impact of climate change on groundwater...

Université de Liège Faculté des Sciences Appliquées

Département ArGEnCo Architecture, Géologie, Environnement et Constructions

Secteur GEO³ Géotechnologies, Hydrogéologie, Prospection Géophysique

Impact of Climate Change on Groundwater Reserves

by Pascal GODERNIAUX

Thesis presented to the University of Liège in fulfilment of the requirements for the degree of

"Docteur en Sciences de l'Ingénieur"

January 2010

This research was carried out by: Pascal GODERNIAUX Research Fellow ("Aspirant du FNRS") Hydrogeology and Environmental Geology Group ArGEnCo Department University of Liège (ULg) Chemin des Chevreuils, 1 Building B52 4000 Liège BELGIUM

This research was financed by: Fonds de la Recherche Scientifique – FNRS (Funds for Scientific Research – FNRS) Rue d'Egmont, 5 1000 Bruxelles BELGIUM

Board of examiners composed of: Prof. CHARLIER Robert, Université de Liège (Belgium) – President Prof. DASSARGUES Alain, Université de Liège (Belgium) – Supervisor Dr. BROUYERE Serge, Université de Liège (Belgium) – Co-supervisor Prof. PIROTTON Michel, Université de Liège (Belgium) Dr. BLENKINSOP Stephen, Newcastle University (United Kingdom) Prof. THERRIEN René, Université Laval (Canada) Prof. VANCLOOSTER Marnik, Université Catholique de Louvain (Belgium)

Citation: Goderniaux P., 2010. Impact of Climate Change on Groundwater Reserves. PhD Thesis. University of Liège, Faculty of Applied Sciences. Liège, Belgium. pp. 190.

Summary

5

SUMMARY

Estimating the impacts of climate change on groundwater represents one of the most difficult

challenges faced by water resources specialists. One difficulty is that simplifying the

representation of the hydrological system, or using too simple climate change scenarios often

leads to discrepancies in projections. Additionally, these projections are affected by uncertainties

from various sources, and these uncertainties are not evaluated in previous studies. In this

context, the objective of this study is to provide an improved methodology for the estimation of

climate change impact on groundwater reserves, including the evaluation of uncertainties. This

methodology is applied to the case of the Geer basin catchment (480 km²) in Belgium.

A physically-based surface-subsurface flow model has been developed for the Geer basin with

the finite element model HydroGeoSphere. The simultaneous solution of surface and subsurface

flow equations in HydroGeoSphere, as well as the internal calculation of the actual

evapotranspiration as a function of the soil moisture at each node of the defined evaporative

zone, improve the representation and calibration of interdependent processes like recharge,

which is crucial in the context of climate change. Fully-integrated surface-subsurface flow models

have recently gained attention, but have not been used in the context of climate change impact

studies.

This surface-subsurface flow model is combined with advanced climate change scenarios for the

Geer basin. Climate change simulations were obtained from six regional climate model (RCM)

scenarios assuming the SRES A2 greenhouse gases emission (medium-high) scenario. These

RCM scenarios were statistically downscaled using two different methods: the 'Quantile Mapping

Biased Correction' technique and a 'Weather Generator' technique. Both of them are part of the

most advanced downscaling techniques. They are able to apply corrections not only to the mean

of climatic variables, but also across the statistical distributions of these variables. This is

Summary

6

important as these distributions are expected to change in the future, with more violent rainfall

events, separated by longer dry periods. The 'quantile mapping bias-correction' technique

generate climate change time series representative of a stationary climate for the periods 2011-

2040, 2041-2070 and 2071-2100. The 'CRU' weather generator is used to generate a large number

of equiprobable scenarios simulating full transient climate change between 2010 and 2085. All

these scenarios are applied as input of the Geer basin model.

The uncertainty is evaluated from different possible sources. Using a multi-model ensemble of

RCMs and GCMs enables to evaluate the uncertainty linked to climatic models. The application

of a large number of equiprobable climate change scenarios, generated with the 'weather

generator', as input of the hydrological model allows assessing the uncertainty linked to the

natural variability of the weather. Finally, the uncertainty linked to the calibration of the

hydrological model is evaluated using the computer code 'UCODE_2005'.

The climate change scenarios for the Geer basin model predict hotter and drier summers and

warmer and wetter winters. Considering the results of this study, it is very likely that groundwater

levels and surface flow rates in the Geer basin will decrease. This is of concern because it also

means that groundwater quantities available for abstraction will also decrease. However, this

study also shows that the uncertainty surrounding these projections is relatively large and that it

remains difficult to state on the intensity of the decrease.

Keywords: Groundwater, Climate change, Geer basin, Chalk, Integrated model,

HydroGeoSphere, Uncertainty, Stochastic scenarios, Weather generator, UCODE_2005.

Acknowledgements

7

ACKNOWLEDGEMENTS

This thesis is the achievement of three years work, during which many people supported and

helped me. I would like to acknowledge all of them gratefully.

Special thanks to Alain Dassargues and Serge Brouyère who supervised this thesis. Thank you to

them for all advices and discussions we had together during these years. Thank you for the time

they spent to read this manuscript and give me some remarks and corrections. I also would like

to acknowledge them for their important support when I applied for a position of 'research

fellow' at FNRS, in 2006. I also remember the times we spent together in meetings and

conferences. These times were both very instructive and pleasant. Finally, I would like to thank

Alain Dassargues for proposing me to join his team and giving us the means to carry out research

in a pleasant work environment.

Thank you to Robert Charlier, Michel Pirotton, René Therrien, Stephen Blenkinsop and Marnik

Vanclooster, who have accepted to join the jury of this thesis and to take time to read this report.

I am very grateful to René Therrien, Daniela Blessent and Jean-Michel Lemieux for helping me to

use 'HydroGeoSphere'. Thank you for their precious advices and recommendations. Thank you

to René Therrien who allowed me to spend 3 months in his team at University Laval (Québec,

Canada). Thanks to him, his family and Daniela Blessent for their warm welcome in Québec.

Similarly, special thanks to Hayley Fowler, Stephen Blenkinsop and Aidan Burton for the work

we have performed together to generate climate change scenarios for the Geer basin. Thank you

to them, as well as Isabella and Micol for the time spent in Newcastle. Thank you to Mary Hill

for giving me the opportunity to teach a short course with her and for answering very quickly to

my e-mails and questions about 'UCODE_2005'.

Acknowledgements

8

I am grateful to FNRS, which funded this research. FNRS and the University of Liège also

financed some presentations in international conferences, as well as scientific stays in University

Laval (Québec, Canada) and Newcastle University (Newcastle Upon Tyne, United Kingdom).

Thank you to the European Union FP6 Integrated Project AquaTerra (Project No. 505428),

which allowed me to meet very interesting people and to have decisive discussions.

Many thanks to the Administration of the Walloon Region, the 'Compagnie Intercommunale

Liégeoise des Eaux' (CILE), the 'Société Wallonne des Eaux' (SWDE), the 'Vlaamse Maatschappij

voor Watervoorziening (VMW), and the 'Afdeling Waterbouwkundig Laboratorium' of the

Flemish Region for all data they have provided about the Geer Basin.

Thank you to Christiane, Martine and Nadia, the secretariat staff, for their availability. Thank you

to Annick Anceau, from the 'Earth Science Library' of the University of Liège, for helping me in

bibliographic research. Thank you to the 'General IT department' (SEGI) of the University of

Liège, for their precious advices about 'NIC3', the super computer for intensive calculation at

University of Liège.

Particular thanks to all my colleagues and friends from the Hydrogeolgy Group and other teams:

Piotr, Philippe, Julie G., Ingrid, Samuel, Pierre J., Matthieu, Fabien, Julie C., Max, Pierre G.,

Nicolas, Cristina, Jordi, Laurent T., Jean -Michel, Tanguy, David, Jean, Frederic and many others.

Special thanks to Piotr who shared my office during more than 3 years.

Finally, I would like to address warm thanks to all my family. Particular thanks to you Céline for

your support and comprehension during these years. Thank you for your patience when I was

abroad. Thank you to have accompanied me in Quebec with the children. Thank you to be there

with me.

Pascal Goderniaux (January 2010)

Table of contents

9

TABLE OF CONTENTS

Summary__________________________________________________________________ 5

Acknowledgements __________________________________________________________ 7

Table of contents ___________________________________________________________ 9

List of figures _____________________________________________________________ 12

List of tables ______________________________________________________________ 15

Knowledge dissemination____________________________________________________ 16

1. Introduction __________________________________________________________ 21

1.1 Introduction ___________________________________________________________ 23 1.2 References_____________________________________________________________ 26

2. Scientific review _______________________________________________________ 27

2.1 Introduction ___________________________________________________________ 29 2.2 Groundwater modelling _________________________________________________ 29 2.3 Climate change modelling ________________________________________________ 34

2.3.1 Global Circulation Models (GCM) _______________________________________________ 35 2.3.2 Climate downscaling__________________________________________________________ 36 2.3.3 Greenhouse gases emissions scenarios ____________________________________________ 40

2.4 Evaluation of the uncertainty linked to climate change impact__________________ 42 2.5 References_____________________________________________________________ 44

3. Methodology __________________________________________________________ 49

3.1 Methodology ___________________________________________________________ 51 3.2 References_____________________________________________________________ 55

4. The Geer basin ________________________________________________________ 57

4.1 Geographical and geological contexts ______________________________________ 59 4.2 Hydrogeological context _________________________________________________ 60 4.3 References_____________________________________________________________ 64

5. Modelling ____________________________________________________________ 65

5.1 Hydrological modelling __________________________________________________ 67 5.1.1 Conceptual model ____________________________________________________________ 67 5.1.2 Mathematical and numerical model ______________________________________________ 67 5.1.3 Discretisation________________________________________________________________ 71 5.1.4 Specified Fluxes _____________________________________________________________ 72 5.1.5 Calibration procedure _________________________________________________________ 73

5.1.5.1 Parameter values ________________________________________________________ 74 5.1.5.2 Evaluation of the calibration _______________________________________________ 78

Table of contents

10

5.2 Sensitivity Analysis _____________________________________________________ 87 5.3 Discussion _____________________________________________________________ 91 5.4 Alternative calibration___________________________________________________ 94 5.5 References_____________________________________________________________ 97

6. Application of climate change scenarios on the Geer basin model _______________ 99

6.1 Climate scenarios ______________________________________________________ 101 6.2 Downscaling of RCM output_____________________________________________ 103 6.3 Projected changes in local climate ________________________________________ 104 6.4 Projected changes in hydrological regime __________________________________ 107 6.5 Discussion ____________________________________________________________ 110 6.6 References____________________________________________________________ 119

7. Application of stochastic climate change scenarios on the Geer basin model _____ 121

7.1 Objectives ____________________________________________________________ 123 7.2 Simulation of the stochastic climate change scenarios ________________________ 124

7.2.1 Generation of precipitations times series using 'RainSim'_____________________________ 125 7.2.1.1 General concepts _______________________________________________________ 126 7.2.1.2 Calibration of the model for the control scenarios (without any climate change) ______ 127 7.2.1.3 Calibration of the model for the climate change scenarios _______________________ 129 7.2.1.4 Generation of the climate change time series for the Geer basin___________________ 134

7.2.2 Generation of PET time series using the 'CRU daily weather generator' _________________ 134 7.2.2.1 General concepts _______________________________________________________ 135 7.2.2.2 Generation of the climate change time series for the Geer basin___________________ 136

7.3 Application of the climate change scenarios on the Geer basin model ___________ 138 7.3.1 Simulation conditions ________________________________________________________ 138 7.3.2 Evolution and uncertainty of projected groundwater levels and surface water flow rates ____ 138 7.3.3 Temporal uncertainty of a specific event _________________________________________ 145

7.4 Verification simulations of modelling hypotheses____________________________ 147 7.4.1 Influence of the number of equiprobable climatic scenarios___________________________ 147 7.4.2 Influence of the time discretisation ______________________________________________ 150

7.5 Discussion ____________________________________________________________ 153 7.6 References____________________________________________________________ 156

8. Uncertainty linked to the calibration of the model and summary of the results ____ 157

8.1 Introduction and objectives______________________________________________ 159 8.2 Estimation of the uncertainty linked to the calibration of the hydrological model _ 160

8.2.1 Methodology _______________________________________________________________ 161 8.2.2 Application of the methodology on the Geer basin hydrological model__________________ 165 8.2.3 Discussions about the confidence intervals________________________________________ 167

8.3 Summary of the results about climate change impact uncertainty ______________ 170

Table of contents

11

8.3.1 Uncertainty linked to the natural variability of the climate____________________________ 170 8.3.2 Uncertainty linked to the climatic models_________________________________________ 171 8.3.3 Uncertainty linked to the downscaling technique ___________________________________ 171 8.3.4 Uncertainty linked to the calibration of the hydrological model________________________ 172

8.4 Conclusions about uncertainty ___________________________________________ 174 8.5 References____________________________________________________________ 177

9. Conclusions and perspectives ___________________________________________ 179

9.1 Conclusions___________________________________________________________ 181 9.1.1 Hydrological modelling_______________________________________________________ 181 9.1.2 Climate change scenarios _____________________________________________________ 182 9.1.3 Uncertainty evaluation _______________________________________________________ 183 9.1.4 Impact for the Geer basin _____________________________________________________ 184 9.1.5 General conclusion __________________________________________________________ 185

9.2 Perspectives __________________________________________________________ 185

Appendix ________________________________________________________________ 187

List of figures

12

LIST OF FIGURES

Figure 2.1. (Figure and legend from IPCC, 2001b). The time evolution of the globally average

temperature change relative to the control run of the CMIP2 simulations. […].........................37 Figure 2.2. (Figure and legend from IPCC, 2000) Total global annual CO2 emissions from all sources

(energy, industry, and land-use change) from 1990 to 2100 (in gigatonnes of carbon

(GtC/yr)) for the families and six scenario groups. […] Each coloured emission band

shows the range of harmonized and non-harmonized scenarios within each group. For

each of the six scenario groups an illustrative scenario is provided (solid and dashed

lines). […] ................................................................................................................................................42 Figure 4.1. Location of the Geer basin and hydrologic limits..............................................................................59 Figure 4.2. Geological cross-section in the Hesbaye aquifer (modified from (modified from Brouyère

et al., 2004b)), with a vertical exaggeration equal to 40....................................................................60 Figure 4.3. Piezometry of the chalk aquifer (in metres above sea level) based on 2088 data (modified

from Orban, 2009; Ruthy, 2009). ........................................................................................................61 Figure 4.4. Evolution of groundwater levels at the observation well 'VIE044', from 1950 to 2006. ............63 Figure 5.1. Spatial discretisation of the Geer basin................................................................................................72 Figure 5.2. Distribution of the hydraulic conductivity zones for the chalk finite elements layers

(results of calibration)............................................................................................................................75 Figure 5.3. (A) Computed steady-state surface water elevations. (B) Computed steady-state subsurface

saturation, with full saturation shown in red (1967-1968)...............................................................79 Figure 5.4. Graphical analysis of the model calibration. (A) Computed values vs. observed values. (B)

Residuals vs. observed values. (C) Weighted residuals vs. observed values. Doted lines

represent increments of the calculated standard error. ....................................................................80 Figure 5.5 : Transient calibration of hydraulic heads for the nine observation wells.......................................85 Figure 5.6. Transient calibration of surface flow rates for the Kanne gauging station (outlet) ......................86 Figure 5.7. Composite Scales Sensitivities (CSS) of the parameters used in the Geer basin model ..............89 Figure 5.8. (A) Aggregated sensitivities for each observation point. (B) Mean leverage statistics for

each observation point. .........................................................................................................................91 Figure 5.9. Transient calibration of hydraulic heads for the 8 observation wells (2nd model).......................95 Figure 5.10. Transient calibration of surface flow rates for the Kanne gauging station (2nd model) ...........95 Figure 5.11. Graphical analysis of the model calibration. (A) Computed values vs. observed values.

(B) Residuals vs. observed values. (C) Weighted residuals vs. observed values...........................96 Figure 6.1. Monthly climatic changes for each climate change scenario (period 2071-2100) relatively to

1961-1990. (A) Temperature - Bierset climatic station. (B) Precipitation - Waremme

climatic station. .................................................................................................................................... 105

List of figures

13

Figure 6.2. Monthly climatic changes for the three time period 2011-2040, 2041-2070, 2071-2100

(climate model RCAO_E) relatively to 1961-1990. (A) Temperature – Bierset climatic

station. (B) Precipitation – Waremme climatic station.................................................................. 106 Figure 6.3. Evolution of hydraulic heads at the eight observation wells for each climate change

scenario, using the model with calibration 1................................................................................... 111 Figure 6.4. Evolution of hydraulic heads at the eight observation wells for each climate change

scenario, using the model with calibration 2................................................................................... 112 Figure 6.5. Evolution of flow rates at gauging station 'Kanne' for each climate change scenario, using

the model with calibration 1.............................................................................................................. 113 Figure 6.6. Evolution of flow rates at gauging station 'Kanne' for each climate change scenario, using

the model with calibration 2.............................................................................................................. 113 Figure 7.1. Schematic of the NSRP stochastic rainfall model (Figure from Burton et al., 2008). The

circles represent the rainfall events. Each star is associated with the beginning of a 'rain

cell'. ........................................................................................................................................................ 126 Figure 7.2. Observed, fitted and simulated precipitation statistics for the Geer basin climate

corresponding to the period 1961 – 1990 ....................................................................................... 129 Figure 7.3. Evolution of the scaling factors between 1975 and 2085 for GCM ECHAM4/OPYCA2..... 131 Figure 7.4. Target climatic statistics of RCAO_E for years 1995, 2025, 2055, 2085 .................................... 132 Figure 7.5. Evolution of target and mean simulated statistics (for successive periods of 1000 years)

between 1975 and 2085 for RCAO_E ............................................................................................ 133 Figure 7.6. Stochastic climate change scenarios. Precipitations of RCAO_E and ARPEGE_H for

February and August .......................................................................................................................... 134 Figure 7.7. Stochastic climate change scenarios. Monthly mean temperature of RCAO_E and

ARPEGE_H for February and August........................................................................................... 137 Figure 7.8. Stochastic climate change scenarios. Monthly mean PET of RCAO_E and ARPEGE_H

for February and August.................................................................................................................... 137 Figure 7.9. Evolution of hydraulic heads at the 8 observation wells for 30 equiprobable climatic

scenarios of ARPEGE_H (2010 – 2085)........................................................................................ 140 Figure 7.10. Evolution of hydraulic heads at the 8 observation wells for 30 equiprobable climatic

scenarios of RCAO_E (2010 - 2085)............................................................................................... 141 Figure 7.11. Mean hydraulic heads at the 8 observation wells for each of the 6 climatic models (30

scenarios) .............................................................................................................................................. 142 Figure 7.12. Evolution of water flow rate at the outlet of the basin for 30 equiprobable climatic

scenarios of ARPEGE_H and RCAO_E (2010 - 2085) .............................................................. 143 Figure 7.13. Mean water flow rate at the outlet of the basin for each of the 6 climatic models (30

scenarios) .............................................................................................................................................. 143

List of figures

14

Figure 7.14. (A) Mean groundwater levels (30 scenarios) and 95% interval at observation point

'OTH002', (B) mean annual water flow rates (30 scenarios) and 95% interval at

'KANNE', for the control simulations and the climatic models ARPEGE_H and

RCAO_E. ............................................................................................................................................. 145 Figure 7.15. Number of outcomes (10 m decrease in OTH002 groundwater level) for each time

interval and each climatic model....................................................................................................... 147 Figure 7.16. Probability density function and 95% interval for HIRHAM_H (OTH002)........................... 147 Figure 7.17. Results comparison when using 30 or 100 equiprobable scenarios of the climatic model

ARPEGE_H. (A) Mean groundwater levels and 95% interval at observation point

'OTH002'. (B) Mean annual water flow rates and 95% interval at 'KANNE'. ......................... 149 Figure 7.18. Comparisons of probability density functions and 95% intervals for ARPEGE_H

(OTH002) using 30 and 100 equiprobable scenarios. The vertical dotted lines represent

the limits of the 95% interval. ........................................................................................................... 149 Figure 7.19. Results comparison when using daily or monthly input solicitations for 30 equiprobable

scenarios of the climatic model ARPEGE_H. (A) Mean groundwater levels and 95%

interval at observation point 'OTH002'. (B) Mean monthly water flow rates and 95%

interval at 'KANNE' (February). (C) Mean monthly water flow rates and 95% interval at

'KANNE' (August). ............................................................................................................................ 152 Figure 8.1. Predictions and 95 % confidence interval around predicted values for 8 years of a

HIRHAM_H downscaled climate change scenario. (A) Absolute groundwater levels. (B)

Groundwater levels change between a scenario without any climate change and the

HIRHAM_H climate change scenario. ........................................................................................... 169 Figure 8.2. Predictions and 95 % confidence interval around predicted values for 8 years of a

HIRHAM_H downscaled climate change scenario. (A) Absolute surface water monthly

flow rates. (B) Change in surface water monthly flow rates between a scenario without

any climate change and the HIRHAM_H climate change scenario. .......................................... 170 Figure 8.3. Summary of all results and uncertainties for the 8 groundwater observation points. The

horizontal red line represents the 'control' mean groundwater level without any climate

change. .................................................................................................................................................. 173 Figure 8.4. Summary of all results and uncertainties for the water flow rate at the outlet of the

catchment ('Kanne'). The horizontal line represents the 'control' mean flow rate without

any climate change. ............................................................................................................................. 174

List of tables

15

LIST OF TABLES

Table 2.1. Selected examples of AOGCMs and spatial resolution (IPCC, 2001b; IPCC, 2007b)..................36 Table 2.2. SRES emissions scenarios (IPCC, 2000)...............................................................................................41 Table 5.1. Parameters used in the flow model........................................................................................................68 Table 5.2. Van Genuchten parameters, total porosity and specific storage.......................................................74 Table 5.3. Full saturated hydraulic conductivities values of the calibrated zones (results of calibration).....75 Table 5.4. Values for the Manning roughness coefficients and coupling length ..............................................76 Table 5.5. Root depths, evaporation depths and Leaf Area Index......................................................................77 Table 5.6. Mean errors between observed and computed heads for the nine observation wells (hobs:

observed hydraulic head, hcomp: computed hydraulic head, N: number of observations). .........82 Table 5.7. Simulated mean water balance terms for the period 1967-2003 .......................................................87 Table 6.1. Climate change scenarios with corresponding RCM and GCM. DMI: Danish

Meteorological Institute, HC: Hadley Center for Climate Prediction and Research, SMHI:

Swedish Meteorological and Hydrological Institute. ..................................................................... 102 Table 6.2. Variations of the mean water balance terms for each climate change scenario and time

interval, using the model with calibration 1. ................................................................................... 114 Table 6.3. Variations of the mean water balance terms for each climate change scenario and time

interval, using the model with calibration 2. ................................................................................... 115 Table 7.1. Random variables used in the NSRP model ..................................................................................... 127 Table 8.1. Summary of the results about climate change impact uncertainty for the period 2071-2100

(calculated with the first hydrological model, calibrated with daily stresses)............................. 170 Table A2. Van Genuchten parameters, total porosity and specific storage.................................................... 189 Table A3. Full saturated hydraulic conductivities values of the calibrated zones (results of calibration) .. 189 Table A4. Values for the Manning roughness coefficients and coupling length ........................................... 189 Table A5. Root depths, evaporation depths and Leaf Area Index................................................................... 190 Table A6. Simulated mean water balance terms for the period 1967-2003 .................................................... 190

Knowledge dissemination

16

KNOWLEDGE DISSEMINATION

Journal articles

- Goderniaux P., Brouyère S., Fowler H.J., Blenkinsop S., Therrien R., Orban Ph., Dassargues

A., 2009. Large scale surface - subsurface hydrological model to assess climate change impacts

on groundwater reserves. Journal of Hydrology, 373, 1-2, 122-138 pp.

- Barth J.A.C., Kalbus E. , Schmidt C. , Bayer-Raich M., Reinstorf F., Schirmer M., Thiéry D.,

Dubus I.G., Gutierrez A., Baran N., Mouvet C., Petelet-Giraud E., Négrel Ph., Banton O.,

Batlle Aguilar J., Brouyère S., Goderniaux P., Orban Ph, Rozemeijer J.C., Visser A., Bierkens

M.F.P., Van der Grift B., Broers H.P., Marsman A., Klaver G., Slobodnik J., Grathwohl P.,

2007. Selected groundwater studies of EU project AquaTerra leading to large-scale basin

considerations. Water Practice & Technology, 2-3, 10 pp.

- Visser A., Dubus I., Broers H.P., Brouyère S., Korcz M., Orban Ph., Goderniaux P., Batlle-

Aguilar J., Surdyk N., Amraoui N., Job H., Pinault J.-L., Bierkens M., 2009. Comparison of

methods for the detection and extrapolation of trends in groundwater quality. Journal of

Environmental Monitoring, 11, 2030-2043 pp.

- Goderniaux et al., 2010. Climate change impact on groundwater reserves using stochastic

climatic scenarios. Manuscript in preparation.

- Orban Ph., Brouyère S., Couturier J., Goderniaux P., Batlle-Aguilar J., Dassargues A., 2010.

Modelling nitrate trends a in chalk aquifer at regional scale. Manuscript in preparation.

- Blenkinsop S., Fowler H.J., Harpham C., Burton A., Goderniaux P., 2010. Modelling

transient climate change with a stochastic weather generator: Projected temperature changes

for the Geer catchment, Belgium. Manuscript in preparation.


17

Conference proceedings


A., 2009. How can large scale surface - subsurface hydrological model be used to evaluate long

term climate change impacts on groundwater reserves. The 7th International Conference on

Calibration and Reliability in Groundwater Modeling. "Managing Groundwater and the

Environment". Proceedings of MODELCARE 2009. Wuhan, China, September 2009, 137-

140 pp.

- Goderniaux P., Brouyère S., Dassargues A., 2007. Integrated approach for assessing climate

change impacts on a regional chalky aquifer in Belgium. Changes in Water Resources Systems:

Methodologies to Maintain Water Security and Ensure Integrated Management. Proceedings

of Symposium HS3006 at IUGG 2007, Perugia (Italy), July 2007, Vol. 315, 100-105 pp.

Conference abstracts


A., 2009. Large Scale Integrated Surface - Subsurface Hydrological Model to Assess Climate

Change Impacts on Groundwater Resources. The 2009 Ground Water Summit - The Science

and Engineering Conference: Adapting to Increasing Demands in a Changing Climate.

National Ground Water Association. Tucson, USA, April 2009.

- Orban Ph., Brouyère S., Couturier J., Wildemeersch S., Goderniaux P., Batlle-Aguilar J.,

Dassargues A., 2009. Assessment of Nitrate Trends in Groundwater Using the Regional Scale

HFEMC Approach. The 2009 Ground Water Summit - The Science and Engineering

Conference: Adapting to Increasing Demands in a Changing Climate. National Ground Water

Association. Tucson, USA, April 2009.

- Orban Ph., Goderniaux P., Batlle-Aguilar J., Brouyère S., 2009. Large-scale flow and

transport modelling for the management of groundwater bodies. AquaTerra Final Conference.

Processes-Data-Models-Future Scenarios. Scientific Fundamental for River Basin

Management. Tubingen, Germany, March 2009.

- Blenkinsop S., Fowler H.J., Burton A., Bovolo I., Van Vliet M., Goderniaux P., Forlin L.,

Harpham C., 2009. State-of-the-art climate change scenarios in AquaTerra. AquaTerra Final


18

Conference. Processes-Data-Models-Future Scenarios. Scientific Fundamental for River Basin

Management Tubingen, Germany, March 2009.

Scientific report – AquaTerra Project

- Goderniaux P., Brouyère S., Orban Ph., Dassargues A., Fowler H.J., Blenkinsop S., 2009.

Report on the development of the Geer hydrological model. Final results about climate

change impacts evaluation. Deliverable R3.30, AquaTerra (Integrated Project FP6 no. 505428),

60 pp.

- Goderniaux P., Brouyère S., Orban Ph., Dassargues A., 2008. Development of the Geer

basin hydrological model for climatic scenarios and first results about impacts evaluation.

Deliverable R3.26, AquaTerra (Integrated Project FP6 no. 505428), 29 pp.

- Goderniaux P., Brouyère S., Orban Ph., Dassargues A., 2007. Intermediate report on the

development of the Geer hydrological model (surface and subsurface water) for climatic

change scenario on that subcatchment. Deliverable R3.21, AquaTerra (Integrated Project FP6

no. 505428), 17 pp.

- Broers H.P., Visser A., Bierkens M., Dubus I., Pinault J.L., Surdyk N., Guyonnet D., Batlle-

Aguilar J., Brouyère S., Goderniaux P., Orban Ph., Korcz M., Bronder J., Dlugosz J.,

Odrzywolek M., 2008. Draft overview paper on trend analysis in groundwater summarizing

the main results of TREND2 in relation to the new Groundwater Directive. Deliverable

T2.12, AquaTerra (Integrated Project FP6 no. 505428), 46 pp.

- Dubus, I., Pinault J.-L., Surdyk N., Guyonnet D., Broers H.P., Visser A., Orban Ph., Batlle-

Aguilar J., Goderniaux P., Brouyère S., 2008. Report with comparison of statistical and

physically deterministic methods of trend assessment and extrapolation in terms of data

requirements, costs and accuracy. Deliverable T2.11, AquaTerra (Integrated Project FP6 no.

505428), 33 pp.

- Broers H.P., Visser A., Heerdink R., Van der Grift B., Surdyk N., Dubus I., Amaoui N.,

Orban Ph., Batlle-Aguilar J., Goderniaux P., Brouyère S., 2008. Report which describes the

physically deterministic determination and extrapolation of time trends at selected test

locations in Dutch part of the Meuse Basin, the Brévilles catchment and the Geer catchment.

Deliverable T2.10, AquaTerra (Integrated Project FP6 no. 505428), 47 pp.


19

- Hérivaux, C., Orban Ph., Batlle-Aguilar J., Brouyère S., Goderniaux P., 2008. Socio-economic

analysis integrating soil-water system modelling for the Geer catchment (Meuse, Walloon

region) - diffuse nitrate pollution in groundwater. Deliverable I3.8, AquaTerra (Integrated

Project FP6 no. 505428), 45 pp.

- Orban, Ph., Batlle-Aguilar J., Goderniaux P., Dassargues A., Brouyère S., 2006. Description

of hydrogeological conditions in the Geer sub-catchment and synthesis of available data for

groundwater modelling. Deliverable R3.16, AquaTerra (Integrated Project FP6 no

505428), 20 pp.

Additional publications

- Brouyère S., Batlle-Aguilar J., Goderniaux P., Dassargues A., 2008. A New Tracer Technique

for Monitoring Groundwater Fluxes: The Finite Volume Point Dilution Method. Journal of

Contaminan Hydrology, 95, 3-4, 121-140 pp.

- Goderniaux P., Brouyère S., Gutierrez A., Baran N., 2010. Persistence of agricultural

groundwater contamination related to hydraulic stratification as shown by long-term tracer

tests. Submitted to Hydrogeology Journal. Under revision.

- Brouyère S., Batlle-Aguilar J., Goderniaux P., Dassargues A., 2007. A new single well tracer

test: The Finite Volume Point Dilution Method. Theory, field application and model

validation – in "Calibration and Reliability in Groundwater Modelling – Credibility in

Modelling – Pre-published proceedings of MODELCARE 2007. Copenhagen, Danemark, 67-

92 pp.


20

1. Introduction

21

1. INTRODUCTION

1. Introduction

22

1. Introduction

23

1.1 Introduction

According to the Fourth Assessment Report (AR4) of the Intergovernmental Panel on Climate

Change (IPCC, 2007a), "Warming of the climate system is unequivocal, as is now evident from

observations of increases in global average air and ocean temperatures, widespread melting of

snow and ice and rising global average sea level". This report also concludes with 'very high

confidence' that this global warming was mostly due to the effect of human activities through

emissions of greenhouse gases in the atmosphere. For more than one decade, issues linked to

climate change have actually raised in importance in societies and political circles. Though

international agreements are difficult to achieve, more and more nations have the reduction of

greenhouse gases emissions in their agenda. In 1997, several countries adopted the 'Kyoto

Protocol' which agrees on an average 5% reduction compared to 1990 emissions over the period

2008-2012 (UNFCCC, 2009). In 2009, the recent 'Copenhagen Accord' "[…] recognises the

scientific view that the increase in global temperature should be maintained below 2 degrees

Celsius […]" to stabilise greenhouse gas concentration in the atmosphere at a level that would

prevent dangerous anthropogenic interference with the climate system (UNFCCC, 2009). Climate

change is likely to bring harmful effects on many ecosystems and human life and activities. Rising

sea levels may submerge inhabited areas, inducing massive migration of population. The

progression of deserts or more frequent occurrences of violent and destructive climatic events

could make life very difficult in some areas, constraining people to adapt themselves to new

conditions. One of the most important indirect issues linked to climate change relates to water

supply, which is obviously essential for life. The availability of water is also necessary for almost

any human activities, including agriculture and associated food security issues. A lot of scientific

research about impact estimation has been carried out in this particular topic. Nevertheless, this

research is often restricted to surface water reserves, neglecting groundwater. However,

groundwater represents an important percentage of total water supply across the world.

1. Introduction

24

According to the United Nations Environment Program about groundwater (UNEP - Morris et

al., 2003), "the contribution from groundwater is vital". Perhaps as many as two billion people

depend directly upon aquifers for drinking water. Forty percent of the world's food is produced

by irrigated agriculture that relies largely on groundwater and over half of the largest megacities in

the world directly rely on local groundwater or make significant use of it. Groundwater often

constitutes the only available resource in arid zones, but is also largely used under more

temperate climates, like in Walloon Region, where approximately 80% of the water supply comes

from aquifers (DGARNE, 2009). Groundwater reserves will continue to be heavily used in the

future, because they constitute an important part of available freshwater in our planet, but also

because groundwater is less sensitive to sudden climatic variations, compared to surface water. In

2001, IPCC experts also reported that "Groundwater is the major source of water across the

world", but they also noticed that "There has been very little research on the potential effects of

climate change" (IPCC, 2001a). Considering this gap, the International Association of

Hydrogeologists (IAH) has created in 2004 a new commission on this particular topic to support

research, studies and international contacts. Though more research has been carried out since the

report of 2001, the lack of knowledge about groundwater and groundwater – surface water

interactions is repeated in the IPCC technical paper about "Climate Change and Water" in 2008

(Bates et al., 2008). Moreover, the few scientific works about climate change impact on

groundwater show variable results. Though differences in the studied climate and aquifer types

surely exert an influence, the way of representing climatic and hydrogeologic systems certainly

contributes to variability in results and uncertainty. Additionally, the uncertainty linked to climate

change impact is not evaluated, or from a very limited number possible uncertainty sources. The

evaluation of this uncertainty is however of major importance to give some credibility to the

climate change impact study. It also enables water managers to analyse risks and take decisions

with full knowledge of projected impact and their degree of confidence.

1. Introduction

25

Considering this general context, the general objectives of this research are the following:

(1) development of a methodology for a reliable estimation of the climate change impacts on

groundwater reserves;

(2) estimation of the uncertainties characterising these projected impacts, considering various

possible uncertainty sources;

(3) pilot application of the two first objectives on the case of the Geer basin catchment

(Belgium).

In Section 2 of this thesis, a review of the scientific studies performed in the field of groundwater

and climate change is performed. In response to the relative weaknesses and strengths identified

in this existing scientific literature, Section 3 presents the methodology developed and used in the

current research. Section 4 describes the geological and hydrological contexts of the Geer basin.

Sections 5 to 8 expose the work performed and the main outcomes of the research, as obtained

following the methodology described in Section 3 ('Methodology'). Finally, Section 9 provides

conclusions and a reflection relative to the research perspectives.

1. Introduction

26

1.2 References

Bates, B.C., Kundzewicz, Z.W., Wu, S. and Palutikof, J., 2008. Climate Change and Water. Technical Paper of the Intergovernmental Panel on Climate Change, IPCC Secretariat, Geneva.

DGARNE, 2009. Etat des nappes d'eau souterraine de la Wallonie. Huitième année. Décembre 2009., Service Public de Wallonie. Direction générale opérationnelle, Agriculture, Ressources naturelles et Environnement. Direction de l'état environnemental. Direction des eaux souterraines.

IPCC (Editor), 2001. Climate Change 2001: Impacts, Adaptation and Vulnerability, Contribution of the working group II to the third assessment report of the Intergovernmental Panel on Climate Change (IPCC). Cambridge Univ. Press (UK), 1000 pp.

IPCC, 2007. Climate Change 2007: Synthesis Report. Contribution of Working Groups I, II and III to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, IPCC, Geneva, Switzerland.

Morris, B.L., Lawrence, A.R.L., Chilton, P.J.C., Adams, B., Calow, R.C. and Klinck, B.A., 2003. Groundwater and its Susceptibility to Degradation: A Global Assessment of the Problem and Options for Management. Early Warning and Assessment Report Series, RS. 03-3. United Nations Environment Programme, Nairoby, Kenya, 138 pp.

UNFCCC, 2009. http://unfccc.int - Official website of the United Nations Framework Convention on Climate Change (UNFCCC).

2. Scientific review

27

2. SCIENTIFIC REVIEW


28


29

2.1 Introduction

A huge number of research studies have been performed about climate change, through various

subjects such as the causes, modelling or impact of climate change on many types of systems.

The scientific review that follows this short introduction does not aimed to provide an inventory

of all of them, but focuses on climate change impact studies performed in the field of

groundwater, and on tools and methods used to achieve them. Particularly, it focuses on the

groundwater modelling work performed in the context of climate change. It also presents the

different modelling approaches to generate climate change scenarios and how they have been

used in groundwater impact studies. Finally, issues relative to uncertainty evaluation are briefly

discussed. This review attempts to highlight the weaknesses and strengths of previous studies,

and constitutes a foreword to the research performed in the framework of this thesis.

2.2 Groundwater modelling

Estimating the possible impacts of climate change on water resources represents one of the most

difficult challenges faced by water managers. Because of the great interest in such projections,

several studies have been recently published on the topic (see for example Christensen et al.,

2004; Fowler et al., 2003; Fowler et al., 2007b; Gellens and Roulin, 1998; VanRheenen et al.,

2004; Wilby et al., 2006). As already mentioned in Section 1, most of these studies focus on

surface water and generally oversimplify or even neglect groundwater, although groundwater is

the main water supply in many parts of the world. Additionally, studies that try to assess climate

change impact on water resources are likely to produce variable results (Jiang et al., 2007). One of

the main reasons for the discrepancy in projections is that simplistic assumptions are often made

to represent the physical processes associated with hydrological systems. This is particularly the

case for the studies that account for groundwater, where the representation of processes


30

associated with subsurface flows and groundwater recharge brings additional complexity. These

assumptions increase the uncertainty associated with model projections and need to be addressed.

A first requirement for estimating the impact of climate change on groundwater systems is a

reliable estimate of the volume of water entering and leaving an aquifer. More specifically, a

reliable estimate of groundwater recharge is needed because it represents the connection between

atmospheric and surface-subsurface processes and is therefore a key element in the context of the

impacts of climate change on groundwater. Similarly, in aquifers strongly influenced by surface

water, groundwater discharge into rivers may be affected by changes in surface water levels, and

consequently affect groundwater levels (Scibek et al., 2007). In previous studies about climate

change impact on groundwater reserves (see for example Brouyère et al., 2004b; Chen et al.,

2002; Holman, 2006; Loáiciga, 2003; Scibek et al., 2007; Serrat-Capdevila et al., 2007;

Woldeamlak et al., 2007), recharge has been estimated with various degrees of complexity,

ranging from simple linear functions of precipitation and temperature (Chen et al., 2002; Serrat-

Capdevila et al., 2007) to the application of "soil models" simulating variably-saturated

groundwater flow and solute transport (Allen et al., 2004; Brouyère et al., 2004b; Scibek and

Allen, 2006b).

Chen et al. (2002) and Chen et al. (2004) developed an empirical model, linking groundwater

levels to climatic variables, for a carbonate rock aquifer in Manitoba (Canada), and proposed to

use it for predicting groundwater levels under climate change conditions. Within this empirical

model, recharge is linked to precipitation and temperature using simple water balance linear

functions.

Serrat-Capdevilla et al. (2007) evaluate climate change impact on groundwater in the San Pedro

basin (Arizona, USA) for the period from 2000 to 2100, using a three dimensional 'MODFLOW'


31

model (Harbaugh, 2005). They calculate the 'mountain-front' recharge into the riparian basin

using an empirical relationship linking recharge to precipitation.

Loaiciga et al. (2000) and Loaiciga (2003) use the 'GWSIM' (Ground Water Simulation Program)

two-dimensional finite-difference model to study the impact of climate change on the Edwards

regional karst aquifer (Texas, USA), under various pumping conditions. Recharge is estimated

from a water balance of the streamflow in the area.

Allen et al. (2004), Scibek and Allen (2006b), Scibek and Allen (2006a), Scibek et al. (2007) study

climate change impact on the 'Grand Forks' Aquifer (British Columbia, Canada) which is an

alluvial aquifer strongly influence by river stages. They use a three dimensional 'MODFLOW'

groundwater model, together with the more complex 'HELP' model (Hydrologic Evaluation of

Landfill Performance – U.S. Environmental Protection Agency) (Schroeder et al., 1994) for

recharge rates calculation, and the 'BRANCH' model (Schaffranek et al., 1981) for river stages

calculation. The physically-based 'HELP' model enables to spatially calculate recharge rates taking

into account processes such as runoff, infiltration, evapotranspiration, surface and moisture

storage, snowmelt, etc.

Brouyère et al. (2004b) modelled groundwater flows under climate change conditions in a chalk

aquifer in Belgium. The saturated groundwater flow model is implemented with the finite

element code 'SUFT3D' (Brouyère, 2001; Carabin and Dassargues, 1999). Recharge rates are

calculated with the soil model 'EPIC-GRID', which performs water budget at the ground surface

level and in the unsaturated zone. Exchange fluxes are unidirectional, from the soil model to the

groundwater model.

Woldeamlak et al. (2007) and Yusoff et al. (2002) performed similar studies for a part of the

Grote-Nete basin (Belgium) and a chalk aquifer (West Norfolk, UK), respectively. In both

studies, the authors developed a groundwater model using 'MODFLOW'. Woldeamlak et al.


32

(2007) calculate groundwater recharge rates with 'WetSpass' (Batelaan and De Smedt, 2001), a

spatially-distributed water balance model. Yusoff et al. (2002) use a more conventional soil water

balance method. Finally, other authors also studied the impact of climate change on groundwater

recharge without assessing the effect on groundwater reserves (see for example Eckhardt and

Ulbrich, 2003; Herrera-Pantoja and Hiscock, 2008; Holman, 2006; Holman et al., 2009; Jyrkama

and Sykes, 2007).

Though groundwater recharge rates were estimated with various degrees of complexity in

previous studies and models, none of them can simulate the feedback, or fluid exchange, between

the surface and subsurface domains. This feedback is an integral component of the water cycle

since groundwater recharge depends on precipitation and evapotranspiration at the surface

domain, evapotranspiration in the vadose zone, evapotranspiration in the saturated zone when

water levels are close to the ground surface, and finally river – aquifer interactions. The

quantitative estimation of the latter four fluxes depends on the simulation of simultaneous

hydraulic conditions in the surface and subsurface domains. Therefore, estimating recharge by

only considering one part of the whole system is unrealistic, inaccurate and potentially unusable

in the context of climate change impact assessments. Similarly, loosely coupled modelling

approaches, where water exchange between surface and subsurface is calculated independently,

do not provide a sufficient level of realism because they do not solve for all the interdependent

processes simultaneously.

To compensate for this lack of interconnection when modelling groundwater, Van Roosmalen et

al. (2007) and Van Roosmalen et al. (2009) used a coupled model that simulates surface water and

groundwater flows simultaneously with water exchanges between both domains. This model,

described by Henriksen et al. (2003) and Sonneborg et al. (2003), and based on the 'MIKE SHE'

code (Graham and Butts, 2006; Refsgaard and Storm, 1995), is used to assess the impact of

climate and land use change on a large scale catchment in Denmark. However, this model uses a


33

relatively simple water balance method (Yan and Smith, 1994) to compute water flows in the

partially saturated zone. While this kind of simplification could be used in area where the

influence of the partially saturated zone is limited (provided some verifications), it constitutes a

serious limitation in other cases. The development and use of physically-based, fully-integrated

hydrological models able to simulate surface- and subsurface-flow in the saturated and partially

saturated zones, with a simultaneous solution of the flow equations in all domains, have recently

gained attention. Currently, the few modelling codes that are able to simulate these processes in

an integrated way include 'HydroGeoSphere' (Therrien et al., 2005), the 'Integrated Hydrology

Model' (InHM) (VanderKwaak, 1999), 'ParFlow' (Kollet and Maxwell, 2006), and 'OpenGeoSys'

(Delfs et al., 2009). As an example of application, Jones (2005) developed such a model for a 75

km² catchment (Laurel Creek Watershed – Ontario, Canada) using 'InHM'. The finite element

grid representing the catchment contained more than 600 000 nodes and transient simulations of

coupled surface and subsurface flow were run over periods of 1 month with specified fluxes

input on an hourly basis. Sudicky et al. (2008) present a similar model for a 17 km² subcatchment

of the Laurel Creek Watershed and use it to simulate contaminant transport issues. Another

example is reported by Li et al. (2008), who modelled, using HydroGeoSphere, surface and

subsurface flows, and evapotranspiration fluxes for a 286 km² catchment (Duffins Creek

Watershed – Ontario, Canada) with more than 700 000 nodes and made transient simulations

over 1 year periods with specified fluxes input on a daily basis. Finally, Kollet & Maxwell (2008)

used 'ParFlow' to implement an integrated surface – subsurface model for the Little Washita

watershed (Oklahoma, United States), which area is approximately 600 km², and performed 1-

year transient simulations with hourly time steps. A drawback of these models is that fully-

integrated simulations typically require substantial computer resources, and most simulations

published have been either limited to small catchments or short time periods. As an example, the

model of Jones et al. (2005) takes more than 4 days of computational time to simulate a period of

1 month with a 3.2 GHz Pentium4 desktop machine equipped with 4.0 Gb RAM. To our


34

knowledge, there is no example of such integrated surface – subsurface models used in the

context of climate change impact evaluation, which requires to simulate longer time periods

(typically 30 years minimum).

A second requirement for estimating the impact of climate change on groundwater systems is

that hydrogeological system models must be capable of consistently representing observed

phenomena, which is not always the case. As already presented above, Chen et al. (2002) propose

to estimate the impact of climate change on a Canadian aquifer with an empirical model that links

piezometric variations and groundwater recharge, where recharge is assumed to be a linear

function of precipitation and temperature. Most studies focussing on surface water, such as

Arnell (2003), also use simplistic transfer functions to represent exchanges between ground- and

surface water. However, such transfer functions often oversimplify the exchange processes.

These functions can still be substituted for more detailed physical representations for specific

conditions if they are verified with calibration, but their use may become uncertain if applied

stresses go beyond the calibration conditions, which is typical for climate change scenarios.

Detailed physically-based and spatially-distributed models that take into account hydrogeologic

processes provide more realistic simulations of groundwater fluxes, including exchanges with

surface water.

2.3 Climate change modelling

In addition to the choice of the hydrological modelling approach, the need for climate change

scenarios, to be used as input of hydrological models, add an additional layer of complexity and

uncertainty to future projections. Before proceeding to any hydrological simulations to evaluate

potential impacts, climatic stresses are needed and specific scenarios have primarily to be

produced.


35

2.3.1 Global Circulation Models (GCM)

General Circulation Models (GCM) constitute an important tool to simulate climatic variables.

According to Grotch and MacCracken (1991), General Circulation Models are "numerical models

that attempt to simulate the global climate by calculating the evolution of the atmosphere in all

three spatial dimensions based on the conservation laws for atmospheric mass, momentum, total

energy, and water vapour". GCMs are actually complex numerical climate models based on

physical laws and implemented at the scale of the earth globe. A GCM is generally the

combination of an 'Atmospheric General Circulation Model' (AGCM) and an 'Oceanic

Circulation Model' (OGCM) to form an 'Atmosphere-Ocean General Circulation Model'

(AOGCM). According to their complexity and continuous developments, AOGCMs are also

coupled with other models such as sea-ice models, land surface processes, biosphere processes,

carbon cycle models, atmospheric chemistry, etc. (IPCC, 2001b; IPCC, 2007b). Numerous GCMs

are available and have been developed during the last decade. Some of them are shown in Table

2.1 along with specific characteristics such as spatial resolution. GCMs can then be used to

generate climate change scenarios by increasing, for example, the emissions of CO2 or other

greenhouse gases. As an illustration, Figure 2.1 (IPCC, 2001b) shows the evolution of simulated

global temperature using different GCMs and considering a CO2 increase of 1% per year.

However, a drawback linked to the use of GCMs is that they are very computationally expensive.

Additionally, GCMs simulate climatic variables at a very large scale, typically for spatial

resolutions between 2° and 5° of latitude and longitude. Hydrological or hydrogeological studies

are on the other hand performed at a much more local scale, typically from a few hundred

squared metres to several squared kilometres. GCMs are unable to accurately simulate climatic

variables for this smaller scale because they do not account for local effects such as topography,

hydrography, land use, etc. As a result, a comparison between climatic observations at a specific

measurement station and the climatic outputs from a particular GCM would probably lead to


36

significant differences. Considering these issues, and particularly the mismatch of scales between

the spatial resolution of GCM outputs and the need for more local hydrological impact studies,

additional work is needed to produce appropriate climatic models and scenarios in this context.

Numerous techniques have been developed for years and are known as 'downscaling' techniques.

All existing techniques have been reviewed in several scientific papers, including Fowler et al.

(2007a) and Wilby and Wigley (1997). A very brief overview of these techniques, based on these

review papers is presented here after.

AOGCM Institute Resolution (atmosph.)

Resolution (ocean)

References

CCSM3 National Centre for Atmospheric Reasearch, USA

1.4° × 1.4° 1.0° × 1.0° (Collins et al., 2004)

CGCM3.1 Canadian Centre for Climate Modelling and Analysis, Canada

1.9° × 1.9° 10.9° × 1.4° (McFarlane et al., 1992)

CNRM-CM3 Météo-France, France 1.9° × 1.9° 2.0° × 2.0° (Deque et al., 1994)

ECHAM4/OPYC3

Max Planck Institute for Meteorology, Germany

2.8° × 2.8° 2.8° × 2.8° (Roeckner et al., 1996)

ECHAM5/MPI-OM

Max Planck Institute for Meteorology, Germany

1.9° × 1.9° 1.5° × 1.5° (Roeckner et al., 2003)

UKMO-HadCM3

Hadley Centre for Climate Prediction and Research/MET Office, UK

2.5° × 3.75° 1.25° × 1.25° (Gordon et al., 2000; Pope et al., 2000)

UKMO-HadGEM1

Hadley Centre for Climate Prediction and Research/MET Office, UK

1.3° × 1.9° 1.0° × 1.0° (Martin et al., 2004)

Table 2.1. Selected examples of AOGCMs and spatial resolution (IPCC, 2001b; IPCC, 2007b).

2.3.2 Climate downscaling

Downscaling techniques can be classified into two main categories: 'dynamical downscaling' and

'statistical downscaling'.


37

Figure 2.1. (Figure and legend from IPCC, 2001b). The time evolution of the globally average temperature

change relative to the control run of the CMIP2 simulations. […]

'Dynamical downscaling' mainly relates to the use of physically-based Regional Climate Models

(RCM). RCMs are climate models with a higher spatial resolution than GCMs. The modelled area

is limited in space and time-dependant boundary conditions are provided and driven by GCMs.

According to Fowler et al. (2007a), RCMs typically provide climatic simulations at a ~0.5°

latitude and longitude scale. First climate applications using RCMs were performed by Dickinson

et al. (1989) and Giorgi (1990). The complexity and number of climate processes included in

RCMs have then progressively increased. Compared to GCMs, RCMs can take into account

some additional regional features such as, for example, topography, vegetation cover, presence of

lakes, etc., which improves the performance of RCMs in reproducing regional climate and

extreme events. The European FP5 PRUDENCE project (Prediction of Regional scenarios and

Uncertainties for defining European Climate change risks and Effect) (Christensen et al., 2007)

produced such kind of high spatial resolution simulations for an ensemble of RCMs, using


38

different GCMs boundary conditions and greenhouse gases emission scenarios (see Section

2.3.3). However, RCMs also suffer from various drawbacks. As GCMs, Regional Climate Models

are computationally expensive and simulations are usually limited in simulated time. In the

context of climate change, simulations are typically performed only for the time slices 1961-1990,

representing a climate without any change, and 2070-2100, representative of a stationary climate

change over the 30-years period of the end of the century. Another major drawback is link to the

fact that RCMs results are strongly dependent on the driving boundary conditions selected from

GCMs. Though more accurate than GCMs, the use of Regional Climate Models is usually not

sufficient to produce climatic scenarios for local hydrological studies and further statistical

downscaling is generally required.

'Statistical downscaling' is based on empirical or statistical relationships betweens GCMs or

RCMs outputs and the local climate to be simulated. Typically, relationships are defined between

a GCM or RCM 'control simulation' without any climate change and observed climatic data from

a local measurement station. To generate climate change scenarios, the same relationships are

then used but with the corresponding GCM or RCM outputs that consider a specific climate

change. Statistical downscaling methods allow providing climatic scenarios at the very local scale

needed for most hydrological studies. They also have the advantage of being computationally

inexpensive and quite flexible. Nevertheless, they can only be applied where observed climatic

data are available in sufficient quantities. According to Fowler et al. (2007a) and IPCC (2001b;

2007b), there exist a huge range of different statistical downscaling methods, from very simple to

highly complex models involving large numbers of variables and parameters. They allow

modelling local climate with various degrees of complexity. Fowler et al. (2007a) classify statistical

downscaling methods into three main categories: the 'regression models', the 'weather typing

schemes' and the 'weather generators' techniques. Briefly, the 'regression' or 'transfer function'

methods use direct relationships between GCM or RCM variables (the 'predictors') and local


39

climate variables (the 'predictants') through the use of regression methods, for example. The

'weather typing scheme' relates to the occurrence of particular 'weather classes' to local climate

(Fowler et al., 2007a). Various methods exist for defining these 'weather classes', specifically for

the downscaling purposes. The 'weather generators' (Wilks and Wilby, 1999) are statistical models

that enable to generate climate scenarios based on the statistical distributions of a series of

climatic variables. Each distribution is determined using observed local climatic data and can be

modified based on GCMs or RCMs outputs. The 'weather generators' have the great advantage

that they enable to easily generate large numbers of stochastic climatic scenarios, that can be used

in subsequent probabilistic analyses of climatic variables or any potential impact.

To date, studies examining the impacts of climate change on groundwater systems have adopted

relatively simple statistical downscaling methods. One of the most straightforward approaches is

the 'perturbation' or 'delta change' method (Prudhomme et al., 2002) which applies 'change

factors' (CFs), calculated as difference between the control and future GCM simulations, to

observations (e.g. Brouyère et al., 2004b; Yusoff et al., 2002). However, since these scenarios

were produced by applying the projected changes to mean temperature and precipitation to the

whole of the corresponding future distribution, they fail to reflect changes in the shape of the

distribution, which is important for extremes or changes in the distribution of wet and dry

periods. In their impact study concerning groundwater in the Grand Forks aquifer, Scibek and

Allen (2006b) use the 'LARS-WG' weather generator (Semenov et al., 1998) to generate a single

100-years scenario representing a stationary climate for each of the periods 2010-2039, 2040-2069

and 2070-2099. (Holman et al., 2009) use the 'CRU' weather generator (Kilsby et al., 2007; Watts

et al., 2004) to produce 100 realisations of 30-years scenarios for the period 2040-2069 and use

them as input of a one-dimensional soil water balance to calculate groundwater recharge.


40

2.3.3 Greenhouse gases emissions scenarios

As briefly explained above, a large panel of climate models is available to produce climatic

simulations in relation with objectives and needs of specific studies. More particularly, they are

widely used to generate climate change scenarios for climate change impact studies, based on

CO2 and other greenhouse gases emissions, among others. The Intergovernmental Panel on

Climate Change (IPCC) developed a set of emissions scenarios known as the SRES scenarios

(Special Report on Emission Scenarios) (IPCC, 2000). All scenarios are regrouped within 4

different storylines which describe the context of the emission scenarios in various domains such

as economy, technology, demography, etc. These four storylines are described in details by IPCC

(2000). Main characteristics are summed up in Table 2.2. Figure 2.2 (IPCC, 2000) shows the

range of projected CO2 emissions between 1990 and 2100 for each of the four storylines. The

sets of greenhouse gases emission scenarios, available for each storyline, can then be used in

GCMs to produce climate change scenarios for specific climatic variables. As an illustration, the

different existing GCMs used with the A2 emission scenarios project a global temperature

increase between 1.3°C and 4.5°C for the period 2070-2100 relatively to the period 1961-1990.

The global temperature increase for the B2 emission scenarios varies between 0.9°C and 3.4°C.

Generally, the increase is expected to be higher for Northern Europe (IPCC, 2000; IPCC,

2001b).


41

Storyline A1

- World that becomes more homogeneous with increased social and cultural interactions

- Very rapid economic growth

- Global population that peaks in mid-century and declines after

- Rapid introduction of new and more efficient technologies

Scenario A1 is divided into 3 sub-categories based on their energy sources: fossil intensive (A1F1), non-fossil energy sources (A1T), balance across all sources (A1B)

Storyline A2

- Very heterogeneous world with self-reliance and preservation of local identities

- Regionally oriented economic development

- Global population that continuously increases

- More fragmented and slower technological change than other scenarios

Storyline B1

- World that becomes more homogeneous

- Rapid change in economic structures toward a service and information economy, with reductions in material intensity

- Global population that peaks in mid-century and declines after

- Introduction of clean and resource-efficient technologies

Storyline B2

- Heterogeneous world

- Emphasis is on local solutions to economic, social and environmental sustainability

- Global population that continuously increases, but at a lower rate than scenario A2

- Less rapid and more diverse technological change than in the scenarios A1 and B1

Table 2.2. SRES emissions scenarios (IPCC, 2000)


42

Figure 2.2. (Figure and legend from IPCC, 2000) Total global annual CO2 emissions from all sources (energy, industry, and land-use change) from 1990 to 2100 (in gigatonnes of carbon (GtC/yr)) for the

families and six scenario groups. […] Each coloured emission band shows the range of harmonized and non-harmonized scenarios within each group. For each of the six scenario groups an illustrative scenario is

provided (solid and dashed lines). […]

2.4 Evaluation of the uncertainty linked to climate change impact

While climate change impact studies focussing on groundwater reserves are not frequent, those

which attempt to assess the uncertainty of the impact estimations are even less common. To be

more specific, most existing studies actually take into account some kind of uncertainty, by using

multiple GCMs or RCMs. As explained above, climatic models can represent atmospheric

processes differently, through different numerical schemes or parameterisations. Testing an

ensemble of climatic models as input of the hydrological model enables to assess the uncertainty

linked to climate modelling. To go further in the analysis, a few studies also use climatic models

representative of different greenhouse gases emissions scenarios, which allows considering

additional uncertainty (e.g. Holman, 2006; Yusoff et al., 2002). However, in all studies, the

analysis is limited to sources of uncertainty related with the modelling of climatic conditions.

Existing climate change impact studies do not consider uncertainty at other levels, such as the


43

uncertainty linked to hydrological modelling, which could have a significant impact. Holman

(2006) also widens the discussion beyond the strict implication of direct climate impacts, by

identifying a series of possible indirect effects mostly related to socio-economic change in human

societies (e.g. land use change, increase of the water demand, change in agricultural practices,

etc.).


44

2.5 References

Allen, D.M., Mackie, D.C. and Wei, M., 2004. Groundwater and climate change: a sensitivity analysis for the Grand Forks aquifer, southern British Columbia, Canada. Hydrogeology Journal, 12(3): 270-290.

Arnell, N.W., 2003. Relative effects of multi-decadal climatic variability and changes in the mean and variability of climate due to global warming: future streamflows in Britain. Journal of Hydrology, 270(3-4): 195-213.

Batelaan, O. and De Smedt, F., 2001. WetSpass: a flexible, GIS based, distributed recharge methodology for regional groundwater modelling. In: H. Gehrels et al. (Editors), Sixth IAHS Scientific Assembly. Impact of Huyman Activity on Groundwater Dynamics. IAHS, Maastrich (The Netherlands), pp. 11-17.

Brouyère, S., 2001. Etude et modélisation du transport et du piégeage des solutés en milieu souterrain variablement saturé (Study and modelling of transport and retardation of solutes in variably saturated media). PhD Thesis, University of Liège, Liège (Belgium), 640 pp.

Brouyère, S., Carabin, G. and Dassargues, A., 2004b. Climate change impacts on groundwater resources: modelled deficits in a chalky aquifer, Geer basin, Belgium. Hydrogeology Journal, 12: 123-134.

Carabin, G. and Dassargues, A., 1999. Modeling groundwater with ocean and river interaction. Water Resources Research, 35(8): 2347-2358.

Chen, Z., Grasby, S.E. and Osadetz, K.G., 2002. Predicting average annual groundwater levels from climatic variables: an empirical model. Journal of Hydrology, 260: 102-117.

Chen, Z., Grasby, S.E. and Osadetz, K.G., 2004. Relation between climate variability and groundwater levels in the upper carbonate aquifer, southern Manitoba, Canada. Journal of Hydrology, 290: 43-62.

Christensen, J.H., Carter, T.R., Rummukainen, M. and Amanatidis, G., 2007. Evaluating the performance and utility of regional climate models: The PRUDENCE project. Climatic Change, 81(Supplement 1): 1-6.

Christensen, N.S., Wood, A.W., Voisin, N., Lettenmaier, D.P. and Palmer, R.N., 2004. Effects of climate change on the hydrology and water resources of the Colorado river basin. Climatic Change, 62: 337-363.

Collins, W.D. et al., 2004. Description of the NCAR Community Atmosphere Model (CAM3.0). Technical Note TN-464+STR, National Center for Atmospheric Research, Boulder, CO, United States.

Delfs, J.O., Park, C.H. and Kolditz, O., 2009. A sensitivity analysis of Hortonian flow. Advances in Water Resources, 32(9): 1386-1395.


45

Deque, M., Dreveton, C., Braun, A. and Cariolle, D., 1994. The ARPEGE/IFS Atmosphere Model - A contribution to the french community climate modeling. Climate Dynamics, 10(4-5): 249-266.

Dickinson, R.E., Errico, R.M., Giorgi, F. and Bates, G.T., 1989. A regional climate model for the western United States. Climatic Change, 15(3): 383-422.

Eckhardt, K. and Ulbrich, U., 2003. Potential impacts of climate change on groundwater recharge and streamflow in a central European low mountain range. Journal of Hydrology, 284(1-4): 244-252.

Fowler, H.J., Blenkinsop, S. and Tebaldi, C., 2007a. Linking climate change modelling to impacts studies: recent advances in downscaling techniques for hydrological modelling. International Journal of Climatology, 27(12): 1547-1578.

Fowler, H.J., Kilsby, C.G. and O'Connell, P.E., 2003. Modelling the impacts of climatic change and variability on the reliability, resilience and vulnerability of a water resource system, Water Resources Research, pp. 10.

Fowler, H.J., Kilsby, C.G. and Stunell, J., 2007b. Modelling the impacts of projected future climate change on water resources in northwest England. Hydrology and Earth System Sciences, 11(3): 1115-1126.

Gellens, D. and Roulin, E., 1998. Streamflow response of Belgian catchments to IPCC climate change scenarios. Journal of Hydrology, 210: 242-258.

Giorgi, F., 1990. Simulation of regional climate using a limited area model nested in a general-circulation model. Journal of Climate, 3(9): 941-963.

Gordon, C., Cooper, C., Senior, C.A., Banks, H., Gregory, J.M., Johns, T.C., Mitchell, J.F.B. and Wood, R.A., 2000. The simulation of SST, sea ice extents and ocean heat transports in a version of the Hadley Centre coupled model without flux adjustments. Climate Dynamics, 16(2): 147-168.

Graham, D.N. and Butts, M.B., 2006. Flexible, integrated watershed modelling with MIKE SHE. In: V.P. Singh and D.K. Frevert (Editors), Watershed Models. CRC Press.

Grotch, S.L. and MacCracken, M.C., 1991. The Use of General Circulation Models to Predict Regional Climatic Change. Journal of Climate, 4(3): 286-303.

Harbaugh, A.W., 2005. MODFLOW-2005, The U.S. Geological Survey modular ground-water model - The Ground-Water Flow Process. U.S. Geological Survey Techniques and Methods 6-A16, U.S. Geological Survey.

Henriksen, H.J., Troldborg, L., Nyegaard, P., Sonnenborg, T.O., Refsgaard, J.C. and Madsen, B., 2003. Methodology for construction, calibration and validation of a national hydrologic model for Denmark. Journal of Hydrology, 280: 52-71.

Herrera-Pantoja, M. and Hiscock, K.M., 2008. The effects of climate change on potential groundwater recharge in Great Britain. Hydrological processes, 22(1): 73-86.

Holman, I., 2006. Climate change impacts on groundwater recharge- uncertainty, shortcomings, and the way forward? Hydrogeology Journal, 14(5): 637-647.


46

Holman, I.P., Tascone, D. and Hess, T.M., 2009. A comparison of stochastic and deterministic downscaling methods for modelling potential groundwater recharge under climate change in East Anglia, UK: implications for groundwater resource management. Hydrogeology Journal, 17(7): 1629-1641.

IPCC, 2000. IPCC Special Report. Emission Scenarios. Cambridge University Press, Cambridge, United Kingdom, 570 pp.

IPCC, 2001. Climate Change 2001: The Scientific Basis. Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, 881 pp.

IPCC, 2007. Climate Change 2007: The Physical Science Basis. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, 996 pp.

Jiang, T., Chen, Y.D., Xu, C.-y., Chen, X., Chen, X. and Singh, V.P., 2007. Comparison of hydrological impacts of climate change simulated by six hydrological models in the Dongjiang Basin, South China. Journal of Hydrology, 336(3-4): 316-333.

Jones, J.-P., 2005. Simulating Hydrologic Systems Using a Physically-based Surface-Subsurface Model: Issues Concerning Flow, Transport and Parameterization. PhD Thesis, University of Waterloo, Waterloo (Canada), 145 pp.

Jyrkama, M.I. and Sykes, J.F., 2007. The impact of climate change on spatially varying groundwater recharge in the grand river watershed (Ontario). Journal of Hydrology, 338(3-4): 237-250.

Kilsby, C.G., Jones, P.D., Burton, A., Ford, A.C., Fowler, H.J., Harpham, C., James, P., Smith, A. and Wilby, R.L., 2007. A daily weather generator for use in climate change studies. Environmental Modelling & Software, 22(12): 1705-1719.

Kollet, S.J. and Maxwell, R.M., 2006. Integrated surface-groundwater flow modeling: A free-surface overland flow boundary condition in a parallel groundwater flow model. Advances in Water Resources, 29(7): 945-958.

Kollet, S.J. and Maxwell, R.M., 2008. Capturing the influence of groundwater dynamics on land surface processes using an integrated, distributed watershed model. Water Resour. Res., 44.

Li, Q., Unger, A.J.A., Sudicky, E.A., Kassenaar, D., Wexler, E.J. and Shikaze, S., 2008. Simulating the multi-seasonal response of a large-scale watershed with a 3D physically-based hydrologic model. Journal of Hydrology, 357(3-4): 317-336.

Loáiciga, H.A., 2003. Climate Change and Ground Water. Annals of the Association of American Geographers, 93(1): 30-41.

Loaiciga, H.A., Maidment, D.R. and Valdes, J.B., 2000. Climate-change impacts in a regional karst aquifer, Texas, USA. Journal of Hydrology, 267: 173-194.

Martin, G. et al., 2004. Evaluation of the Atmospheric Performance of HadGAM/GEM1, Hadley Center for Climate Change and Research/MET Office, Exeter, UK.


47

McFarlane, N.A., Boer, G.J., Blanchet, J.P. and Lazare, M., 1992. The Canadian Climate Center 2nd-Generation General-Circulation Model and its Equilibrium Climate. Journal of Climate, 5(10): 1013-1044.

Pope, V.D., Gallani, M.L., Rowntree, P.R. and Stratton, R.A., 2000. The impact of new physical parametrizations in the Hadley Centre climate model: HadAM3. Climate Dynamics, 16(2): 123-146.

Prudhomme, C., Reynard, N. and Crooks, S., 2002. Downscaling of global climate models for flood frequency analysis: where are we now? Hydrological processes, 16(6): 1137-1150.

Refsgaard, J.C. and Storm, B., 1995. MIKE SHE. In: V.P. Singh (Editor), Computer Models of Watershed Hydrology. Water Resources Publications, Highlands Ranch, Colorado, pp. pp. 809-846.

Roeckner, E. et al., 1996. The atmospheric general circulation model ECHAM-4: model description and simulation of present-day climate. Report n° 218, Max-Planck Institute for Meteorology, Hamburg, Germany.

Roeckner, E. et al., 2003. The Atmospheric General Circulation Model ECHAM5. Part I: Model Description. MPI Report 349, Max-Planck Institute for Meteorology, Hamburg, Germany.

Schaffranek, R.W., Baltzer, R.A. and Goldberg, D.E., 1981. A model for simulation of flow in singular and interconnected channels. U.S. Geological Survey, Techniques of Water-Resources Investigations, Book 7, Chapter C3, U.S. Geological Survey, USA.

Schroeder, P.R., Dozier, T.S., Zappi, P.A., McEnroe, B.M., Sjostrom, J.W. and Peyton, R.L., 1994. The Hydrologic Evaluation of landfill Performance (HELP) model: Engineering documentation for version 3. Rep. EPA/600/R-94/168b, U.S. Environmental Protection Agency, Washington (D.C., USA).

Scibek, J. and Allen, D.M., 2006a. Comparing modelled responses of two high-permeability, unconfined aquifers to predicted climate change. Global and Planetary Change, 50(1-2): 50-62.

Scibek, J. and Allen, D.M., 2006b. Modeled impacts of predicted climate change on recharge and groundwater levels. Water Resources Research, 42: 18.

Scibek, J., Allen, D.M., Cannon, A.J. and Whitfield, P.H., 2007. Groundwater-surface water interaction under scenarios of climate change using a high-resolution transient groundwater model. Journal of Hydrology, 333(2-4): 165-181.

Semenov, M.A., Brooks, R.J., Barrow, E.M. and Richardson, C.W., 1998. Comparison of the WGEN and LARS-WG stochastic weather generators for diverse climates. Climate Research, 10(2): 95-107.

Serrat-Capdevila, A., Valdés, J.B., Pérez, J.G., Baird, K., Mata, L.J. and Maddock Iii, T., 2007. Modeling climate change impacts - and uncertainty - on the hydrology of a riparian system: The San Pedro Basin (Arizona/Sonora). Journal of Hydrology, 347(1-2): 48-66.


48

Sonnenborg, T.O., Christensen, B.S.B., Nyegaard, P., Henriksen, H.J. and Refsgaard, J.C., 2003. Transient modeling of regional groundwater flow using parameter estimates from steady-state automatic calibration. Journal of Hydrology, 273(1-4): 188-204.

Sudicky, E.A., Jones, J.-P., Park, Y.-J., Brookfield, E.A. and Colautti, D., 2008. Simulating complex flow and transport dynamics in an integrated surface-subsurface modeling framework. Geosciences Journal, 12(2): 107-122.

Therrien, R., McLaren, R.G., Sudicky, E.A. and Panday, S.M., 2005. HydroGeoSphere. A three-dimensional numerical model describing fully-integrated subsurface and surface flow and solute transport, 343 pp.

Van Roosmalen, L., Christensen, B.S.B. and Sonnenborg, T.O., 2007. Regional Differences in Climate Change Impacts on Groundwater and Stream Discharge in Denmark. Vadose Zone J, 6(3): 554-571.

Van Roosmalen, L., Sonnenborg, T.O. and Jensen, K.H., 2009. Impact of climate and land use change on the hydrology of a large-scale agricultural catchment. Water Resources Research, 45.

VanderKwaak, 1999. Numerical Simulation of Flow and Chemical Transport in Integrated Surface-Subsurface Hydrologic Systems, University of Waterloo, Waterloo, Ontario, Canada, 242 pp.

VanRheenen, N.T., Wood, A.W., Palmer, R.N. and Lettenmaier, D.P., 2004. Potential Implications of PCM Climate Change Scenarios for Sacramento–San Joaquin River Basin Hydrology and Water Resources. Climatic Change, 62(1): 257-281.

Watts, M., Goodess, C.M. and Jones, P.D., 2004. Validation of the CRU daily weather generator. BETWIXT Technical Briefing Note 4, Climatic Research Unit, University of East Anglia.

Wilby, R.L., Whitehead, P.G., Wade, A.J., Butterfield, D., Davis, R.J. and Watts, G., 2006. Integrated modelling of climate change impacts on water resources and quality in a lowland catchment : River Kennet, UK. Journal of Hydrology, 330(1-2): 204-220.

Wilby, R.L. and Wigley, T.M.L., 1997. Downscaling general circulation model output: a review of methods and limitations. Progress in Physical Geography, 21(4): 530-548.

Wilks, D.S. and Wilby, R.L., 1999. The weather generation game: a review of stochastic weather models. Progress in Physical Geography, 23(3): 329-357.

Woldeamlak, S., Batelaan, O. and De Smedt, F., 2007. Effects of climate change on the groundwater system in the Grote-Nete catchment, Belgium. Hydrogeology Journal, 15(5): 891-901.

Yan, J.S. and Smith, K.R., 1994. Simulation of integrated surface-water and ground-water systems - Model formulation. Water Resources Bulletin, 30(5): 879-890.

Yusoff, I., Hiscock, K.M. and Conway, D., 2002. Simulation of the impacts of climate change on groundwater resources in eastern England. Geological Society, London, Special Publications, 193(1): 325-344.

3. Methodology

49

3. METHODOLOGY

3. Methodology

50

3. Methodology

51

3.1 Methodology

The objective of this thesis is to provide improved methods for the estimation of the impacts of

climate change on groundwater reserves, by developing an innovative modelling approach that

alleviates the simplifying assumptions presented in Section 2 through the review of the scientific

literature. This new approach includes providing improvements and innovations at three levels:

1. Hydrological modelling.

2. Use of climate change modelling and downscaling techniques.

3. Methods that allow evaluating uncertainties of various types.

1. To demonstrate the approach, the numerical model 'HydroGeoSphere' (Therrien et al., 2005)

has been used to develop catchment-scale simulation of coupled surface and subsurface water

flow in the Geer basin located in the Walloon Region of Belgium. 'HydroGeoSphere', which

enables to perform physically-based and spatially-distributed simulations, provides a realistic

representation of the system, compared to simplified models that are inadequate if the water

fluxes extrapolated in the climate change scenarios and imposed to the hydrologic system are not

included in the intervals of values used in the calibration procedure. The model developed in this

study fully integrates surface- and subsurface- flow in the saturated and partially saturated zones,

with a simultaneous solution of the flow equations in all domains using finite elements. As

already pointed out in Section 2, this simultaneous solution enables a better representation of the

whole system because water flow in one domain is interconnected with flow in the other

domains. Water exchange between the surface and subsurface is calculated internally at each time

step. Similarly, the actual evapotranspiration is calculated internally as a function of the soil

moisture at each node of the defined evaporative zone and at each time step. Integrating

evapotranspiration, surface, and subsurface flow calculations in the same model does not only

increase the complexity of the model, which would not guarantee more robust predictions (Ebel

3. Methodology

52

and Loague, 2006), but also increases the number of observed data available for calibration.

Because both surface and subsurface data are used for calibration, parameter values are better

constrained, and the uncertainty in the estimation of some components of the global water

balance is reduced. In particular, recharge and surface water – groundwater interactions are better

represented, which is crucial in the context of a climate change impact study on groundwater

reserves.

2. The climate change scenarios, to be applied as input of the Geer basin hydrological model,

have been generated from state-of-the-art RCMs simulations, using two different downscaling

techniques: the 'quantile mapping bias-correction' technique (Wood et al., 2004) and the 'CRU

weather generator' (Kilsby et al., 2007; Watts et al., 2004). These two downscaling techniques

have the great advantage, compared to other downscaling techniques usually used in hydrological

studies, that they not only apply a correction to the mean of climatic variables, but also across the

statistical distributions of these variables. This seems crucial to achieve a reliable climate change

impact study as the distribution of precipitation, for example, is expected to change in the future,

with more intense rainfall events separated by longer dry periods. This change in distribution may

actually have a significant impact on the dynamic of runoff and groundwater infiltration. The

'quantile mapping bias-correction' technique is used to generate climatic variables time series

representative of a stationary climate for the periods 2011-2040, 2041-2070 and 2071-2100. The

'CRU' weather generator is used to generate a large number of equiprobable transient climate

change scenarios between 2010 and 2085. All these downscaled scenarios are then applied as

input of the Geer basin hydrological model to evaluate climate change impact on groundwater

resources.

3. Uncertainty is evaluated regarding different possible sources, at different levels between the

modelling of the climate at the scale of the earth globe and the modelling of hydrological

conditions at the scale of the catchment.

3. Methodology

53

- First, climate change scenarios are downscaled from various RCMs, with boundary conditions

from different GCMs. Using data from several climate models allows evaluating the

uncertainty related to the modelling of atmospheric conditions at the scale of the earth globe

(GCMs) and at the regional scale (RCMs).

- Second, two different methods (the 'quantile mapping bias-correction' and 'CRU weather

generator' techniques) are used to downscale climatic data from large to catchment scale.

Using both types of downscaled climatic scenarios as input of the Geer basin model will

enable to highlight possible differences due to the downscaling method.

- Third, large numbers of equiprobable climate change scenarios from 2010 to 2085 are

generated using the 'CRU weather generator' technique and applied as input of the Geer basin

hydrological model. The objective is to estimate the uncertainty of projected groundwater

levels, in relation to the natural variability of the climate. Because the number of climate

change scenarios is large, statistical tools can then be used to calculate, for each specific time

between 2010 and 2085, the variance of simulated groundwater levels and associated

confidence intervals.

- The previous steps described above relate to the uncertainty linked to the modelling of

climatic conditions. However, additional source of uncertainty is also linked to the use to the

Geer basin hydrological model. Numerical models are actually simplifications of the reality.

No calibration is perfect and it has some implications on the accuracy of models predictions.

In this study, the computer code 'UCODE_2005' (Hill and Tiedeman, 2007; Poeter et al.,

2005) is used to evaluate this kind of uncertainty linked to the model calibration.

UCODE_2005 has been developed by the 'U.S. Geological Survey' (USGS), and can be

coupled with other modelling codes to perform sensitivity analysis, data needs assessment,

automatic inverse calibration, prediction and uncertainty analysis. In the context of this study,

3. Methodology

54

linear confidence intervals are calculated for predicted groundwater levels under climate

change conditions. To corroborate the analysis, climate change impact on groundwater levels

are also evaluated using two Geer basin hydrological models with unequal calibrations.

Discrepancies are then examined and compared with the linear confidence intervals calculated

with 'UCODE_2005'.

The implementation of the integrated surface – subsurface model with 'HydroGeoSphere' is

presented with many details in Section 5. An important part of the section is also devoted to the

evaluation of the model calibration. Section 6 describes the used climate models (GCMs and

RCMs), the 'quantile mapping bias-correction' downscaling technique, and the application of the

downscaled scenarios as input of the Geer basin model. Section 7 describes the 'CRU weather

generator', the application of stochastic climate change scenarios on the Geer basin model, and

subsequent interpretations of the results. The first part of Section 8 presents the coupling of

'UCODE_2005' with the Geer basin model to estimate the uncertainty linked to the calibration

of the hydrological model. The second part of Section 8 summarises and discusses all results

about impact uncertainty.

3. Methodology

55

3.2 References

Ebel, B.A. and Loague, K., 2006. Physics-based hydrologic-response simulation: Seeing through the fog of equifinality. Hydrological processes, 20(13): 2887-2900.

Hill, M.C. and Tiedeman, C.R., 2007. Effective groundwater model calibration. With Analysis of data, sensitivities, predictions and uncertainty. John Wiley & Sons, New Jersey, 455 pp.


Poeter, E.P., Hill, M.C., Banta, E.R., Mehl, S. and Christensen, S., 2005. UCODE_2005 and Six Other Computer Codes for Universal Sensitivity Analysis, Calibration, and Uncertainty Evaluation: U.S. Geological Survey Techniques and Methods 6-A11, 283 pp.



Wood, A.W., Leung, L.R., Sridhar, V. and Lettenmaier, D.P., 2004. Hydrologic Implications of Dynamical and Statistical Approaches to Downscaling Climate Model Outputs. Climatic Change, 62(1): 189-216.

3. Methodology

56

4. The Geer basin

57

4. THE GEER BASIN

4. The Geer basin

58

4. The Geer basin

59

4.1 Geographical and geological contexts

The Geer sub-catchment is located in eastern Belgium, north-west of the city of Liège, in the

intensively cultivated 'Hesbaye' region. The hydrological basin extends over approximately 480

km², on the left bank of the Meuse River (Figure 4.1).

Figure 4.1. Location of the Geer basin and hydrologic limits.

The geology of the Geer basin essentially consists of Cretaceous chalky formations that dip

northward and that are bounded at their base by 10 metres of smectite clays of very low hydraulic

conductivity (Figure 4.2). The chalk formation consists of a series of chalk layers, whose

thicknesses range from a few meters up to 70 m. This formation can be divided into two main

groups, the Campanian lower chalks and the Maastrichtian upper chalks. A flint conglomerate of

dissolved chalk residues overlies the chalk, with a maximum thickness of 10 m. Tertiary sand

lenses of small extension are found locally above this conglomerate and a thick layer (up to 20 m)

of Quaternary loess is observed throughout the catchment. Tertiary sands and clays entirely cover

the chalk formations north of the Geer River (Figure 4.2) (Hallet, 1998; Orban et al., 2006).

4. The Geer basin

60

Figure 4.2. Geological cross-section in the Hesbaye aquifer (modified from (modified from Brouyère et al.,

2004b)), with a vertical exaggeration equal to 40.

4.2 Hydrogeological context

The main aquifer in the region is the ‘Hesbaye’ aquifer, which corresponds to the chalk layers and

is unconfined over most of the basin. Subsurface flow is from south to north and the aquifer is

mainly drained by the Geer River that flows from west to east, as shown at Figure 4.3 (Orban,

2009; Orban et al., 2006; Ruthy, 2009). The chalk porous matrix, whose total porosity is

estimated equal to 44%, enables the storage of large quantities of groundwater, while fast

preferential flow occurs through fractures, which represent approximately 1% of the total

porosity (Brouyère, 2001; Hallet, 1998). At a macroscopic scale, the hydraulic properties of the

chalk formations vary vertically and laterally. The lower Campanian chalks are usually less

permeable than the upper Maastrichtian chalks. Laterally, zones of higher hydraulic conductivity

are observed and associated with 'dry valleys', mostly oriented south to north. These zones,

characterized by a higher degree of fracturing, are associated with a slight lowering of hydraulic

4. The Geer basin

61

heads. For the larger part of the Geer catchment, the saturated zone is exclusively located in the

chalk formations. The thick loess layer located above the chalk controls the water infiltration rate

from the land surface to the chalky aquifer, resulting in smoothed recharge fluxes at the

groundwater table and attenuation of seasonal fluctuations of hydraulic heads that are better

characterised by multi-annual variations (Brouyère et al., 2004a). As an illustration, Figure 4.4

shows the evolution of groundwater levels at the observation well 'VIE004' (Figure 4.1) from

1950 to 2006. In the northern part of the catchment, near the Geer River, water levels are closer

to the ground surface and semi-confined conditions may prevail because of the loess Quaternary

deposits. North of the Geer River, Tertiary deposits become thicker and contain some clearly

clayey layers. These layers are responsible for the confined nature of the chalky aquifer at this

location (Figure 4.2).

Figure 4.3. Piezometry of the chalk aquifer (in metres above sea level) based on 2088 data (modified from

Orban, 2009; Ruthy, 2009).

The ‘Hesbaye’ aquifer is largely exploited for drinking water, primarily through a network of

pumping galleries of more than 40 km that is located in the saturated chalk formation (Figure

4. The Geer basin

62

4.1). Most of this water is used to supply the city of Liège and its suburbs. A part of drained

water is transported by gravity, but some pumping is also carried out directly in the galleries to

provide drinking water to local communities or to lift up water into more elevated aqueducts.

The draining galleries can also be "switched on" or "switched off" in function of the water needs

or nitrate content (Hodiaumont et al., 1999; Orban, 2009). According to Hallet (1998), extracted

groundwater volumes represent between 6% and 11% of annual precipitations. The groundwater

budget (Hallet, 1998) also indicates groundwater losses mostly through the northern catchment

boundary, and partly resulting from groundwater extraction in the Flemish region of Belgium

located directly north of the Geer basin. According to (Hallet, 1998), the water budget of the

Geer basin for the period 1975-1994, is distributed in 63% of evapotranspiration, 18% of river

discharge at the outlet of the catchment, 11% of groundwater abstraction and 7% of groundwater

losses through the northern boundary.

The Hesbaye aquifer suffers from severe nitrate contamination problems, due to intensive

agricultural activities. In the Geer basin, 77% of the area is actually devoted to crops, mainly for

the cultivation of cereals and sugar beets. In many locations in the unconfined part of the aquifer,

nitrate concentrations are frequently over 45 mg/L, approaching the drinking water limit of 50

mg/L (Batlle-Aguilar et al., 2007; Hallet, 1998).

4. The Geer basin

63

Figure 4.4. Evolution of groundwater levels at the observation well 'VIE044', from 1950 to 2006.

4. The Geer basin

64

4.3 References

Batlle-Aguilar, J., Orban, P., Dassargues, A. and Brouyère, S., 2007. Identification of groundwater quality trends in a chalk aquifer threatened by intensive agriculture in Belgium. Hydrogeology Journal, 15(8): 1615-1627.


Brouyère, S., Carabin, G. and Dassargues, A., 2004a. Climate change impacts on groundwater resources: modelled deficits in a chalky aquifer, Geer basin, Belgium. Hydrogeology Journal, 12: 123-134.

Brouyère, S., Dassargues, A. and Hallet, V., 2004b. Migration of contaminants through the unsaturated zone overlying the Hesbaye chalky aquifer in Belgium: a field investigation. Journal of Contaminant Hydrology, 72(1-4): 135-164.

Hallet, V., 1998. Etude de la contamination de la nappe aquifère de Hesbaye par les nitrates: hydrogéologie, hydrochimie et modélisation mathématique des écoulements et du transport en milieu saturé (Contamination of the Hesbaye aquifer by nitrates: hydrogeology, hydrochemistry and mathematical modeling). PhD Thesis, University of Liège, Liège (Belgium), 361 pp.

Hodiaumont, A., Cantillana, R. and Compère, J.M., 1999. Les eaux souterraines de la CILE : Contexte, captage et qualité. Tribune de l'eau, 52(600-601): 31-50.

Orban, P., 2009. Solute transport modelling at the groundwater body scale: Nitrate trends assessment in the Geer basin (Belgium). PhD Thesis, University of Liège, Faculty of Applied Sciences, Liège, 219 pp.

Orban, P., Batlle-Aguilar, J., Goderniaux, P., Dassargues, A. and Brouyère, S., 2006. Description of hydrogeological conditions in the Geer sub-catchment and synthesis of available data for groundwater modelling. Deliverable R3.16, AquaTerra (Integrated Project FP6 no 505428).

Ruthy, I., 2009. Carte hydrogéologique de Wallonie et notice explicative, Tongeren-Herderen 34/5-6. Service public de Wallonie, DGARNE, Namur.

5. Modelling

65

5. MODELLING

5. Modelling

66

5. Modelling

67

5.1 Hydrological modelling

5.1.1 Conceptual model

The Geer hydrological catchment defines the boundaries of the modelled area (Figure 4.1). The

smectite clay (Figure 4.2) is considered impervious and the contact between the clay and the

chalk represents the lower boundary of the model. The western, southern and eastern boundaries

correspond to surface water divides and it is assumed that there is no water exchange across these

boundaries for either surface or subsurface flow. On the other hand, groundwater fluxes through

the northern boundary must be taken into account. Along this border, hydrogeological and

hydrographical limits differ, and groundwater flows northwards towards the adjacent basin.

The Geer River at the level of the 'Kanne' gauging station, located 4 kilometres upstream from

the confluence with the Meuse River, is considered as the outlet of the catchment (Figure 4.1).

Surface water exchanges are not observed elsewhere along the model boundaries, since they

correspond to topographical limits.

Pumping wells operated by water supply companies or farmers are distributed over the whole

basin but water collected through the network of draining galleries is the largest component of

the total of groundwater abstraction in the Geer basin.

5.1.2 Mathematical and numerical model

The Geer basin hydrological model has been developed with the HydroGeoSphere finite element

model (Therrien et al., 2005). The spatially-distributed model simulates fully coupled 3D variably

saturated groundwater flow in granular or fractured aquifers and 2D overland flow, as well as

solute transport in the surface and subsurface domains. HydroGeoSphere simulates the dynamic

interactions between all sub-domains at each time step. It partitions rainfall into components

such as evapotranspiration, runoff and infiltration. The model also allows the calculation of water

5. Modelling

68

infiltration or exfiltration between rivers and aquifers. These interactions are of great interest in

the context of climate change as recharge is very sensitive to climatic variations and represent

crucial elements for impacts projections.

K Full Saturated hydraulic conductivity [L.T-1]

n Total porosity [-]

Ss Specific storage [L-1]

α Van van Genuchten parameter [-]

β Van van Genuchten parameter [L-1] Subs

urfa

ce d

omain

Swr Residual water saturation [-]

Lc Coupling length [L]

nx Manning roughness coefficient [L-1/3T]

Surf

ace

dom

ain

ny Manning roughness coefficient [L-1/3T]

Le Evaporation depth [L]

θe1 θe2 Evaporation limiting saturations [-]

LAI Leaf Area Index [-]

Lr Root depth [L]

C1, C2, C3 Transpiration fitting parameters [-]

θt1 θt2 Transpiration limiting saturations [-]

Eva

potra

npisr

atio

n

Cint Canopy storage parameter [L]

Table 5.1. Parameters used in the flow model

HydroGeoSphere uses the control volume finite element approach to simultaneously solve

Richards' equation describing 3D variably-saturated subsurface flow and a 2D depth-averaged

surface flow equation, which is the diffusion-wave approximation of the Saint Venant equation.

In the subsurface domain, the hydraulic head, the degree of saturation, and the water Darcy flux

are calculated at each node in the grid. In the surface domain, water depth (≈ height of water

above ground surface) and fluid flux are calculated for each node of the 2D grid. The stream

5. Modelling

69

locations can be implicitly retrieved by considering the surface nodes where the water depth is

greater than zero. Transport processes include advection, dispersion, retardation and decay.

Newton-Raphson iterations are used for solving non-linear equations. More information on the

model and equations solved is available in Therrien et al. (2005) and in Li et al (2008).

Hydrologic parameters required for the fully-coupled simulation are listed in Table 5.1 along with

their domain of application. It should be noted that fractures are not represented explicitly in the

Geer basin model, and equivalent porous media properties are assigned to the elements

representing the aquifer.

The Geer basin model uses a ‘dual-node approach’ to calculate water exchanges between the

surface and subsurface domains. In this approach, surface nodes have to coincide with nodes of

the subsurface grid topmost layer. Water flux between each corresponding surface and subsurface

nodes is calculated as the hydraulic head difference between the two domains multiplied by a

leakage factor (coupling length – Le [L]) characterising the properties of the soil. In

HydroGeoSphere, the model of Kristensen and Jensen (1975) is used to calculate the actual

transpiration Tp [LT-1] and evaporation Es [LT-1] as a function of the potential evapotranspiration

Ep [LT-1], the soil moisture at each node belonging to the specified evaporative and root zones,

and the 'Leaf Area Index' (LAI [-]) that represents the cover of leaves over a unit area (Equation

5.1 to 5.6) (Therrien et al., 2005). Equation 5.2 expresses the vegetation term, as a function of

LAI, and parameters C1 and C2. Full transpiration can occur if water saturation θ [-] is higher

than θt1 and there is no transpiration if water saturation is lower than θt2. Between these two

limiting saturations, transpiration decreases following a law governed by the parameters C3

(Equations 5.1 and 5.3). RDF(Lr) is the 'Root Distribution Function' that distributes the water

extracted from the root zone, along the root depth Lr [L], following a quadratic law. The quantity

of extracted water is more important near the surface and decreases with depth until zero at the

root depth Lr. The 'canopy evaporation Ecan [LT-1] corresponds to the evaporation of water

5. Modelling

70

intercepted by the canopy. Full evaporation can occur if water saturation is higher than θe1 and

there is no evaporation if water saturation is lower than θe2. Between these two limits, evaporation

decreases following a law governed by parameter C3 (Equations 5.4 and 5.5). EDF(Le) is the

'Evaporation Distribution Function' that distributes the water extracted from the evaporative

zone, along the evaporation depth (Le), following a quadratic law. The interception of

precipitation by the canopy is simulated by the bucket model, where precipitation in excess of

interception storage and evapotranspiration reaches the ground surface. The 'interception storage

capacity' maxintS [L] represents the maximum quantity of water that can be intercepted by the

canopy. It depends on LAI and the 'canopy storage parameter' cint [L] (Equation 5.6).

( ) [ ]canprp EELRDFfLAIfT −×××= )()( 21 θ (5.1)

( )[ ]{ }LAICCLAIf ×+= 121 ,1min,0max)( (5.2)

⎪⎪⎩

⎪⎪⎨

⎧

≤

≤≤⎥⎦

⎤⎢⎣

⎡−−

−

≤≤

=

θθ

θθθθθθθ

θθ

1

1221

1

2

2

1

1

003

t

tt

C

tt

t

t

for

for

for

f (5.3)

( ) [ ] ( )ecanps LEDFLAIfEEE ×−×−×= )(1* 1α (5.4)

⎪⎪⎩

⎪⎪⎨

⎧

>

≤≤−−

<

=

1

1221

2

2

1

0

*

e

eeee

e

e

for

for

for

θθ

θθθθθθθ

θθ

α (5.5)

LAIcS ×= intmaxint (5.6)

5. Modelling

71

5.1.3 Discretisation

A three-dimensional finite element mesh, composed of several layers of 6-node triangular

prismatic elements (Figure 5.1), was generated based on the conceptual model presented

previously. The elements have lateral dimensions equal to approximately 500 m. The top and

bottom layers of nodes represent the soil surface and the contact between smectite clay and

chalk, respectively. Subsurface formations are discretised using 11 finite element layers. Five

layers are used for the first five meters below the ground surface, with each layer having a

thickness of one meter. The finer vertical discretisation near ground surface represents more

accurately river – aquifer interactions as well as recharge processes at the interface between the

surface and subsurface domains. In particular, distributing several nodes vertically within the first

few meters below the ground surface enables the variation of evaporative and root depths, as well

as the vertical distribution of evapotranspiration rates, according to the land use and soil type.

The remaining lower six finite element layers are uniformly distributed vertically between the fifth

and bottom layers. A material is assigned to each 3D finite element based on data from more

than 120 boreholes distributed throughout the catchment. The ground surface is discretised using

one layer of 2D finite elements (Figure 5.1). The elevation of the surface nodes are calculated

using the Geer basin DTM (Digital Terrain Model), whose pixels have dimensions equal to

30 × 30 m. In the dual node approach, the nodes forming the surface domain correspond to the

node of the top layer of the subsurface domain. The total number of nodes for the subsurface

and surface domains is equal to 9420 and 785, respectively.

No-flow boundaries are applied to subsurface nodes belonging to the western, southern, eastern

and bottom boundaries. Cauchy conditions (head dependent flux) are applied on the subsurface

nodes along the northern boundary to take into account groundwater losses in the direction of

the adjacent catchment located northward from the Geer basin. For the surface flow domain, no-

flow Neumann boundary conditions are prescribed along the hydrographical limits of the Geer

5. Modelling

72

basin. Critical-depth boundary conditions are prescribed at the nodes corresponding to the

catchment outlet, at the level of the 'Kanne' gauging station. A critical-depth boundary condition

forces the water elevation at the boundary to be equal to the 'critical depth'. The 'critical depth' is

the water elevation for which the energy of the flowing water relatively to the stream bottom is

minimum (Hornberger et al., 1998; Therrien et al., 2005).

Figure 5.1. Spatial discretisation of the Geer basin

5.1.4 Specified Fluxes

Specified hydrological fluxes within the Geer catchment consist of precipitation,

evapotranspiration and groundwater abstraction by draining galleries and pumping wells.

5. Modelling

73

Historical climatic data are available for several weather stations located inside or near the Geer

basin1 (more details in Orban et al., 2006). The stations shown in Figure 4.1 have complete

precipitation (P) time series from 1960 to 2005. Temperature (T) and potential

evapotranspiration (PET) data, for the same time period, are available for the Bierset station only.

Data from these weather stations are used as inputs to the model and are applied on the surface

node layer as transient specified fluxes. Precipitation data from each station are distributed using

Thiessen polygons. Potential evapotranspiration data available only for the Bierset station are

assumed to be applicable to the whole catchment.

Extracted groundwater volumes, from the draining galleries and from the most important

production wells (Figure 4.1), have been collected by the Walloon administration and are updated

annually (Orban et al., 2006). Transient volumetric flow rates are prescribed at each node of the

3D grid corresponding to the draining galleries or the pumping wells locations.

5.1.5 Calibration procedure

The model was calibrated to observed hydraulic heads and surface flow rates during the period

1967-2003. A preliminary calibration was performed in steady state conditions, using the mean

data of the hydrologic year 1967-1968, and the results were used as initial conditions for the

transient simulations. Calibration further showed that inaccuracies in these initial conditions only

affect the simulation results on a short-term basis. Even with initial conditions very different

from reality, such as a fully saturated subsurface domain, induced differences are reduced within a

few days for surface water flow rates and within 2 years for groundwater hydraulic heads. The

transient flow model is calibrated to surface flow rates measured at the 'Kanne' gauging station

1 Historical climatic data for the Geer catchment were obtained from the Royal Meteorological Institute of Belgium (RMI).

5. Modelling

74

located on the Geer river at the catchment outlet, and to hydraulic heads from 8 observation

wells selected according to their location and the availability of measured hydraulic heads during

the calibration period (Figure 4.1). Specified fluxes are input on a daily basis, using total daily

precipitation, evapotranspiration and groundwater abstraction rates. Adaptive time-stepping is

used so that groundwater hydraulic heads and surface water elevations do not vary by more than

0.5 m and 0.01 m, respectively, during one time step. For the Geer basin model, time steps

commonly vary between a few minutes to a few hours.

5.1.5.1 Parameter values

In the subsurface domain, the Van Genuchten parameters are prescribed according to Brouyère

(2001) and Brouyère et al. (2004a). Table 5.2 summarizes the values used for the chalk and loess

formations. Saturated hydraulic conductivities are adjusted during calibration, taking into account

the extension of the geological units and the zones of higher hydraulic conductivity associated

with 'dry valleys'. The chalky aquifer is also vertically divided into 3 zones, namely 'upper chalk',

'intermediate chalk' and 'lower chalk'. This enables the represention of the decrease of saturated

hydraulic conductivity with depth. Adjusted values are also kept within ranges provided by the

measurements from laboratory and field tests conducted in the geologic formations of the Geer

basin, and by ranges of hydraulic conductivity values given by Hallet (1998), Brouyère (2001),

Brouyère et al. (2004a), and Dassargues and Monjoie (1993) for the Geer basin formations. Table

5.3 and Figure 5.2 summarize all saturated hydraulic conductivity values at the end of the

calibration.

Van Genuchten parameters Residual water saturation Total porosity Specific storage

α [m-1] β [-] Swr [-] n [-] Ss [m-1]

Chalk formations 0.099 1.10 0.023 0.44 1×10-4

Loess formations 7.57 1.16 0.024 0.41 1×10-4

Table 5.2. Van Genuchten parameters, total porosity and specific storage

5. Modelling

75

Name K [m.s-1]

Chalk 1 4×10-5 Chalk 2 1×10-3 Chalk 3 3×10-5 Chalk 4 2×10-6 Chalk 5 5×10-6 Lo

wer

cha

lk

Chalk – Dry valleys 2×10-4

Chalk 1 1×10-4 Chalk 2 1×10-3 Chalk 3 3×10-5 Chalk 4 1×10-4 Chalk 5 5×10-6

Inte

rmed

iate

chalk


Chalk 1 1×10-4 - 4×10-4 Chalk 2 1×10-3 Chalk 3 1×10-4 Chalk 4 1×10-4 Chalk 5 1×10-4 U

pper

cha

lk

Chalk – Dry valleys 2×10-4 - 5×10-4

Quaternary loess 1×10-8

Tertiary deposits 0.3×10-7 - 1×10-7

Table 5.3. Full saturated hydraulic conductivities values of the calibrated zones (results of calibration)

Figure 5.2. Distribution of the hydraulic conductivity zones for the chalk finite elements layers (results of

calibration)

5. Modelling

76

In the surface domain, the coupling length and the friction coefficients were adjusted according

to the soil2 and land use3 maps, respectively. The soil mean characteristics and thicknesses are

quite homogeneous at the scale of a 2D surface element in this model, since these characteristics

vary at a much smaller scale. The coupling length was therefore assumed constant everywhere

and equal to 0.01 m. Calibration later showed that the results were insensitive to the value of the

coupling length. Three categories of land-use, namely ‘rural’, ‘urban’ and ‘forested’, have been

identified and Manning's roughness coefficients were initially defined for each category. The

values of Manning's roughness coefficients obtained at the end of the calibration (Table 5.4) vary

between 0.03 and 0.6 [L-1/3T] which is consistent with values more commonly used in

hydrological models (Hornberger et al., 1998; Jones, 2005; Li et al., 2008).

nx - ny [s.m-1/3] Lc [m]

Rural 0.4 0.01

Urban 0.03 0.01

Forested 0.6 0.01

Table 5.4. Values for the Manning roughness coefficients and coupling length

The parameters used to calculate the actual evapotranspiration (Kristensen and Jensen, 1975)

were defined using values found in the scientific literature and are summarised in Table 5.5 for

four land-use categories (rural crop, rural grassland, rural broadleaf deciduous forested, urban).

Root depths range between 0 m and 5.2 m, according to values given by Canadell et al. (1996). A

uniform evaporation depth value of 2 m is assumed over the whole catchment. Values for the

maximum Leaf Area Index (LAI) are given by Scurlock et al. (2001), Asner et al. (2003), Vasquez

and Feyen (2003) and Li et al. (2008). Breuer et al. (2003) give maximum and minimum values of

2 © Direction Générale de l’Agriculture (Ministère de la Région Wallonne). Projet de Cartographie Numérique des sols de Wallonie (PCNSW). Projet du Gouvernement Wallon (GW VIII/2007/Doc.58.12/12.07/B.L & GW VII/2000/Doc.1331/07.12/JH.)

3 European Environment Agency (http://www.eea.europa.eu). Corine Land Cover Project. Copyright EEA, Copenhagen, 2007.

5. Modelling

77

the LAI throughout the year. For the Geer basin model, maximum LAI varies from 0.40 to 5.12.

In absence of information about minimum LAI for the vegetation of the Geer basin, LAI is

arbitrarily reduced by 50 % during the winter months. However, the results are insensitive to the

value of Min. LAI given that evapotranspiration is very low during winter months anyway. Values

for the empirical transpiration fitting parameters C1, C2 and C3, as well as for the canopy storage

interception Cint can be found in Kristensen and Jensen (1975) and Li et al. (2008). Used values of

C1, C2, C3 and Cint are equal to 0.3, 0.2, 10 and 1×10-5 m, respectively. The limiting saturations,

corresponding to the wilting point and field capacity, are specified as the saturations

corresponding to pF values4 equal to 4.2 and 2.5, as found in Brouyère (2001).

Rural crop (temperate)

Rural grassland (temperate)

Rural broadleaf deciduous forested

(temperate) Urban

Root depth Lr [m] 2.1 2.6 5.2 0.0

Evaporation depth Le [m] 2.0

Max. LAI [-] 4.22 2.50 5.12 0.40

Cint [m] 1×10-5

C1 [-] 0.3

C2 [-] 0.2

C3 [-] 10

Table 5.5. Root depths, evaporation depths and Leaf Area Index

4 pF=log(-hydraulic pressure)

5. Modelling

78

5.1.5.2 Evaluation of the calibration

Results of the steady state and transient simulations, using the calibrated parameters, are shown

in Figure 5.3 to Figure 5.6. Figure 5.3A presents the computed steady-state subsurface saturations

for the hydrological year 1967-1968. Similarly, Figure 5.3B shows the computed steady-state

water elevation at each node of the surface domain. The locations of the Yerne and the Geer

Rivers are clearly seen and correspond to the highest water elevations. The three graphs of Figure

5.4 present, in different ways, the computed and observed values following the methodology and

guidelines proposed by Hill & Tiedeman (2007) for a global assessment of the calibration. Figure

5.5 presents the measured and simulated transient hydraulic heads for the 8 selected observation

wells. Figure 5.6 presents the observed and simulated transient flow rates for the ‘Kanne’ gauging

station located at the outlet of the basin. Table 5.6 shows the mean absolute residuals, the mean

residuals and the weighted residuals for each observation point.

Data presented at Figure 5.4 consists in monthly mean hydraulic heads and flow rates. In this

first graph, monthly data are used because they are more representative of the general

fluctuations of groundwater reserves, which are the focus of this study. Daily data, and especially

daily surface flow rates, are however also presented in the next figures. Figure 5.4A shows

computed monthly hydraulic heads and flow rates along a line with slope equal to 1. This way of

presenting calibration results, commonly used by modellers, is not adequate because the large

ranges of simulated values tend to hide the potential calibration errors. Additionally, units and

order of magnitude can also differ if different kinds of observation data are used in the

calibration procedure. An alternative is to plot the residuals instead of computed values (Figure

5.4B) but it does not solve the problem of potentially different units. Hill & Tiedeman (2007)

propose to plot weighted residuals to assess the quality of the calibration and to evaluate some

important properties (Figure 5.4C). For each observation, weighted residuals are calculated

following Equation 5.1.

5. Modelling

79

Figure 5.3. (A) Computed steady-state surface water elevations. (B) Computed steady-state subsurface

saturation, with full saturation shown in red (1967-1968).

5. Modelling

80

Figure 5.4. Graphical analysis of the model calibration. (A) Computed values vs. observed values. (B)

Residuals vs. observed values. (C) Weighted residuals vs. observed values. Doted lines represent increments of the calculated standard error.

5. Modelling

81

( )( ) ( )( ) ( )( )byybyybyyw iii

iii

iii '1'1'2

1

22

1−×=−×⎟⎟

⎠

⎞⎜⎜⎝

⎛=−×

σσ (5.1)

iw : Weight for observation i

b : Vector of parameters

iy : Observed value i

iy' : Computed value i as a function of ( )b

2iσ : Variance of the true measurement errors of observation i

Using the inverse of the true measurement errors variance for weighting the residuals offers three

main advantages. First, weighted residuals are non dimensional and can be compared more easily.

Second, the weights enable to integrate in the analysis the uncertainty linked to the observations.

According to Equation 5.1, the weights associated with uncertain or inaccurately measured

observations will be less important and will then reduce the influence of the associated residual.

Third, this weighting scheme can be used in the calculation of the objective function when

performing a parameter optimisation. It allows calculating and interpreting very useful additional

statistics for sensitivity or prediction analysis, as used in Section 5.2 and Section 8.2. More details

in Hill & Tiedeman (2007). For the Geer basin model, the weights of the groundwater level

observations have been uniformly set to 3.84, considering the lack of information or particular

study about the true error of each observation. This weight value is based on the assumption that

the 95% confidence interval including the true groundwater level is within 1 metre around the

measured groundwater level. This confidence interval considers all types of possible errors:

measurement of the well top elevation, measurement of the groundwater depth, errors linked to

the precision of the measurement devices, encoding errors. Assuming that measurement errors

are normally distributed, the 95% confidence interval can easily be related to the variance. The

weights associated with the surface water flow rates at the outlet of the catchment vary between

5. Modelling

82

3.7 and 141.7. They have been calculated using a coefficient of variation of the measurement

errors equal to 10%. According to the team responsible of the measurement, this value is

considered as pertinent considering the measurement method used (personal communication).

Mean absolute residuals

( )N

hhN

i

compi

obsi∑ −

Mean residuals

( )N

hhN

i

compi

obsi∑ −

Mean weighted residuals

( )N

hhwN

i

compi

obsii∑ −2

1

A7-PL37 [m] 8.2 -8.2 -16.2

BOR009 [m] 4.0 -3.5 -7.0

CEL167 [m] 1.9 -0.3 -0.7

MOM001 [m] 3.0 1.1 2.2

OTH002 [m] 4.1 -2.1 -4.2

SLI006 [m] 3.3 2.6 5.3

VIE044 [m] 3.7 -2.9 -5.8

WIH014 [m] 5.3 -4.8 -9.5

Kanne [m³.s-1] 0.71 0.04 0.44

Table 5.6. Mean errors between observed and computed heads for the nine observation wells (hobs: observed hydraulic head, hcomp: computed hydraulic head, N: number of observations).

As shown in Table 5.6, the mean residual for all hydraulic heads is equal to -2.3 m, varying from -

8.2 m (A7-PL37) to 2.6 m (SLI006), and the mean residual for flow rate observations is equal to

0.04 m³/s. Similarly, the mean weighted residual for hydraulic heads and flow rate observations

are equal to -6.2 and 0.44, respectively. The mean weighted residual aggregating all types of

observations is equal to -5.2. Ideally, for the model being a consistent representation of the whole

hydraulic system, weighted residuals must be randomly distributed around zero (Hill and

Tiedeman, 2007). Here, the mean weighted residuals indicate that there is a negative bias in the

simulated groundwater levels. According to Figure 5.4C and Table 5.6, this negative bias is mainly

due to the simulated groundwater levels at A7-PL37 which severely underestimate observed

groundwater levels. These higher model errors at A7-PL37 could be explained by the proximity

5. Modelling

83

of the model borders, where the boundary conditions may not be verified locally, or by the

proximity of the Geer valley, which the more abrupt topography may not be well represented by

large finite elements. In particular, groundwater losses through the northern catchment boundary

may be variable along this border, while they are simulated in the model using a uniform 'head

dependent flux' boundary condition. This bias is then more representative of inadequacies in the

conceptual assumptions than in the calibration procedure. Although this kind of model errors

could have non negligible implications on model predictions, they are in this case quite limited

and local, as discussed later in Section 5.3, regarding the results of the sensitivity analysis.

Finally, Hill and Tiedeman (2007) propose to evaluate the 'Calculated error variance' or

'Calculated standard error' using Equation 5.2, as a measure of the overall model fit. If the

weighting scheme described above is used, a 'calculated standard error' equal to 1 means that the

model errors or residuals are consistent with the observation measurement errors. In other

words, it means that, in average, the model errors will be equivalent to the observation

measurement errors. For the Geer basin model, the 'calculated standard error' is equal to 10.2

with a 95% confidence interval between 9.9 and 10.55. This value indicates that, in average, the

model errors are greater than the measurement errors, which is common with hydrological

models. A 'calculated standard error' higher than 1 is not an indicator of badly construct models

with regard to the conceptual assumptions, but have some implications on the accuracy of the

predictions made with the calibrated model. The evaluation of the uncertainty related to the

model predictions is the subject of Section 8.2.

5 This confidence interval is calculated as suggested in Hill and Tiedeman (2007): 2

2

2

2

;LU

snsnχχ××

, where n is the

degree of freedom equal to the difference between the number of observation and the number of parameter, s is the 'calculated standard error', 2

Uχ and 2Lχ are the upper and lower tail value of a chi-square distribution with n degree

of freedom.

5. Modelling

84

Calculated error variance : ( )[ ]

( )NParNObs

byyws

NObs

iiii

−

−=∑ 2

2'

(5.2a)

Calculated standard error : ( )[ ]

( )

21

2'

⎟⎟⎟⎟

⎠

⎞

⎜⎜⎜⎜

⎝

⎛

−

−=∑

NParNObs

byyws

NObs

iiii

(5.2b)

NObs : Number of observations

NPar : Number of parameters

While the method exposed above enables to evaluate the overall fit of the model, the graphs of

Figure 5.5 and Figure 5.6 enable to evaluate the calibration in a 'transient' context. Figure 5.5

shows that groundwater seasonal variations, as calculated by the model, are slightly too high at

some observation wells, especially for ‘VIE044’, where the groundwater level is close to the

ground surface. However, simulated heads satisfactorily reproduce the multi-annual variations in

groundwater levels, which is important is the context of long term climate change impact on

groundwater reserves. These multi-annual variations are better reproduced in the upstream part

of the catchment (VIE044, BOR009, CEL167) than in the downstream part (SLI006, WIH014),

where they are overestimated. The observations are however more limited at WHI014 and

SLI006 which makes the evaluation of the calibration more hazardous. At Figure 5.6, the

simulated flow rates at the outlet of the basin are of the same order of magnitude as the observed

flow rates at the 'Kanne' gauging station. However, some discharge peaks present too high values

in comparison to the observed flow rates, and the associated recession periods are often too long.

This is especially noticed for high water discharge months (e.g. Figure 5.6C) for which these too

high and too long peaks result in overestimated monthly flow rates. Representative examples are

for December 1981, February 1984 and March 1988 when simulated flow rates are more than

twice as important as observed flow rates. Though it could not explain alone such a difference, it

5. Modelling

85

must be noted that the observed high water flow rates are characterised by a higher measurement

uncertainty given the lack of validation gauging for occasional high discharge events.

Nevertheless, simulated annual flow rates match quite well the observed annual flow rates and

the water balance analysis shows that the model overestimates by 0.3% of the total precipitation

the water flow rates at the 'Kanne' gauging station over the period 1975-2003. Table 5.7 shows

the main components of the water budget for the simulation performed between 1967 and 2003.

Figure 5.5 : Transient calibration of hydraulic heads for the nine observation wells

5. Modelling

86

Figure 5.6. Transient calibration of surface flow rates for the Kanne gauging station (outlet)

5. Modelling

87

Rain Actual evapotransp.

North boundary

Outlet (‘Kanne’)

Water abstraction

Water balance error (∆ res.)

mm/year 803.5 -558.4 -45.0 -159.1 -51.4 -10.4

% of rainfall 100 -69.5 -5.6 -19.8 -6.4 -1.3

Table 5.7. Simulated mean water balance terms for the period 1967-2003

5.2 Sensitivity Analysis

At the same time as the calibration procedure, a sensitivity analysis for all parameters used in the

model was performed. When implementing and calibrating a model, performing a sensitivity

analysis is fundamental as it enables to identify which parameters are sensitive to the variables of

interest. Thanks to these results, the insensitive or nearly insensitive parameters can be ignored in

the calibration procedure as any change in their value will have very little effect on the variables

used in the analysis. This enables to drastically reduce computing times dedicated for calibration,

especially when automatic calibration codes are used. Here, the sensitivity analysis has been

performed using 'UCODE_2005' (Hill and Tiedeman, 2007; Poeter et al., 2005) in combination

with 'HydroGeoSphere'. UCODE_2005 is a computer code, developed by the 'U.S. Geological

Survey' (USGS), which can be coupled with other modelling codes to perform sensitivity analysis,

data needs assessment, calibration, prediction and uncertainty analysis. The sensitivity analysis of

the Geer basin model parameters (Figure 5.7) has been performed using the 'perturbation

method', which consists in perturbing one parameter by a small value – the others being kept

constant - and looking the effect on selected variables. UCODE_2005 is used to perturb each

parameter one by one, and to run one HydroGeoSphere simulation for each perturbed

parameter. The 'Composite Scaled Sensitivity' is then calculated for each parameter using

Equation 5.3 (Hill and Tiedeman, 2007) and the as weighting scheme described in Section 5.1.5.2

5. Modelling

88

and Equation 5.1. The 'Composite Scaled Sensitivity' (CSS) aggregates the sensitivities of all

selected variables relatively to one parameter.

21

2.

2/1

⎟⎟⎟⎟⎟⎟

⎠

⎞

⎜⎜⎜⎜⎜⎜

⎝

⎛

⎟⎟⎠

⎞⎜⎜⎝

⎛××

∂∂

=∑

NObs

wbby

CSS

NObs

iij

j

i

j (5.3)

jCSS : Composite Scaled Sensitivity of parameter j

For the Geer basin model, selected variables correspond to monthly mean hydraulic heads at the

8 observation points and monthly mean flow rates at the outlet of the basin (Figure 4.1). In order

to limit computing times, the sensitivity of some similar parameters but for different zones are

calculated jointly by the means of unity multipliers. UCODE_2005 perturbs each multiplier that

acts simultaneously on several parameters, rather then to perturb all the parameters one by one in

different iterations. For the case of the Geer basin, the parameters sensitivities are generally

calculated jointly for all zones of the model. The hydraulic conductivity sensibilities are calculated

for each zone of Figure 5.2 independently of the layer and for each layer independently of the

horizontal zonation. As an example, the parameter related to the hydraulic conductivity of the

'dry valleys' in the 'upper chalk' layer is calculated using two different multipliers as shown in

Equation 5.4.

smMULTMULTK valleysDryChalkK

ChalkUpperK

ChalkUppervalleysDryChalk /102 4___

_−−

− ×××= (5.4)

As shown in Figure 5.7, the most sensitive parameters are the hydraulic conductivity of the 'dry

valleys', the field capacity used to calculate the actual evapotranspiration, the hydraulic

conductivity of the zones 'Chalk 1' (Figure 5.2), the hydraulic conductivities in the upper chalk

layer and the total porosity of the chalk. The hydraulic conductivity of the loess, parameters C2

5. Modelling

89

and C1 used to calculate the transpiration, the conductance used in the Cauchy condition along

the North boundary, the hydraulic conductivity of the zone 'Chalk 5' located besides the outlet of

the catchment, the 'Leaf Area Index' and the wilting point present medium sensitivities to

observations. The other parameters show low sensitivities to observations. The specific storage

parameter has a very low sensitivity, in accordance with the fact that the aquifer is unconfined

over most of the catchment. The sensitivity of the Van Genuchten parameters could not be

calculated using 'UCODE_2005' because unsaturated functions are directly specified using tables

of values (Pressure – Relative hydraulic conductivity – Saturation) instead of the parameter

values. This constitutes a limitation of 'UCODE_2005'. However, the experience gained during

the calibration of the model showed that the parameters α and β (Table 5.2) also present a

significant sensitivity.

Figure 5.7. Composite Scales Sensitivities (CSS) of the parameters used in the Geer basin model

The same kind of analysis has been performed inversely, by aggregating the sensitivities of all

parameters relatively to one observation point (Equation 5.5). This aggregated value traduces the

relative importance of the information provided by each observation point in the calibration of

the parameters. Figure 5.8A presents the aggregated sensitivities for each of the 8 observation

5. Modelling

90

wells and for the 'Kanne' gauging station at the outlet of the catchment. Though all statistics have

more or less the same order of magnitude, this graphs shows that the simulated data at 'Kanne'

and A7-PL37 are the least sensitive to the parameters variations.

21

.2

.2/1

..

⎟⎟⎟⎟⎟⎟

⎠

⎞

⎜⎜⎜⎜⎜⎜

⎝

⎛

⎟⎟⎠

⎞⎜⎜⎝

⎛××

∂∂

=∑ ∑

NParNObs

wbby

CSS K

NObs

i

NPar

jij

j

i

K

K

(5.5)

KCSS : Composite Scaled Sensitivity for all observations of the observation point K

KNObs : Number of transient observations for observation point K

As proposed by Hill and Tiedeman (2007), the statistics of Figure 5.7 and Figure 5.8A can be put

in parallel with the 'leverage' statistics shown at Figure 5.8B. The 'leverage' statistic jointly

expresses the importance of an observation relatively to its parameters sensitivity and relatively to

its capability to reduce parameter correlations. In groundwater models, typical correlations occur

between recharge and hydraulic conductivity for which simultaneous inverse variations may have

little effects on groundwater levels. More details about the calculation of the 'leverage' statistic

can be found in Hill and Tiedeman (2007). For the Geer basin model, the flow rates at 'Kanne'

have the highest 'leverage', three times more important than the other observation points. Given

the relatively low aggregated sensitivity of the flow rates (Figure 5.8A), this high 'leverage' values

then traduces the capability of these observations to reduce the correlations between parameters

of the Geer basin model, when used in the calibration procedure. This analysis reinforces the

importance of using flow rate observations when calibrating the model, and demonstrates the

attractiveness of integrated surface – subsurface models.

5. Modelling

91

Figure 5.8. (A) Aggregated sensitivities for each observation point. (B) Mean leverage statistics for each

observation point.

5.3 Discussion

The modelling approach integrating surface flow, subsurface flow and evapotranspiration better

represents the interdependent aspect of recharge processes between surface and subsurface

domains compared to classical or externally coupled models, which is a key element in the

context of assessing potential climate change impacts on groundwater.

Spatial and temporal discretisations have been chosen to allow the study of long-term variations

of groundwater levels and water balance terms under changing climate. A 30-year simulation with

the Geer basin model using daily input data takes more than 12 days on a 3.0 GHz Pentium4

desktop machine equipped with 4 Gb RAM. Using a discretisation as fine as reported by Jones

(2005) and Li et al. (2008) (see Section 2), would lead to excessively large simulation times, mostly

5. Modelling

92

because of the much longer period covered by the climate change scenarios. However, the

objective of the model is not to simulate surface water at the river bed scale, but to provide an

accurate representation of the components of water balance at any time during the simulation.

Using a model with a coarser discretisation is assumed to be appropriate to study climate change

impacts while keeping the computational demand low. The grid used was developed according to

this objective.

Even with a limited number of nodes, computational times remain very large and make the

completion of an automatic calibration dramatically long. Automatic calibration or inversion

procedures minimize the objective function through several iterations, each of them including a

complete sensitivity analysis of all parameters. This implies running dozens of simulations and

leads to computing times as large as several years. The calibration performed here was then

performed by trial-an-error and, as all other calibrations, remains imperfect as shown in Section

5.1.5.2. Remaining uncertainty may translate into the results of climate change impact studies and

needs to be addressed. This is the subject of Section 8, where some 'UCODE_2005' capabilities

are used to calculate confidence intervals for predicted variables. Additionally to the uncertainty

around simulated values, the evaluation of the calibration performed at Section 5.1.5.2 also

showed a non negligible bias for simulated groundwater levels at A7-PL37, probably induced by

the North boundary condition which could be non adequate close to this observation well.

Although this kind of biases could have implications on subsequent prediction analysis, the

results of the sensitivity analysis show that they should be in this case quite limited. Simulated

values at A7-PL37 are the least sensitive to parameters and using observations at this well do not

play an important role in the reduction of parameters correlation (see Figure 5.8). It is then

expected that the effect of the bias at A7-PL37 will only have a local influence on subsequent

prediction analysis and will not spread elsewhere in the model. Results at A7-PL37 should

however be considered with caution. Finally, it must be noted that the calibration performed with

5. Modelling

93

the Geer basin model is original as it is performed using both observed hydraulic heads and

surface water flow rates. Most studies where fully integrated surface – subsurface hydrological

models are used do not present any calibration results for observed subsurface hydraulic heads

(Jones, 2005; Li et al., 2008; Sudicky et al., 2008). Van Roosmalen et al. (2007) only use one

observation per well to calibrate their model. Additionally, they only present global performance

criteria values aggregating hydraulic head error from all observation wells. Consequently, it is

impossible to evaluate the quality of the calibration regarding spatial and temporal variations.

Complete sensitivity analyses, as presented at Section 5.2, are usually performed in the context of

model automatic calibrations. They are also very useful even with more classic calibrations as the

sensitivity analysis helps the modeller to better understand the system and eliminate insensitive

parameters. It should be noted that the sensitivity results depend on parameters values because

the hydrological model is actually non-linear. As a consequence, a parameter which appears to be

nearly insensitive under specific conditions, might gain in importance under different conditions.

In the case of the Geer basin model, the 'leverage' statistics clearly show the role of the surface

flow rates observations in reducing parameters correlations. With a simple subsurface model,

high groundwater levels could be explained by low hydraulic conductivities, high recharge rates

or low discharge rates. The use of an integrated surface-subsurface hydrological model and the

analysis of all water balance terms then enable to better identify the origin of errors provided by

the model simulations. This better understanding gives some reliability to the interpretations,

reinforces the importance of using both flow rates and groundwater level observations when

calibrating the model and, more generally, gives some credibility to the methodology of using

surface-subsurface integrated models.

5. Modelling

94

5.4 Alternative calibration

The results presented in Section 5.1.5 and the associated discussion showed that the calibration

of the Geer basin model is not perfect. As already said here above, large computing times and the

impossibility to use automatic calibration codes is one of the most important difficulties to be

faced when calibrating such catchment scale integrated models. Rather than spending huge time

quantities to achieve a better calibration, an alternative is to use several models with different

calibration results or different hypothesis in the conceptual model (Hill and Tiedeman, 2007).

While an – hopefully – better calibrated model may give more accurate prediction results, using

several different models will give a range of variation for predictions. This range of variation

traduces the uncertainty linked to the use of a model, and is sometimes more demanded by water

managers than a unique but more accurate prediction value. Following this idea, a second

calibrated model of the Geer basin is presented here after, and will be used in the following

chapters for the climate change impact study. Clearly, two models are not sufficient to evaluate

the uncertainty link to the use of hydrological models, but they will give a first idea about the

possible range of variation of climate change impacts.

This second model was developed previously to the first one and, in order to limit computational

time, specified fluxes were input on a monthly basis, using mean monthly precipitation,

evapotranspiration and groundwater abstraction rates. Adaptive time steps commonly vary

between 1 h and 1 day and a 30-year simulation takes a bit more than 1 day on a 3.0 GHz

Pentium4 desktop machine equipped with 4 Gb RAM. Figure 5.9, Figure 5.10 and Figure 5.11

present the calibration results for hydraulic heads and surface water flow rates. Generally,

computed heads are higher than observed heads, except in A7-PL37. Seasonal variations are still

too high and multi-annual variations in groundwater levels are underestimated at some

observation points. The simulated flow rates match well to observed values in summer, for low

flow rates and recession periods. Differences remain for all winter months, where simulated flow

5. Modelling

95

rates are too high. Over the period 1967-2003, simulated water flow rates at the 'Kanne' gauging

station overestimate by 6% of the total precipitation the observed flow rates. This overestimation

also implies that other water balance terms – actual evapotranspiration and water flux through

the North boundary – are underestimated. This model then presents important positive biases in

weighted residuals for the flow rates at the outlet of the aquifer and for all observation wells,

except for A7-PL37 which presents a negative bias. These biases have some implications on the

accuracy of the simulation results and make hazardous the evaluation of model uncertainty (Hill

and Tiedeman, 2007), as performed in Section 8 for the first hydrological model. Further

discussion about the use of this model is provided in Section 6. More details about the calibration

of this second model can be found in Goderniaux et al. (2009) and in Annex 1.

Figure 5.9. Transient calibration of hydraulic heads for the 8 observation wells (2nd model)

Figure 5.10. Transient calibration of surface flow rates for the Kanne gauging station (2nd model)

5. Modelling

96

Figure 5.11. Graphical analysis of the model calibration. (A) Computed values vs. observed values. (B)

Residuals vs. observed values. (C) Weighted residuals vs. observed values.

5. Modelling

97

5.5 References

Asner, G.P., Scurlock, J.M.O. and Hicke, J.A., 2003. Global synthesis of leaf area index observations: implications for ecological and remote sensing studies. Global Ecology & Biogeography, 12(3): 191-205.

Breuer, L., Eckhardt, K. and Frede, H.-G., 2003. Plant parameter values for models in temperate climates. Ecological Modelling, 169: 237-293.


Brouyère, S., Dassargues, A. and Hallet, V., 2004. Migration of contaminants through the unsaturated zone overlying the Hesbaye chalky aquifer in Belgium: a field investigation. Journal of Contaminant Hydrology, 72(1-4): 135-164.

Canadell, J., Jackson, R.B., Ehrlinger, J.R., Mooney, H.A., O.E., S. and Schulze, E.D., 1996. Maximum rooting depth of vegetation types at the global scale. Oecologia, 108: 583-595.

Dassargues, A. and Monjoie, A., 1993. The chalk in Belgium. In: R.A. Downing, M. Price and G.P. Jones (Editors), The hydrogeology of the chalk of the North-West Europe. Oxford University Press, Oxford, UK, pp. Chapter 8: 153 - 269.

Goderniaux, P., Brouyère, S., Fowler, H.J., Blenkinsop, S., Therrien, R., Orban, P. and Dassargues, A., 2009. Large scale surface - subsurface hydrological model to assess climate change impacts on groundwater reserves. Journal of Hydrology, 373(1-2): 122-138.

Hallet, V., 1998. Etude de la contamination de la nappe aquifère de Hesbaye par les nitrates: hydrogéologie, hydrochimie et modélisation mathématique des écoulements et du transport en milieu saturé (Contamination of the Hesbaye aquifer by nitrates: hydrogeology, hydrochemistry and mathematical modeling). PhD Thesis, University of Liège, Liège (Belgium), 361 pp.


Hornberger, M.G., Raffensperger, J.P., Wilberg, P.L. and Eshleman, K.L., 1998. Elements of Physical Hydrology. JUH Press, 312 pp.

Jones, J.-P., 2005. Simulating Hydrologic Systems Using a Physically-based Surface-Subsurface Model: Issues Concerning Flow, Transport and Parameterization. PhD Thesis, University of Waterloo, Waterloo (Canada), 145 pp.

Kristensen, K.J. and Jensen, S.E., 1975. A Model For Estimating Actual Evapotranspiration From Potential Evapotranspiration. Nordic Hydrology, 6: 170-188.

5. Modelling

98

Li, Q., Unger, A.J.A., Sudicky, E.A., Kassenaar, D., Wexler, E.J. and Shikaze, S., 2008. Simulating the multi-seasonal response of a large-scale watershed with a 3D physically-based hydrologic model. Journal of Hydrology, 357(3-4): 317-336.

Orban, P., Batlle-Aguilar, J., Goderniaux, P., Dassargues, A. and Brouyère, S., 2006. Description of hydrogeological conditions in the Geer sub-catchment and synthesis of available data for groundwater modelling. Deliverable R3.16, AquaTerra (Integrated Project FP6 no 505428).


Scurlock, J.M.O., Asner, G.P. and Gower, S.T., 2001. Worldwide Historical Estimates of Leaf Area Index, 1932-2000, prepared for the Oak Ridge National Laboratory. ORNL/TM-2001/268, Oak Ridge, Tennessee.

Sudicky, E.A., Jones, J.-P., Park, Y.-J., Brookfield, E.A. and Colautti, D., 2008. Simulating complex flow and transport dynamics in an integrated surface-subsurface modeling framework. Geosciences Journal, 12(2): 107-122.


Van Roosmalen, L., Christensen, B.S.B. and Sonnenborg, T.O., 2007. Regional Differences in Climate Change Impacts on Groundwater and Stream Discharge in Denmark. Vadose Zone J, 6(3): 554-571.

Vázquez, R.F. and Feyen, J., 2003. Effect of potential evapotranspiration estimates on effective parameters and performance of the MIKE SHE-code applied to a medium-size catchment. Journal of Hydrology, 270(3-4): 309-327.

6. Application of climate change scenarios on the Geer basin model

99

6. APPLICATION OF CLIMATE CHANGE SCENARIOS

ON THE GEER BASIN MODEL


100


101

As stated previously, the integrated Geer basin model has been specially developed to assess the

possible impacts of climate change on groundwater resources. As a next step climate change

scenarios are therefore applied to the basin model and projected changes and uncertainties are

assessed. Here, 6 climate change scenarios for 3 different time periods (2011-2040, 2041-2070,

2070-2100) are downscaled from Regional Climate Models (RCMs) to the station scale using a

'quantile mapping bias-correction' technique.

6.1 Climate scenarios

In order to assess the likely impacts of climate change on water resources for the Geer

catchment, Regional Climate Model (RCM) output from the European Union Fifth Framework

Programme (FP5) PRUDENCE project (Prediction of Regional scenarios and Uncertainties for

Defining EuropeaN Climate change risks and Effects) (Christensen et al., 2007) was used. These

dynamic climate models provide a series of high-resolution simulations of European climate for a

control simulation (1961-1990) and for a future time period (2071-2100). These are the results of

a series of “time-slice” experiments, each representing a stationary climate over the selected 30-

year period, whereby a climate model is allowed to fully adjust to an equilibrium state in response

to a prescribed radiative forcing, i.e. the simulations reflect variability about an equilibrium state

over a 30-year period. In addition to the uncertainty introduced by the choice of RCM, each

model derives its boundary conditions from a different GCM, with each GCM representing

atmospheric processes differently, either through different numerical schemes or different

parameterisations. One way of addressing these uncertainties is through the use of multi-model

ensembles. Here, we use an ensemble of 6 RCM simulations (Table 6.1) with boundary

conditions derived from what may be considered as two different GCMs, the HadAM3H

atmosphere only model (Gordon et al., 2000; Pope et al., 2000) and the ECHAM4/OPYC

coupled atmosphere-ocean model (Roeckner et al., 1996). The HadRM3P and ARPEGE RCM

simulations derive boundary conditions from HadAM3P and the coupled atmosphere-ocean


102

model HadCM3 respectively. Both HadAM3H and HadAM3P are dynamically downscaled to an

intermediate resolution from the HadCM3 coupled atmosphere-ocean model and are thus closely

related. Further details on the RCMs used within PRUDENCE may be found in Jacob et al.

(2007).

Here, only projections using the SRES A2 emissions (medium-high) scenario (Nakicenovic et al.,

2000) are examined as recent observed increases in atmospheric carbon dioxide concentrations

are in accordance with projections from high emissions scenarios (Rahmstorf et al., 2007).

However, significant divergence in greenhouse gas concentrations between scenarios in the

second half of the 21st century generates uncertainty in future climate forcing. Although this

uncertainty arising from future emissions is not examined here it is discussed within the context

of the final results in Section 6.5. For each RCM, mean daily temperature and daily total

precipitation were extracted for the control and future time periods for the RCM grid cells

overlying the meteorological stations shown in Figure 4.1.

A2 SCENARIO

INST RCM GCM PRUDENCE ACRONYM

AQUATERRA ACRONYM

DMI HIRHAM HadAM3H A2 HS1 HIRHAM_H

DMI HIRHAM ECHAM4/OPYCA2 ecscA2 HIRHAM_E

HC HadRM3P HadAM3P A2 adhfa HAD_P_H

SMHI RCAO HadAM3H A2 HCA2 RCAO_H

SMHI RCAO ECHAM4/OPYCA2 MPIA2 RCAO_E

Météo-France Arpège HadCM3 A2 DE6 ARPEGE_H

Table 6.1. Climate change scenarios with corresponding RCM and GCM. DMI: Danish Meteorological Institute, HC: Hadley Center for Climate Prediction and Research, SMHI: Swedish Meteorological and

Hydrological Institute.


103

6.2 Downscaling of RCM output

Even the relatively high-resolution RCMs (approximately 0.5° grids) used in this study are

too coarse to be effectively applied in hydrological impacts studies and a further downscaling step

is therefore required. One of the simplest downscaling methods that has been applied in

hydrological impacts assessment is the bias-correction approach (e.g. Fowler and Kilsby, 2007;

Kleinn et al., 2005). In this approach, biases in climate model control simulations of the mean

monthly climatology for the relevant grid cell relative to station observations are calculated

(calculated as a simple difference for temperature and a ratio for precipitation). This bias is

assumed to be the same for the future simulations and so corrected climate change scenarios may

therefore be obtained by applying the same bias corrections additively to daily temperature and as

a scalar to daily precipitation values for future time periods. However, this method only applies

the correction to the mean and does not take account of model deficiencies in reproducing

observed variability. We therefore adopt the quantile-based mapping approach to bias correction

described by Wood et al. (2004) which has been previously used in hydrological impacts studies

(e.g. Salathé et al., 2007). This mapping approach uses an empirical transfer function (e.g.

Panofsky and Brier, 1968) to force the probability distributions of the control simulations of daily

temperature and precipitation to match the observed distributions. Separate mapping functions

are calculated on a monthly basis for each station using the appropriate grid cell from each

model. Thus, for each RCM, the distributions of daily temperature and precipitation for the

control simulation are corrected to match those of the observed data, and are identical for each

model. Under the assumption that model biases are stationary in time, the same transfer

functions are therefore applied to adjust the temperature and precipitation scenarios for the

2071-2100 time period.

For many hydrologists and water resource planners, scenarios for the end of the 21st century do

not adequately reflect the most appropriate timescales for decision-making and planning. More


104

frequently management decisions are made for the near-future, on decadal rather than century

timescales. To address these needs, scenarios were also produced for two additional time

periods: 2011-2040 and 2041-2070. To produce these we adopted a conventional pattern scaling

approach (Mitchell, 2003; Santer et al., 1990), assuming that changes to mean climate parameters

will occur in proportion to the projected change in global mean temperature. This method has

been used to construct climate change scenarios for hydrological impact studies (e.g. Salathé,

2005) and has been applied to the scaling of changes in different climatic variables for different

geographic regions and time periods (e.g. Mitchell et al., 1999; Santer et al., 1994; Tebaldi et al.,

2004). The changes in mean monthly temperature and total monthly precipitation were therefore

scaled for the relevant time periods in proportion to the mean global temperature change

projected by the GCM which provided lateral boundary conditions for each RCM simulation

(either HadCM3 or ECHAM4) using data available from the IPCC data distribution centre6.

6.3 Projected changes in local climate

The climate change scenarios for the 2071-2100 time period show a general increase in

temperature throughout the year (Figure 6.1A). The annual mean temperature increase for

Bierset ranges from +3.5°C (HIRHAM_H) to +5.6°C (RCAO_E) with the projected change

strongly influenced by the GCM used to drive the RCM simulations. Simulations driven by GCM

ECHAM4/OPYCA2 (scenarios HIRHAM_E and RCAO_E, see Table 6.1) project the greatest

increases, particularly during spring and summer. Although all models project the largest

temperature increases during summer with the maximum increases in August, those by

HIRHAM_E (~7.5°C) and RCAO_E (~9.5°C) are larger than those projected by the other

6 http://www.ipcc-data.org/


105

models (+4.7°C to +6.4°C). All models project the smallest temperature increases during late

winter/early spring ranging from ~+2°C (HIRHAM_H; March) to ~+5.5°C (RCAO_E; March).

Figure 6.1. Monthly climatic changes for each climate change scenario (period 2071-2100) relatively to 1961-

1990. (A) Temperature - Bierset climatic station. (B) Precipitation - Waremme climatic station.

The RCMs consistently project a decrease in annual precipitation for the 2071-2100 time period

but there is a large range from −1.9 % (ARPEGE_H) to −15.3 % (HAD_P_H) (Figure 6.1B).

These precipitation decreases are a consequence of large projected decreases during summer

months but are partly offset by increases in winter precipitation. The largest summer decreases

are projected by RCAO_E but these are also offset by the largest winter increases projected by


106

any of the models. The large annual decrease projected by HAD_P_H however arises as a

consequence of moderate decreases in summer precipitation that persist throughout autumn and

are only offset by comparatively small increases during winter.

Figure 6.2. Monthly climatic changes for the three time period 2011-2040, 2041-2070, 2071-2100 (climate model RCAO_E) relatively to 1961-1990. (A) Temperature – Bierset climatic station. (B) Precipitation –

Waremme climatic station.

Figure 6.2 shows monthly mean temperature and precipitation change for the three time periods

2011-2040, 2041-2070, 2071-2100 and for the climate model RCAO_E (GCM

ECHAM4/OPYCA2). In accordance with the pattern scaling approach used to downscale the


107

climate change scenarios, the temperature and precipitation change gradually increases from the

first time period to the third one, with a greater change between the second and third time

periods than between the first and second ones. Conclusions are similar for the climate models

driven by the other GCM (HadAM3).

All models therefore suggest that by the end of the century, the climate of the Geer basin will

consist of warmer, wetter winters and much hotter, drier summers, with a more pronounced

annual cycle of temperature and precipitation. Given the decreased summer rainfall, higher

evapotranspiration driven by higher temperatures and the projected regional increase in the

frequency of summer droughts (Blenkinsop and Fowler, 2007), increased stress is likely to be

placed on water resources during summer. During winter, higher evapotranspiration could be

offset by increased rainfall. The main form of uncertainty lies in the magnitude of the annual

groundwater recharge change and how quickly significant impacts on groundwater reserves will

be felt.

6.4 Projected changes in hydrological regime

Using the two calibrated flow models presented in Section 5 and the six downscaled RCM

scenarios, hydrological simulations were run to evaluate the direct climate change impacts on the

groundwater system of the Geer catchment for the three time periods 2011-2040, 2041-2070 and

2071-2100 using the bias-corrected precipitation and evapotranspiration scenarios. As the bias-

correction of each climate scenario reflects control simulation biases relative to observations,

future changes are expressed relative to an additional hydrological ‘control simulation’ driven by

the observed climate data. Daily PET are derived from temperature data using the Blaney-Criddle

method (Blaney and Criddle, 1950), which is a regression based approach that use relationships

between observed temperature and PET. Groundwater abstraction flow rates (from wells and

from the draining galleries) are firstly kept equal to zero through all simulations to avoid nodes


108

desaturation where abstraction flow rates are prescribed. The influence of groundwater

abstraction is discussed later in this section. As for the calibration procedure, initial conditions for

each time period and climate change scenario are obtained by running a preliminary steady state

simulation with the mean climatic data of the corresponding time period and climate change

scenario. Figure 6.3 and Figure 6.4 present the mean hydraulic head, the standard deviation,

minimum and maximum values for each time period, each RCM scenario and each observation

well. Similarly, Figure 6.5 and Figure 6.6 present the flow statistics at gauging station ‘Kanne’ for

each time period and each RCM scenario. These flow statistics are also presented for summer

and winter separately. Figure 6.3 and Figure 6.5 relate to climate change impact evaluated using

the hydrological model presented at Section 5.1.5 (calibration 1, with daily stresses). Figure 6.4

and Figure 6.6 relate to climate change impact calculated using the hydrological model presented

at Section 5.4 (calibration 2, with monthly stresses).

The results of the climate change impact study are quite different depending on the hydrological

model and calibration used. Using the first model, significant decreases are projected for almost

all groundwater levels and surface water flow rates, from the first to the third time period with a

gradual increase in intensity. By 2071-2100, mean groundwater levels are expected to decrease by

4-19 m depending on location in the Geer basin and the climate change scenario analysed (Figure

6.3). For an equivalent unsaturated zone depth, which smoothes recharge fluxes, the variability of

the groundwater levels is projected to increase. For the same period, flows at 'Kanne' are

expected to decrease by 37% - 65% depending on the climate change scenario, with a slightly

more important decrease during summer (Figure 6.5). Generally, the greatest changes are

projected by HAD_P_H which predicts large precipitation decreases during almost the whole

year, and by HIRHAM_E and RCAO_E which combine high increases in temperature and

medium decreases in precipitation. The smallest changes are projected by ARPEGE_H, which

combines a small increase in temperature with a small decrease in precipitation. The climate


109

change impacts achieved when using the second hydrological model and calibration are

significantly less important. During 2011-2040, no clear change from the observed control

simulation can be identified, with large uncertainties projected in the direction of change for both

surface flow rate and mean groundwater hydraulic heads. By 2071-2100, mean groundwater levels

are expected to decrease by 2-7 m (Figure 6.4). Annual surface flow rates at 'Kanne' are expected

to decrease between 9% and 32%, with a more important decrease during summer and no

significant change during winter.

Table 6.2 and Table 6.3 present the changes in each of the water balance terms for each time

period and each RCM scenario. Table 6.2 and Table 6.3 relate to climate change impact calculated

using the two hydrological models presented at Section 5.1.5 (calibration 1) and 5.4 (calibration

2), respectively. Generally, both tables show a decrease in surface water flow rates at 'Kanne' (as

previously shown at Figure 6.5 and Figure 6.6), a decrease in water fluxes through the North

boundary, and an increase in actual evapotranspiration. Over the three time periods, the actual

evapotranspiration gains an increasing importance compared to the annual rainfall flux, which is

expected to decrease in the future. It results in a decreasing importance of the surface water flow

rates at 'Kanne' compared to annual rainfall. The main difference between both hydrological

models lies in the magnitude of the actual evapotranspiration change. The first hydrological

model enables large increases in relative and absolute actual evapotranspiration over the three

time periods. At equal rainfall and potential evapotranspiration rates, the second hydrological

model projects a lower increase in actual evapotranspiration relatively to the rainfall fluxes.

Moreover, the absolute actual evapotranspiration rates decrease, as the general increase in

temperature is offset by the decrease in water quantities available for evapotranspiration. This

analysis shows that the simulated actual evapotranspiration is strongly dependent on the

calibration of the models themselves. As already discussed in Section 2, the water quantities


110

available for evapotranspiration will indeed depend on how water circulates on the ground

surface, in the partially and fully saturated zones.

6.5 Discussion

The application of climate change scenarios as inputs of two hydrological models with different

calibrations leads to significantly different results. The differences are linked to the simulated

actual evapotranspiration rates which are more important with the first hydrological model.

According to the analysis performed in Section 5.4, the calibration of the second hydrological

model presents substantial biases in groundwater levels and surface water flow rates. Particularly,

the flow rates at the outlet of the catchment are significantly overestimated, due to an

underestimation of the actual evapotranspiration and/or water fluxes through the North

boundary. However, it is difficult to know which of these two terms is overestimated and in what

proportions. In view of these biases, the results of the impact study using the first hydrological

model should be considered as more reliable. Particularly, the first model offers the opportunity

to perform an uncertainty analysis (see Section 8), what would be hazardous with the second

model. Nevertheless, (though they have to be considered with caution,) the results from the

second model should not be rejected, as there is no indication that they are far from hypothetical

true predictions. Additionally, both models present imperfections regarding multi-annual

groundwater variations which are overestimated by the first model and underestimated by the

second one. Consequently, though more confidence is attached to the first hydrological model,

the impact study performed in this chapter gives a first range of variation linked to the calibration

of the hydrological models. A subsequent analysis concerning this kind of uncertainty is

presented at Section 8.


111

Figure 6.3. Evolution of hydraulic heads at the eight observation wells for each climate change scenario,

using the model with calibration 1.


112

Figure 6.4. Evolution of hydraulic heads at the eight observation wells for each climate change scenario,

using the model with calibration 2.


113

Figure 6.5. Evolution of flow rates at gauging station 'Kanne' for each climate change scenario, using the

model with calibration 1.

Figure 6.6. Evolution of flow rates at gauging station 'Kanne' for each climate change scenario, using the

model with calibration 2.


114

Rain Actual

evapotransp.

Flux out of

North boundary

Flux out of

main outlet

(‘Kanne’)

Water

abstraction

mm/year 803.0 -498.8 -49.3 -254.0 0.0 Control period % of rainfall 100 -62.1 -6.1 -31.6 0.0

HIRHAM_H mm/year 774.6 -530.1 -47.8 -207.8 0.0

% of rainfall 100 -68.4 -6.2 -26.8 0.0

HIRHAM_E mm/year 776.9 -548.5 -46.9 -191.0 0.0

% of rainfall 100 -70.6 -6.0 -24.6 0.0

HAD_P_H mm/year 769.6 -532.8 -47.4 -199.5 0.0

% of rainfall 100 -69.2 -6.2 -25.9 0.0

RCAO_H mm/year 786.1 -533.3 -47.8 -215.0 0.0

% of rainfall 100 -67.8 -6.1 -27.4 0.0

RCAO_E mm/year 786.4 -553.0 -47.2 -199.7 0.0

% of rainfall 100 -70.3 -6.0 -25.4 0.0

ARPEGE_H mm/year 798.9 -532.5 -48.7 -239.1 0.0

2011

-204

0

% of rainfall 100 -65.5 -6.1 -29.9 0.0

HIRHAM_H mm/year 743.1 -559.1 -46.1 -154.5 0.0

% of rainfall 100 -75.2 -6.2 -20.8 0.0

HIRHAM_E mm/year 755.5 -576.9 -45.2 -146.7 0.0

% of rainfall 100 -76.4 -6.0 -19.4 0.0

HAD_P_H mm/year 733.0 -558.0 -45.3 -146.4 0.0

% of rainfall 100 -76.1 -6.2 -20.0 0.0

RCAO_H mm/year 767.4 -562.7 -46.5 -171.6 0.0

% of rainfall 100 -73.3 -6.1 -22.4 0.0

RCAO_E mm/year 772.6 -591.0 -45.4 -152.2 0.0

% of rainfall 100 -76.5 -5.9 -19.7 0.0

ARPEGE_H mm/year 794.7 -558.1 -47.6 -202.0 0.0

2041

-207

0

% of rainfall 100 -70.2 -6.0 -25.4 0.0

HIRHAM_H mm/year 697.9 -575.9 -43.0 -101.3 0.0

% of rainfall 100 -82.5 -6.2 -14.5 0.0

HIRHAM_E mm/year 719.1 -595.1 -42.4 -100.5 0.0

% of rainfall 100 -82.8 -5.9 -14.0 0.0

HAD_P_H mm/year 680.6 -563.8 -42.4 -99.7 0.0

% of rainfall 100 -82.8 -6.2 -14.7 0.0

RCAO_H mm/year 740.4 -590.6 -44.3 -123.3 0.0

% of rainfall 100 -79.8 -6.0 -16.7 0.0

RCAO_E mm/year 749.6 -618.8 -42.6 -107.2 0.0

% of rainfall 100 -82.6 -5.7 -14.3 0.0

ARPEGE_H mm/year 787.9 -592.4 -46.3 -164.2 0.0

2071

-210

0

% of rainfall 100 -75.2 -5.9 -20.8 0.0

Table 6.2. Variations of the mean water balance terms for each climate change scenario and time interval, using the model with calibration 1.


115

Rain Actual

evapotransp.

Flux out of

North boundary

Flux out of

main outlet

(‘Kanne’)

Water

abstraction

mm/year 803.0 -475.5 -40.9 -278.3 0.0 Control period % of rainfall 100 -59.2 -5.1 -34.7 0.0

HIRHAM_H mm/year 774.6 -462.7 -40.5 -266.7 0.0

% of rainfall 100 -59.7 -5.2 -34.4 0.0

HIRHAM_E Mm/year 776.9 -471.0 -40.1 -258.8 0.0

% of rainfall 100 -60.6 -5.2 -33.3 0.0

HAD_P_H Mm/year 769.6 -449.9 -40.5 -273.9 0.0

% of rainfall 100 -58.5 -5.3 -35.6 0.0

RCAO_H mm/year 786.1 -471.7 -40.5 -267.9 0.0

% of rainfall 100 -60.0 -5.1 -34.1 0.0

RCAO_E mm/year 786.4 -463.7 -40.4 -277.5 0.0

% of rainfall 100 -59.0 -5.1 -35.3 0.0

ARPEGE_H mm/year 798.9 -472.2 -40.9 -285.3 0.0

2011

-204

0

% of rainfall 100 -59.1 -5.1 -35.7 0.0

HIRHAM_H mm/year 743.1 -458.4 -39.8 -240.2 0.0

% of rainfall 100 -61.7 -5.4 -32.3 0.0

HIRHAM_E mm/year 755.5 -468.9 -39.5 -239.0 0.0

% of rainfall 100 -62.1 -5.2 -31.6 0.0

HAD_P_H mm/year 733.0 -450.7 -39.9 -237.6 0.0

% of rainfall 100 -61.5 -5.4 -32.4 0.0

RCAO_H mm/year 767.4 -470.7 -39.9 -250.0 0.0

% of rainfall 100 -61.3 -5.2 -32.6 0.0

RCAO_E mm/year 772.6 -465.7 -39.8 -260.7 0.0

% of rainfall 100 -60.3 -5.2 -33.7 0.0

ARPEGE_H mm/year 794.7 -485.9 -40.4 -266.7 0.0

2041

-207

0

% of rainfall 100 -61.1 -5.1 -33.6 0.0

HIRHAM_H mm/year 697.9 -436.7 -39.0 -216.8 0.0

% of rainfall 100 -62.6 -5.6 -31.1 0.0

HIRHAM_E mm/year 719.1 -449.5 -38.8 -221.7 0.0

% of rainfall 100 -62.5 -5.4 -30.8 0.0

HAD_P_H mm/year 680.6 -440.0 -39.0 -197.1 0.0

% of rainfall 100 -64.7 -5.7 -29.0 0.0

RCAO_H mm/year 740.4 -455.1 -39.3 -238.3 0.0

% of rainfall 100 -61.5 -5.3 -32.2 0.0

RCAO_E mm/year 749.6 -448.8 -39.1 -253.1 0.0

% of rainfall 100 -59.9 -5.2 -33.8 0.0

ARPEGE_H mm/year 787.9 -489.8 -39.9 -255.2 0.0

2071

-210

0

% of rainfall 100 -62.2 -5.1 -32.4 0.0

Table 6.3. Variations of the mean water balance terms for each climate change scenario and time interval, using the model with calibration 2.


116

As stated in Section 6.1, adopting a multi-model approach for the climate scenarios enables the

uncertainty derived from climate model selection to be incorporated into the assessment of the

impacts of climate change on the Geer catchment. The full range of uncertainties in future

climate scenarios is not represented in this study, as only six regional climate models from the

larger PRUDENCE ensemble have been used. However, the same framework could readily be

applied to a larger ensemble size given adequate computational resources. Furthermore, the

uncertainty in future emissions is not addressed in this study. Whilst the PRUDENCE project

does provide some RCM simulations for the same future time period (2071-2100) for the B2

emissions (medium-low) scenario, the application of these to the groundwater model is unlikely

to provide a greater understanding of future uncertainties in the response of the Geer basin. A

comparison of the contribution of the various sources of uncertainty within the PRUDENCE

model simulations indicates that emission scenario is the most important source only for summer

temperatures over southern Europe (Déqué et al., 2007). Generally, the uncertainty introduced

by the GCM boundary conditions is larger than that for the other sources whilst the RCM

introduces uncertainty of a similar magnitude to that of the GCM boundary conditions only for

summer precipitation. Here, the full range of uncertainty generated by the choice of GCM

boundary conditions is necessarily constrained by the experimental combinations provided by the

PRUDENCE project and has been maximised in terms of the subset of experiments selected in

this analysis. However, it is evident that the limited GCM selection applied in PRUDENCE

constrains the uncertainty measured from this source (Déqué et al., 2007). It is noted that the A2

and B2 scenarios examined by PRUDENCE only constitute 50% of the spread of greenhouse

gas concentrations from all SRES scenarios (Déqué et al., 2007) and that impact studies using a

larger range of emission scenarios suggest a greater contribution to total uncertainty generated by

emissions relative to RCM choice (e.g. Olesen et al., 2007). Nonetheless, this study provides a

major advance in the assessment of the uncertainty of the impact of climate change on

groundwater systems.


117

The impact study performed in this chapter does not consider any water abstraction though it

represents a non negligible water balance term for the period 1967-2003. Including some water

abstraction terms in the analysis will cause a decrease in absolute groundwater levels and surface

water flow rates, but the possible effect on groundwater drawdown linked to climate change is

less certain. To test the influence of pumping wells and draining galleries, additional climate

change simulations were performed, including groundwater abstraction flow rates. For different

reasons explained in Section 4, the draining galleries in the Geer basin can not be considered as

typical drains. Using a 'third type' (head dependant flux) boundary condition to represent them is

then not adequate, and they are actually represented with 'second type' (specified flux) boundary

condition in the model. The problem with this boundary condition is that the abstraction

volumes for the next decades are not known. The actual abstraction volumes for the period

1967-2003 show a relative correlation with the variations of the observed multi-annual

groundwater levels, but in the context of climate change, these groundwater levels are precisely

not known in advance. In this additional impact study, the abstraction flow rates have then been

fixed to the mean abstraction flow rates for the period 1967-2003 and are kept constant through

all simulations. The fact that the specified flow rates are constant constitutes a limitation because

the mean flow rate value may be too important in periods of low groundwater levels, inducing

node desaturation where negative flow rates are specified and the end of the simulation. To avoid

these numerical problems, the influence of groundwater abstraction on the climate change impact

has been tested using the second hydrological model, less prone to exhibit node desaturation.

Using this model also allows saving computing times. According to these new simulations,

groundwater levels, for the period 2071-2100 are expected to decrease by 2-8 m, instead of 2-7 m

when not considering water abstraction. The decrease is then slightly more important with active

wells and draining galleries, but the difference between both cases is small compared to the

difference achieved when using the two alternative calibrated models. Complete results


118

concerning the simulations with groundwater abstraction can be found in Goderniaux et al.

(2009).

Finally, the influence of the temporal discretisation on the climate change impact results was also

examined. Using the first hydrological model, identical climate change simulations were

subsequently run, using mean daily and mean monthly specified fluxes (potential

evapotranspiration and precipitation). Results (not presented here) show very small differences

between both temporal discretisations regarding the climate change impact on groundwater levels

and surface water flow rates at the outlet of the aquifer. The influence of the temporal

discretisation of the specified fluxes is also examined and discussed in more details in the next

Section.


119

6.6 References

Blaney, H.F. and Criddle, W.D., 1950. Determining water requirements in irrigated areas from climatological irrigation data. Technical Paper n°96, US Department of Agriculture, Soil Conservation Service, Washington, D.C.

Blenkinsop, S. and Fowler, H.J., 2007. Changes in European drought characteristics projected by the PRUDENCE regional climate models. International Journal of Climatology, 27: 1595-1610.

Christensen, J.H., Carter, T.R., Rummukainen, M. and Amanatidis, G., 2007. Evaluating the performance and utility of regional climate models: The PRUDENCE project. Climatic Change, 81(Supplement 1): 1-6.

Déqué, M. et al., 2007. An intercomparison of regional climate simulations for Europe: assessing uncertainties in model projections. Climatic Change, 81(0): 53-70.

Fowler, H. and Kilsby, C., 2007. Using regional climate model data to simulate historical and future river flows in northwest England. Climatic Change, 80(3): 337-367.

Goderniaux, P., Brouyère, S., Fowler, H.J., Blenkinsop, S., Therrien, R., Orban, P. and Dassargues, A., 2009. Large scale surface - subsurface hydrological model to assess climate change impacts on groundwater reserves. Journal of Hydrology, 373(1-2): 122-138.

Gordon, C., Cooper, C., Senior, C.A., Banks, H., Gregory, J.M., Johns, T.C., Mitchell, J.F.B. and Wood, R.A., 2000. The simulation of SST, sea ice extents and ocean heat transports in a version of the Hadley Centre coupled model without flux adjustments. Climate Dynamics, 16(2): 147-168.

Jacob, D. et al., 2007. An inter-comparison of regional climate models for Europe: model performance in present-day climate. Climatic Change, 81(Supplement 1): 31-52.

Kleinn, J., Frei, C., Gurts, J., Lüthi, D., Vidale, P.L. and Schär, C., 2005. Hydrologic simulations in the Rhine basin driven by a regional climate model. Journal of Geophysical Research, 110: 18.

Mitchell, J.F.B., Johns, T.C., Eagles, M., Ingram, W.J. and Davis, R.A., 1999. Towards the Construction of Climate Change Scenarios. Climatic Change, 41(3): 547-581.

Mitchell, T.D., 2003. Pattern Scaling: An Examination of the Accuracy of the Technique for Describing Future Climates. Climatic Change, 60(3): 217-242.

Nakicenovic, N. et al., 2000. Emissions Scenarios. A Special Report of Working Group III of the Intergovernmental Panel on Climate Change, Cambrigde University Press, Cambridge.

Olesen, J. et al., 2007. Uncertainties in projected impacts of climate change on European agriculture and terrestrial ecosystems based on scenarios from regional climate models. Climatic Change, 81(0): 123-143.


120

Panofsky, H.A. and Brier, G.W., 1968. Some applications of statistics to meteorology. The Pennsylvania State University, University Park, 224 pp.

Pope, V.D., Gallani, M.L., Rowntree, P.R. and Stratton, R.A., 2000. The impact of new physical parametrizations in the Hadley Centre climate model: HadAM3. Climate Dynamics, 16(2): 123-146.

Rahmstorf, S., Cazenave, A., Church, J.A., Hansen, J.E., Keeling, R.F., Parker, D.E. and Somerville, R.C.J., 2007. Recent Climate Observations Compared to Projections. Science, 316(5825): 709-709.

Roeckner, E. et al., 1996. The atmospheric general circulation model ECHAM-4: model description and simulation of present-day climate. Report n° 218, Max-Planck Institute for Meteorology, Hamburg, Germany.

Salathé, E.P., 2005. Downscaling simulations of future global climate with application to hydrologic modelling. International Journal of Climatology, 25(4): 419-436.

Salathé, E.P., Mote, P.W. and Wiley, M.W., 2007. Review of scenario selection and downscaling methods for the assessment of climate change impacts on hydrology in the United States pacific northwest. International Journal of Climatology, 27(12): 1611-1621.

Santer, B.D., Brüggemann, W., Cubasch, U., Hasselmann, K., Höck, H., Maier-Reimer, E. and Mikolajewica, U., 1994. Signal-to-noise analysis of time-dependent greenhouse warming experiments. Climate Dynamics, 9(6): 267-285.

Santer, B.D., Wigley, T.M.L., Schlesinger, M.E. and Mitchell, J.F.B., 1990. Developing climate scenarios from equilibrium GCM results. Rpt. No. 47, Max-Plandk-Institut-fur-Meteorologie, Hamburg (Germany).

Tebaldi, C., Nychka, D. and Mearns, L.O., 2004. From global mean responses to regional signals of climate change: simple pattern scaling, its limitations (or lack of) and the uncertainty in its results. Proceedings of the 18th Conference of Probability and Statistics in the Atmospheric Sciences, AMS Annual Meeting, Seattle (U.S.A.).

Wood, A.W., Leung, L.R., Sridhar, V. and Lettenmaier, D.P., 2004. Hydrologic Implications of Dynamical and Statistical Approaches to Downscaling Climate Model Outputs. Climatic Change, 62(1): 189-216.

7. Application of stochastic climate change scenarios on the Geer basin model

121

7. APPLICATION OF STOCHASTIC CLIMATE CHANGE

SCENARIOS ON THE GEER BASIN MODEL


122


123

7.1 Objectives

In Section 6, climate change scenarios downscaled using the 'Quantile Mapping Bias Correction'

technique were used to evaluate the impact of climate change on groundwater reserves in the

Geer basin. This technique presents however two important limitations. First, simulations are

performed for the three time slices 2011-2040, 2041-2070 and 2071-2100, with climatic time

series representing a stationary climate over each of these time slices. In reality, the change in

climatic conditions is expected to continuously increase over the years. Second, each RCM is

represented by a unique climatic time series for each time slice. However, while mean climatic

statistics can be projected in the future, it is not possible to predict the weather or the exact

values of climatic variables at a specific time in the future. In the context of a climate change

impact study, it implies that it is difficult to project groundwater levels at a specific date by using

only one climatic scenario. A given specific future year, simulated for example as a 'dry year' in a

climatic scenario, could indeed correspond to a 'dry year' in reality.

In this section, new developments are performed to overcome these two limitations. The

objectives of this additional analysis are twofold:

(1) to simulate groundwater flow conditions using climate change scenarios that simulate the

change of mean climatic statistics in a fully transient way;

(2) to evaluate the possible range of variation of projected flow conditions, due to the natural

variability of the weather around mean climatic statistics. The idea is to use a large

number of equiprobable climatic scenarios as input of the Geer basin model, and to

calculate possible impact on groundwater reserves in a probabilistic way.

To reach these objectives, a 'weather generator' is used to downscale and generate 100

equiprobable climatic time series representing continuously changing climatic conditions from


124

2010 to 2085, for each of the 6 climate model used in Section 6 (see Table 6.1). Similarly to other

statistical downscaling techniques, the 'weather generator' is more accurate than RCMs and it

enables to generate climatic scenarios at the catchment scale, with limited computing resources.

Additionally, the 'weather generator' used in this study represents more realistically the weather

variability and climatic extremes, in comparison with the other statistical or dynamical

downscaling techniques. Finally, it enables to simulate data time series for all variables used in the

Penmann-Montheith equation and to calculate the related potential evapotranspiration time series

in a more direct and accurate way than with the Blaney-Criddle method used in Section 6. The

'weather generator' used in this study is part of the most advanced downscaling techniques, but it

has never been used in a climate change impact study on groundwater reserves.

Section 7.2 describes the functioning of the 'weather generator' downscaling technique used in

this study, and presents the climatic scenarios generated for the Geer basin. Section 7.3 analyses

the results of the application of all stochastic scenarios as input to the Geer basin hydrological

model. Section 7.4 examines what are the implications on the results when applying different

numbers of scenarios on the hydrological model, and when using different time discretisations.

Section 7.5 provides additional discussions.

7.2 Simulation of the stochastic climate change scenarios

For the Geer basin case study, 100 equiprobable climatic scenarios from 2010 to 2085 were

generated for each of the 6 Regional Climate Models (RCM) and for a control period without any

climate change. The generation of these climatic scenarios is performed in two main steps. First,

the daily precipitation time series from 2010 to 2085 are generated using the rainfall model

'RainSim' (Burton et al., 2010; Burton et al., 2008). Second, the precipitation time series generated

with 'RainSim' are used as input of the 'CRU daily weather generator' (Kilsby et al., 2007; Watts et

al., 2004) to produce associated data time series for the other variables (temperature max.,


125

temperature min., relative humidity, sunshine hours, wind speed, atmospheric pressure). The

associated evapotranspiration time series are then calculated using the Penman-Montheith

equation.

Observed data used to calibrate 'RainSim' correspond to precipitation from Waremme climatic

station (1961-1990), which has been identified as the most representative across the Geer basin.

Observed data used to calibrate the 'CRU weather generator' (temperature, relative humidity,

sunshine hours, wind speed, vapour pressure, solar radiation) are from Bierset station (1961-

1990), where long continuous data time series are available. Except for 7 years of daily

temperature data, the additional variables required for the 'CRU weather generator' are not

available at Waremme station, from where precipitation data are used. However, Bierset and

Waremme climatic stations are located only 15 kilometres away. Daily precipitation data for both

stations present a very good correlation with slightly higher mean precipitations at Bierset. Daily

temperature data for the 7 common years present a quasi perfect correlation and differences in

mean monthly temperatures do not exceed 0.2°C.

The main concepts used by the 'RainSim' rainfall model and the 'CRU weather generator', as well

as results concerning the data time series generated for the Geer basin are presented in sections

7.2.1 and 7.2.2.

7.2.1 Generation of precipitations times series using 'RainSim'

RainSim is a stochastic rainfall generator that enables, among others, to perform statistical

downscaling of climate change scenarios and to generate large number of stochastic

precipitations time series. The concepts and equations used in RainSim are presented in Burton et

al. (2008) and Burton et al. (2010). These concepts are summarized here after.


126

7.2.1.1 General concepts

RainSim is based on the Neyman-Scott rectangular pulses (NSRP) model (Cowpertwait, 1991;

Neyman and Scott, 1958) which conceptualises each rainfall event as the aggregation of several

"rain cells" (or rain rectangular pulses) characterised by a rainfall intensity and duration (Figure

7.1). The properties of all rainfall events and their associated "rain cells" are determined by

several random variables governed by specific statistical distributions and specified parameters.

All these random variables, their corresponding statistical distribution and parameters are

presented in Table 7.1. Figure 7.1 presents the main concepts of the NSRP model.

Figure 7.1. Schematic of the NSRP stochastic rainfall model (Figure from Burton et al., 2008). The circles

represent the rainfall events. Each star is associated with the beginning of a 'rain cell'.


127

Random variable Statistical distribution of the random variable

Parameter defining the statistical distribution

Time interval between rain events [T] Poisson λ

Number of "rain cells" for each rain event [-] Poisson ν

Time interval between the origin of each "rain cell" and the origin of the corresponding rain event [T] Exponential β

Intensity of each "rain cell" [LT-1] Exponential η

Duration of each "rain cell" [T] Exponential ξ

Table 7.1. Random variables used in the NSRP model

7.2.1.2 Calibration of the model for the control scenarios (without any climate

change)

Different rainfall time series can therefore be generated using the NSRP model given the

specification of the parameters defining the statistical distributions and different outcomes of the

random variables presented in Table 7.1. These parameters are usually calibrated by fitting the

NSRP model to observed data and a distinct calibration is usually performed for each month of

the calendar. More particularly, RainSim performs automatic calibration for each month of the

calendar by fitting specific statistics derived from simulated rainfall time series to the equivalent

statistics derived from observed time series. In this study, the following 6 observed rainfall

statistics are used to calibrate the model:

- daily mean

- daily variance

- monthly variance

- probability of a dry day (< 1 mm)

- daily lag-1 autocorrelation.

- daily skewness coefficient = ( )( )[ ]

3

3

dY

dd YEYEσ−


128

Yd : Daily rain accumulation

2dYσ : Variance of the daily rain accumulation

The probability of a dry day is evaluated considering that a day is dry if rain accumulation is less

than 1 mm. The daily lag-1 autocorrelation is the cross correlation of the rainfall time series with

itself considering a time-lag of 1 day. The daily skewness coefficient in the calibration is

particularly important to get a good representation of extremes events.

Though it is automatically performed by 'RainSim', the user can control the calibration by

weighting each of these 6 statistics in the objective function and by specifying some upper and

lower bounds to the parameters. The set of weights is chosen to get the best fit of simulated to

observed statistics, in accordance with the objectives of the underlying study. As an example, in a

study where the representation of climate extremes is important (drought, floods), the user can

decide to increase the weight of the daily skewness coefficient.

Figure 7.2 presents observed, fitted and simulated statistics for the Geer basin climate

corresponding to the period 1961 – 1990. The black lines correspond to observed data. The black

crosses refer to the statistics as fitted by 'RainSim' after the calibration process. The red circles

refer to 10 different stochastic simulations for time intervals of 100 years under the same climate

conditions than the observation period 1961 – 1990. Generally, the resulting fit as obtained with

'RainSim' is good although deviations are noticed for the daily skewness coefficient and monthly

variance. The range of variation displayed by the 100-years stochastic simulations almost always

include the corresponding observed statistic value.


129

Figure 7.2. Observed, fitted and simulated precipitation statistics for the Geer basin climate corresponding

to the period 1961 – 1990

7.2.1.3 Calibration of the model for the climate change scenarios

The parameters values achieved with this calibration (Figure 7.2) relate to the observed climatic

data of the period 1961 – 1990. In the context of climate change, where climatic statistics are by

definition expected to change, these parameter values can not be used and new parameter value

have to be calculated for any different "climatic state". More particularly, to simulate a continuous

and gradual change between 2010 and 2085, the parameter values should also vary continuously


130

with time and climate change intensity. In practical terms, new parameter values have to be

calculated for each year of the time period and for each month of the calendar. To generate

complete stochastic simulations between 2010 and 2085, it then implies the calculation of 912

different sets of 5 parameters (see Table 7.1).

'RainSim' can routinely calculate all these parameters based on "observed", or "target", climatic

statistics for all months and years between 2010 and 2085. These target climatic statistics can be

achieved, in two steps, thanks to data provided by RCMs and GCMs. First, the climatic statistics

for 2085 are calculated using the Geer basin observed climatic statistics for the period 1961 –

1990, and the statistics of a RCM 30-year time slices for the periods 1961 – 1990 (control period)

and 2071 – 2100 (Equation 7.1). Second, the climatic statistics for each year between 2010 and

2085 are interpolated considering that rainfall statistics vary in the same proportions as the

temperatures from the GCM used to drive the RCM boundary conditions. A scaling factor is

calculated according to Equation 7.2 for the years 2025 and 2055. The scaling factor of the years

1975 and 2085 is equal to 0 and 1, respectively. Between 1975, 2025, 2055 and 2085, the scaling

factor is interpolated linearly. Finally, the rainfall statistics are calculated according to Equation

7.3.

( ) 19901961,__,

_

21002071,_2086

_−

−

×= obsmonthiperiodcontrolRCM

monthi

RCMmonthi

monthi statstat

statRCMstat (7.1)

GCMperiodcontrol

GCM

GCMperiodcontrol

GCMyearGCM

year TTTT

SF_21002071

_

−−

=−

(7.2)

( ) ( )( )19901961_

2086_

19901961__

−− −×+= monthimonthiGCMyearmonthi

yearmonthi statRCMstatSFstatRCMstat (7.3)


131

As an illustration, Figure 7.3 presents the scaling factor evolution between 1975 and 2085 for the

GCM 'ECHAM4/OPYCA2' which drives the boundary conditions of RCMs 'HIRHAM_E' and

'RCAO_E' (Table 6.1). According to this graph, the rate of change is expected to be more

important at the end of the 21st century. Figure 7.4 also shows the target climatic statistics of

'RCAO_E', which projects the largest change (Figure 6.1), for 4 years (1995, 2025, 2055 and

2085). As expected, the daily mean precipitations and the probability of a dry day increase during

winter and decrease during summer. The daily skewness coefficient, which is important to get a

good representation of extremes events, increases for almost all months. The variances and

autocorrelation have more unpredictable variations, depending on the month of the calendar.

More details can be found in (Burton et al., 2010; Burton et al., 2008).

Figure 7.3. Evolution of the scaling factors between 1975 and 2085 for GCM ECHAM4/OPYCA2


132

Figure 7.4. Target climatic statistics of RCAO_E for years 1995, 2025, 2055, 2085


133

This procedure for calibrating the parameters and calculating the 'target' statistics between 2010

and 2085 was repeated for the 6 RCMs listed in Figure 6.1. To validate the calibration, target and

simulated mean climatic statistics have then been compared. For each year between 1975 and

2085, a 1000-year rainfall time series corresponding to a stationary climate equivalent to the

climatic conditions of the year, was generated using the related set of parameters. Mean statistics

were then calculated for each 1000-year time series, and compared to the corresponding target

statistics. Using 1000-years time series enables to calculate robust mean statistics, what is less

evident with shorter time periods. Some results of this comparison are presented at Figure 7.5 for

February and August of 'RCAO_E'. As shown on the graphs, the 'daily mean precipitation' and

'probability of a dry day' target statistics match closely the equivalent mean simulated statistics.

Conclusions are similar for the other statistics and calendar months.

Figure 7.5. Evolution of target and mean simulated statistics (for successive periods of 1000 years) between

1975 and 2085 for RCAO_E


134

7.2.1.4 Generation of the climate change time series for the Geer basin

Following this validation, 100 equiprobable rainfall daily time series were then generated from

2010 to 2085 by 'RainSim' for each RCM, by using the successive sets of parameters for each year

and each calendar month. Figure 7.6 presents the stochastic climate change scenarios related to

ARPEGE_H and RCAO_E. For the sake of clarity, only 30 time series are plotted on each

graph. In accordance with Figure 6.1, precipitations increase during winter and decrease during

summer, with larger changes for RCAO_E than for ARPEGE_H.

Figure 7.6. Stochastic climate change scenarios. Precipitations of RCAO_E and ARPEGE_H for February

and August

7.2.2 Generation of PET time series using the 'CRU daily weather generator'

Once the precipitation time series have been generated, the second step in the production of the

climatic scenarios consists in the generation of the other variables using the 'CRU daily weather

generator'. The methodology for generating these variables is described in detail in Kilsby et al

(2007). The complete analysis for the Geer basin is presented in Blenkinsop et al. (2010).


135

7.2.2.1 General concepts

The method, which is summarised here after, is based on observed correlation and auto-

correlation relationships between climatic variables. Daily precipitation time series are generated

separately from the other variables because precipitation is conceptualised as a clustered rainfall

event, simulated using the more adequate 'Neyman-Scott rectangular pulses' model (see Section

7.2.1), and the other meteorological variables, considered as continuous phenomena, are more

easily simulated by regression procedures. Observed data for all variables, except precipitation,

are firstly normalised (zero mean and unit standard deviation of the statistical distribution). This

normalisation is performed separately for each half month of the calendar (12 × 2), and for four

different transition types between days ('wet-wet', dry-dry', 'wet-dry' and 'dry-wet') determined

depending on the wet/dry status of the preceding and current day. A regression relationship

similar to Equations 7.4 (Kilsby et al., 2007) is then calibrated for each variable and for each of

the 96 distributions (12 × 2 × 4).

111111 dPcPbTaT iiii +++= −−

221212 dPcPbRaR iiii +++= −−

jijijijijij eRdTcPbRaX ++++= −1 (7.4)

iT : Normalised daily temperature on day i

iP : Normalised daily precipitation on day i

iR : Normalised daily temperature range (TMAX- TMIN) on day i

ijX : Remaining normalised variables on day i (j=3, #variables)

jjjjj edcba ,,,, : Correlation and auto-correlation coefficients evaluated for each

variable and each regression


136

To generate new climatic data time series, these correlation and auto-correlation coefficients are

then used in Equations 7.5, where 'r' is a random normal variable scaled depending on each

regression. These equations have the advantage of preserving the correlation and auto-correlation

between all variables. Normalised new values are then scaled back to absolute values using the

corresponding means and standard deviations (Kilsby et al., 2007).

rdPcPbTaT iiii ++++= −− 111111

rdPcPbRaR iiii ++++= −− 221212

reRdTcPbRaX jijijijijij +++++= −1 (7.5)

To generate future climate change times series, all coefficients are kept constant but the means

and standard deviations used to scale back normalised values are adapted depending on the year.

The scaling scheme used here is similar to what is done for precipitation (see Figure 7.3 and

Section 7.2.1.3).

7.2.2.2 Generation of the climate change time series for the Geer basin

For the Geer basin case study, 100 time series from 2010 to 2085 were generated for each of the

6 climatic models of Table 6.1, and for a 'control period' without any climate change, similarly to

the case of precipitation. Simulated variables are the daily mean temperature, the daily

temperature range (TMAX – TMIN), the daily mean vapour pressure, the daily mean wind speed, the

daily sunshine hours, and the daily mean solar radiation. The potential evapotranspiration is

calculated using the Penman-Montheith equation. Figure 7.7 and Figure 7.8 show the

temperature and PET time series for February and August, related to the climatic models

ARPEGE_H and RCAO_E. These graphs are consistent with the general increase in

temperature and PET during the whole year, with larger increases for RCAO_E and during

summer months (Figure 6.1).


137

Figure 7.7. Stochastic climate change scenarios. Monthly mean temperature of RCAO_E and ARPEGE_H

for February and August

Figure 7.8. Stochastic climate change scenarios. Monthly mean PET of RCAO_E and ARPEGE_H for

February and August


138

7.3 Application of the climate change scenarios on the Geer basin model

7.3.1 Simulation conditions

One hundred equiprobable climate change scenarios (2010-2085) have been generated for each

of the 6 climatic models (RCM – GCM). To save computing times, which are very large for each

hydrological simulation, only 30 scenarios were firstly applied as input of the Geer basin

hydrological model (calibration 1 with daily stresses, see Section 5). 30 additional scenarios

relative to a 'control period' (2010-2085), without any climate change, were also used as a

referential. The influence of the number of equiprobable climatic scenarios used as input of the

hydrological model is examined later in this chapter (Section 7.4.1)

In Section 7.3.2, projected groundwater levels and surface water flow rates from 2010 to 2085 are

examined, and the uncertainty linked to the natural variability of the climate is evaluated through

the calculation of confidence intervals. In Section 7.3.3, it is taken advantage of the fully transient

climate change scenarios to calculate the time of occurrence of a specific event and the

uncertainty surrounding it.

7.3.2 Evolution and uncertainty of projected groundwater levels and surface

water flow rates

Results concerning the evolution of projected groundwater levels and surface water flow rates at

the outlet of the catchment are presented from Figure 7.9 to Figure 7.13. Figure 7.9 and Figure

7.10 present the evolution of groundwater levels between 2010 and 2085 for the climatic models

ARPEGE_H and RCAO_E respectively. Figure 7.11 shows the mean groundwater levels

between 2010 and 2085 for the 6 climatic models and for the 'control period'. Figure 7.12 and

Figure 7.13 present a similar analysis for the monthly surface water flow rates at the outlet of the

catchment. At Figure 7.13, the evolution of flow rates are also shown separately for February and


139

August. For each climatic model, mean groundwater levels and mean surface water flow rates are

calculated for each day between 2010 and 2085, using the 30 scenarios results. These means

express the average behaviour of groundwater levels and flow rates given 30 outcomes. For all

climatic models and observation points, mean groundwater levels and flow rates present a

decreasing trend between 2010 and 2085. By the year 2085, mean groundwater levels are

expected to decrease by 7-20 m, and water flow rates at 'Kanne' are expected to decrease between

44% and 70%, in comparison to the mean groundwater levels of the 'control simulations'. This

decrease is slightly higher than what was expected using the same hydrological model with the

'quantile mapping bias corrected' climatic scenarios, which projects a decrease of 4-19 m and 37-

67%, respectively (see Section 6). The different curves of mean groundwater levels and surface

water flow rates also show clear seasonal variations, which are the combination of higher or

lower seasonal fluctuations visible on each individual scenario. The quick evolution of

groundwater levels and surface water flow rates during the first years of the simulations show

that the initial conditions are sometimes inadequate. This is of little importance since induced

differences are quickly reduced, within two or three years. Additionally, the change in

groundwater level and water flow rate is not calculated in comparison to this initial state but in

comparison to 'control simulations' without any climate change.


140

Figure 7.9. Evolution of hydraulic heads at the 8 observation wells for 30 equiprobable climatic scenarios of

ARPEGE_H (2010 – 2085)


141

Figure 7.10. Evolution of hydraulic heads at the 8 observation wells for 30 equiprobable climatic scenarios

of RCAO_E (2010 - 2085)


142

Figure 7.11. Mean hydraulic heads at the 8 observation wells for each of the 6 climatic models (30 scenarios)


143

Figure 7.12. Evolution of water flow rate at the outlet of the basin for 30 equiprobable climatic scenarios of

ARPEGE_H and RCAO_E (2010 - 2085)

Figure 7.13. Mean water flow rate at the outlet of the basin for each of the 6 climatic models (30 scenarios)

The added value of using a large number of stochastic climatic scenarios is that it enables to

evaluate the uncertainty linked to the natural variability of climate. Figure 7.14 present the mean

groundwater level at 'OTH002' and the mean water flow rate at the outlet of the catchment

('Kanne'), as well as the enveloping intervals containing 95% of the climate change scenarios for


144

the control simulations and the two contrasted climatic models ARPEGE_H and RCAO_E. For

each climatic model and for each particular time between 2010 and 2085, it means that the 'true'

future scenario has a probability of 0.95 to be included within the respective enveloping interval,

when considering only the uncertainty linked to the natural variability of climate. The 95%

intervals are calculated considering that the distributions of groundwater levels and surface water

flow rates at each specific time are normal and lognormal, respectively. The results presented at

Figure 7.14 show that the enveloping intervals related to the different climatic models and the

'control simulations' are overlapping. This is observed just after 2010, but also at the end of the

century when climate changes are higher. The uncertainty linked to the natural variability of the

climate is then high, around 10 metres concerning groundwater levels at 'OTH002'. Nevertheless,

by the year 2085, the 95% intervals of ARPEGE_H and RCAO_E, which are expected to give

the lowest and highest decreases (see Figure 7.11), are entirely situated below the mean curve

related to the 30 'control' simulations. It indicates that, even if the uncertainty of projections

remains high, it is very likely that groundwater levels and surface water flow rates will decrease.

Finally, it should also be noted that the 95% interval tends to decrease with groundwater levels.

This is due to the increasing importance of the partially saturated zone which acts as a buffer and

attenuates all climatic fluctuations from the ground surface to the water table.


145

Figure 7.14. (A) Mean groundwater levels (30 scenarios) and 95% interval at observation point 'OTH002', (B) mean annual water flow rates (30 scenarios) and 95% interval at 'KANNE', for the control simulations

and the climatic models ARPEGE_H and RCAO_E.

7.3.3 Temporal uncertainty of a specific event

As already mentioned, another advantage of using the 'weather generator' downscaling technique

is that it allows simulating the change of mean climatic statistics in a fully transient way. It is then

possible to evaluate the impact of 'cumulated' climate change on groundwater reserves, rather

than using stationary climatic models. Using fully transient climatic models also enables to make

additional interesting analyses for water management. With usual downscaling techniques, it was

possible to answer the question about the increase or decrease of groundwater levels for three

stationary climates, representatives of three specific time slices (2011-2040, 2041-2070, 2071-

2100). With the downscaling technique used in this chapter, it is also possible to make it in

reverse and answer the question about when is it expected to reach a specified decrease in


146

groundwater levels, and to evaluate the associated uncertainty. As an illustration, a specific

application is performed here for the Geer basin where the following question is answered:

"When is it expected to have a continuous 10 m decrease of the groundwater level at OTH002?"

For each climate change scenario, the year of occurrence of a first 10 m decrease, relatively to the

mean groundwater level of the control simulations, was pointed out. An additional criterion was

that the 10 m decrease must be observed during 10 consecutive years after its first occurrence. To

overcome the effect of seasonal variations which could bias the analysis, yearly groundwater

levels were used. At Figure 7.15, the different years of occurrence are regrouped into 8 time

intervals between 2010 and 2085, and the number of outcomes for each time interval is plotted

separately for each climatic model. The climatic models RCAO_E and HIRHAM_E project the

fastest decreases. Inversely, ARPEGE_H shows a wider distribution with the highest numbers of

realisations occurring latter in the century. Using these distributions, it is then possible to evaluate

the uncertainty of the time of occurence and to calculate some confidence intervals. At Figure

7.16, a probability density function (pdf) of mean and standard deviation equal to 2040 and 15,

respectively, is associated to the distribution of HIRHAM_H. According to this normal pdf, the

95% interval for year of occurrence would be included between 2010 and 2069. This interval is

wide and is also probably influenced by the large natural multi-annual variability of groundwater

levels in the Geer basin.

To conclude, this kind of analysis can be easily reproduced for each climatic model, for each

location in the basin and for any particular impact.


147

Figure 7.15. Number of outcomes (10 m decrease in OTH002 groundwater level) for each time interval and

each climatic model.

Figure 7.16. Probability density function and 95% interval for HIRHAM_H (OTH002)

7.4 Verification simulations of modelling hypotheses

7.4.1 Influence of the number of equiprobable climatic scenarios

In the previous analysis, only 30 stochastic scenarios for each climatic model (GCM – RCM)

have been used as input of the hydrological model, though 100 scenarios have actually been

generated for each of them. This choice was made because of very large computing times. A

unique simulation between 2010 and 2085 with daily input precipitation and potential


148

evapotranspiration takes more than 20 days on a 3.0 Ghz Pentium4 desktop machine equipped

with 4 Gb RAM. Considering the number of scenarios and climatic models, it results a huge

volume of calculation to be performed by computers, and the question arises about the adequate

number of climatic scenarios to be used for each climatic model. To answer this question, 100

scenarios of ARPEGE_H have been used as input of the hydrological model and the results are

here compared with the previous ones, performed using 30 scenarios only. These comparisons

are shown at Figure 7.17 and Figure 7.18. Figure 7.17A shows the mean groundwater level and

the 95% interval at the observation point 'OTH002' between 2010 and 2085, calculated using 30

and 100 climatic scenarios of ARPEGE_H. The difference between both cases is very small.

Although the multi-annual slope of the mean groundwater curve and of the 95% interval limits

are more regular when using 100 scenarios, results and underlying interpretations remains

basically identical. Conclusions are similar concerning the annual flow rates at the outlet of the

catchment (see Figure 7.17B). At Figure 7.18, a similar comparison is performed with the years of

occurrence of a first 10 m decrease in groundwater level at 'OTH002' during a minimum period

of 10 years. The statistical distributions are evaluated successively for 30 and 100 scenarios of

'ARPEGE_H'. The probability density functions are then associated to these distributions, using

the calculated means and standard deviations, and the intervals containing 95% of the outcomes

are finally calculated. Using 30 climatic scenarios, the 95% interval is included between 2015 and

2077 around a mean of 2046. Using 100 climatic scenarios, the 95% interval is included between

2015 and 2082 around a mean of 2048. Although there is a difference between both cases,

especially for the upper limit of the 95% interval, this difference remains low in comparison to

the width of the interval. Nevertheless, using more scenarios may help identifying more easily the

type of distribution that characterises the different outcomes of the test. At Figure 7.18, the

distribution obtained from the 100 scenarios fits better a normal law than the distribution

obtained using only 30 scenarios.


149

Figure 7.17. Results comparison when using 30 or 100 equiprobable scenarios of the climatic model

ARPEGE_H. (A) Mean groundwater levels and 95% interval at observation point 'OTH002'. (B) Mean annual water flow rates and 95% interval at 'KANNE'.

Figure 7.18. Comparisons of probability density functions and 95% intervals for ARPEGE_H (OTH002)

using 30 and 100 equiprobable scenarios. The vertical dotted lines represent the limits of the 95% interval.


150

7.4.2 Influence of the time discretisation

Another possibility for reducing computing times is to input all stresses (precipitations, potential

evapotranspiration and water abstraction volumes) on a monthly basis, using monthly means,

instead of daily means. Using smoothed stress factors allows expecting to have longer adaptive

time steps in HydroGeoSphere and thus shorter total simulation times. However, it may also be

expected that smoothed variations can also induce discrepancies in the modelling results as

compared to daily stress factors. Indeed, if precipitation and potential evapotranspiration are

smoother with lower extremes, this may have a significant impact on run-off and groundwater

recharge and, consequently, on groundwater levels and 'base flow'. As an example, a heavy rain

occurring during a unique day will not have the same effect on groundwater recharge than the

same amount of precipitation distributed on a whole month. The objective here is thus to

evaluate (1) the benefit in simulation time and (2) the impact of using smoothed stress factors.

For 30 equiprobable scenarios of ARPEGE_H, climate change impacts on groundwater levels

and water flow rates at the outlet of the catchment were then evaluated using respectively daily

and monthly mean precipitation and potential evapotranspiration. When using monthly stress

factors, a simulation between 2010 and 2085 is reduced to approximately three days instead of

more than 20 days. The saved CPU time is then considerable. Means and 95% intervals between

2010 and 2085 are presented at Figure 7.19 for both cases. Figure 7.19A shows groundwater

levels at the observation point 'OTH002'. Figure 7.19B and C show monthly water flow rates at

'Kanne' for February and August respectively. Concerning groundwater levels, very little

differences between both cases are observed and mean groundwater levels from all 30 scenarios

are globally similar. The upper limit of the 95% interval is slightly higher when using monthly

inputs, indicating that the simulated groundwater recharge is slightly more important, especially in

periods of high groundwater levels. Differences appear when considering water flow rates at the

outlet of the catchment ('Kanne'). The upper limit of the 95% interval and, to a lesser extent, the


151

mean water flow rates are generally lower when using monthly precipitation and potential

evapotranspiration inputs. The 95% interval is reduced, indicating a decrease in the flow rate

variability and a lower dependance of flow rates at the outlet of the catchment to surface run-off.

For February and August, the maximum decreases of the 95% interval are equal to -30.6% and -

27.3%, respectively. This observation is in accordance with the use of mean monthly stresses

which smooth all precipitations events and favour water infiltration rather than surface run-off.

The lower limits of the 95% intervals are however unchanged when using daily or monthly

inputs. This is due to the fact that the lowest monthly flow rates occur during very dry periods

with none or very few precipitation events. During these periods, flow rates in rivers are basically

equal to the groundwater discharge or 'base flow' and the influence of using daily or monthly

precipitations is therefore quite limited.

Considering all these observations and interpretations, the impact of using monthly mean stresses

remains quite limited with the Geer basin model. Although significant decreases in the upper

limit of the 95% intervals can be observed, mean groundwater levels and mean flow rates from

all scenarios are very similar. This small influence can be explained by geographical and

hydrogeological considerations. First, the thick layer of unsaturated loess located above the chalk

aquifer smoothes infiltration between the ground surface and the top of the water table (see

Section 4). The influence of daily precipitations is then already limited. Second, the Geer basin is

characterised by very flat areas, where run-off is limited. Water in the Geer River is then mostly

fed by groundwater discharge, less sensitive to daily precipitations.


152

Figure 7.19. Results comparison when using daily or monthly input solicitations for 30 equiprobable

scenarios of the climatic model ARPEGE_H. (A) Mean groundwater levels and 95% interval at observation point 'OTH002'. (B) Mean monthly water flow rates and 95% interval at 'KANNE' (February). (C) Mean

monthly water flow rates and 95% interval at 'KANNE' (August).


153

7.5 Discussion

The objectives of the study presented in this section were to (1) simulate flow conditions using

climatic scenarios that simulate climate change in a full transient way between 2010 and 2085; and

(2) to evaluate the uncertainty of projected groundwater levels and surface flow rates, due to the

natural variability of the weather. To reach these objectives, a state-of-the-art 'weather generator'

downscaling technique have been used to generate a large number of equiprobable stochastic

scenarios simulating full transient climate change between 2010 and 2085, for each of 6 different

climatic models (GCM – RCM) and for an additional 'control' climatic model without any climate

change. These climate change stochastic scenarios have then been applied as input of the Geer

basin hydrological model. This methodology constitutes a real innovation since such stochastic

climate change scenarios have never been used in combination with hydrological models. Here,

the advantages of these climatic scenarios are additionally combined with those of a fully

integrated surface – subsurface hydrological model, and this innovative methodology is one of

the most advanced in the fields of groundwater modelling and climate change.

The application of 30 climate change scenarios, for each climatic model (GCM – RCM), as input

of the Geer basin model enabled to calculate 95% confidence intervals around projected

groundwater levels and surface flow rates, for each specific time between 2010 and 2085. Results

show that these confidence intervals are relatively wide, especially when adding the uncertainty

linked to the use of different regional and global climatic model (GCM – RCM), as shown in

Figure 7.14 for observation point 'OTH002'. Although this total range of uncertainty is quite

high, it is however entirely located below the mean groundwater level of the 'control simulation'

at the end of the century. This indicates that it is very likely that groundwater levels will decrease,

even if the decrease intensity remains uncertain.


154

As already mentioned, the main drawback linked to this methodology is the very large computing

time required to perform the hydrological simulations between 2010 and 2085. In this study, all

simulations were performed using 'NIC3', a super computer for intensive calculation at

University of Liège. This computer is equipped with 1296 different cores and enables to run large

numbers of simulations at the same time. Nevertheless, these kinds of computer are not always

available and several options when running hydrological simulations can be used to reduce

computing times. In Section 7.4.1, the influence of the number of climatic scenarios used as input

of the hydrological model is examined. The study showed that using only 30 equiprobable

scenarios for each climatic model is sufficient to get a satisfactory estimation of the mean

expected groundwater levels or surface water flow rates and the range of uncertainty around

these mean values. Using only 30 scenarios by climatic model then allows limiting computing

times as the total number of simulations is also reduced. Another way for reducing computing

times is to play on the time discretisation of the input stresses. With the Geer basin model, using

monthly mean precipitation and potential evapotranspiration rates instead of daily means enables

to increase the general size of the adaptive time steps and to reduce total computing time for

each simulation by a factor of 8 approximately. At section 7.4.2, a comparison of the impact

results, achieved with daily and monthly stresses, shows that the differences between both cases

are quite limited for the Geer basin hydrological model. The climate change impact calculated

with daily and monthly mean stresses are almost similar except a small increase of the uncertainty

intervals.

The results and conclusions achieved in this study are specific to the case of the Geer basin, and

have to be verified for each particular hydrogeological context and model. As an example, the

same analysis performed with the second hydrological model (calibrated using monthly mean

stresses) for which multi-annual variations are lower, would probably lead to different intervals.

As a consequence, and for a comprehensive analysis, the uncertainty linked to the natural


155

variability of the climate has to be compared with other kind of uncertainty, such as the

uncertainty linked to the use of the hydrological model. It is the objective of Section 8, where all

uncertainty intervals are regrouped and compared.


156

7.6 References

Blenkinsop, S., Fowler, H.J., Harpham, C., Burton, A. and Goderniaux, P., 2010. Modelling transient climate change with a stochastic weather generator. Projected temperature changes for the Geer catchment, Belgium. Journal of hydrology, submitted.

Burton, A., Fowler, H.J., Blenkinsop, S. and Kilsby, C.G., 2010. Downscaling transient climate change using a Neyman-Scott Rectangular Pulses stochastic rainfall model. Journal of hydrology, in press.

Burton, A., Kilsby, C.G., Fowler, H.J., Cowpertwait, P.S.P. and O'Connell, P.E., 2008. RainSim: A spatial-temporal stochastic rainfall modelling system. Environmental Modelling & Software, 23(12): 1356-1369.

Cowpertwait, P.S.P., 1991. Further developments of the Neyman-Scott clustered point process for modeling rainfal. Water Resources Research, 27(7): 1431-1438.


Neyman, J. and Scott, E.L., 1958. Statistical approach to problems of cosmology. Journal of the Royal Statistical Society, Series B 20 (1): 1-43.


8. Uncertainty linked to the calibration of the model and summary of the results

157

8. UNCERTAINTY LINKED TO THE CALIBRATION OF

THE MODEL AND SUMMARY OF THE RESULTS


158


159

8.1 Introduction and objectives

The objective of this chapter is to discuss and classify the diverse sources of uncertainty that may

affect the predictions of climate change impact on groundwater resources. In Section 7, the

'weather generator' downscaling technique enabled to generate a large number of equiprobable

climate change scenarios. The application of these scenarios as input of the Geer basin

hydrological model allowed assessing the uncertainty linked to the natural variability of the

climate. This kind of uncertainty relates to the physics of climatic phenomena. On the other

hand, additional uncertainty is linked to the modelling of these climatic and hydrologic

phenomena. The use of 6 different regional climate model experiments (GCM – RCM) allowed

evaluating the uncertainty linked to the existence of diverse general and regional atmospheric

models. Climate change impact on the Geer basin groundwater resources was also evaluated

using climatic scenarios generated with two different downscaling methods. An additional source

of uncertainty is linked to the use of hydrological models. As all kinds of numerical models,

hydrological models are actually simplified representations of the reality, developed for satisfying

specific objectives. They can be used for predictive purposes if they are able to satisfactorily

reproduce the reality, in accordance with the objectives of the study. This ability of models needs

to be carefully evaluated.

Section 8.2 presents the evaluation of the uncertainty linked to the calibration of the Geer basin

hydrological model. Section 8.3 summarises and compares all results about climate change impact

uncertainty in the Geer basin. Section 8.4 gives some conclusions.


160

8.2 Estimation of the uncertainty linked to the calibration of the

hydrological model

The ability of hydrological models to reproduce the reality is determined by the model

calibration. The calibration of the Geer basin model has been evaluated in Section 5. However,

although this evaluation is needed, it is also not sufficient because it does not necessarily give

information about the accuracy and precision of predictions. A model which presents a good fit

to observed data may give poor predictions under stress conditions that are different from those

used in the calibration procedure (Henriksen et al., 2003; Hill and Tiedeman, 2007; Konikow and

Person, 1985; Refsgaard, 1997). This is particularly the case here where precipitation and

potential evapotranspiration vary under climate change conditions. Consequently, it is really

important to evaluate objectively the quality of the simulated predictions.

Several methods exist to evaluate the uncertainty linked to the calibration of models. The

'random sampling Monte Carlo' method consists in running many simulations with different

parameters values and examining what are the impacts on predictions (Bedford and Cook, 2001).

This method is quite straightforward but becomes difficult to implement when computing time

for each simulation is large. Additionally, it is sometimes hazardous to choose the ranges of

variations and statistical distributions that govern parameters values. In this study, the prediction

uncertainty linked to model calibration is quantified using the 'inferential methods' described in

details in Hill and Tiedeman (2007). Such methods are usually applied following a successful

termination of an automatic inverse calibration. They use the statistics describing the model fit to

observations and the results of the parameters sensitivity analysis (Sections 5.1.5.2 and 5.2) to

evaluate the uncertainty related to the estimated parameter values. This uncertainty calculated for

parameters is then related to the predictions uncertainty thanks to the calculated sensitivity of the

predicted values to the parameters. The general methodology and theory are described in Section


161

8.2.1. The application of this methodology with the Geer basin hydrological model is presented is

Section 8.2.2. Results about climate change predictions uncertainty are presented in Section 8.2.2.

8.2.1 Methodology

Main concepts and equations of the method used in this chapter are briefly presented here after.

They come from Hill & Tiedeman (2007), where the method is described in details.

The conditions to apply the method are: (1) the conceptual model must be correct, (2) the

observation measurement errors must be unbiased, and (3) the calibrated model and simulated

residuals must be unbiased. While the first two conditions are verified before the model

construction or assumed, the third condition has to be verified through the evaluation of the

model fit at the end of the calibration procedure. The second step to estimate the parameters and

predictions uncertainty consist in evaluating the 'calculated error variance', using Equation 8.1, as

made in Section 5.1.5.2 for the Geer basin model. The weight for each observation is calculated

as the inverse of the true measurement error variance. As explained in Section 5.1.5.2, this

enables to account for the errors linked to the observations measurements and gives higher

weights to more certain observations. The theory presented here after is based on the assumption

that this weighting scheme is followed.

Calculated error variance : ( )[ ]

( )NParNObs

byyws

NObs

iiii

−

−=∑ 2

2'

(8.1)

iw : Weight for observation i

b : Vector of parameters

iy : Observed value i

iy' : Computed value i as a function of ( )b


162

NObs : Number of observations


The 'calculated error variance' expresses the variance of model errors (i.e. differences between

observed and computed values) for the calibration data set. This variance or uncertainty is related

to the parameters through the sensitivities of simulated values to parameters. In Equation 8.2,

0bijX are the components of the sensitivity matrix 0b

X calculated for the set of parameters 0b

from the optimised model. The matrix X has a number of rows and columns equal to the

number of observations and parameters, respectively.

j

ibij b

yX

∂∂

='

0 (8.2)

ijX : Sensitivity of the computed value i to the parameter j

iy' : Computed value i

jb : Parameter j

The parameter variance – covariance matrix ( )bV (NPAR × NPAR) is then calculated using

Equation 8.3. Linear confidence intervals on parameter values can be calculated with Equation

8.4, where the standard deviation of parameter j can be easily calculated using the corresponding

variance in the variance – covariance matrix ( )bV . A 95% linear confidence interval on a specific

parameter is the interval that has a 95% probability of containing the 'true' value of the

parameter.


163

( ) ( ) 12 −⋅⋅×= XwXsbV T (8.3)

2s : Calculated error variance

X : Sensitivity matrix

w : Matrix containing all weights

( )jbj sntb ×−± 2/0.1, α (8.4)

( )2/0.1, α−nt : Student t-statistic

n : Degree of freedom (NObs - NPar)

α : Level of significance

jb : Parameter j

jbs : Standard deviation of the parameter j

These confidence intervals are useful when numerical models are used to get good estimation of

parameter values. However, it is usually not the case, and numerical models are more often used

for predictive purposes. In this context, it is more interesting to calculate confidence intervals on

predicted values. The uncertainty on parameter values can be related to the predictions by using

the sensitivities of the predicted values to the parameters. The prediction standard deviations can

be calculated using Equation 8.5. Linear 'individual' and 'simultaneous' predictions confidence

intervals are calculated using Equations 8.6 and 8.7 respectively. 'Individual' intervals have a given

probability of including the 'true' predicted value considering a unique prediction. 'Simultaneous'

intervals have the same given probability of including the 'true' predicted valued simultaneously

for all predictions. To illustrate the difference between both kinds of intervals, Hill and Tiedeman

(2007) take the example of 100 equivalent models with different sets of parameters. This example

is here adapted to the modelling of the Geer basin. A 95% 'individual' confidence interval for a


164

predicted groundwater level at specific location and time will contain 95 of 100 simulated

groundwater levels independently of the other predictions (at other locations and times).

Concerning the 95% 'simultaneous' confidence intervals at each location and time, simulated

groundwater levels or flow rates from the same 95 of 100 models will be included in the intervals.

'Simultaneous' intervals are always equal or larger than the 'individual' intervals.

( )2

1

'''

⎥⎥⎦

⎤

⎢⎢⎣

⎡

∂∂

××∂∂

= ∑ ∑NPAR

i i

lNPAR

jij

j

lp b

pbV

bp

sl

(8.5)

lps ' : Standard deviation of the prediction l

lp' : Predicted value l

jb : Parameter j

( )bV : Parameters variance – covariance matrix


( )lpsl sntp '2/0.1,' ×−± α (8.6)

( )lpBl skntp '2/0.1,' ×−± α (8.7)

( )2/0.1, α−nts : Student t statistic

( )kntB 2/0.1, α− : Bonferroni t statistic

n : Degree of freedom (# observations - # parameters)

α : Level of significance

lp' : Predicted value l

k : Number of simultaneous intervals

lps ' : Standard deviation of the prediction l


165

Equations 8.6 to 8.7 allow calculating linear confidence interval for estimated parameters and

predictions. These linear intervals involves that the model is linear in the vicinity of parameter

values and, more particularly, that parameter sensitivities are constant within calculated intervals.

Linear intervals can be sufficient for moderately non-linear models. However, they become less

accurate as the degree of non linearity increases. Methods exist to calculate non-linear confidence

intervals and are described by Vecchia and Cooley (1987) and Hill & Tiedeman (2007). However,

they are very computationally intensive and they are impossible to apply with models that require

huge computing times. The appropriateness of calculating linear confidence intervals with the

Geer basin model is discussed later in Section 8.2.3.

8.2.2 Application of the methodology on the Geer basin hydrological model

The concepts and equations presented here above have been applied in the context of the climate

change impact study on the Geer basin groundwater resources. The objective is then to evaluate

the uncertainty of simulated groundwater levels and flow rates due to the use of the hydrological

model, under climate change conditions.

The methodology has been applied to the case of the Geer basin model calibrated with daily

stresses (calibration 1, see Section 5). The predictions correspond to monthly groundwater levels

and surface water flow rates at the same observation points as for the calibration procedure (8

observation wells and the 'Kanne' gauging station, see Figure 4.1), but under climate change

conditions. In this study, the predictions uncertainty has been evaluated for climate change

corresponding to the climatic model 'HIRHAM_H - 2071-2100' (Figure 6.1), which presents

medium climate change compared to the other RCMs. The predictions uncertainty is calculated

for absolute groundwater levels and flow rates and for the associated decreases in comparison

with the equivalent groundwater levels and flow rates without any climate change. When

calculating predictions sensitivities, this implies running twice more simulations for each


166

parameter, as hydraulic conditions without climate change are needed to calculate the equivalent

decreases. To limit computing times, the 'predictive' period was limited to 6 years. It corresponds

to the solicitations of the period 1984 – 1990, scaled by the change factors of 'HIRHAM_H -

2071-2100' for the simulations with climate change. Predictions are evaluated after two years of

simulation to let the hydrological model accommodate the initial conditions.

Practically, it implies evaluating the calibrated model, calculating the observations sensitivities (i.e.

computed values corresponding to the observations of the calibration step with identical

solicitations) and calculating the predictions sensitivities. These analyses have been performed

using the computer code 'UCODE_2005' (Hill et al., 2003; Poeter et al., 2005). The evaluation of

the calibrated model and the observations sensitivities have been calculated for the period 1967 –

2003 and are presented in Section 5.2. As explained in Section 5.1.5.2, the weights of

groundwater level and surface water flow rates observations have been assigned so that the 95%

confidence interval including the true groundwater level is within 1 metre around the measured

groundwater level, and the coefficient of variation of the flow rate measurement errors is equal to

10%. Given these assumptions, the weights for observed groundwater levels are equal to 3.84m-2,

and the weights for surface water flow rates observations vary between 3.7 and 141.7 m-6s2. Using

these weights and the simulated variables, the 'calculated standard error' s, used in Equation 8.3,

is equal to 10.2. The observations sensitivities allow calculating the parameters variance –

covariance matrix7 (Equation 8.3), which is used in Equation 8.5 to calculate the prediction

uncertainty.

7 The parameters linked to the calculation of the actual evapotranspiration have not been adjusted during the calibration stage and are specified on the basis of prior information from the scientific literature. This prior information helps reducing the uncertainty on the parameters values and has been incorporated in the uncertainty evaluation by adding lines and columns in the matrices of Equation 8.3. More details about how to use prior information can be found in Hill and Tiedeman (2007).


167

Predicted values and 95 % confidence intervals are presented in Figure 8.1 and Figure 8.2 for

groundwater levels at the 8 observation points and surface water flow rates at the outlet of the

catchment ('Kanne'), respectively. The mean range of the 'simultaneous' 95 % intervals around

absolute groundwater levels varies between 6.9 m at the observation point A7-PL37 and 28.9 m

at MOM001. The mean range of the 'individual' 95 % intervals around absolute groundwater

levels varies between 2.4 m and 9.8 m for the same observation points. Concerning surface water

flow rates, the mean 'simultaneous' and 'individual' 95% intervals are equal to 3.9 m³/s and 1.4

m³/s, respectively. The intervals are however larger for winter than for summer, with the

uncertainty also spreading in the area corresponding to flow rates increase. Generally, the

confidence intervals around predicted absolute values (Figure 8.1A) are generally similar to

equivalent intervals around predicted changes (Figure 8.1B). Though some variations of the range

of these intervals are observed during the 6-years time period, no clear trend can be identified, for

most of the observation points. At 'MOM001', however, the range of the confidence interval

clearly increases with time.

8.2.3 Discussions about the confidence intervals

The confidence intervals presented at Figure 8.1 and Figure 8.2 have 95 % chance of including

the 'true' predicted value considering the model calibration and the stresses specified for the

period of interest. These are linear confidence intervals, evaluated given the calibrated parameters

values and considering that all sensitivities are constant. As already explained, these sensitivities

can vary significantly with non-linear models, and this is obviously the case for the surface –

subsurface model used in this study. If non-linearity is severe, the linear confidence intervals can

become inaccurate. This is well illustrated at Figure 8.2A, where the linear 95% confidence

intervals of predicted flow rates spill out onto the area of negative flow rates. This situation is

obviously not realist and is due to the non-linear nature of the hydrological model, especially

when conditions become very dry. In this case, the lower limit of the flow rate linear confidence


168

intervals is inaccurate and the lower range of the interval is overestimated. Vecchia & Cooley

(1987) developed an alternative method for calculating non-linear confidence intervals. However,

the method requires scanning ranges of variation of parameters values and performing several

sensitivity analyses to find to minimum and maximum predicted values. According to Hill and

Tiedeman (Hill and Tiedeman, 2007), calculating prediction non-linear intervals is often longer

and more difficult than performing a complete non-linear automatic regression of the parameters.

Considering the computing times for a single transient simulation with the Geer basin

hydrological model, this task was not feasible within reasonable times.

Additional source of potential inaccuracy of the confidence intervals relates to the fact that the

Van Genuchten parameters are not included in the analysis, due to technical incompatibility of

the input tables with 'UCODE_2005' processes (see Section 5.2). Though the Van Genuchten

parameters are not adjusted during the calibration stage and specified on the basis of laboratory

experiments (Brouyère, 2006), a limited uncertainty remains for these parameters. Neglecting it in

the evaluation of the uncertainty of the predictions could induce an underestimation of the

confidence intervals. Nevertheless, though inaccurate in some cases, the linear confidence

interval is a good indicator of the predictions uncertainty. They are particularly useful when

compared with uncertainty of other types, what is made in Section 8.3.

The different linear confidence intervals have been calculated with the first hydrological model

(Calibration 1). Results are specific for this model and could hardly be transposed to the case of

the second hydrological model (calibration 2) described in Section 5.4, or any other model.

However, performing the same uncertainty analysis with this second model is not possible as the

calibration results do not satisfy the conditions required for such an analysis. Actually, as shown

in Figure 5.9 to Figure 5.11, the second model presents a bias, with an overestimation of almost

all groundwater levels and surface water flow rates.


169

Figure 8.1. Predictions and 95 % confidence interval around predicted values for 8 years of a HIRHAM_H

downscaled climate change scenario. (A) Absolute groundwater levels. (B) Groundwater levels change between a scenario without any climate change and the HIRHAM_H climate change scenario.


170

Figure 8.2. Predictions and 95 % confidence interval around predicted values for 8 years of a HIRHAM_H downscaled climate change scenario. (A) Absolute surface water monthly flow rates. (B) Change in surface

water monthly flow rates between a scenario without any climate change and the HIRHAM_H climate change scenario.

8.3 Summary of the results about climate change impact uncertainty

In this study, several tools have been used to assess the uncertainty surrounding the climate

change impact estimations for the groundwater reserves of the Geer basin. The results achieved

are summarised and compared in Table 8.1, Figure 8.3 and Figure 8.4, for the period 2071-2100.

Uncertainty linked to Method Type of interval

calculated

Range of the interval for groundwater levels

(Min – Max)

Range of the interval for annual flow rates

('Kanne') (Min – Max)

the natural variability of the weather

use of 180 equiprobable climatic scenarios 95% confidence interval 2.3 – 18.6 m 1.01 – 2.55 m³/s

the climatic models use of 6 RCMs

experiments total range of variation 3.2 – 11.3 m 0.99 m³/s

the calibration of the hydrological model

inferential statistical methods

('UCODE_2005')

95% linear confidence interval 6.9 – 28.7 m 3.92 m³/s

the downscaling method

use of 2 different downscaling method total range of variation 0.7 – 4.8 m 0.05 – 0.35 m³/s

Table 8.1. Summary of the results about climate change impact uncertainty for the period 2071-2100 (calculated with the first hydrological model, calibrated with daily stresses).

8.3.1 Uncertainty linked to the natural variability of the climate

The uncertainty linked to the natural variability of the climate has been evaluated using a large

number of equiprobable climate change scenarios generated using the 'CRU weather generator'

downscaling technique. The whole set of projected groundwater levels and flow rates has then


171

been used to calculate 95% confidence intervals (see Section 7). The range of groundwater levels

confidence intervals varies between 2.3 m and 18.6 m depending on the observation point and

the climatic model. Annual flow rates confidence intervals vary between 1.01 and 2.55 m³/s,

depending on the climatic model (Table 8.1), with larger intervals during winters than during

summers. Generally, the range of the confidence intervals tends to decrease as the importance of

the partially saturated zone, which smoothes the climatic variations, becomes higher.

8.3.2 Uncertainty linked to the climatic models

Uncertainty has been evaluated from various possible sources, by using several different input

possibilities or by using more complex methods allowing the calculation of confidence intervals.

The first source of uncertainty is linked to the general and regional climatic models (GCMs and

RCMs) used before applying statistical downscaling. This source is the only one considered in

most existing studies about groundwater and climate change. Here, the possible range of

variation for projected groundwater levels has been evaluated using 6 different RCMs with

boundary conditions from 2 different GCMs (see Table 6.1). This number is limited but has been

maximised from the PRUDENCE ensemble by the choice of the climatic models, as explained in

Section 6. The ranges of variation of groundwater levels and flow rates due to this source of

uncertainty are mainly dependent on the location in the Geer basin and the hydrological model

used. The highest and lowest decreases are usually projected by 'HAD_P_H' and 'ARPEGE_H',

respectively.

8.3.3 Uncertainty linked to the downscaling technique

In this study, two statistical downscaling techniques have been used to downscale RCMs outputs:

the 'Quantile Mapping Bias Correction' technique and the 'CRU weather generator'. The

projected mean groundwater levels achieved from both kinds of downscaled climatic data are

very similar, though decreases are slightly more important with the 'weather generator'


172

downscaled data. This difference remains anyway very small compared to the uncertainty from

other possible sources. However, it should be noted that the two downscaling techniques used

here are part of the most sophisticated, both of them involving corrections not only to the mean

of climatic variables but also across their statistical distribution. Therefore, it is finally consistent

that the difference is small. Discrepancies would probably be higher if comparisons between

sophisticated downscaling methods and more simple ones were performed.

8.3.4 Uncertainty linked to the calibration of the hydrological model

Finally, the uncertainty linked to the calibration of the hydrological model was estimated by

coupling 'UCODE_2005' with 'HydroGeoSphere' (Section 8.2). The mean range of the 95%

confidence intervals around predicted groundwater levels varies between 6.9 m and 28.7 m. The

mean confidence interval around predicted flow rates is equal to 4.1 m³/s (Table 8.1), with larger

intervals during winters than during summers. This kind of uncertainty was also previously

'tested' by using two hydrological models with different calibrations. As already mentioned, the

second model (calibration 2) gives significantly less important decreases for the period 2071-

2100. As shown in Figure 8.3, these differences between both models are sometimes more

important than the range of the equivalent confidence interval calculated with 'UCODE_2005'.

Several reasons can explain these greater differences. (1) The confidence intervals are linear and

may be inaccurate where the model presents too important non-linearity (see Section 8.2.3). (2)

The calibration of the second hydrological model presents a significant bias (see Section 5.4) and

does not satisfy the conditions necessary to apply the statistical methods used in 'UCODE_2005'.

Therefore, it is also hazardous to compare the results of this model with confidence intervals

produced by 'UCODE_2005'. (3) The range of the 95% confidence interval may be

underestimated because the Van Genuchten parameters are not included in the analysis (see

Section 8.2.3). (4) Finally, the 95% confidence intervals cover only 95% of the whole set of

possibilities, by definition.


173

Figure 8.3. Summary of all results and uncertainties for the 8 groundwater observation points. The

horizontal red line represents the 'control' mean groundwater level without any climate change.


174

Figure 8.4. Summary of all results and uncertainties for the water flow rate at the outlet of the catchment

('Kanne'). The horizontal line represents the 'control' mean flow rate without any climate change.

8.4 Conclusions about uncertainty

The research works performed in the framework of this PhD thesis enabled to evaluate the

uncertainty from various sources. Following the summary of the results presented here above, it

is possible to rank these sources of uncertainty from the most to the less significant. The

uncertainties linked to the calibration of the hydrological model and the natural variability of the

climate appear to be the most important in this case. The uncertainty linked to the use of

different general and regional climatic models (GCMs and RCMs) comes in third position,

followed by the uncertainty linked to the use of downscaling methods. This evaluation of

uncertainties and their ranking is specific to this study and could hardly be transposed to other

models or case studies, without a detailed analysis. Nevertheless, it could constitute a guide in the

arduous and difficult task of uncertainty evaluation. Particularly, it shows the importance and the

usefulness of evaluating the uncertainty linked to the calibration of the model.


175

Among all uncertainties summarised here, two different kinds have to be differentiated. First, the

uncertainty linked to the natural variability of the climate is inherent to the climatic phenomenon,

and can not be eliminated whatever happens. Calculated intervals do not traduce a model 'error'

but express how groundwater levels or surface water flow rates will inevitably vary with time

around an average position, due to the occurrence of dry and wet periods. At the other hand, the

other uncertainties are linked to 'errors' or 'approximations' made in climate or hydrological

modelling. The direction of the simulated 'error' does not necessarily vary with time. These

uncertainties can be reduced by improving the models or the knowledge necessary to implement

them. As an example, a model that fits better the observations will also gives more confident

predictions.

Finally, additional sources of uncertainty have not been considered in this study. All climatic

scenarios used here correspond to the A2 greenhouse gases emissions scenario (medium-high)

(see Section 2.3.3). Considering lower emissions, such as the B2 scenarios (medium-low), would

lead to less important changes. Déqué et al. (2007) however showed that, when using climate

models, the uncertainty from emissions scenarios was lower than the uncertainty from GCMs

and RCMs over Belgian latitudes. Another uncertainty not considered in this study is linked to

the conceptual model used when implementing the hydrological model. This uncertainty relates

to potential changes that would be induced by other boundary conditions, different numbers of

finite elements in the grid or alternative zonations of the hydraulic properties, for example. The

evaluation of such uncertainty has been studied by Rojas (2009) and Rojas et al. (2009) for a

regional aquifer in Chile, but not in the context of climate change impact.

Considering all these results about impacts and uncertainties, it is very likely that groundwater

levels in the Geer basin will decrease by the end of the century, due to climate change. Actually,

calculated impacts from both hydrological models project significant decreases over the whole

basin for this period. Additionally, most part of the confidence intervals or ranges of variations


176

(Figure 8.3) are comprised below the 'control' groundwater level corresponding to the 1980s

climate without any change. Considering all uncertainties, it is however more difficult to evaluate

what will be the intensity of the decrease. Some confidence intervals are actually large compared

to the projected decreases. Furthermore, the uncertainties related to modelling 'errors' or

approximations (calibration of the hydrological model, climatic downscaling techniques,

modelling of general or regional models, etc.) may superimposed, inducing larger global

uncertainty around final predicted values. Concerning surface water flow rates at the outlet of the

catchment, conclusions are similar for annual flow rates. Uncertainty is however larger for winter

flow rates than for summer flow rates, with the uncertainty also spreading in the area

corresponding to flow rates increase.


177

8.5 References

Bedford, T. and Cook, R.M., 2001. Probabilistic Risk Analysis, Foundations and Methods. Cambridge University Press., Cambridge, UK.

Brouyère, S., 2006. Modelling the migration of contaminants through variably saturated dual-porosity, dual-permeability chalk. Journal of Contaminant Hydrology, 82: 195-219.

Déqué, M. et al., 2007. An intercomparison of regional climate simulations for Europe: assessing uncertainties in model projections. Climatic Change, 81(0): 53-70.

Henriksen, H.J., Troldborg, L., Nyegaard, P., Sonnenborg, T.O., Refsgaard, J.C. and Madsen, B., 2003. Methodology for construction, calibration and validation of a national hydrologic model for Denmark. Journal of Hydrology, 280: 52-71.

Hill, M.C., Poeter, E., Zheng, C. and Doherty, J., 2003. MODFLOW 2001 and other modeling odysseys. Ground Water, 41(2): 113.


Konikow, L.F. and Person, M., 1985. Assessment of Long-Term Salinity Changes in an Irrigated Stream-Aquifer System. Water Resour. Res., 21.


Refsgaard, J.C., 1997. Parameterisation, calibration and validation of distributed hydrological models. Journal of Hydrology, 198(1-4): 69-97.

Rojas, R., 2009. Uncertainty analysis in groundwater modelling: An integrated approach to account for conceptual model uncertainty. PhD Thesis, Katholiek Universiteit Leuven. Faculty of Science, Leuven, 150 pp.

Rojas, R., Batelaan, O., Feyen, L. and Dassargues, A., 2009. Assessment of conceptual model uncertainty for the regional aquifer pampa del Tamarugal. Hydrology and Earth System Sciences Discussions, 6(5): 5881-5935.

Vecchia, A.V. and Cooley, R.L., 1987. Simultaneous confidence and prediction intervals for nonlinear regression models with application to a groundwater flow model. Water Resources Research, 23(7): 1237-1250.


178

9. Conclusions and perspectives

179

9. CONCLUSIONS AND PERSPECTIVES


180


181

9.1 Conclusions

The objectives of this research are the following:

(4) development of a methodology for a reliable estimation of the climate change impacts on

groundwater reserves;

(5) estimation of the uncertainty characterising these projected impacts, considering various

possible uncertainty sources;

(6) pilot application of the two first objectives on the case of the Geer basin catchment

(Belgium).

These objectives have been set in response to the lack of research carried out in the fields of

climate change and groundwater. They also respond to some gaps or oversimplifications made in

previous studies. This study then presents a robust methodology and guidelines that can be used

to assess impacts of climate change on groundwater reserves and the uncertainties surrounding

these. The methodology provides new outcomes at three levels: hydrological modelling, climate

change scenarios, and methods for uncertainty evaluation. This new methodology has been

applied to the case of the Geer basin groundwater reserves.

9.1.1 Hydrological modelling

To assess the climate change impact on groundwater reserves of the Geer basin, a 'physically-

based', 'spatially-distributed', integrated surface – subsurface hydrological model has been

developed with the finite element code 'HydroGeoSphere'. This model enables a more realistic

representation of the system for three reasons:


182

- Interconnected flow processes, such as groundwater recharge, are better represented because

the flow equations are solved simultaneously in all domains (surface, partially saturated zone,

saturated zone).

- Integrated models allow using both surface and subsurface observed data in the calibration

procedure. Compared to stand-alone groundwater models, this reduces correlations between

parameters and enables to check more easily that some components of the water balance are

well represented.

- The model is 'physically-based', which is more indicated especially when extrapolations are

made.

This study also outlines the importance and usefulness of a reliable and objective evaluation of

the calibration. The calibration of the Geer basin model has been evaluated using weighted

residuals which enable to better identify possible biases in the model.

9.1.2 Climate change scenarios

The climate change scenarios used in this study have been statistically downscaled from RCMs

using two different downscaling techniques: the 'Quantile Mapping Bias Correction' technique

and the 'Weather Generator' ('RainSim' and 'CRU' WG). Both of them are sophisticated

techniques and are able to project changes not only to the mean values of climatic variables but

also to the shape of their statistical distribution. The 'Quantile Mapping Bias Correction'

technique is used to generate one climatic scenario for 30-years time periods representing a

stationary climate over the selected period. The 'weather generator' technique is even more

sophisticated. It enables to generate large numbers of equiprobable climatic scenarios,

representative of a transient climate change between 2010 and 2085. When applying these

scenarios as input of hydrological models, they offer the opportunity to evaluate impact from a


183

probabilistic point of view. Because they simulate full transient climate change, they also enable

to analyse the timing of potential impact. Nevertheless, running many climate change scenarios as

input of hydrological models requires important computer resources. Both techniques then

present specific advantages. The choice of the technique to be used depends on the objectives of

the study and the computing resources available.

In this study, it was also examined the effect of using larger time steps, which could substantially

reduce the total time of calculation, particularly when applying a large number of different climate

change scenarios. With the Geer basin model, it was shown that using mean monthly input

stresses instead of daily inputs implies limited changes in the range of confidence intervals. It was

also shown that applying 100 or 30 equiprobable climate change scenarios as input of the

hydrological model leads to very similar results.

9.1.3 Uncertainty evaluation

The uncertainty affecting projected groundwater levels has been evaluated regarding various

uncertainty sources.

- The uncertainty linked to the climatic models has been evaluated by using a multi-model

ensemble RCMs and GCMs.

- The uncertainty linked to the natural variability of the climate has been estimated by applying

a large number of equiprobable climatic scenarios as input of the Geer basin hydrological

model. Climatic scenarios have been generated with 'RainSim' and the 'CRU Weather

Generator'.

- The uncertainty linked to the downscaling procedure has not been studied extensively. Only

two different techniques, among the most advanced, have been tested to generated climate

change scenarios and apply them as input of the Geer basin hydrological model


184

- The uncertainty linked to the calibration of the hydrological model has been evaluated using

the capabilities of the computer code 'UCODE_2005'.

9.1.4 Impact for the Geer basin

The climatic models relative to the Geer basin area project a pattern of much hotter and drier

summers and warmer and wetter winters. According to these models, the annual temperature

mean change varies between +3.1°C and +5.6°C, and the annual precipitation mean change

varies between -1.9% and -15.3%, for the period 2070-2100. Considering these climatic models

and the results achieved in this study, it is very likely that groundwater levels will significantly

decrease by the end of the century. For the period 2070-2100, groundwater levels are expected to

decrease by 2-19m depending on the location in the Geer basin, the climatic and hydrological

models. For the same period, annual water flow rates at the outlet of the catchment are expected

to decrease by 9% - 65%, with a more pronounced change during summers than during winters.

This is of concern because it also means that the groundwater quantities available for abstraction

will also decrease, while the aquifer is already intensively exploited. Furthermore, the water

demand may increase in the future, due for example to an intensification of irrigation practices.

For the period 1967-2003, mean groundwater abstraction represents 6.4% of precipitation or

~16% of the 'effective water'. 'Effective water' is equal to precipitation minus actual

evapotranspiration, and should then be partitioned between groundwater recharge and surface

runoff. Considering the results of Table 6.2 and Table 6.3, and the assumption that groundwater

abstraction will remain constant over the 21st century, it is expected that the proportion of

groundwater abstraction relatively to the 'effective water' term would be included between 16%

and 44% by the end of the century, depending on the climatic and hydrological models used.

However, this study also showed that the uncertainty surrounding projected groundwater levels

in the Geer basin are relatively large. Consequently, it remains difficult to state on the intensity of


185

the decrease with high confidence. Main uncertainty is linked to the natural variability of the

climate and the calibration of the hydrological model. While the first one expresses how

groundwater levels will vary around a mean position, the second one relates to modelling

'approximations' and could be reduced by improving the model calibration, if needed.

9.1.5 General conclusion

The methodology developed in this study combines the advantages of a fully-integrated surface –

subsurface model, sophisticated climate change scenarios and methods to evaluate uncertainties

from various sources. The tools applied in this study have never been used previously in the

context of climate change impact on groundwater reserves. Their use and combination

constitutes an innovation and advances the study of climate change impacts on groundwater

reserves. Particularly, the fact that projections are calculated along with uncertainties gives

credibility to the study and constitutes a very important tool for helping water managers to take

decisions. Results achieved in this thesis are specific to the case study of the Geer basin but the

methodology applied and the main conclusions about it can be reproduced in other catchments

and contexts.

9.2 Perspectives

The integrated surface – subsurface model of the Geer basin offer numerous additional

applications for river basin management. The present study focuses on the direct impacts of

climate change on groundwater reserves but other factors may also affect indirectly but

importantly, the groundwater reserves in the context of climate change. Examples of such factors

are the evolution of land use or changes to agricultural practices. Through the use of appropriate

scenarios, it should be possible to test the influence of such indirect factors on the Geer basin

groundwater reserves. Similarly, the model could also be used to examine what would be the

effect and usefulness of mitigation measures such as artificial recharge for example. Additionally,


186

problems of contaminant accumulation (e.g. salts, pesticides, fertilisers) are commonly observed

in basins characterised by intensive agriculture activities. These issues offer new opportunities to

further use and develop the model to address contaminant transport problems.

One difficulty of using catchment scale integrated models lies in the fact that computing times

can become very large. This issue prevents of makes difficult some applications. As examples, an

inverse automatic calibration or the calculation of non-linear confidence intervals with

'UCODE_2005' was not possible in this study, due to unworkable computing times.

Nevertheless, a lot of research is currently performed in the field of integrated modelling, and

future new developments may help reducing these times. Particularly, most of the computing

codes which perform integrated surface – subsurface flow simulations will shortly be able to run

simulations on several processors in parallel. Other developments also include the simplification

of flow calculation in the partially saturated zone. Calculating one-dimensional vertical flow

instead of three-dimensional flow in this zone might significantly reduce computing times

without affecting too much the performance of the model. Such new developments that will

facilitate the calibration and other applications should also improve the attractiveness of long-

term catchment-scale integrated modelling.

Appendix

187

APPENDIX

Appendix

188

Appendix

189

Parameterisation of the second Geer basin hydrological model, calibrated

with monthly stresses

Van Genuchten parameters Residual water saturation Total porosity Specific storage

α [L-1] β [-] Swr [-] n [-] Ss [L-1]

Chalk formations 0.099 1.10 0.023 0.44 1×10-4

Loess formations 7.57 1.16 0.024 0.41 1×10-4

Table A2. Van Genuchten parameters, total porosity and specific storage

Name K [LT-1]

Chalk 1 4×10-5 Chalk 2 1×10-3 Chalk 3 3×10-5 Chalk 4 2×10-6 Chalk 5 2×10-5 Lo

wer

cha

lk


Chalk 1 1×10-4 Chalk 2 1×10-3 Chalk 3 1×10-5 Chalk 4 1×10-4 Chalk 5 5×10-5

Inte

rmed

iate

cha

lk


Chalk 1 1×10-4 Chalk 2 1×10-3 Chalk 3 1×10-4 Chalk 4 1×10-4 Chalk 5 1×10-4 U

pper

cha

lk


Quaternary loess 1×10-8

Tertiary deposits 0.3×10-7 - 1×10-7

Table A3. Full saturated hydraulic conductivities values of the calibrated zones (results of calibration)

X-Y friction [L-1/3T] Coupling length [L-1/3T]

Rural 3 0.01

Urban 0.3 0.01

Forested 6 0.01

Table A4. Values for the Manning roughness coefficients and coupling length

Appendix

190

Rural crop (temperate)

Rural grassland (temperate)

Rural broadleaf deciduous forested

(temperate) Urban

Root depth Lr [L] 2.1 2.6 5.2 0.0

Evaporation depth Le [L] 2.0

Max. LAI [-] 4.22 2.50 5.12 0.40

Cint [L] 1×10-5

C1 [-] 0.3

C2 [-] 0.2

C3 [-] 10

Table A5. Root depths, evaporation depths and Leaf Area Index

Rain Actual evapotransp. North boundary Outlet (‘Kanne’) Water abstraction Water balance

error

mm/year 798.6 -502.3 -37.5 -209.2 -51.1 1.5

% of rainfall 100 -62.9 -4.7 -26.2 -6.4 0.2

Table A6. Simulated mean water balance terms for the period 1967-2003

Goderniaux P., 2010. Impact of Climate Change on Groundwater Reserves. PhD Thesis. University of Liège, Faculty of Applied Sciences. Liège, Belgium. pp. 190.

impact of climate change on groundwater...

Documents