Artikel wurde von der Zeitschrift Grundwasser zur Veröffentlichung angenommen, online publiziert und mit Erlaubnis von Springer<br />
reproduziert. Copyright © 2007 Springer. Der Artikel ist online abrufbar via SpringerLink:<br />
http://www.springerlink.com/openurl.asp?genre=journal&eissn=1432-1165<br />
1
Einleitung und Zielsetzung<br />
In Deutschl<strong>and</strong> fallen jährlich etwa 250 Mio. t mineralischer Abfälle an, die in erheblichem Umfang im Erd-,<br />
Straßen- und Verkehrsflächenbau verwertet werden. Recycling-Baust<strong>of</strong>fe und Rückstände aus der<br />
industriellen Produktion (z. B. Hoch<strong>of</strong>enschlacken) oder aus der Abfallbeh<strong>and</strong>lung (z. B. Hausmüllverbrennungsaschen)<br />
finden unter <strong>and</strong>erem in Trag- und Frostschutzschichten beim Straßenbau, zum Bau von<br />
Lärmschutzwällen oder als Verfüllmaterial zunehmende Verwendung (Krass et al. 2004a; 2004b). Die<br />
Verwertung von Recycling-Baust<strong>of</strong>fen (RCB) in technischen Bauwerken wird in Deutschl<strong>and</strong> in<br />
Übereinstimmung mit der diesbezüglichen Politik der Europäischen Union gegenüber einer Deponierung<br />
prinzipiell vorgezogen (KrW-/AbfG 1994). Aufgrund der häufig vorliegenden Belastung von RCB durch<br />
organische und anorganische Schadst<strong>of</strong>fe (z.B. PAK, Salze, Schwermetalle) muss bei der Verwertung in<br />
technischen Bauwerken jedoch berücksichtigt werden, dass Inhaltsst<strong>of</strong>fe durch das Sickerwasser<br />
ausgewaschen werden und das Grund- und Oberflächenwasser belasten können. Aus diesem Grund ist<br />
eine Bewertung von Verwertungsmaßnahmen hinsichtlich ihrer Umweltauswirkungen notwendig. Das<br />
Schadst<strong>of</strong>faustragsverhalten technischer Bauwerke ist deshalb ein sowohl auf nationaler als auch auf<br />
internationaler Ebene aktuelles und intensiv bearbeitetes Forschungsgebiet. So führten Hjelmar et al. (2007)<br />
großskalige Feldstudien an einem Straßenabschnitt in Dänemark durch und beobachteten ein zeitliches<br />
Abklingen der Schadst<strong>of</strong>fquellstärke, welches bei der Umweltwirkungsprognose berücksichtigt werden sollte.<br />
Für einige Schadst<strong>of</strong>fe ergäben sich so weniger konservativen Prognosen, die weniger restriktive<br />
Grenzwerte erlauben würden, ohne den Schutz des Grundwassers zu gefährden (Hjelmar et al. 2007). Die<br />
Autoren wiesen zudem das Auftreten von Kapillarsperren in Straßendämmen nach, die zu einem Umfließen<br />
des Verwertungsmaterials und somit zu reduzierten Wasserflüssen durch die Schadst<strong>of</strong>fquelle führten.<br />
Kapillarsperreneffekte wurden auch von Hansson et al. (2006) bei numerischen Simulationen der<br />
Wasserströmung in Straßendämmen beobachtet. Susset (2007) führte Freil<strong>and</strong>lysimeteruntersuchungen mit<br />
RCB über Löss- und geringsorptiven S<strong>and</strong>bodenmonolithen durch und stellte einen Rückhalt von PAK über<br />
bisher 4 Jahre nach, wobei die Wasserdurchsatzraten übertragen auf Feldbedingungen mehreren<br />
Jahrzehnten entsprechen. Sulfatkonzentrationen an der Unterkante des RCB gingen im<br />
Beobachtungszeitraum kaum zurück und brachen in voller Höhe durch. Für Chlorid hingegen konnte eine<br />
dispersive Verdünnung der Maximalkonzentration beobachtet werden, da die „Lebensdauer“ der<br />
Chloridquelle deutlich kleiner als die Transportzeit durch die Lysimeter war. In umfangreichen Säulenexperimenten<br />
mit RCB über Böden unter feldnahen Bedingungen zeigten Stieber et al. (2006), dass<br />
organische Kontaminanten wie PAK in der ungesättigten Zone effektiv durch mikrobiellen Abbau eliminierbar<br />
sind, s<strong>of</strong>ern die dafür physiologisch notwendigen Milieubedingungen nicht durch das Sickerwasser nachteilig<br />
verändert werden.<br />
Die Vielzahl der zu berücksichtigenden hydraulischen, geochemischen und mikrobiologischen Prozesse<br />
sowie die Komplexität ihrer Interaktionen macht deutlich, dass die Anwendung von prozessbasierten<br />
Modellen bei der Bewertung von umwelt<strong>of</strong>fen verwerteten Recyclingmaterialien große Vorteile bietet. Vor<br />
diesem Hintergrund untersucht diese Arbeit in Anlehnung an die Methodik der Sickerwasserprognose nach<br />
BBodSchV (1999) anh<strong>and</strong> numerischer reaktiver St<strong>of</strong>ftransportsimulationen, welche Schadst<strong>of</strong>feinträge von<br />
im Straßenbau eingesetztem RCB über die ungesättigte Zone ins Grundwasser unter Berücksichtigung<br />
möglicher zeitabhängiger Festlegungs- und Abbauprozesse (Sorption, Intrapartikeldiffusion, Bioabbau)<br />
erfolgen können. Dazu wird hier erstmals eine Kopplung des Stromröhrenmodells SMART (Finkel 1998,<br />
Finkel et al. 1998) mit dem Finite-Elemente-Modell GeoSys/Rock<strong>flow</strong> (Kolditz & Bauer 2004; Kolditz et al.<br />
2006) eingesetzt. Die Kombination verschiedener repräsentativer Verwertungsszenarien (Parkplatz,<br />
Lärmschutzwall, Straßendamm), RCB-typischer Schadst<strong>of</strong>fklassen und regionaltypischer Unterböden führt<br />
zu einem umfangreichen Satz von Typ-Szenarien-Simulationen, deren Ergebnisse es erlauben,<br />
• den aus den Verwertungsszenarien zu erwartenden charakteristischen zeitlichen Verlauf der<br />
Schadst<strong>of</strong>feinträge in Boden und Grundwasser darzustellen,<br />
• die während des Transports für Schadst<strong>of</strong>frückhalt und Konzentrationsminderung relevanten<br />
Prozesse zu identifizieren,<br />
• Einflüsse der Unterbodeneigenschaften hinsichtlich des Filter- und Puffervermögens herauszustellen<br />
und<br />
• das Potential der modellbasierten Sickerwasserprognose praxisnah zu demonstrieren.<br />
Modellkonzept<br />
Die Berechnung der Schadst<strong>of</strong>feinträge aus dem RCB über die ungesättigte Zone ins Grundwasser wird in<br />
dieser Studie erstmalig mit Hilfe einer Kopplung zwischen dem Finite-Elemente-Modell GeoSys/Rock<strong>flow</strong><br />
2
und dem stochastischen Stromröhrenmodell SMART durchgeführt. Die Simulationsaufgabe wird hier in die<br />
zwei Teilprobleme Hydraulik/Strömung und reaktiver Transport zerlegt. Diese Vorgehensweise wurde<br />
gewählt, um die Auswirkung der räumlich variablen Sickerwasserströmung in den Verwertungsszenarien auf<br />
den Schadst<strong>of</strong>ftransport berücksichtigen sowie physikalisch realistische Schadst<strong>of</strong>ffreisetzungsprozesse in<br />
der Quelle und die durch Intrapartikeldiffusion zeitabhängige Sorption beim Transport quantifizieren zu<br />
können.<br />
Die Kopplung zwischen der Strömungssimulation mit GeoSys/Rock<strong>flow</strong> und der reaktiven<br />
Transportsimulation mit SMART erfolgt über Verteilungsfunktionen (pdf, "probability density function") der<br />
Verweilzeiten des Sickerwassers, welche die Einflüsse der hydraulischen Heterogenität auf das<br />
Strömungsfeld beschreiben (vgl. Utermann et al. 1990; Bold 2004). Die pdf sind Modellergebnis der<br />
Strömungssimulation und werden von SMART als Modelleingabe benötigt, um in der reaktiven<br />
Transportsimulation Massendurchbruchskurven an einer unterstromigen Kontrollebene (hier die<br />
Grundwasseroberfläche) zu berechnen. Da das SMART-Konzept zwar Transportsimulationen in heterogen<br />
zusammengesetzten Materialien erlaubt (Lithokomponenten-Ansatz nach Kleineidam et al. 1999a), entlang<br />
der Stromröhre jedoch eine homogene Materialverteilung vorausgesetzt wird, muss für jede Material- bzw.<br />
Modellschicht der aus mehreren Baumaterialien und Unterbodenhorizonten bestehenden Typ-Szenarien<br />
eine separate SMART-Simulation gerechnet werden. Das Modell ist somit als eine Reihe von hinterein<strong>and</strong>er<br />
geschalteten 1D-Stromröhren konzipiert (siehe Abb. 1).<br />
GeoSys/Rock<strong>flow</strong> SMART<br />
35cm<br />
Charakterisierung<br />
der Strömung<br />
konst. Infiltration<br />
Tragschicht<br />
RCB<br />
Körnung<br />
0/32<br />
Unterboden<br />
Untergrund<br />
Grundwasseroberfläche<br />
(GWO)<br />
Tracereingabe<br />
BTC Tragschicht<br />
Tracereingabe<br />
BTC Unterboden<br />
Tracereingabe<br />
BTC Untergrund<br />
pdf Tragschicht<br />
pdf Unterboden<br />
pdf Untergrund<br />
3<br />
reaktiver<br />
Transport<br />
Stromröhre<br />
Tragschicht<br />
Stromröhre<br />
Untergrund<br />
Modellergebnis<br />
BTC an GWO<br />
Abb. 1: Veranschaulichung des Modellkonzepts anh<strong>and</strong> des Parkplatzszenarios: Simulation der<br />
Tracerdurchbruchskurven zur Ableitung von Laufzeitverteilungen (pdf) der einzelnen Modellschichten mit<br />
GeoSys/Rock<strong>flow</strong>; reaktive Transportsimulationen mit SMART (BTC = Durchbruchskurve); Kopplung beider<br />
Modelle durch die pdf.<br />
Die modellierte Massendurchbruchskurve an der unterstromigen Kontrollebene jeder Stromröhre<br />
repräsentiert jeweils die oberstromige zeitabhängige Konzentrationsr<strong>and</strong>bedingung der nachfolgenden<br />
Stromröhre bzw. im Fall der untersten Modellschicht den Schadst<strong>of</strong>feintrag ins Grundwasser.<br />
Dementsprechend muss für jede Modellschicht eine eigene pdf abgeleitet werden. Aus Gründen der<br />
Modellvereinfachung werden die einzelnen pdf als unabhängig vonein<strong>and</strong>er betrachtet (vgl. V<strong>and</strong>erborght et<br />
al. 2007). Zur Ableitung der pdf wird in GeoSys/Rock<strong>flow</strong> für jede einzelne Materialschicht an der<br />
Schichtobergrenze ein konservativer Tracer mit konstanter Konzentration C0 in das Strömungsfeld<br />
eingegeben und dessen Durchbruch C(t) an der Schichtuntergrenze zeitlich registriert (siehe Abb. 1). Die pdf<br />
der Modellschicht ergibt sich als Ableitung der Durchbruchskurve dC/dt aufgetragen über die Zeit t.<br />
Die Kopplung beider Simulationsmodelle wurde als sogenannte lose Kopplung realisiert, d.h. die Quellcodes<br />
wurden nicht kombiniert sondern kommunizieren über Batch-Aufrufe und automatisch generierte Eingabe-<br />
/Ausgabedateien während bzw. nach der Programmausführung. Diese Art der Einbindung hat im Hinblick auf
die Anzahl Simulationsläufe den Vorteil, dass pro Kombination aus Verwertungsmaßnahme und<br />
Unterbodentyp nur eine Simulation des Strömungsfeldes zur Bestimmung aller pdf notwendig ist und diese<br />
jeweils für die darauf folgenden Transportsimulationen der einzelnen Schadst<strong>of</strong>fklassen mit SMART<br />
verwendet werden können.<br />
Numerische Simulation der Strömung und des reaktiven Transports<br />
Die Simulation der ungesättigten Strömung mit GeoSys/Rock<strong>flow</strong> zur Ableitung der pdf erfolgt auf Grundlage<br />
der Richards-Gleichung, zu deren Lösung das Van-Genuchten-Mualem-Modell (van Genuchten 1980;<br />
Mualem 1976) verwendet wird (Du et al. 2005). Auf eine umfassende Ausführung der allgemein bekannten<br />
Modellgleichung wird hier und im Folgenden verzichtet, eine ausführliche Erläuterung der verwendeten<br />
mathematischen Modelle ist in Grathwohl et al. (2006) zu finden. Die Berechnung der Tracerbewegung im<br />
ungesättigten Strömungsfeld zur Ableitung der pdf erfolgt auf Grundlage der Konvektions-<br />
Dispersionsgleichung. Die Beschreibung des reaktiven Transports mit SMART erfolgt dagegen wie oben<br />
beschrieben nicht entlang von Raumkoordinaten sondern in Abhängigkeit der Verweilzeit entlang einer<br />
Stromröhre (Finkel, 1998). Die Rückhalteprozesse beim Transport werden durch das in SMART integrierte<br />
numerische Modell BESSY (Jäger & Liedl 2000) quantifiziert. Die Berücksichtigung der den<br />
Sorptionsprozess zeitlich limitierenden Intrapartikel-Porendiffusion ist insbesondere für Böden mit hohem<br />
Kies- oder Skelettanteil notwendig (Grathwohl 1998; Rügner et al. 1997; 1999). Die Sorptionskinetik wird<br />
unter Annahme von sphärischen Partikeln durch das 2. Fick´sche Gesetz in Radialkoordinaten beschrieben:<br />
2<br />
∂ w ⎡∂<br />
Cw<br />
2 ∂ C<br />
= Da⎢<br />
+ 2<br />
∂ t ⎣ ∂ r r ∂ r<br />
C w<br />
r [m] bezeichnet den radialen Abst<strong>and</strong> vom Kornmittelpunkt, CW [kg m -3 ] die Konzentration in der wässrigen<br />
Phase der Intrapartikelporen und Da [cm 2 s -1 ] den scheinbaren Diffusionskoeffizienten, der gegenüber der<br />
Diffusion in Wasser (Daq) vermindert ist:<br />
D<br />
D<br />
ε<br />
4<br />
⎤<br />
⎥<br />
⎦<br />
(1)<br />
aq<br />
a = (2)<br />
( ε + Kd<br />
ρk<br />
) τ f<br />
ε [-] bezeichnet die Intrapartikelporosität, τf [-] den Tortuositätsfaktor, Kd [m -3 kg -1 ] den Gleichgewichts-<br />
Verteilungskoeffizienten und ρk [kg m -3 ] die Partikeldichte gemäß ρk = ( 1−<br />
ε )ρ mit ρ [kg m -3 ] der<br />
Mineraldichte.<br />
Die Verteilung der Schadst<strong>of</strong>fe zwischen dem in den Intrapartikelporen gelösten und dem sorbierten Anteil<br />
wird durch eine lineare Sorptionsisotherme beschrieben, für den Schadst<strong>of</strong>fabbau im mobilen Porenwasser<br />
wird eine Kinetik erster Ordnung angenommen. Somit ergibt sich unter stationären Fließbedingungen ein<br />
insgesamt lineares mathematisches Modell für die im Sickerwasser auftretenden Konzentrationen C und<br />
eine Proportionalität zwischen C und den anfänglich aus der Schadst<strong>of</strong>fquelle eluierenden Schadst<strong>of</strong>fkonzentrationen<br />
C0, bzw. den anfanglich sorbierten St<strong>of</strong>fmengen.<br />
Parametrisierung der Typ-Szenarien<br />
Im Rahmen dieser Studie werden drei repräsentative Verwertungsszenarien betrachtet: Parkplatz (PP),<br />
Lärmschutzwall (LSW) und Straßendamm (SD). Das in diesen zum Einsatz kommende Verwertungsmaterial<br />
besteht aus granularem RCB. Als Schadst<strong>of</strong>fe werden vier Modellsubstanzen verwendet, die in ihren<br />
Eigenschaften für RCB typische Schadst<strong>of</strong>fe repräsentieren: Naphthalin (NAP) und Phenanthren (PHE) als<br />
schwach bzw. mäßig sorbierende organische Schadst<strong>of</strong>fe, ein stark sorbierender für das durchschnittliche<br />
Verhalten der 15 EPA-PAK repräsentativer Summenparameter (Σ15 EPA-PAK) sowie ein konservativer<br />
Tracer als Beispiel leicht löslicher Salze wie Chlorid. Für die Transportstrecken unterhalb der Quellterme<br />
werden sechs regionaltypische Unterbodeneinheiten Deutschl<strong>and</strong>s berücksichtigt, um Charakteristika und<br />
Unterschiede hinsichtlich ihrer Filter- und Pufferkapazitäten herauszustellen. Die folgenden Abschnitte<br />
erläutern die Parametrisierung der Typ-Szenarien (für detailliertere Ausführungen siehe Grathwohl et al.<br />
2006).
Verwertungsszenarien und -materialien<br />
Das PP-Szenario wird als vertikales 1D-Modell betrachtet und orientiert sich am Querschnitt für<br />
Verkehrsflächen der Bauklasse VI mit Pflasterdecke und ungebundener Tragschicht auf einer<br />
Frostschutzschicht (FSS) nach FGSV (2001). Für Tragschicht und FSS, die im Modell als einheitliche<br />
Schicht von 0.35 m betrachtet werden, wird RCB mit einer Körnung der Sieblinie 0/32 (FGSV 2004)<br />
eingesetzt. Das Pflaster selbst wird nicht berücksichtigt, da die geforderte Versickerungsleistung von 2.7*10 -5<br />
m s -1 (FGSV 1998) deutlich über den hier angenommenen Infiltrationsraten liegt (s.u.). Abb. 1 zeigt neben<br />
dem Modellkonzept auch den für die Modellierung vereinfachten PP-Aufbau.<br />
Die Szenarien LSW und SD werden als 2D-Vertikalschnitte betrachtet, bei denen die Simulationen aus<br />
Symmetriegründen nur für jeweils eine Hälfte des Querschnitts durchgeführt wurden. Das LSW-Modell (Abb.<br />
2) orientiert sich an einem von Mesters (1993) experimentell untersuchten LSW. Als Bodenabdeckung des<br />
Wallkerns aus RCB wird ein Lehmboden verwendet.<br />
0.5m<br />
3.5m<br />
1m 6m<br />
2m<br />
RCB<br />
Körnung 0/32<br />
Unterboden<br />
Untergrund<br />
Symmetrieachse<br />
1:1. 5<br />
Planum<br />
Abb. 2: Lärmschutzwall mit Wallkern aus Recycling-Baust<strong>of</strong>f (RCB).<br />
0.5m<br />
5<br />
Bodenabdeckung<br />
kulturfähiger<br />
Lehmboden<br />
Grundwasseroberfläche<br />
(GWO)<br />
Das SD-Modell (Abb. 3) orientiert sich am Regelquerschnitt RQ 26 für vierstreifige Autobahnen (FGSV<br />
1996). Das Bankett wird nach FGSV (2005) als schwach durchlässiger Boden SŪ* ausgeführt (entspr.<br />
Bodenart Su3), die Bodenabdeckung der Böschung als stark durchlässiger S<strong>and</strong> (entspr. Bodenart Ss). Die<br />
hydraulischen Eigenschaften der für Bankett und Böschung verwendeten Böden sind in Tab. 1 aufgeführt.<br />
Unterhalb der Asphaltdeckschicht schließt sich eine ungebundene Tragschicht an, für die wie für den<br />
darunter liegenden Dammkern RCB der Sieblinie 0/32 angenommen wird.<br />
1.3m<br />
0.3m<br />
RCB<br />
Körnung 0/32<br />
Unterboden<br />
Untergrund<br />
Symmetrieachse<br />
10m 1.5m<br />
Asphaltdeckschicht<br />
undurchlässig<br />
3%<br />
Planum 4%<br />
12%<br />
schwach durchlässiger<br />
Boden<br />
1:1. 5<br />
2.3m<br />
10 cm stark durchlässiger<br />
Boden<br />
2m<br />
Grundwasseroberfläche<br />
(GWO)<br />
Abb. 3: Straßendamm mit Frostschutzschicht/Tragschicht aus Recycling-Baust<strong>of</strong>f (RCB).<br />
Zu den ungesättigten hydraulischen Eigenschaften von Recyclingmaterialien im Straßenbau finden sich in<br />
der Literatur kaum experimentelle Angaben. Deshalb wurden die Van-Genuchten-Parameter für den RCB in<br />
den einzelnen Verwertungsszenarien mit einem Ansatz von Arya & Paris (1981) bzw. Mishra et al. (1989) auf
Grundlage des Korngrößenspektrums der 0/32-Siebline, der Lagerungsdichte ρb = ( V ⋅ ρ p ) und der<br />
Porosität η = ( 1−<br />
ρb<br />
/ ρ)<br />
abgeleitet. V [-] ist der Verdichtungsgrad und ρp die Proctordichte [g cm -3 ] (siehe<br />
Tab. 1). Dieser Ansatz wurde für natürliche Böden entwickelt, sollte nach Hansson et al. (2006) jedoch auch<br />
für relativ grobkörnige Materialien wie Tragschichtschotter im Straßenbau geeignet sein. Für die gesättigte<br />
Leitfähigkeit Ks von Tragschichten wird ein Mindestwert von 5.4*10 -5 m s -1 gefordert (FGSV 1998). Aufgrund<br />
der großen Spannbreite experimentell in Labor und in-situ bestimmter Durchlässigkeiten (z.B. Wörner et al.<br />
2001; Stoppka 2002; Kellermann 2003) wird dieser Wert hier in allen Verwertungsszenarien für den RCB<br />
angenommen. Abb. 4 zeigt die θ-ψ- sowie die θ-K-Beziehungen des RCB für PP, LSW und SD. Die<br />
Kurvenverläufe sind durchweg sehr ähnlich und durch sehr geringe Kapillarität gekennzeichnet, was zu einer<br />
schnellen Drainage der Tragschichten erforderlich ist.<br />
Tab. 1: Hydraulische Eigenschaften der Baumaterialien für Parkplatz, Lärmschutzwall<br />
und Straßendamm<br />
Parkplatz Lärmschutzwall<br />
RCB RCB Böschung ‡<br />
6<br />
Straßendamm<br />
RCB Bankett † Böschung †<br />
θs 0.27 0.31 0.43 0.25 0.36 0.37<br />
θr 0.00 0.00 0.08 0.00 0.00 0.04<br />
n 1.29 1.29 1.56 1.29 1.28 1.57<br />
α [m -1 ] 166 154 3.60 146 2.64 8.74<br />
l 0.50 0.50 0.50 0.50 0.50 0.50<br />
ρ [g cm -3 ] ¶ 2.66 2.66 2.65 2.66 2.65 2.65<br />
ρb [g cm -3 ] 1.94 1.84 1.51 2.00 1.69 1.67<br />
ρp [g cm -3 ] ¶ 1.94 1.94 - 1.94 - -<br />
V [-] § 1 0.95 - 1.03 - -<br />
Ks [m s -1 ] 5.40*10 -5 5.40*10 -5 1.00*10 -6 5.40*10 -5 1.00*10 -6 2.90*10 -6<br />
‡ : Carsell & Parish 1988; † : Hennings 2000;<br />
§ : FGSV 2002 (PP), 1997 (LSW), 2004 (SD); ¶ : Kellermann 2003<br />
α, n, l: empirische Van-Genuchten-Parameter<br />
θr, θs: residualer und gesättigter Wassergehalt<br />
ρ, ρb, ρp: Mineral-, Lagerungs- und Proctordichte<br />
V: Verdichtungsgrad; Ks: gesättigte hydraulische Leitfähigkeit<br />
ψ [m]<br />
10 2<br />
10 1<br />
10 0<br />
10 -1<br />
10 -2<br />
10 -3<br />
10 -4<br />
Parkplatz<br />
Lärmschutzwall<br />
Straßendamm<br />
0 0.1 0.2 0.3 0.4<br />
θ [-]<br />
Krel [-]<br />
10 0<br />
10 -1<br />
10 -2<br />
10 -3<br />
10 -4<br />
10 -5<br />
10 -6<br />
10 -7<br />
10 -8<br />
10 -9<br />
10 -10<br />
0 0.1 0.2 0.3 0.4<br />
θ [-]<br />
Abb. 4: Wassergehalts-SAugspannungs- (links) sowie Wassergehalts-Leitfähigkeits-Beziehungen (rechts) für<br />
Recycling-Baust<strong>of</strong>f in den drei Verwertungsszenarien (ψ = Matrixpotential, Krel = relative Leitfähigkeit, θ =<br />
Wassergehalt).
Die Kd-Werte von NAP, PHE und Σ15 EPA-PAK im RCB von 106, 496 bzw. 1333 l kg -1 sowie die<br />
Intrapartikelporosität ε = 0.015 wurden von Henzler (2004) experimentell bestimmt. Daq für PHE und NAP<br />
wurde mit 7.86*10 -10 bzw. 9.15*10 -10 m² s -1 nach Hayduk & Laudie (1974) abgeschätzt. Die heterogene<br />
Zusammensetzung des RCB aus verschiedenen Korngrößenfraktionen wurde durch den Lithokomponenten-<br />
Ansatz nach Kleineidam et al. (1999a) berücksichtigt. Wegen sehr langer Rechenzeiten wird der RCB durch<br />
zwei Korngrößenklassen modelliert (siehe auch Rügner et al. 2005), für die die kinetische Sorption jeweils<br />
separat berechnet wird. Auf die Fein- und Grobfraktion entfallen Anteile von 32.8 % bzw. 67.2 % mit<br />
„effektiven“ Kornradien von a = 0.25 mm bzw. a = 8 mm (siehe Henzler (2004)).<br />
Auswahl und Klassifizierung der Unterbodenpr<strong>of</strong>ile<br />
Die Charakterisierung der Unterbodenpr<strong>of</strong>ile erfolgte auf Basis der nutzungsdifferenzierten Bodenübersichtskarte<br />
1:1.000.000 (BÜK1000; BGR 2006). Aus deren 672 Referenzpr<strong>of</strong>ilen wurden sechs Pr<strong>of</strong>ile nach<br />
Flächenrepräsentanz und dem zu erwartenden charakteristischen Transportverhalten für die numerischen<br />
Simulationen des reaktiven St<strong>of</strong>ftransports ausgewählt (Braunerde und Podsol aus S<strong>and</strong>, Fahlerde aus<br />
Geschiebelehm, Schwarzerde und Parabraunerde aus Löss, Pelosol aus verwittertem Mergel- und<br />
Tonstein). Zur Vereinfachung der Simulationen wurde eine Reduktion der vertikalen Pr<strong>of</strong>ildifferenzierung<br />
durch Zusammenfassung mehrerer Horizonte vorgenommen, soweit eine einheitliche Betrachtung nach<br />
bodenkundlichen und geologischen Aspekten möglich schien. Bei allen Pr<strong>of</strong>ilen wurden somit Daten aus<br />
verschiedenen Horizonten nach deren Mächtigkeit gewichtet gemittelt (Tab. 2). Als zu berechnende<br />
Pr<strong>of</strong>iltiefe wurde für die Böden die in der BÜK1000 (BGR 2006) beschriebene Pr<strong>of</strong>iltiefe ohne den<br />
Oberboden angesetzt, da dieser bei der Bebauung abgetragen wird. Die resultierenden Pr<strong>of</strong>iltiefen (Tab. 2)<br />
sind im Sinne von Mindesttiefen zu verstehen, da die in der Praxis relevanten Grundwasserflurabstände<br />
häufig deutlich größer sein dürften.<br />
Zur Festlegung der Ks-Werte und Van-Genuchten-Parameter wurden Pedotransferfunktionen von Wösten et<br />
al. (1998) angewendet. Der Sättigungswassergehalt θs wurde für alle Bodenschichten der Porosität<br />
gleichgesetzt. Ks bezieht sich auf den Feinbodenanteil (< 2 mm) und wurden bei höheren Skelettgehalten um<br />
die Reduktion der Leitfähigkeit durch den Skelettanteil mit einem Verfahren von Brakensiek & Rawls (1994)<br />
korrigiert. Für die Simulation der Tracerversuche in GeoSys/Rock<strong>flow</strong> zur Ableitung der pdf wurden die<br />
Dispersivitäten mit αL = 0.1 m und αT = 0.01 m festgelegt.<br />
Die Kd–Werte wurden aus dem Gehalt an organischem Kohlenst<strong>of</strong>f foc [-] (= 0.01 Corg; vgl. Tab. 2) und dem<br />
K = K f abgeschätzt. Für<br />
auf den Corg-Gehalt normierten Verteilungskoeffizienten KOC [l kg -1 ] durch d OC oc<br />
Horizonte, die nach BÜK1000 (BGR 2006) als „humusfrei“ ausgewiesen sind, wurde als konservative<br />
Annahme für den Feinboden ein minimaler Corg von 0.01 % angenommen. Zur Abschätzung des KOC wurde<br />
die auf der Wasserlöslichkeit S [mol l -1 ] beruhende Korrelation von Seth et al. (1999) verwendet (für<br />
vergleichbare Ansätze siehe auch Allen-King et al. 2002):<br />
log KOC = − 0.<br />
88 log S + 0.<br />
07 (3)<br />
S beträgt für PHE 3.46*10 -5 , für NAP 8.74*10 -4 mol l -1 . Für Σ15 EPA-PAK wurde S über ein mittleres<br />
Molgewicht von 202 g mol −1 und einer effektiven Wasserlöslichkeit von 2.5 g l -1 (Grathwohl 2004) mit 1.14<br />
*10 -5 mol l -1 abgeschätzt. Für die verschiedenen Böden ergeben sich so Kd-Werte für NAP zwischen 0.06<br />
und 5.10 l kg -1 , für PHE zwischen 0.99 und 87.2 l kg -1 sowie für Σ15 EPA-PAK zwischen 2.45 und 217 l kg -1<br />
Für die Skelettanteile wurden KOC-Werte von 4.0 l kg -1 (NAP), 5.2 l kg -1 (PHE) und 5.7 l kg -1 (Σ15 EPA-PAK)<br />
angenommen (Rügner et al. 2005). Mit diesen vergleichsweise hohen Werten wird berücksichtigt, dass es<br />
sich beim Corg dieser Komponenten idR. um gealtertes Material mit höherer Sorptionskapazität h<strong>and</strong>elt<br />
(Kleineidam et al. 1999b). Der foc wurde mit 0.0005 angesetzt (Kleineidam et al. 1999b).<br />
Die Sorptionskapazität und -kinetik für den Feinboden wird vor allem durch das partikuläre organische<br />
Material bestimmt. Es wurden folgende effektive Parameter zugrunde gelegt: a = 11.7 µm, ε = 0.00175,<br />
ρ = 2.65 g cm -3 und die Tortuosität τ f = 1/<br />
ε (Grathwohl, 1992; Rügner et al. 1999). Für die Grobfraktion (> 2<br />
mm) wurden a = 1 cm und ε = 0.01 angenommen (Rügner et al. 1999; Kleineidam et al. 1999a).<br />
Die Transportsimulationen für NAP, PHE und Σ15 EPA-PAK wurden jeweils mit und ohne Berücksichtigung<br />
von Bioabbau durchgeführt, wobei nur die in Lösung vorliegenden Anteile als abbaubar betrachtet wurden.<br />
Zur Beurteilung der langfristigen Filterwirkung des Bodens bedarf es repräsentativer Langzeit-<br />
Ratenkonstanten (Henzler et al. 2006), deren quantitative Abschätzung in der ungesättigten Zone jedoch<br />
durch die Komplexität der Wechselwirkungen zwischen schadst<strong>of</strong>fspezifischen Eigenschaften, klimatischen<br />
und geochemischen R<strong>and</strong>bedingungen (Grathwohl et al. 2003, Höhener et al. 2006) bisher kaum möglich<br />
ist. Aus diesem Grund wurde ein relativ niedriger Wert von 1.15*10 -7 s -1 (Halbwertzeit = 70 d) angenommen.<br />
7
Dies ist als konservative Abschätzung zu betrachten, da viele Schadst<strong>of</strong>fe in der ungesättigten Zone unter<br />
Feldbedingungen deutlich schneller abgebaut werden können (Maier & Grathwohl 2005, Rügner et al. 2005).<br />
Tab. 2: Bodeneigenschaften der sechs betrachteten Unterböden aus der BÜK1000 (BGR 2006).<br />
Für vertikal aggregierte Horizonte wurden die Parameter über die Mächtigkeit der einzelnen<br />
Horizonte gewichtet gemittelt.<br />
Bodentyp Braunerde Podsol Fahlerde<br />
8<br />
Schwarzerde <br />
Parabraunerde<br />
Pelosol<br />
Fläche [km²] 1672 7237 6933 3455 14428 5614<br />
Einteilung Unterboden Unterboden Unterboden Untergrund Unterboden Unterboden Unterboden<br />
von - bis [m] 0.00-1.70 0.00-1.70 0.00-0.90 0.90-1.70 0.00-1.70 0.00-1.70 0.00-1.75<br />
Ton [%] 3.82 10.40 31.85 31.85 19.69 19.78 39.88<br />
Schluff [%] 14.54 20.48 20.48 20.48 69.40 70.28 37.00<br />
S<strong>and</strong> [%] 57.41 63.21 38.68 38.68 8.44 8.43 16.48<br />
Skelett [%] 24.24 5.91 9.00 9.00 2.47 1.50 6.64<br />
Corg [%] 0.01 ‡ 0.21 0.01 ‡ 0.01 ‡ 0.88 0.01 ‡ 0.03<br />
ρb [g cm -3 ] 1.71 1.61 1.62 1.75 1.42 1.63 1.60<br />
θs 0.29 0.36 0.35 0.34 0.45 0.38 0.39<br />
†<br />
θr<br />
0.01 0.01 0.01 0.01 0.01 0.01 0.01<br />
n † 1.34 1.26 1.17 1.08 1.15 1.13 1.16<br />
α † [m -1 ] 5.93 5.98 6.21 4.95 1.76 1.24 2.73<br />
l † 1.77 -0.30 -0.66 -3.57 -1.91 -0.32 -2.75<br />
Ks † [m s -1 ] 1.52*10 -6 3.37*10 -6 3.09*10 -6 8.46*10 -7 3.83*10 -6 1.38*10 -6 9.48*10 -7<br />
‡ : 0.00 nach BÜK1000; Annahme eines minimalen Corg-Gehaltes des Feinbodens von 0.01%<br />
† : abgeleitet nach Wösten et al. (1998); Ks korrigiert um Skelettanteil nach Brakensieck & Rawls (1994)<br />
θr, θs: residualer und gesättigter Wassergehalt; ρb: Lagerungsdichte<br />
α, n, l: empirische Van-Genuchten-Parameter; Ks: gesättigte hydraulische Leitfähigkeit<br />
Abschätzung der Sickerwasserrate<br />
Für eine langfristige Beurteilung der Schadst<strong>of</strong>fverlagerung ist es sinnvoll und zulässig von vereinfachten<br />
Fließverhältnissen auszugehen (Grathwohl & Susset 2001; Henzler et al. 2006). Aus diesem Grund wird eine<br />
stationäre ungesättigte Strömung angenommen. Die jährlichen mittleren Infiltrationsraten I [mm a -1 ] der<br />
Verwertungsszenarien wurden als Mittelwerte für Deutschl<strong>and</strong> abgeschätzt. Hierzu wurde das BAGLUVA-<br />
Verfahren (Glugla et al. 2003) angewendet. Als mittlerer korrigierter Niederschlag Njahr wurde ein Wert von<br />
859 mm a −1 angenommen (BMU 2000). I ergibt sich als Differenz von Njahr und der Jahressumme der<br />
Evapotranspiration ET. Diese wird als Funktion der maximalen ET und st<strong>and</strong>ortspezifischer Parameter<br />
bestimmt. Für den PP wurde I so mit 583 mm a -1 abgeschätzt, was im oberen Bereich für sickerfähige<br />
Pflasterdecken liegt (Richter 2003). Die Annahme wird jedoch durch Freil<strong>and</strong>lysimeter-Untersuchungen von<br />
Flöter (2006) gestützt, in denen der Anteil von I am Niederschlag über 4 a im Mittel rund 80 % betrug. Für<br />
den LSW wurde I mit 313 mm a -1 abgeschätzt. Für den SD ergibt sich die zu infiltrierende Wassermenge als<br />
Summe des Oberflächenabflusses der Asphaltdecke (Abflussbeiwert = 0.9; FGSV 2005) und des auf<br />
Bankett und Böschung fallenden Niederschlags abzüglich der ET. Für die räumliche Verteilung der Infiltration<br />
mussten vereinfachende Annahmen getr<strong>of</strong>fen werden, da der SD-Querschnitt eine neue Bauweise nach<br />
FGSV (2005) darstellt, bei der der Straßenabfluss zu größeren Teilen über die Böschung infiltrieren soll,<br />
bisher jedoch keine experimentellen Daten dazu vorliegen. Aus diesem Grund wurde angenommen, dass die<br />
Versickerung gleichmäßig über Bankett und Böschung erfolgt. Auf den Querschnitt bezogen ergibt sich I mit<br />
2318 mm a -1 . Für den Böschungsfuß wurden wie für den LSW I = 313 mm a -1 angenommen.
Ergebnisse und Diskussion<br />
Die im Folgenden vorgestellten zeitlichen Konzentrationsverläufe der Modellsubstanzen an der GWO stellen<br />
über den unteren Modellr<strong>and</strong> integrierte Durchbruchskurven dar. Dabei ist für das LSW- und SD-Szenario<br />
zwischen den mit SMART berechneten, auf den Transportpfad bezogenen (graue dicke Kurven, Abb. 5, 8, 9)<br />
und den über den gesamten Querschnitt gemittelten Konzentrationen (schwarze dünne Kurven, Abb. 5, 8, 9)<br />
zu unterscheiden (siehe auch Abb. 6).<br />
Die Tracer-Durchbruchskurven an der GWO für die drei Szenarien PP, LSW und SD sind in Abb. 5 (a)-(c)<br />
dargestellt. Für den PP erfolgt der Durchbruch des Tracers am frühesten bei Braunerde und Podsol, später<br />
bei Pelosol, Parabraunerde und Schwarzerde (Abb. 5 (a)), die eine geringere<br />
Porenwasserfließgeschwindigkeit als die S<strong>and</strong>böden aufweisen. Dispersion führt zu Verminderungen der<br />
Durchbruchskonzentrationen auf C/C0 zwischen 0.23 (Braunerde) und 0.16 (Schwarzerde). Die hier<br />
betrachteten Transportstrecken von bis zu 1.75 m ab der RCB-Unterkante dürften idR. kürzer als die in<br />
vielen Praxisfällen relevanten Grundwasserflurabstände sein, wodurch unter Umständen eine stärkere<br />
dispersive Konzentrationsreduktion möglich wäre.<br />
C/C0 [-]<br />
0.3<br />
0.2<br />
0.1<br />
a) Parkplatz<br />
Braunerde<br />
Podsol<br />
Schwarzerde<br />
Parabraunerde<br />
Fahlerde<br />
Pelosol<br />
0<br />
0 1 2 3<br />
Zeit [a]<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0<br />
0 5 10 15<br />
Zeit [a]<br />
9<br />
b) Lärmschutzwall<br />
Legende siehe<br />
Straßendamm<br />
0.3<br />
0.2<br />
0.1<br />
0<br />
c) Straßendamm<br />
Braunerde (verd.)<br />
Podsol (verd.)<br />
Schwarzerde (verd.)<br />
Braunerde<br />
Podsol<br />
Schwarzerde<br />
0 1 2 3<br />
Zeit [a]<br />
Abb. 5: Durchbruchskurven des Tracers an der Grundwasseroberfläche für (a) Parkplatz-, (b)<br />
Lärmschutzwall- und (c) Straßendamm-Szenario. Gezeigt werden Durchbruchskurven für den<br />
Schadst<strong>of</strong>ftransportpfad ((b) und (c), graue dicke Kurven) bzw. auf das Gesamtbauwerk bezogene<br />
Durchbruchskurven unter Berücksichtigung der Verdünnung durch nicht belastetes Sickerwasser ((b) und<br />
(c), schwarze dünne Kurven).<br />
Die über den unteren Modellr<strong>and</strong> integrierten Tracer-Durchbruchskurven für das LSW-Szenario (Abb. 5 (b))<br />
zeigen im Vergleich zum PP ein stärkeres Tailing und über längere Zeiträume anhaltende hohe<br />
Konzentrationen C/C0. Aus Anschaulichkeitsgründen werden die Ergebnisse hier nur für Braunerde, Podsol<br />
und Schwarzerde vorgestellt. Das ausgeprägte Tailing resultiert aus der geringeren Infiltrationsrate sowie<br />
aus der Geometrie des LSW. Die Überdeckung des grobkörnigen RCB mit einem feinkörnigen Lehm führt an<br />
der Böschung zu einer Kapillarsperre, welche die Infiltration in den RCB vermindert und ein Umströmen des<br />
Wallkerns verursacht (siehe Abb. 6 (a)).<br />
Die Kapillarsperre bildet sich aus, da der RCB aufgrund geringer Kapillarität bereits bei geringen<br />
Saugspannungen einen Großteil des Porenwassers verliert (siehe Abb. 4) und die hydraulische Leitfähigkeit<br />
gegenüber dem Lehm deutlich stärker abnimmt. Das Sickerwasser kann so entlang der geneigten<br />
Materialgrenze in dem nun besser wasserleitenden Lehmboden abfließen. Mit zunehmender Distanz von der<br />
Wallkrone erhöhen sich die Wasserflüsse im Lehm, sodass in Abhängigkeit der Druckverhältnisse<br />
zunehmend Wasser infiltrieren kann (Ross 1990). Die räumliche Variabilität der Sickerwasserströmung führt<br />
zu Bereichen mit stark reduzierten Fließgeschwindigkeiten und so zu einem langsameren Auswaschen des<br />
Tracers aus dem RCB. Die Maximalkonzentrationen erreichen die GWO nach etwa 2 (Braunerde) bis 3.5 a<br />
(Schwarzerde) (Abb. 5 (b)). Im Vergleich zum PP kommt es zudem zu höheren Peakkonzentrationen C/C0<br />
für den Transportpfad (Braunerde: 0.43, Schwarzerde: 0.35), da zum Zeitpunkt des Peak-Durchbruchs an<br />
der GWO die Eluatkonzentrationen der Quelle noch deutlich höher als beim PP sind (siehe Abb. 7 (a)) und<br />
so die dispersive Konzentrationsminderung weniger effektiv ist. Der länger anhaltende hohen<br />
Eluatkonzentrationen resultieren aus den geringeren Sickerwassermengen, der Reduktion der Infiltration in<br />
den RCB durch Kapillarsperren und der größere Mächtigkeit des RCB gegenüber dem PP. Auf das gesamte<br />
Bauwerk bzw. auf die gesamte infiltrierende Wassermenge bezogen reduzieren sich die<br />
Maximalkonzentrationen durch kleinräumige Mittelung jeweils um ca. 40 %. Zu beachten ist in diesem<br />
Zusammenhang, dass die Ausprägung und Effektivität einer Kapillarsperre von einer Reihe von
R<strong>and</strong>bedingungen abhängig ist. Aus diesem Grund sind die berechneten Verdünnungsfaktoren nur für die<br />
hier betrachteten Geometrien, Infiltrationsraten und Materialkombinationen quantitativ gültig.<br />
(a) Lärmschutzwall (b) Straßendamm<br />
Modellergebnisse bezogen auf den<br />
Transportpfad (SMART-Output)<br />
über Gesamtbreite gemittelte Modellergebnisse über Gesamtbreite gemittelte Modellergebnisse<br />
10<br />
Modellergebnisse bezogen auf den<br />
Transportpfad (SMART-Output)<br />
Abb. 6: Geschwindigkeitsvektoren der ungesättigten Strömung an allen Elementknoten im Lärmschutzwall<br />
(a) und in einem Ausschnitt des Straßendammes (b). Die Konzentration des Flusses in den<br />
Bodendeckschichten ist Resultat des Kapillarsperreneffektes, wird durch die feinere Netzdiskretisierung<br />
jedoch überzeichnet.<br />
C/C0 [-]<br />
1<br />
0.1<br />
0.01<br />
a) Tracer<br />
Parkplatz<br />
Lärmschutzwall<br />
Straßendamm<br />
0.01 0.1 1 10<br />
Zeit [a]<br />
1<br />
0.9<br />
0.8<br />
0.7<br />
0.6<br />
0.5<br />
0.4<br />
0.3<br />
0.2<br />
0.1<br />
0<br />
b) Parkplatz<br />
Naphthalin<br />
Phenanthren<br />
Σ15 EPA PAK<br />
0.1 1 10 100<br />
Zeit [a]<br />
Abb. 7: Zeitliche Entwicklung der aus der Schadst<strong>of</strong>fquelle eluierenden Konzentrationen an der Recycling-<br />
Baust<strong>of</strong>f-Unterkante (RCB) für ausgewählte Schadst<strong>of</strong>f-Verwertungsszenario-Kombinationen: (a)<br />
konservativer Tracer für Parkplatz-, Lärmschutzwall- und Straßendammszenario; (b) Naphthalin,<br />
Phenanthren und Σ15 EPA-PAK für das Parkplatzszenario.<br />
Auch für das SD-Szenario ist ein Kapillarsperreneffekt zu beobachten (Abb. 6(b)). Während sich im<br />
Bankettbereich eine relativ gleichförmige Verteilung der Sickerwasserströmung zeigt, strömt in der<br />
Bodenabdeckung der Böschung ein Teil des Sickerwassers am RCB vorbei und infiltriert konzentriert am<br />
Böschungsfuß. Bei kleinräumiger Mittelung der Durchbruchskonzentrationen ergibt sich eine Verdünnung um<br />
ca. 31 %. Die Durchbruchskurven des Tracers (Abb. 5 (c)) zeigen ein steileres Ansteigen als beim LSW und<br />
fallen innerhalb von fünf Jahren auch deutlich schneller ab. Die maximalen Durchbruchskonzentrationen<br />
C/C0 liegen im kleinräumigen Mittel zwischen 0.20 (Braunerde) und 0.18 (Schwarzerde).<br />
Abb. 8 zeigt die Durchbruchskurven für NAP, PHE und Σ15 EPA-PAK jedoch ohne Abbau. Für das PP-<br />
Szenario zeigt sich in Abb. 8 (a)-(c) deutlich der Effekt der vom Corg des Unterbodens abhängigen Sorption.<br />
Während für die Corg-armen Böden NAP bereits nach kurzer Zeit durchbricht, wird es beim mäßig Corghaltigen<br />
Podsol leicht, bei der Corg-reichen Schwarzerde stärker retardiert (Abb. 8 (a)). Für PHE und<br />
Σ15 EPA-PAK (Abb. 8 (b) und (c)) zeigen auch die Corg-armen Böden eine deutliche Retardation. Der<br />
flachere Verlauf der Durchbruchskurven für Braunerde im oberen Konzentrationsbereich deutet zudem auf<br />
einen Einfluss der langsamen Sorptionskinetik des Skelettanteils hin. Für alle Böden bis auf die Schwarzerde<br />
wird innerhalb der Simulationszeit der Durchbruch der PHE-Peaks mit C/C0 zwischen 0.70 und 0.96 und eine<br />
500
anschließende Konzentrationsabnahme auf Grund der zeitlichen Abnahme der Quellstärke (siehe Abb. 7 (b))<br />
beobachtet. Für die Schwarzerde steigt C/C0 nach 500 a noch deutlich an.<br />
C/C0 [-]<br />
C/C0 [-]<br />
C/C0 [-]<br />
1<br />
0.1<br />
0.01<br />
1<br />
0.1<br />
0.01<br />
1<br />
0.1<br />
0.01<br />
a) Parkplatz<br />
Naphthalin<br />
d) Lärmschutzwall<br />
Naphthalin<br />
g) Straßendamm<br />
Naphthalin<br />
0.1 1 10 100<br />
Zeit [a]<br />
b) Parkplatz<br />
Phenanthren<br />
e) Lärmschutzwall<br />
Phenanthren<br />
h) Straßendamm<br />
Phenanthren<br />
0.1 1 10 100<br />
Zeit [a]<br />
11<br />
c) Parkplatz<br />
Σ15 EPA PAK<br />
Braunerde<br />
Podsol<br />
Schwarzerde<br />
Parabraunerde<br />
Fahlerde<br />
Pelosol<br />
f) Lärmschutzwall<br />
Σ15 EPA PAK<br />
Braunerde (verd.)<br />
Podsol (verd.)<br />
Schwarzerde (verd.)<br />
Braunerde<br />
Podsol<br />
Schwarzerde<br />
i) Straßendamm<br />
Σ15 EPA PAK<br />
Braunerde (verd.)<br />
Podsol (verd.)<br />
Schwarzerde (verd.)<br />
Braunerde<br />
Podsol<br />
Schwarzerde<br />
0.1 1 10 100<br />
Zeit [a]<br />
Abb. 8: Durchbruchskurven der sorbierbaren St<strong>of</strong>fe Naphthalin (schwach sorptiv, links) und Phenanthren<br />
(mäßig gut sorptiv, Mitte) und des Summenparameters Σ15 EPA-PAK (stark sorptiv, rechts) für die drei<br />
Szenarien Parkplatz ((a)-(c)), Lärmschutzwall ((d)-(f)) und Straßendamm ((g)-(i)). Gezeigt werden für<br />
Lärmschutzwall und Straßendamm jeweils Durchbruchskurven für den Schadst<strong>of</strong>ftransportpfad (graue dicke<br />
Kurven) bzw. auf das Gesamtbauwerk bezogene Durchbruchskurven unter Berücksichtigung der<br />
Verdünnung durch nicht belastetes Sickerwasser (schwarze dünne Kurven).<br />
Abb. 8 (d)-(f) zeigt die NAP-, PHE- und Σ15 EPA-PAK- Durchbruchskurven des LSW-Szenarios. Ein<br />
deutlicher Unterschied zum PP ist das längere Anhalten hoher Konzentrationen, welches aus den im<br />
Durchschnitt niedrigeren Strömungsgeschwindigkeiten resultiert. Für NAP werden unter Berücksichtigung<br />
der Verdünnung durch am RCB vorbeiströmendes Sickerwasser Maximaldurchbrüche von C/C0 = 0.56<br />
erreicht (Abb. 8 (d)). Für PHE und Σ15 EPA-PAK zeigen sich bei allen betrachteten Böden nach 500 a noch<br />
deutlich ansteigende Konzentrationen (Abb. 8 (e) und (f)).<br />
Im Vergleich dazu zeigen sich für das SD-Szenario aufgrund der stark erhöhten Infiltrationsraten steilere<br />
Durchbruchskurven und somit ein früheres Erreichen der Konzentrationspeaks (Abb. 8 (g)-(i)). NAP zeigt<br />
seine Maximaldurchbrüche bei der Braunerde bereits nach 19 a (C/C0 = 0.59), bei der Schwarzerde nach 46<br />
a (C/C0 = 0.60).<br />
Ohne Abbau erfolgt langfristig der Eintrag der gesamten Schadst<strong>of</strong>fmasse des RCB ins Grundwasser. Wird<br />
Abbau bei der Simulation mit berücksichtigt (Abb. 9 (a)– (i)), ist die Massenreduktion bei gegebener<br />
Abbauratenkonstante nur von der Verweilzeit der Substanzen im Bodenwasser abhängig. Retardation hat
keinen Einfluss auf die abgebaute Masse, da Mikroorganismen idR. nur zum Abbau in Lösung vorliegender<br />
Schadst<strong>of</strong>fe fähig sind. So zeigen Abb. 9 (a)-(c) für den PP, dass der Abbau bei der Parabraunerde trotz<br />
früherer Durchbruchszeiten effektiver ist, als z.B. bei der Braunerde oder dem Podsol. Dies liegt an der<br />
geringeren Abst<strong>and</strong>sgeschwindigkeit va = q / θ (= Transportgeschwindigkeit des Tracers) bzw. dem höheren<br />
Wassergehalt der feinkörnigen gegenüber den s<strong>and</strong>igen Böden bei vorgegebener Infiltrationsrate (=<br />
Darcyfluss q), woraus sich eine längere effektive Aufenthaltszeit der Substanzen im Bodenwasser des Pr<strong>of</strong>ils<br />
ergibt. Insgesamt ist bei der angenommenen Halbwertzeit von 70 d eine Reduktion der<br />
Durchbruchkonzentrationen für den PP um einen Faktor von bis zu 30 möglich.<br />
C/C0 [-]<br />
C/C0 [-]<br />
C/C0 [-]<br />
1<br />
0.1<br />
0.01<br />
0.001<br />
1<br />
0.1<br />
0.01<br />
0.001<br />
1<br />
0.1<br />
0.01<br />
0.001<br />
a) Parkplatz<br />
Naphthalin<br />
d) Lärmschutzwall<br />
Naphthalin<br />
Braunerde (verd.)<br />
Podsol (verd.)<br />
Schwarzerde (verd.)<br />
Braunerde<br />
Podsol<br />
Schwarzerde<br />
g) Straßendamm<br />
Naphthalin<br />
Braunerde (verd.)<br />
Podsol (verd.)<br />
Schwarzerde (verd.)<br />
Braunerde<br />
Podsol<br />
Schwarzerde<br />
0.1 1 10 100<br />
Zeit [a]<br />
b) Parkplatz<br />
Phenanthren<br />
Braunerde<br />
Podsol<br />
Schwarzerde<br />
Parabraunerde<br />
Fahlerde<br />
Pelosol<br />
e) Lärmschutzwall<br />
Phenanthren<br />
h) Straßendamm<br />
Phenanthren<br />
0.1 1 10 100<br />
Zeit [a]<br />
12<br />
c) Parkplatz<br />
Σ15 EPA PAK<br />
f) Lärmschutzwall<br />
Σ15 EPA PAK<br />
i) Straßendamm<br />
Σ15 EPA PAK<br />
0.1 1 10 100<br />
Zeit [a]<br />
Abb. 9: Durchbruchskurven der sorbierbaren St<strong>of</strong>fe Naphthalin (schwach sorptiv, links) und Phenanthren<br />
(mäßig gut sorptiv, Mitte) und des Summenparameters Σ15 EPA-PAK (stark sorptiv, rechts) unter<br />
Berücksichtigung des biologischen Abbaus für die drei Szenarien Parkplatz ((a)-(c)), Lärmschutzwall ((d)-(f))<br />
und Straßendamm ((g)-(i)). Gezeigt werden für Lärmschutzwall und Straßendamm jeweils<br />
Durchbruchskurven für den Schadst<strong>of</strong>ftransportpfad (graue dicke Kurven) bzw. auf das Gesamtbauwerk<br />
bezogene Durchbruchskurven unter Berücksichtigung der Verdünnung durch nicht belastetes Sickerwasser<br />
(schwarze dünne Kurven).<br />
Aufgrund der langsameren Strömungsgeschwindigkeiten ist beim LSW die durchschnittliche Verweilzeit der<br />
gelösten Schadst<strong>of</strong>fe im Boden höher als beim PP. Dies führt zu einer stärkeren Konzentrationsreduktion bei<br />
Abbau (Minderungsfaktoren zwischen 25 und 150; Abb. 9 (d)-(f)). Umgekehrt zeigen sich beim SD auf Grund<br />
höherer Strömungsgeschwindigkeiten deutlich geringere Aufenthaltszeiten, sodass sich trotz des Abbaus nur<br />
geringe Minderungsfaktoren zwischen 2.5 und 5 ergeben (Abb. 9 (g)–(i)).
Schlussfolgerungen<br />
Die in dieser Studie durchgeführten umfangreichen Typ-Szenarien-Simulationen ergänzen die im BMBF-<br />
Verbundprojekt „Sickerwasserprognose“ erarbeiteten wissenschaftlichen Grundlagen und methodischen<br />
Instrumentarien durch die Anwendung auf praxisrelevante Fallbeispiele. Aus den dabei gewonnenen<br />
Ergebnissen lassen sich folgende Schlussfolgerungen ableiten:<br />
• Dispersive Konzentrationsminderung ist nur bedingt wirksam. Eine nennenswerte Abschwächung der<br />
Durchbruchskonzentrationen ist nur für Fälle möglich, in denen das Abklingen der Quellkonzentrationen<br />
deutlich vor dem Durchbruch des Schadst<strong>of</strong>fpeaks an der GWO erfolgt. Bei länger anhaltenden<br />
Quellstärken (z.B. für Salze wie Sulfat) ist dagegen mit weitgehend unvermindertem<br />
Konzentrationsdurchbruch zu rechnen.<br />
• Für stärker sorbierende Schadst<strong>of</strong>fe (PHE, Σ15 EPA-PAK) zeigt sich schon bei geringem Corg der<br />
Unterböden eine Retardation der Peak-Durchbruchszeitpunkte um viele Jahrzehnte bis Jahrhunderte.<br />
Mit signifikanten St<strong>of</strong>feinträgen (in Abhängigkeit der Quellstärke) ins Grundwasser ist so bei „natürlichen“<br />
Grundwasserneubildungsverhältnissen erst nach sehr langen Zeiträumen zu rechnen.<br />
• Auch bei relativ niedrigen Ratenkonstanten ist eine deutliche Reduktion der Konzentrationen organischer<br />
Schadst<strong>of</strong>fe durch mikrobiellen Abbau möglich. Bezüglich der Effektivität des Abbaus ergeben sich<br />
jedoch erhebliche Unterschiede zwischen den drei betrachteten Verwertungsszenarien, die aus den<br />
unterschiedlichen Strömungs- bzw. Transportgeschwindigkeiten resultieren.<br />
• Die St<strong>of</strong>fverlagerung in den in zwei Raumdimensionen betrachteten Szenarien Lärmschutzwall und<br />
Straßendamm zeigt darüber hinaus einen ausgeprägten Einfluss der räumlich variablen<br />
Sickerwassermenge. Bereiche hoher Strömungsgeschwindigkeit führen zu früheren Ankunftszeiten der<br />
Schadst<strong>of</strong>fe im Grundwasser. Transportprognosen unter Vernachlässigung dieser Effekte können<br />
deshalb zu einer Unterschätzung der Schadst<strong>of</strong>fverlagerung führen und sind nicht konservativ.<br />
• Die Strömungsbilanzen für Lärmschutzwall und Straßendamm legen nahe, dass mit geeigneten<br />
Bauwerksgeometrien und Materialien eine Reduktionen der Infiltration in das Verwertungsmaterial und<br />
des Austrags von Schadst<strong>of</strong>fen ins Grundwasser durch das Ausnutzen von Kapillarsperreneffekten zu<br />
erzielen ist. Der hier vorgestellte Modellansatz schafft grundsätzlich die Möglichkeit, das Design von<br />
Verwertungsszenarien ohne großen Mehraufw<strong>and</strong> in der Umsetzung durch numerische Simulationen in<br />
Richtung einer möglichst geringen Umweltbelastung bei hoher Verwertungsquote optimieren zu können.<br />
Danksagung: Die Untersuchungen wurden im Rahmen des BMBF-Förderschwerpunkts<br />
„Sickerwasserprognose“ (Projektnummer 02WP0517) durchgeführt. Dem BMBF sei für die Förderung<br />
gedankt. Des Weiteren danken wir Herrn Dr. Utermann und Herrn Dr. Duijnisveld (BGR, Hannover), Herrn<br />
Dr. Susset und Herrn Dr. Leuchs (LANUV, Recklinghausen), Herrn Dr. Henzler (UFZ, Leipzig) und Frau Dr.<br />
Kocher (BaST, Bergisch Gladbach) für ausführliche und hilfreiche Diskussionen sowie Frau Dr. Kouznetsova<br />
und Herrn Duran für Hilfe bei der Durchführung der Simulationen.<br />
Literaturangaben<br />
Allen-King, R. M. Grathwohl, P., Ball, W. P.: New modelling paradigms for the sorption <strong>of</strong> hydrophobic organic<br />
chemicals to heterogeneous carbonaceous matter in soils, sediments <strong>and</strong> rocks.- Adv. Wat. Res. 25, 985-<br />
1016 (2002)<br />
Arya, L.M., Paris, J.F.: A physicoempirical model to predict soil moisture characteristics from particle-size<br />
distribution <strong>and</strong> bulk density data. Soil Sci. Soc. Am. J. 45, 1023-1030 (1981)<br />
BBodSchV: Bundes-Bodenschutz- und Altlastenverordnung vom 16. Juli 1999. Bundesgesetzblatt Jahrgang<br />
1999, Teil I Nr.36, 1554-1682 (1999)<br />
BGR (Bundesanstalt für Geowissenschaften und Rohst<strong>of</strong>fe): Nutzungsdifferenzierte Bodenübersichtskarte der<br />
Bundesrepublik Deutschl<strong>and</strong> 1:1.000.000 (BÜK 1000 N2.3). Auszugskarten Acker, Grünl<strong>and</strong>, Wald; Digit.<br />
Archiv FISBo BGR; Hannover und Berlin (2006)<br />
BMU (Bundesministerium für Umwelt, Naturschutz und Reaktorsicherheit): Hydrologischer Atlas von<br />
Deutschl<strong>and</strong>, 3. Lieferung 2003; Berlin (2000)<br />
Bold, S.: Processed-based prediction <strong>of</strong> long-term risk <strong>of</strong> groundwater pollution by organic non-volatile<br />
contaminants. Dissertation, Tübinger Geowissenschaftliche Arbeiten (TGA), C72, 76 S.; Tübingen (2004)<br />
Brakensiek, D.L., Rawls, W.J.: Soil containing rock fragments: Effects on infiltration.- Catena 23 (1-2), 99–110<br />
(1994)<br />
13
Carsell, R. F., Parish, R. S.: Developing joint probability distributions <strong>of</strong> soil water retention characteristics.<br />
Water Resour. Res. 24 (5), 755-769 (1988)<br />
Du, Y., Wang, W. Kolditz, O.: Richards Flow Modelling. 5. Workshop “Porous Media" Proceedings CD, 2.<br />
Ausgabe, Center for <strong>Applied</strong> Geosciences, University <strong>of</strong> Tübingen (2005)<br />
FGSV: RAS-Q 96, Richtlinien für die Anlage von Straßen – Teil: Querschnitte. 62 S.; Köln (1996)<br />
FGSV: ZTV E-StB 94, Zusätzliche Technische Vertragsbedingungen und Richtlinien für Erdarbeiten im<br />
Straßenbau. 108 S.; Köln (1997)<br />
FGSV: Merkblatt für wasserdurchlässige Befestigung von Verkehrsflächen. 36 S.; Köln (1998)<br />
FGSV: RStO 01, Richtlinien für die St<strong>and</strong>ardisierung des Oberbaues von Verkehrsflächen. 52 S.; Köln (2001)<br />
FGSV: ZTV T-StB 95, Zusätzliche Technische Vertragsbedingungen und Richtlinien für Tragschichten im<br />
Straßenbau. Ausgabe 1955 / Fassung 2002. 126 S.; Köln (2002)<br />
FGSV: ZTV SoB-StB 04, Zusätzliche Technische Vertragsbedingungen und Richtlinien für den Bau von<br />
Schichten ohne Bindemittel im Straßenbau. 48 S.; Köln (2004)<br />
FGSV: RAS-EW, Richtlinien für die Anlage von Straßen – Teil: Entwässerung. 82 S.; Köln (2005)<br />
Finkel, M.: Quantitative Beschreibung des Transports von polyzyklischen aromatischen Kohlenwasserst<strong>of</strong>fen<br />
(PAK) und Tensiden in porösen Medien. Tübinger Geowissenschaftliche Arbeiten (TGA), C47. 98 S.;<br />
Tübingen (1998)<br />
Finkel M., Liedl R., Teutsch G.: Modelling surfactant-enhanced remediation <strong>of</strong> polycyclic aromatic<br />
hydrocarbons. J. Environ. Modelling & S<strong>of</strong>tware 14, 203-211 (1998)<br />
Flöter, O.: Wasserhaushalt gepflasterter Straßen und Gehwege. Lysimeterversuche an drei Aufbauten unter<br />
praxisnahen Bedingungen unter Hamburger Klima. Dissertation, Hamburger Bodenkundliche Arbeiten,<br />
58, 329 S.; Hamburg (2006)<br />
Glugla, G., Jankiewicz, P., Rachimow, C., Lojek, K., Richter, K., Fürtig, G., Krahe, P.:<br />
Wasserhaushaltsverfahren zur Berechnung vieljähriger Mittelwerte der tatsächlichen Verdunstung und<br />
des Gesamtabflusses. BfG-Bericht, Nr. 1342, 106 S.; (2003)<br />
Grathwohl, P.: Diffusion controlled desorption <strong>of</strong> organic contaminants in various soils <strong>and</strong> rocks.- In Kharaka,<br />
K.Y., Maest, A.S. (Hrsg.): Water rock Interaction - Proceedings <strong>of</strong> the Utah Conference, 283-286 (1992)<br />
Grathwohl, P.: Diffusion in Natural Porous Media: Contaminant Transport, Sorption/Desorption <strong>and</strong><br />
Dissolution Kinetics. 224 S.; Kluwer Academic Publishers, Boston (1998)<br />
Grathwohl, P.: Gutachten zur Beschreibung von fachlichen Eckpunkten für die Festlegung von<br />
Zuordnungswerten der Einbauklasse 1.1 (Z 1.1) für organische Schadst<strong>of</strong>fe in mineralischen Abfällen.<br />
Umweltbundesamt (Hrsg.) Texte 37/2004. Umweltforschungsplan des Bundesministeriums für Umwelt,<br />
Naturschutz und Reaktorsicherheit, Forschungsbericht 36301047, UBA-FB 000721. 71 S.; Berlin (2004)<br />
Grathwohl, P., Susset, B.: Sickerwasserprognose - Grundlagen, Möglichkeiten, Grenzen, Modelle. In<br />
"Belastung von Böden und Gewässern" Gemeinschaftstagung ATV-DVWK, 28./29.05.01 Hannover<br />
(2001)<br />
Grathwohl, P., Liedl, R., Beyer, C. Konrad, W.: Übertragung der Ergebnisse des BMBF -<br />
Förderschwerpunktes „Sickerwasserprognose“ auf repräsentative Fallbeispiele. Abschlussbericht des<br />
Teilprojekts 1a des Forschungsvorhabens „Entwicklung und Validierung eines Modells zur Abschätzung<br />
der St<strong>of</strong>fkonzentration am Beurteilungsort“ im Rahmen des BMBF-Förderschwerpunkts<br />
„Sickerwasserprognose“ (Förderkennzeichen 02WP0517) (2006)<br />
Grathwohl, P., Halm, D., Bonilla, A., Broholm, M., Burganos, V., Christophersen, M., Comans, R., Gaganis, P.,<br />
Gorostiza, I., Höhener, P., Kjeldsen, P. <strong>and</strong> van der Sloot, H.: Guideline for Groundwater Risk<br />
Assessment at Contaminated Sites.- Final Report EU-Project EVK1-CT-1999-00029; (2003)<br />
Hansson, K., Lundin, L.C., Simunek, J.: Modeling water <strong>flow</strong> patterns in flexible pavements. Transportation<br />
Research Record 1936, 133-141 (2006)<br />
Hayduk, W., Laudie, H.: Prediction <strong>of</strong> diffusion coefficients for nonelectrolytes in dilute aqueous solutions.<br />
AIChE Journal 20 (3), 611-615 (1974)<br />
Hennings, V.: Methodendokumentation Bodenkunde. Auswertungsmethoden zur Beurteilung der<br />
Empfindlichkeit und Belastbarkeit von Böden. In: Geologisches Jahrbuch/ Sonderheft Reihe G, B<strong>and</strong> 1, 2.<br />
Auflage, 232 S.; (2000)<br />
Henzler R.: Quantifizierung und Modellierung der PAK-Elution aus verfestigten und unverfestigten<br />
Abfallmaterialien. Dissertation, Tübinger Geowissenschaftliche Arbeiten (TGA), C76, 118 S.; Tübingen<br />
(2004)<br />
Henzler, R., Rügner, H., Grathwohl, P.: Bewertung der Filter- und Pufferfunktion von Unterböden für<br />
organische Schadst<strong>of</strong>fe. Bodenschutz 1/2006, 8-14 (2006)<br />
Hjelmar, O., Holm J., Crillesen, K.: Utilisation <strong>of</strong> MSWI bottom ash as sub-base in road construction: First<br />
results from a large-scale test site. J. Hazard. Mater. 139 (3), 471-480 (2007)<br />
Höhener, P, Dakhel, N., Christophersen, M., Broholm, M, Kjeldsen, P.: Biodegradation <strong>of</strong> hydrocarbons<br />
vapors: Comparison <strong>of</strong> laboratory studies <strong>and</strong> field investigations in the vadose zone at the emplaced fuel<br />
source experiment, Airbase Værløse, Denmark. J. Contam. Hydrol. 88 (3-4), 337-358 (2006)<br />
14
Jäger, R., Liedl, R.: Prognose der Sorptionskinetik organischer Schadst<strong>of</strong>fe in heterogenem Aquifermaterial.<br />
Grundwasser 2, 57-66 (2000)<br />
Kellermann, C.: Zur Bewertung des Infiltrationsverhaltens von Tragschichten ohne Bindemittel. Dissertation,<br />
Schriftenreihe des Lehrstuhls für Verkehrswegebau der Ruhr-Universität Bochum, Heft 19, 165 S.;<br />
Bochum (2003)<br />
Kleineidam, S., Rügner, H., Grathwohl, P.: Impact <strong>of</strong> grain scale heterogeneity on slow sorption kinetics.<br />
Environ. Tox. Chem. 18 (8), 1673–1678 (1999a)<br />
Kleineidam, S., Rügner, H., Ligouis, B., Grathwohl, P.: Organic matter facies <strong>and</strong> equilibrium sorption <strong>of</strong><br />
phenanthrene. Environ. Sci. Tech. 33 (10), 1637–1644 (1999b)<br />
Kolditz, O., Bauer, S.: A process-oriented approach to computing multifield problems in porous media. J.<br />
Hydroinf. 6 (3), 225-244 (2004)<br />
Kolditz, O., Xie, M., Kalbacher, T., Bauer, S., Wang, W., McDermott, C., Chen, C., Beyer, C., Gronewold, J.,<br />
Kemmler, D., Legeida, D., Walsh, R., Du, Y., Park, C.H.: GeoSys/Rock<strong>flow</strong> version 4.3.21 - Theory <strong>and</strong><br />
users manual, Center for <strong>Applied</strong> Geoscience, University <strong>of</strong> Tübingen (2006)<br />
Krass, K., Brüggemann, M., Görener, E.: Anfall, Aufbereitung und Verwertung von Recycling-Baust<strong>of</strong>fen und<br />
industriellen Nebenprodukten im Wirtschaftsjahr 2001 – Teil 1; Recycling-Baust<strong>of</strong>fe. Straße + Autobahn<br />
4, 193-202 (2004a)<br />
Krass, K., Brüggemann, M., Görener, E.: Anfall, Aufbereitung und Verwertung von Recycling-Baust<strong>of</strong>fen und<br />
indstriellen Nebenprodukten im Wirtschaftsjahr 2001 – Teil 2; Industrielle Nebenprodukte. Straße +<br />
Autobahn 5, 275-281 (2004b)<br />
KrW/AbfG: Gesetz zur Förderung der Kreislaufwirtschaft und Sicherung der umweltverträglichen Beseitigung<br />
von Abfällen (Kreislaufwirtschafts- und Abfallgesetz – KrW/AbfG). BGBl I 1994, 2705 (1994)<br />
Maier, U., Grathwohl, P.: Natural Attenuation in the un<strong>saturated</strong> zone <strong>and</strong> shallow groundwater: coupled<br />
<strong>modeling</strong> <strong>of</strong> vapor phase diffusion, biogeochemical processes <strong>and</strong> transport across the capillary fringe. In:<br />
Nützmann, G., Viotti, P., Aagard, P.: Reactive transport in soil <strong>and</strong> groundwater., 141-155; Springer,<br />
Berlin (2005)<br />
Mesters, K.: Abschätzung der Mobilisierbarkeit von leichtlöslichen Salzen aus Müllverbrennungsasche am<br />
Beispiel eines Lärmschutzwalles.- Dissertation, Schriftenreihe des Lehrstuhls für Verkehrswegebau der<br />
Ruhr-Universität Bochum, Heft 5, 170 S.; Bochum (1993)<br />
Mishra, S., Parker, J. C. , Singhal, N.: Estimation <strong>of</strong> soil hydraulic properties <strong>and</strong> their uncertainty from particle<br />
size distribution data. J. Hydrol. 108, 1-18 (1989)<br />
Mualem, Y.: A new model for predicting the hydraulic conductivity <strong>of</strong> un<strong>saturated</strong> porous media. Water<br />
Resour. Res. 12, 513–522 (1976)<br />
Richter, T.: Regenwasserversickerung durch Pflasterdecken. Zement-Merkblatt Straßenbau S15<br />
http://www.vdz-online.de/fileadmin/gruppen/vdz/3LiteraturRecherche/<br />
Zementmerkblaetter/S15.pdf (2003)<br />
Ross, B.: The diversion capacity <strong>of</strong> capillary barriers. Water Resour. Res. 26 (10), 2625–2629 (1990)<br />
Rügner, H., Henzler, R., Grathwohl, P.: Beurteilung der Empfindlichkeit der Filter- und Pufferfunktion von<br />
Böden (i.B. Unterböden) nach Maßstäben des vorsorgenden Bodenschutzes für organische Schadst<strong>of</strong>fe -<br />
Abschlussbericht. Projekt: „LABO 2003 B 2.03" (Programm: "Wasser und Boden" 2002 der Bund/Länder-<br />
Arbeitsgemeinschaft Bodenschutz) (2005)<br />
Rügner, H., Kleineidam, S., Grathwohl, P.: Sorptions- und Transportverhalten organischer Schadst<strong>of</strong>fe in<br />
heterogenen Materialien am Beispiel des Phenanthrens. Grundwasser 3, 133-138 (1997)<br />
Rügner, H., Kleineidam, S., Grathwohl, P.: Long-term sorption kinetics <strong>of</strong> phenanthrene in aquifer materials.<br />
Environ. Sci. Tech. 33 (10), 1645–1651 (1999)<br />
Seth, R., Mackay, D., Muncke, J.: Estimating <strong>of</strong> organic carbon partition coefficient <strong>and</strong> its variability for<br />
hydrophobic chemicals. Environ. Sci. Technol. 33 (14), 2390-2394 (1999)<br />
Stieber, M., Kraßnitzer, S., dos Santos Coutinho, C., Tiehm, A.: Untersuchung der Bedeutung des<br />
mikrobiellen Abbaus für den Transport persistenter organischer Schadst<strong>of</strong>fe in der ungesättigten Zone.<br />
Abschlussbericht des Teilprojekts 3 des Forschungsvorhabens „Entwicklung und Validierung eines<br />
Modells zur Abschätzung der St<strong>of</strong>fkonzentration am Beurteilungsort“ im Rahmen des BMBF-<br />
Förderschwerpunkts „Sickerwasserprognose“ (Förderkennzeichen 02WP0515) (2006)<br />
Stoppka, B.: Prognose des Salzaustrags aus Straßenbaust<strong>of</strong>fen in der Sickerzone.- Dissertation,<br />
Schriftenreihe des Lehrstuhls für Verkehrswegebau der Ruhr-Universität Bochum, Heft 16, 232 S.;<br />
Bochum (2002)<br />
Susset, B.: Materialien zur Sickerwasserprognose.- In: Förstner, U., Grathwohl, P.: Ingenieurgeochemie,<br />
Technische Geochemie - Konzepte und Praxis. Textbeitrag und Folien im CD-Anhang der 2. neu<br />
bearbeiteten Auflage, 471 S.; Springer, Berlin (2007)<br />
Utermann, J., Kladivko, E.J., Jury, W.A.: Evaluation pesticide migration in tile drained soils with a transfer<br />
function model. J. Environ. Quality 19, 707-714 (1990)<br />
V<strong>and</strong>erborght, J., Vereecken, H.: One-Dimensional Modeling <strong>of</strong> Transport in Soils with Depth-Dependent<br />
Dispersion, Sorption <strong>and</strong> Decay. Vadose Zone J. 6, 140-148 (2007)<br />
15
van Genuchten, M.T.: A closed form equation for predicting the hydraulic conductivity <strong>of</strong> un<strong>saturated</strong> soils.<br />
Soil Sci. Soc. Am. J. 44, 892-898 (1980)<br />
Wörner, T., Löcherer, L., Westiner, E.: Untersuchung über das Verhalten von Recycling-Baust<strong>of</strong>fen in<br />
Tragschichten ohne Bindemittel unter längerer Verkehrsbeanspruchung. Beitrag zum Statusseminar<br />
Verbundvorhaben "Restst<strong>of</strong>fverwertung im Straßenbau", TV 4, FKZ 147 10 43 0, Bochum (2001)<br />
Wösten, J. H. M., Lilly, A., Nemes, A., Le Bas, C.: Using existing soil data to derive hydraulic parameters for<br />
simulation models <strong>and</strong> in l<strong>and</strong> use planning. – Final Report on the European Union Funded project. DLO-<br />
Staring Centre, Report 156, Wageningen /The Netherl<strong>and</strong>s (1998)<br />
16