Groundwater Model of the Seeland Aquifer

Amt für Wasser Office des eaux und Abfall et des déchets

Bau -, Verkehrs- Direction des travaux undErreur Energiedirektion ! Nom de publics, des transports despropriété Kantons deBern document et de l’énergie inconnu. du canton de Berne Erreur ! Nom de propriété de document inconnu.

GROUNDWATER MODEL OF THE

SEELAND AQUIFER

Dr. Fabien Cochand

Rolf Tschumper

Prof. Philip Brunner

Prof. Daniel Hunkeler

Neuchâtel, le 19.07.2019

Table of contents

1 Introduction ...... 1

2 General consideration...... 1

2.1 Previous studies ...... 1

2.2 Seeland aquifer ...... 2

2.3 Model geometries ...... 3

2.4 Simultaneous field measurements ...... 5

2.5 General modelling methodology ...... 5

3 Model development ...... 6

3.1 Model mesh development ...... 6

3.2 Steady-state model boundary conditions ...... 6

3.2.1 River BCs ...... 6

3.2.2 Fixed GW head BCs and well BCs ...... 7

3.2.3 Lateral inflow BCs ...... 7

3.2.4 Recharge from rainfall ...... 10

3.3 Transient model boundary conditions ...... 11

4 Steady-state model ...... 13

4.1 Calibration procedure ...... 13

4.2 Calibrated value of the hydraulic conductivity ...... 14

4.3 Reproduction of groundwater heads ...... 15

4.4 Water balance ...... 16

4.5 River-aquifer interactions ...... 18

4.6 Uncertainties ...... 19

5 Transient model ...... 21

5.1 Calibration procedure and initial conditions ...... 21

5.2 Transient reproduction of the groundwater heads ...... 21

5.3 Minimum, mean and maximum maps ...... 25

5.4 Summary of model weakness or uncertainties ...... 25

6 Model applications ...... 26

6.1 Pumping test at the Gimmiz pumping station ...... 27

6.2 Sugar factory pollution ...... 30

6.2.1 Advective behaviour of the contaminant ...... 30

6.2.2 Transport model ...... 32

7 Model improvement and conclusion ...... 33

8 References ...... 34

9 Appendices ...... 36

Table of figures

Figure 1: Seeland geology (from WWA (2004)) ...... 3

Figure 2: Model area and boundaries ...... 4

Figure 3: Model mesh ...... 6

Figure 4: Implemented BCs ...... 8

Figure 5: Lateral inflow calculation: (a) Monthly mean precipitation and monthly mean calculated PET, (b) Net precipitation and (c) Monthly mean lateral inflow ...... 9

Figure 6: Soil types within the Seeland (left) and percentage of agriculture (right)...... 10

Figure 7: Implemented temporary pumping wells ...... 12

Figure 8: Pilot points and measured GW heads used to calibrate the model...... 13

Figure 9: Calibrated hydraulic conductivity ...... 14

Figure 10: Observed vs. simulated GW head scatter plot ...... 15

Figure 11: Observed vs. simulated GW head errors at observation wells ...... 16

Figure 12: Simulated river-aquifer interactions ...... 18

Figure 13: Relative uncertainty reduction calculated by PEST ...... 20

Figure 14: Observed transient groundwater head observation wells ...... 23

Figure 15: Weakness areas of model ...... 26

Figure 16: Gimmiz pumping test hydrographs ...... 28

Figure 17: Observed vs simulated drawdown ...... 29

Figure 18: a) Expected particle tracking using observed groundwater head contours

(KELLERHALS+HEAFELI, 2018) and b) small-scale simulated particle tracking ...... 31

1 Introduction

According to articles 39 to 41 of the cantonal law for water use (WNG) the Office for Water and Waste of the canton of Berne (AWA) is in charge to collect the fundamental information for the use and the conservation of any surface and subsurface water.

To fulfil this aim:

 The hydrogeological archive (SousSol) and available datasets (Blaue Berichte, local studies,

HydroPro, WAWIKO, …) are of great value. This Information should be better evaluated and

documented for the public.

 Updated information of groundwater must be provided in an easy way to involved parties and

interested people, in particular departments and responsible persons.

 The most important information must be provided to the public as thematic maps in the

Geoportal of the Canton Bern and must be updated.

 Monitoring the ongoing situation of the groundwater table in a quantitative and qualitative

way according to the Wasserversorgungsstrategie 2010.

In order to achieve general objectives described above, a groundwater flow FEFLOW model of the Seeland aquifer was developed.

This report presents the technical development and results of the Seeland aquifer model. The model development is based on the preliminary study entitled “GROUNDWATER MODELLING

TASK OF THE SEELAND AQUIFER - PRELIMINARY STUDY - 12.05.2017”. In this report, we will focus on technical modelling aspects and some key points will be repeated.

2 General consideration

2.1 Previous studies

Numerous geologic and hydrogeological studies have been dedicated to the Seeland aquifer.

The first comprehensive hydrogeology study of the Seeland can be found in WEA (1976) (WEA is now AWA) and includes a strategy for a sustainable groundwater resources management. Another

report (WEA 1989) is dedicated to the interactions between the aquifer, the former Aare and the

Hagneck Canal. To date, the most comprehensive general studies is WEA (1998). Other studies investigated more specific aspects; Jammet (2011) investigated the interactions between the Hagneck

Canal, the aquifer and the pumping wells of Gimmiz during different pumping conditions, Wanner and

Böhi (2011) carried out a pumping test at Gimmiz and Jordan (2000) developed a 2D Feflow model of the northern part (north of the Hagneck Canal) of the Seeland aquifer to determine the capture zone of the Worben pumping well. Before this study, WEA (1998) developed also a numerical model of the entire Seeland aquifer (northern part and southern part) to establish water balances and evaluate its hydrodynamic behaviour. There are also several studies regarding the water quality and hydrochemistry, such as Hoffmayer (1995), Ullrich (1998) or Baillieux et al. (2014). Finally, WWA

(2004), published by the AWA, compiled reports and data regarding the Seeland aquifer and presented the state of knowledge. All these studies provide many needed information and were intensively consulted.

2.2 Seeland aquifer

Figure 1 presents the geological context of the Seeland. The aquifer consists of gravels which are known as Aareschotter. The aquifer also contains low-permeable finer sediments

(“Verlandungsböden und feinkörnige Sedimente” in Figure 1) but the Aareschotter constitute the most important sediment of the productive (permeable) aquifer.

Figure 1: Seeland geology (from WWA (2004))

2.3 Model geometries

The model was developed with the computer program FEFLOW 7 using the finite element analysis. In our case, the numerical simulator was configured to solve the equation of a 2D horizontal saturated unconfined groundwater flow. The top of the model is represented by the ground surface and the bottom is defined by the granular aquifer bottom which was calculated in 2016 by the AWA

(AWA, 2016). The model boundaries are defined by the boundaries between granular aquifer and the molasse or moraine or small valleys (lateral boundaries in Figure 2 and along the Aare valley between

Aarberg and Mühleberg). The northern boundary is located along the Nidau-Büren Canal. Although

the river probably does not drain the entire groundwater flow and does not constitute the end of the aquifer, the river represents a pragmatic boundary because the river water level is known and consequently, a 1st kind boundary condition (BC) can be easily implemented. In addition, the aquifer part beyond the canal is less conductive and less productive which makes this part less important according to AWA. The Treiten-Fräschels boundary represents the end of the productive aquifer

(Figure 2). Beyond this boundary, the aquifer is less productive and consists of fine sediments (WEA,

1976). Only 15% of the groundwater flow (see Appendix 1 and Appendix 2 for aquifer water balances) of the aquifer flow across this boundary. Additionally, a groundwater observation well is located along this transect and therefore, a 1st kind BC may be easily implemented (see next section for further details about model BCs).

Figure 2: Model area and boundaries

2.4 Simultaneous field measurements

On the 4th April 2017 and on the 20th November 2017 two simultaneous field measurement campaigns (Simultanmessung) were carried out (AWA, 2017). The main goal of these campaigns is to provide a large amount of groundwater head data representing the “hydro-geological state” of the study area for calibrating the model.

During these campaigns, groundwater heads in more than 200 piezometers and water levels in numerous rivers were measured (well locations are presented in Figure 8). On the 4th April 2017, also extensive chemical analysis found place in 35 boreholes and wells of all water supply companies in the Seeland.

2.5 General modelling methodology

Firstly, a steady-state model was calibrated using PEST (PEST, 2015) to reproduce the hydrogeological state at the date of the first Simultanmessung (4th April 2017). To evaluate the capability of the model, the simulated GW heads were compared with the GW head observations collected during the first Simultanmessung and the simulated water balance was compared with estimations of other studies (mainly WEA (1976) and WWA (2004)). Secondly, the calibrated steady- state model was used as starting point for the transient model and was run for the period from January

2010 to December 2017. The main goal of this step was to reproduce the aquifer dynamics and to represent the minimum, maximum and mean groundwater heads during the simulated periods. To achieve this, the hydraulic conductivities and porosities were adjusted to have the best fit with the transient GW head observations. Both calibration steps will be further described in chapter 4.1 and

5.1.

3 Model development

3.1 Model mesh development

A 2D mesh (Figure 3) was developed in the Feflow interface with the mesh generator

“Triangle” to solve the 2D saturated unconfined flow equation. A model mesh with equilateral (as equilateral as possible) elements facilitates the numerical convergence and ensures good numerical stability. Therefore, the mesh generator was constrained to create triangles with a minimum angle of

34°. Additionally, the mesh was refined near the main rivers and pumping wells where high hydraulic gradients are expected to ensure an adequate number of calculation node. The elements located on rivers have a maximum edge of 10 m and near pumping wells a maximum edge of 4 m. The total number of nodes is 127’706 and the number of elements is 251’128.

Figure 3: Model mesh

3.2 Steady-state model boundary conditions

3.2.1 River BCs

Cauchy BCs were used to reproduce river-aquifer interactions (green lines in Figure 4). To implement this type of BCs, river water levels and riverbed elevations have to be known. To define water levels in rivers, data of BKW, cantonal and federal gauging stations were used (pink triangles in

Figure 4). The water levels at the date of the first “Simulatanmessung” were implemented and between gauging stations, river water levels were linearly interpolated. For riverbed elevations the DTM- dataset “Gerinneschläuche” was used to extract the riverbed elevations. (note that the

Gerinneschläuche dataset was also used to define the ground surface (Geo7, 2011)). However, these data were not available for all rivers (Appendix 3). In rivers where this information was not available, field measurements taken during the “Simulatanmessung” were used (Appendix 4).

3.2.2 Fixed GW head BCs and well BCs

The following data were used to implement fixed groundwater head BCs presented in Figure 4:

 Treiten-Fräschels boundary: Observed groundwater heads in cantonal piezometer G104 at the

date of the first “Simulatanmessung”.

 Nidau-Büren-Kanal boundary: Observed water level at the federal gauging station “Aare –

Brügg #2029” at the date of the first “Simulatanmessung” and value obtained during the

Simultanmessung”.

 Saane boundary: Observed groundwater head during the Simultanmessung”

 Mühleberg boundary: No data were available in this area. Therefore, the elevation of the river

water level at the date of the first “Simulatanmessung” was used to implement the fixed

groundwater head BC.

Well BCs were implemented with annual mean pumping rates to reproduce municipal pumping wells.

3.2.3 Lateral inflow BCs

To calculate lateral inflows, the same methodology developed for the model of the Emmental aquifer (Cochand et al., 2016) was used. First of all, all subcatchments of the model boundary were defined by using 25x25m DEM and the hydrological tools of the ArcMap-GIS package. Then subcatchments were selected if any significant rivers drain the subcatchment. This means that all subcatchments presented in Figure 4 have a diffuse groundwater inflow in the aquifer. For example,

the Lyssbach catchment was not selected because we assumed that the major part of the water is drained by the river and consequently the subsurface groundwater inflow negligible.

Figure 4: Implemented BCs

Then, monthly mean water balances of each subcatchment were calculated using weather data of the

Mühleberg weather station. To illustrate the water balance calculation methodology, the Figure 5 shows each calculation step of the subcatchment #8. In Figure 5a, the potential evapotranspiration

(PET) and the precipitation (of the subcatchment #8) are presented. The PET (equ.1) was calculated using the Oudin equation (Oudin et al., 2005) given as:

푅 푇 +5 푃퐸푇 = 푒 ∙ 푎 (1) 휆휌 80

where 푅푒 is the extraterrestrial radiation depending only on latitude and Julian day, 휆 is the latent heat of evaporation for water, 휌 is water density, 푇푎 is the daily mean temperature measured at Mülheberg weather station. Then, 851% of the PET was subtracted from precipitation to obtain the monthly mean net precipitation (Figure 5b). Finally, the net precipitation was converted in suitable units to obtain lateral inflow. The monthly mean value was implemented in the steady-state model and a 6-month moving mean (Figure 5c) was implemented in the transient model. Results obtained with this methodology will be discussed in chapter 4.4.

Figure 5: Lateral inflow calculation: (a) Monthly mean precipitation and monthly mean calculated PET, (b) Net precipitation and (c) Monthly mean lateral inflow

1 we assumed that PET is not 100% effective. 85% of the PET is a frenquently observed value (see the underway recharge project of the FOEN carried out by the CHYN)

3.2.4 Recharge from rainfall

To calculate the recharge from the rainfall and the irrigation demand, an agricultural crop modelling task was carried out using a CropSyst model. CropSyst is a multi-year, multi-crop, daily time step cropping systems simulation model developed to serve as an analytical tool to study the effect of climate, soils, and management on cropping systems productivity, recharge and irrigation demand (Stöckle et al., 2003). The necessary data to run CropSyst are soil properties, crop types and weather data. Therefore fieldwork was carried in the spring (2017) to identify large-scale soil properties within the Seeland as part of a master thesis (Rüfenacht, 2017). Three main types of soil were identified (Figure 6 left) and for each soil type, a vertical cross-section with representative soil properties (porosity, hydraulic conductivity etc.) was created. Then, crop types were gathered from the

Federal Office for Agriculture (Figure 6 right) and weather data of the Mühleberg weather station were used to run CropSyst models for the 2010-2018 period.

Figure 6: Soil types within the Seeland (left) and percentage of agriculture (right).

Table 1 summarises CropSyst results. These results will be also further discussed in a later chapter.

Table 1: Simulated recharge and irrigation demand.

Soil Soil Soil 1 2 3 Recharge (mm/y) 412 401 409 Irrigation (mm/y) 4 5 4 3.3 Transient model boundary conditions

The transient model was run for the period from the January 1st 2010 to December 31th 2017.

Consequently, BCs have to be adapted and each BC modification is presented in Table 2.

Table 2: Transient BC modifications

River BCs

 Daily mean water levels in rivers were implemented. However, the numbers of gauging stations or measuring points used to interpolate river elevation were less important in transient conditions than in steady-state. Therefore, some slight differences in river BCs may be observed between the steady-state and transient model.  Canals water levels in the southern part of the aquifer were assumed constant.  A sinusoidal function was used to fill data gap of the Mühleberg gauging station (see time- series #35 in the transient model). GW head BCs

 Daily mean values also were implemented for fixed GW head BCs.  A linear regression between piezometers G106 and G104 was used to fill the data gap of the observation well G104 implemented as BC of the Treiten-Fräschels boundary (see time-series #21 in the transient model).  At the Saane boundary, one meter was subtracted from the Saane river water levels to define the BCs.  At the Mühleberg boundary, it was assumed that the groundwater head does not fluctuate because of the dam upstream. Therefore, the same BC as for the steady-state model was employed.  Downstream the Mühleberg power plant, the groundwater level was maintained at about 461m during a construction work. A temporary fixed groundwater head BC was implemented to reproduce this groundwater level (see time serie#901) Well BCs

 Monthly mean pumping rates were used to set well BCs.  During the Gimmiz pumping test (see chapter 6.1) daily mean values were used at Gimmiz and Römerstrasse pumping wells.  Additional well BCs (presented in Figure 7) were implemented to represent temporary pumping occurring during construction works near Studen, Aarberg and Lyss.

Lateral inflow BCs

 At first, monthly mean values were implemented as BCs in the model but results showed too big variations. Therefore the lateral inflows were step by step smoothed using moving mean averages. The best result was obtained using a moving mean of the last 6 months (Figure 5c). Recharge

 Daily mean recharge and irrigation outputs of the CropSyt model were implemented for the three zones of soil.

Figure 7: Implemented temporary pumping wells

4 Steady-state model

4.1 Calibration procedure

To calibrate the model, the pilot point methodology available in FEPEST was used. Figure 8 shows the location of the pilot points as well as measurement of GW heads used to calibrate the model. The emphasis of the calibration has been on the hydraulic conductivity and the river-aquifer transfer rate adjustments with the aim of reproducing observed groundwater heads and river-aquifer interactions.

Figure 8: Pilot points and measured GW heads used to calibrate the model.

4.2 Calibrated value of the hydraulic conductivity

Figure 9 shows the calibrated hydraulic conductivity. The calibrated values have the same order of magnitude as values presented in WWA (2004). In addition, the calibration carried out with pilot point allowed the reproduction of the heterogeneity of the aquifer. It can be seen that areas with a higher hydraulic conductivity, probably due to former meanders of the Aare, are reproduced in the middle of the aquifer between the Hagneck Kanal and the Nidau-Bürren-Kanal. Conversely, areas with lower hydraulic conductivity on the outskirts and in the southern part of the study area made of finer lacustrine sediments are also well reproduced with pilot point calibration procedure.

Figure 9: Calibrated hydraulic conductivity

4.3 Reproduction of groundwater heads

Figure 10 shows the observed vs. simulated scatter plot. As presented, the model reproduces the observed GW heads with a reasonable error. Except five observation wells, all observed groundwater heads are reproduced in a range of +/- 50 cm (histogram of errors are presented in

Appendix 5).

Figure 10: Observed vs. simulated GW head scatter plot

The simulated groundwater heads with an error higher than +/- 50 cm are located close to other wells accurately reproduced (Figure 11) and have not an error exceeding 80 cm. This suggests that the model misrepresents local aquifer heterogeneities. However, one point in lyss identified in Figure 11, the simulated groundwater head is 3.6 m below the observed value. During the calibration procedure, additional pilot points were added at this location (see Figure 8) to have a greater heterogeneity but this approach has not worked. This indicates clearly a deeper problem which may be due to several causes. The aquifer thickness may be incorrect, the observed hydraulic head value may be wrong, the lateral flux may be largely misestimate or the interaction with the Lyssbach incorrectly reproduced.

Despite several attempts to correct the simulated groundwater heads, consisting in varying locally the aquifer thickness, modifying lateral flux or changing the transfer rate coefficient between river and aquifer, the simulated groundwater head remained wrongly reproduced at this location. As suggested by the AWA, this error may also be caused by an error of GPS measurement in this enclosed valley.

Figure 11: Observed vs. simulated GW head errors at observation wells

4.4 Water balance

Table 3 presents the simulated water balance of the steady-state model. These values are compared with values calculated or estimated in WEA (1998) and WWA (2004) reported in Table 3 and presented in detail in Appendix 1, Appendix 2 and from Appendix 6 to Appendix 9.

Table 3: Simulated water balance

Simulated WEA (1998) and WWA Model component (2004) Rate in [l/s] Rate out [l/s] Rate in [l/s] Rate out [l/s] Recharge taking into 691 0 760 - account irrigation Lateral inflow 375 0 380 - River infiltration: 719 887 Alte Aare (Aarberg-Lyss) 160 0 190 400 Alte Aare (Lyss-Meienried) 213 21 22-229 0-400 Aare (Aarberg-Mühleberg) 56 103 - - Lyssbach 4 0 12-28 0 Hagneck-Kanal 157 28 100-305 34-148 Binnenkanal 1 193 0 156-190 Sägibach 43 0 1-32 0 Unterwasserkanal 0 190 40 50 Southern kanal 27 190 30-70 150-330 Saane 0 15 - - Northern boundary 0 700 870 Southern boundary 0 43 86 Saane boundary 18 0 Mühleberg boundary 0.35 0 Wells 0 218

Recharge calculated with CropSyst and the lateral inflow calculated by doing subcatchment water balances are very close to the previously estimated values in WEA (1998) and WWA (2004).

This demonstrates the efficacy and reliability of approaches employed to estimate these fluxes.

Regarding river-aquifer interactions, results also seem in accordance with the observations although, the observations may have significant uncertainties. The order of magnitude is similar and the two most important river sections, the Hagneck Kanal and the Alte Aare (from Aarberg to Meienried) are well reproduced. Finally, fluxes at the two main boundaries have the same order of magnitude of the observed values which means that the overall water balance is accurate. The water balance is presented in maps (created by the AWA) in Appendix 10 and Appendix 11.

4.5 River-aquifer interactions

Figure 12 presents the simulated river-aquifer interactions. These simulated interactions

(infiltration or exfiltration) are similar as those presented in WEA (1998) except for the main canal in the southern part of the aquifer. The difference in water level of the canal to change the infiltration of the canal in the aquifer into the exfiltration of the aquifer in the canal is very small, about 10 to 20 cm.

This small variation is impossible to reproduce at the scale of the model. However, groundwater heads and water balance are well reproduced in the southern part of the aquifer, therefore, the model was not modified in this area.

Figure 12: Simulated river-aquifer interactions

4.6 Uncertainties

The use of pilot points for calibration also allows for an uncertainty analysis of the model. The

Figure 13 shows the reduction of uncertainty induced by the calibration of the model. A value of 1

(green) means a full reduction of uncertainties and 0 (red) means no reduction of uncertainties. In other words, red areas show areas with more uncertainties and green with less uncertainty. Figure 13 shows a large area in the middle of the aquifer in which uncertainties are important. This is due to the very low hydraulic gradient in this area which prevents the calibration of the hydraulic conductivity with a low uncertainty. Any values of hydraulic conductivity above a certain threshold reproduce a water table with a low hydraulic gradient. Consequently, the calibrated hydraulic conductivity has a significant uncertainty.

This uncertainty analysis may be also used to define areas in which fieldwork (e.g., implementation of a new borehole) would be the most profitable in term of uncertain reduction. In our case, the most profitable area according to the uncertain analysis and our expert knowledge, would be the area surrounded in blue in Figure 13. In this area, the calibration was problematic and the analysis indicates that uncertainties remain in this area. Note that the calibration and the uncertainty analysis were kindly supervised and validated by Mr Doherty, the developer of the PEST software.

Figure 13: Relative uncertainty reduction calculated by PEST

5 Transient model

5.1 Calibration procedure and initial conditions

The new boundary conditions described in chapter 3.3 were applied to the model and the groundwater heads simulated by the steady-state model were used as initial conditions. Then the model was run for the period from the January 1st 2010 to December 31th 2017. However, the initial conditions do not represent the January 1st 2010 but April 4th 2017 and consequently, the first simulated months may be not accurately reproduced. Therefore, the first simulated year (i.e., 2010) is used to obtain accurate initial conditions and was not considered in the calibration procedure.

The aim of the transient calibration is to adjust slightly the porosity in order to reproduce transient groundwater head dynamics. Unfortunately, the simulation time of the model was too long to use pilot point methodology to calibrate the model. Therefore, the calibration was carried out manually by considering two zones: the plain and the valley between Aarberg and Mülheberg.

5.2 Transient reproduction of the groundwater heads

Figure 14 shows the observation wells used to validate the transient dynamics of the model. and the observed and simulated transient groundwater heads are presented in Appendix 12. Except for few piezometers (further detailed), all piezometers which represent different dynamics of the aquifer are well reproduced. This satisfactory reproduction is validated by the transient simulated groundwater head error statistic summarized in Table 4 in which all mean error and RMSE are lower than 50 cm

(except PW Vieleroltigen)

 In the southern part of the aquifer which consists of fine and low conductive deposits

represented by piezometers G106 and G107 are very well reproduced.

 Piezometers Römerstrasse, Gimmiz 1 to Gimmiz 5, G118 and G116 reproduce correctly

observed hydraulic heads. This means that the interaction between the aquifer and the

Hagneck-Kanal which is a key process in the aquifer dynamics is correctly simulated.

 Other piezometers located in the northern part of the aquifer (G282, G122, G286, G124,

G125, PW Kappelen, G121, P4, P7, P9 P109 Worben, PW SWG, G126, P26, RB1, P33,

G127, P41, P57,G129) which are mainly influenced by the Alte Aare and the Hagneck-Kanal

to a lesser extent are accurately simulated.

 Finally, piezometers located between Aarberg and Mülheberg (and the small part of the Saane

valley) also are correctly reproduced.

Few piezometers however present some discrepancies. Firstly piezometer G124, located in an area (near Lyss) in which the calibration of the steady-state model was already difficult. Although the gap between simulated and observed hydraulic heads is not prohibitive, the model hardly reproduces the general dynamics. As already evocated in chapter 4.3, the conceptual model at this location may be wrong. However, the extent of the misrepresentation of the groundwater dynamics is likely small because the dynamics in piezometer G125 which is close to G124 is accurately reproduced. Secondly the simulated dynamics of the piezometer G121 located along the Sägibach is flatter than the observed dynamics. The low hydraulic heads are accurately reproduced but high hydraulic heads are systematically underestimated. The transfer rate coefficients of the Sägibach were modified to make the model more sensitive to the water levels of the river but this modification was unsuccessful. The causes of this error may be various such as a wrong river water level, a local overestimation of the hydraulic conductivity. Nevertheless, piezometer P109 Worben and PW Worben located also along the Sägibach and PW Kappelen in the centre of the aquifer is very accurately reproduced. This suggests that the misrepresentation is restricted to the proximity of the well. Finally, although piezometers were installed between Aarberg and the confluence of the Aare and the Saane (CHYN1 to

CHYN5), the duration of the observations are too short to validate the model with certainty. Model simulates significant amplitudes in groundwater heads at each piezometer. Nevertheless, these amplitudes have not been yet observed. Therefore, data loggers in piezometer CHYN1 to CHYN5 should be not removed before two or three years to have longer observations. If it appeared that model overestimated groundwater head amplitudes, the model could be easily corrected by adjusting transfer rate coefficient of the Aare. Note that the updated observed groundwater heads at the CHYN1 to

CHYN5 piezometers are presented in Appendix 13 to provide an initial idea of the dynamics.

Figure 14: Observed transient groundwater head observation wells

Table 4: Statistic on simulated groundwater head errors

MEAN Standard Piezometers RMSE ERROR deviation G106 -0.06 0.09 0.16 G107 -0.22 0.14 0.14 Römerstrasse -0.24 0.21 0.32 GIMMIZ1 -0.11 0.12 0.29 GIMMIZ2 -0.02 0.11 0.29 GIMMIZ3 -0.18 0.13 0.28 GIMMIZ4 -0.37 0.19 0.34 GIMMIZ5 -0.37 0.19 0.33 G116 -0.24 0.24 0.35 G118 -0.08 0.16 0.27 G282 0.16 0.20 0.32 G122 -0.11 0.18 0.31 G286 0.12 0.04 0.12 G124 0.15 0.20 0.33 G125 0.15 0.13 0.19 P4 -0.04 0.07 0.13 P7 -0.02 0.05 0.09 P9 0.10 0.08 0.10 PW Kappelen 0.00 0.05 0.23 G121 0.39 0.26 0.25 P109 Worben 0.16 0.09 0.08 PW SWG -0.03 0.08 0.07 G126 0.26 0.15 0.09 P26 0.17 0.21 0.12 RB1 0.35 0.11 0.13 P33 0.29 0.20 0.20 G127 -0.16 0.18 0.29 P41 0.31 0.19 0.12 P57 -0.08 0.16 0.26 G129 -0.06 0.13 0.23 CHYN5 0.05 0.03 0.21 CHYN4 0.10 0.03 0.21 CHYN3 -0.21 0.04 0.13 CHYN2 -0.34 0.03 0.07 CHYN1 0.10 0.42 0.41 PW REWAGAU 0.35 0.12 0.27 DSK1/15 0.15 0.08 0.26 PW Wieleroltigen -0.77 0.38 0.32 S1 -0.60 0.35 0.10 S10 -0.01 0.05 0.18

5.3 Minimum, mean and maximum maps

To generate minimum, mean and maximum maps, the DHI-WASY plugin was employed. This plugin allows for extracting minimum, mean and maximum simulated groundwater heads at each calculation node of the transient model. To avoid wrong values biased by inaccurate initial conditions, the first year of the transient model was not considered for the extraction. Therefore, minimum, mean and maximum hydraulic heads are representative for the period 2011-2017. Finally, temporary pumping wells presented in Figure 7 were removed to avoid temporary low hydraulic heads in the map. The mean GW heads, the depth to groundwater and the delta H (Hmax-Hmin) maps are presented in

Appendix 14, Appendix 15 and Appendix 16.

5.4 Summary of model weakness or uncertainties

Figure 15 summarizes weakness areas of the model classified in two categories; weaknesses due to uncertainty of BCs (area 1 in green) and weaknesses due to uncertainties of conceptual model

(area 2 in purple). Uncertainties of area near Lyss are probably due to a poor understanding of local geology and were already discussed in section 4.3. In the valley between Aarberg and Oltigen, the lack of long term groundwater head observations does not allow the validation of the calibration. This uncertainty will be reduced in the next few years when more data will be available. Near Dotzigen, an area in which the simulated groundwater heads are upper than the ground surface but is not observed in the reality. However, the simulated steady-state groundwater heads are well reproduced in this area.

It is likely due to a drain network installed in the area but unfortunately, we don’t have the confirmation. Therefore model results in this area should be carefully used. The few first hundred meters after the Saane boundary have uncertainties due to the implemented BCs. As presented in the

Figure 11 and in the simulated hydrograph of the PW Wileroltigen, simulated groundwater heads are a slightly out of range. Therefore, we suggest a careful use of model results in this area. Finally, in areas close to the model boundaries and boundary conditions, the model results may be also subject to uncertainties because of the simplification of processes. We suggest also a careful use of model results in these areas.

Figure 15: Weakness areas of model 6 Model applications

In addition to the validation of the model calibration by comparing simulated and observed groundwater heads, model validation may be carried out by assessing the capability of the model to reproduce field experimentations or the reproduction of a contamination. In our case, two events were selected to validate the model capability; 1) a pumping test carried out at Gimmiz pumping station and

2) the accidental infiltration in the aquifer of sugar solution by the sugar factory of Aarberg.

6.1 Pumping test at the Gimmiz pumping station

During January and February 2011, a pumping test detailed in KELLERHALS+HEAFELI

(2011) was carried out at the pumping station of Gimmiz. Table 5 summarizes each phase of the pumping test and the amount of pumped water and the location of pumping wells are presented in

Figure 14.

Table 5: Amount of pumped water (l/min) and phases of the pumping test

The aim of this model application is to evaluate the capability of the model to reproduce the observed groundwater heads in each pumping well during the pumping test. Because the pumping test took place during the transient calibration period, the specific pumping rates of each well were already implemented in the transient model. Therefore, a detailed analysis of the Gimmiz and Römerstrasse hydrographs allows for assessing the ability of the model to reproduce the pumping test dynamics

(Figure 16). It seems that the model can reproduce the pumping test with a reasonable error given that the model was not calibrated specifically to reproduce this pumping test. The drawdowns during the pumping test, the slopes of the drawdown and the recovery period after the pumping test are similar and the differences between observed and simulated groundwater heads are about 20-30 cm.

Figure 16: Gimmiz pumping test hydrographs

Figure 17 shows the interpolated observed and the simulated drawdowns in the vicinity of the pumping wells. The observed drawdown looks more circular around the wells than the simulated drawdown but the model simulates also the diminution of the drawdown beyond the Hagneck Kanal.

This means that the simulated interactions between the Hagneck Kanal and aquifer are realistic.

However, the extent of the drawdown is more important in model and reach 0.9 m instead of 0.6 near

Walperswil. This may be due to a misrepresentation of the heterogeneity of the aquifer or the underestimation of the lateral flux near Walperswil. It is not surprising that the model does not reproduce perfectly the pumping test because it was not calibrated for this purpose. However, the general behaviour of the aquifer is reproduced by the model.

Figure 17: Observed vs simulated drawdown

6.2 Sugar factory pollution

During the night of 30/31 August 2014, the sugar factory in Aarberg released accidentally about 300 m3 of sugar solution (melasse) with a sugar content of 67 %. The main part of the spilt melasse was recovered, but an unknown amount infiltrated in the aquifer (details in

KELLERHALS+HEAFELI (2018)). This accident may be reproduced by the model to demonstrate its reliability. To achieve this, the steady-state model was employed using to two different approaches; the first one by considering only the advective processes for the transport and the second by integrating dispersive and advective processes in the transport.

6.2.1 Advective behaviour of the contaminant

To analyse the advective transport of the sugar contaminant, forward particle tracking may be used in the model and compared with the expected evolution of the contamination detailed in

KELLERHALS+HEAFELI (2018). Figure 18a shows the expected contaminant flow direction

(KELLERHALS+HEAFELI, 2018) and Figure 18b the simulated traces at different scales. The expected and simulated behaviours of the contaminant are similar and follow the same pattern: after the infiltration, the contaminant flows to the right side of the aquifer perpendicularly to the Alte Aare and then flows in direction to the downstream part of the aquifer by following the aquifer boundary.

At a larger scale, the contaminant crosses the Alte Aare and reaches Kappelen. Because the model reproduces accurately groundwater contour, it is able to reproduce the general transport dynamics of the contamination that follows streamlines. Although, the particle tracking do not consider degradation and sorption which play a significant role in the transport dynamics, the particle tracking provides a good insight into general behaviour and evolution of the contamination.

Figure 18: a) Expected particle tracking using observed groundwater head contours (KELLERHALS+HEAFELI, 2018) and b) small-scale simulated particle tracking

6.2.2 Transport model

To have a more detailed insight, diffusion and sorption processes have to be added to the advective processes. Appendix 17 shows the simulated concentration at different time steps by considering dispersion and retardation in the transport. In this application, the longitudinal dispersivity is 50 m and the transverse dispersivity is 5 m according to Gelhar et al. (1992) and the retardation factor is 4 (-) and it was assumed that 1600 kg in one day of melasse infiltrates the aquifer.

This example was not accurately calibrated and various processes such as biological degradation were not taken into account. The main aim of this application is to demonstrate the applicability of the model. However, Appendix 17 gives a rough idea of the contaminant behaviour and may be compared to the advective behaviour. Note that the dark blue values in the appendix represent a very low concentration due to numerical dispersion and have to be interpreted as a concentration of 0 mg/l. After eight years the front part of the contaminant reaches the Alte Aare and it can be seen that the dispersivity leads to a lateral contamination which reaches the boundaries of the model. By integrating more data and spending more time in the calibration, the model may be used to reproduce more accurately the contamination dynamics and make predictions.

7 Model improvement and conclusion

This report showed the development of a steady-state and transient models of the Seeland aquifer to extract minimum, mean and maximum groundwater heads within the aquifer. The model was successfully calibrated in steady-state with a mean error of 0.16 m between simulated and observed groundwater. Then the model was run and calibrated to reproduce the daily groundwater dynamics of the period 2010 – 2017. Here too, the model behaviour was very satisfying. All different groundwater dynamics of the aquifer such as for example near the Hagneck Kanal or in the southern part of the aquifer were accurately reproduced. To have an additional validation of the ability of the model to reproduce aquifer dynamics, the model was employed to reproduce a pumping test and a contamination of the aquifer. Here again, the model was able to reproduce the dynamics. All these validations ensure a high level of certainty of model results and allow the extraction of minimum, mean and maximum groundwater heads of the aquifer to produce map which was the main goal of this modelling task. However, the model could be improved if more observation data were available mainly between Aarberg and Mühleberg. In this area, the groundwater observation data are not long enough to validate the long term model behaviour. Finally, a better understanding of the groundwater dynamics near Dotzigen could allow the improvement of the model in this area where the model simulate a groundwater heads high than the ground surface.

8 References

AWA: Dokumentation Stauerkarte, 2016.

AWA: Nachführung der Grundwasserkarte im Gebiet Seeland, Dokumentation Simultanmessung, 2017.

Baillieux, A., Campisi, D., Jammet, N., Bucher, S., and Hunkeler, D.: Regional water quality patterns in an alluvial aquifer: Direct and indirect influences of rivers, J. Contam. Hydrol. (2014), http://dx.doi.org/10.1016/j.jconhyd.2014.09.002, 2014.

Cochand, F., Brunner, P., and Hunkeler, D.: Groundwater modelling task of the Emmental aquifer.Centre for Hydrogeology and Geothermics (CHYN), Université de Neuchâtel, 2016.

Gelhar, L. W., Welty, C., and Rehfeldt, K. R.: A critical review of data on field-scale dispersion in aquifers, Water Resources Research, 28, 1955-1974, doi:10.1029/92WR00607, 1992.

Geo7: Qualitätsanalyse DTM Gerinneschläuche, 2011.

Hoffmayer, P.: Hydrogéologie et hydrochimie dans l'aquifère du Seeland (Berne, Suisse). Postgraduate thesis in Hydrogeology. Centre for Hydrogeology and Geothermics (CHYN), Université de Neuchâtel, 1995.

Jammet, N.: Caractérisation et suivi de la relation entre le canal d'Hagneck et les captages de Gimmiz (aquifère Nord du Seeland) sous différentes conditions de pompages. Masterthesis CHYN, Université de Neuchâtel, 2011.

Jordan, P.: Modélisation de la partie nord de l'auquifère du Seeland (BE) – Methode de détermination des aires d'alimentation Zu milieu poreux. Postgraduate thesis in Hydrogeology. Centre for Hydrogeology and Geothermics (CHYN), Université de Neuchâtel, 2000.

KELLERHALS+HEAFELI: Grundwasserfassungen Gimmiz Pumpversuch "Januar / Februar 2011", 22, 2011.

KELLERHALS+HEAFELI: Versickerung von Zuckerkonzentrat Grundwasserbeeinflussung: Stand Winter 2017/2018, 8, 2018.

Oudin, L., Hervieu, F., Michel, C., Perrin, C., Andréassian, V., Anctil, F., and Loumagne, C.: Which potential evapotranspiration input for a lumped rainfall–runoff model?: Part 2— Towards a simple and efficient potential evapotranspiration model for rainfall–runoff modelling, Journal of Hydrology, 303, 290-306, 2005.

PEST: http://www.pesthomepage.org/, 2015.

Rüfenacht, F.: Soil Hydraulic Properties and Groundwater, Recharge in the Seeland, Master Thesis in Environmental Engineering - ETH Zürich, 2017.

Stöckle, C. O., Donatelli, M., and Nelson, R.: CropSyst, a cropping systems simulation model, European Journal of Agronomy, 18, 289-307, https://doi.org/10.1016/S1161- 0301(02)00109-0, 2003.

Ullrich, S.: Distribution spatiale de l'état redox dans l'aquifère du Seeland entre Aarberg et Worben. Postgraduate thesis in Hydrogeology. Centre for Hydrogeology and Geothermics (CHYN), Université de Neuchâtel, 1998.

Wanner, J., and Böhi, D.: Grundwasserfassungen Gimmiz : Pumperversuch Januar/Februar 2011. KELLERHALS + HAEFELI AG, GEOLOGEN, 3011 BERN 2011.

WEA: Hydrogeologie Seeland. Geologen SIA, Bern, 1976.

WEA: Nutzungs-, Schutz- und Uberwachungskonzept Seeland - Technischer Bericht. Geotechnisches institut AG, Bern, 1998.

WWA: Hydrogeologie Seeland - Stand 2004. Geotechnisches institut AG, Bern, 2004.

9 Appendices

Appendix 1: Southern aquifer water balance (from WWA (2004))

Appendix 2: Northern aquifer water balance (from WWA (2004))

Appendix 3: Available riverbed elevations

Appendix 4: Additional field measurements.

Appendix 5: Error distribution of the simulated groundwater heads (steady-state model)

Histogram of simulated errors

70 60 50 40 30 20 10 0 -0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8

NB OF OBSERVATION WELLSOBSERVATION OF NB ERRORS (M)

Appendix 6: Northern aquifer river-aquifer interaction

Appendix 7: Southern aquifer river-aquifer interaction

Appendix 8: Amount of water exchanged between Hagneck Canal and aquifer (from (WWA, 2004))

Appendix 9: Amount of water exchanged between Alte Aare and aquifer (from (WWA, 2004))

Appendix 10: Simulated water balance of the northern part

Appendix 11: Simulated water balance of the southern part

Appendix 12: Simulated (in red) and observed (in black) hydraulic heads

Appendix 13: Simulated (in red) and observed (in black) hydraulic heads for piezometer CHYN1 to CHYN5 with updated observations

Appendix 14: Simulated mean groundwater heads

Appendix 15: Depth to groundwater (Ground elevation – mean groundwater heads)

Appendix 16: Difference in groundwater head elevation (Hmax-Hmin)

Appendix 17: Simulated concentration at different times.