Riverbank Filtration Within The Context of River Restoration and Climate Change

University of Neuchtel
Faculty of Science
Centre for Hydrogeology

and Geothermics
Ph.D. Thesis
Riverbank Filtration
within the Context of
River Restoration and Climate Change
presented to the Faculty of Science of the University of Neuchtel to satisfy
the requirements of the degree of Doctor of Philosophy in Science
by
Samuel Diem
Thesis defense date: 20.08.2013

Public presentation date: 19.09.2013
Ph.D. Thesis evaluation committee:

Prof. Dr. Daniel Hunkeler, University of Neuchtel (Director of the thesis)
Prof. Dr. Mario Schirmer, University of Neuchtel (Co-director of the thesis)
Prof. Dr. Philippe Renard, University of Neuchtel
Prof. Dr. John Molson, University Laval, Qubec, Canada
Acknowledgments
The completion of this Ph.D. Thesis would not have been possible without the support of
colleagues, friends and family.
I would like to start by thanking my supervisor, Mario Schirmer. He gave me the opportunity
to write this Ph.D. Thesis, supported me in any situation, had always time for discussions,
provided helpful inputs, and never refused a request (course, conference, field equipment,).
I am grateful to Daniel Hunkeler for the possibility to do my Ph.D. at the CHYN and for
many helpful discussions. I thank John Molson for his commitment as external examiner. I
further would like to thank Philippe Renard and Olaf Cirpka who supported me with
substantial ideas and instructive discussions. I am also thankful to Eduard Hoehn and Daniel
Kser for their advice in various situations and stages of my Ph.D. and for motivating me in
times of local and global minima in my motivation space.
Many thanks to the RIBACLIM-Team: Urs von Gunten for leading and initiating the project
and for many fruitful discussions; Matthias Rudolf von Rohr for his continuous support in the
field, his introduction to lab work and his optimism; Sabrina Bahnmller for her help in the
lab and the field and for always reminding me that we are all in the same boat; Janet Hering,
Hans-Peter Kohler and Silvio Canonica for their uncomplicated exchange and their
collaboration.
A big THANK YOU to the Hydrogeology Group (Anne-Marie, Ben, Dirk, Jana, Mehdi,
Stefano, Vidhya), my office mates (Ryan, Sabine, Stephan) and the RoKi Group (Lars, Lina,
Matthias, Simon, Yama). You all contributed to this thesis in some way and created an
enjoyable working environment at Eawag. I thank Lena Froyland (Master student) and Roger
Mgroz (Intern) for the opportunity to supervise their work. I also thank the Eawag-Werkstatt
(Andy, Peter, Richi) for their technical support in the field, the Aua-Lab for analyzing all the
samples, as well as the IT- and Empfang-Team for their excellent service.
Many thanks to all my friends; you helped me to release and re-adjust my mind. Finally, I
would like to thank my family, especially my parents, my famous four sisters and my
parents-/brother-/sisters-in-law, who always supported and encouraged me. Last but certainly
not least, I would like to thank Murielle, my beloved wife. She was my backbone, supportive
and understanding through it all she showed me true love.
Abstract
Drinking water derived by riverbank filtration is generally of high quality and is an important
source of drinking water in several European countries. In the future however, riverbankfiltration systems will face two major challenges river restoration and climate change. The
goal of this Ph.D. Thesis was to deepen the understanding of physical and biogeochemical
processes that occur during riverbank filtration and develop new tools in order to facilitate the
assessment of potential adverse effects of river restoration and climate change on the quality
of river-recharged groundwater.
River restoration measures can lead to shorter residence times between the river and the
pumping well and therefore can increase the risk of drinking water contamination by bacteria
or pollutants. Numerical groundwater models provide quantitative information on
groundwater flow paths and residence times, but require a rigorous definition of the spatial
and temporal river water level distribution. In this thesis, two new interpolation methods were
developed to generate time-varying 1D and 2D river water level distributions. The methods
were implemented at the partly restored Niederneunforn field site at the peri-alpine Thur
River (NE-Switzerland), and were applied to a 3D groundwater flow and transport model. The
results confirmed the methods suitability for accurately simulating groundwater flow paths
and residence times.
The increased occurrence of heat waves due to climate change likely favors the development
of anoxic conditions in the infiltration zone, which may significantly deteriorate the quality of
river-recharged groundwater. Results from field sampling campaigns and column experiments
suggest that particulate organic matter (POM) degradation mainly accounted for the
variability of dissolved oxygen (DO) consumption during riverbank filtration. Furthermore,
DO consumption was found to positively correlate with temperature and discharge. The latter
was attributed to an enhanced trapping of POM within the riverbed during high-discharge
conditions. To quantify the temperature and discharge dependence of DO consumption during
riverbank filtration, a new semi-analytical model was developed and successfully applied to
the Niederneunforn field site. The modeling approach can be transferred to other riverbankfiltration systems to efficiently estimate groundwater DO concentrations under various
climatic and hydrologic conditions and, hence, to assess the risk of arising anoxic conditions.
Kurzfassung
Mittels Uferfiltration gewonnenes Trinkwasser ist generell von hoher Qualitt und stellt eine
wichtige Trinkwasserressource fr mehrere europische Lnder dar. In der Zukunft werden
Uferfiltrationssysteme jedoch mit zwei bedeutenden Herausforderungen konfrontiert
Flussrevitalisierung und Klimanderung. Das Ziel dieser Doktorarbeit war es, das Verstndnis
physikalischer und biogeochemischer Prozesse whrend der Flussinfiltration zu vertiefen und
Instrumente zu entwickeln, um potentielle negative Auswirkungen von Flussrevitalisierung
und Klimanderung auf die Qualitt des Uferfiltrats besser zu erfassen.
Massnahmen der Flussrevitalisierung knnen zu verkrzten Fliesszeiten zwischen Fluss und
Trinkwasserfassung fhren, was das Risiko einer Trinkwasserkontamination mit Bakterien
und Schadstoffen erhhen kann. Numerische Grundwassermodelle liefern quantitative
Informationen ber Grundwasserfliesspfade und Fliesszeiten, bentigen aber eine genaue
Definition der rumlichen und zeitlichen Flusswasserstandsverteilung. In dieser Arbeit
wurden zwei neue Interpolationsmethoden entwickelt, um zeitlich variable 1D und 2D
Flusswasserstandsverteilungen zu generieren. Die Methoden wurden am teilweise
revitalisierten Feldstandort Niederneunforn am voralpinen Fluss Thur (Nordostschweiz)
implementiert und auf ein 3D Grundwasserstrmungs- und Transportmodell angewandt. Die
Resultate
besttigten
die
Eignung
der
Methoden
zur
przisen
Simulation
von
Grundwasserfliesspfaden und Fliesszeiten.

Das vermehrte Auftreten von Hitzewellen aufgrund der Klimanderung begnstigt
mglicherweise die Ausbildung anoxischer Verhltnisse in der Infiltrationszone, was die
Qualitt des Uferfiltrats deutlich verschlechtern wrde. Die Resultate aus Feldprobennahmen
und Sulenversuchen deuten darauf hin, dass der Abbau von partikulrem organischem
Material (POM) hauptschlich fr die Variabilitt der Sauerstoffzehrung whrend der
Flussinfiltration verantwortlich war. Zustzlich wurde eine positive Korrelation zwischen der
Sauerstoffzehrung und der Temperatur sowie dem Abfluss festgestellt. Letztere wurde einem
erhhten Eintrag von POM in das Flussbett whrend Hochwasserbedingungen zugeschrieben.
Um die Temperatur- und Abflussabhngigkeit der Sauerstoffzehrung whrend der
Flussinfiltration zu quantifizieren, wurde ein neues semi-analytisches Modell entwickelt und
erfolgreich am Feldstandort Niederneunforn angewandt. Der Modellansatz lsst sich auf
weitere Uferfiltrationssysteme bertragen, um Sauerstoffkonzentrationen im Grundwasser
effizient abzuschtzen, und somit das Risiko aufkommender anoxischer Bedingungen zu
beurteilen.
ii
Rsum
Leau de consommation provenant de la filtration sur berge est gnralement de bonne qualit
et constitue une source deau potable dans plusieurs pays de lUnion Europenne. A lavenir
pourtant, la filtration sur berge devra faire face deux dfis majeurs: les programmes de
restauration de rivire et les changements climatiques. La prsente thse vise, dune part,
approfondir la comprhension des processus physiques et biogochimiques lis la filtration
sur berge et, dautre part, dvelopper de nouveaux outils pour valuer les impacts potentiels
des programmes de restauration et des changements climatiques sur la qualit de leau
souterraine provenant de ces techniques de filtration.
Les mesures de restauration de rivire peuvent conduire un raccourcissement des temps de
rsidence de leau entre la rivire et le puits de pompage, et ainsi augmenter les risques de
contamination chimique et microbiologique. La modlisation hydrodynamique est une
approche quantitative privilgie pour identifier les chemins dcoulements souterrains et
quantifier les temps de rsidence; elle requiert cependant une dfinition rigoureuse de la
variabilit spatiale et temporelle des niveaux de rivire. Dans cette thse, deux mthodes
dinterpolation sont dveloppes pour gnrer une reprsentation temporelle adquate des
niveaux de rivires en 1D et 2D. Ces mthodes sont testes sur le site exprimental et
partiellement restaur de Niederneunforn (NE de la Suisse), situ sur la rivire pralpine Thur,
et sont implmentes dans un modle numrique dcoulement et de transport souterrain 3D.
Les rsultats confirment la pertinence de ces mthodes pour la simulation prcise des chemins
dcoulements et des temps de rsidence.
Avec les changements climatiques, laugmentation de la frquence des vagues de chaleur
favorisera probablement le dveloppement de conditions anoxiques dans les zones
dinfiltration de la rivire, ce qui tendra dtriorer la qualit de leau souterraine recharge
par filtration de berge. Les rsultats de campagnes dchantillonnage et dexpriences sur
colonne suggrent que la dgradation de la matire organique particulaire (MOP) est la
principale cause de variabilit de la consommation en oxygne dissous (OD) lie au processus
de filtration sur berge. En outre, la consommation de lOD apparat positivement corrle la
temprature et au dbit de la rivire. Cette seconde corrlation est attribue au pigeage accru
de la MOP dans le lit du cours deau pendant les priodes de hauts dbits. Finalement, afin de
quantifier linfluence de la temprature et du dbit sur la consommation en OD dans un
contexte de filtration sur berge, un modle semi-analytique original est dvelopp et valu,
avec succs, sur les donnes du site de Niederneunforn. Cette approche de modlisation est
iii
transfrable dautres sites de filtration sur berge o des outils performants sont ncessaires
pour estimer les teneurs en OD dans diverses conditions climatiques et hydrologiques, et
valuer ainsi le risque de dveloppement de conditions anoxiques.
iv
Keywords
Riverbank filtration; river restoration; climate change; groundwater flow and reactive
transport modeling; stochastic-convective reactive transport; river water level distribution;
particulate organic matter; oxygen consumption
Stichwrter
Uferfiltration; Flussrevitalisierung; Klimanderung; Grundwasserstrmungs- und reaktive
Transportmodellierung; stochastisch-konvektiver reaktiver Transport; Flusswasserstandsverteilung; partikulres organisches Material; Sauerstoffzehrung
Mots cls
Filtration sur berge; restauration de rivire; changements climatiques; modlisation numrique
dcoulement et de transport souterrain; transport stochastique-convectif ractif; distribution
des niveaux de rivire; matire organique particulaire; consommation en oxygne dissous
Content
Chapter 1
Introduction .......................................................................................................... 1
1.1
Motivation and background ......................................................................................... 1
1.2
Objectives and structure of this thesis ......................................................................... 8
Chapter 2
New methods to estimate 2D water level distributions of dynamic rivers ......... 11
2.1
Introduction ............................................................................................................... 12
2.2
Interpolation methods to estimate river water levels ................................................. 13
2.3
Application to the Niederneunforn field site ............................................................. 17
2.4
Discussion .................................................................................................................. 22
2.5
Conclusions ............................................................................................................... 23
2.6
Supporting information.............................................................................................. 25
Chapter 3
Assessing the effect of different river water level interpolation schemes on

modeled groundwater residence times ............................................................... 37
3.1
Introduction ............................................................................................................... 38
3.2
Interpolation methods ................................................................................................ 39
3.3
Method implementation ............................................................................................. 41
3.4
Groundwater flow and transport model ..................................................................... 47
3.5
Results and discussion ............................................................................................... 51
3.6
Conclusions ............................................................................................................... 58
3.7
Chapter 4
NOM degradation during river infiltration: Effects of the climate variables

temperature and discharge .................................................................................. 65
4.1
Introduction ............................................................................................................... 66
4.2
Materials and methods ............................................................................................... 68
4.3
4.4
Conclusions ............................................................................................................... 84
vii
4.5
Chapter 5
Modeling the dynamics of oxygen consumption during riverbank filtration by a

stochastic-convective approach .......................................................................... 89
5.1
Introduction ............................................................................................................... 90
5.2
Theory........................................................................................................................ 92
5.3
Materials and methods ............................................................................................... 95
5.4
5.5
Conclusions ............................................................................................................. 111
Chapter 6
Conclusions and Outlook ................................................................................. 115
6.1
Conclusions ............................................................................................................. 115
6.2
Outlook .................................................................................................................... 120
Bibliography ........................................................................................................................... 123
viii
Chapter 1
Introduction
1.1 Motivation and background
Riverbank filtration provides an inexpensive, sustainable and yet effective means to improve
the quality of surface water (Tufenkji et al., 2002). During passage through the riverbed and
the aquifer, the infiltrated river water is exposed to processes like sorption, physicochemical
filtration and biodegradation. These natural attenuation processes efficiently remove
suspended particles, bacteria, viruses, parasites, micropollutants, and other organic and
inorganic compounds usually present in surface waters, such as natural organic matter and
ammonium (Kuehn and Mueller, 2000; Sacher and Brauch, 2002). The effectiveness of
riverbank filtration has long been recognized in Europe, where several countries cover
considerable fractions of their drinking water demand by riverbank filtration (~50% in France
and Slovak Republic, ~48% in Finland, ~45% in Hungary, ~25% in Switzerland and ~16% in
Germany (Hiscock and Grischek, 2002; Tufenkji et al., 2002)). Hence, riverbank filtration
provides an important drinking water resource, which needs to be maintained and protected,
especially within the context of upcoming challenges such as river restoration and climate
change. Even though experience of more than a century in the operation of riverbankfiltration systems exists, the current understanding of bank-filtration schemes is primarily
based on empirical knowledge (Hiscock and Grischek, 2002). Therefore, the adequate
management of riverbank-filtration systems in a changing environment requires a better
understanding of the physical and biogeochemical processes that occur during riverbank
filtration (Hoehn and Meylan, 2009), which is the motivation for this Ph.D. Thesis.
1.1.1
Riverbank filtration
Riverbank filtration can occur under natural conditions or can be induced by a lowering of the
groundwater table below surface water levels either by hydraulic boundaries such as side
channels, or by groundwater abstraction at pumping wells (Fig. 1.1). Infiltration occurs
mostly under saturated conditions, in direct hydraulic connection to the groundwater table
(Hiscock and Grischek, 2002). If the groundwater table is more than a few decimeters below
the riverbed, the hydraulic connection is lost and river water infiltrates through an unsaturated
zone, which is aerated (Brunner et al., 2009; Hoehn and Meylan, 2009). Riverbank-filtration
systems are typically positioned in alluvial valley aquifers, which predominantly consist of
1
gravels and sands allowing for high abstraction rates. However, floodplain deposits are
complex geologic systems that exhibit heterogeneity on different scales and can include
highly conductive open framework gravels forming preferential groundwater flow paths, as
well as layers of silt and clay (Huggenberger et al., 1998).
Besides the dilution of the freshly infiltrated river water with ambient groundwater at the
pumping well, the residence time of the bank filtrate has been identified as one of the key
factors determining the removal efficiency of riverbank filtration. Several studies have shown
that concentrations of natural organic matter and organic contaminants decreased with
increasing travel time (Sontheimer, 1980; Grnheid et al., 2005). Also, the removal of
pathogenic microbes was found to be strongly related to the residence time. A breakthrough
of E. Coli at a pumping station at the Rhine River was observed after a flood event, due to
shortened groundwater residence times (Eckert and Irmscher, 2006). Therefore, to ensure an
adequate quality of produced water, minimal residence times are often required by law and
are regulated via the definition of protection zones (FOEN, 2012b). The required minimal
groundwater residence time between the river and a pumping well is 10 d in Switzerland
(BUWAL, 2004) and 50 d in Germany (DVGW, 2000).
Infiltrating river water is subject to strong geochemical changes that are related to redox
reactions, dissolution/precipitation of minerals, ion exchange and gas exchange (Jacobs et al.,
1988; Tufenkji et al., 2002). Numerous studies have demonstrated that the most significant
geochemical changes were associated with the biodegradation of natural organic matter
(Jacobs et al., 1988; Bourg and Bertin, 1993; Brugger et al., 2001b; Sobczak and Findlay,
2002; Tufenkji et al., 2002). These degradation reactions were found to occur dominantly
within the first meters of infiltration, where the microbial abundance and activity was highest
(von Gunten et al., 1994; Brugger et al., 2001a). This interface between the river and alluvial
groundwater, the hyporheic zone, has been identified as a distinct environment playing a
crucial functional role in the biogeochemical cycling of nutrients and organic matter (Brunke
and Gonser, 1997; Boulton, 2007).
Natural organic matter (NOM) in river systems originates from both allochthonous
(terrestrially-derived) sources and generally more biodegradable autochthonous sources
(periphyton) (Pusch et al., 1998; Leenheer and Croue, 2003). NOM is composed of dissolved
organic matter (DOM) and particulate organic matter (POM). During infiltration of river
water, DOM is transported through the riverbed as a mobile substrate, whereas POM is
retained in the riverbed sediments as a stationary substrate (Pusch et al., 1998). The
2
retention and storage of POM within the riverbed was found to depend on the grain-size
distribution of the riverbed sediments and the hydrologic conditions (Brunke and Gonser,
1997).
The microbial degradation of NOM in riverbed sediments leads to a consumption of dissolved
or solid terminal electron acceptors such as oxygen (O2), nitrate (NO3-) and Mn(III/IV)/Fe(III)(hydr)oxides. In case of oxygen depletion, the redox sequence proceeds to
denitrification followed by reductive dissolution of manganese and ferric oxyhydroxides
releasing dissolved Mn(II) and Fe(II). These species are undesired and need to be removed
from pumped water by additional cost-intensive treatment steps (Kuehn and Mueller, 2000).
Furthermore, re-oxidation of dissolved Mn(II) and Fe(II) when mixing with aerobic water at
the pumping well can cause clogging of the well screen. It also has been recognized that redox
conditions can significantly influence the removal efficiency of riverbank filtration. The
degradation of DOM and several organic micropollutants was found to be less effective under
anoxic conditions (Schwarzenbach et al., 1983; Grnheid et al., 2005; Massmann et al., 2006;
Maeng et al., 2010).
The temperature dependence of the microbially mediated NOM degradation is well known
(O'Connell, 1990; Kirschbaum, 1995). As a result, redox conditions at various bank-filtration
systems were observed to undergo seasonal variations with the occurrence of anoxic
conditions in summer (von Gunten et al., 1991; Greskowiak et al., 2006; Massmann et al.,
2006; Massmann et al., 2008). Besides temperature, the redox conditions in the infiltration
zone are also influenced by the availability of electron donors (NOM, ammonium) and
electron acceptors (dissolved oxygen, nitrate). During industrialization in the 1950s and 1960s,
rivers in urban catchments carried higher loads of organic substances and ammonium (Eckert
and Irmscher, 2006). Combined with lower oxygen concentrations in river water, strongly
reducing conditions prevailed in the infiltration zones of many riverbank-filtration systems,
which caused the mobilization of manganese and iron. Since the 1970s, river water quality
has improved significantly and oxic conditions in infiltration zones re-established (Kuehn and
Mueller, 2000).
The retention of fine suspended particles (<2 mm) during riverbank filtration contributes to
the clogging of the riverbed and the development of a colmation layer (Fig. 1.1), which is
characterized by a reduced permeability causing lower infiltration rates and higher residence
times of infiltrated river water (Brunke, 1999). The colmation layer may be subject to strong
spatial heterogeneity and temporal variability. Bed-moving floods can rework the riverbed
3
and cause a decolmation with an inherent increase in riverbed permeability (Schubert, 2002;
Doppler et al., 2007). The higher permeability during floods was found to promote the import
of POM into gravelly riverbed sediments, which in turn enhanced the microbial respiration
rate (Naegeli et al., 1995). The colmation of the riverbed can have a positive effect on the
removal of organic and inorganic pollutants and bacteria due to higher residence times
(Hiscock and Grischek, 2002). On the other hand, a colmation layer may adversely affect
groundwater quality, as higher residence times also favor the development of anoxic
conditions and the mobilization of Mn(II) and Fe(II) (Tufenkji et al., 2002).
Fig. 1.1. Schematic illustration of riverbank filtration. Infiltrated river water passes the
riverbed and the aquifer and is extracted at a pumping station, where it mixes with ambient
groundwater and may undergo secondary treatment steps before being supplied to the
drinking water distribution system. Adapted from Schirmer (2013).
1.1.2
Challenge 1: River restoration
Over the past few decades, there has been a shift in perspective of water authorities from river
corridor channelization toward restoration. As a consequence, several laws have been
established such as the European Water Framework Directive (European-Commission, 2000)
or the Swiss Water Protection Law (GSchG, 2011; GSchV, 2011) that require engineering
tasks in river courses to jointly improve flood protection, ecological status and water quality.
In Switzerland, 15000 km of river length are strongly engineered, channelized and in a bad
ecological condition, of which 4000 km will undergo river restoration during the next 80
years (FOEN, 2012c).
River restoration measures may comprise removal of the bank stabilization and the colmation
layer, widening of the riverbed, as well as the establishment of braided river sections, remeandering stream reaches and gravel bars (Fig. 1.2). These actions enhance the exchange
between the river and the hyporheic zone and increase the habitat diversity, both being
essential for the ecological health of a river system (Brunke and Gonser, 1997; Baumann et al.,
2009). The higher exchange with the hyporheic zone is thought to increase the interstitial
microbial activity and therefore the removal of various contaminants. However, the widening
of the riverbed and the higher riverbed permeability might decrease groundwater residence
times and therewith increase the risk of drinking water contamination by pollutants or bacteria
(Hoehn and Scholtis, 2011). Due to a lack of process understanding and following the
precautionary principle, river restoration measures are not allowed within the protection zones
of drinking water wells (FOEN, 2012b).
To mitigate this conflict of interest between river restoration and drinking water protection, a
good knowledge of the physical and biogeochemical processes is needed. More specifically,
the well-capture zone, groundwater flow paths, flow velocities, groundwater residence times
and the fraction of freshly infiltrated river water need to be known (Hoehn and Meylan, 2009).
Several methods based on natural tracers such as temperature and electrical conductivity were
developed to quantify exchange rates (Anderson, 2005; Kalbus et al., 2006; Vogt et al.,
2010b), travel times and the fraction of riverbank filtrate in abstracted water (Hoehn and
Cirpka, 2006; Cirpka et al., 2007; Vogt et al., 2010a). However, these methods are not able to
provide information on groundwater flow paths and flow velocities. To this end, groundwater
flow and transport modeling is a powerful tool, as it provides a quantitative link between
groundwater residence times, flow paths and flow velocities (Wondzell et al., 2009).
It is well known from modeling studies and field observations that riverbed morphology
affects the river water level distribution, which in turn affects or drives the exchange with
groundwater (Harvey and Bencala, 1993; Woessner, 2000; Cardenas et al., 2004; Cardenas,
2009; Kser et al., 2009). Accordingly, one of the prerequisites for the setup of a groundwater
flow model of a real river-groundwater system is an accurate description of the water level
distribution in the river (Kser et al., 2013). However, restored river systems may have
complex water level distributions that need to be characterized by their full spatial (i.e. two
horizontal dimensions) and temporal extent (i.e. for any discharge condition). Ideally, such
water level information is extracted from hydraulic models (Doppler et al., 2007; Derx et al.,
2010; Engeler et al., 2011), but their setup is time consuming and requires a considerable
amount of data input, that is, the riverbeds bathymetry and water level information for the
5
calibration and validation process. Therefore, there is a need for alternative methods to
describe surface water level distributions of dynamic rivers that allow for reliable simulations
of the groundwater flow field and accurate predictions of groundwater residence times.
Additionally, the sensitivity of these predictions on the method selection and the level of
detail in the water level distribution related to these methods should be assessed.
Fig. 1.2. Schematic illustration of a restored riverbank-filtration system. The riverbed was
widened, the colmation layer removed and gravel bars were established. These river
restoration measures promote hyporheic exchange (orange arrows) and the development of
new habitats. On the other hand, flow paths between the river and the pumping well may be
shortened, which can cause a contamination of the drinking water by pollutants or bacteria.
Beaver dams in side channels can lead to higher water levels that may reverse the hydraulic
gradient. Adapted from Schirmer (2013).
1.1.3
Challenge 2: Climate change
Today, most peri-alpine alluvial gravel-and-sand aquifers are oxic owing to the low loads of
organic matter and nutrients in rivers. Oxic conditions allow a beneficial minimal treatment of
pumped groundwater before supplying it as drinking water to the distribution system.
However, during the hot summer of 2003, which was accompanied by low river discharges
and river temperatures of more than 25C (OcCC, 2005), the redox conditions in numerous
riverbank-filtration systems turned anoxic over a period of up to three months (Eckert et al.,
2008) and the onset of denitrification was observed (Sharma et al., 2012). Denitrification may
result in elevated concentrations of nitrite (NO2-), which is of concern due to its toxicity.
Hoehn and Scholtis (2011) even reported a case where the redox sequence proceeded to
Mn(IV)- and Fe(III)-reducing conditions. As mentioned in Section 1.1.1, the appearance of
dissolved Mn(II) and Fe(II) in the abstracted water requires additional treatment processes,
and precipitation of Mn(IV)-/Fe(III)(hydr)oxides can lead to clogging of the filter screen.
6
Furthermore, compounds that are better degraded under oxic conditions such as DOM,
ammonium and certain micropollutants are more persistent under anoxic conditions and may
appear at the pumping well in elevated concentrations (Section 1.1.1).
For Europe, climate models predict an increase in air temperature, a decrease in summer
rainfall and an increase in winter rainfall, as well as an increased frequency and intensity of
weather extremes such as heat waves or flood events (IPCC, 2007), which is consistent with
the climate-change scenarios for Switzerland (CH2011, 2011). River discharges are expected
to decrease in summer and increase in winter. River water temperature, which is decisive for
the degradation processes within the riverbed, is likely to increase by the same extent as air
temperature, especially during low-flow conditions (FOEN, 2012a).
The effects of climate-related changes in temperature and discharge on the quality of surface
water and groundwater are manifold (Murdoch et al., 2000; Zwolsman and van Bokhoven,
2007; Park et al., 2010; Green et al., 2011). Riverbank-filtration systems are most vulnerable
to changes in residence time and redox conditions (Section 1.1.1). An increased occurrence of
flood events may raise the risk of drinking water contamination by pathogenic bacteria or
other pollutants. However, flood events are temporally limited to a few days, facilitating their
management.
Low discharges are expected to increase groundwater residence times, as well as the loads of
organic matter and nutrients in rivers due to less dilution of wastewater effluent (Sprenger et
al., 2011). Combined with higher temperatures, summer heat waves tend to promote the
development of anoxic conditions in the infiltration zone, as observed in summer 2003. An
increased frequency and intensity of long-lasting heat waves is therefore likely to increase the
risk for riverbank-filtration systems to become denitrifying or even Mn(IV)- or Fe(III)reducing (Fig. 1.3) over periods of several months (Sprenger et al., 2011). Such long-lasting
anoxic conditions would require complementary drinking water treatment technologies, the
implementation of which is expensive and needs to be planned in advance to ensure the
supply of high-quality drinking water. To assess the risk of arising anoxic conditions more
accurately, it is crucial to understand the dynamics of NOM degradation during riverbank
filtration and its dependence on the climate-related variables temperature and discharge. The
quantification of these dependencies might allow for model-based estimates of oxygen
concentrations in river-recharged groundwater, which would facilitate the anticipation of
possible changes in redox conditions under various climatic and hydrologic conditions.
Fig. 1.3. Higher temperatures during future summer heat waves will enhance the microbial
degradation of NOM, which might lead to a complete consumption of oxygen. In case of
anoxic conditions, denitrification and eventually reductive dissolution of manganese and
ferric oxyhydroxides can take place. The occurrence of nitrite and especially dissolved
manganese and iron would require additional drinking water treatment steps. Furthermore, the
re-oxidation of dissolved manganese and iron at the pumping well can lead to a clogging of
the filter screen. Adapted from Schirmer (2013).
1.2 Objectives and structure of this thesis

The overall goal of this Ph.D. Thesis is to deepen the understanding of physical and
biogeochemical processes affecting the quality of river-recharged groundwater, and to
develop tools that may help to assess potential impacts of river restoration measures and
climatic changes on the effectiveness of riverbank-filtration systems. This overall goal of the
thesis was subdivided into four main objectives (i-iv), which will be addressed in Chapters 25, respectively. To achieve these objectives, an integrative approach was employed that
combines field investigations, laboratory experiments, as well as groundwater flow and
reactive transport modeling.
This thesis was written within the RIBACLIM project (RIverBAnk filtration under CLIMate
change scenarios) in the framework of the National Research Program Sustainable Water
Management (NRP61). Field investigations, method development and modeling work were
performed at a partly restored field site in Niederneunforn (northeastern Switzerland) at the
peri-alpine Thur River. Chapters 2-5 contain detailed descriptions of the field site.
Within the context of river restoration, the objectives of this Ph.D. Thesis were (i) to develop
alternative methods to generate water level distributions of dynamic rivers for the application
in groundwater models, and (ii) to assess the predictive capability of these methods with
respect to the simulated groundwater residence time.
Chapter 2 presents two new alternative interpolation methods to estimate water level
distributions of highly dynamic rivers in the context of modeling riverbank-filtration systems
at scales in the order of kilometers. These methods were implemented at the Niederneunforn
field site and water level time series at multiple locations within the river reach were
generated. These water level predictions were compared with those of a third reference
method, which is based on an existing hydraulic model, to test the accuracy of the alternative
interpolation methods.
Chapter 3 investigates the impact of the method selection and the considered level of detail in
the river water level distribution on the simulated groundwater residence time. Thereto,
steady-state surface water level distributions were generated using both alternative methods,
the reference method, as well as two simplified methods, and were assigned to a 3D
groundwater flow and transport model of the field site. After calibration against groundwater
heads, each of the models was used to predict the spatial groundwater residence time
distribution within the modeling domain.
Within the context of climate change, the objectives of this Ph.D. Thesis were (iii) to
investigate the dependence of NOM degradation during riverbank filtration on the climaterelated variables temperature and discharge and (iv) to develop a model that simulates oxygen
consumption during riverbank filtration as a function of river temperature and discharge.
Chapter 4 elucidates the contribution of DOM consumption to the overall consumption of
electron acceptors and its dependence on the climate-related variables temperature and
discharge. Two detailed groundwater sampling campaigns were performed during typical
summer and winter conditions to capture different temperature ranges. The results were
compared to those of laboratory column experiments that were conducted at temperatures that
span the range of the field conditions. Additionally, data of periodic field samplings that
covered a wide range of temperature and discharge conditions over a period of five years were
evaluated.
Chapter 5 presents a newly developed semi-analytical model that allows estimating oxygen
concentrations in river-recharged groundwater under various climatic and hydrologic
conditions. The model is based on the stochastic-convective reactive approach and
incorporates a dependence of the oxygen consumption rate on river temperature and discharge.
These dependencies were inferred from high-resolution oxygen time series measured in the
Thur River and an adjacent observation well.
Chapter 2
New methods to estimate 2D water level distributions of
dynamic rivers
Published in Ground Water
Diem, S., Renard, P., Schirmer, M., 2012. New methods to estimate 2D water level
distributions of dynamic rivers. Ground Water, doi: 10.1111/gwat.12005.
Abstract
River restoration measures are becoming increasingly popular and are leading to dynamic
riverbed morphologies that in turn result in complex water level distributions in a river.
Disconnected river branches, nonlinear longitudinal water level profiles and morphologically
induced lateral water level gradients can evolve rapidly. The modeling of such rivergroundwater systems is of high practical relevance in order to assess the impact of restoration
measures on the exchange flux between a river and groundwater or on the residence times
between a river and a pumping well. However, the model input includes a proper definition of
the river boundary condition, which requires a detailed spatial and temporal river water level
distribution. In this study, we present two new methods to estimate river water level
distributions that are based directly on measured data. Comparing generated time series of
water levels with those obtained by a hydraulic model as a reference, the new methods proved
to offer an accurate and faster alternative with a simpler implementation.
11
2.1 Introduction
Over the past few decades there has been a shift in focus from river corridor channelization
toward restoration. It has been recognized that rivers need more space for the purpose of flood
protection (Woolsey et al., 2007). Furthermore, restoration measures such as widening of the
riverbed, re-meandering stream reaches, and constructing gravel bars, should increase the
exchange between rivers and groundwater, which is essential for the ecological health of a
river (Brunke and Gonser, 1997). On the other hand, riverbed widening can decrease travel
times between rivers and pumping wells, which may increase the risk of drinking water
contamination by pollutants or bacteria (Hoehn and Scholtis, 2011).
Groundwater flow and transport modeling is a valuable tool to obtain a process-based
understanding of (restored) surface water-groundwater systems. Compared to tools developed
to work with artificial and natural tracers, a calibrated model provides quantitative
conclusions on flow paths, mixing ratios and travel times (Wondzell et al., 2009). It is well
known that riverbed morphology affects the river water level distribution, which in turn
affects or drives the exchange with groundwater (Woessner, 2000; Cardenas et al., 2004;
Cardenas, 2009). Therefore, one of the prerequisites for the setup of a groundwater flow
model of a real river-groundwater system is an accurate description of the water level
distribution in the river.
Restored river systems may have complex water level distributions that need to be
characterized by their full spatial (i.e. two horizontal dimensions) and temporal extent (i.e. for
any discharge condition). Past small-scale field and modeling studies (10 to 100 m) applied
one or two sets of detailed water level measurements in one or two dimensions (Wroblicky et
al., 1998; Storey et al., 2003; Lautz and Siegel, 2006; Wondzell et al., 2009). On the scale in
the order of kilometers, which is more relevant for practical problems, this approach is not
applicable. Instead, the extraction of river water level information from a hydraulic model
might be a proper solution (Doppler et al., 2007; Derx et al., 2010; Engeler et al., 2011).
However, the setup of a hydraulic model is time consuming and requires a large amount of
data input, that is, the riverbed bathymetry and water level information for the calibration and
validation process.
In this study, we present two alternative methods to estimate water level distributions of
highly dynamic rivers in the context of modeling river-groundwater systems at scales in the
order of kilometers. The two methods combine continuous and periodic water level
measurements from different locations in order to account for the spatial and temporal
12
variability of the water level distribution. We predicted water levels at several locations at a
restored reach of the peri-alpine Thur River and compared them with water level predictions
of a reference method, which is based on an existing hydraulic model. Finally, we point out
the advantages and disadvantages of the different methods and discuss the optimal use of each
method when assigning the water level distribution of restored and dynamic river systems to
groundwater flow and transport models.
2.2 Interpolation methods to estimate river water levels

2.2.1
Problem setup
Instrumentation of a 1 km river reach is typically comprised of two water level gauges, one
upstream and one downstream of the field site. For groundwater flow modeling, river water
levels need to be estimated at each river boundary node between the two gauging stations.
The simplest method assumes a constant gradient (linear interpolation) in the longitudinal
direction and a zero gradient in the lateral (transverse) direction. Although this might be
adequate for a channelized system, for restored corridors the dynamic riverbed morphology
might lead to a nonlinear longitudinal water level profile and to lateral water level gradients.
Furthermore, the water level distribution might change as a function of the discharge.
The basic idea of the two new interpolation methods is to combine continuous water level
records from water level gauges with water levels measured periodically under different
discharge conditions. The latter are measured at fixpoints, which are distributed throughout
the river reach to refine the water level distribution at locations where the installation of a
water level gauge is technically difficult or simply too expensive. A fixpoint is defined as a
reference point in the river, for example, a point on a construction rock or a steel rod, whose
altitude is known. By establishing a mathematical relationship between the water level data at
the gauging stations and the fixpoint, the water level at the fixpoint can be estimated for any
measured water level at the gauging station.
We consider the river as a two-dimensional (2D) domain, which is discretized by multiple
lines parallel to the main flow direction of the river and several sections of support points ( )
perpendicular to the flow direction (Fig. 2.1). The key task of the interpolation methods is to
estimate a water level
at each support point from any water level measured at the gauging
station (i.e. for any discharge condition). Support points are placed at a location where a
fixpoint ( ) or a gauging station ( ) exists. One fixpoint per section is enough unless lateral
water level gradients are observed, in which case a fixpoint must be defined on both sides of
13
the river (Fig. 2.1). The periodic water level measurements at the fixpoints are denoted as
while the continuous water level measurements at the gauging station are denoted as
estimation of
. The
consists of three steps:
1. Establish a mathematical relationship

2. Use
to compute
between the measurements
and
for any time step or time series:
3. Estimate the water levels or water level time series
for the support points located
on the same section as the fixpoint .

Different options for these steps have been considered for the two new methods, which are
referred to as alternative methods. In order to compare these alternative methods with a
reference, we developed a third method that is based on a hydraulic model and is referred to
as the reference method. These three methods are described in the following sections. The
explanation is descriptive in order to convey the main idea. A thorough mathematical
development of the methods and their application to real data is presented in the Supporting
information (Section 2.6).
Fig. 2.1. Schematic representation of a river system with multiple lines and sections of
support points ( , black dots). Gauging stations ( ) and fixpoints ( ) are shown as black
circles.
Once the set of lines and support points with water levels
has been obtained using one of
the three methods, the final interpolation of the water levels from the support points to the
river boundary nodes of the numerical model has to be performed. This step is identical for all
three methods and is accomplished by a one-dimensional linear interpolation along the lines,
which implies that each of the lines is mapped by a curvilinear system of longitudinal
coordinates to account for different curvatures of the lines. More precisely, each of the river
boundary nodes of the groundwater model is projected perpendicularly onto the closest line
and the water level is linearly interpolated between the upstream and downstream support
point. Some simulation codes (e.g. FEFLOW) offer tools to accomplish this final interpolation
step.
14
2.2.2
Method 1: Regression of measured data (RM)
The first alternative method uses a polynomial regression technique to obtain the
mathematical relationship . The method requires a continuous water level time series at one
gauging station
and periodic water level measurements at a fixpoint
(Fig. 2.2). Each time
a measurement is made at the fixpoint, the corresponding water level at the gauging station
can be extracted from the continuous water level time series. A polynomial equation is then
fitted to the data pairs, which constitutes the mathematical relationship . The polynomial
order has to be chosen according to the range and the characteristics of the data. A guideline
to defining the polynomial order is given in the Supporting information (Section 2.6).
Applying
to any water level time series measured at the gauging station produces
predictions of corresponding water level time series at the fixpoint. If there are two fixpoints
on the same section to capture lateral gradients, a separate polynomial equation is fitted to the
corresponding data pairs. The gauging station
is denoted as determining gauging station
as its water level uniquely defines the water level at the fixpoint.
Fig. 2.2. Illustration of the regression approach. (a) Continuous water level time series
measured at the gauging station. (b) Periodic water level measurements
at a fixpoint (black
dots). For the same measurement times, water levels at the gauging station are extracted in (a).
(c) is established between
and
by polynomial regression and is used to estimate the
water level time series (dashed line) at the fixpoint in (b).
15
The estimation of the water level at the support points from the water level at the fixpoint is
made in the simplest possible manner. If no lateral gradients exist, the water level of the
fixpoint is assigned to all of the support points located on the same section. However, if a
second fixpoint was installed on the same section to capture a lateral gradient, assigning the
water levels to the support points should be based on field observations. For example, if a
discrete step forms the lateral gradient, the water level at each support point can be
determined from the most representative fixpoint. Otherwise, a linear interpolation between
the two fixpoints might be an appropriate solution.
2.2.3
Method 2: Interpolation of measured data (IM)
The second alternative method uses an interpolation approach that requires two gauging
stations. One of them has to be defined as determining gauging station
. The fixpoints can
be located between or outside of the two gauging stations.

This IM method is based on a conceptual behavior model. According to the model, the
riverbed morphology exerts a high influence on the river water level distribution under lowflow conditions, potentially leading to nonlinear longitudinal water level profiles or lateral
water level gradients. As the water level rises, the influence of the riverbed morphology on
the water level distribution decreases, and at some point lateral water level gradients
disappear and a linear longitudinal water level profile is reached. In general, the IM method
describes the water level at a fixpoint by an interpolation function ( ) that consists of a linear
interpolation term between the water levels at the two gauging stations (
and
) and a
deviation
from the linear trend (Fig. 2.3). Following the conceptual behavior model, the
deviation
is at a maximum,
are smaller than
, when water levels at the determining gauging station
. On the other hand, the deviation goes to zero for water levels higher
than a threshold water level
. Between
and
, the deviation
is assumed to
decrease linearly (Fig. 2.3). Similarly, we can describe the water level at a second fixpoint on
the same section by the water level of the first fixpoint plus a lateral difference
. Again,
is at a maximum for low water levels and goes to zero above a threshold water level.
To estimate the parameters ,
and
, the deviation from the linear trend has to be
calculated for each measurement at a fixpoint and plotted against the corresponding water
level at the determining gauging station. More details on how to estimate the parameters for
the IM method are given in the Supporting information (Section 2.6). The estimation of the
water level at the support points is identical to the procedure described in the previous section
for the RM method.
16
Fig. 2.3. Illustration of the interpolation approach. (a) Two longitudinal water level profiles
for the conditions
and
. (b) Deviation from the linear trend ( ) as a
function of
.
2.2.4
Method 3: Regression of hydraulic model data (RH)
To compare the two alternative methods with an independent reference, we developed a third
method that is based on data from an existing hydraulic model of the main river channel in the
section of interest. The hydraulic model output consists of 2D water level distributions, each
corresponding to a specific discharge condition. Similar to the RM method, the RH method
applies a polynomial regression technique to obtain the mathematical relationship
However, the relationships are based on water levels extracted from the hydraulic model
output at each support point and at the location of the determining gauging station, and not on
measured water levels.
2.3 Application to the Niederneunforn field site

2.3.1
Field site
The peri-alpine Thur River drains a catchment area of 1730 km2 and originates in an alpine
region that reaches its highest point on Mount Sntis (2502 m above sea level, m asl). The
Thur River is the largest river in Switzerland without a retention basin. This leads to a very
dynamic discharge regime ranging from 3 to 1100 m3/s with an average of 47 m3/s. The field
site (Fig. 2.4) is located approximately 12 km upstream of the confluence with the Rhine
River. In the western part of the field site, restoration measures were realized in 2002.
Restoration measures were forbidden in the upstream section to protect the water quality at
the nearby pumping station, where a pumping well supplies the community of Nieder- and
Oberneunforn with drinking water. The field site was instrumented with more than 80
piezometers (2) during the interdisciplinary RECORD project (Restored corridor dynamics,
http://www.cces.ethz.ch/projects/nature/Record; Schneider et al. (2011)). The aquifer has a
17
thickness of 5.31.2 m and its hydraulic conductivities were estimated to range from 410-3 to
410-2 m/s (Diem et al., 2010; Doetsch et al., 2012). The silty sand of the alluvial fines on top
of the aquifer has a much lower hydraulic conductivity and can be regarded as the semiconfining unit, with a thickness of 0.5-3 m.
The Thur River has a width of 50-100 m (Fig. 2.4). In the restored section a large gravel bar
has evolved during the past few years. To the north, a disconnected branch of the river exists,
which is only flooded at high river stages (>200 m3/s). The longitudinal river water level
profile does not have a linear shape for low-flow conditions. In the upstream 400 m of the
river the gradient is 0.5 and in the downstream 800 m it is 2. In the middle of the river
reach, lateral water level gradients occur during low-flow conditions. These lateral surface
water level differences exist due to the asymmetrical riverbed morphology and can reach up
to 0.4 m. Two side channels (north and south) flow parallel to the river and have an average
width of 4-8 m. Two beaver dams are located in the northern side channel. The upstream dam
has a significant effect on water levels, creating differences of up to 0.5 m.
A 2D hydraulic model of the Thur River was developed by Pasquale et al. (2011) and forms
the basis for the reference RH method (Section 2.2.4). Riverbed cross sections measured in
September 2009 with an average spacing of 50 m were interpolated using the technique
presented by Schppi et al. (2010) to obtain the river bathymetry. The hydrodynamic
simulations were performed using the 2D model BASEMENT, which applies the finite
volume method to integrate the shallow water equations (Pasquale et al., 2011). The model
deals with both sub- and supercritical flow regimes, thus providing a suitable tool to simulate
2D water level distributions of dynamic rivers. The modeling results comprise 19 steady-state
simulations for flows ranging between 10 and 650 m3/s, and provide water level altitudes at
each raster cell (22 m). The hydraulic model does not include the side channels and the
disconnected branch and is considered to be valid until the major flood events of June 2010.
Fig. 2.4. Field site located in Niederneunforn at the Thur River with indicated lines, support
points, fixpoints and gauging stations. Selected piezometers are shown as well. The white
polygon shows the perimeter of method implementation.
18
2.3.2
Data collection and method implementation
We installed two water level gauges in the main channel of the river, as well as in both side
channels, and one in the river branch (Fig. 2.4). Three of these water level gauges are
maintained by the Agency for the Environment of the canton Thurgau. The sensors (DL/N 70,
STS AG, Switzerland) have been continuously measuring pressure, temperature and electrical
conductivity (EC) at 15-min intervals since April 2010 (error of single measurement: 0.1%
for pressure, 0.25% for temperature and 2% for EC, according to the manufacturers
manual). We installed sensors of the same type in selected piezometers shown in Fig. 2.4. The
raw data were processed in order to remove outliers, to subtract the barometric air pressure
and to transform the pressure data to absolute water levels (m asl). To have more information
on the water level distribution between the water level gauges, we installed several fixpoints
along the river and the northern side channel. We chose the fixpoint positions according to the
location of piezometer transects and visibly steep water level gradients (e.g. hydraulic jumps
at beaver dams) or lateral gradients (central part of the river). We defined an indexing system
that allows a distinct identification of each point. The first index refers to the section number
and the second index to the line number. We leveled the absolute height of the fixpoints using
a high-precision differential GPS (Leica GPS1200, Leica Geosystems AG, Switzerland) and a
leveling device (Sprinter 100M, Leica Geosystems AG, Switzerland). We measured water
levels at these fixpoints periodically between February and May 2011 covering a discharge
range of 10 to 100 m3/s.
Based on the resulting data set, we implemented both alternative methods and the reference
method, each of which covered the three steps described in Section 2.2.1. As the hydraulic
model did not include the disconnected branch and the side channels, we coupled the RH
method to the RM method. More details on the method implementation can be found in the
Supporting information (Section 2.6).
2.3.3
Comparison of the interpolation methods
To evaluate the performance of the two alternative methods, we applied each of the
interpolation methods (RM, IM, and RH) to generate water levels for a 1-month period (May
26, 2010 to June 30, 2010, vertical lines in Fig. 2.5) at each fixpoint and gauging station
within the river domain. The upstream gauging station
was used as the determining
gauging station for each of the methods. We plotted the generated time series of water levels
(30-min intervals) of the reference method (RH) against the time series of both alternative
methods (RM and IM). Fig. 2.6 shows examples for one gauging station and two fixpoints. If
19
the generated water levels were the same for each time step, the dots of the scatter plots would
be located on a line with a slope of one. Deviations from this line correspond to deviations of
the RM/IM methods from the RH method and were quantified by a mean and a standard
deviation (Fig. 2.6, Table 2.1).
Fig. 2.5. (a) Water level time series in the Thur River at fixpoint
predicted with the RH
method and measured groundwater heads in a nearby piezometer between May and October
2010. (b) Calculated water level difference between the river and groundwater. The vertical
lines indicate the 1-month period used for creating the scatter plots of Fig. 2.6.
The scatter plots for the gauging station
are identical for the RM and the IM method,
which both used the gauging station data directly for this point. The water levels of the RH
method however, were generated based on the hydraulic model. Therefore, these two plots
actually compare the measured water level data with the water level predictions of the
hydraulic model. The match was good for lower water levels but deviations of up to 0.5 m
occurred for the peak flows during flood events. As only a small portion of the data was
subject to such large errors, the mean error of water level prediction with the RH method was
small (1 cm). The uncertainty of the water level prediction is reflected in the standard
deviation of 10 cm (Table 2.1).
The scatter plots of the fixpoint
revealed a difference in the behavior of the two
alternative methods. The water level predictions for the study period (Fig. 2.5) exceeded the
range of water levels measured at
(371.6-372.5 m asl). Depending on the polynomial fit,
the RM method is likely to fail for predictions outside of the measured range, as there are no
data points constraining the polynomial equation. In our case, the parabola that was fitted to
the data at the fixpoint
was obviously too narrow in order to reliably predict water levels
beyond the measured range. Correspondingly, the deviations with respect to the RH method
20
showed a mean error of -11 cm and a standard deviation of 42 cm. The IM method performed
better for the peak flow water levels. As the water level predictions at a fixpoint are always
bounded by two measured water levels at the upstream and downstream gauge, the stability
beyond the measured range was better for the IM method. For the fixpoint
, both the RM
and the IM method performed well despite the systematic positive offset for the RM method
at high water levels. This offset might be attributed to the problem described above for the
fixpoint
Most of the mean errors and their standard deviations for the fixpoints within the river domain
varied between 1 and 10 cm (Table 2.1). We considered this level of error to be acceptable as
the reference RH method itself had an accuracy of 10 cm at
. As
was the determining
gauging station for all fixpoints in the river, predictions were identical for all methods,
explaining the values of zero in Table 2.1. The high standard deviations for the fixpoints
and
for the RM method can be blamed on the failure of the regression approach beyond
the measured data.
Fig. 2.6. Scatter plots for one gauging station ( ) and two fixpoints ( ,
) within the
river domain. The time series of water levels generated by the reference method RH are
plotted against the time series of water levels generated by the RM (top three figures) and the
IM method (bottom three figures). Mean (error) and standard deviations (error) of the
residuals between the alternative methods and the reference method are indicated.
21
The alternative methods especially the IM method tend to show higher deviations (>10 cm)
from the RH method at the fixpoints on sections 1 to 4, which are located in the restored part
of the river. We found evidence that the major flood event on June 17 through 24, 2010 (Fig.
2.5) led to a change in the riverbed morphology in the restored river section and to a
corresponding change in the relationship between the water levels at the determining gauging
station and at the fixpoints. First, the scatter plot for the gauging station
(Fig. 2.6) reveals
two distinct regimes at the lower end of the water level spectrum for the 1-month period that
covers the major flood event. Second, after the flood event there is a clear and permanent
decrease (~10 cm) in the difference between the water levels at
predicted with the RH
method, and groundwater heads measured at a nearby piezometer (Fig. 2.5). The drop in the
water level difference could be due to a change in the riverbed morphology after the flood,
resulting in an underprediction of water levels by the RH method, which assumed a constant
morphology. Because the data for the RM and the IM method were collected after this
morphologically active flood event, they might have captured the higher water levels and
accordingly led to an overestimation of water levels compared to the RH method. Therefore,
the differences in predicted water levels among the different methods were presumably not
only caused by structural artifacts, but also by real differences due to morphology changes in
the river.
Table 2.1. Mean (error) and standard deviations (error) of the residuals [m] between the time
series of water levels generated by both alternative methods and the reference method at the
gauging stations and fixpoints within the river domain.
RM
error
error
IM
error
error
F18
0.07
0.12
F18
0.11
0.16
G21
-0.01
0.10
G21
-0.01
0.10
F36
0.05
0.04
F36
0.16
0.09
F31
0.06
0.04
F31
0.17
0.08
F46
-0.08
0.42
F46
0.12
0.08
F41
-0.11
0.42
F41
0.06
0.09
F51
-0.01
0.07
F51
-0.04
0.04
F66
0.06
0.05
F66
0.02
0.02
G76
0.00
0.00
G76
0.00
0.00
2.4 Discussion
A hydraulic model has the advantage of being physically based, which allows a wide range of
discharge conditions to be simulated. This makes water level predictions by the RH approach
very robust. However, setting up a hydraulic model is time consuming, and needs both a large
data set and to be thoroughly calibrated. At our field site, the hydraulic model was not able to
include the water levels of the disconnected river branch, because during low-flow conditions
this branch is fed by groundwater. Furthermore, the hydraulic model did not cover the side
22
channels. The RH method had to be coupled to the RM method in order to include the full
surface water level distribution required for the assignment of boundary conditions in a
groundwater model.
Compared to the RH method, both alternative methods presented in this chapter (RM and IM)
provide a more efficient way of predicting water level distributions in a hydraulically and
morphologically varying environment. First, the accuracy of the water level predictions with
the alternative methods was in the same range as the accuracy of the reference RH method
itself. Second, the alternative methods require minimal data and computational effort, making
them simpler and faster to implement than the hydraulic model.
In comparison to the IM method, the RM method benefits from the regression approach,
which is fast and straightforward in its implementation and its application. However, the RM
method does not provide reliable water level predictions when they exceed the range of water
levels measured at the fixpoints. This in turn is the strength of the IM method, whose water
level predictions are always bounded by water levels measured at two gauging stations.
Furthermore, the IM method is based on a conceptual behavior model that is physically
consistent. Even though the IM method is empirical and more complex in its implementation,
the measured data at most of the fixpoints supported the methods underlying assumptions.
2.5 Conclusions
Several field and modeling studies have shown that the water level distribution in rivers exerts
an important influence on the exchange between rivers and groundwater. In this study, we
presented two new methods to define spatial and temporal river water level distributions for
the purpose of modeling surface water-groundwater systems. The basic idea is to record water
levels continuously at water level gauges and measure water levels periodically under
different discharge conditions at fixpoints to refine the water level distribution at locations
where it is technically difficult or too expensive to install a gauging station. The RM method
applies a polynomial regression approach for the prediction of water levels at fixpoints as a
function of the corresponding water levels at the determining gauging station, while the IM
method uses an interpolation approach between two gauging stations. To compare these
alternative methods to a reference method, we developed a third method, which is based on
water level data from a 2D hydraulic model and also applies a regression approach (RH). The
hydraulic model has the clear advantage of being physically based and covering a wide range
of discharge. On the other hand, the alternative methods are simpler and faster in their
implementation, while still being able to account for typical hydromorphological features of
23
dynamic (restored or natural) river sections (e.g. nonlinear longitudinal water level
distributions, lateral water level gradients, disconnected river branches and hydraulic jumps).
We compared water level time series generated by both alternative methods with those
generated by the reference method at all of the fixpoints located in the 1.2 km long river reach
of our field site. For most cases, the accuracy of the water level predictions of the alternative
methods was comparable to the accuracy of the reference method itself. In addition, we found
evidence that the riverbed, and hence the water level distribution for a given discharge
condition, changed between the implementation of the reference and the alternative methods.
This change in riverbed morphology might have contributed to some of the larger deviations
among water level predictions.
The results of this study allow us to recommend both alternative methods for the river water
level assignment in future modeling studies of river-groundwater systems at scales in the
order of kilometers. The RM method is straightforward in its implementation, but is limited to
water level predictions within the range of measurements made at the fixpoints. If discharge
conditions beyond the measured range have to be simulated, we recommend the use of the IM
method instead.
Each of the presented methods has limitations in terms of accuracy in water level predictions.
Even though we consider each of the methods to be accurate, water level predictions will
differ for a specific discharge condition. The impact of the river water level uncertainty on
key predictions of groundwater models as exchange flux or groundwater residence time, both
in steady state and transient conditions, could be the basis for future research.
Acknowledgments
This study was accomplished within the National Research Program Sustainable Water
Management (NRP61) and funded by the Swiss National Science Foundation (SNF, Project
No. 406140-125856). Many thanks to Matthias Rudolf von Rohr, Lena Froyland, and Urs von
Gunten for their support. We would like to thank Nicola Pasquale (IfU, ETH Zurich) for
having provided the results of the hydraulic model. We thank John Molson and Ryan North
for many helpful discussions. The Agency for the Environment of the canton Thurgau
provided data, logistics and financial support. Additional support was provided by the
Competence Center Environment and Sustainability (CCES) of the ETH domain in the
framework of the RECORD project (Assessment and Modeling of Coupled Ecological and
Hydrological Dynamics in the Restored Corridor of a River (Restored Corridor Dynamics)).
We thank the three anonymous reviewers for their constructive comments.
24
2.6 Supporting information

This section contains a detailed description of the interpolation methods and their
implementation at the Niederneunforn field site.
2.6.1
Problem setup
At our field site, we considered the river as a two-dimensional domain and the two side
channels as a one-dimensional domain, based on their different widths. The river geometry
was represented by a set of six lines for the main channel plus one line for the disconnected
branch. Both side channels were represented by one line (Fig. 2.7). A more general and
schematic river system is illustrated in Fig. 2.8. We consider now that each of the lines is
mapped by a curvilinear system of longitudinal coordinates
an index
. The lines were numbered with
for the main river channel in our example. Similarly, sections
perpendicular to the main course of the river were considered and numbered with another
index
. The same numbering system was used for the river branch and the side
channels with the line indices
Fig. 2.7. Field site located in Niederneunforn at the Thur River with indicated lines, support
points, fixpoints and gauging stations. The colors of the support points indicate the
line/fixpoint, from which the water level was transferred. The general flow direction in the
river and the side channels is from right to left.
The intersections between the lines and the sections (Fig. 2.8) define the locations of support
points where water levels will be estimated using the interpolation methods described below.
Support points are identified in this coordinate system with a letter and two indices,
. The
first index always refers to the section number and the second index always refers to the line
number.
25
Fig. 2.8. Schematic representation of a curvilinear coordinate system with the lines
and the sections
. The dots at intersections between the lines and the
sections represent support points
and the circles represent gauging stations
or fixpoints
.
represents a gauging station located on section
measured continuously.
and line
where water levels are
represents a fixpoint on section and line with periodic manual

and
measurements. Their curvilinear coordinates are denoted

The measured water levels are denoted as
, respectively.
for those continuously recorded at
for those recorded periodically at fixed times
at
, and
. The longitudinal position of
the sections perpendicular to the main river axis are chosen such that each section contains at
least one fixpoint or gauging station.
The aim of the interpolation methods is to estimate the water levels at any support point
, where only limited or no water level data is available, by combining the measured
water levels at a gauging station
have the same meaning as
The estimation of
with those at a fixpoint
. The indices
but refer to a different location within the coordinate system.
is comprised of three steps.
1. Establish a mathematical relationship , such that

2. Use
to compute
for any time step :
3. Estimate the water levels

lines
and
and
, based on
.
.
for the support points located on section
and
Different options for these steps have been considered for the two new methods, which are
both based on measured data at gauging stations and fixpoints and are referred to as
alternative methods. In order to compare these alternative methods with a reference, we
developed a third method, which is based on a hydraulic model and is referred to as reference
method. These three methods are described in the following sections. All three methods have
been implemented in a MATLAB program. The final interpolation of the water levels from
26
the support points to the river boundary nodes of the numerical model is identical for all the
methods and is accomplished by a one-dimensional linear interpolation along the lines .
2.6.2
Method 1: Regression of measured data (RM)
This first method is based on using a polynomial regression technique to obtain the
mathematical relationship . For each fixpoint
, we identify the gauging station
2.9, Fig. 2.10) whose extracted water levels

best with the water levels
for the measurement times
(Fig.
correlate
Fig. 2.9. Illustration of the notations used for describing the RM (Regression of measured
data) method.
The polynomial regression is expressed as follows:
(2.1)
The gauging station
is denoted as the determining gauging station as its water level
uniquely determines the water level at
. The coefficients
square fitting to the measured data for the times
are obtained by standard least-
. To determine the maximum order
of the
polynomial Eq. (2.1), we conducted a systematic analysis. We increased the polynomial order
step by step and chose the one for which the following increase would have led to a RMSE
(root mean square error (Mayer and Butler, 1993)) reduction of less than 0.001 m. This
procedure led to a second-order (quadratic) and a first-order (linear) polynomial equation for
the fixpoints in the river and in the northern side channel, respectively (Fig. 2.11).
Applying
to any measured time series
allows estimating the time series
.
(2.2)
27
Fig. 2.10. Schematic representation of the regression approach. (a) Continuous water level
time series measured at the determining gauging station. (b) Measured water levels at a
fixpoint at the times , for which the water levels at the determining gauging station are
extracted in (a). (c) is established between
and
by polynomial regression. is used
to estimate the time series
in (b).
At section three and four, the water levels were measured at two fixpoints (
and
) on both shores to capture the lateral gradients. The lateral water level differences vary
depending on the river stage. These differences are in the order of 0.4 m for low water levels
and gradually decrease as the water level rises (Fig. 2.11). This suggests an asymmetric
riverbed morphology inducing lateral water level gradients for low-flow conditions. As the
water level rises, the influence of the riverbed morphology decreases until the lateral
differences are negligible.
The estimation of the water level
the water level
(or
at the support points of section (Fig. 2.9) based on
) is made in the simplest possible manner. In general, it is
assumed that there is no lateral gradient and therefore we have the following expression:
(2.3)
For the support points
and
at section three and four however, we assigned the water
levels of the northern fixpoints to the support points of the lines
28
and the water levels
of the southern fixpoints to the support points on the lines
, based on field
observations.
Fig. 2.11. Application of the regression approach to establish between the water levels at
the determining gauging station
and the water levels at the fixpoints
and
according to the RM (Regression of measured data) method. The RMSE of the polynomial
equations are indicated.
2.6.3
Method 2: Interpolation of measured data (IM)
The second method uses an interpolation function

fixpoint
that estimates the water level at a
, located between a determining gauging station
and a second gauging station
. The identification of the determining gauging station is not strictly defined; it can be
either the upstream or the downstream gauging station. We assume for the development of the
equations that the upstream gauging station is the determining gauging station and that
and
are located on the same line but on different sections ,
and
(Fig. 2.12).
Fig. 2.12. Illustration of the notations used for describing the IM (Interpolation of measured
data) method.
29
According to the IM method, the water levels
at a fixpoint between
generally be described by a linear interpolation term

trend
and
can
plus a deviation from this linear
.
(2.4)
with
(2.5)
According to the conceptual behavior model, on which the IM method relies,
maximum
for water levels
smaller than a specific water level
is at a
due to a strong
influence of the riverbed morphology (Fig. 2.13a).

(2.6)
As the water level rises, the influence of the riverbed morphology decreases and the
longitudinal water level distribution approaches a linear shape, with
2.13a) for water levels
higher than a threshold water level
being zero (Fig.

.
(2.7)
To account for the decrease of
between
and
is multiplied by an
inhibition term, which is assumed to decrease linearly from one to zero (Fig. 2.13b).
(2.8)
30
Fig. 2.13. Schematic illustration of the interpolation approach. (a) Two hypothetical
and
. (b)
longitudinal water level profiles for the situations
as a function of
.
To estimate
according to Eq. (2.4), the parameters
and
have to be
obtained at each fixpoint. To obtain these parameters, we calculate the deviation from the
for each measurement
linear trend
.
(2.9)
According to Eq. (2.9), plotting
against
between the water levels
(Fig. 2.13b), which was confirmed by the data at
and
should reveal a linear dependence
most of our fixpoints (Fig. 2.14). The parameters can then be estimated according to Fig.
2.13b. Typically, we use the maximum value of
of
for
. The threshold value
for
and the corresponding value
can be determined by using the value
is zero. In case the measured water level spectrum did not include high
for which
enough values of
for
going to zero, the linear trend can be extrapolated to the
intersection with the abscissa to estimate
To account for lateral water level gradients in a two-dimensional river system, the IM method
extends the conceptual behavior model, which is described above for the longitudinal
direction, to the lateral direction. Accordingly, the water level at a fixpoint
located on a
second line (Fig. 2.12) can be described by the water level at the fixpoint
plus a lateral
difference
.
(2.10)
31
is at a maximum
for water levels
smaller than a specific water level
, due to
the influence of an asymmetric riverbed.

(2.11)
For water levels
higher than a threshold water level
, the lateral gradients
disappear.
(2.12)
In between, a linear decrease of
is assumed.
(2.13)
The measured lateral difference

(2.14)
can be plotted against the water level
in order to estimate the required parameters by
Eq. (2.10). The same procedure can be applied as described above for estimating the
parameters required for Eq. (2.4) (Fig. 2.13b). The linear dependence of
on
as
expected by Eq. (2.14) was confirmed by our data (Fig. 2.14).

In our example, we determined the water levels at almost all fixpoints by using the
formulation for
according to Eq. (2.4). The determining and the second gauging
stations as well as the parameters describing
are listed in Table 2.2.
Eq. (2.4) assumes that both gauging stations are located on the same line as the fixpoint,
which is not the case for our river system. However, as we assume equal water levels across
each section (Eq. (2.3)) except at section three and four, this assumption is still valid. At
section three and four, we determined the water levels at the southern fixpoints
using the formulation for
describing
and
by
according to Eq. (2.10). In Table 2.2, the parameters
are written in italic. The estimation of the water level at the support points
is identical to the procedure described for the RM method in the previous section.
32
Fig. 2.14. Application of the interpolation approach to estimate the required parameters to
establish according to the IM (Interpolation of measured data) method. Dark blue:
Measured deviations from the linear trend for the fixpoint
. Light blue: Measured water
level differences between the fixpoints
and
. The lines indicate the linear trend as
expected by Eqs. (2.8) and (2.13).
Table 2.2. Determining gauging stations
of each fixpoint and gauging station along the
(river main channel) and
(northern side channel) for all methods. For the
lines
IM method, the second gauging station
and the resulting parameters of Eqs. (2.4) and
(2.10) are also given. The dashes indicate that the data of the gauging station are used directly.
The italic numbers refer to the description of lateral gradients according to Eq. (2.10). The
brackets for the fixpoints and gauging stations in the northern side channel for the reference
RH method (see next section) indicate the coupling to the RM method.
F18
G21
F36
F31
F46
F41
F51
F66
G76
F28
F38
G48
F58
F68
G78
RM
Gkld
G76
G76
G76
G76
G76
G76
G76
G48
G48
G48
G78
-
Gkld
G76
G76
G76
G76
G76
G76
G76
F18
F18
G48
G48
-
Gmn
G21
G21
G21
G21
G21
G21
G21
G48
G48
G78
G78
-
aij / aijL [m]

-0.20
0.47
-0.25
0.47
-0.32
0.44
0.28
0.46
0.48
-0.29
0.20
-
IM
ha / haL [m asl]
372.43
372.43
372.43
372.43
372.43
372.43
372.43
370.35
370.35
371.57
371.57
-
33
hthresh / hthreshL [m asl]

373.20
373.60
372.90
373.60
372.90
373.60
373.60
371.31
371.60
373.00
371.83
-
RH
Gkld
G76
G76
G76
G76
G76
G76
G76
G76
(G48)
(G48)
(-)
(G48)
(G78)
(-)
2.6.4
Method 3: Regression of hydraulic model data (RH)
To compare the two alternative methods with an independent reference, we developed a third
method that is based on data from an existing hydraulic model of the main river channel in the
section of interest. This reference method applies a polynomial regression technique to obtain
the mathematical relationship
, as the RM method does. In contrast to the RM method
however, the RH method allows the extraction of water level information directly at each
support point within the river domain
from the hydraulic model output, each
corresponding to one specific discharge condition

point
. We identify the determining support
that is located at the position of the determining gauging station
(Fig. 2.15),
whose water levels correlate best with the water levels within the river domain.
Fig. 2.15. Illustration of the notations used for describing the RH (Regression of hydraulic
model data) method.
The polynomial regression is performed by standard least-square fitting in order to obtain the
parameters
according to the following equation:

(2.15)
Applying the systematic analysis described in the section of the RM method, we determined a
maximal order
for the polynomial Eq. (2.15) (Fig. 2.16). The range of extracted water
levels from the hydraulic model (Fig. 2.16) is much bigger than the range of measured water
levels used for the RM method (Fig. 2.11). This difference explains the different polynomial
order for the RH and RM method as a second-order polynomial approximates a fourth-order
polynomial over a small range. The hydraulic model confirmed the existence of lateral water
level gradients and their decrease for increasing water levels (Fig. 2.16), as included in the
RM and IM method.
The application of
to a measured time series

34
allows estimating the time series
(2.16)
As the hydraulic model did not include the disconnected branch and the side channels, we
coupled the RH method to the RM method (Table 2.2).
Fig. 2.16. Application of the regression approach to establish between the water levels at
the determining support point
and the water levels
and
according to the RH
method (Regression of hydraulic model data). The RMSE of the polynomial equations are
indicated.
35
Chapter 3
Assessing the effect of different river water level interpolation
schemes on modeled groundwater residence times
Submitted to Journal of Hydrology
Diem, S., Renard, P., Schirmer, M., submitted to Journal of Hydrology. Assessing the
effect of different river water level interpolation schemes on modeled groundwater
residence times.
Abstract
Modeling river-groundwater interactions requires knowledge of the rivers spatiotemporal
water level distribution. A quantitative understanding of these interactions is of particular
importance in the context of river restoration. The dynamic nature of riverbed morphology in
restored river reaches might result in complex river water level distributions, including
disconnected river branches, nonlinear longitudinal water level profiles and morphologically
induced lateral water level gradients. Recently, two new methods were proposed to accurately
and efficiently capture 2D water level distributions of dynamic rivers. In this study, we
assessed the predictive capability of these methods with respect to simulated groundwater
residence times. Both methods were used to generate surface water level distributions of a
1.2 km long partly restored river reach of the Thur River in northeastern Switzerland. We then
assigned these water level distributions as boundary conditions to a 3D steady-state
groundwater flow and transport model. When applying either of the new methods, the
calibration-constrained groundwater flow field accurately predicted the spatial distribution of
groundwater residence times; deviations were within a range of 30% when compared to
residence times obtained using a reference method. We further tested the sensitivity of the
simulated groundwater residence times to a simplified river water level distribution. The
negligence of lateral river water level gradients of 20-30 cm on a length of 200 m caused
errors of 40-80% in the calibration-constrained groundwater residence time distribution
compared to results that included lateral water level gradients. The additional assumption of a
linear water level distribution in longitudinal river direction led to deviations from the
complete river water level distribution of up to 50 cm, which caused wide-spread errors in
37
simulated groundwater residence times of 200-500%. For an accurate simulation of

groundwater residence times, it is therefore imperative that the longitudinal water level
distribution is correctly captured and described. Based on the confirmed predictive capability
of the new methods to estimate 2D river water level distributions, we can recommend their
application to future studies that model dynamic river-groundwater systems.
3.1 Introduction
Groundwater flow and transport modeling is a valuable and frequently applied tool to gain a
process-based understanding of surface water-groundwater systems, providing quantitative
information on flow paths, mixing ratios and residence times (Wondzell et al., 2009). It is
well known from synthetic modeling studies that riverbed morphology affects the river water
level distribution, which in turn drives the exchange with groundwater (Woessner, 2000;
Cardenas et al., 2004; Cardenas, 2009). Therefore, an important prerequisite for the setup of a
groundwater flow and transport model of a real surface water-groundwater system is an
accurate description of the water level distribution along the surface water boundary
conditions.
A quantitative assessment of groundwater flow paths and residence times is of particular
interest at riverbank filtration systems (Tufenkji et al., 2002). Groundwater residence time is
an important parameter in determining the effectiveness of the natural attenuation processes
that occur during riverbank filtration (Eckert and Irmscher, 2006). River restoration measures,
such as riverbed enlargements, potentially lead to reduced groundwater residence times, and
hence, to an increased risk of drinking water contamination (Hoehn and Scholtis, 2011) that
competes with the original purpose of river restoration (Brunke and Gonser, 1997; Woolsey et
al., 2007). Groundwater flow and transport modeling could help to mitigate this conflict of
interest, by providing a quantitative assessment of the groundwater flow paths and residence
times (Hoehn and Meylan, 2009).
Restored river systems may have complex water level distributions characterized by nonlinear
longitudinal water level distributions, morphologically induced lateral water level gradients,
disconnected river branches and hydraulic jumps. Such water level distributions need to be
characterized by their full spatial (i.e. two horizontal dimensions) and temporal extent (i.e. for
any discharge condition) and ideally are extracted from hydraulic models (Doppler et al.,
2007; Derx et al., 2010; Engeler et al., 2011). However, the setup of a hydraulic model is time
consuming and requires a considerable amount of data input. In Chapter 2, we proposed two
new alternative interpolation methods to estimate time-varying one- and two-dimensional (1D,
38
2D) surface water level distributions of dynamic rivers based directly on measured water level
data. Even though transient water level predictions with the alternative methods are
considered to be accurate when compared to those of a third reference method, water level
predictions will differ for a specific discharge condition (Chapter 2).
In this study, we assess the predictive capability of the new alternative methods with respect
to the simulated groundwater residence time and the effect of reducing the considered level of
detail in the surface water level distribution. Thereto, steady-state surface water level
distributions at a partly restored riverbank filtration system are generated with both alternative
methods and the reference method, as well as with two simplified methods. The resulting
water level distributions are then assigned to a 3D groundwater flow and transport model.
After calibration against groundwater heads, each of the model scenarios is used to predict the
spatial groundwater residence time distribution within the modeling domain.
3.2 Interpolation methods

The interpolation methods used in this study are based on those established in Chapter 2. A
brief description of the methods is provided in this section, but for a more detailed description
the reader is referred to Chapter 2. The alternative methods and the reference method are
referred to as complete interpolation methods, as they cover the full level of detail including
lateral water level gradients and nonlinear longitudinal water level distributions.
3.2.1
Complete interpolation methods
Both new alternative interpolation methods are based on the concept of combining continuous
water level records (
(
) from water level gauges ( ) with periodic water level measurements
) at fixpoints ( ) in between. By combining this data, the water level distribution between
the water level gauges is obtained at a higher resolution. Fixpoints are defined as reference
points in the river whose altitude is known. The first alternative RM method (Regression of
measured data) applies a polynomial regression technique to predict water levels at fixpoints
from any measured water level at a specific water level gauge, while the second alternative
IM method (Interpolation of measured data) uses a nonlinear interpolation approach
between two water level gauges.
Depending on the lateral extent, the river might be considered as a 1D or a 2D domain. In the
latter case, the river is discretized by multiple lines parallel to the main flow direction of the
river and several sections of support points ( ) perpendicular to the flow direction (Fig. 3.1).
Sections of support points are defined at locations where a water level gauge or a fixpoint
39
exists. One fixpoint per section is enough unless lateral water level gradients are observed, in
which case a fixpoint should be defined on both shorelines.
The water levels at the support points (
) are estimated from the water levels at the fixpoint
in the simplest possible manner. If no lateral water level gradient exists, the water level of the
fixpoint is assigned to all support points on the same section. If a second fixpoint was defined
to capture lateral gradients, assigning water levels to the support points should be based on
field observations. The final interpolation of water levels from the support points to the river
boundary nodes of the numerical model is identical for all the interpolation methods and is
performed by a linear interpolation along the set of lines.
The third complete RH method (Regression of hydraulic model data) applies a polynomial
regression technique, similar to the RM method, but is based on water levels extracted from a
hydraulic model at each support point directly. The RH method is therefore considered as
reference method among the complete interpolation methods.
3.2.2
Simplified interpolation methods
In addition to the predictive comparison of the complete interpolation methods described

above, we assessed the error in residence time prediction that evolves when the water level
distribution of the river is simplified. Thereto, we applied two progressively simplified
methods, both based on the complete IM method. The first simplified method ignores lateral
water level gradients and is denoted as Interpolation of measured data without lateral
gradients (IM_wo_lat). The second simplification additionally assumes a linear interpolation
between the river water level gauges and is called Interpolation of measured data assuming a
linear interpolation (IM_lin).
Fig. 3.1. Schematic illustration of a river system with multiple lines and sections of support
points ( , filled black circles). The open black circles indicate the water level gauges ( ) and
fixpoints ( ).
40
3.3 Method implementation

This section provides a description of the Niederneunforn field site (Section 3.3.1) and a
review of the implementation of the interpolation methods at this field site (Section 3.3.2).
Section 3.3.3 presents the generated surface water level distributions, which we assigned to
the groundwater flow and transport model (see Section 3.4) to simulate the spatial
groundwater residence time distribution.
3.3.1
Field site
The Niederneunforn field site (Fig. 3.2) is located at the Thur River in NE-Switzerland,
approximately 12 km upstream of the confluence with the Rhine River. The Thur River is a
peri-alpine river draining a catchment area of 1730 km2. It is the largest river in Switzerland
without a retention basin and therefore has a very dynamic discharge regime. Discharges
range from 3 to 1100 m3/s, with an average discharge of 47 m3/s.
Fig. 3.2. Niederneunforn field site at the Thur River in NE-Switzerland. Fixpoints and water
level gauges in the river and the side channels are shown as open black circles. Based on their
position, the set of lines and sections of support points (filled circles) were defined for the
implementation of the interpolation methods. The colors of the lines will be used in Fig. 3.4
again. The colors of the support points in the river indicate the shore line or the fixpoint/water
level gauge on that shore line from which the water levels were transferred. The white
polygon represents the modeling domain. The general flow direction of the river and the side
channels is from right to left.
The field site was instrumented with more than 80 piezometers (2) during the
interdisciplinary
RECORD
project
(Restored
corridor
dynamics,
http://www.cces.ethz.ch/projects/nature/Record; Schirmer (2013); Schneider et al. (2011)) in

the context of restoration measures that were realized in 2002. The river restoration was
constrained to the northwestern part of the river reach (Fig. 3.2). Restoration measures were
forbidden in the northeastern part in order to protect the water quality of the nearby pumping
station, which supplies the community of Nieder- and Oberneunforn with drinking water. This
vertical well produces a total of 36 m3, split into two daily periods of 1 h and 2 h. At the
41
southern bank of the Thur River, the bank stabilization was maintained to protect the 4 mhigh dam, which prevents flooding of nearby farms and agricultural land.
Based on 57 drilling profiles, the gravel-and-sand aquifer has a thickness of 5.31.2 m at the
field site. Hydraulic conductivities were estimated to range from 410-3 to 410-2 m/s by slug
tests, a pumping test and a salt tracer test (Diem et al., 2010; Doetsch et al., 2012). The
aquifer is underlain by a lacustrine clay layer, which forms the lower hydraulic boundary. On
top of the aquifer is a 0.5-3 m thick layer of silty sand from the alluvial fines that can be
regarded as the semi-confining unit. The aquifer varies both spatially and temporally between
confined and unconfined. Cross-borehole georadar travel-time tomography revealed an
average porosity of 203% (Schneider et al., 2011).
At the field site, the width of the Thur River varies between 50 and 100 m (Fig. 3.2). After the
completion of the restoration measures, a large gravel bar has evolved at the downstream end
of the river reach. At the same time, a partly disconnected branch of the river developed,
which is only flooded at high river stages (>200 m3/s) and is otherwise fed by groundwater.
During low-flow conditions, the river water level profile in the longitudinal direction is
nonlinear. In the upstream 400 m, the gradient is 0.5 and in the downstream 800 m, it is 2.
In the central part of the river reach, lateral water level gradients occur during low-flow
conditions. These lateral surface water level differences are caused by the asymmetrical
riverbed morphology and can reach up to 0.4 m. Two side channels (north and south) flow
parallel to the river with widths ranging between 4 and 8 m. Two beaver dams are located in
the northern side channel. The upstream dam has a more pronounced effect on water levels,
resulting in changes of up to 0.5 m.
3.3.2
Data collection
Two water level gauges were installed in the main channel of the river, two in each side
channel, and one in the river branch (Fig. 3.2). Several fixpoints were added between the
water level gauges to increase the spatial resolution of the water level distribution. In the
northern side channel, the fixpoints were placed upstream and downstream of both beaver
dams to capture the hydraulic jumps. Fixpoints in the river were placed close to piezometer
transects, either on the northern or on the southern shore. In the central portion of the river
reach, where lateral water level gradients were observed, a fixpoint was added on either shore.
The southern side channel is very straight, has a uniform width, and does not have any
obstacles. Therefore, a linear water level distribution was assumed between the water level
gauges, which was confirmed by one set of measurements along the channel.
42
Water levels were measured periodically at the fixpoints between February and May 2011
covering a discharge range of 10 to 100 m3/s. The sensors of the water level gauges (DL/N 70,
STS AG, Switzerland) have been continuously measuring pressure, temperature and electrical
conductivity (EC) at 15-min intervals since April 2010 (error of single measurement: 0.1%
for pressure, 0.25% for temperature and 2% for EC, according to the manufacturers
manual). For model calibration, the same type of sensor was placed in each of the observation
wells shown in Fig. 3.2. The raw data of the water level gauges and the observation wells
were processed to correct for the barometric air pressure, and to transform the pressure data to
absolute water levels (m asl).
The system of lines and support points was defined based on the location of water level
gauges and fixpoints. Each point can be identified by a uniquely defined indexing system. The
first index
(
refers to the section number and the second index
to the line number
). The river was considered to be a 2D domain, described by a set of six lines for
the main channel, and one additional line for the disconnected branch. Sections of support
points were defined wherever a water level gauge or a fixpoint was located. The colors of the
support points in Fig. 3.2 indicate the shore line or the fixpoint/water level gauge on that
shore line, from which the water levels were transferred. A zero lateral gradient was assumed
everywhere except for the river sections
, where the generally lower water levels of
the southern fixpoints were assigned to the support points on lines
and the generally
higher water levels of the northern fixpoints to the support points on lines
(Fig. 3.2).
Because the width of the northern and southern side channel is much smaller relative to the
river width, they were considered as a 1D domain and described by a single line.
For the implementation of the RH method, an existing 2D hydraulic model of the Thur River
was used, which was developed based on the bathymetry measured in September 2009 and
covered a discharge range of 10-650 m3/s (Schppi et al., 2010; Pasquale et al., 2011). The
hydraulic model did not include the side channels and the disconnected river branch.
Therefore, the RH method was coupled to the RM method to cover the full surface water level
distribution at the Niederneunforn field site.
Fig. 3.3a shows a 3-month water level time series at the support point
, determined using
the reference RH method (black line). The time series of measured groundwater head at an
observation well located 100 m from
(gray line, Fig. 3.3a) illustrates the quasi-
instantaneous reaction of the groundwater heads to changes in the river water level, with a
propagation speed of about 0.2 m/s or 19600 m/d. Minima in EC, which corresponded with
43
peak flows in the river, were caused by the diluting effect of rain events (Fig. 3.3b). The
characteristic EC signal was identified in all observation wells close to the river, both on its
northern and its southern side, which indicates losing conditions. The EC signal in the river
was transported into groundwater and was used as a natural tracer. By analyzing the EC time
series with nonparametric deconvolution (Cirpka et al., 2007; Vogt et al., 2010a), we obtained
estimates of local residence time distributions, characterized by a mean and a standard
deviation (Supporting information; Section 3.7: Fig. 3.9, Fig. 3.10, Table 3.2).
Fig. 3.3. (a) Water level time series (May to August 2010) generated by the RH method at
in the Thur River (black line) and measured groundwater head time series at
support point
a nearby observation well (gray line). The black vertical line indicates the point in time (May
26, 2010, 17:00) for which the surface water level distribution at the field site was generated
using each of the five methods (Fig. 3.4). (b) Time series of electrical conductivity (EC)
measurements in the Thur River at the same observation well used in (a).
3.3.3
Generated water level distributions
The surface water level distributions used for the steady-state model simulations were
generated with all interpolation methods for the conditions on May 26, 2010 (vertical line in
Fig. 3.3a), which was at the end of a relatively short period of low flow (23 m3/s). As
groundwater heads are highly correlated with the river water levels (Fig. 3.3a), a steady-state
assumption is reasonable.
Fig. 3.4a shows the spatial water level distributions generated by the three complete methods.
For better clarity, we only plotted the results from one line in the main river channel (
dark blue), which illustrates the nonlinear longitudinal water level distribution. The southern
side channel (
, black line) had a much lower water level than the river and therefore
complied with its purpose of draining groundwater. Water levels in the northern side channel
44
, green line) were considerably higher because the northern channel flows back to the
river at the western end of the field site (river section
). The northern side channel could
only drain groundwater in one ~400 m segment, where water levels were below river water
levels. This segment was located downstream of the 50 cm water level drop caused by the
eastern beaver dam (
). The disconnected branch of the river (
, red line) showed
slightly lower water levels than the main river channel. The disconnected branch and the
exfiltrating segment of the northern side channel seemed to be responsible for the river
infiltration that occurred to the northern side of the river.
Even though the alternative methods are considered to be accurate in their water level
predictions (Chapter 2), single realizations of the spatial water level distribution showed
deviations from the reference RH method of mostly 10 to 20 cm (Fig. 3.4a). However, errors
of more than 30 cm occurred at the river section
. As this section was located on a bend
in the river, the section did not cross the river perpendicularly. The RH method accounted for
the water level gradient across this section by assigning a water level to each of the support
points individually, based on the hydraulic model. In contrast, the alternative methods
assumed a constant water level across section
available from one fixpoint (
, as water level information was only
).
Other deviations in the output from the complete methods were caused by a different data
basis (hydraulic model vs. measured data) and/or the different structure of the interpolation
methods. The coupling of the RH method to the RM method for the northern and the southern
side channel led to identical water level distributions for lines
. The IM method
showed an identical water level distribution for the southern side channel as well, while
deviations from the RM/RH method in the northern side channel reached a maximum of
10 cm at
In Fig. 3.4b, water level distributions determined from the complete IM method and from the
two simplified versions are depicted. The main river channel is now represented by the
southern shoreline (
across sections
lines
, light blue). The IM method considered lateral water level gradients

where water levels on the southern shoreline
(representative for
, see Fig. 3.2) were 20-30 cm lower than on the northern shoreline
(representative for lines
, see Fig. 3.2). The first simplified IM method (IM_wo_lat)
ignored these lateral water level gradients and the water levels of the northern fixpoints were
assigned to all support points on sections
southern lines
. As a consequence, water levels on the
increased by 20 and 30 cm, respectively (Fig. 3.4b, Fig. 3.2). The

45
second simplification of the IM method (IM_lin) additionally assumed a linear interpolation

between the two water level gauges, which are located at river sections
. This
assumption caused deviations in water levels of 30-50 cm at the support points in between
(
).
Fig. 3.4. (a) Spatial water level distribution along the lines
, generated with the three
complete interpolation methods. The figure is orientated for a viewer looking towards NNE,
with the flow from right to left. (b) Spatial water level distribution along the lines
,
generated with the IM method and its two simplified versions. The water level distributions
were generated for the conditions on May 26, 2010, 17:00 (vertical line in Fig. 3.3a, discharge
23 m3/s). The colors of the lines correspond to those of Fig. 3.2 and textures of the lines
correspond to the different interpolation methods (see legend).
46
3.4 Groundwater flow and transport model

3.4.1
Numerical model setup
We set up a 3D finite-element groundwater flow and transport model using FEFLOW

(version 6.0, DHI-WASY GmbH). The modeling domain is shown in Fig. 3.2. The northern
boundary is defined by the northern end of the aquifer and the southern boundary by the
southern side channel. The vertical model extent was restricted to the gravel-and-sand aquifer,
whose top and bottom elevations were determined from 26 drilling profiles, from which the
entire model domain was interpolated using a kriging technique. The horizontal discretization
length of the triangular elements varied between 1 and 5 m around observation wells and
along boundaries, including the river and the side channels. In the remaining model domain,
the maximal horizontal length of the elements was 10 m. In the vertical direction, the aquifer
was subdivided into five layers. The top four layers had a thickness of 1 m and the bottom
layer had a variable thickness.
The definition of the boundary conditions is depicted in Fig. 3.5. We applied an influx
boundary condition on the eastern and northern borders (q1 = 0.18 m/d, q2 = 0.0043 m/d), and
an outflux boundary condition on the western border (q3 = -0.57 m/d, q4 = -2.3 m/d). These
2nd Type boundary conditions were applied to all layers. We determined the absolute
groundwater flux from measured hydraulic gradients and estimated hydraulic conductivities.
Recharge was neglected, as no rainfall occurred for ~10 d before the simulation time (Fig.
3.3a). At the location of the pumping well we assigned an average extraction rate of 36 m3/d
(0.4 L/s).
Within the modeling domain, the southern side channel is exfiltrating along its entire length
due to water levels well below those in the river (Fig. 3.4a). The channel bed sediments are
gravelly and tracer tests revealed a good connection to groundwater. We therefore chose a 1D
fixed-head boundary condition (1st Type) for the top layer along the southern side channel.
A colmation layer of unknown thickness was identified in the bed of the Thur River (Hoehn
and Meylan, 2009; Schneider et al., 2011). The bed of the northern side channel was covered
by a thick (0.5-1 m) silt and clay colmation layer, except in the middle exfiltrating segment
located downstream of the eastern beaver dam, where the bed sediments were gravelly. To
account for the effect of colmation, we assigned a Cauchy boundary condition (3rd Type) to
the river and the northern side channel. In FEFLOW, the colmation layer is characterized by a
transfer rate
corresponds to the hydraulic conductivity of the colmation layer

47
and
to its thickness. The river was described by a 2D Cauchy boundary condition on the
top layer, which we subdivided into two zones of transfer rates. Zone L1 (Fig. 3.5) covered
the restored part of the river, including the disconnected branch. The remaining channelized
part was covered by zone L2. We split the 1D Cauchy boundary condition along the northern
side channel into three zones of transfer rates (L3-L5) to separate the middle exfiltrating
segment (L4) (with its gravelly bed sediment) from the upstream and downstream heavily
clogged segments.
FEFLOW describes each transfer rate zone by a transfer rate for infiltration (Lin) and
exfiltration (Lout). Our system of Cauchy boundary conditions (L1-L5) was therefore
characterized by 10 parameters (Lin1-Lin5, Lout1-Lout5). Lout is typically larger than Lin, as the
exfiltrating clean groundwater flushes the pore space. This effect probably explains the
gravelly bed sediments in the southern side channel as well as in the exfiltrating segment of
the northern side channel (L4). On the other hand, suspended particles in infiltrating surface
water tend to clog the pore space, as was the case in the main river channel (L1, L2) and in
the upstream and downstream part of the northern side channel (L3, L5).
Fig. 3.5. Spatial definition of the boundary conditions (BCs) and the hydraulic conductivity
zonation within the modeling domain. Groundwater head observations were available at all 19
observation wells, while experimentally determined residence times were restricted to 9
observation wells (black filling). The rectangle at the bottom of the figure illustrates the true
ratio between the vertical and the longitudinal extent of the model.
48
3.4.2
Calibration procedure
To initially obtain a realistic spatial groundwater residence time distribution when using the
reference RH water level distribution (Fig. 3.4a), we jointly estimated the transfer rates and
the hydraulic conductivity distribution by fitting both, measured groundwater heads and
experimentally determined groundwater residence times from nonparametric deconvolution of
EC time series (Section 3.3.2, Fig. 3.5). A list of these residence times together with their
standard deviations is presented in the Supporting information (Table 3.2, Fig. 3.9). To reduce
the number of 10 adjustable transfer rates, pilot model runs were performed using estimated
parameter values from field observations. These model runs revealed that the infiltrating main
river channel (Lin1, Lin2), the exfiltrating disconnected river branch (Lout1) and the exfiltrating
segment of the northern side channel (Lout4) were responsible for most of the groundwater
flux at the Cauchy boundary conditions. Lout1 was additionally tied to Lin1 with a factor of 10
to eliminate parameter correlation. For each manual adjustment of the hydraulic conductivity
distribution, the remaining three adjustable transfer rates were estimated using PEST (Doherty,
2005) by fitting measured groundwater heads (i.e. minimizing the sum of squared errors
between simulated and measured heads). This procedure was iterated until the simulated
groundwater residence times (see Section 3.4.3) were within 1 standard deviation of the
experimentally determined residence times. The resulting hydraulic conductivity distribution
comprised four different zones (Fig. 3.5) and their hydraulic conductivities were within a
range of 410-3 - 610-2 m/s (Table 3.1), which corresponds well to the measured values
(Section 3.3.1).
Based on this initial parameterization for the reference RH model scenario (Table 3.1), the
water level distributions generated with both alternative and both simplified methods (Fig.
3.4a, b) were assigned to the river and the side channel boundary conditions of the model. For
each of the four model scenarios, the three adjustable transfer rates were calibrated against
measured groundwater heads using PEST to ensure that the basis of the groundwater
residence time simulation was a calibration-constrained groundwater flow field. The
remaining transfer rates and the hydraulic conductivity distribution were kept constant. Fig.
3.11 (Supporting information) visually summarizes the initial and the subsequent calibration
procedure by a flow chart.
49
Table 3.1. Hydraulic conductivities and transfer rates of the corresponding parameter zones
(Fig. 3.5) resulting from the initial model calibration using the RH water level distribution.
The transfer rates in bold were estimated using PEST. The remaining transfer rates were kept
at their initial values, which in turn were estimated from field observations. Lout1 (in brackets)
was tied to Lin1 with a factor of 10. The three adjustable transfer rates were estimated for the
remaining four model scenarios as well (Fig. 3.6), while the hydraulic conductivity
distribution was kept constant.
3.4.3
K1 [m/s]
610-2
K2 [m/s]
210-2
K3 [m/s]
110-2
K4 [m/s]
410-3
Lin1 [1/d]
Lin2 [1/d]
Lin3 [1/d]
Lin4 [1/d]
Lin5 [1/d]
11.9
2.1
0.5
1.0
0.5
Lout1 [1/d]
Lout2 [1/d]
Lout3 [1/d]
Lout4 [1/d]
Lout5 [1/d]
(119)
70.0
5.0
24.7
5.0
Simulation of groundwater age
Based on the calibration-constrained groundwater flow field, we simulated the spatial

distribution of groundwater residence time, hereafter referred to as groundwater age. We
performed a steady-state transport simulation of a tracer with a zero-order source term.
According to Goode (1996), the groundwater age (i.e. the time since entering the model
domain) is obtained by a steady-state transport simulation of a tracer with an appropriate
definition of the boundary conditions and a zero-order source term equal to the porosity. A
fixed age (concentration) of zero was defined at inflowing boundaries and a natural 2nd Type
boundary condition at outflowing boundaries. The latter is described by
, i.e.
the age ( ) in normal direction ( ) to the boundary does not change. The zero-order source
term was set to 0.2 mg L-1 d-1 for each element according to the mean porosity of 0.2 (Section
3.3.1).
We set the longitudinal dispersivity to 10 m for our model on the scale of 1000 m, according
to Gelhar et al. (1992). We assigned a value of 1 m to both the horizontal and vertical
transverse dispersivity, as differentiating the two was not possible in FEFLOW. As a result,
the vertical transverse dispersivity was obviously too high, which caused an excessive vertical
dispersion and hence, small vertical age differences. However, as the vertical extent of the
modeling domain was very small compared to its horizontal extent (Fig. 3.5), the groundwater
age was mainly controlled by horizontal transport.
50
To visualize and compare the spatial age distributions from the different model scenarios, we
first calculated vertically averaged age distributions, weighted by the element thickness.
Additionally, we produced spatial distributions of relative age differences to highlight spatial
differences among the age distributions. The relative age difference
age
and a reference mean age
between a mean
was calculated at each node position as follows:

(3.1)
3.5 Results and discussion

3.5.1
Calibration results
The estimated adjustable transfer rates for all the model scenarios are plotted in Fig. 3.6.
Model scenarios that used the new alternative methods (RM, IM) had small changes in the
estimated transfer rates compared to results of the reference RH model scenario (10-40%).
Neglecting lateral water level gradients in the first simplified IM method (IM_wo_lat) led to
adjustments of the transfer rates by a factor of 2-3 compared to the complete IM method. The
largest changes, at 1-2 orders of magnitude, were made during model calibration of the linear
river water level distribution scenario for the second simplified method (IM_lin).
The post-calibration root mean square error (RMSE) between measured and simulated
groundwater heads was 7.7 cm for the reference RH method and varied between 7.2-7.3 cm
for the RM, the IM as well as for the IM_wo_lat model scenario. The model efficiencies
(Mayer and Butler, 1993) were 0.98 for all of the complete methods and the IM_wo_lat
method. The IM_lin model scenario resulted in a RMSE of 11 cm and a model efficiency of
0.95. The improvement of the pre-calibration RMSE (using the estimated transfer rates of the
RH model scenario; Table 3.1) over the post-calibration RMSE was small (0.1-0.3 cm) for the
alternative methods (RM, IM). For the simplified IM_wo_lat and IM_lin methods, the RMSE
was reduced by 2 and 5 cm, respectively.
The inspection of the inversion statistics revealed that the inversion problem was well posed
and not infected by a high degree of non-uniqueness. First, the three eigenvalues of the
parameter covariance matrix had a maximum to minimum ratio of 102, well below the
maximum acceptable value of 107 (Doherty, 2005). Second, the ratio of the maximum and the
minimum composite sensitivity was 9, while the acceptable maximum is 100. The highest
correlation of 0.8 was found between Lin2 and Lout4. The remaining two correlation
coefficients were less than 0.5.
51
Fig. 3.6. Post-calibration transfer rates of the reference (RH), the alternative (RM, IM) and the
simplified (IM_wo_lat, IM_lin) model scenarios.
3.5.2
Groundwater flow field and absolute age distribution
Before being able to compare the age distributions of different model scenarios, we have to
become familiar with the general characteristics of the groundwater flow field at our field site.
Fig. 3.7 presents the calibration-constrained groundwater flow field and the corresponding
groundwater age distribution from the reference RH model scenario (Fig. 3.4a). For
groundwater recharged by the river or the northern side channel, which was mostly the case
within the modeling domain, the groundwater age reflects the real age since infiltration.
Along the northern and northeastern inflow boundaries (Fig. 3.5) however, where a fixed age
of zero was assigned as well, the groundwater age rather refers to the time since entering the
modeling domain.
Infiltration at the main river channel occurred to both the northern and the southern sides,
with an overall infiltration rate of 300 L/s. Most of the infiltrated water (about 80%) flowed
towards the southern side, due to the high water level gradient between the river and the
southern side channel. The corresponding exfiltration rate along the southern side channel
(240 L/s) was validated by the measured discharge difference between the water level gauges
and
(Fig. 3.2). Only about 20% of the river infiltration occurred to the northern side,
induced by the exfiltrating segment of the northern side channel (L4) and the disconnected
branch of the river (Fig. 3.4a). The model estimated an exfiltration rate of 25 L/s at the
exfiltrating segment of the northern side channel, which was consistent with the measured
difference in discharge downstream and upstream of the beaver dam. Exfiltration at the
disconnected branch occurred at a rate of about 30 L/s.
52
The groundwater flow field on the southern side of the river was relatively uniform.
Infiltration occurred at a steep angle to the river and the groundwater age reached a maximum
of 4-8 d. The high water level difference between the river and the southern side channel
caused an asymmetric groundwater flow field underneath the river, which significantly
contributed to the much higher infiltration towards the southern side.
On the northern side of the river, the flow field was more complex. In the upstream part at
river sections
, infiltration occurred at a relatively steep angle and groundwater flow
was directed towards the exfiltrating segment (L4) of the northern side channel. To the north
of L4, the low groundwater head gradients and the long flow paths led to groundwater ages of
up to 90 d. On the southern side of L4, the groundwater age reached only 10-20 d due to the
presence of direct pathways linking the river with the side channel. The pumping well had no
significant impact on the groundwater flow field due to its low average pumping rate (Section
3.4.1).
Further downstream, at river sections
, the groundwater flow field on the northern
side of the river became parallel to the river as the water levels in the northern side channel
became higher than those in the main river channel (Fig. 3.4a). Therefore, groundwater that
was not drained by the exfiltrating segment of the northern side channel (L4) was deflected
towards and drained by the disconnected branch of the river. The further upstream infiltration
at river sections
occurred, the longer and the more arc-shaped the flow paths became.
Accordingly, the groundwater age increased from the river towards the northern side channel,
reaching a maximum of 50 d.
Fig. 3.7. Calibration-constrained groundwater flow field (shown as groundwater isopotentials

with an equidistance of 15 cm) and vertically averaged groundwater age from the reference
model scenario using the RH surface water level distribution (Fig. 3.4a). The three adjustable
transfer rate zones are indicated.
53
3.5.3
Predictive comparison of the complete interpolation methods
To assess the predictive capability of both new alternative methods (RM, IM), we calculated
the spatial distribution of the relative age difference
according to Eq. (3.1), using the
age distribution of the RH method (Fig. 3.7) as a reference (
). Both the RM and the IM
model scenario showed similar results (Fig. 3.8a, b). To the north of the river, two zones with
lower ages (10-20%) relative to the RH model scenario were identified between the river
sections
and
. These small deviations were caused by a combination of minor
water level differences with respect to the RH method at river sections
(Fig. 3.4a)
and the related changes in transfer rates that occurred during calibration, mostly within 1040%. These changes actually provided a slightly better fit to the measured groundwater heads
(Section 3.5.1). As an example, the higher water levels of the IM method along river sections
were compensated by a 40% reduction in the Lin1 transfer rate relative to the RH
model scenario (Fig. 3.6).
The highest adjustment during calibration of the IM model scenario was made for Lout4,
which was doubled to balance the higher water levels in the exfiltrating segment of the
northern side channel (Fig. 3.4a, Fig. 3.6). A zone of higher groundwater age remained
between the river and the northern side channel (Fig. 3.8b), but deviations were restricted to
an upper bound of 30%.
An additional common element of both alternative model scenarios was the zone between the
main river channel and the disconnected branch, where groundwater age differed from the
reference model scenario by 40-100%. These deviations can be blamed on the failure of the
zero lateral gradient assumption at river section
for both the RM and the IM method
(Section 3.3.3). The too low water levels compared to the RH method could not be
compensated during calibration of the transfer rates as no observations were available in this
region, and therefore led to a pronounced increase in groundwater age. The failure of the zero
lateral gradient assumption was also responsible for the positive deviations in groundwater
age at the western part of the southern side of the river. Apart from that, errors in groundwater
age on the southern side of the river were generally small (<10%) because river water level
deviations were small compared to the difference in water level between the river and the
southern side channel.
Besides the large differences between the main river channel and the disconnected branch,
both alternative methods (RM, IM) were able to predict the groundwater age within 30% of
the reference RH method. The error of 30% is small compared to the uncertainty (standard
54
deviation) of the experimentally determined residence times of 60-80% (Supporting

information: Table 3.2). The calibration of the transfer rates compensated for minor
differences among the surface water level distributions and the resulting flow fields provided
an accurate prediction of groundwater age. These results confirm the capability of the new
alternative methods to efficiently capture the relevant characteristics of the surface water level
distribution allowing for a reliable simulation of the groundwater age distribution.
Fig. 3.8. Spatial distribution of the relative difference in mean groundwater age (Eq. (3.1)),
for the two alternative methods (RM, IM) with respect to the RH method (a, b), and for the
two simplified versions of the IM method with respect to the complete IM method (c, d). (c)
and (d) contain an additional isoline (black dashed line) of relative age difference of 60% and
200%, respectively. The three adjustable transfer rate zones are indicated.
55
3.5.4
Predictive comparison of the simplified interpolation methods
We used the groundwater age distribution from the IM model scenario as a reference (
) to
compare the groundwater age distributions of the two simplified model scenarios. Fig. 3.8c
shows the results for the model scenario that applied the first simplified river water level
distribution, which ignored lateral water level gradients (IM_wo_lat). The only difference in
the water level distribution compared to the full IM implementation was water levels 2030 cm higher at the support points on river sections
on the southern lines
(Fig. 3.4b, Fig. 3.2). Even though these changes in the water level distribution seem to be
small, the effect on the calibration-constrained groundwater age distribution was considerable.
To the north of the river, a band of higher age was identified that extended from the river at
sections
, along the northern side channel, to the disconnected river branch, with
deviations of up to 40-80% compared to the IM model scenario.

As described in Section 3.5.2, the groundwater flow field underneath the river was generally
asymmetric because groundwater was largely withdrawn by the strong gradient between the
river and the southern side channel. This effect was intensified by the lateral gradient at
sections
, which was also directed towards the south. Therefore, the negligence of the
lateral gradient reduced the asymmetric characteristics and more water infiltrated to the
northern side of the river. To counteract this additional groundwater flux and the higher
groundwater heads, the Lin1 transfer rate was lowered by 50% during the calibration
procedure (Fig. 3.6). However, groundwater heads at the upstream observation wells were
affected as well, which in turn was compensated by a reduction of the Lout4 transfer rate by a
factor of 3. On the one hand, these adjustments reduced the RMSE to nearly the same level as
for the complete IM method (Section 3.5.1). On the other hand, the direction of the resulting
flow field on the northern side of the river was slightly more parallel to the river direction,
which caused longer flow paths and longer travel times at the nodes identified by the bluish
domain in Fig. 3.8c.
Fig. 3.8d depicts the spatial distribution of the relative age difference for the second
simplified model scenario that used the IM_lin water level distribution. The assumed linear
interpolation between the two water level gauges at river sections
led to lower water
levels by up to 50 cm relative to the complete IM implementation (Fig. 3.4b). Accordingly,

the gradient between the river and the exfiltrating segment of the northern side channel
decreased substantially, which led to a more river-parallel groundwater flow field on the
northern side of the river and a widespread increase of 200-500% in groundwater age. In
56
attempt to compensate for the errors in river water levels during calibration against
groundwater heads, the Lin1 and Lin2 transfer rates were increased by 2 and 1 orders of
magnitude, respectively. Additionally, the Lout4 transfer rate was decreased by 2 orders of
magnitude compared to the IM model scenario. Even though the RMSE was reduced from
16 cm to 11 cm, the groundwater flow field and hence the errors in groundwater age
prediction could not be improved on the northern side of the river. Instead, the very high Lin1
transfer rate caused a 20-80% decrease in groundwater age between the river and the
disconnected river branch, where the water level differences were similar compared to the
complete IM method (Fig. 3.4b).
The errors in the prediction of groundwater age on the southern side of the river were <10%
for the IM_wo_lat model scenario and ranged from -30% in the western part to 30% in the
middle part for the IM_lin model scenario. These small changes in groundwater age compared
to those on the northern side of the river can be attributed to the high absolute water level
difference between the river and the southern side channel, compared to which the errors in
river water levels were small. For instance, the error of 30 cm at the river section
for the
simplified IM_lin method reduced the water level gradient between the river and the southern
side channel by 30%, compared to the complete IM method (Fig. 3.4b). In contrast, the same
error of 30 cm reduced the water level gradient between the river and the exfiltrating segment
of the northern side channel by about a factor of 2.5. This clearly demonstrates that
uncertainty in surface water levels has the highest impact on the simulated groundwater age in
zones where the water level gradients between infiltrating and exfiltrating boundaries are
small.
The description of the river as a 2D domain allowed for including lateral water level gradients.
Lateral gradients in river water levels were described in the literature, but were explained by
stream curvature and the centrifugal force (Cardenas et al., 2004). In restored river reaches,
morphologically induced lateral water level gradients are likely to occur quite frequently. At
our field site, lateral water level differences of 20-30 cm were restricted to a ~200 m long
section (Fig. 3.4b) and their negligence caused errors in groundwater age prediction of 40-80%
(Fig. 3.8c). When compared to the uncertainty of the experimentally determined residence
times of 60-80% (Supporting information: Table 3.2), the errors associated with neglecting
lateral gradients might be acceptable, depending on the purpose and requirements of the study.
At other field sites, however, lateral water level gradients can be more pronounced and their
inclusion might be crucial for an accurate groundwater age prediction.
57
The explicit consideration of the river as a 2D domain was also essential to reliably represent
the rivers lateral extent and therewith the length of the groundwater flow paths as well as the
groundwater residence times. Furthermore, the 2D representation of the river in its full lateral
extent allowed us to account for asymmetric groundwater flow underneath the river, which
was identified as an important feature influencing the infiltration rates towards the northern
and the southern side of the river (Fig. 3.7, Section 3.5.2). In previous modeling studies, the
river was described as a 2D domain as well, but was cut in approximately the middle where a
no-flow boundary condition was assumed (Derx et al., 2010). This assumption suppresses
potential asymmetric groundwater flow underneath the river and might lead to a bias in the
flow field and the water budget.
Linearly interpolating water levels between water level gauges separated by one or several
kilometers is common practice when assigning river water levels to models of rivergroundwater systems. Our results revealed that assuming a linear water level distribution can
lead to considerable errors in the river water level distribution that translate into inacceptable
errors in the simulated groundwater age of >200%. Hence, the accurate description of the
longitudinal water level distribution is of major importance for a reliable groundwater age
prediction.
3.6 Conclusions
In this study, we assessed the predictive capability of two new alternative methods for the
estimation of 1D and 2D river water level distributions of dynamic rivers (RM, IM), with
respect to the simulated groundwater residence time (groundwater age). Surface water levels
generated with both alternative methods and with a reference method (RH) were assigned to
the river and side channel boundary conditions of a 3D groundwater flow and transport model
of a partly restored riverbank-filtration system in NE-Switzerland. Steady-state model
calibration against measured groundwater heads was performed for each of these model
scenarios by an automated adjustment of selected transfer rates using PEST. The age
predictions of the calibration-constrained groundwater flow fields lay within a range of 30%
compared to the reference RH model scenario. This relatively low error confirmed the
predictive capability of the alternative methods when applied to real and complex rivergroundwater systems.
We also investigated the sensitivity of the modeled groundwater age distribution to reduced
complexity in the river water level distribution. For the first scenario, we modified the IM
method to ignore lateral gradients, which led to errors in groundwater age prediction of 40-80%
58
over a considerable area to the north of the river. In the second scenario, we further simplified
the IM method by assuming a linear longitudinal water level distribution. As a result, errors in
groundwater age of 200-500% were widespread, which demonstrates the importance of an
accurate longitudinal water level distribution for the modeled groundwater age.
The results of this study allow us to recommend both alternative approaches presented in
Chapter 2 for the river water level assignment in future modeling studies of river-groundwater
systems at the kilometer scale. To implement either of the alternative methods at a specific
river-groundwater system, the placement of the water level gauges and the fixpoints should be
carefully assessed. First, it is important to note that an accurate description of the surface
water levels is most important in zones where the gradients between infiltrating and
exfiltrating boundaries are small. Second, our results indicate that the longitudinal water level
distribution should be captured in detail to reliably simulate the groundwater flow field and
the groundwater age distribution. We suggest that, if feasible, water level gauges should be
installed at 1 km intervals. In between, fixpoints (e.g. an armor stone or a steel rod) should be
installed and leveled with a spacing that is inversely proportional to the change in surface
water level gradient and might range from 50-200 m. For instance, a segment with a linear
water level profile can be captured by two fixpoints, while a segment with a changing
gradient requires three or more fixpoints. Additionally, fixpoints should be installed upstream
and downstream of a hydraulic jump, for example at beaver dams or weirs. To maximize the
accuracy in groundwater age prediction, we recommend the inclusion of lateral water level
gradients by defining two fixpoints on the same section.
This study demonstrates that a reduced level of detail in the river water level distribution can
lead to considerable errors in simulated groundwater flow paths and residence times.
Therefore, it is essential to capture the river water level distribution in its full spatial and
temporal extent. To this end, the new methods proposed in Chapter 2 proved to offer an
accurate and efficient alternative compared to using a hydraulic model. The application of
these interpolation methods when modeling riverbank-filtration systems will, for instance,
help to reliably assess the impact of river restoration measures on groundwater residence
times and hence, to mitigate the conflict of interest between river restoration and drinking
water protection.
59
Acknowledgments
This study was conducted within the National Research Program Sustainable Water
No. 406140-125856). We would like to thank Matthias Rudolf von Rohr, Lena Froyland and
Urs von Gunten for their support. We warmly thank Ryan North and John Molson for many
helpful discussions. The Agency for the Environment of the Canton Thurgau provided data,
logistics and financial support. Additional support was provided by the Competence Center
Environment and Sustainability (CCES) of the ETH domain in the framework of the
RECORD (Assessment and Modeling of Coupled Ecological and Hydrological Dynamics in
the Restored Corridor of a River (Restored Corridor Dynamics)) and RECORD Catchment
projects.
60
Fig. 3.9. Vertically averaged spatial groundwater age distribution of the calibrated RH model
scenario. At the labeled observation wells, estimates of the local groundwater residence time
distribution were obtained by nonparametric deconvolution of EC time series (Table 3.2).
Fig. 3.10. Estimated residence time distributions g() from nonparametric deconvolution of
the EC time series measured in the Thur River and in groundwater (a) at an observation well
close to the river (R050) and (b) at an observation well close to the pumping well (F3) (Fig.
3.9). Solid line: mean of 1000 conditional realizations, gray area: 16 and 84 percentiles of the
conditional statistical distributions, dotted lines: minimum and maximum values obtained in
all realizations. The mean residence time (center of gravity, tc(g)) and the standard deviation
((g)) of both residence time distributions are indicated.
61
Table 3.2. Mean residence times (center of gravity, tc(g)) and standard deviations ((g)) of the
residence time distributions obtained by nonparametric deconvolution of the EC time series at
the labeled observation wells in Fig. 3.9 (Froyland, 2011). These residence times were used as
observations for the initial model calibration (Section 3.4.2). The standard deviations ((g))
typically amounted to 60-80% of the mean residence times (tc(g)).
R050
R073
R051
R059
R006
F3
R069
R081
R049
tc(g) [d]
0.6
1.1
3.5
3.8
3.2
8.9
1.5
2.3
4.5
62
(g) [d]
0.5
0.7
2.4
3.5
1.8
6.0
1.3
1.8
3.8
Fig. 3.11. Flow chart describing the initial model calibration using the reference RH water
level distribution and the subsequent model calibration using the water level distributions of
the alternative and the simplified methods.
63
Chapter 4
NOM degradation during river infiltration: Effects of the
climate variables temperature and discharge
Published in Water Research
Diem, S., Rudolf von Rohr, M., Hering, J.G., Kohler, H.P.E., Schirmer, M., von Gunten,
U., 2013. NOM degradation during river infiltration: Effects of the climate variables
temperature and discharge. Water Research, doi: 10.1016/j.watres.2013.08.028.
Abstract
Most peri-alpine shallow aquifers fed by rivers are oxic and the drinking water derived by
riverbank filtration is generally of excellent quality. However, observations during past heat
waves suggest that water quality may be affected by climate change due to effects on redox
processes
such
as
aerobic
respiration,
denitrification,
reductive
dissolution
of
manganese(III/IV)- and iron(III)(hydr)oxides that occur during river infiltration. To assess the
dependence of these redox processes on the climate-related variables temperature and
discharge, we performed periodic and targeted (summer and winter) field sampling campaigns
at the Thur River, Switzerland, and laboratory column experiments simulating the field
conditions. Typical summer and winter field conditions could be successfully simulated by
the column experiments. Dissolved organic matter (DOM) was found not to be a major
electron donor for aerobic respiration in summer and the DOM consumption did not reveal a
significant correlation with temperature and discharge. It is hypothesized that under summer
conditions, organic matter associated with the aquifer material (particulate organic matter,
POM) is responsible for most of the consumption of dissolved oxygen (DO), which was the
most important electron acceptor in both the field and the column system. For typical summer
conditions at temperatures >20C, complete depletion of DO was observed in the column
system and in a piezometer located only a few meters from the river. Both in the field system
and the column experiments, nitrate acted as a redox buffer preventing the release of
manganese(II) and iron(II). For periodic field observations over five years, DO consumption
showed a pronounced temperature dependence (correlation coefficient r = 0.74) and therefore
a seasonal pattern, which seemed to be mostly explained by the temperature dependence of
65
the calculated POM consumption (r = 0.7). The river discharge was found to be highly and
positively correlated with DO consumption (r = 0.85), suggesting an enhanced POM input
during flood events. This high correlation could only be observed for the low-temperature
range (T<15C). For temperatures >15C, DO consumption was already high (almost
complete) and the impact of discharge could not be resolved. Based on our results, we
estimate the risk for similar river-infiltration systems to release manganese(II) and iron(II) to
be low during future average summer conditions. However, long-lasting heat waves might
lead to a consumption of the nitrate buffer, inducing a mobilization of manganese and iron.
4.1 Introduction
Riverbank filtration is a widely applied technique to produce drinking water and contributes
substantially to the overall drinking water production in several European countries (France
~50%, Germany ~16% (Tufenkji et al., 2002), Switzerland ~25%). Natural attenuation
processes during river infiltration efficiently remove particles, bacteria, viruses, parasites and,
to a lesser extent, organic contaminants, such as pharmaceuticals (Kuehn and Mueller, 2000;
Sacher and Brauch, 2002; Grnheid et al., 2005). During river infiltration, biogeochemical
processes can alter the composition of the infiltrating water significantly (Jacobs et al., 1988).
The most important biogeochemical process during river infiltration is the biodegradation of
natural organic matter (NOM), which occurs within bacterial biofilms in riverbed sediments
(Pusch et al., 1998). NOM in river systems originates from both allochthonous (terrestriallyderived) sources and generally more biodegradable autochthonous sources (periphyton)
(Pusch et al., 1998; Leenheer and Croue, 2003). NOM is composed of dissolved organic
matter (DOM) and particulate organic matter (POM), usually quantified as dissolved organic
carbon (DOC) and particulate organic carbon (POC), respectively (Leenheer and Croue,
2003). During infiltration of river water, DOM is transported through the riverbed as a
mobile substrate, whereas POM is retained in the riverbed sediments as a stationary
substrate (Pusch et al., 1998).
The biodegradation of NOM in riverbed sediments leads to a consumption of dissolved or
solid terminal electron acceptors, such as oxygen (O2), nitrate (NO3-), Mn(III/IV)- and
Fe(III)(hydr)oxides and sulfate (SO42-). Redox conditions in riverbank-filtration and artificialrecharge systems were observed to undergo seasonal variations with the formation of anoxic
conditions during summer due to the temperature dependence of NOM degradation
(Greskowiak et al., 2006; Massmann et al., 2006; Sharma et al., 2012).
66
In the Swiss context, riverbank filtration is often the only barrier between river water and
drinking water. This is possible because of the usually high dilution of wastewater effluents in
receiving rivers and the generally oxic conditions of shallow groundwater. However, during
the hot summer of 2003, the redox conditions in several riverbank-filtration systems turned
anoxic. Hoehn and Scholtis (2011) reported a case at the Thur River, where the redox
sequence even proceeded to Mn(IV)- and Fe(III)-reducing conditions. The subsequent reoxidation of dissolved Mn(II) and Fe(II) at the pumping station led to clogging of the filter
screen and to rusty water.
Climate models predict an increase in summer air temperatures (4-5 K) and a decrease in
precipitation (25%) inducing lower discharges in rivers during summer months in northern
Switzerland by 2085 (CH2011, 2011; FOEN, 2012a). Lower discharges may give rise to less
dilution of wastewater effluents and accordingly to higher DOC concentrations. Combined
with higher temperatures, the risk for riverbank-filtration systems to become anoxic or even
develop Mn(IV)- or Fe(III)-reducing conditions is likely to increase (Sprenger et al., 2011).
To assess this risk more accurately, the dynamics of NOM degradation and its dependence on
climate variables need to be better understood (Eckert et al., 2008; Green et al., 2011).
Besides the direct influence of temperature and discharge, the biogeochemical processes
during river infiltration and hence the redox conditions in the infiltration zone might also be
affected by indirect climate-related changes in river water quality. The effect of climate
change on river water quality (e.g. dissolved oxygen, nutrients (nitrate, ammonium), DOC and
major ions) in relation to hydrologic, terrestrial and resource-use factors has been addressed in
many studies (Murdoch et al., 2000; Zwolsman and van Bokhoven, 2007; Park et al., 2010).
However, Senhorst and Zwolsman (2005) conclude that the impact of climate change on
surface water quality is quite site specific and cannot be generally transferred to other
watersheds and hence, should be assessed case by case. Therefore, we focused on the direct
effect of the climate-related variables temperature and discharge on NOM degradation and the
related consumption of electron acceptors during river infiltration.
The objectives of the present study were to assess the contribution of DOM consumption to
the overall consumption of electron acceptors and to examine the effects of the climate-related
variables temperature and discharge by means of field investigations and column experiments.
To capture different temperature ranges, the field investigations consisted of two detailed
sampling campaigns performed during typical summer and winter conditions at the perialpine Thur River. The column experiments were performed at temperatures that span the
67
range of typical field conditions. Additionally, the data of periodic field samplings that
covered a wide range of temperature and discharge conditions over a period of five years were
evaluated.
Firstly, we assessed the temperature dependence of the consumption of dissolved oxygen (DO)
and DOM by comparing the data of the field campaigns with those of column experiments
and verified the findings by a correlation analysis of the periodic data. Secondly, we
investigated the impact of the discharge conditions on DO and DOM consumption using the
periodic field data. Finally, we discuss the implications of our findings on the redox-related
groundwater quality at riverbank-filtration systems in a changing climate.
4.2 Materials and methods

4.2.1
Field site
The field site of our investigation is located in NE-Switzerland (Niederneunforn) at the perialpine Thur River, which drains a catchment of 1700 km2 (Fig. 4.1a). As no retention basin is
located along the whole course of the river, the discharge behaves very dynamically with a
range of 3-1100 m3/s. During the interdisciplinary RECORD project (Restored corridor
dynamics, http://www.cces.ethz.ch/projects/nature/Record; Schneider et al. (2011); Schirmer
(2013)), 80 piezometers were installed (Fig. 4.1b). To the northern side of the river, the
piezometers are arranged in two transects, the forest transect and the pumping station transect.
Most of the piezometers are fully screened, covering the full aquifer thickness (5.31.2 m).
The gravel-and-sand aquifer, which is underlain by lacustrine clay and overlain by 0.5-3 m
alluvial fines, is highly conductive with hydraulic conductivities between 410-3 and
610-2 m/s (Chapter 3). The pumping well at the pumping station transect is operated during a
daily period of only 3 h extracting a total volume of 36 m3.
A groundwater flow and transport model was set up for low-flow conditions (23 m3/s) and
was calibrated against groundwater heads and experimentally determined travel times
(Chapter 3). According to the resulting groundwater flow field (Fig. 4.1b, c), river water is
naturally infiltrating into the aquifer, both at the forest and the pumping station transect. The
low abstraction rate of the pumping well was found not to have any significant effect on the
infiltration rate or the groundwater flow field. Groundwater flow velocities were in the order
of 5-10 m/d at the pumping station transect and ranged between 20-50 m/d at the forest
transect.
68
Fig. 4.1. (a) Catchment of the Thur River and location of the Niederneunforn field site, NESwitzerland. (b) Schematic representation of the Niederneunforn field site. Groundwater head
isolines (head equidistance 10 cm, light gray) were extracted from a groundwater flow model
(Chapter 3). (c) Enlargement of the area indicated by the dashed rectangle in (b), which
contains the locations of the piezometers used in this study. The black lines in (c) indicate the
advective flow paths from the river to the piezometers according to the resulting flow field of
the groundwater flow model.
Neither the pumping station transect nor the forest transect can be considered as a typical
riverbank-filtration system with high abstraction rates considerably changing the groundwater
flow field. Yet, both transects qualify for studying the microbial degradation processes during
river infiltration and their dependencies on the climate-related variables temperature and
discharge. Compared to the pumping station transect, the forest transect has the advantage of
higher flow velocities and shorter residence times to the closest piezometers, which allows a
more detailed assessment of the NOM degradation dynamics. Therefore, in this chapter, we
focus on results from the forest transect.
4.2.2
Sampling campaigns and periodic samplings
To capture different temperature ranges, we conducted two detailed sampling campaigns

during summer and winter 2011. We sampled the river and adjacent groundwater on five days
between August 19-26, 2011 (summer campaign) and November 23-29, 2011 (winter
campaign). The periodic samplings were carried out during a period of five years (2008-2012)
and covered a wide temperature and discharge range (Supporting information; Section 4.5:
69
Fig. 4.9). Samples were taken in different piezometers at the forest transect. For better clarity,
we only present results from a representative subset of the sampled piezometers (Fig. 4.1c).
The groundwater samples were pumped by a submersible electric pump (Whale, Bangor,
Northern Ireland) with an average pumping rate of 10 L/min. Before taking the in-situ
measurements and the samples, we pumped at least 20 L (twice the piezometer volume) and
waited until the electrical conductivity (EC) of the pumped groundwater was stable. We
measured DO concentrations (LDO10115 (optical sensor), Hach Lange GmbH, Berlin,
Germany, accuracy 0.1 mg/L), pH and temperature (PHC10115, Hach Lange GmbH, Berlin,
Germany, accuracy pH 0.1, T0.3 K) and EC (Cond 340i, WTW GmbH, Weilheim,
Germany) in a 10 L bottle, which was constantly flushed with groundwater. The small
opening on top of the bottle minimized the gas exchange with the atmosphere and thus
guaranteed reliable DO measurements at near-zero concentrations. Groundwater and river
water was filled into polypropylene bottles (1 L), filtered within 24 h through a 0.45 m
cellulose nitrate filter (Sartorius AG, Gttingen, Germany) and stored at 4C until analysis.
Concentrations of DOC (to quantify DOM), nitrate, ammonium and major ions were
measured in the river water and groundwater samples (for analytical method see Section
4.2.4).
DO, as well as temperature and discharge were continuously measured (10-min intervals) in
the Thur River at a gauging station of the Federal Office for the Environment (FOEN), which
is located in Andelfingen, 10 km downstream of our field site (Fig. 4.1a). The river DO
concentrations in Andelfingen were found to agree well with those at our field site (Hayashi et
al., 2012). DO concentrations in the river underwent diurnal fluctuations. Such diurnal DO
fluctuations are mainly caused by a combination of photosynthesis of periphyton during
daytime and respiration during the night (Hayashi et al., 2012).
We calculated the DO consumption that occurred during infiltration by subtracting the
measured DO concentration in groundwater from the daily mean DO concentration in the
river. The latter was calculated based on the continuous DO time series measured at the
gauging station in Andelfingen to minimize the bias in calculated DO consumption due to
diurnal DO fluctuations in the river. As the diurnal fluctuations of the DOC concentrations in
the river were not significant, we subtracted the measured DOC concentration in groundwater
from the DOC concentration in the river to calculate the DOM consumption.
Eight grab samples were taken from the riverbed at a depth of 0-20 cm close to the first
piezometer R050 (Fig. 4.1c). The samples were dried and sieved; the riverbed sediment
70
mainly consists of sandy gravel with little silt and clay. To quantify the POM contained in the
riverbed sediment, we measured the POC concentrations for three grain-size fractions
<0.25 mm. The POC concentration was highest for the fraction <0.063 mm (1.40.3% w/w),
and lowest for the fraction 0.125-0.25 mm (0.50.3% w/w) (for analytical method see Section
4.2.4).
4.2.3
Column experiments
A schematic representation of the setup of the column experiments is shown in Fig. 4.2. The
column casing consisted of a Plexiglas tube (length 30 cm, inner diameter 5.2 cm) and was
packed with fractionated sand (0.125-0.25 mm grain size) from a gravel bar at the field site
close to the forest transect. The sand was dried at room temperature and sieved afterwards
before being dry-filled into the column in form of a sand rain (von Gunten and Zobrist,
1993). The sand was mainly composed of calcite and quartz (40% and 25%, respectively) and
the POC concentration was about 0.30.2% (w/w). The sand fraction 0.125-0.25 mm was
chosen because it is well defined and represents the available reactive surfaces with a
considerable amount of POM.
Filtered Thur River water (0.45 m, cellulose nitrate, Sartorius AG, Gttingen, Germany) was
stored in a 2 L tank and was used as feed water for the column. It was pumped from the
bottom to the top of the column at a flow rate of 0.4 L/d by means of an HPLC pump (Jasco
PU-2080, Jasco Corporation, Tokyo, Japan) (Fig. 4.2). Every three days, the storage tank was
replenished with fresh Thur River water stored at 5C. In case of the experiment at 20C,
Thur River water was allowed to equilibrate for about 6 hours before replenishment. No
measurable DOM degradation in the storage tank was observed under these conditions. To
assess the hydraulics in the column, a tracer test with a 8.55 mM NaCl solution was
conducted and the EC was measured at the end of the column. We estimated an effective
porosity of 0.32 and a dispersivity of 0.08 cm by inverse modeling with the software CXTFIT
(Toride et al., 1995). The corresponding total pore volume of the column was 0.2 L. Hence,
the residence time in the column was 0.5 d at a flow rate of 0.4 L/d.
The column was first operated at 20C in a climatised chamber for 19 days after an
equilibration time of about 2 months with Thur River water taken on July 4, 2012
(composition 1, Supporting information: Table 4.1). After that, the column was operated at
5C with the same feed water in an incubator for 26 days including an equilibration time of 20
days. Thereafter, as a control experiment, the column was operated again at 20C in an
incubator for 28 days including an equilibration time of 22 days (water composition 2,
71
Supporting information: Table 4.2). This control experiment was conducted to test if any of
the column properties relevant for the bacterial degradation processes (sand composition,
POM concentration/composition) had changed during the operation period at 5C.
Fig. 4.2. Schematic representation of the setup of the column experiment with 13 sampling
ports along the column (SP1-13) and one at the inlet (SP0) and one at the outlet (SP14). River
water was pumped from the bottom to the top of the column. DO, temperature (T) and pH
were measured in flow-through cells (F).
The column featured 15 sampling ports; one at the inlet, 13 along the column and one at the
outlet (Fig. 4.2). DO and temperature were continuously measured at the inlet and at the outlet
of the column (LDO101 (optical sensor), Hach Lange GmbH, Berlin, Germany, accuracy
0.1 mg/L, T0.3 K), while pH was measured after the column (PHC 301, Hach Lange
GmbH, Berlin, Germany, accuracy pH 0.1) in flow-through cells (Fig. 4.2). For each of the
experiments at different temperatures, we measured three DO concentration profiles after an
equilibration time of 20 d. DO was measured directly at each sampling port in a flowthrough cell. After that, about 40 mL of sample volume was taken at the column inlet (SP0)
and outlet (SP14) for the DOC and nitrate analyses by connecting a rinsed regenerated
0.45 m cellulose filter (National Scientific Company, Rockwood, USA) to the port. The
sampling procedure was conducted from the outlet to the inlet of the column, following the
opposite direction of the water flow.
72
The DO consumption in the column was calculated by subtracting the measured DO

concentration at sampling port SP14 from the DO concentration at SP0 0.5 d before. The
DOM consumption was determined from the difference of DOC concentrations at SP14 and
SP0.
4.2.4
Analytical methods
DOC concentrations were measured with a Shimadzu TOC-V CPH (Shimadzu Corporation,
Kyoto, Japan). Nitrate and the other major ions were analyzed by means of a Metrohm 761
Compact IC (Metrohm Schweiz AG, Zofingen, Switzerland). Ammonium was measured with
a Spectrophotometer Varian Cary 50 Bio (Varian BV, Middelburg, The Netherlands). The
POC concentrations in sediment samples were determined by subtracting the inorganic carbon
fraction, measured with a CO2 Coulometer CM5015 (UIC Inc., Joilet, USA), from the total
carbon fraction, measured with a CNS analyzer Eurovector EA3000 (Hekatech GmbH,
Wegberg, Germany).

4.3.1
Hydraulic conditions and concentration profiles of redox-active compounds
The winter and the summer sampling campaigns were conducted during low-flow conditions.
The daily mean discharge of the Thur River during the field sampling campaigns varied
between 13.7 and 24.8 m3/s in summer and between 5.4 and 6.1 m3/s in winter (Fig. 4.3). The
daily mean temperatures in the river ranged from 20.9 to 22.9C in summer and from 5.9 to
6.7C in winter (Fig. 4.3).
The groundwater residence times between the river and the piezometers were estimated based
on a groundwater flow model as well as the analysis of EC time series (Chapter 3) and ranged
from 0.5 d at the piezometer closest to the river (R050) to 13 d at the piezometer furthest from
the river (R023) (Fig. 4.1c, Fig. 4.4). As the mean residence time in the column was 0.5 d, the
column well represented the situation at piezometer R050. The column experiments
additionally allowed resolving the microbially-mediated redox processes on a scale that was
not accessible in the field system.
73
Fig. 4.3. Discharge (black line) and temperature (gray line) time series measured at the FOEN
gauging station in Andelfingen covering the two detailed summer and winter sampling
campaigns (August 19-26, 2011 and November 23-29, 2011). Each of the sampling days is
shown as a dashed vertical line.
The temperature and the concentration profiles of selected redox-active compounds are shown
in Fig. 4.4 for the sampling campaigns and in Fig. 4.5 for the column experiments. Most of
the DO and DOM consumption in the field occurred between the river and the first
piezometer (R050), both during summer and winter conditions. This observation indicates that
most of the degradation processes took place within the first meters of the infiltration zone
and shows that, in agreement with other studies (Bourg and Bertin, 1993; Brugger et al.,
2001a; Sobczak and Findlay, 2002), the microbial activity was highest in this zone. For the
piezometers further away from the river, an additional decrease in DO and DOC was observed
in winter. However, considering the longer travel times to these piezometers, the
corresponding degradation processes occurred much slower.
74
Fig. 4.4. Temperature and concentration profiles of DO, DOC and nitrate during (a) the
summer and (b) the winter campaign. Mean and standard deviations are shown as symbols
and error bars, respectively (n=5). Estimated groundwater residence times () between the
river and the piezometers are also indicated.
The DO concentration profiles for the column experiments revealed that the highest DO
consumption rate occurred between the first two sampling ports. This is in accordance with
other column experiments, for which the highest DO consumption rate was observed within
the first centimeter of the column (von Gunten and Zobrist, 1993; von Gunten et al., 1994). At
20C, the initial decrease of DO was followed by a linear decrease in the column, while at
5C the DO concentration remained constant after the initial decrease (Fig. 4.5a). The linear
decrease at 20C suggests a zero-order degradation rate, which means that the DO
consumption in the column was not limited by the substrates DO and NOM. The DO profile
of the control experiment (performed after the 5C experiment) was almost identical to the
DO profile of the first experiment at 20C (Supporting information: Fig. 4.11). This indicates
that the column properties decisive for the bacterial degradation processes did not change
during the operation period at 5C.
We consider aerobic respiration of NOM as the only process responsible for the consumption
of DO. Based on measured ammonium concentrations in river water, nitrification (oxidation
of ammonium) potentially accounted for 2% of the DO consumption in the field campaigns
and the column experiments, and can therefore be neglected. The removal of DOM during
75
river infiltration might be attributed to both microbial degradation and abiotic sorption
processes (Brugger et al., 2001b). However, the summer and winter field sampling campaigns
were conducted during relatively stable temperature and discharge conditions. It is therefore
reasonable to assume that the sorption processes were in steady state and did not affect the
removal of DOM. The same is true for the column system, which was equilibrated for 20 d
before the measurements were taken. Therefore, we assume that microbial degradation
processes dominated the abiotic sorption processes, which is in agreement with Sobczak and
Findlay (2002).
The DO consumption between the river and the first piezometer (R050) was larger in summer
than in winter and that between the column inlet and column outlet was larger at 20 than at
5C. This is a clear indication of the temperature dependence of the microbially mediated
degradation of NOM by aerobic respiration. In contrast, the consumption of DOM was similar
during summer and winter conditions with about 0.7 mg C/L (Fig. 4.4) during the field
sampling campaigns and about 0.3 mg C/L (Fig. 4.5) in the column experiments. Furthermore,
the DOM was not consumed completely, both during the field sampling campaigns and in the
column experiments. Only a fraction of about 30-50% of the total DOM was degraded,
corresponding to the biodegradable DOM (BDOM). Similar values for the BDOM fraction
were reported in several field studies (Sobczak and Findlay, 2002; Sharma et al., 2012). 50-70%
of the DOM remained, presumably due to its recalcitrant nature.
During summer conditions, DO was nearly completely consumed at the first piezometer R050
(Fig. 4.4). Similarly, almost anoxic conditions were observed at the column outlet at 20C
(Fig. 4.5). However, denitrification was not observed either in the field, or in the column.
Apparently, under summer conditions, there is enough nitrate available to act as a redox
buffer preventing Mn(IV)- and Fe(III)-reducing conditions in the infiltration system.
76
Fig. 4.5. Concentration profiles of DO, DOC and nitrate in the column at (a) 20C and (b) 5C.
Each symbol represents a sampling port. The first data point corresponds to the sampling port
at the inlet of the column (SP0), the last data point to the sampling port at the outlet of the
column (SP14). Mean and standard deviations are shown as symbols and error bars,
respectively (n=3).
4.3.2
Impact of temperature on DO and DOM consumption
4.3.2.1 Sampling campaigns and column experiments

To assess the impact of temperature on the aerobic NOM degradation, we compared the DO
and DOM consumption between the river and the first piezometer (R050), and between the
column inlet and outlet for summer and winter conditions. The resulting mean DO and DOM
consumption for the field campaigns and column experiments are shown in Fig. 4.6. For a
simplified version of aerobic respiration, one mole of DO (O2) is used to oxidize one mole of
organic carbon (CH2O) (Eq. (4.1)):
(4.1)
Accordingly, if DOM consumption would explain the entire DO consumption, the
corresponding molar consumptions should be equal (bars in Fig. 4.6). However, during
summer conditions (both in the field and in the column), DOM consumption explained only
10-20% of DO consumption (Fig. 4.6). The remaining 80-90% of the reduction capacity to
77
explain the DO consumption must therefore have been provided by other sources. POM in
riverbed sediments and the column sand is an obvious source of additional reduction capacity
(Brugger et al., 2001b; Sharma et al., 2012).
As mentioned in Section 4.3.1, DO consumption in the field and the column system was much
smaller for lower than higher temperatures, while the DOM consumption remained at about
the same level (Fig. 4.6) and did not seem to be significantly affected by temperature. As a
result, DOM consumption accounted for larger fractions of 50% and 100% of the DO
consumption at lower temperatures in the column and the field, respectively. Furthermore, the
different behavior during summer and winter conditions suggests that the variability and
temperature dependence of DO consumption are determined by the POM, rather than the
DOM consumption.
Fig. 4.6. Comparison of DO and DOM (expressed as DOC) consumption for the summer (2023C) and winter (5-7C) field campaigns and for the column experiments at 20 and 5C,
respectively. The consumption refers to the concentration difference between the Thur River
and the piezometer R050 in the field and between column inlet and outlet for the column
experiments. Standard deviations are shown as error bars.
4.3.2.2 Periodic samplings
To test the dependence of the DO and DOM consumption on temperature more systematically,
we compiled the corresponding data of all the samplings taken between 2008 and 2012 at the
Thur River and the piezometer R050. We additionally calculated the POM consumption
according to Eq. (4.2) assuming that the POM consumption corresponds to the difference
between DO and DOM consumption. This implies that the consumption of DO (O2) and
DOM (CH2O) is solely related to aerobic respiration (Section 4.3.1) and occurs at a molar
ratio of one (Eq. (4.1)).
(4.2)
78
Fig. 4.10 (Supporting information) shows the DOM and the calculated POM consumption of
all the 45 periodic samplings, including the two sampling campaigns of summer and winter
2011. Fig. 4.7 shows a plot of the DO consumption, the DOM consumption and the calculated
POM consumption as a function of the daily mean river water temperature in scatter plots
together with the calculated correlation coefficients. The DO consumption showed a high and
significant correlation (r = 0.74, Fig. 4.7a) with the daily mean river water temperature. Thus
temperature explained the variation in DO consumption to a high degree. However, DOM
consumption was not correlated with river water temperature (Fig. 4.7b). This result coincides
with the data shown in Fig. 4.6 and is in agreement with findings from Brugger et al. (2001b).
Since the calculated POM consumption (Fig. 4.7c) is based on the difference between one
highly-correlated parameter (DO consumption) and one uncorrelated parameter (DOM
consumption), it is necessarily the case that POM consumption will also be highly correlated
with temperature (r = 0.7).
This result supports the conjecture that the POM consumption is primarily responsible for the
variability and temperature dependence of DO consumption, as stated in Section 4.3.2.1. The
divergence between the temperature dependencies observed for DO and DOM consumption
also have the consequence that DOM consumption accounts for a larger proportion of DO
consumption at lower than at higher temperatures. At lower temperatures (<10C), DOM
consumption accounted for 50-100% of DO consumption, while at high temperatures (>15C),
DOM consumption accounted for 0-40% of DO consumption (Fig. 4.7). Correspondingly, the
calculated POM consumption makes a more important contribution at high temperatures,
suggesting that POM is the most important electron donor under these conditions.
79
Fig. 4.7. Scatter plots between (a) DO consumption (DO), (b) DOM consumption (DOC),
(c) calculated POM consumption (POC =DO-DOC) at the piezometer R050, and daily
mean river water temperature (T). r is the correlation coefficient between two variables with
the significance levels: *** = p<0.001, ** = p<0.01, * = p<0.05, (-) = p>0.05. p is the
probability of two variables being uncorrelated.
4.3.2.3 Conceptual explanation
The fact that no correlation was observed between DOM consumption and temperature does
not mean that this process is not temperature dependent. One conceptual explanation is based
on the fact that consumption of the biodegradable fraction of DOM (BDOM) occurs generally
at high degradation rates as it is present in a dissolved and bioavailable form. Even though the
BDOM degradation rates might be lower at lower temperatures, bacteria most likely were
able to completely consume the full BDOM fraction within the residence times of our field
and column systems. Accordingly, on our scales of observation, the DOM consumption was
limited by the amount of BDOM rather than by temperature-dependent rates. In contrast,
POM consumption involves the cleavage of organic macromolecules into soluble monomers
by extracellular hydrolytic enzymes (Egli, 1995; Pusch et al., 1998). Therefore, degradation of
POM generally occurs at lower rates than degradation of BDOM (Greskowiak et al., 2006;
80
Sharma et al., 2012) and the temperature dependence of POM degradation could be resolved
on the timescales of our experimental systems (field, column).
This conceptual explanation is based on the assumption that DOM resulting from hydrolysis
of POM in bacterial biofilms does not contribute significantly to the DOM pool in the
sampled groundwater, and hence did not affect the observed DOM consumption (Fig. 4.7b).
In literature, the hydrolysis of POM is considered to be the rate-limiting step in POM
degradation, which implies that the DOM resulting from hydrolysis undergoes fast
biodegradation (Valentini et al., 1997; Henze et al., 1999; Vavilin et al., 2008). Furthermore,
there is evidence that DOM uptake and transformation by bacteria is very efficient in biofilms
(Pusch et al., 1998). Bacterial biofilms in sediments act thus as effective sinks of DOM that
originated either from the river (as transported/mobile substrate) or from the hydrolysis of
POM (stationary substrate).
4.3.3
Impact of discharge on DO and DOM consumption
The temperature dependence of DO consumption might be superimposed by a dependence on

hydrologic conditions. High-discharge conditions or flood events are generally not considered
to affect the redox conditions in the infiltration zone (Sprenger et al., 2011), but rather to
increase the risk of a breakthrough of contaminants or pathogens at the pumping well due to
shorter groundwater residence times (Schubert, 2002). However, it is well known that the
suspended POM concentration in the river increases with river discharge, as a result of an
increased mobilization of allochthonous and autochthonous POM (Meybeck, 1982). The
contribution of periphyton (autochthonous POM) to the suspended POM pool in a river was
found to positively correlate with discharge, because of the increased abrasive action at the
riverbed (Uehlinger, 2006; Akamatsu et al., 2011). Furthermore, it has been observed that the
POM import into gravelly riverbed sediments mostly occurred during flood events, which
increased the POM concentrations and the microbial respiration rate in the riverbed sediments
(Naegeli et al., 1995; Brunke and Gonser, 1997). Hence, the higher POM availability in the
riverbed during high-discharge conditions might give rise to an increased DO consumption
during river infiltration.
The overall correlation between DO consumption and the daily mean discharge (measured in
Andelfingen) was small and not significant (Supporting information: Table 4.3). However,
DO consumption increased with increasing discharge up to 60 m3/s and leveled off at 200300 mol/L beyond this point (Fig. 4.8). As the data for discharges below 60 m3/s covered a
wide range of temperatures, DO consumption showed only a weak correlation with discharge
81
(r = 0.31). To compensate for the temperature dependence, we defined two temperature

ranges, T<15C (circles) and T>15C (triangles). Following this separation, we found a
significant correlation between DO consumption and discharge for the low-temperature range
(r = 0.85). DOM consumption was not correlated with discharge (Supporting information:
Table 4.3) suggesting that the increase in DO consumption is probably caused by an enhanced
POM consumption, which supports the notion described above.
For the high-temperature range, the correlation between DO consumption and discharge was
small and not significant. This result can be explained by the fact that DO consumption for the
high-temperature range was already high (200-300 mol/L) and nearly complete at low
discharges. Hence, an increase in discharge could not lead to a significant increase in DO
consumption.
Another process that might explain the increased DO consumption at higher discharges is the
leaching of DOM from the soil and vegetation (root zone) at high groundwater tables. At our
field site, this effect was found to be restricted to a zone dominated by the pioneer plant salix
viminalis (willow bush) (Peter et al., 2012). The piezometer R050 was located close to the
river, where the gravel-and-sand aquifer is covered by a relatively thin layer of alluvial fines
(0.3 m) with no or only little grass vegetation. For this zone, the input of DOM from the soil
and vegetation was found to be negligible (Peter et al., 2012).
Fig. 4.8. Scatter plot between DO consumption at the piezometer R050 (DO) and daily mean
river discharge (Q). r is the correlation coefficient between the two variables with the
significance levels: *** = p<0.001, ** = p<0.01, * = p<0.05, (-) = p>0.05. p is the probability
of two variables being uncorrelated.
Low-discharge conditions are expected to promote the DO consumption as a result of
increased groundwater residence times and higher loads of DOM in the river due to less
dilution of wastewater effluent (Sprenger et al., 2011). In our observations, however, DO
consumption did not reveal a significant increasing trend for decreasing discharges, regardless
82
of the considered temperature range (Fig. 4.8). Furthermore, the DOM consumption did not
reveal any significant correlation with discharge, but was positively correlated with DOC
concentrations in the river (Supporting information: Table 4.3). However, DOC
concentrations in the river increased rather than decreased with discharge; the correlation was
weak but significant. This suggests DOM sources other than wastewater treatment plants.
4.3.4
Implications for groundwater quality at riverbank-filtration systems
The latest climate change scenarios predict an increase in summer and winter air temperature
of 4-5 K for NE-Switzerland by 2085. Furthermore, extreme events, such as summer heat
waves, are expected to occur with a higher frequency and intensity (CH2011, 2011). River
discharges are expected to decrease in summer and increase in winter. River water
temperature, which is decisive for the degradation processes in the riverbed, is likely to
increase by the same extent as air temperature, especially during low-flow conditions (FOEN,
2012a). During the hot summer of 2003, the mean river water temperature of the Thur River
was 22C for a period of 70-80 d. In the future, we might have to deal more frequently with
extreme events of similar or longer duration and even higher mean river water temperatures.
According to our results, the calculated POM consumption showed a pronounced temperature
dependence and accounted for most of the DO consumption during summer. Consumption of
DO was enhanced during high-flow conditions, probably due to an increased import of POM
into the riverbed. On our scales of observation, we could not identify any temperature
dependence of the DOM consumption, which was rather limited by the fraction of BDOM.
Moreover, the expectation that low-discharge conditions would lead to higher DOC
concentrations in the river due to less dilution of wastewater effluents, and hence to an
increased DOM and DO consumption during river infiltration, is not supported by our data.
Sprenger et al. (2011) assessed the vulnerability of riverbank-filtration systems to climate
change (drought and flood scenarios) with respect to several water quality parameters (DOM,
nutrients, pathogens). The vulnerability of a riverbank-filtration system to climate change
with respect to the redox milieu in groundwater might depend on three major factors. (1) The
hydraulic connection between the river and groundwater. If infiltration occurs through an
unsaturated zone, which is aerated, complete DO depletion is not expected (Huggenberger et
al., 1998; Hoehn and Scholtis, 2011). During infiltration in direct hydraulic connection to the
groundwater table (saturated), which was the case at our field site, DO is not replenished and
might become depleted, as observed during summer. (2) Based on our results, we consider the
catchment characteristics as another decisive factor. Rivers in catchments without a retention
83
basin (e.g. a lake) have a more dynamic discharge regime and are likely to have a higher POM
load, especially during flood events. (3) The grain-size distribution of the riverbed sediments.
In gravelly riverbeds, the import of POM is possible to deeper layers, while in sandy or
clogged riverbeds, the import of POM might be physically hindered (Brunke and Gonser,
1997). We therefore anticipate the highest vulnerability to climate change with respect to the
redox milieu for riverbank-filtration systems at which infiltration occurs through a gravelly
riverbed in direct hydraulic connection to the groundwater table and which are located in
catchments without a retention basin.
For such riverbank-filtration systems, we consider future summer heat waves to be critical for
the redox-related groundwater quality. A substantial increase in river water temperature
during future heat waves will enhance the POM turnover, leading to a complete consumption
of DO and potentially nitrate. As a consequence, Mn(II) and Fe(II) could be released, as
observed during the hot summer 2003 (Hoehn and Scholtis, 2011). An additional POM input
into the riverbed during flood events induces a higher consumption of DO and possibly nitrate,
which may enhance the risk for Mn(III/IV)-/Fe(III)-reducing conditions in the infiltration
zone.
The subsequent re-oxidation of Mn(II) and Fe(II) and precipitation of Mn(IV)- and
Fe(III)(hydr)oxides can lead to clogging problems and deterioration of the drinking water
quality (rusty water) at pumping stations of a drinking water supply. For the removal of
dissolved Mn(II) and Fe(II), conventional pump-and-treat techniques can be applied, which
are based on physical-chemical or biological processes (Mouchet, 1992). Alternatively, in-situ
techniques can be used, in which the oxidation processes take place directly in the aquifer
(Mettler et al., 2001).
4.4 Conclusions
We performed field investigations and laboratory column experiments to investigate the
dynamics of redox processes that occur during river infiltration and their dependence on the
climate-related variables temperature and discharge. The observations of the summer and
winter sampling campaigns in the field could be successfully reproduced by column
experiments. Particulate organic matter (POM) was identified as the main electron donor for
dissolved oxygen (DO) consumption during summer conditions and both the DO and the
calculated POM consumption revealed a pronounced temperature dependence. The DO
consumption was enhanced during flood events, presumably due to an additional POM input
into the riverbed.
84
In our field and column systems, DO was the most important electron acceptor. In summer
(T>20C), DO concentrations in groundwater were close to zero, but denitrification was not
observed. Similarly, most of the Swiss aquifers fed by rivers are (sub)oxic under todays
summer conditions and nitrate buffers the redox system before it becomes Mn(III/IV)- and
Fe(III)-reducing. Therefore, currently there is no need to implement demanganation and
deferisation. However, during future summer heat waves, an enhanced POM turnover could
lead to a full consumption of DO and nitrate, enabling the release of Mn(II) and Fe(II). As the
source, quality and quantity of POM and its input into the riverbed are very difficult to assess,
it is nearly impossible to find direct intervention strategies.
We recommend long-term monitoring of the redox conditions at riverbank-filtration systems,
which are characterized by an infiltration that occurs through a gravelly riverbed in direct
hydraulic connection to the groundwater table, and by catchments without a retention basin.
Long-term data will allow taking adequate measures in time, while accounting for the sitespecific hydrogeological conditions.
Acknowledgments
This study was accomplished within the National Research Program Sustainable Water
No. 406140-125856). We would like to thank Sabrina Bahnmller, Ryan North, Sebastian
Huntscha, Simone Peter and Lena Froyland for their help in the field and the AUA Laboratory,
Jacqueline Traber, Sabrina Bahnmller, Elisabeth Salhi and Irene Brunner for the analytical
work. Moreover, we would like to express our gratitude to Eduard Hoehn and Silvio Canonica
for helpful discussions. The Federal Office for the Environment (FOEN) provided data of the
gauging station in Andelfingen. Additional support was provided by the Competence Center
Environment and Sustainability (CCES) of the ETH domain in the framework of the
projects.
85
Fig. 4.9. Discharge time series covering all the samples taken at the Niederneunforn field site
between 2008 and 2012. Each sampling is shown as vertical dashed line.
86
Fig. 4.10. (a) DOM consumption (DOC) and calculated POM consumption (POC=DODOC) between the river and the piezometer R050 for the periodic samplings between 2008
and 2012 (n=45) including the sampling campaigns of summer and winter 2011. The DO
consumption (DO) corresponds to the sum of the dark and light gray bars. On the x-axis, the
month and the year of each sampling are indicated. The data are sorted by increasing daily
mean river water temperature, which is shown in (b) for each of the samplings.
Fig. 4.11. DO concentration profiles in the column measured during the main experiment
(white circles) and during the control experiment (gray circles) at 20C. The control
experiment was conducted after the experiment at 5C. The mean and the standard deviations
are shown as circles and error bars, respectively (n=3).
87
Table 4.1. Composition 1 of Thur River water used in the column experiments (collected on
July 4, 2012).
DOC
pHa El. conductivityb Alkalinityc NO3-
[mg/L]
[S/cm]
[mmol/L]
Na+
Ca2+
Mg2+
PO42- d
[mg N/L] [mg/L] [mg/L] [mg/L] [g P/L]
NH4+ d
[g N/L]
2.1
8
407
4.08
2.2
11.9
68.0
12.1
35.5
11.6
a
measured with a pH meter PHC 301, Hach Lange GmbH, Berlin, Germany
b
measured with a Metrohm 712 Conductometer, Metrohm Schweiz AG, Zofingen, Switzerland
c
measured with a Metrohm 809 Titrando, Metrohm Schweiz AG, Zofingen, Switzerland
d
measured with a Spectrophotometer Varian Cary 50 Bio, Varian BV, Middelburg, The Netherlands
Table 4.2. Composition 2 of Thur River water used in the column experiments (collected on
August 13, 2012).
DOC
pH El. conductivity Alkalinity NO3-
[mg/L]
2.1
Na+
Ca2+
Mg2+
PO42-
NH4+
[S/cm]
[mmol/L] [mg N/L]
[mg/L]
[mg/L] [mg/L] [g P/L] [g N/L]
410
3.83
18.2
58.9
2.1
13.7
14.3
19.0
Table 4.3. Correlation coefficients between the parameters DO consumption (DO), DOM
consumption (DOC), DOC concentration in the river (DOC river), calculated POM
consumption (POC =DO-DOC), daily mean river water temperature and daily mean river
discharge for the periodic samplings at the piezometer R050 (n=45). Significance levels: ***
= p<0.001, ** = p<0.01, * = p<0.05, (-) = p>0.05. p is the probability of two variables being
uncorrelated.
DO
DOC
DOC river
POC
DOC
0.15 (-)
DOC river
0.35*
0.61***
POC
0.87***
-0.35*
0.029 (-)
88
Temperature
0.74***
-0.0038 (-)
0.14 (-)
0.70***
Discharge
0.23 (-)
-0.066 (-)
0.46**
0.25 (-)
Chapter 5
Modeling the dynamics of oxygen consumption during
riverbank filtration by a stochastic-convective approach
In revision for Journal of Hydrology
Diem, S., Cirpka, O.A., Schirmer, M., in revision for Journal of Hydrology. Modeling the
dynamics of oxygen consumption upon riverbank filtration by a stochastic-convective
approach.
Abstract
Dissolved oxygen (DO) is an important groundwater-quality parameter. In riverbank
sediments, a strong decrease of DO over the distance of a few meters has frequently been
observed. The consumption rates may vary in time, which puts the representativeness of
common, sporadic DO measurements in groundwater, based on monthly or even yearly
sampling, into question. We present a new modeling approach that allows efficiently
estimating DO concentrations in alluvial groundwater from measured DO concentrations in
the river under various temperature and discharge conditions. The model is based on the
stochastic-convective reactive approach and assumes a time-invariant lognormal travel-time
distribution of the streamline ensemble connecting the river and a groundwater observation
well. DO consumption, resulting from aerobic respiration, is modeled by zero-order kinetics.
The DO consumption rate depends on river temperature and discharge. While the temperature
dependence of aerobic respiration is well known, the discharge dependence is probably
related to an increased trapping of particulate organic matter (POM) within the riverbed
during high-discharge events, thus enhancing the POM availability and DO consumption rate.
We propose an empirical equation that quantifies the dependence between discharge and the
DO consumption rate, which we infer from high-resolution DO time series measured in the
Thur River (NE-Switzerland) and an adjacent observation well. The estimated
parameterization at our field site suggests that an increasing discharge within the narrow
window of 20-50 m3/s enhances the DO consumption rate by a factor of 4. By considering the
measured DO in the river and including the dependence of the DO consumption rate on both
discharge and temperature, the model was able to capture the diurnal, short-term (days to
89
weeks), and seasonal dynamics of the observed DO within the alluvial aquifer. The
temperature dependence of the DO consumption rate was found to be more important on a
seasonal time scale, while the effect of discharge dominated the DO behavior during
hydrological events extending over a few days to weeks. The presented modeling approach
can be transferred to other riverbank-filtration systems to efficiently estimate DO
concentrations in alluvial aquifers under various climatic and hydrologic conditions and,
hence, assess the risk of approaching anoxic conditions in a changing climate.
5.1 Introduction
Drinking water originating from riverbank filtration provides considerable fractions (10-50%)
of the overall drinking water production in several European countries (Tufenkji et al., 2002).
The redox conditions in the infiltration zone are mainly controlled by microbial oxidation of
natural organic matter (NOM), which leads to a consumption of terminal electron acceptors,
such as oxygen, nitrate and Mn(IV)-/Fe(III)(hydr)oxides. Redox conditions have been
observed to undergo seasonal variations due to the temperature dependence of NOM
degradation (von Gunten et al., 1991; Massmann et al., 2006).
In the Swiss lowlands, the peri-alpine alluvial shallow gravel-and-sand aquifers are mostly
oxic, which allows a minimal treatment of pumped groundwater before supplying it as
drinking water to the distribution system. Under anoxic conditions, as observed in the hot
summer of 2003 (Hoehn and Scholtis, 2011), the redox sequence proceeds to denitrification
and eventually Mn(IV)-/Fe(III)oxide-reduction, leading to the occurrence of undesired
dissolved species such as nitrite or Mn(II)/Fe(II) (Jacobs et al., 1988; Massmann et al., 2008).
Additionally, it has been shown that degradation of micropollutants is sensitive to the
presence or absence of oxygen (Greskowiak et al., 2006; Massmann et al., 2006; Maeng et al.,
2010).
As a result of climate change, river temperatures as well as the frequency and intensity of heat
waves are expected to increase, while river discharge is expected to decrease during summer
months (CH2011, 2011; FOEN, 2012a). The higher microbial activity at increased
temperatures and the longer groundwater residence times at lower discharge are supposed to
increase the frequency of anoxic conditions in riverbank-filtration systems (Sprenger et al.,
2011). To assess this risk, it is crucial to understand the dynamics of oxygen consumption
during riverbank filtration as a function of temperature and discharge. While continuous
monitoring of dissolved oxygen (DO) concentrations is routinely performed in rivers by
environmental agencies, the monitoring of DO concentrations in groundwater is often
90
restricted to sampling on a monthly or even yearly basis. A quantitative understanding of the

temperature and discharge dependence of DO consumption is therefore required, allowing for
model-based estimates of DO concentrations in groundwater from those measured in the river.
Spatially explicit numerical reactive transport models have successfully been applied to
simulate the dynamics of NOM degradation and DO consumption during river infiltration
(Matsunaga et al., 1993; Horner et al., 2007; Sharma et al., 2012). Sharma et al. (2012)
simulated the seasonal variation of groundwater DO concentrations by including a
temperature-dependent NOM degradation rate. They also showed that discharge-related
changes in the residence time caused changes in groundwater DO concentrations. However,
spatially explicit numerical reactive transport models are often very complex and site specific,
they have long run times and require a large data set for model setup and calibration.
In this study, we present a new semi-analytical model to predict DO concentrations in riverfed groundwater under various climatic and hydrologic conditions. The implementation of the
model is easier and faster and model run times are shorter compared to spatially explicit
numerical reactive transport models. The model is based on the stochastic-convective reactive
approach (Simmons, 1982; Simmons et al., 1995) and incorporates a dependence of the DO
consumption rate on river temperature and discharge. In former modeling studies, discharge
was considered to affect groundwater residence times (Sharma et al., 2012) but not the
degradation rate itself. We hypothesize that discharge fluctuations can also affect the NOM
availability and hence, the DO consumption rate.
After an introduction to the theory of stochastic-convective reactive transport and the
temperature and discharge dependence of DO consumption, we describe the model setup and
its implementation for a riverbank-filtration system in NE-Switzerland at the peri-alpine Thur
River. We then present a 9-month long high-resolution DO time series measured in the Thur
River and an adjacent groundwater observation well. Based on these data, we test the
dependence between discharge and the DO consumption rate and quantify this relationship by
an empirical equation. We incorporate the estimated discharge and a literature-based
temperature dependence of the DO consumption rate in the model formulation and simulate
the 9-month long groundwater DO time series. We present the modeling results and discuss
the model performance and the strengths and limitations of the modeling approach. Finally,
we give advice for future applications of the model.
91
5.2 Theory
5.2.1
Stochastic-convective reactive transport
The transport system between the river and a groundwater observation well is conceptualized
as ensemble of convective-reactive streamlines, each characterized by a distinct travel time.
Diffusive exchange among the streamlines is not considered and dispersive effects are treated
as a component of the randomness in the travel time probability-density function (pdf) of the
streamline ensemble. Hence, the reactive processes are separated from physical transport
(Simmons, 1982).
For each streamline, DO consumption is assumed to be solely related to aerobic respiration
and is described by zero-order kinetics:
(5.1)
which is a good approximation to Michaelis-Menten kinetics, as typical half-saturation
constants of aerobic respiration are small (0.03-0.3 mg/L according to Greskowiak et al.
and
(2006) and Schfer et al. (1998)). In Eq. (5.1),

dissolved oxygen and NOM, respectively, and
are molar concentrations of

is the effective zero-order DO
consumption rate, depending on the state of the system (temperature/discharge). If we assume

as the elemental composition of NOM and complete aerobic degradation, the
consumption rates of DO and NOM are identical.
The DO concentration
within each streamline with residence time is described by:

(5.2)
with
being the input DO concentration of the river. Fig. 5.1 shows the normalized DO
concentration distribution
times
for a set of streamlines with dimensionless travel
for three different effective consumption rates (colored dashed lines).
The DO concentration at an observation well (
) results from the stochastic-convective

over the travel-time pdf
averaging of the concentration distribution
:
(5.3)
in which
is the maximal occurring travel time in
. The bold solid colored lines in
Fig. 5.1 indicate the normalized DO concentrations
at the observation well
92
resulting from the stochastic-convective averaging over the travel-time pdf (solid black line)
for the different effective consumption rates.
Fig. 5.1. Schematic illustration of the stochastic-convective averaging of the normalized

concentration distribution
(colored dashed lines, Eq. (5.2)) over the travel-time
pdf of the streamline ensemble
(black solid line) for three different effective
consumption rates
. The colored dotted lines represent the incremental integration to each
according to the right axis label. The bold solid colored lines indicate the resulting
at the observation well after full integration to
normalized DO concentrations
(Eq. (5.3)).
The DO concentration
at a certain observation well and a certain point in time is
described by convolution of the DO concentration distribution of the streamline ensemble

with the travel-time pdf
:
(5.4)
In contrast to Eqs. (5.2) and (5.3), the DO concentrations along the individual streamlines are
affected by time-dependent input DO concentrations
consumption rates
5.2.2
and dynamic effective
Effective DO consumption rate
The degradation of NOM during riverbank filtration is performed by microbes attached to the
riverbed sediments (Pusch et al., 1998), mainly within the first meters of infiltration (Brugger
et al., 2001a). NOM consists of dissolved organic matter (DOM) and particulate organic
matter (POM). In contrast to DOM, POM is retained in the riverbed sediments by filtration.
Chapter 4 of this thesis and other studies (Brugger et al., 2001b; Sharma et al., 2012) have
highlighted that the degradation of POM accounted for most of the variability in DO
consumption, while DOM degradation did not show a significant dependence on temperature
93
and discharge. In the following, we therefore consider POM as the main electron donor for
DO consumption during riverbank filtration.
The rate of aerobic POM degradation, and hence the effective DO consumption rate
depend on temperature and POM availability (concentration and composition) (Pusch et al.,
1998). The suspended POM concentration in the river water is known to increase with
increasing discharge as a result of an increased mobilization of allochthonous (terrestriallyderived) and autochthonous (aquatically-derived) POM (Meybeck, 1982; Sugimoto et al.,
2006; Besemer et al., 2009). The increase in autochthonous POM, which is generally highly
biodegradable (Leenheer and Croue, 2003), was found to be related to the shearing off of
periphyton due to the increasing abrasive action at the riverbed (Biggs and Close, 1989;
Uehlinger, 2006; Akamatsu et al., 2011). Furthermore, it has been shown that the POM import
into gravelly riverbeds mostly occurred under high-discharge conditions, which increased the
POM concentration and the microbial respiration rate in the riverbed (Bretschko and Moser,
1993; Naegeli et al., 1995; Brunke and Gonser, 1997). Based on these observations, we
hypothesize that discharge can be used as a proxy for the POM availability in the riverbed
sediments, which in turn affects the effective DO consumption rate.
In order to separate the dependence of the effective consumption rate on temperature from
that on discharge, we conceptualize
consumption rate
as the product of a discharge-dependent
and a temperature factor
:
(5.5)
which implies that
is assumed to depend on the river temperature
at the time of infiltration
and the discharge
The dependence of the DO consumption rate on discharge is not well established in the
literature. Therefore, the assessment and quantification of this relationship constitutes an
important part of this chapter (Sections 5.3.3 and 5.4.3). On the other hand, the temperature
dependence of the POM degradation rate has been intensively studied. We apply the
expression for the temperature factor
introduced by OConnell (1990):

(5.6)
in which
and
describe the temperature dependence, while
denotes the optimal temperature for degradation.
94
is a scale parameter.
5.3 Materials and methods

5.3.1
Field site and data collection
The Niederneunforn field site is located in the Thur catchment in northeastern Switzerland
(Fig. 5.2a). Because no retention basin (e.g. a lake) is located along the course of the river, its
discharge is very dynamic, ranging from 3 to 1100 m3/s, the average discharge being 47 m3/s.
The field site was instrumented with numerous observation wells during the interdisciplinary
RECORD project (Schneider et al., 2011; Schirmer, 2013) in the context of river restoration
measures realized in the year 2002 (Fig. 5.2b). A 3D groundwater flow and transport model
was setup and calibrated against measured groundwater heads and experimentally determined
groundwater travel times (Chapter 3). The groundwater flow field for low-flow conditions
(Fig. 5.2b) indicates losing conditions along the main river channel.
Fig. 5.2. (a) Thur catchment with indicated location of the Niederneunforn field site, NESwitzerland. (b) Niederneunforn field site. The Thur River flows from right to left.
Groundwater isopotentials at low-flow conditions (equidistance 10 cm) are taken from the 3D
steady-state groundwater flow model of Chapter 3. T1 and T2 indicate the downstream and
upstream gauging stations, respectively. (c) Enlargement of the area indicated by the dashed
rectangle in (b). The black lines/arrows indicate the advective flow paths from the river to
selected observation wells according to the groundwater flow model. Continuous DO
measurements were conducted in the observation well R050.
95
To gain insight into the dynamics of DO consumption during river infiltration at the field site,
we measured DO concentrations in the observation well R050 (Fig. 5.2c) at a temporal
resolution of 15 min with an optical sensor (ROX/YSI 600OMS, YSI Inc., USA). The sensor
was located in a depth of 2.3 m, in the center of the 2 m screen of the observation well. The
groundwater table was always at least 1 m above the sensor.
DO concentrations in the river are routinely measured in Andelfingen (10 km downstream of
Niederneunforn, Fig. 5.2a) by the Federal Office for the Environment at 10-min intervals
using an optical sensor (SC100, Hach Lange GmbH, Germany). To validate the applicability
of these data to our field site in Niederneunforn, we measured river DO concentrations at the
gauging station T2 (Fig. 5.2b) with an optical sensor (LDO10115, Hach Lange GmbH,
Germany) during three control periods, two in summer and one in winter. These control
measurements of river DO concentrations revealed a good agreement between Andelfingen
and Niederneunforn, including the phase and the amplitude of the diurnal fluctuations (Fig.
5.3).
Temperature and specific electrical conductivity (EC) were measured at the gauging station
T1 (Fig. 5.2b) and in the observation well R050 at 15-min intervals (DL/N 70, STS AG,
Switzerland). River discharge was measured at the gauging station T2 (Fig. 5.2b) at 15-min
intervals by the Agency for the Environment of the Canton Thurgau. The data acquisition
covered a period of 9 months, from August 2011 to April 2012. The groundwater DO time
series contained a one-month gap in November 2011 and two smaller gaps in January and
February 2012. Outliers in the data set were removed and intervals were equalized to 15 min.
Fig. 5.3. Comparison of DO time series measured in the Thur River in Andelfingen (solid line)
and in Niederneunforn (NNF, dashed line) during a five-day control period in summer 2012.
96
5.3.2
Estimation of travel-time pdf
The stochastic-convective reactive approach separates the physical transport from the reactive
processes (Section 5.2.1). The travel-time pdf should therefore be obtained from a
conservative tracer. Vogt et al. (2010a) propose to use fluctuations in the electrical
conductivity (EC), which are transported from the river into the aquifer without retardation.
The transport of EC fluctuations from the river to an observation well can be described by
convolution of the input EC signal
with a travel-time pdf
:
(5.7)
in which the offset
accounts for an increase in EC by the mineralization of groundwater.
We parameterized the travel-time pdf

the mean
by a lognormal pdf
, characterized by
and standard deviation . We assumed that the parameters of the travel-time pdf
are time invariant, i.e. the travel-time pdf does not change over time. While discharge-related
changes in travel time have been emphasized at other riverbank-filtration systems (Sharma et
al., 2012), the assumption of a discharge-independent travel-time pdf seems to be reasonable
at our field site. Previous studies at this site have shown that water level fluctuations in the
river propagate into the aquifer quasi-instantaneously and that the change in head difference
between the river and the groundwater is relatively small during flood events (Chapter 2, 3).
We estimated the mean
and the standard deviation of the lognormal travel-time pdf at the
observation well R050 by parametric deconvolution of the measured EC input and output
signals in the river and the observation well, respectively. Deconvolution corresponds to the
minimization of the following objective function
:
(5.8)
All minimization problems described in this chapter were solved using the nonlinear leastsquare trust-region-reflective algorithm implemented as lsqnonlin function in MATLAB
(Coleman and Li, 1996). To validate the resulting lognormal travel-time pdf, we additionally
conducted a nonparametric deconvolution (Cirpka et al., 2007) of a 9-month long EC time
series (April-December 2010). In contrast to parametric deconvolution, the nonparametric
deconvolution does not assume a predefined shape of the travel-time pdf.
97
5.3.3
Discharge and temperature dependence of the DO consumption rate
To assess the dependence of the zero-order DO consumption rate

(Section 5.2.2), we estimated
on river discharge
separately for seven periods of 10-20 d with different but
relatively stable discharge conditions, by minimizing the following objective function
(5.9)
with
(5.10)
in which the temperature factor
is given in Eq. (5.6) with parameters listed in Table 5.1
(Greskowiak et al., 2006).
Table 5.1. Comparison of different parameterizations of the temperature factor (Eq. (5.6)).
Estimated
Literature
[-]
-1.5
-1.5a,b, -1.16c, -3.43d
[C-1]
0.19
0.18a,b, 0.22c, 0.19d
Topt [C]
26.3
35a,b, 33.8c, 36.9d, 437e
Greskowiak et al. (2006): used in this study, bSharma et al. (2012), cOConnell (1990),
Kirschbaum (1995), eHeitzer et al. (1991)
Based on the resulting consumption rates
of the seven periods at different discharges, we
established a quantitative and empirical relationship
, see Section 5.4.3. This
relationship was incorporated in the model formulation (Eqs. (5.4) and (5.5)) and the
corresponding parameters were estimated by fitting the 9-month long DO time series (Eq. (5.9)
combined with Eq. (5.5)).
To validate the parameterization of
we estimated
and
(according to Eq. (5.6)) from Greskowiak et al. (2006),
by fitting the DO time series of all the seven periods (Eq. (5.9)
combined with Eq. (5.10)). The estimate for

the one for
was within the range of literature values, while
was unrealistically low (Table 5.1). The parameterization of Greskowiak et al.
(2006) is in better agreement with other literature values. Furthermore, the estimated
parameterization matched the one of Greskowiak et al. (2006) in the range of daily mean river
temperatures covered by our data set (0-22C, Fig. 5.4), which validates the parameterization
98
of Greskowiak et al. (2006) for our study. This parameterization has also been successfully
applied by Sharma et al. (2012).
Fig. 5.4. Temperature factor ( ) that accounts for the temperature dependence of the DO
consumption rate (Eq. (5.6)). Solid line: parameterization of the temperature factor found by
Greskowiak et al. (2006), which was adopted for this study. Dashed line: estimated
temperature factor from our data.

5.4.1
Description of measured data
The 9-month long river discharge time series covered a range of 5-450 m3/s and was
characterized by periods of low flow (<25 m3/s) punctuated by flood events (Fig. 5.5a). The
frequency of the floods was highest during December 2011 and January 2012. During
snowmelt in March and April 2012, base flow was generally elevated to 40-50 m3/s.
The DO concentrations in the river underwent strong diurnal fluctuations of up to 6 mg/L (Fig.
5.5b), resulting from the interplay of photosynthesis and respiration by periphyton (Reichert
et al., 2009; Hayashi et al., 2012). The diurnal fluctuations revealed a repeated pattern of
steadily increasing amplitude, truncated by floods. Complete reduction of diurnal fluctuations
occurred after peak flows of >150 m3/s, while partial reductions already occurred at
discharges >30 m3/s. These findings are consistent with those of Uehlinger (2006), who
investigated the dynamics of gross primary production (GPP) in the Thur River and found that
GPP reduction, resulting from periphyton removal from the riverbed, started at 30-50 m3/s
and was most significant during bed-moving floods (>150 m3/s). Diurnal fluctuations were
not present during December and January, partly due to the high frequency of flood events,
and partly due to low temperatures and low solar radiation.
99
Fig. 5.5. (a) DO, discharge ( ) and temperature ( ) time series in the Thur River and R050
during the 9-month long measurement period. The seven periods (P1-P7) used for the
estimation of the consumption rates at different discharges are represented by gray bars. (b)
Detailed plot of diurnal DO fluctuations in the river and in groundwater as a function of
discharge for the 1-month long period delineated by the black rectangle in (a).
The diurnal DO fluctuations propagated from the river into the aquifer and could be resolved
at the observation well R050 (Fig. 5.5b). As a result of the transport processes, the
fluctuations were time shifted and had smaller amplitudes. The microbial degradation
processes caused an overall decrease in DO concentrations. The DO consumption between the
river and the observation well R050 showed a pronounced seasonal pattern, which reflects the
temperature dependence of microbial POM degradation (Fig. 5.5a). During the temperature
maximum in August 2011, groundwater DO concentrations varied between 0-4 mg/L. With
decreasing temperatures in autumn and winter, the DO concentrations in groundwater
increased up to nearly the level in the river itself, as observed during the temperature
minimum in February 2012.
100
The seasonal pattern in groundwater DO concentrations was overlain by shorter-term

variations that were related to river discharge (Fig. 5.5a, b). During flood events, the DO
concentrations in groundwater dropped steeply and recovered again when river discharge
declined. At first glance, this behavior is against intuition, as one would expect a
breakthrough of higher DO concentrations during flood events due to higher infiltration
velocities and shorter travel times. A similar pattern with decreasing DO concentrations at
high discharges was observed in the riverbed of an upland stream in Scotland (Malcolm et al.,
2006; Soulsby et al., 2009). However, at the Scotland site the drop in DO concentrations
occurred after the peak flow and was explained by an upwelling of anoxic groundwater due to
a reversal in the hydraulic gradient. This explanation does not apply to our field site, as losing
conditions prevailed permanently. Furthermore, vertical DO profiles in the observation well
R050 and adjacent wells did not reveal vertical differences in DO concentrations. This
excludes that the observed drops in DO concentrations were caused by an uplift of older
groundwater with lower DO concentrations during the rise of the groundwater table.
The conceptual explanation for the observed decrease in DO concentrations during high
discharge is based on results from previous studies, as outlined in Section 5.2.2. First,
suspended POM concentrations in rivers are known to be positively correlated with river
discharge (Meybeck, 1982). Second, the POM import into gravelly riverbed sediments was
found to occur mostly under high-discharge conditions (Naegeli et al., 1995; Brunke and
Gonser, 1997). Finally, the increased POM availability in the riverbed allowed for faster POM
degradation and faster DO consumption (Naegeli et al., 1995). Therefore, the observed
dynamics in DO consumption (Fig. 5.5) illustrates the need of accounting for dependence of
the DO consumption rate on both temperature and discharge in the model formulation.
5.4.2
Travel-time pdf
We estimated the parameters of the lognormal travel-time pdf by deconvolution of the diurnal
EC fluctuations (Section 5.3.2, Eq. (5.8)) that occurred during the low-flow conditions of
periods P1 and P3 (Fig. 5.5, Fig. 5.6). The mean
and the standard deviation
were
estimated jointly for both periods, while the offset was estimated separately. The resulting
lognormal travel-time pdf at the observation well R050, shown in Fig. 5.7 (dashed line), is
characterized by a mean of 0.4 d and a standard deviation of 0.3 d. The maximal travel time
was set to 2 d, in order to obtain a recovery rate of at least 99%. The offset amounted
to 36 S/cm for both periods. When applying the estimated lognormal travel-time pdf, the
model simulations (Eq. (5.7)) well reproduced the measured EC time series with Nash101
Sutcliffe model efficiencies (Mayer and Butler, 1993) of 0.9 and 0.8 for periods P1 and P3,
respectively (Fig. 5.6).
Fig. 5.6. Measured and simulated EC time series in groundwater at the observation well R050
for periods P1 and P3 (Fig. 5.5), using the estimated lognormal travel-time pdf (Fig. 5.7).
Fig. 5.7 additionally shows the statistical distribution of 1000 conditional realizations of the
travel-time pdf resulting from nonparametric deconvolution of a 9-month long EC time series.
Even though the 9-month long time series contained diurnal and event-driven EC fluctuations
during low- and high-flow conditions, respectively, the nonparametric travel-time pdf did not
show a shift to shorter travel times than the lognormal travel-time pdf; the modes were
identical. This supports the use of a time-invariant travel-time pdf for our system, despite the
strong discharge fluctuations. Furthermore, apart from the secondary peak, which may be an
artifact caused by diurnal periodicity in EC time series (Vogt et al., 2010a), the lognormal
travel-time pdf captured the main characteristics of the nonparametric travel-time pdf.
102
Fig. 5.7. Lognormal travel-time pdf at the observation well R050 (dashed line) resulting from
deconvolution of the EC time series of periods P1 and P3 (Fig. 5.5, Fig. 5.6) according to Eq.
(5.8). Additionally, the results from nonparametric deconvolution of a 9-month long EC time
series are plotted. Solid line: mean of 1000 conditional realizations, gray area: 16 and 84
percentiles of the conditional statistical distributions, dotted lines: minimum and maximum
values obtained in all realizations.
5.4.3
Quantifying the discharge dependence of the DO consumption rate
Fig. 5.8 plots the estimated zero-order DO consumption rates
(Section 5.3.3) for each of the
seven separate periods against the corresponding mean discharge. These estimates of the DO
consumption rates were compensated for the temperature dependence according to Eq. (5.10)
and can be interpreted as effective DO consumption rates at a temperature of 10C, at which
the temperature factor
equals one (Fig. 5.4). The consumption rates remained on a level of
7-10 mg L-1 d-1 up to discharges of 20 m3/s and steeply increased to 28 mg L-1 d-1 between
20 m3/s and 50 m3/s. These results confirm the dependence between the DO consumption rate
and discharge, and suggest that discharge is a good proxy for the POM availability in the
riverbed at our field site.
103
Fig. 5.8. Circles: estimated zero-order DO consumption rates as a function of the mean
relationship (Eq. (5.14))
discharge of each of the seven periods. Red line: estimated
with parameters estimated by fitting the complete 9-month long DO time series (Table 5.2).
The DO consumption rates in Fig. 5.8 were estimated for separate periods of 10-20 d in length
with different, but relatively stable discharge conditions (Fig. 5.5). An average estimation of
the consumption rates for discharges >50 m3/s was not feasible, as these discharges were
related to flood events of short duration with fast changes in discharge. To estimate the
consumption rate for a changing discharge within the full discharge spectrum, an empirical
expression
is required.
The concentration of suspended sediment and suspended POM in river water is usually
described as a power function of river discharge (Meybeck, 1982; Hedges et al., 1986;
Syvitski et al., 2000). If we assume that the POM import into the riverbed sediment is
proportional to the POM concentration in the river, we may describe the discharge
dependence of the substrate (POM) concentration
in the riverbed sediment by a power
function as well:
(5.11)
The relation between the degradation rate and the substrate concentration
may be
described by Michaelis-Menten kinetics:

(5.12)
104
in which
denotes the maximal degradation rate and
the half-saturation constant.
Hence, even though the POM concentration (availability) in the riverbed might steadily
increase with increasing discharge (Eq. (5.11)), we expect the DO consumption rates to level
off at some point.
If we substitute Eq. (5.11) in Eq. (5.12) and reformulate the resulting equation, we can
describe the discharge-dependent consumption rate by:
(5.13)
For an exponent
of zero for
, Eq. (5.13) describes an S-shaped curve, which increases from a level

to a level of
for
. However, the estimated DO consumption
rates (Fig. 5.8) indicate a level different from zero at low discharges (7-10 mg L-1 d-1). We
may address this by an offset at
, resulting in:
(5.14)
in which
describes the minimum, and
, respectively. The exponent

and
the maximum consumption rate, approached at
defines how fast the transition from
represents the discharge, at which the increase in
to
occurs,
is the highest. As already stated in
Section 5.2.2, it is assumed that the consumption rate depends on the discharge
at
the time of infiltration.

We incorporated the
relationship of Eq. (5.14) in the model formulation (Section 5.3.3)
and estimated the four parameters by fitting the 9-month long DO time series, which covered
a discharge range of 5-450 m3/s (Fig. 5.5). The resulting parameters are listed in Table 5.2 and
the corresponding
relationship is shown in Fig. 5.8. The resulting DO consumption rates
remained at a level of 7 mg L-1 d-1 up to a discharge of 20 m3/s, then increased steeply and
leveled off at 27 mg L-1 d-1 at a discharge of 50 m3/s.
The offset of 7 mg L-1 d-1 at low discharges may be explained by a consumption of POM that
is either stored in the riverbed or imported from autochthonous sources. The steep increase in
DO consumption rate above 20 m3/s is probably related to an increased POM concentration in
the river and the riverbed, partly due to the shearing off of periphyton (Section 5.4.1,
Uehlinger (2006)). Even though the POM concentration (availability) in the river and the
riverbed is likely to further increase at discharges >50 m3/s (Eq. (5.11)), the DO consumption
rate seems to be constrained by a microbial limitation (Eq. (5.12)).
105
Overall, the resulting
relationship indicates that a discharge increase within the narrow
window of 20-50 m3/s enhances the consumption rate at our field site by a factor of about 4. It
also shows that the effect of discharge on the DO consumption rate is already close to the
maximum at a discharge of 50 m3/s. The probability that the discharge
at the Thur River
exceeds 50 m /s is 30%. Typical situations are during snowmelt or short thunderstorms in the
Thur catchment.
Table 5.2. Estimated parameters of the
9-month long DO time series (see Fig. 5.8).
relationship (Eq. (5.14)) obtained by fitting the
Estimated parameters
-1
-1
7.5
-1
-1
A [mg L d ]
19.2
27.3
4.3
C [mg L d ]
KQ [m /s]
b [-]
The combination of the discharge dependence (Eq. (5.14), Fig. 5.8) with the temperature
dependence (Eq. (5.6), Fig. 5.4) according to Eq. (5.5) allows the visualization of the effective
DO consumption rate
for a range of river temperatures and discharges (Fig. 5.9).
Accordingly, the highest effective DO consumption rates at our field site occur at high
temperatures and at discharges >50 m3/s. The highest risk for the development of anoxic
conditions in the infiltration zone of a riverbank-filtration system is usually anticipated for
conditions of high river temperatures and low river discharges (Sprenger et al., 2011),
assuming that low river discharges increase groundwater residence times. At our site,
however, river discharge hardly affects groundwater residence times. Therefore, we anticipate
the highest risk for the development of anoxic conditions in the infiltration zone at our site
during summer heat waves combined with elevated discharges >50 m3/s.
106
Fig. 5.9. Effective DO consumption rate

(Eq. (5.5)) for a range of temperature and
(Eq. (5.14), Table 5.2) and
discharge conditions using the estimated parameter set for
the literature-based parameter set for (Eq. (5.6), Table 5.1).
5.4.4
Simulation of DO concentrations in groundwater
We applied the literature-based temperature dependence (Eq. (5.6), Fig. 5.4) and the estimated
discharge dependence (Eq. (5.14), Fig. 5.8) of the effective DO consumption rate to simulate
the 9-month long groundwater DO time series according to Eqs. (5.4) and (5.5). The
simulated DO concentrations at the observation well R050 are plotted in Fig. 5.10 together
with the measured DO concentrations at R050 and those measured in the Thur River. We
additionally plotted the discharge time series, the discharge-dependent DO consumption rate
, the river temperature time series and the effective DO consumption rate
The seasonal temperature variations led to a seasonal variation of the effective DO

consumption rate
that translated into a seasonal pattern in the simulated DO
concentrations. The seasonal trend in
was overlain by the effect of discharge. An
increase in discharge above 20 m3/s caused a fast increase in

the simulated groundwater DO concentrations. The
and an associated drop in
time series illustrates the fast
increase of the consumption rate at increasing discharges, the truncation at a maximum of

27 mg L-1 d-1 above 50 m3/s, and the decrease during the recession of the flood event. The
impact of discharge variations on
was not only restricted to short periods of a few days.
During the flood events in December 2011/January 2012 and during the snowmelt period in
March/April 2012, discharge was above 40-50 m3/s during several weeks, which caused a
persistent increase in
.
107
Fig. 5.10. Measured and simulated DO concentrations in groundwater at the observation well
R050. The simulations were performed using the estimated parameter set for
(Table 5.2)
and the literature-based parameter set for (Table 5.1). The figure additionally shows the
time series of discharge ( ), the discharge-dependent DO consumption rate (
), the
temperature ( ) in the Thur River and the effective DO consumption rate (
). The
periods P1-P7, which are enlarged in Fig. 5.11, are delineated by gray bars.
The comparison of the simulated and the measured groundwater DO time series revealed a
very good model performance, with a Nash-Sutcliffe model efficiency of 0.94. By accounting
for both the temperature dependence and the discharge dependence of the DO consumption
rate, the model captured the full dynamics in DO consumption during riverbank filtration on
different temporal scales. The simulated groundwater DO concentrations tightly followed the
measured DO concentrations across the complete temperature range (0-22C) and the
complete discharge range (5-450 m3/s). The good model performance also validates the
108
proposed expression for the empirical
relationship (Eq. (5.14)) and supports the
underlying concept of using discharge as a proxy for the POM availability.

Fig. 5.11 shows the simulation results of the periods P1-P7 to assess the model performance
on a daily to sub-daily time scale. For the periods P1-P4, the model successfully reproduced
the DO time series in groundwater with respect to both the daily mean DO concentrations and
the phase and amplitude of the diurnal fluctuations. The model also captured the depressions
in groundwater DO concentrations during P1, which were related to two minor discharge
events during this period (Fig. 5.10).
Besides an initial overestimation in P7, the model reasonably reproduced the daily mean DO
concentrations in periods P5-P7 (March/April 2012). However, the amplitudes of the diurnal
groundwater DO fluctuations were overestimated, especially in P6. This overestimation might
be attributed to a bias in the input DO concentrations in the river, which were measured in
Andelfingen, 10 km downstream of our field site (Section 5.3.1). While we demonstrated that
the river DO concentrations in Andelfingen were representative for those at our field site
during low-flow conditions (Section 5.3.1, Fig. 5.3), this may not be the case during longlasting elevated discharges of 40-50 m3/s that prevailed during the snowmelt period in
March/April 2012.
The effective DO consumption rates
ranged from 2 mg L-1 d-1 under low-flow conditions
during winter temperatures (February, P4, Fig. 5.10) to 85 mg L-1 d-1 during a minor discharge
event of 43 m3/s at high summer temperatures (August, P1, Fig. 5.10). This range is within
the spectrum of typical DO consumption rates of 0.05-96 mg L-1 d-1 (Matsunaga et al., 1993;
Sharma et al., 2012). The seasonal temperature variations caused changes in
by a factor
of 10, while variations in the discharge caused changes by a factor of 4. On the seasonal time
scale, the temperature-related changes in
for 70% of the overall variability in
discharge-related changes in
were therefore more important and accounted

. However, on the time scale of days to weeks,
dominated its variability.
109
Fig. 5.11. Measured and simulated DO concentration time series in groundwater at the
observation well R050 for the seven periods P1-P7 (Fig. 5.10).
5.4.5
Transferability of the model approach to other sites
The modeling approach proposed in this chapter may be applicable in other riverbankfiltration systems. Three main elements are required: The travel-time pdf, the temperature
dependence, and the discharge dependence of the effective zero-order DO consumption rate.
Once these elements are obtained, the model (Eqs. (5.4) and (5.5)) can be used to efficiently
predict DO concentrations in groundwater from those measured in the river under various
temperature and discharge conditions.
The travel-time pdf at an observation well may be obtained by deconvolution of a
conservative signal, e.g. an EC or chloride time series measured in the river and the
observation well. In this study, we have used a parametric deconvolution assuming a
lognormal travel-time pdf (Section 5.3.2), but other parametric models such as the gamma
distribution may be more appropriate at other sites. For our system, the assumption of a timeinvariant travel-time pdf has shown to be reasonable despite very dynamic discharge
110
fluctuations (Section 5.4.2). At other riverbank-filtration systems however, the travel-time pdf
might change as a function of discharge, which also affects predicted DO concentrations. A
definition of a lognormal travel-time pdf with discharge-dependent parameters (
, ) is
possible, but not straightforward.

To account for the temperature dependence of the effective DO consumption rate according to
Eq. (5.5), we suggest to use the formulation of the temperature factor
proposed by
OConnell (1990) and the corresponding parameter values found by Greskowiak et al. (2006).
This parameterization has been successfully applied in a previous modeling study and has
been validated in the present study (Section 5.3.3).
The presented relationship between discharge and the DO consumption rate
(Eq. (5.14))
is based on the concept of using discharge as a proxy for the POM availability in the riverbed.
However, this concept will probably not be generally applicable, as the dynamics in the POM
availability in the riverbed also depends on factors such as the catchment characteristics, the
discharge regime and the grain-size distribution of the riverbed sediments (Naegeli et al.,
1995; Brunke and Gonser, 1997; Pusch et al., 1998). Therefore, the inclusion of a dischargedependent consumption rate
in the functional form of Eq. (5.5) may not be adequate for
all riverbank-filtration systems. We suppose that an increase in the POM availability and
hence in the DO consumption rate with increasing discharge only occurs at riverbankfiltration systems with a gravelly riverbed. In coarse-sediment rivers, the import of POM to
deeper layers is possible, while in sand-bottomed or clogged rivers, the import of POM is
physically prevented (Brunke and Gonser, 1997).
The four parameters of the
relationship according to Eq. (5.14) are site specific and need
to be estimated by fitting measured DO time series in the river and in the observation well
(Section 5.3.3). The parameterization obtained for our field site (Table 5.2) might only be
applicable to riverbank filtration within the Thur Valley, which extends 30 km upstream from
our field site (Fig. 5.2).
5.5 Conclusions
We have presented a new semi-analytical model to simulate the dynamics of dissolved
oxygen (DO) consumption during riverbank filtration. The model implementation is fast and
simple and model run times are short compared to spatially explicit numerical reactive
transport models. Hence, the modeling approach provides an efficient tool to estimate DO
111
concentrations in groundwater from those measured in the river under various climatic and
hydrologic conditions.
We have developed and implemented the model at a field site in northeastern Switzerland at
the peri-alpine Thur River on the basis of a 9-month long high-resolution DO time series
measured in the river and adjacent groundwater. The model successfully reproduced the
seasonal, short-term (days to weeks) and diurnal variations in groundwater DO concentrations
by considering the measured DO concentrations in the river and accounting for both the
temperature and the discharge dependence of the zero-order DO consumption rate.
The good model performance validates the proposed empirical expression that quantifies the
relationship between discharge and the DO consumption rate (
relationship) and supports
the underlying conceptual model. According to this conceptual model, high-discharge

conditions increase the suspended POM concentration in the river. During river infiltration,
POM is retained in the riverbed sediments within the first meters of infiltration. The higher
POM import during high-discharge conditions increases the POM availability in the riverbed
sediments, which allows for a faster POM degradation and a faster DO consumption within
microbial biofilms. The estimated parameterization of the
relationship from our data set
revealed a steep increase of the DO consumption rates between 20-50 m3/s and a leveling off
beyond. As the POM availability in the riverbed is likely to further increase at
discharges >50 m3/s, the observed saturation may reflect the maximal DO consumption rate
for the existing microbial community in the riverbed sediments.
High-discharge conditions can also reduce the residence time of infiltrated river water, which
counteracts the effect of an increased POM availability on observed groundwater DO
concentrations. At our field site, however, river discharge hardly affected groundwater
residence times and therefore, the effect of a discharge-dependent DO consumption rate
dominated the observed DO concentration dynamics in groundwater during high-discharge
conditions.
Our results revealed that the combination of high river temperatures and elevated discharges
of >50 m3/s maximize the DO consumption rate and hence, the risk of anoxic conditions at
our field site. Anoxic conditions at riverbank-filtration systems are of concern because of the
appearance of undesired species as nitrite and dissolved Fe(II) and Mn(II). Furthermore, the
redox state of the infiltration zone is decisive for the degradation of emerging micropollutants.
Summer heat waves are expected to occur more frequently and more intensively in future due
to climate change, increasing the risk of anoxic conditions in the infiltration zone. Our
112
modeling approach can contribute to a better assessment of this risk at individual riverbankfiltration systems and potentially reduces expensive long-term monitoring of groundwater DO
concentrations.
Acknowledgments
This study was conducted within the National Research Program Sustainable Water
No. 406140-125856). We would like to thank Matthias Rudolf von Rohr, Michael Dring,
Chris Robinson, Andreas Raffainer, Roger Mgroz and Urs von Gunten for their support. We
are grateful to Eduard Hoehn and Peter Reichert for many helpful discussions. The Federal
Office for the Environment (FOEN) provided data of the gauging station in Andelfingen, and
the Agency for the Environment of the Canton Thurgau provided discharge data of their
gauging station in Niederneunforn. Additional support was provided by the Competence
Center Environment and Sustainability (CCES) of the ETH domain in the framework of the
projects.
113
Chapter 6
Conclusions and Outlook
6.1 Conclusions
The main objectives of this Ph.D. Thesis were to improve the understanding of physical and
biogeochemical processes that occur during riverbank filtration, and the factors controlling
them, by combining laboratory and field investigations as well as groundwater flow and
reactive transport modeling. Moreover, this work aimed at developing tools that facilitate the
model-based assessment of potential impacts of river restoration and climate change on riverrecharged groundwater quality.
In the framework of river restoration, reductions in groundwater residence times as a result of
riverbed widening and higher riverbed permeability may lead to a breakthrough of pathogenic
bacteria or pollutants at the pumping well. To anticipate the effect of river restoration
measures on residence times, a thorough understanding of the groundwater flow paths and
flow velocities is indispensable. While methods based on the analysis of natural tracers
provide direct estimates of the residence time, a calibrated numerical groundwater flow and
transport model quantitatively links groundwater flow paths, flow velocities and residence
times.
The surface water level distribution is an important prerequisite for modeling rivergroundwater systems. In this thesis, two alternative interpolation methods were developed that
are simpler and faster in their implementation compared to a hydraulic model, while still
being able to account for typical hydromorphological features of dynamic (restored or natural)
river sections, such as nonlinear longitudinal water level distributions, lateral water level
gradients, disconnected river branches and hydraulic jumps. Both interpolation methods allow
the definition of time-varying 1D or 2D surface water level distributions by combining
continuous water level records at gauging stations with periodic water level measurements at
fixpoints in between. The RM method applies a polynomial regression approach for the
prediction of water levels at fixpoints as a function of the corresponding water levels at the
determining gauging station, while the IM method uses a nonlinear interpolation approach
between two gauging stations. Compared to a third reference method (RH), which is based on
water level data from a 2D hydraulic model, the alternative methods proved to be accurate
with respect to their water level predictions. The RM method benefits from the
115
straightforward implementation, but was found to provide reliable water level predictions
only within the range of measurements made at the fixpoints. For the simulation of discharge
conditions beyond the measured range, the use of the IM method is recommended instead.
To assess the impact of the method selection and the considered level of detail in the surface
water level distribution on the simulated groundwater residence time, a 3D groundwater flow
and transport model of the partly restored Niederneunforn field site was developed. Steadystate surface water level distributions were generated with both alternative methods (RM, IM)
and the reference method (RH), as well as with two simplified versions of the IM method, and
were assigned to the 1D side channel (1st/3rd Type) and the 2D river (3rd Type) boundary
conditions of the model. Model calibration against measured groundwater heads was
performed for each of these model scenarios by an automated adjustment of selected transfer
rates (conductance of colmation layer). Based on the calibration-constrained groundwater
flow fields, groundwater age (groundwater residence time) was simulated according to the
method of Goode (1996). This method is very convenient, as it provides a spatial distribution
of groundwater age and accounts for advective and dispersive age fluxes. Goodes method
was successfully applied in this thesis and its application to other riverbank-filtration systems
is recommended when simulating groundwater residence times.
To initially obtain a realistic groundwater residence time distribution at the field site, the
transfer rates and the hydraulic conductivity distribution were jointly estimated by fitting both,
measured groundwater heads and experimentally determined residence times from
nonparametric deconvolution of EC time series. This procedure is highly recommended, as
calibration against groundwater heads alone provides non-unique information on flow
velocities and residence times.
The age predictions of the calibration-constrained groundwater flow fields using both
alternative methods lay within a range of 30% compared to the reference RH model scenario.
This error is small compared to the uncertainty of experimentally determined residence times
of 60-80%, which confirms the predictive capability of the alternative methods when applied
to real and complex river-groundwater systems. The simplification of the river water level
distribution by neglecting lateral water level differences of 20-30 cm in the central part of the
river caused moderate errors in groundwater age prediction of 40-80%. The additional
simplification by assuming a linear longitudinal river water level distribution led to deviations
in water levels of up to 50 cm and induced widespread errors in groundwater age of 200500%.
116
Even though these specific errors in groundwater age prediction only apply to our field site,
they emphasize the importance of an accurate (especially longitudinal) surface water level
distribution for a reliable simulation of the groundwater flow field and groundwater residence
times. To this end, the application of both alternative interpolation methods presented herein
can be recommended, and may help to reliably assess the impact of river restoration measures
on groundwater residence times and hence, to mitigate the conflict of interest between river
restoration and drinking water protection.
In the context of climate change, the development of anoxic conditions in the infiltration zone
of riverbank-filtration systems is of concern due to the mobilization of nitrite and especially
Mn(II) and Fe(II), which would require additional treatment of the abstracted water. Climate
models predict an increased frequency and intensity of summer heat waves, characterized by
long-lasting periods of low discharges and high temperatures, which likely increase the risk
for riverbank-filtration systems to become anoxic or even Mn(IV)- or Fe(III)-reducing. In this
thesis, periodic and targeted (summer and winter) field sampling campaigns and column
experiments simulating the field conditions were performed to improve the understanding of
the dependence of dissolved oxygen (DO) and dissolved organic matter (DOM) consumption
during riverbank filtration on the climate-related variables temperature and discharge.
The column experiments consistently reproduced the observations of the summer and winter
field sampling campaigns. In both systems, DO consumption was found to strongly depend on
temperature, while DOM consumption was essentially the same under summer and winter
conditions. In summer, DOM consumption only explained 10-20% of DO consumption. The
remaining 80-90% of the reduction capacity to explain the DO consumption is likely provided
by particulate organic matter (POM) associated with the riverbed sediments.
These findings were confirmed by the results of 45 periodic field samplings that covered a
wide temperature and discharge range over a period of five years. DO consumption was
highly correlated with temperature, while the DOM consumption did not reveal a significant
temperature dependence. Therefore, the variability and temperature dependence of DO
consumption seemed to be explained by POM degradation to a large extent. The divergence
between the temperature dependencies observed for DO and DOM consumption also have the
consequence that POM is the most important electron donor under summer conditions. The
importance of POM degradation for explaining the seasonal variability of the DO
consumption during riverbank filtration has also been described in previous studies and
presumably applies to most riverbank-filtration systems.
117
The periodic samplings also allowed for assessing the effect of discharge conditions on DO
and DOM consumption. The general concept that low-flow conditions enhance DO and DOM
consumption as a result of increased groundwater residence times and higher loads of DOM
in the river due to less dilution of wastewater effluent is not supported by the results of this
thesis. DO consumption increased rather than decreased with increasing discharge and DOM
consumption did not reveal any significant correlation with discharge. It is hypothesized that
the increased DO consumption at higher discharges is related to an increased import of POM
into the riverbed. Furthermore, the DOC concentrations in the river increased rather than
decreased with discharge, suggesting DOM sources other than wastewater treatment plants.
During summer conditions (T>20C), groundwater DO concentrations at the field site were
close to zero, but denitrification was not observed. Similarly, most peri-alpine aquifers fed by
rivers are (sub)oxic under todays summer conditions. Therefore, currently there is no need to
implement additional water treatment steps such as demanganation and deferisation. However,
a substantial increase in river water temperature during future heat waves will enhance the
POM turnover, leading to a complete consumption of DO and potentially nitrate. As a
consequence, nitrite and possibly Mn(II) and Fe(II) could be released. In order to take
adequate measures in time, the long-term monitoring of DO and other redox parameters in
groundwater is recommended.
However, while continuous monitoring of DO concentrations is routinely performed in rivers
by environmental agencies, the monitoring of DO concentrations in groundwater is often
restricted to sampling on a monthly or even yearly basis. In this thesis, a new semi-analytical
model was developed that allows efficiently estimating DO concentrations in river-fed
groundwater from measured DO concentrations in the river under various temperature and
discharge conditions on a weekly, daily or even sub-daily time scale.
The model is based on the stochastic-convective reactive approach. The transport system
between the river and a groundwater observation well is conceptualized as ensemble of
convective-reactive streamlines, whose travel times are assumed to be lognormally distributed
and time invariant. The lognormal travel time probability-density function (pdf) for a selected
observation well was obtained from parametric deconvolution of diurnal EC fluctuations
measured in the Thur River and the observation well. For the Niederneunforn field site, the
assumption of a time-invariant travel-time pdf has shown to be reasonable despite very
dynamic discharge fluctuations. At other riverbank-filtration systems however, extending the
model by a non-stationary formulation of the travel-time pdf might be required.
118
DO consumption during riverbank filtration is simulated by zero-order kinetics and the

consumption rate depends on river temperature and discharge. The temperature dependence of
aerobic respiration is well known. A literature-based expression was adopted and the
corresponding parameters were successfully validated for the Niederneunforn field site. The
same parameterization has also been validated in a previous modeling study at a different site
(Sharma et al., 2012) and is presumably applicable to account for the temperature dependence
of aerobic respiration at most riverbank-filtration systems.
The discharge dependence of the DO consumption rate is probably related to an increased
trapping of POM within the riverbed during high-discharge events, thus enhancing the POM
availability and DO consumption rate. An empirical equation that quantifies the discharge
dependence was proposed (
relationship). The inferred parameterization from high-
resolution DO time series measured in the river and the observation well suggests an increase
of the DO consumption rate by a factor of 4 within the narrow window of 20-50 m3/s and a
leveling off beyond. This finding agrees well with the observations of the periodic samplings;
DO consumption increased up to a discharge of 60 m3/s and leveled off beyond. Therefore,
the highest risk for the development of anoxic conditions in the infiltration zone at the
Niederneunforn field site is anticipated during summer heat waves combined with elevated
discharges >50 m3/s. Such discharges typically result from short thunderstorms in the Thur
catchment.
By considering the measured DO concentrations in the river and accounting for both the
temperature and the discharge dependence of the zero-order DO consumption rate, the model
successfully reproduced the seasonal, short-term (days to weeks) and diurnal variations in
groundwater DO concentrations with a Nash-Sutcliffe model efficiency of 0.94. The good
relationship and the underlying concept of
model performance validates the proposed
using discharge as a proxy for the POM availability in the riverbed. However, this concept is
probably not generally applicable and the inclusion of a discharge-dependent DO
consumption rate in the model formulation might only be adequate for riverbank-filtration
systems with a gravelly riverbed and within a catchment without retention basins (e.g. a lake).
Rivers in catchments without retention basins typically have a dynamic discharge regime with
strong, rapid and frequent discharge fluctuations, which favors the mobilization of POM. In
gravelly riverbeds, the import of POM during high-discharge conditions is possible to deeper
layers, while in sandy or clogged riverbeds, the import of POM might be physically hindered.
119
The presented modeling approach can be transferred to other riverbank-filtration systems. The
literature-based temperature dependence of the DO consumption rate seems to be generally
applicable and may be directly adopted in future model applications. On the other hand, the
travel-time pdf and the discharge dependence of the DO consumption rate are site specific and
need to be estimated case by case. Once these elements are obtained, the model can be used to
efficiently estimate groundwater DO concentrations at riverbank-filtration systems under
various climatic and hydrologic conditions and, hence, assess the risk of approaching anoxic
conditions in a changing climate.
6.2 Outlook
The effectiveness of riverbank filtration depends on the interplay of physical and
biogeochemical processes, both being subject to three-dimensional heterogeneity and
temporal variability. This Ph.D. Thesis is one step towards a better understanding of these
processes and their dependencies, and provides methods and models that may facilitate the
characterization and risk assessment of riverbank-filtration systems within the context of a
changing and challenging environment. However, the investigations and conclusions of this
Ph.D. Thesis led to more questions and the developed tools may require further refinement
and validation. Potential research questions and recommendations that may be applicable to
future research projects are outlined below.
River restoration. The effect of river restoration measures on river-fed groundwater quality is
not well understood yet. For instance, higher riverbed permeability is likely to reduce the
residence time within the riverbed, which may lower the retention capacity for bacteria and
pollutants. Conversely, an increased hyporheic exchange is considered to promote interstitial
microbial activity, which in turn enhances the removal of organic matter and pollutants.
Furthermore, the higher infiltration rate and the higher oxygen flux may help to maintain oxic
conditions in the infiltration zone, thus being favorable for river-recharged groundwater
quality as well. What is the relative importance of these contrasting effects? Does the
enhanced microbial activity compensate or even dominate the shorter residence times with
respect to the removal efficiency? Answering these questions would help to further mitigate
the conflict of interest between river restoration and drinking water protection. Thereto,
research projects that investigate hydraulic, microbiologic and biogeochemical aspects of
river-groundwater systems before and after river restoration are highly recommended.
POM. In this thesis, the importance of POM as an electron donor under summer conditions is
emphasized (Chapter 4). It has been shown that POM accounted for most DO consumption
120
under summer conditions and hence, will probably play an important role in the development
of anoxic or even Mn(IV)-/Fe(III)-reducing conditions in a changing climate. Furthermore,
DO consumption was found to increase with increasing discharge, which was attributed to an
enhanced import of POM into the riverbed. The importance of POM as electron donor has
been recognized in previous studies, but the factors controlling POM dynamics in riverbed
sediments are highly variable and not well understood yet (Naegeli et al., 1995; Brunke and
Gonser, 1997). Hence, it might be worthwhile to further investigate the source, composition
and bioavailability of POM in river systems, as well as its retention, storage and consumption
within the riverbed. A better assessment of the POM dynamics in riverbed sediments is a
complex task due to the riverbed heterogeneity and variability, but may help to identify
potential intervention strategies that allow reducing the POM availability and hence, the risk
of anoxic and Mn(IV)-/Fe(III)-reducing conditions during future heat waves.
Further model application and validation. The presented alternative interpolation methods
(Chapter 2) and the developed stochastic-convective reactive modeling approach (Chapter 5)
were successfully but exclusively applied at the Niederneunforn field site. Using these
tools at other riverbank-filtration systems would help to validate their applicability and
versatility.
Interpolation methods. In this thesis, the application of the interpolation methods for
simulating groundwater flow and transport at the field site was restricted to steady state
(Chapter 3). However, the new interpolation methods are capable of generating time-varying
1D and 2D surface water level distributions (Chapter 2) and may be used for transient model
simulations to assess the impact of water level fluctuations on key predictions at riverbankfiltration systems such as groundwater flow paths, flow velocities and residence times. In this
context, it might be crucial to consider not only the change in water levels, but also in the
lateral extent of the river. In their current implementation, the interpolation methods assume a
constant river extent. Nevertheless, depending on the groundwater modeling code, accounting
for a changing river width with changing river water levels is possible (e.g. in FEFLOW using
the Interface Manager), but may require a considerable amount of work.
Non-stationary formulation of travel-time pdf. The current implementation of the stochasticconvective reactive model (Chapter 5) assumes a time-invariant travel-time pdf, implying that
discharge does not affect groundwater residence times. While this assumption has shown to
be reasonable at the Niederneunforn field site, other studies demonstrated a significant effect
of discharge fluctuations on groundwater residence times (Sharma et al., 2012). To maximize
121
the applicability of the modeling approach, the inclusion of a non-stationary travel-time pdf in
the model formulation is recommended. Relating the parameters of travel-time pdfs (e.g. the
mean and the standard deviation of the lognormal distribution) to discharge is possible, but
not straightforward and will require considerable effort.
Climate change scenarios. The stochastic-convective reactive modeling approach
incorporates a quantitative description of DO consumption during riverbank filtration as a
function of the climate-related variables temperature and discharge, and has the particular
strength of very short model run times. Therefore, the modeling approach is well suited for
long-term simulations of groundwater DO concentrations in river-recharged groundwater.
Swiss climate models provide projections of temperature and precipitation for 30-year periods
centered around 2035, 2060 and 2085 with a daily resolution, which can be transferred to a
specific site either by the delta change method or statistical downscaling (CH2011, 2011).
These climate projections at least for temperature can readily be applied as input to the
stochastic-convective model, which would provide valuable information on the frequency and
intensity (duration) of anoxic conditions at riverbank-filtration systems. However, it is
recommended to extend the model formulation in order to include both DO and nitrate
consumption during riverbank filtration. The inclusion of nitrate as a second electron acceptor
in the model can be achieved with a minor programming effort and would allow for
quantitatively determining the risk of Mn(IV)-/Fe(III)-reducing conditions in the infiltration
zone of riverbank-filtration systems.
122
Bibliography
Akamatsu, F., Kobayashi, S., Amano, K., Nakanishi, S., Oshima, Y., 2011. Longitudinal and
seasonal changes in the origin and quality of transported particulate organic matter
along a gravel-bed river. Hydrobiologia 669 (1), 183-197.
Anderson, M.P., 2005. Heat as a ground water tracer. Ground Water 43 (6), 951-968.
Baumann, M., Jordan, P., Hoehn, E., Geisser, H., 2009. Ein neues Grundwassermodell fr das
Thurtal. Band 63 der Mitteilungen der Thurgauischen Naturforschenden Gesellschaft.
Thurgauische Naturforschende Gesellschaft, Frauenfeld, 240 pp.
Besemer, K., Luef, B., Preiner, S., Eichberger, B., Agis, M., Peduzzi, P., 2009. Sources and
composition of organic matter for bacterial growth in a large European river
floodplain system (Danube, Austria). Org. Geochem. 40 (3), 321-331.
Biggs, B.J.F., Close, M.E., 1989. Periphyton biomass dynamics in gravel bed rivers: the
relative effects of flows and nutrients. Freshwater Biol. 22 (2), 209-231.
Boulton, A.J., 2007. Hyporheic rehabilitation in rivers: restoring vertical connectivity.
Freshwater Biol. 52 (4), 632-650.
Bourg, A.C.M., Bertin, C., 1993. Biogeochemical Processes during the infiltration of River
Water into an Alluvial Aquifer. Environ. Sci. Technol. 27 (4), 661-666.
Bretschko, G., Moser, H., 1993. Transport and retention of matter in riparian ecotones.
Hydrobiologia 251 (1-3), 95-101.
Brugger, A., Reitner, B., Kolar, I., Queric, N., Herndl, G.J., 2001a. Seasonal and spatial
distribution of dissolved and particulate organic carbon and bacteria in the bank of an
impounding reservoir on the Enns River, Austria. Freshwater Biol. 46 (8), 997-1016.
Brugger, A., Wett, B., Kolar, I., Reitner, B., Herndl, G.J., 2001b. Immobilization and bacterial
utilization of dissolved organic carbon entering the riparian zone of the alpine Enns
River, Austria. Aquat. Microb. Ecol. 24 (2), 129-142.
Brunke, M., 1999. Colmation and depth filtration within streambeds: Retention of particles in
hyporheic interstices. Int. Rev. Hydrobiol. 84 (2), 99-117.
Brunke, M., Gonser, T., 1997. The ecological significance of exchange processes between
rivers and groundwater. Freshwater Biol. 37 (1), 1-33.
Brunner, P., Cook, P.G., Simmons, C.T., 2009. Hydrogeologic controls on disconnection
between surface water and groundwater. Water Resour. Res. 45, 13.
BUWAL, 2004. Groundwater protection guide. Bundesamt fr Umwelt, Wald und Landschaft,
Bern, pp. 141.
Cardenas, M.B., 2009. Stream-aquifer interactions and hyporheic exchange in gaining and
losing sinuous streams. Water Resour. Res. 45, 1-13.
123
Cardenas, M.B., Wilson, J.L., Zlotnik, V.A., 2004. Impact of heterogeneity, bed forms, and
stream curvature on subchannel hyporheic exchange. Water Resour. Res. 40 (8), 1-13.
CH2011, 2011. Swiss Climate Change Scenarios CH2011. C2SM, MeteoSwiss, ETH, NCCR
Climate and OcCC, Zurich, pp. 88.
Cirpka, O.A., Fienen, M.N., Hofer, M., Hoehn, E., Tessarini, A., Kipfer, R., Kitanidis, P.K.,
2007. Analyzing bank filtration by deconvoluting time series of electric conductivity.
Ground Water 45 (3), 318-328.
Coleman, T.F., Li, Y., 1996. An Interior, Trust Region Approach for Nonlinear Minimization
Subject to Bounds. SIAM J. Optimiz. 6, 418-445.
Derx, J., Blaschke, A.P., Blschl, G., 2010. Three-dimensional flow patterns at the riveraquifer interface -- a case study at the Danube. Adv. Water Resour. 33 (11), 13751387.
Diem, S., Vogt, T., Hoehn, E., 2010. Spatial characterization of hydraulic conductivity in
alluvial gravel-and-sand aquifers: a comparison of methods. Grundwasser 15 (4), 241251.
Doetsch, J., Linde, N., Vogt, T., Binley, A., Green, A.G., 2012. Imaging and quantifying salttracer transport in a riparian groundwater system by means of 3D ERT monitoring.
Geophysics 77 (5), B207-B218.
Doherty, J., 2005. Pest. Model-independant parameter estimation. Watermark Numerical
Computing.
Doppler, T., Franssen, H.J.H., Kaiser, H.P., Kuhlman, U., Stauffer, F., 2007. Field evidence
of a dynamic leakage coefficient for modelling river-aquifer interactions. J. Hydrol.
347 (1-2), 177-187.
DVGW, 2000. Richtlinien fr Trinkwasserschutzgebiete. Deutsche Vereinigung des Gas- und
Wasserfachs, DVGW, Bonn.
Eckert, P., Irmscher, R., 2006. Over 130 years of experience with Riverbank filtration in
Dusseldorf, Germany. J. Water Supply Res. T. 55 (4), 283-291.
Eckert, P., Lamberts, R., Wagner, C., 2008. The impact of climate change on drinking water
supply by riverbank filtration. Water Sci. Technol.: Water Supply 8 (3), 319-324.
Egli, T., 1995. The ecological and physiological significance of the growth of heterotrophic
microorganisms with mixtures of substrates. In: Jones, G.N. (Ed.), Advances in
Microbial Ecology. Plenum Press, New York, pp. 305-386.
Engeler, I., Hendricks Franssen, H.J., Mller, R., Stauffer, F., 2011. The importance of
coupled modelling of variably saturated groundwater flow-heat transport for assessing
river-aquifer interactions. J. Hydrol. 397 (3-4), 295-305.
European-Commission, 2000. Directive 2000/60/EC of the European Parliament and of the
Council establishing a framework for Community action in the field of water policy.
European Commission, Brussels, pp. 72.
124
FOEN, 2012a. Effects of climate change on water resources and watercourses. Synthesis
report on the Climate Change and Hydrology in Switzerland (CCHydro) project.
Federal Office for the Environment FOEN, Bern, pp. 76.
FOEN, 2012b. Groundwater protection zones in unconsolidated rocks. A module for the
enforcement guide groundwater protection. Federal Office for the Environment FOEN,
Bern, pp. 58.
FOEN, 2012c. Revitalization of water courses. Strategic planning. Federal Office for the
Environment, FOEN, Bern, pp. 42.
Froyland, L., 2011. Characterizing the dynamics of groundwater flow in a losing stream.
Master Thesis. Norwegian University of Science and Technology, Trondheim, pp. 58.
Gelhar, L.W., Welty, C., Rehfeldt, K.R., 1992. A critical review of data on field-scale
dispersion in aquifers. Water Resour. Res. 28 (7), 1955-1974.
Goode, D.J., 1996. Direct simulation of groundwater age. Water Resour. Res. 32 (2), 289-296.
Green, T.R., Taniguchi, M., Kooi, H., Gurdak, J.J., Allen, D.M., Hiscock, K.M., Treidel, H.,
Aureli, A., 2011. Beneath the surface of global change: Impacts of climate change on
groundwater. J. Hydrol. 405 (3-4), 532-560.
Greskowiak, J., Prommer, H., Massmann, G., Ntzmann, G., 2006. Modeling seasonal redox
dynamics and the corresponding fate of the pharmaceutical residue phenazone during
artificial recharge of groundwater. Environ. Sci. Technol. 40 (21), 6615-6621.
Grnheid, S., Amy, G., Jekel, M., 2005. Removal of bulk dissolved organic carbon (DOC)
and trace organic compounds by bank filtration and artificial recharge. Water Res. 39
(14), 3219-3228.
GSchG, 2011. Bundesgesetz ber den Schutz der Gewsser, 814.20. Die Bundesversammlung
der Schweizerischen Eidgenossenschaft, Bern, pp. 32.
GSchV, 2011. Gewsserschutzverordnung, 814.201. Der Schweizerische Bundesrat, Bern, pp.
68.
Harvey, J.W., Bencala, K.E., 1993. The effect of streambed topography on surface-subsurface
water exchange in mountain catchments. Water Resour. Res. 29 (1), 89-98.
Hayashi, M., Vogt, T., Machler, L., Schirmer, M., 2012. Diurnal fluctuations of electrical
conductivity in a pre-alpine river: Effects of photosynthesis and groundwater
exchange. J. Hydrol. 450, 93-104.
Hedges, J.I., Clark, W.A., Quay, P.D., Richey, J.E., Devol, A.H., Santos, U.D., 1986.
Compositions and fluxes of particulate organic material in the amazon river. Limnol.
Oceanogr. 31 (4), 717-738.
Heitzer, A., Kohler, H.P.E., Reichert, P., Hamer, G., 1991. Utility of Phenomenological
Models for Describing Temperature Dependence of Bacterial Growth. Appl. Environ.
Microbiol. 57 (9), 2656-2665.
125
Henze, M., Gujer, W., Mino, T., Matsuo, T., Wentzel, M.C., Marais, G.V.R., Van Loosdrecht,
M.C.M., 1999. Activated Sludge Model No.2d, ASM2d. Water Sci. Technol. 39 (1),
165-182.
Hiscock, K.M., Grischek, T., 2002. Attenuation of groundwater pollution by bank filtration. J.
Hydrol. 266 (3-4), 139-144.
Hoehn, E., Cirpka, O.A., 2006. Assessing residence times of hyporheic ground water in two
alluvial flood plains of the Southern Alps using water temperature and tracers. Hydrol.
Earth Syst. Sc. 10 (4), 553-563.
Hoehn, E., Meylan, B., 2009. Measures to protect drinking-water wells near rivers from
hydraulic engineering operations in peri-alpine flood-plains. Grundwasser 14 (4), 255263.
Hoehn, E., Scholtis, A., 2011. Exchange between a river and groundwater, assessed with
hydrochemical data. Hydrol. Earth Syst. Sc. 15 (3), 983-988.
Horner, C., Holzbecher, E., Ntzmann, G., 2007. A coupled transport and reaction model for
long column experiments simulating bank filtration. Hydrol. Process. 21 (8), 10151025.
Huggenberger, P., Hoehn, E., Beschta, R., Woessner, W., 1998. Abiotic aspects of channels
and floodplains in riparian ecology. Freshwater Biol. 40 (3), 407-425.
IPCC, 2007. Fourth Assessment Report: Climate Change 2007: Impacts, Adaptation and
Vulnerability. Summary for Policy-Makers. Intergovernmental Panel on Climate
Change, IPCC, New York.
Jacobs, L.A., Von Gunten, H.R., Keil, R., Kuslys, M., 1988. Geochemical changes along a
river-groundwater infiltration flow path: Glattfelden, Switzerland. Geochim.
Cosmochim. Ac. 52 (11), 2693-2706.
Kalbus, E., Reinstorf, F., Schirmer, M., 2006. Measuring methods for groundwater - Surface
water interactions: A review. Hydrol. Earth Syst. Sc. 10 (6), 873-887.
Kser, D.H., Binley, A., Heathwaite, A.L., 2013. On the importance of considering channel
microforms in groundwater models of hyporheic exchange. River Res. Appl. 29 (4),
528-535.
Kser, D.H., Binley, A., Heathwaite, A.L., Krause, S., 2009. Spatio-temporal variations of
hyporheic flow in a riffle-step-pool sequence. Hydrol. Process. 23 (15), 2138-2149.
Kirschbaum, M.U.F., 1995. The temperature dependence of soil organic matter
decomposition, and the effect of global warming on soil organic C storage. Soil Biol.
Biochem. 27 (6), 753-760.
Kuehn, W., Mueller, U., 2000. Riverbank filtration - An overview. J. Am. Water Works Ass.
92 (12), 60-69.
Lautz, L.K., Siegel, D.I., 2006. Modeling surface and ground water mixing in the hyporheic
zone using MODFLOW and MT3D. Adv. Water Resour. 29 (11), 1618-1633.
126
Leenheer, J.A., Croue, J.P., 2003. Characterizing aquatic dissolved organic matter. Environ.
Sci. Technol. 37 (1), 18A-26A.
Maeng, S.K., Ameda, E., Sharma, S.K., Grtzmacher, G., Amy, G.L., 2010. Organic
micropollutant removal from wastewater effluent-impacted drinking water sources
during bank filtration and artificial recharge. Water Res. 44 (14), 4003-4014.
Malcolm, I.A., Soulsby, C., Youngson, A.F., 2006. High-frequency logging technologies
reveal state-dependent hyporheic process dynamics: implications for hydroecological
studies. Hydrol. Process. 20 (3), 615-622.
Massmann, G., Greskowiak, J., Dnnbier, U., Zuehlke, S., Knappe, A., Pekdeger, A., 2006.
The impact of variable temperatures on the redox conditions and the behaviour of
pharmaceutical residues during artificial recharge. J. Hydrol. 328 (1-2), 141-156.
Massmann, G., Nogeitzig, A., Taute, T., Pekdeger, A., 2008. Seasonal and spatial distribution
of redox zones during lake bank filtration in Berlin, Germany. Environ. Geol. 54 (1),
53-65.
Matsunaga, T., Karametaxas, G., von Gunten, H.R., Lichtner, P.C., 1993. Redox chemistry of
iron and manganese minerals in river-recharged aquifers: A model interpretation of a
column experiment. Geochim. Cosmochim. Ac. 57 (8), 1691-1704.
Mayer, D.G., Butler, D.G., 1993. Statistical validation. Ecol. Model. 68 (1-2), 21-32.
Mettler, S., Abdelmoula, M., Hoehn, E., Schoenenberger, R., Weidler, P., von Gunten, U.,
2001. Characterization of iron and manganese precipitates from an in situ ground
water treatment plant. Ground Water 39 (6), 921-930.
Meybeck, M., 1982. Carbon, nitrogen, and phosphorus transport by world rivers. Am. J. Sci.
282 (4), 401-450.
Mouchet, P., 1992. From conventional to biological removal of iron and manganese in France.
J. Am. Water Works Ass. 84 (4), 158-167.
Murdoch, P.S., Baron, J.S., Miller, T.L., 2000. Potential effects of climate change on surfacewater quality in North America. J. Am. Water Resour. As. 36 (2), 347-366.
Naegeli, M.W., Hartmann, U., Meyer, E.I., Uehlinger, U., 1995. POM-dynamics and
community respiration in the sediments of a floodprone prealpine river (Necker,
Switzerland). Arch. Hydrobiol. 133 (3), 339-347.
O'Connell, A.M., 1990. Microbial decomposition (respiration) of litter in eucalypt forests of
south-western Australia: an empirical-model based on laboratory incubations. Soil
Biol. Biochem. 22 (2), 153-160.
OcCC, 2005. Heat Summer 2003. OcCC/ProClim, Bern, pp. 28.
Park, J.H., Duan, L., Kim, B., Mitchell, M.J., Shibata, H., 2010. Potential effects of climate
change and variability on watershed biogeochemical processes and water quality in
Northeast Asia. Environ. Int. 36 (2), 212-225.
127
Pasquale, N., Perona, P., Schneider, P., Shrestha, J., Wombacher, A., Burlando, P., 2011.
Modern comprehensive approach to monitor the morphodynamic evolution of a
restored river corridor. Hydrol. Earth Syst. Sc. 15 (4), 1197-1212.
Peter, S., Koetzsch, S., Traber, J., Bernasconi, S.M., Wehrli, B., Durisch-Kaiser, E., 2012.
Intensified organic carbon dynamics in the ground water of a restored riparian zone.
Freshwater Biol. 57 (8), 1603-1616.
Pusch, M., Fiebig, D., Brettar, I., Eisenmann, H., Ellis, B.K., Kaplan, L.A., Lock, M.A.,
Naegeli, M.W., Traunspurger, W., 1998. The role of micro-organisms in the
ecological connectivity of running waters. Freshwater Biol. 40 (3), 453-495.
Reichert, P., Uehlinger, U., Acuna, V., 2009. Estimating stream metabolism from oxygen
concentrations: Effect of spatial heterogeneity. J. Geophys. Res.-Biogeosci. 114, 15.
Sacher, F., Brauch, H.J., 2002. Experiences on the fate of organic micropollutants during
riverbank filtration. In: Ray, C. (Ed.), Riverbank filtration: understanding contaminant
biogeochemistry and pathogen removal. Kluwer, Dordrecht, pp. 135-151.
Schfer, D., Schfer, W., Kinzelbach, W., 1998. Simulation of reactive processes related to
biodegradation in aquifers - 2. Model application to a column study on organic carbon
degradation. J. Contam. Hydrol. 31 (1-2), 187-209.
Schppi, B., Perona, P., Schneider, P., Burlando, P., 2010. Integrating river cross section
measurements with digital terrain models for improved flow modelling applications.
Computers and Geosciences 36 (6), 707-716.
Schirmer, M., 2013. Das RECORD-Projekt - Flussrevitalisierung, eine kologische
Massnahme in einem komplexen Umfeld. Aqua und Gas 3, 22-28.
Schneider, P., Vogt, T., Schirmer, M., Doetsch, J., Linde, N., Pasquale, N., Perona, P., Cirpka,
O.A., 2011. Towards improved instrumentation for assessing river-groundwater
interactions in a restored river corridor. Hydrol. Earth Syst. Sc. 15 (8), 2531-2549.
Schubert, J., 2002. Hydraulic aspects of riverbank filtration - Field studies. J. Hydrol. 266 (34), 145-161.
Schwarzenbach, R.P., Giger, W., Hoehn, E., Schneider, J.K., 1983. Behavior of organiccompounds during infiltration of river water to groundwater - field studies. Environ.
Sci. Technol. 17 (8), 472-479.
Senhorst, H.A.J., Zwolsman, J.J.G., 2005. Climate change and effects on water quality: a first
impression. Water Sci. Technol. 51 (5), 53-59.
Sharma, L., Greskowiak, J., Ray, C., Eckert, P., Prommer, H., 2012. Elucidating temperature
effects on seasonal variations of biogeochemical turnover rates during riverbank
filtration. J. Hydrol. 428, 104-115.
Simmons, C.S., 1982. A stochastic-convective transport representation of dispersion in onedimensional porous-media systems. Water Resour. Res. 18 (4), 1193-1214.
Simmons, C.S., Ginn, T.R., Wood, B.D., 1995. Stochastic-convective transport with nonlinear
reaction: Mathematical framework. Water Resour. Res. 31 (11), 2675-2688.
128
Sobczak, W.V., Findlay, S., 2002. Variation in bioavailability of dissolved organic carbon
among stream hyporheic flowpaths. Ecology 83 (11), 3194-3209.
Sontheimer, H., 1980. Experience with riverbank filtration along the rhine river. J. Am. Water
Works Ass. 72 (7), 386-390.
Soulsby, C., Malcolm, I.A., Tetzlaff, D., Youngson, A.F., 2009. Seasonal and inter-annual
variability in hyporheic water quality revealed by continuous monitoring in a

salmon spawning stream. River Res. Appl. 25 (10), 1304-1319.
Sprenger, C., Lorenzen, G., Hulshoff, I., Grutzmacher, G., Ronghang, M., Pekdeger, A., 2011.
Vulnerability of bank filtration systems to climate change. Sci. Total Environ. 409 (4),
655-663.
Storey, R.G., Howard, K.W.F., Williams, D.D., 2003. Factors controlling riffle-scale
hyporheic exchange flows and their seasonal changes in a gaining stream: A threedimensional groundwater flow model. Water Resour. Res. 39 (2), SBH8-1 - SBH8-17.
Sugimoto, R., Kasai, A., Yamao, S., Fujiwara, T., Kimura, T., 2006. Short-term variation in
behavior of allochthonous particulate organic matter accompanying changes of river
discharge in Ise Bay, Japan. Estuar. Coast. Shelf Sci. 66 (1-2), 267-279.
Syvitski, J.P., Morehead, M.D., Bahr, D.B., Mulder, T., 2000. Estimating fluvial sediment
transport: The rating parameters. Water Resour. Res. 36 (9), 2747-2760.
Toride, N., Leij, F.J., van Genuchten, M.T., 1995. The CXTFIT Code for Estimating
Transport Parameters from Laboratory or Field Tracer Experiments, Version 2.0.
Research Report No. 137, US Salinity Laboratory, Riverside, CA, pp. 131.
Tufenkji, N., Ryan, J.N., Elimelech, M., 2002. The promise of bank filtration. Environ. Sci.
Technol. 36 (21), 422A-428A.
Uehlinger, U., 2006. Annual cycle and inter-annual variability of gross primary production
and ecosystem respiration in a floodprone river during a 15-year period. Freshwater
Biol. 51 (5), 938-950.
Valentini, A., Garuti, G., Rozzi, A., Tilche, A., 1997. Anaerobic degradation kinetics of
particulate organic matter: A new approach. Water Sci. Technol. 36 (6-7), 239-246.
Vavilin, V.A., Fernandez, B., Palatsi, J., Flotats, X., 2008. Hydrolysis kinetics in anaerobic
degradation of particulate organic material: An overview. Waste Manage. 28 (6), 939951.
Vogt, T., Hoehn, E., Schneider, P., Freund, A., Schirmer, M., Cirpka, O.A., 2010a.
Fluctuations of electrical conductivity as a natural tracer for bank filtration in a losing
stream. Adv. Water Resour. 33 (11), 1296-1308.
Vogt, T., Schneider, P., Hahn-Woernle, L., Cirpka, O.A., 2010b. Estimation of seepage rates
in a losing stream by means of fiber-optic high-resolution vertical temperature
profiling. J. Hydrol. 380 (1-2), 154-164.
von Gunten, H.R., Karametaxas, G., Keil, R., 1994. Chemical Processes in Infiltrated
Riverbed Sediments. Environ. Sci. Technol. 28 (12), 2087-2093.
129
von Gunten, H.R., Karametaxas, G., Krahenbuhl, U., Kuslys, M., Giovanoli, R., Hoehn, E.,
Keil, R., 1991. Seasonal biogeochemical cycles in riverborne groundwater. Geochim.
Cosmochim. Ac. 55 (12), 3597-3609.
von Gunten, U., Zobrist, J., 1993. Biogeochemical changes in groundwater-infiltration
systems: Column studies. Geochim. Cosmochim. Ac. 57 (16), 3895-3906.
Woessner, W.W., 2000. Stream and fluvial plain ground water interactions: Rescaling
hydrogeologic thought. Ground Water 38 (3), 423-429.
Wondzell, S.M., LaNier, J., Haggerty, R., 2009. Evaluation of alternative groundwater flow
models for simulating hyporheic exchange in a small mountain stream. J. Hydrol. 364
(1-2), 142-151.
Woolsey, S., Capelli, F., Gonser, T., Hoehn, E., Hostmann, M., Junker, B., Paetzold, A.,
Roulier, C., Schweizer, S., Tiegs, S.D., Tockner, K., Weber, C., Peter, A., 2007. A
strategy to assess river restoration success. Freshwater Biol. 52 (4), 752-769.
Wroblicky, G.J., Campana, M.E., Valett, H.M., Dahm, C.N., 1998. Seasonal variation in
surface-subsurface water exchange and lateral hyporheic area of two stream-aquifer
systems. Water Resour. Res. 34 (3), 317-328.
Zwolsman, J.J.G., van Bokhoven, A.J., 2007. Impact of summer droughts on water quality of
the Rhine River - a preview of climate change? Water Sci. Technol. 56 (4), 45-55.
130

Riverbank Filtration Within The Context of River Restoration and Climate Change

Uploaded by

Copyright:

Available Formats

Riverbank Filtration Within The Context of River Restoration and Climate Change

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Riverbank Filtration Within The Context of River Restoration and Climate Change

Uploaded by

Copyright:

Available Formats

University of Neuchtel

Centre for Hydrogeology

Thesis defense date: 20.08.2013

Ph.D. Thesis evaluation committee:

Grundwasserfliesspfaden und Fliesszeiten.

Motivation and background ......................................................................................... 1

Objectives and structure of this thesis ......................................................................... 8

New methods to estimate 2D water level distributions of dynamic rivers ......... 11

Interpolation methods to estimate river water levels ................................................. 13

Application to the Niederneunforn field site ............................................................. 17

Assessing the effect of different river water level interpolation schemes on

Interpolation methods ................................................................................................ 39

Method implementation ............................................................................................. 41

Groundwater flow and transport model ..................................................................... 47

Results and discussion ............................................................................................... 51

NOM degradation during river infiltration: Effects of the climate variables

Materials and methods ............................................................................................... 68

Results and discussion ............................................................................................... 73

Modeling the dynamics of oxygen consumption during riverbank filtration by a

Materials and methods ............................................................................................... 95

Results and discussion ............................................................................................... 99

Conclusions ............................................................................................................. 111

Conclusions and Outlook ................................................................................. 115

Conclusions ............................................................................................................. 115

Outlook .................................................................................................................... 120

Bibliography ........................................................................................................................... 123

Challenge 1: River restoration

Challenge 2: Climate change

1.2 Objectives and structure of this thesis

2.2 Interpolation methods to estimate river water levels

consists of three steps:

1. Establish a mathematical relationship

between the measurements

for any time step or time series:

3. Estimate the water levels or water level time series

for the support points located

on the same section as the fixpoint .

has been obtained using one of

Method 1: Regression of measured data (RM)

and periodic water level measurements at a fixpoint

(Fig. 2.2). Each time

is denoted as determining gauging station

Method 2: Interpolation of measured data (IM)

. The fixpoints can

be located between or outside of the two gauging stations.

are smaller than

, when water levels at the determining gauging station

than a threshold water level

, the deviation from the linear trend has to be

Method 3: Regression of hydraulic model data (RH)

2.3 Application to the Niederneunforn field site

Data collection and method implementation

Comparison of the interpolation methods

was used as the determining

are identical for the RM and the IM method,

revealed a difference in the behavior of the two

(371.6-372.5 m asl). Depending on the polynomial fit,

was obviously too narrow in order to reliably predict water levels

was the determining