Rhine at risk?

Impact of climate change on low-probability floods in the Rhine basin and the effectiveness of flood management measures

Aline te Linde

Cover design and picture: Sophie Valkenier Stamps from the collection of: Johan te Linde Painting: ‘Blick vom Isteiner Klotz rheinaufw¨rts richtung Basel’, Peter Birmann, 1830 a Printed by: W¨hrmann Print Service, Zutphen o ISBN 978-90-8570-742-4

VRIJE UNIVERSITEIT

Rhine at risk?
Impact of climate change on low-probability floods in the Rhine basin and the effectiveness of flood management measures

ACADEMISCH PROEFSCHRIFT

ter verkrijging van de graad Doctor aan de Vrije Universiteit Amsterdam, op gezag van de rector magnificus prof.dr. L.M. Bouter, in het openbaar te verdedigen ten overstaan van de promotiecommissie van de faculteit der Aard- en Levenswetenschappen op donderdag 12 mei 2011 om 13.45 uur in de aula van de universiteit, De Boelelaan 1105

door

Asselina Hermina te Linde geboren te Deventer

promotor: copromotoren:

prof.dr J.C.J.H. Aerts prof.dr. A.J. Dolman dr. J.C.J. Kwadijk

leescommissie:

prof.dr A. Bronstert dr.ing. R. Lammersen prof.dr. H. Middelkoop prof.dr.ir. N.C. van de Giesen prof.dr. B.J.J.M. van den Hurk prof.dr.ir. P. Vellinga

This research was carried out at: Institute for Environmental Studies (IVM) Faculty of Earth and Life Sciences VU University Amsterdam De Boelelaan 1085 (visiting address) De Boelelaan 1087 (postal address) 1081 HV Amsterdam The Netherlands and Deltares Rotterdamseweg 185 (visiting address) P.O. Box 177 (postal address) 2600 MH Delft The Netherlands This research was carried out in the framework of the Dutch National Research Project ‘Climate changes Spatial Planning’ (wwww.klimaatvoorruimte.nl).

De zon smelt in de bergen de sneeuw om tot het water dat glinsterend in beken de flanken langs krioelt Zo kringelend haar weg zoekt tot het zich weer verzamelt en stampend de rivier wordt die de rivier zich voelt

Kom dan blauwe regen Val maar naar benee Kom dan blauwe regen Spoel me met je mee Kom dan blauwe regen Stroom me in de zee

Door dammen en met sluizen laat de rivier zich temmen neemt boten zwaarbeladen en geschiedenissen mee Ik zit hier aan de kade en luister naar de stemmen die aan de einder opgaan in het water van de zee

Kom blauwe regen, De Dijk, 2005

Opgedragen aan wijlen Jaap Griede († 08–11–2006)

viii

Preface

Fluvial flooding is a natural phenomenon creating the agricultural conditions and water availability relied on by humans. Fertile soils in river valleys are deposited by regular flooding, and river flows maintain the groundwater levels necessary for crop growth. Rivers provide water for drinking, industry, irrigation, sanitation, and transportation. Therefore, for centuries river valleys have attracted human settlement and economic development. However, flooding can be problematic in areas having such economic value. Consequently, there are efforts to mitigate the hazard through the development of flood defense measures. Large rivers are often canalized to constrain their outwash, and valuable land may be protected by dikes. In highly-developed and densely populated river basins, such as the Rhine basin, flood defense measures have become sophisticated over the years. The question remains, however, whether current flood management practices are adequate in the face of climate change and further socioeconomic development. Flood frequency is low in river basins with existing flood management practices, leading to a certain level of safety. However, a flood event caused by dike failure or overtopping can still have major consequences. Flood mitigation approaches must optimize the balance between the investment costs of the flood defense system and the level of protection it affords. However, these costs and benefits vary both temporally and spatially. Public perception and socio-economic conditions influence this balance, but these factors and local policy change over time and may vary throughout a region, particularly in cross-boundary river basins. History has shown that a flooding event that causes damage and casualties often triggers or accelerates decision making of flood risk management. These decisions are then based on the best available knowledge at the time.

ix

The future, however, is inherently uncertain and both the probability of a flood and the potential damage it may cause will change over time. This increases the challenge of designing flood defense measures. The potential for property damage and casualties depends on population increases, economic growth, and land-use changes in flood prone areas. The probability of flooding will vary according to changes in the river network and the hydraulic profile, the addition or removal of flood defense measures, and land-use changes such as deforestation and afforestation. Finally, climate change and variability have an influence on the flooding probability, primarily through altered temperature and precipitation, and thus needs to be considered in flood risk assessment and flood management. Erroneous or missing information may lead to undesirably low protection levels and possibly large hazards, while over-protection might lead to excess investment and possible unnecessary and undesirable landscape changes. Gathering essential information and knowledge to aid flood management requires extensive research and analysis. The Rhine is a unique river basin for conducting research on the impact of climate change on extreme flood peaks and flood risk, and on the effectiveness of flood management measures. Seven countries share its basin and the Rhine has high economic value. Consequently, research and data on the river network and discharge, related land use, population wealth and other characteristics, soil properties, and meteorology are available for a considerable period of time. In addition, models exist for simulating the regions climate, hydrology and hydraulics, land use and flood damage potential. It is expected that climate change will have a major impact on the discharge regime of the Rhine. Global warming may cause increased precipitation and earlier snow melt, with peak discharges likely to advance from spring to winter. This can increase the frequency of flooding and thus the flood risk. Therefore, for the riparian countries within the Rhine basin, establishing the effect of climate change on flood risk is an urgent issue. Flood defense measures are designed to prevent the occurrence of extreme water levels at the so-called design discharge. However, there are methodological challenges in the design process due to the high safety standards in the Rhine basin (varying from 1/200 to 1/1250 per year). These high safety standards imply that defense measures need to protect against flood events that most likely have not yet been observed to date. Thus, the estimation of the probability and duration of design flood events is subject to scientific debate, in particular when considering climate change scenarios. In Europe, water management is moving away from a flood defense approach to a more risk management based approach, which takes both the probability and the potential consequences of flooding into account. To aid in the planning of a flood risk management approach, basin-wide information is needed on the current probability of flooding, maximum water levels of extreme flood peaks, and the potential damx

age from flooding. In addition, methods are needed to estimate the effect of climate and socio-economic changes on future trends in flood risk. However, modeling the impact of climate change in order to estimate the probability of flooding and test the effectiveness of flood management measures entails various difficulties and uncertainties. Improvements to this modeling process have received a great deal of attention in current research. Nonetheless, to date, no attempt has been made to estimate future flood risks on a basin-wide scale for the Rhine basin. In short: is the Rhine at risk? In this thesis, I will describe methods for optimizing the simulation of the discharge regime and flood-peak probability in the Rhine basin, for present conditions and under future climate change. This thesis will also provide estimates for future flood risk related to different climate change and socio-economic scenarios. In addition, I will show how these methods can be applied to test the effectiveness of flood management measures in the Rhine basin. Finally, I will reflect on the implications of the results and provide new insight into the uncertainties in flood risk estimation in the Rhine basin. These are different to how these uncertainties are currently perceived. The conclusions are relevant for flood management decision making. Before specifying the objective and research questions of this thesis, the background, key questions, and available research methods are introduced and specified for the Rhine basin.

xi

xii

Contents

Preface 1 Introduction 1.1 Problem definition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.1 1.1.2 1.1.3 1.2 1.2.1 1.2.2 1.2.3 1.3 1.3.1 1.3.2 1.4 1.5 Economic loss due to floods . . . . . . . . . . . . . . . . . . . . . . . . Flood frequency and climate change . . . . . . . . . . . . . . . . . . . Towards flood risk management . . . . . . . . . . . . . . . . . . . . . . Basin characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . Discharge characteristics and flood peak generation . . . . . . . . . . . Flood defense measures . . . . . . . . . . . . . . . . . . . . . . . . . . Simulating the effect of climate change on extreme flood peaks . . . . Estimating potential damage and future flood risk . . . . . . . . . . .

ix 1 1 1 2 5 6 6 7 7 10 10 15 16 17 19 20 21 23 23 25

The Rhine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Available research and remaining challenges . . . . . . . . . . . . . . . . . . .

Objective and research questions . . . . . . . . . . . . . . . . . . . . . . . . . Thesis outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

2 Comparing model performance of two rainfall-runoff models 2.1 2.2 2.3 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Divergent concepts in rainfall-runoff modeling . . . . . . . . . . . . . . . . . . Model description and study area . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1 2.3.2 VIC . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . HBV . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xiii

2.3.3 2.4 2.4.1 2.4.2 2.4.3 2.5 2.5.1 2.5.2 2.6

Rhine basin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Forcing data comparison . . . . . . . . . . . . . . . . . . . . . . . . . . Model performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Forcing data comparison . . . . . . . . . . . . . . . . . . . . . . . . . . Model performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

25 26 26 27 27 31 31 31 39 43 44 46 46 47 48 51 53 54 56 56 59 59 60 62 64 68 72 75 77 78 80 82 82 86

Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

Discussion and conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3 Simulating low probability peak discharges 3.1 3.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The Rhine basin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 3.2.2 3.3 Geographical Characteristics . . . . . . . . . . . . . . . . . . . . . . . Hydrological Characteristics . . . . . . . . . . . . . . . . . . . . . . . .

Methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.1 3.3.2 3.3.3 Generating long meteorological time series . . . . . . . . . . . . . . . . Hydrological models . . . . . . . . . . . . . . . . . . . . . . . . . . . . Extreme value analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3.4

Evaluation of changes in precipitation and temperature 3.4.1 3.4.2

Comparing data for the control climate . . . . . . . . . . . . . . . . . Comparing ECHAM5-RACMO with the delta change approach . . . . . . . . . . . .

3.5

Evaluation of simulated discharge and flood-peak probabilities 3.5.1 3.5.2 3.5.3 3.5.4

Effect of meteorological bias correction . . . . . . . . . . . . . . . . . . Performance of simulated discharge at sub-basin scale . . . . . . . . . Climate change impact on monthly mean discharge . . . . . . . . . . . Climate change impact on flood-peak probabilities . . . . . . . . . . .

3.6 3.7

Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Conclusions and further work . . . . . . . . . . . . . . . . . . . . . . . . . . .

4 Future flood risk estimates 4.1 4.2 4.3 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Case study area: The Rhine basin . . . . . . . . . . . . . . . . . . . . . . . . Data and method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.1 4.3.2 Current and future land use . . . . . . . . . . . . . . . . . . . . . . . . Inundation map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

xiv

4.3.3 4.3.4 4.4 4.4.1 4.4.2 4.4.3 4.4.4 4.5

Flood damage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Climate change scenarios for changes in flood probabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Discharges and probabilities . . . . . . . . . . . . . . . . . . . . . . . . Land-use change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Flood damage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Flood risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

87 87 88 88 89 91 94 97 101

Simulation results

Discussion and conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5 Effectiveness of flood management measures 5.1 5.2

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102 Rhine basin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 5.2.1 5.2.2 General description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 Flood management in the Rhine basin . . . . . . . . . . . . . . . . . . 105 Hydrological modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 Simulating long discharge series (Steps 1, 2 and 3) . . . . . . . . . . . 108 Selection of 16 flood waves (Steps 4 and 5) . . . . . . . . . . . . . . . 109 Hydraulic modeling and description of measures (Steps 6 and 7) . . . 110 Basin-wide effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122

5.3

Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 5.3.1 5.3.2 5.3.3 5.3.4

5.4

Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 5.4.1 5.4.2 5.4.3 5.4.4 Local effects of flood management measures . . . . . . . . . . . . . . . 123 Longitudinal profiles of peak water levels . . . . . . . . . . . . . . . . 126 Flood-peak probability at Lobith . . . . . . . . . . . . . . . . . . . . . 128 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130 Effectiveness of measures . . . . . . . . . . . . . . . . . . . . . . . . . 132 Further work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 135

5.5

Discussion and conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130 5.5.1 5.5.2 5.5.3

6 Synthesis and recommendations 6.1 6.2

Summary of results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 Research questions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 138 6.2.1 6.2.2 6.2.3 6.2.4 Comparing model performance . . . . . . . . . . . . . . . . . . . . . . 138 Simulating climate change . . . . . . . . . . . . . . . . . . . . . . . . . 139 Future flood risk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141 Effectiveness of flood management measures . . . . . . . . . . . . . . . 142

xv

6.3

Implications and recommendations . . . . . . . . . . . . . . . . . . . . . . . . 143 6.3.1 6.3.2 6.3.3 6.3.4 6.3.5 The consequences of ignoring uncertainty . . . . . . . . . . . . . . . . 143 How to implement uncertainty in the calculations of the design discharge146 Lowering peak water levels effective at local scale . . . . . . . . . . . . 147 Risk-based approach can be beneficial basin-wide . . . . . . . . . . . . 148 Toward communicating uncertainty and adaptive management . . . . 150

6.4

Remaining challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150 153

7 Summaries 7.1 7.2

English summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 Nederlandse samenvatting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156 161 165 167 169 171 193 197 201

A The HBV model B The SOBEK model C The GEV distribution D Performance indicators Bibliography Dankwoord Publications related to this thesis Curriculum Vitae

xvi

List of Figures

1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9

Map displaying the flood extent of the 1926 flooding in the Netherlands . . . Maps of the Rhine basin: a) sub-basins and major cities, b) elevation, and c) land use. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Inland waterway transport flows in Europe in 2007 . . . . . . . . . . . . . . . Canalization of the Upper Rhine. . . . . . . . . . . . . . . . . . . . . . . . . . Measures in the Dutch Spatial Planning Key Decision ‘Room for the River’. . IPCC SRES scenarios for global warming . . . . . . . . . . . . . . . . . . . . Hydrological response to precipitation . . . . . . . . . . . . . . . . . . . . . . Design discharge at Lobith . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3 4 7 8 10 11 12 14 15 17 24 30 32 33 35 36 38 46 49

Inundation depth map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

1.10 Outline of the thesis. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 2.2 2.3 2.4 2.5 2.6 2.7 3.1 3.2 3.3 Map of the Rhine basin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ERA15 versus CHR versus CRU precipitation . . . . . . . . . . . . . . . . . . Monthly precipitation values for the Rhine basin . . . . . . . . . . . . . . . . Daily simulation results of the HBV model . . . . . . . . . . . . . . . . . . .

Scatter plots of observed and simulated discharge . . . . . . . . . . . . . . . . Performance criteria daily discharge values . . . . . . . . . . . . . . . . . . . . Performance criteria monthly discharge values . . . . . . . . . . . . . . . . . . Map of the Rhine basin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Flowchart displaying all modeling steps . . . . . . . . . . . . . . . . . . . . .

Box-whisker plots of monthly biases of the 134 sub-basins in mean precipitation 55

xvii

3.4 3.5 3.6 3.7 3.8 3.9

Box-whisker plots of monthly biases of the 134 sub-basins in the coefficient of variation of the precipitation . . . . . . . . . . . . . . . . . . . . . . . . . . . Box-whisker plots of basin-wide average 10-day precipitation . . . . . . . . .

56 57 58

Extreme value plot of resampled 10-day precipitation values . . . . . . . . . .

Box-whisker plots of observed and simulated discharges for the control climate 60 Monthly mean change in discharge according to KNMI’06 scenarios and biascorrected RACMO output . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Mean change in winter discharge . . . . . . . . . . . . . . . . . . . . . . . . . 65 66 67 71 81 83 90 91 93 95 96

3.10 Extreme value distributions and GEV fits of yearly maximum discharge at Lobith . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.11 Simulated change in discharge at T =200 year at sub-basin scale . . . . . . . . 4.1 4.2 4.3 4.4 4.5 4.6 4.7 Maps of the Rhine basin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Flowchart of the method used for estimating future flood risk. . . . . . . . . . Extreme value distributions and GEV fits of annual maximum discharge at Lobith . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Land-use maps for 2000 and 2030 . . . . . . . . . . . . . . . . . . . . . . . . . Potential damage (a) and flood risk (b), aggregated to seven regions along the Rhine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Basin-wide annual expected flood damage (risk) for 2030, compared to the reference situation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Annual expected flood damage, for the reference situation and projections for 2030 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8

Rhine basin and flood prone areas . . . . . . . . . . . . . . . . . . . . . . . . 104 Discharge wave at Worms before and after canalization of the Upper Rhine . 106 Flowchart describing all steps of the method . . . . . . . . . . . . . . . . . . . 107 Effect of strategies and the Wplus climate change scenario on peak discharge 116 Effect of strategies and the W-plus climate change scenario on peak water level117 Effect of APF2020 retention measures on flood peaks with different return periods at Lobith, with and without the simulation of flooding . . . . . . . . 124 Longitudinal profile of the change in peak water level . . . . . . . . . . . . . . 127 Extreme value plots for the yearly discharge maxima at Lobith for different climate conditions and measures . . . . . . . . . . . . . . . . . . . . . . . . . 129

A.1 Illustration of discharge formation in the HBV model . . . . . . . . . . . . . . 163

xviii

List of Tables

2.1 2.2 2.3 2.4 2.5 2.6 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 4.1 4.2 4.3 4.4

Basin and sub-basin characteristics . . . . . . . . . . . . . . . . . . . . . . . . Performance criteria daily and monthly discharge values at Lobith . . . . . . Observed and simulated mean, minimum and maximum discharge . . . . . . Performance criteria daily discharge values . . . . . . . . . . . . . . . . . . . . Performance criteria monthly discharge values . . . . . . . . . . . . . . . . . . Analysis of peak flows and low flows at the outlet of the basin . . . . . . . . . Basin and sub-basin characteristics . . . . . . . . . . . . . . . . . . . . . . . . Delta values of climate change scenarios for the year 2050 . . . . . . . . . . . Observed temperature and precipitation, compared to simulated historical temperature by RACMO . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

26 28 34 37 39 40 47 50 55

Basin and sub-basin discharge characteristics for the control climate (1961–1995) 63 Seasonal change in discharge at Lobith for 2050 . . . . . . . . . . . . . . . . . Statistical parameters of peak discharges for the control climate and two climate change scenarios for 2050 . . . . . . . . . . . . . . . . . . . . . . . . . . Discharge at different return periods for the control climate and under climate change . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Relative change in discharge at Lobith at different return periods . . . . . . . Relative change in discharge at all sub-basins at different return periods . . . Suitability maps used for the ‘Land Use Scanner’. . . . . . . . . . . . . . . . . Climate change scenarios for increased flooding probabilities in 2030 . . . . . Surface percentages of different land-use classes in the flood prone area . . . 64 66 68 68 72 86 89 90 92

Expected damage for different regions in 2000 and 2030 . . . . . . . . . . . .

xix

4.5 4.6 4.7

Expected damage for different land-use categories in 2000 and 2030 . . . . . . Basin-wide annual expected damage (risk) in e million per year . . . . . . . . Annual expected damage (risk) in e million per year for different regions in 2000 and 2030 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

93 94 95

5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9

Measures along the Rhine . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 Current land use and the measure land-use change to forest . . . . . . . . . . 113 Properties of the measure restored abandoned meanders . . . . . . . . . . . . 114 Effect of planned measures for 2020 and W-plus climate change scenario on peak discharge . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 Effect of planned measures for 2020 and W-plus climate change scenario on peak water level . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 Effect of additional measures on peak discharge . . . . . . . . . . . . . . . . . 120 Effect of additional measures on peak water level . . . . . . . . . . . . . . . . 121 Effect of the bypass around Cologne . . . . . . . . . . . . . . . . . . . . . . . 126 Estimated return periods at Lobith . . . . . . . . . . . . . . . . . . . . . . . . 130

xx

Chapter

1

Introduction

1.1
1.1.1

Problem definition
Economic loss due to floods

According to the United Nations, within two decades nearly 60% of humanity will live in urban areas (UN-HABITAT, 2008). More than two-thirds of the world’s large cities are vulnerable to rising sea levels and river discharges, exposing ∼ 400 million people to the risk of extreme floods and storms (Nicholls et al., 2008). In the past decades, the number of fatalities and the economic loss caused by floods have increased considerably worldwide (Munich Re AG, 2008). This increase can be attributed to either climate change (Vellinga et al., 2001), population growth, or the increase in wealth in low-lying and often densely populated deltas (Bouwer et al., 2007; Pielke Jr. et al., 2008), or a combination of these causes. In Europe, floods cause about 38% of the total economic losses due to natural hazards, making it, together with windstorms, the most costly type of natural disaster (EEA, 2008). In recent decades, Europe suffered over 100 major damaging fluvial floods, including catastrophic floods along the Danube and Elbe rivers in 2002 and 2006 (Thieken et al., 2005), the summer floods in 2007 in Great-Britain, and more recently in 2010 along the Wisla and Oder in Poland and Germany. Between 1970

1

2

Chapter 1. Introduction

and 2006, floods have caused some 1300 fatalities, the displacement of more than half a million people and at least e 110 billion of economic losses (normalized to 2006 values) (Barredo, 2009). The Rhine has a long flood history that caused casualties and severe damage (Glaser and Stangl, 2003; Blackbourn, 2006). Floods in 1993 and 1995 resulted in e 1.4 and e 2.7 billion damage, respectively (Engel, 1997; Brakenridge and Anderson, 2008). The probabilities of occurrence for the 1993 and 1995 events are relatively high and have been estimated at 1/25 and 1/50 per year, respectively (Te Linde et al., 2010). The peak discharges in those events resulted from heavy, long lasting rainfall in the uplands of the Middle and Lower Rhine, combined with mild temperatures that caused premature melting of snow (Caspary, 1996). The largest riverine flood in the Netherlands in the 20st century occurred in 1926 (12 600 m3 /s at Lobith, Figure 1.1). In January 1995, the peak discharge reached 11 885 m3 /s at Lobith (Figure 1.2), and parts of Germany flooded. Within the Netherlands, 250 000 people were evacuated, but against expectations, the Dutch dikes held (Bezuyen and van Duin, 1998; Disse and Engel, 2001). To illustrate how hazardous floods can be in the Rhine basin, Herget and Meurs (2010) used historical descriptions of flood events and marks of maximum water levels on buildings to estimate that an exceptional event in the year 1374 reached almost 24 000 m3 /s at Cologne (Figure 1.2). Sprong (2009), however, estimated a peak discharge maximum of ∼ 18 000 m3 /s for the same event. Regardless, this level is still 50% higher than the discharge during the peak event of 1995. The question remains: what is the probability that such a flood event reoccurs, and what will then be the consequences?

1.1.2

Flood frequency and climate change

The Fourth Assessment report by the Intergovernmental Panel on Climate Change indicates that global warming will intensify the global hydrological cycle (IPCC, 2007b). As a consequence, the magnitude and frequency of extreme precipitation events is expected to increase, which may lead to increased probability of floods (Milly et al., 2005; Christensen and Christensen, 2007; Kundzewicz et al., 2007). In North-West Europe, both from observations and prediction models a climate change-induced trend can be derived towards wetter winter conditions and increasing flood frequencies (EEA, 2008). In several rivers, the frequency of what are now considered 100-year floods may double in the future (Dankers and Feyen, 2008). In the Lower Rhine, the peak discharge is likely to advance from spring to winter, as warming temperatures increase precipitation volumes and cause earlier snow melts (Middelkoop et al., 2001; Pfister et al., 2004). These impacts from climate change are projected to increase the mean winter discharge by 5–30% and decrease the mean

1.1. Problem definition

3

Figure 1.1: Map displaying the flood extent of the 1926 flooding in the Netherlands. The pictures are for the same event.

4

Chapter 1. Introduction

! ( ! ( ! ( ! ( ! ( ! ( ! ( ( Amsterdam ! ! ( ! ( ! ( ! ( ! (

! (

! !( ( ! (

75

! (

! (

Kilometers
! (

§

!( ( ! !Rotterdam! ! (( ! ! ! (( ( ( ! ( ! ( !! (( ! (

! ( ! ( Den Haag !Netherlands ! ! ( ( ( ! (

! ( ! (

! (

! ! Essen( ( ( !( Duisburg! !! ! ! ! ( ( ! (( ( ! !!! ! !(! ! ! ( ((( ( ( ( ( ( ! Dortmund ( ( ! ( Dusseldorf ! ! ! ( ( ! (

! (

!! (( ! ( ! (

!! ((

! (

! (

! (

! ! !! ( ( (( !! (( ! ! ! ( ( ( ! ( ! ! ( ( ! ( ! ( ! (

! (

! (

Lower Rhine ( Köln !

! (

Germany
! (
! (

! (

Belgium

! (

! (

Lahn ! ! ( (
! (

! (
! (

Frankfurt am Main
! ( ! (

! ( ! ( ! ( Luxembourg
! ( ! (

! (

! ( ! ! ( ( ! ( ! (

Main

! (

! (

! ( Middle Rhine ! ( ! (

! ( ! ( ! (
! (

! ( ! (
! ( ! (

Nurnberg

! (

! (

! ( Moselle

! ( ! (
! (

! (

! ( ! ( Stuttgart Neckar ! ( ! ( ! (

! (

!! ((
! (

! (

! (

! (

! (

! (

France

! ( ! (

! ( ! ( Upper Rhine ! (

! (

! (

Austria
! (

! (

! (

Switzerland

Legend

Residential high density
Residential low density Commercial Port areas Infrastructure Construction Recreation Forest / nature Agriculture Cultivation Pasture Water courses Sea and ocean

Population ! (
! (
! (

> 500,000 100,001 - 500,000 < 100,000

Elevation (m)
High : 4059

Low : 0

Figure 1.2: Maps of the Rhine basin: a) sub-basins and major cities, b) elevation, and c) land use.

1.1. Problem definition

5

summer discharge by 0–45% (e.g. Kwadijk, 1993; Buishand and Lenderink, 2004). As a consequence, the 1/1250 per year flood event (which is the design discharge for the Dutch embankments) at the gauging station of Lobith is estimated to increase from 16 000 m3 /s at present to somewhere between 16 500 and 19 500 m3 /s by 2050 (Kwadijk and Middelkoop, 1994; Grabs, 1997; Shabalova et al., 2003; Vellinga et al., 2008). This will increase the flood risk (Hooijer et al., 2004; Pinter et al., 2006). However, existing studies on flood risk in the Rhine basin have several shortcomings related to addressing low probability flood events. This research, therefore, aims to provide new methods and approaches to better understand the mechanisms and impacts of low probability floods in the Rhine basin.

1.1.3

Towards flood risk management

In several countries such as the Netherlands and Germany, there is a trend in flood management to convert from flood prevention (limited to controlling the water) to flood risk management which in addition considers the consequences of a potential flooding event (Plate, 2002; Merz et al., 2004; B¨chele et al., 2006; ISFD4, 2008; Merz u et al., 2010). Flood risk is defined in this thesis as the product of probability and potential damage, i.e., the expected loss per year (Smith, 2001). This trend arose from awareness that continued socio-economic development behind flood defenses increases the flood risk. De Moel et al. (2011) explain that not only the amount of economic and social capital has increased, but that developments behind flood defenses are occurring increasingly in more dangerous places, because the safer locations are already occupied. In Europe, the tendency towards flood risk management is stimulated by the EU Flood Directive (EU, 2007a), which was enacted on 26 November 2007. It sets out several actions for member states such as initiating flood risk assessments, the development of flood risk maps, and the preparation of basin-wide risk management plans, which have to be completed by the end of 2015. Many member states in Europe have already collated information and maps for flood hazard (areas at risk of damage), but information and maps for flood risk are still rare (De Moel et al., 2009). The EU Green Paper on adaptation to climate change (EU, 2007b) also stresses the importance of looking beyond defensive measures. The Dutch government has focused for many years on flood protection (Ten Brinke et al., 2008), with a recent cautious policy shift that considers the probabilities and consequences of flooding due to failure of flood defenses (Ministry of Transport, Public Works and Water Management, 2006a). In Germany, Samuels et al. (2006) found that at the federal level, flood risk management is also shifting away from flood protection to a more holistic approach that includes spatial planning and damage reduction. In practice, though, this new approach is difficult to implement, as most Bundesl¨nder a

6

Chapter 1. Introduction

draw up their own water management plans and use conventional methods of flood defense (Becker et al., 2007). Historical research revealed that river canalization and the implementation of flood defense measures in upstream parts of the Rhine basin directly affected and often increased flood risk downstream (Blackbourn, 2006). Neighboring countries in the Rhine basin see the advantage of cross-boundary cooperation, but so far they have been reluctant to transfer their political influence to the International Commission for the Protection of the Rhine (ICPR) (Aerts and Droogers, 2004; Hooijer et al., 2004). Additional research on flood risk, climate change, and the identification of effective adaptation measures in the Rhine basin can optimize future and cross-boundary river basin management.

1.2
1.2.1

The Rhine
Basin characteristics

With a length of 1320 km and a river basin area of 185 000 km2 , the Rhine is one of the larger rivers of North-West Europe (Belz et al., 2007). The river basin covers parts of Switzerland, Austria, Liechtenstein, France, Luxembourg, Germany, and the Netherlands. The main tributaries of the Rhine are the Neckar, Main, Mosel, Lahn, and Sieg (Figure 1.2). The river basin can be subdivided in five sections using geographical and geological features (Preusser, 2008). The first section is the Alpine mountains where the river originates. Second is the Upper Rhine from Basel (where the flood plain is used for agriculture) up to Mainz. Several cities exist along the Rhine branch in this section. Third is the Middle Rhine between Mainz en Bonn, which is hilly. In this section the Rhine flows through a narrow gorge. Fourth is the Lower Rhine, which is densely urbanized. From here, the flood prone area widens until it becomes a river delta in the Netherlands (fifth section) (Hooijer et al., 2004). About 50% of the basin is used for agriculture, and 15% for urban or suburban uses. The remainder is forest and otherwise fallow lands (Wessel, 1995; Eberle et al., 2005) (Figure 1.2). The Rhine is extensively used for inland shipping (Jonkeren et al., 2007; CCNR, 2009) and connects one of the world’s largest sea ports, Rotterdam, to the inland European markets (Figure 1.3). The river also provides water for cooling energy plants, and for industrial, domestic, and agricultural purposes (Grabs, 1997). The Rhine basin is one of the most heavily industrialized areas in the world. It has 58 million inhabitants, of which 10.7 million live in flood prone areas that are protected by dikes (ICPR, 2001).

1.2. The Rhine

7

Figure 1.3: Inland waterway transport flows in Europe in 2007. Percentages show the increase in shipped tons in 2007 from 2006 (Source: CCNR (2009)).

1.2.2

Discharge characteristics and flood peak generation

The average discharge of the Rhine at the German Dutch border is 2200 m3 /s and the maximum discharge since observations began in 1901 was 12 600 m3 /s, at Lobith in 1926. Both rainfall and melt water contribute to discharge generation, depending on the season (Pinter et al., 2006; Uehrlinger et al., 2009). Extreme floods in the Lower Rhine mainly occur during the winter period. Research on historical floods in the Rhine basin shows that the most extreme peaks resulted from large meteorologic low-pressure systems that slowly crossed the basin while releasing great amounts of rainfall over several days (Beersma et al., 2008). Frozen soil can contribute to more extreme peaks, as it prevents rainfall from infiltrating the ground (Engel, 1997). In addition, the sudden melting of snow and frozen soil has been reported to increase flooding (Disse and Engel, 2001).

1.2.3

Flood defense measures

In order to protect vital economic activities in the Rhine basin, flood risk management has been primarily limited to flood defense measures. Dikes, first constructed in the 17th century, have been widened and heightened (Lammersen et al., 2002). To force incision and reduce the flooding of the Upper Rhine, the river was straightened between 1817 and 1890 (Blackbourn, 2006), which shortened the river between Basel and Worms from 354 to 273 km (Silva et al., 2001).

8

Chapter 1. Introduction

Figure 1.4: Canalization of the Upper Rhine. The drastic rectification to tame the Upper Rhine was designed by an engineer named Tulla in the early 19th century. Although his plans had been ready for several decades, to his frustration they were not agreed upon by authorities, even though technology was sufficiently advanced for their implementation. His great project was realized finally with the changing institutional and political background in the riparian states of the Upper Rhine after the French Revolution (Blackbourn, 2006). From 1928 to 1977, the Upper Rhine was further canalized to aid shipping and the construction of hydro-electric power stations (Figure 1.4). As a result, floods moved faster through this section of the Rhine, and peak water levels rose higher. To counteract this undesired effect, Germany and France agreed in 1982 to implement retention measures in and along the Upper Rhine. From 1851 to 1990 dramatic changes were also made in the Lower Rhine. Dikes were built, meanders were cut-off, and the river was secured with groynes (Silva et al., 2001). Since then, flood management in the Netherlands and Germany has gone through several phases of institutional cooperation, which have been mainly triggered by extreme

1.2. The Rhine

9

events (Becker et al., 2007). The safety levels of the embankments vary between the different riparian countries of the Rhine basin. In the eastern part of the Netherlands, embankments are designed to withstand floods that occur on average once in 1250 years (now estimated at 16 000 m3 /s at Lobith (Ministry of Transport, Public Works and Water Management, 2006b); whereas in Germany, safety levels are set at 1/200 to 1/500 per year (Lammersen, 2004). Thus, floods may occur upstream in Germany, while being prevented by the Dutch dike system downstream (Gudden, 2004; Apel et al., 2006). The flood peaks in 1993 and 1995 were an important wake-up call in the Netherlands. During these flood peaks, dike stability was considered weak and reached a point where safety could no longer be guaranteerd which led to the evacuation of 250 000 people, even though the observed discharges were much lower than the design discharge for those areas (12 000 m3 /s vs. 16 000 m3 /s). Since then, dikes have been strengthened and heightened up to their design levels (Ministry of Transport, Public Works and Water Management, 1995). After the floods in 1993 and 1995, discussions on flood frequency and protection increased basin wide. A Ministers conference in February 1995, attended by representatives of all countries sharing the Rhine basin, resulted in the Action Plan on Floods, which is under the mandate of the International Commission for the Protection of the Rhine (ICPR, 1998). This plan has four goals: (1) to reduce flood risk by 10% in 2005 and by 25% in 2020, as compared with 1995; (2) to reduce extreme flood stages by 30 cm in 2005 and by 70 cm in 2020, as compared with 1995; (3) to raise awareness of flood risks; and (4) to improve flood forecasting. Measures to reduce flood stages were targeted mainly at upstream retention methods in the tributaries and the creation of extra retention volume by inundation polders along the main Rhine branch. In the Netherlands, instead of new dike reinforcements and dike raise, the government decided to create ‘Room for the River’, by widening and deepening flood planes (Ministry of Transport, Public Works and Water Management, 2006c). Figure 1.5 shows an overview of the flood reduction measures that can be implemented along the Rhine in the Netherlands. In 2005, an evaluation of the Action Plan on Floods revealed that, with current climate conditions, neither the lowered flood risk nor the reduced flood stages targets were met (ICPR, 2005a). It was also uncertain whether the original plan adequately addressed the latest climate change projections. Hooijer et al. (2004) suggested that a lower flood risk in the Rhine basin could be achieved easier through damage reduction and spatial planning than by flood defense measures. However, this assumption is not yet confirmed by research, and there is a need to test the effectiveness of additional flood management measures by considering the impact of climate change on extreme flood peaks.

10

Chapter 1. Introduction

Deepening of the forelands

Removal of obstacles

Lowering of groynes

Enlarging of summer beds

Flood channel

Strengthening of dikes

Displacement of dikes or depoldering

Figure 1.5: Measures in the Dutch Spatial Planning Key Decision ‘Room for the River’.

1.3
1.3.1

Available research and remaining challenges
Simulating the effect of climate change on extreme flood peaks

To estimate the impact of climate change on future flood-peak probabilities, researchers must understand relevant and often complex relationships, processes, and thresholds in climate and hydrological systems, while accounting for the shortcomings and uncertainties in their modeling tools. The prediction of the impacts of climate change on the probabilities of extreme flood peaks requires many steps, including downscaling the output of general circulation models, simulating runoff, and conducting an extreme value analysis. All steps are concisely explained below. GCMs and RCMs In its Special Report on Emission Scenarios (SRES), the IPCC used different assumptions for economic development to define a set of greenhouse gas (GHG) emission scenarios (IPCC, 2000). They range from a world with regional economies and local solutions for environmental sustainability (low scenario, B1), to a global market economy with high growth rates and fossil fuel intensive energy technologies (high scenario, A2). Based on these emission scenarios, General Circulation Models (GCMs) were used to model weather characteristics at a global scale up to the year 2100 (Covey et al., 2003). This resulted in different scenarios for global warming (Figure 1.6).

1.3. Available research and remaining challenges

11

Figure 1.6: IPCC SRES scenarios for global warming (IPCC, 2000). However, for subcontinental basins like the Rhine, the spatial resolution of a GCM is inadequate for forcing a hydrological model, and downscaling is required. The grid size of a GCM varies between 1 ◦ and 2 ◦ , which at the latitude of the Rhine is about 100–200 km north-south and 80–160 km east-west. Consequently, the Rhine basin is only covered by a few GCM grid cells. Downscaling can be done by perturbing the historical series with the average change in climate parameters; this is referred to as the ‘delta change’ approach (Lenderink et al., 2007a; Te Linde, 2007). Downscaling can also be done statistically based on a reference period (Jacob et al., 2008), or dynamically using regional climate models (RCMs) that are nested within a GCM (e.g. Wilby et al., 2000; Jacob, 2001; Kay et al., 2006). The latter is known as the direct forcing approach. Different climate models typically result in different climate projections and often simulate different changes in weather characteristics; this provides a first indication of the uncertainties involved (Horton et al., 2006; Menzel et al., 2006; Prudhomme, 2006; Lenderink et al., 2007b). The performance of climate models can be tested against observational data, and it is found that most RCMs underestimate the high rainfall extremes (Pitman and Perkins, 2007). A bias-correction on RCM output data using historical time series data is necessary to obtain meaningful input for hydrological models (Leander et al., 2008; Bakker and Van den Hurk, 2009). Methods for downscaling and bias-correction are continuously improving and different methods now exist to transform RCM output to hydrological input. There is not yet a consensus on which method is preferred: the robust but simple delta change approach, or the less robust (bias-corrected) direct RCM output, which provides more detailed

12

Chapter 1. Introduction

Figure 1.7: Hydrological response to precipitation (Bitesize, 2010). information on the temporal and spatial changes (Wilby et al., 2000; Lenderink et al., 2007b). Runoff simulation The meteorological data series obtained from models describe both a reference situation and future scenarios of precipitation and temperature characteristics. The next step is to simulate the hydrological response to these climate change scenarios. However, this is not as straightforward as it seems. RCMs and hydrological models overlap each other in simulating land surface processes such as evaporation and discharge generation. Hydrologists and meteorologists are investigating methods to match the observed complexity in hydrological and meteorological processes with the necessary simplification in modeling tools at different scales (e.g. Koster et al., 2000). Similar to climate models, there are many hydrological models to choose from. The hydrological response to a precipitation event can be visualized by a hydrograph (Figure 1.7). The hydrological community has devoted a great deal of attention in the last few decades to the understanding of hydrological processes and their representation in rainfall-runoff models (e.g. Refsgaard, 1996; Seibert, 1999; Perrin et al., 2000; Liu and Todini, 2002; Booij, 2003; Wagener et al., 2003; Troy et al., 2007). For an overview of benchmark papers since 1933 describing discharge generation processes, see Beven (2006b). However, in the Rhine basin, our understanding of the discharge generating processes is still inadequate, and modeling results for the current hydrological situation at basin scale are only of moderate quality (Weerts, 2003; Bogaard et al., 2005). For example, there is an ongoing debate on the utility of more complex distributed models that at-

1.3. Available research and remaining challenges

13

tempt to describe all physical processes at high resolutions, including soil-atmosphere feedback processes, as opposed to relatively simple lumped model approaches (e.g. Beven, 2001; Savenije, 2001; Bergstr¨m et al., 2002; Beven, 2006a; Sivakumar, 2008). o In theory, physically-based or land surface models that describe soil-atmosphere feedback processes are preferred in climate impact analyses (Hurkmans et al., 2008). The question remains whether in practice they are easily applied and perform better than lumped model approaches. The flood routing scheme of most hydrological models is generally insufficient for providing detailed information on flood wave propagation and peak water levels in low gradient river stretches, where floodplain inundation plays an important role (Chow, 1959). Engineers have developed hydrodynamic models that are capable of simulating the necessary detailed information on flood peak behavior in river channels much better (e.g. Delft Hydraulics, 2005). These models are used for planning of hydraulic structures, they are well calibrated using observations and scale models, and their performance is less subject to scientific debate than is the mathematical description of hydrological processes. However, these models require detailed information on hydraulic profiles and roughness parameters and involve intensive computing. Thus, hydrodynamic models are not applied within the vast majority of climate impact studies on discharge behavior, creating a gap in our understanding when planning flood management measures that aim to consider future climate change. Also, the hydrological models currently applied in climate impact studies generally disregard the consequences of upstream flooding in Germany and France. Lammersen (2004) showed that extreme flood peaks can be topped off and delayed significantly in the downstream parts of the Rhine basin due to flooding in upstream areas having lower safety levels than the Dutch dike system. However, it is sometimes argued that as a response to flooding, safety levels and dikes in upstream areas might be raised in the future (Silva, 2003). Consequently, this effect has not been further investigated and is ignored at present in flood management strategies in the Lower Rhine in Germany and the Netherlands. Extreme flood peaks To test the effectiveness of flood management measures, the understanding of peak flow generation processes is essential. The consideration of climate change requires estimates of changes in the probability and magnitude of extreme flood peaks, which is far from trivial. For very rare extreme events (i.e., a probability of 1/200 per year or less), this often requires hydrological or statistical modeling of substantially higher discharges than are observed. As a result, these models are used far outside their calibrated range, introducing considerable uncertainties (Refsgaard, 1997; Klemeˇ, s 2000a,b; Katz, 2002; Sivakumar, 2005; Garrett and M¨ller, 2008). u

14

Chapter 1. Introduction

24000

Discharge at Lobith (m3/s)

22000 20000 18000 16000 14000 12000 10000 8000 6000 4000 2000 1 2 5 20 50 100 200 1000 10000
1/1250

Return period (years)

Figure 1.8: Statistical extrapolation to obtain the design discharge at Lobith at a return period of 1250 years and the 90% confidence intervals of the extreme value distribution fit (Silva et al., 2001). Two approaches exist to estimate discharge volumes at design standards (i.e. 1/200 per year or less) that exceed historical observation records (i.e. 100 years). The most commonly used approach is to extrapolate an extreme value distribution fit of observed peak events to the desired safety standards (Shaw, 2002, Figure 1.8). This assumes stationarity of the observed data record, which implies fluctuations within an unchanging natural system as described by a static probability density function. However, in 100 years, both meteorological conditions and the river basin change, violating the principle of stationarity and increasing the uncertainty associated with the statistical extrapolation (Milly et al., 2008). In an alternative approach, a weather generator can be used to create long, resampled time series of climate change scenarios as input for the hydrological models. These long time series then contain extreme events with very low probabilities (1/200 per year or less). Although research on the performance of such an approach to estimate design discharges is available (e.g. De Wit and Buishand, 2007; Leander, 2009), the design discharge at 1/1250 per year in the Netherlands is estimated by statistical extrapolation (Figure 1.8), which results in 90% confidence intervals at 16 000 m3 /s of ± 3000 m3 /s. These confidence intervals, however, are ignored in decision making (Ministry of Transport, Public Works and Water Management, 2006b). Human intervention and the nonlinear and complex system of climate change further undermine the basic assumption of stationarity in water systems (Pielke Sr. et al., 2009). In response, Sivapalan and Samuel (2009) proposed a shift from traditional flood frequency analysis, based on stationarity, towards a new process-based approach (McMillan and Brasington, 2008) for flood estimation and the assessment of flooding probabilities. In this approach, the impact of climate change and the changes in river geometry and land use due to human influence are parameterized in models, in an

1.3. Available research and remaining challenges

15

Figure 1.9: Inundation depth map between Ludwigshafen and Worms. attempt to physically describe extreme situations and the consequences of changed conditions. Only recently have efforts been made to apply a process-based approach to climate impact studies (Raff et al., 2009).

1.3.2

Estimating potential damage and future flood risk

In flood risk management, the combined influence of socio-economic development and climate change presents a major challenge to water managers. Not only is it necessary to limit flood probabilities, but managers must also assess measures to alleviate the consequences of a flood. Examples of these measures are spatial planning (to avoid urban development in risky places), flood damage insurance, and early warning systems (Schanze et al., 2006). Spatial adaptation and flood defense measures may need to last for several decades; thus, their planning and management requires insight in likely future necessities. Also, flood risk is dynamic, due to changes in both land use and flood probability; this requires consideration of future impacts on flood risk. Worldwide, only a few studies have examined the potential future increase in disaster losses, accounting for both climate and socio-economic change (Bouwer, 2010).

16

Chapter 1. Introduction

To date, the most comprehensive overview on flood damage in the Rhine basin is the Rhine Atlas of the ICPR (Figure 1.9), which maps the overall direct damage potential (ICPR, 2001). However, it is not designed to derive information on current flood risk (probability times damage), nor future basin-wide flood risk estimations under socio-economic, land use, and climate changes. The consequences due to flooding can be extensive and varied, and may be categorized into direct and indirect damage (Merz et al., 2010). Another distinction is made between tangible and intangible loss, depending on whether or not they can be monetized. Tangible direct loss includes damage to private and public property and infrastructure. Human suffering due to loss of life or injuries is an example of an intangible direct loss. Damage due to evacuation and disruption of economic activities is referred to as an indirect loss (Jonkman, 2007). Another example of an indirect loss is the loss of value suffered by any remaining real estate due to the flooding event (Daniel et al., 2009). Direct economic damage is an important indicator for the severity of a flooding event, and is often monetized by insurance companies or others. In this thesis I focus on direct economic damage as a part of flood risk. The potential damage in flood prone areas is calculated by combining land-use and damage models, which may vary widely in scale and accuracy (e.g. Thieken et al., 2005; Klijn et al., 2007).

1.4

Objective and research questions

The central aim of this thesis is to assess the effect of climate and socio-economic change on flood risk in the Rhine basin and to improve existing simulation methods. The focus is on the estimation of the probability of flooding and the development of cross-boundary flood management measures to cope with expected changes. This objective will be achieved by addressing the following questions: 1. Does a physically-based land-surface model perform better than a conceptual rainfall-runoff model when simulating the effect of climate change on the discharge regime? 2. How can we optimally use climate change scenarios of precipitation and temperature to estimate expected changes in low-probability flood peak events? 3. What is the present basin-wide flood risk, how will it change up to 2030, and what is the relative contribution of climate and socio-economic change to this risk? 4. Which flood management measures are most effective in reducing flood stages and flood frequency on a basin-wide scale?

1.5. Thesis outline

17

Climate change scenarios of precipitation and temperature

Chapter 2

Chapter 3

Chapter 4

Socio-economic scenarios of land-use change and economic growth

Hydrological model comparison

Impact of climate change

Future ood risk

Formulation of ood management measures

Selection of hydrological model

Future peak discharges & frequency

Chapter 5

Effectiveness of ood management measures

Figure 1.10: Outline of the thesis.

1.5

Thesis outline

A schematic of the chapters of this thesis is provided in Figure 1.10. Chapter 1 provides the problem definition and available research, concisely describes the Rhine basin, and sets out the main objectives and research questions. Chapter 2 compares the performance of two hydrological models of different complexity and process description. The models are evaluated on their ability to simulate (peak) discharge on a sub-basin and basin-wide scale, using observed discharge series and different meteorological datasets. The result determines which rainfall-runoff model is best used in the remainder of the research. In Chapter 3, future changes to the discharge regime are simulated and future floodpeak probabilities are estimated, using resampled climate modeling data. The resampling was performed to obtain a long time series of 1000 (when possible 10 000) years of daily discharge, representing hydrological conditions in the reference climate and in the climate predicted for 2050. This allows us to describe a certain state of the system using a process-based approach, in which changes in rainfall-runoff processes are

18

Chapter 1. Introduction

simulated due to altered meteorological conditions from climate change. These long time series reduce statistical uncertainty when estimating low probability flood-peak events. Different methods to transform the regional climate model (RCM) output to a hydrological model input were used and compared. Chapter 4 provides estimates of future flood risk (the product of probability and potential damage) in the Rhine basin. Socio-economic scenarios, a land-use model and a damage model are used together to estimate current and future damage in flood prone areas. Only direct economic losses are considered, since most available data pertains to direct damage. Combined with earlier results on how flood peak probabilities change with climate, flood risk is assessed and spatially differentiated along the main Rhine branch. In Chapter 5, the effectiveness of flood management measures on peak discharges is evaluated assuming a climate change scenario for the year 2050. Here, we extend the process-based simulation of discharge with a hydrodynamic component. Also, the effect of upstream flooding in Germany on flood stages downstream in the Netherlands is estimated using an ensemble of synthetic flood waves and a combination of a hydrological and hydrodynamic model. Chapter 6 summarizes the results of previous chapters and addresses the four research questions. The main conclusions are presented and recommendations are made for further research.

Chapter

2

Comparing model performance of two rainfall-runoff models using different atmospheric forcing data sets

Te Linde, A.H., Aerts, J.C.J.H., Hurkmans, R.T.W. and Eberle, M., 2008. Comparing model performance of two rainfall-runoff models in the Rhine basin using different atmospheric forcing data sets. Hydrololgy and Earth System Sciences, 12: 943–957. Abstract Due to the growing wish and necessity to simulate the possible effects of climate change on the discharge regime on large rivers such as the Rhine in Europe, there is a need for well performing hydrological models that can be applied in climate change scenario studies. There exists large variety in available models and there is an ongoing debate in research on rainfall-runoff modeling on whether or not physically based distributed models better represent observed discharges than conceptual lumped model approaches do. In addition, it is argued that Land Surface Models (LSMs) carry the potential to accurately estimate hydrological partitioning, because they solve the coupled water and energy balance. In this Chapter, the hydrological models HBV and VIC were compared for the Rhine basin by testing their performance in simulating discharge. Overall, the semi-distributed conceptual HBV model performed much better than the distributed land surface model VIC (E=0.62, r2 =0.65 vs. E=0.31, r2 =0.54 at Lobith). It is argued here that even for a well-documented river basin such as the Rhine, more complex modeling does not automatically lead to better results. 19

20

Chapter 2. Comparing model performance of two rainfall-runoff models

Moreover, it is concluded that meteorological forcing data has a considerable influence on model performance, irrespectively to the type of model structure and the need for ground-based meteorological measurements is emphasized.

2.1

Introduction

It is expected that climate change will have major implications for the discharge regime of many rivers around the world (Kundzewicz et al., 2007). Changes in seasonal discharge are projected for river basins in mid-latitude regions, such as the Rhine basin in Europe. Seasonal discharge will most likely shift to more discharge in winter and less discharge in summer, and the frequencies of floods and droughts are expected to increase (Kwadijk, 1993; Middelkoop et al., 2001; Buishand and Lenderink, 2004). Recent climate change research focuses on simulating changes in the magnitude and frequencies of flood events using different models that are either developed for scenario studies, real time flood forecasting, or both (Van Deursen, 2006; Te Linde, 2007). Our understanding of the discharge generating processes in the Rhine basin, though, is still deficient and modeling results for describing the current hydrological situation at basin scale are of moderate quality. For instance, extreme events inside the calibrated range are both over and underestimated and it is difficult to separate the effects of errors in input data and model structure (Weerts, 2003). This increases the inherent uncertainty when using models outside their calibrated range, as is common practice in climate scenarios studies. Thus there is a need for a well performing hydrological model on extreme events that can be applied in various climate scenario studies, but there exists large variety in available models. Since these issues are common in applications of hydrological modeling in other regions as well, we chose to compare two rainfall-runoff models for the Rhine basin with divergent model structures. The semi-distributed conceptual model HBV (Hydrologiska Byr˚ Vattenbalansavdelans ning) (Bergstr¨m, 1976; Lindstr¨m et al., 1997) has been applied in multiple studies o o for the Rhine basin since 1999 by both the German Federal Institute of Hydrology and the Dutch Ministry of Transport, Public Works and Water Management (M¨lders u et al., 1999; Weerts and Van der Klis, 2004; Eberle et al., 2005). However, the HBV model does not exactly describe all the physical processes that are believed to be of major importance for the simulation of timing and magnitude of extreme flood and drought events (Sch¨r et al., 1998; Ward and Robinson, 1990). Potential evaporation, a for example, is calculated using the Penman-Wendling approach based on temperature and sunshine duration (Eberle et al., 2005) while more innovative methods are available using coupled water and energy balance simulations. Recently the state of the art distributed land surface model (LSM) VIC (Variable Infiltration Capacity) (Liang et al., 1994) has been applied on the Rhine basin (Hurkmans et al., 2008), which does describe all relevant land surface processes, including the energy balance,

2.2. Divergent concepts in rainfall-runoff modeling

21

and therefore carries the potential to estimate hydrological partitioning more accurately than the HBV model does. Because of a realistic representation of evaporation processes in land surface models such as done within VIC, Troy et al. (2007) argue that these types of models are inevitable when performing climate and land use change scenario studies. However, the application of a distributed land surface model such as VIC at a macroscale river basin, such as the Rhine basin, is still a highly simplified representation because of its spatial resolution. Even when using a very fine grid, in the order of tens or hundreds of meters and by that sabotaging calculation time, it will never represent actual processes that vary at a scale of trees and ditches (Uhlenbrook, 2003) and the actual heterogeneity of hydrological processes. Considering the required input data and computer capacity, the question remains whether more complex and demanding models such as VIC can be preferred over simpler, conceptual water balance models such as HBV. A better understanding of the use and capacity of different hydrological models would enhance the confidence in future climate scenario studies using these hydrological models. An uncertainty analysis of all processing steps from climate scenarios via downscaling methods to hydrological modeling is required. Estimating uncertainty of model simulations starts with analyzing model performance using historical data. In this view, the goal of this Chapter is to compare the hydrological models HBV and VIC by testing their performance for simulating historical discharge. Based on the performance of both models, a recommendation can be made for the type of hydrological model to be preferred for climate change scenario studies. Since both models have a different physical structure resulting from a different theoretical background, the divergent concepts in rainfall-runoff modeling are first addressed in Section 2.2. In Section 2.3, the models and study area are described. In Section 2.4, the methods that are used for comparing atmospheric forcing data and model performance are explained, whereupon the results are presented in Section 2.5. Finally, the results are discussed and several conclusions are drawn in Section 2.6.

2.2

Divergent concepts in rainfall-runoff modeling

There is an ongoing debate in research on rainfall-runoff modeling on the utility of more complex distributed models that aim to describe all physical processes, including soil-atmosphere feedback processes. In the last decades the hydrologic community has devoted a great deal of attention to the understanding of hydrological processes and their representation by means of physically based, distributed models. The general idea of physically based, distributed modeling is that it represents reality better than lumped model approaches, as it takes into account spatial information and even more important, it uses physical law (mass balance and energy equations) to describe the

22

Chapter 2. Comparing model performance of two rainfall-runoff models

hydrological processes (Refsgaard, 1996; Reggiani and Schellekens, 2003). However, it is well recognized that the available approaches are often still far from providing a satisfactory representation of rainfall-runoff transformation (Bergstr¨m et al., 2002). o A lot of work remains in identifying different runoff response mechanisms and to characterize the key state variables during calibration (Perrin et al., 2000; Uhlenbrook et al., 1999; Wagener et al., 2003). This should be done by extensive and long duration field observations, using the growing availability of radar and space highresolution datasets, improving physical descriptions and refining grid size. Examples of physically based, distributed models are SHE (Abbott et al., 1986), FLOWSIM (Rientjes and Zaadnoordijk, 2000), WASIM-ETH (Schulla and Kaspar, 2006), LARSIM (Ludwig and Bremicker, 2006), REW (Reggiani et al., 1998, 1999), LISFLOOD (De Roo et al., 1998, 2000), TOPKAPI (Liu and Todini, 2002) and tRIBS (Vivoni, 2003). Related to these models are land surface model (LSMs). The original purpose of LSMs was to represent the land surface in (regional) climate simulations used for climate models and numerical weather prediction (Liang et al., 1994). Recently, LSMs have been used for streamflow forecasting as well. By solving both the water and energy balance, LSMs are able to exploit a larger part of the information provided by regional climate model output (Hurkmans et al., 2008), which is an advantage in climate change scenario studies where regional climate model (RCM) output is used. But because of the complex model structure and the large number of parameters in LSMs, they are generally difficult to parameterize. Furthermore, in most of these distributed modeling approaches, it remains difficult to represent processes occurring at scales smaller than the grid or element scale. The VIC model therefore offers sub-grid scale variation in vegetation and soil characteristics (Liang et al., 1994). On the other hand, some researchers advocate a more straightforward hydrologic approach claiming that more complex modeling does not always lead to better results. Depending on dominant processes, data availability, scale and application of the model, one should select the appropriate modeling approach which can result in using a very simple model (Booij, 2003; Seibert, 1999). When formulating their famous and widely used performance criterion, Nash and Sutcliffe (1970) already warned for the risk of over-parameterized models. In recent years, the debate on model complexity versus model performance has intensified again and Beven (2001, 2002a,b) goes a step further and critically analyzed the constraints of distributed modeling. The perfect hydrological model that represents reality accurately will never exist, as there will always remain necessary approximations of processes and parameters at the model element scale. Beven (2001) claims that the ongoing pursue to a realistic representation has led to unjustified determinism in many distributed modeling applications and a lack of recognition of the problems of distributed modeling such as nonlinearity, scale and equifinality (which arises when many different parameter sets give equally good results). Furthermore, Savenije (2001) states that the large number of parameters in distributed models make it possible to represent hydrological behav-

2.3. Model description and study area

23

ior well for the current situation, but due to over-parameterization these models are not the right tools to describe what will happen if certain characteristics of the basin change, such as land use or soil characteristics. Savenije (2001) suggests to further develop a new data-based top down approach (Jothityangkoon et al., 2001) in which relatively simple basin response functions describe complex hydrological processes at scales with sufficient level of aggregation. It consists of beginning with a large time step and gradually introducing the complexity required to meet the needs of shorter time steps. This resembles the conceptual approach of already long-existing water balance models like Sacramento, HBV and RhineFlow (Van Deursen and Kwadijk, 1993). Bogaard et al. (2005) argue that the main challenges in understanding discharge generating processes appear to be related to the scale of the processes. Micro scale hydrological processes are highly heterogeneous, non-linear and interconnected, with the consequence that upscaling from micro- to basin scale and subsequent parameterization is practically impossible. In conclusion, hydrologists are looking for answers to match the observed complexity at the plot-scale, with the apparent simplicity that arises at the basin scale. Comparing the HBV and VIC models, having opposed model structures, for their performance in a well-documented river basin like the Rhine basin, will add to the debate on divergent concepts in hydrological modeling.

2.3
2.3.1

Model description and study area
VIC

The Variable Infiltration Capacity (VIC) model (Liang and Zhenghui, 2001; Liang et al., 1994) is a distributed, macro-scale land surface model with a physically based soil-vegetation-atmosphere transfer scheme (SVATS), which solves both the water and energy balance. It is distinguished from other SVATS by its focus on runoff processes. These are represented through the variable infiltration curve, a parameterization of the effects of sub-grid variability in soil moisture holding capacity, from which the model takes its name, and a representation of non-linear baseflow. Routing of surface runoff and baseflow is done by the algorithm developed by Lohmann et al. (1996). A more extensive description of the modeling scheme is available in Hurkmans et al. (2008), who recently developed the VIC model for the Rhine basin at a spatial resolution of 0.05×0.05 degree. The seven required atmospheric input time series are derived from a re-analysis dataset and are described in Section 2.4.1.

24

Chapter 2. Comparing model performance of two rainfall-runoff models

a)

b)

c)

Lobith

Figure 2.1: Map of the Rhine basin showing (a) 134 HBV sub-catchments; (b) the calculation grid used in VIC (0.05×0.05 degree); and (c) discharge measurement locations and sub-basins used in the analysis.

2.3. Model description and study area

25

2.3.2

HBV

The HBV-96 model (Hydrologiska Byr˚ Vattenbalansavdelning) (Bergstr¨m, 1976; ans o Lindstr¨m et al., 1997) model is a semi-distributed conceptual model. The model that o is used in this study simulates discharge on a daily basis for 134 sub-basins of the Rhine. The model simulates snow accumulation, snowmelt, actual evapotranspiration, soil moisture storage, groundwater depth and runoff. The required forcing data are precipitation, temperature, and potential evaporation. The model consists of different routines in which snowmelt is computed by a day-degree relation, and groundwater recharge and actual evaporation are functions of actual water storage in a soil box. Discharge formation is represented by a linear reservoir for base flow and a non-linear approach for fast runoff components. Appendix A provides more information and an illustration of discharge formation in the HBV model. The sub-basins are linked together with a simplified Muskingum approach (Shaw, 2002) to simulate routing processes. The HBV model was developed for the Rhine in several steps since 1997 by the Dutch Institute for Inland Water Management and Waste Water Treatment (RIZA) and the German Federal Institute of Hydrology (BfG). A complete description of the HBV calculation scheme and the model structure for the Rhine basin is found in Eberle et al. (2005).

2.3.3

Rhine basin

The study area includes the Rhine basin (Figure 2.1) upstream of the Dutch-German border and covers an area of 160 800 km2 . The Rhine originates in the Alpine mountains that comprise almost 36 000 km2 upstream of Basel, with maximum elevations of more than 4000 m a.s.l. Air temperatures are below zero during the winter season due to this height, and a substantial part of the precipitation is stored as snow. Land cover in the Alps is characterized by agricultural land in the lower regions and by forest, shrubs, meadows, unvegetated areas and glaciers on the higher slopes. The area of the Upper Rhine between Basel and Bingen is hilly, with elevations reaching over 1000 m a.s.l., but with flood plains along the main rivers. In the flood plains there is urban development, while the hills are mainly forested. The main tributaries Neckar, Main, Moselle, Lahn and Sieg have a mixed land use pattern, with agriculture and vineyards on the valley slopes, and forest on the hillslopes and mountains. The Middle Rhine has incised in higher grounds, which resulted in a deep narrow valley without floodplains. The relatively flat and low-lying Lower Rhine area downstream of Cologne until the Dutch-German border is an urbanized area with a mixture of agriculture, meadows and some forest. Overall, the Rhine basin is densely settled, with an average population density of 270 persons per km2 (Early, 2001). About 50% of the basin is used for agriculture, 15–20% is urban or suburban land, and the remainder is forest and otherwise natural lands (Wessel, 1995).

26

Chapter 2. Comparing model performance of two rainfall-runoff models

Table 2.1: Basin and sub-basin characteristics. Surface area (km2 ) is defined by the basin area upstream of the gauging station.
Basin Gauge Surface area (km2 ) 160 800 139 549 27 088 5304 27 142 12 710 50 624 Mean Q (m3 /s) 2 206 2116 334 48 176 141 1297 Min. Q (m3 /s) 788 618 10 0 44 3 379 Max. Q (m3 /s) 11 885 10 406 4020 730 1991 2105 4430 Mean ann. max. Q (m3 /s) 7473 6494 2190 364 1043 1133 3191 Data period

Rhine Rhine Mosel Lahn Main Neckar Rhine

Lobith Andernach Cochem Kalkofen Raunheim Rockenau Maxau

1989–2005 1961–2004 1961–2004 1961–2004 1989–2005 1971–1990 1961–2004

The average discharge of the Rhine at Lobith is 2206 m3 /s (1989–1995). The mean annual maximum discharge is 7473 m3 /s, while the maximum discharge in the period since 1961 is 11 885 m3 /s, which occurred in January 1995 and caused floods in Germany. Parts of the Netherlands where evacuated, but against expectations the dikes held. Earlier considerable and some catastrophic floods in history are 1421, 1845, 1882 and 1926 (Disse and Engel, 2001). The surface area of the sub-basins under consideration in the present study vary from 5304 km2 to 27 142 km2 , as can be seen from Table 2.1 among other basin characteristics.

2.4
2.4.1

Methods
Data

Both the HBV and VIC models were forced using downscaled ECMWF ERA15 atmospheric re-analysis data, which is provided by the Max Planck Institute for Meteorology (MPI), Hamburg, Germany. The regional climate model REMO (Jacob, 2001) was used for downscaling and this dataset will be further referred to as ERA15. The ERA15 data set comprises the years 1993 through 2003, at a 3-hourly time step, with a grid resolution of 0.088 degrees and provides the following forcing data: precipitation, temperature, specific humidity, air pressure, downward radiation (shortwave and longwave) and wind speed. These input data are all required to run the VIC model. To compare this data to observations, two additional meteorological datasets are available. First, a historical data set is available from the International Commission for the Hydrology of the Rhine basin (CHR). This data set is referred to as CHR and contains daily values of precipitation and temperature for the years 1961 through 1995, which are based on 36 measurement stations throughout the basin (Sprokkereef,

2.4. Methods

27

2001). Second, a historical dataset using interpolated measured data is available from the Climate Research Unit (CRU) where they develop a number of global datasets widely used in climatic research. This data set is referred to as CRU (Mitchell and Jones, 2005) and contains precipitation and temperature values at a monthly time step and comprises the years 1900 through 1998, with a grid resolution of 0.5 degrees. HBV was also forced by CHR precipitation and temperature data and VIC only by CHR precipitation. VIC could not be forced by CHR temperature, because the models needs daily variation of temperature and therefore requires 3 or 6 hourly values of minimum and maximum temperature data. HBV only needs daily values of these forcing parameters, and at least monthly mean values of potential evaporation as input data. As a consequence of the detailed data input requirements of the VIC model, the ERA15 data still provided the remaining forcing parameters in combination with the CHR precipitation values. Combining measured values with RCM output data disturbs the water balance of the RCM output. It creates a figurative, but false forcing data set. Precipitation values of the CRU data set were only used for comparison of forcing data. Additional spatial information on altitude, soil types and land cover is derived from a GIS database available at Federal Institute of Hydrology in Germany (Eberle et al., 2005). Historical discharge data was provided by the Dutch governmental Institute for Inland Water Management and Waste Water Treatment (RIZA).

2.4.2

Forcing data comparison

Rainfall amounts of the three forcing datasets were compared for the period of 1993– 1995; the only three years the three datasets all overlap. A first comparison was made for basin wide mean values at a daily basis between the ERA15 and CHR values. For the second comparison, the ERA15 and CHR data sets were aggregated to weekly and monthly values and then compared to the CRU data.

2.4.3

Model performance

Calibration at Lobith As is explained in Section 2.4.1, HBV and VIC were forced both by ERA15 and CHR precipitation values for comparison reasons. For VIC this results in a forcing data set containing a combination of measured data for precipitation and RCM output for the remaining seven input parameters. This forcing dataset is considered incorrect and therefore both models were calibrated using only ERA15 output.

28

Chapter 2. Comparing model performance of two rainfall-runoff models

Table 2.2: Performance criteria daily and monthly discharge values at Lobith for the calibration period (March 1993–December 1993) and the validation period (1994– 2003).
Calibration period daily monthly 0.47 0.26 0.49 −0.08 0.44 0.35 0.85 0.73 0.64 0.58 0.75 0.54 0.81 0.88 0.97 0.96 0.23 0.23 0.32 0.32 −0.55 −0.55 0.19 0.19 Validation period daily monthly 0.31 0.40 0.62 0.60 – – – – 0.54 0.67 0.65 0.64 – – – – 0.08 0.08 −0.04 −0.04 – – – –

E

r2

VE

VIC ERA15 HBV ERA15 VIC CHR HBV CHR VIC ERA15 HBV ERA15 VIC CHR HBV CHR VIC ERA15 HBV ERA15 VIC CHR HBV CHR

We forced both models with ERA15 data and calibrated for the discharge gauge at the Dutch-German border at Lobith (see Figure 2.1c) using observed discharge at Lobith for the year 1993. Only one year was used in order to limit the amount of calibration time for the VIC model. Because 1993 contains a relatively dry summer, as well as an extreme peak in winter, it was considered representative of the extremes for the total period. The model simulations were initialized using model states of October 1993 and also the first two months of 1993 are considered as a ‘warm-up’ period, hence model results for this period were not used in the calibration process. To calibrate VIC, former applications of VIC (Liang et al., 1994) were followed in that seven parameters were selected for calibration using an automated approach. These seven parameters describe the layer depths, relations between soil moisture content and baseflow and the infiltration capacity. For a complete description, see Hurkmans et al. (2008). The original calibration process for the HBV model of the Rhine basin is described by Eberle et al. (2005). HBV was recalibrated for the year 1993 in a stepwise approach using the ERA15 dataset. Based on results of a parameter sensitivity analysis by Passchier and Stone (2003), for HBV, only the parameters fc (field capacity that represents the total water storage capacity of the soil) and khq (describing the quick runoff function) were adjusted for recalibration.

2.4. Methods

29

Sub-basin scale validation performance The calibrated models are validated using the remaining period of the ERA15 data set, the years 1994 through 2003. There is a large number of efficiency criteria to choose from for model validation, such as those presented by Krause et al. (2005) and each criterion may place different emphasis on different types of simulated and observed behaviors. The objective performance criteria used in the current study to compare the integral time series for the locations, are the coefficient of efficiency (E) (Nash and Sutcliffe, 1970), the coefficient of determination (r2 ) and the volume error (V E). For a description of these performance criteria, see Appendix D. Model performance differs with the scale on which it is applied. In the present study we are interested in discharges at Lobith (the outlet of the basin), discharges upstream in the main Rhine channel and model performance at the sub-basin scale. The discharge gauges that were used in the analysis are Lobith, Andernach and Maxau along the Rhine branch, and tributary gauging stations at Cochem (Moselle), Kalkofen (Lahn), Raunheim (Main) and Rockenau (Neckar). These locations are shown in Figure 2.1 and characteristics of the sub-basins upstream of those gauges are presented in Table 2.1. Peak flows and low flows Periods with extreme discharges are often of most interest both in impact studies and real time flow predictions. A good representation by the model of the absolute amount, the timing and duration of the peak and low flows is very relevant. Subsequently, just for the gauge at Lobith, we selected five peak flow and five low flow periods, and chose additional performance indicators that relate to magnitude and timing of peak flows, together with minimum values and duration of low flows. These indicators are observed maximum discharge (max. Qobs ), relative difference between observed and simulated maximum discharge (dmax. Qsim ), difference in peak timing (dT ), observed minimum discharge (min. Qobs ), relative difference between observed and simulated minimum discharge (dmin. Qsim ) and duration of the low flow period under a threshold of 1300 m3 /s (DUT). A discharge of 1300 m3 /s at Lobith is a critical value in summer periods; lower discharges affect shipping industry, agricultural supply, electricity production and drinking water supplies.

30

Chapter 2. Comparing model performance of two rainfall-runoff models

100 ERA15 (mm/week) ERA15 (mm/day) 30 20 10 r =0.64 0 0 250 ERA15 (mm/month) 200 150 100 50 0 0 250 CRU (mm/month) 200 150 100 50 0 0 100 200 CHR (mm/month) r2=0.84 100 200 CHR (mm/month) r =0.74
2 2

80 60 40 20 0 0 250 r =0.61 50 CHR (mm/week) 100
2

10 20 30 CHR (mm/day) ERA15 (mm/month)

200 150 100 50 0 0 100 200 CRU (mm/month) r =0.65
2

Figure 2.2: ERA15 versus CHR versus CRU precipitation. The period 1993–1995 was used for the comparison.

2.5. Results

31

2.5
2.5.1

Results
Forcing data comparison

The difference between measured precipitation data (CHR and CRU) and reanalysis data (ERA15) provides an indication for the error or bias in the reanalysis data set. The assumption here is that measured data better represents actual values than reanalysis data and to test this assumption both measured datasets are compared with ERA15 data. Figure 2.2 illustrates the correlation between the precipitation data at different time steps. Daily values of ERA15 and CHR correlate poorly (r2 =0.41) while the correlation coefficient increases with increasing time step length. The precipitation values of the ERA15 data do not show a constant bias that can be corrected. The correlation between monthly values of ERA15 and CHR is reasonably well (r2 =0.74) and slightly higher than between ERA15 and CRU (r2 =0.65). The correlation between CHR and CRU, however, has an r2 value of 0.84, which indicates that these two databases are most alike and that ERA15 probably has a larger error than the measured data.

2.5.2

Model performance

Calibration and validation period at Lobith Daily values of all performance criteria for Lobith are displayed in Table 2.2, where a distinction is made between the calibration and the validation period. The additional six locations will be discussed below. At Lobith after calibration, the results of the HBV model forced with ERA15 show a moderate performance (E=0.49, r2 =0.75), whereas the VIC model fits less well (E=0.47, r2 =0.64). This is mainly caused by an overestimation of the volume, by 23% (VIC) and 32% (HBV), respectively. VIC forced by CHR shows an increased correlation (r2 =0.81) when compared to its performance when forced by ERA15, but a decrease on the other performance criteria (E=0.44, V E=−0.55). However, the HBV model forced with CHR fits well when compared to observed discharges (E=0.85, r2 =0.97). Figure 2.4 depicts the results of the period 1993–2003 at Lobith, respectively for the VIC and the HBV models both forced with ERA15 data. The HBV model shows a better fit of the simulated discharge to the observed discharge than VIC, which is confirmed by the efficiency coefficients as shown in Table 2.2. The coefficient of efficiency (E) of HBV is 0.62, where VIC shows 0.31 and coefficient of determination (r2 ) of HBV is 0.65, where VIC displays 0.54. The volume error of both models is low (−4% by HBV and 8% by VIC). A visual analysis of the hydrographs at multiple peak flow events, shows that both models simulate the recession curve well. Errors arise

32

Chapter 2. Comparing model performance of two rainfall-runoff models

300 P ERA15 P CHR P (mm) 200

100

0 12000 10000 Q (m3/s) 8000 6000 4000 2000 0 12000 10000 Q (m3/s) 8000 6000 4000 2000 0 Mar Apr May Jun Jul 1993 Aug Sep Oct Nov Dec VIC CHR HBV CHR Observed VIC ERA15 HBV ERA15 Observed

Figure 2.3: Monthly precipitation values for the Rhine basin according to different datasets (top) and daily discharge values at Lobith (bottom); model simulation results for the calibration period compared to the observed discharge in the period March– December 1993. at most extreme peak events (see section below) on peak flows and low flows) where the quick flow component either is too small or too large. VIC tends to overestimate more peaks than HBV does and shows a delayed peak at many peak events. Medium flows are mostly well represented by HBV, whereas VIC substantial over estimates discharges, sometimes for a period of several months. Low flow periods are simulated well for a length of time up to 2 or 3 months, and when drought periods are more lengthy, both models tend to underestimate baseflow. The changeable reaction of both models to different meteorological conditions suggests that the storage capacity in the upper layers is very irregular, resulting in variable estimates of direct runoff. Also, the depletion factor controlling drainage from the lower layers seems to be too large during lengthy drought events. A further explanation for these moderately successful results might be that at a short time step like a daily basis, errors in timing of simulated high and low flows have a considerable negative influence on the performance indicators. Nonetheless, when

2.5. Results

33

5000 dQ (m/s) 0 −5000 12000 10000 8000 Q (m/s) 6000 4000 2000 0 5000 0 −5000 12000 10000 8000 Q (m3/s) 6000 4000 2000 0 Observed Simulated HBV Observed Simulated VIC

500

1000

1500

2000

2500

3000

3500

4000

dQ (m3/s)

500

1000

1500

2000 T (days)

2500

3000

3500

4000

Figure 2.4: Daily simulation results of the HBV model (a) and the VIC model (b) compared to the observed river discharge for the period 1993–2003 (4017 days).

34

Chapter 2. Comparing model performance of two rainfall-runoff models

Table 2.3: Observed and simulated mean, minimum and maximum discharge (in m3 /s), their standard deviation (SD) and skewness for the period March 1993 through December 2003.
Basin Rhine Gauge Lobith Observed VIC HBV Observed VIC HBV Observed VIC HBV Observed VIC HBV Observed VIC HBV Observed VIC HBV Observed VIC HBV Mean Q (m3 /s) 2387 2811 2339 2197 2474 2054 355 325 263 48 42 33 183 234 180 150 216 144 1322 1631 1335 Min Q (m3 /s) 788 773 746 630 734 593 31 49 21 0 6 1 51 39 44 27 34 19 400 374 407 Max Q (m3 /s) 11 885 11 394 11 228 10 500 10 487 11 092 4020 2463 3644 598 350 506 1,991 1885 1946 2140 2490 2291 4330 5222 5137 SD (m3 /s) 1300 1468 1244 1182 1258 1104 416 282 274 61 42 40 197 227 189 142 201 167 530 739 629 Skewness (–) 2.29 1.45 1.99 2.29 1.46 2.08 3.20 2.63 4.37 3.86 2.44 4.21 3.65 2.48 3.89 5.26 3.17 4.93 1.38 0.98 1.18

Rhine

Andernach

Mosel

Cochem

Lahn

Kalkofen

Main

Raunheim

Neckar

Rockenau

Rhine

Maxau

monthly values of simulated discharge are evaluated they display similar or slightly worse results, as can be seen from Table 2.2; VIC and HBV forced with ERA15 perform moderately and HBV forced with CHR fits well, which is about equal to the HBV simulations at a daily basis. The difference in coefficient of efficiency (E) between daily and monthly values of the HBV model forced with ERA15 in the calibration period stands out though, a moderate 0.49 for daily values and a dramatic −0.08 for monthly values. Instead of the expected damping effect on performance, aggregating to a bigger time step indeed causes the observed and modeled peak value of several days in December 1993 (shown in Figure 2.3) to damp, but does not effect the more or less consistent over estimation during the months May until July. Since the coefficient of efficiency (E) is sensitive to peak values, in this case the absolute observed and modeled discharge values are damped, but the relative error by time step increases which causes the coefficient to drop. These results indicate that forcing data largely influence the performance values of both models. A closer examination of the precipitation values in both forcing data sets during the calibration period is depicted in Figure 2.3, together with observed

2.5. Results

35

Lobith 12000 10000 8000 6000 4000 2000 0 0 4000 8000 12000 Q (m3/s) HBV Q (m3/s) VIC 12000 10000 8000 6000 4000 2000 0 0

Lobith

4000

8000

12000

Q (m3/s) Observed

Q (m3/s) Observed Raunheim 2000 Q (m3/s) HBV Q (m3/s) VIC 1500 1000 500 0 0 4000
3

Andernach 12000 10000 8000 6000 4000 2000 0 0 4000
3

Andernach 12000 10000 Q (m3/s) HBV 8000 6000 4000 2000

Raunheim 2000 1500 1000 500 0

Q (m3/s) VIC

8000

12000

0

8000

12000

0

500

1000

1500

2000

0

500

1000

1500

2000

Q (m /s) Observed Cochem 4000 Q (m3/s) VIC 3000 2000 1000 0 Q (m3/s) VIC 4000 3000 2000 1000 0 1000 2000 3000 4000 Q (m3/s) Observed 0

Q (m /s) Observed Cochem

Q (m3/s) Observed

Q (m3/s) Observed

Rockenau 2500 Q (m3/s) HBV 2000 Q (m3/s) VIC 1500 1000 500 0 0 500 1000 1500 2000 2500 Q (m3/s) Observed 2500 2000 1500 1000 500 0 0

Rockenau

0

1000 2000 3000 4000 Q (m3/s) Observed

500 1000 1500 2000 2500 Q (m3/s) Observed

Kalkofen 600 Q (m3/s) HBV Q (m3/s) VIC 600

Kalkofen 6000 5000 400 4000 3000 2000 1000

Maxau 6000 5000 Q (m3/s) HBV 4000 3000 2000 1000 0 2000 4000 6000 0 0

Maxau

400

200

200

0

0 0 200 400 600 Q (m3/s) Observed

0

200

400

600

Q (m3/s) VIC

0

Q (m3/s) Observed

2000

4000

6000

Q (m3/s) Observed

Q (m3/s) Observed

Figure 2.5: Scatter plots of observed and simulated discharge Q (m3 /s) at a daily basis. The results for VIC are displayed on the left side and for HBV on the right side. and simulated discharge values. The figure shows that during the months May, June and July, both HBV and VIC forced with ERA 15 consistently overestimate discharge by 25–100%, whereas HBV forced with CHR also overestimates discharge, but to a lesser degree. VIC forced by CHR shows an even better fit for these months. This can be explained by the equally consistent higher ERA15 precipitation values when compared to the CHR data. In August, ERA15 again displays higher values than the CHR data, while observed and simulated discharge agree quite well. This lack of reaction in modeled discharge in August can be explained by higher evaporation values during summer than spring, which neutralize the precipitation surplus, next to the fact that absolute precipitation values are lower in summer than in springtime.

36

Chapter 2. Comparing model performance of two rainfall-runoff models

1 0.5 E (−) E (−) 600 kmr 500 400 300 0 −0.5 −1 −1.5 900 VIC ERA15 HBV ERA15 VIC CHR HBV CHR 800 700

1 0.5 0 −0.5 −1 900

800

700

600 kmr

500

400

300

1 0.8 r (−) r2 (−) 800 700 600 kmr 500 400 300 0.6 0.4 0.2 0 900

1 0.8 0.6 0.4 0.2 0 900

2

800

700

600 kmr

500

400

300

0.7 0.5 0.3 VE (−) 0.1 −0.1 −0.3 −0.5 −0.7 900 800 700 600 kmr 500 400 300 VE (−)

0.7 0.5 0.3 0.1 −0.1 −0.3 −0.5 −0.7 900 800 700 600 kmr 500 400 300

Figure 2.6: Performance criteria daily discharge values. kmr represents the length of the Rhine from the Bodensee. The calibration period is displayed at the left side and the validation period at the right side. Sub-basin scale performance Several statistical parameters for the complete simulation period are presented in Table 2.3. The mean and minimum simulated discharges agree reasonable well for the HBV model, whereas VIC overestimates those values, except for the gauges at Cochem and Kalkofen. The maximum discharges, though, are underestimated for most locations, except for the most upstream gauges Rockenau and Maxau. The values for the standard deviation (SD) based on daily values are high for both simulated and observed values. This can be explained by the skewed distribution of the discharge values. Based on this information it can be concluded that the probability density function of the observed values at Lobith is best represented by the simulated discharges by HBV. For the remaining six gauges upstream of Lobith, scatter plots of the daily observed and simulated discharges are displayed (Figure 2.5) for the validation period. The

2.5. Results

37

Table 2.4: Performance criteria daily discharge values. kmr represents the length of the Rhine from the Bodensee.
Calibration period kmr E VIC ERA15 HBV ERA15 VIC CHR HBV CHR r2 VIC ERA15 HBV ERA15 VIC CHR HBV CHR V E VIC ERA15 HBV ERA15 VIC CHR HBV CHR Validation period kmr E r2 VE VIC ERA15 HBV ERA15 VIC ERA15 HBV ERA15 VIC ERA15 HBV ERA15 857 Lobith 0.47 0.49 0.44 0.85 0.64 0.75 0.81 0.97 0.23 0.32 −0.55 0.19 857 Lobith 0.31 0.62 0.54 0.65 0.08 −0.04 613 Andernach 0.52 0.59 0.68 0.91 0.61 0.75 0.81 0.97 0.14 0.23 −0.07 0.10 613 Andernach 0.30 0.55 0.48 0.60 0.02 −0.09 592 Cochem 0.45 0.81 0.63 0.92 0.57 0.82 0.69 0.95 −0.31 0.17 −0.14 0.13 592 Cochem 0.38 0.48 0.43 0.56 −0.25 −0.29 586 Kalkofen 0.24 0.30 0.50 0.86 0.50 0.37 0.67 0.86 −0.49 −0.31 −0.52 −0.08 586 Kalkofen 0.32 0.44 0.43 0.51 −0.36 −0.31 497 Raunheim 0.64 0.60 0.46 0.89 0.76 0.83 0.78 0.96 0.43 0.62 −0.34 0.21 497 Raunheim 0.05 0.07 0.27 0.23 0.02 −0.05 428 Rockenau −0.16 0.31 0.14 0.77 0.40 0.48 0.52 0.88 0.50 0.28 0.06 0.12 428 Rockenau −0.46 0.17 0.32 0.45 0.21 −0.07 363 Maxau −1.20 −0.40 0.64 0.78 0.47 0.54 0.71 0.95 0.27 0.22 −0.41 0.09 363 Maxau −0.62 0.28 0.39 0.49 0.18 −0.01

accessory r2 values are presented in Table 2.4. Table 2.4 and 2.5 show the results of all performance criteria for daily and monthly values respectively, for all locations. Above the location name, the kmr number is displayed. This number represents the length of the Rhine from the Bodensee in Switzerland and Germany. For example, the gauging station at Lobith is located 857 km downstream form the Bodensee. In the current study, the gauges that are not located exactly along the Rhine, but along tributaries draining the sub-basins, have kmr numbers that represent locations where the side rivers enter the Rhine. The kmr number is used to illustrate all performance criteria as presented in Table 2.4 and 2.5, in a graphical way in Figure 2.6 and Figure 2.7. In Figure 2.7, the volume error is not displayed, since the volume error does not change when the time step is adjusted (see Table 2.4 and 2.5). The Nash & Sutcliffe efficiency coefficient (E) decreases in the upstream direction, sometimes even below zero at Rockenau and Maxau for VIC results. An efficiency lower than zero indicates that the mean value of the observed time series would have been a better predictor than the model. From all graphs on the left side representing the calibration period, it is obvious that HBV forced with CHR performs considerably better than VIC forced with CHR and than both models forced with ERA15. Nonetheless, VIC forced with CHR performs better than VIC forced with ERA15, especially in upstream direction. Moreover, for the ERA15 forcing, HBV performs

38

Chapter 2. Comparing model performance of two rainfall-runoff models

1 0.5 E (−) E (−) 600 kmr 500 400 300 0 −0.5 −1 −1.5 900 VIC ERA15 HBV ERA15 VIC CHR HBV CHR 800 700

1 0.5 0 −0.5 −1 −1.5 900

800

700

600 kmr

500

400

300

1 0.8 r (−) r2 (−) 800 700 600 kmr 500 400 300 0.6 0.4 0.2 0 900

1 0.8 0.6 0.4 0.2 0 900

2

800

700

600 kmr

500

400

300

Figure 2.7: Performance criteria monthly discharge values. kmr represents the length of the Rhine from the Bodensee. The calibration period is displayed at the left side and the validation period at the right side. marginally better than VIC at a daily basis, whereas VIC performs marginally better than HBV at a monthly basis. When studying the validation period on the right side, however, HBV performs substantially better than VIC, which indicates that HBV is more robust in its performance. Peak flows and low flows Table 2.6 shows the five highest daily discharges and the five lowest monthly discharges, as observed and simulated at Lobith. Observed volumes over threshold and maximum peak discharges reveal that both models overestimate and underestimate the same peaks. Furthermore, it shows that VIC tends to delay flood peaks, for some peaks even up to 6 days, while HBV simulates the timing of the peaks very well. Two factors in VIC explain this delaying of peak flows: first, the routing algorithm that is used in VIC might delay arrival at Lobith slightly compared to the internal routing algorithm in HBV. This was also noted in Hurkmans et al. (2008) where runoff from another conceptual water balance model (STREAM) was routed with different algorithms. Second, the degree to which peaks are delayed also depends on calibration parameters, particularly depths of the upper layers and the infiltration capacity factor (see for details on VIC calibration Hurkmans et al. (2008)). When the resulting infiltration capacity is higher, there is less direct runoff and, in case of near-saturation, excess water is, with a small delay, transported as baseflow. For the peak of 1993, which is included in the calibration period, the simulated timing by VIC was rather

2.6. Discussion and conclusions

39

Table 2.5: Performance criteria monthly discharge values. kmr Represents the length of the Rhine from the Bodensee.
Calibration period kmr E VIC ERA15 HBV ERA15 VIC CHR HBV CHR r2 VIC ERA15 HBV ERA15 VIC CHR HBV CHR V E VIC ERA15 HBV ERA15 VIC CHR HBV CHR Validation period kmr E r2 VE VIC ERA15 HBV ERA15 VIC ERA15 HBV ERA15 VIC ERA15 HBV ERA15 857 Lobith 0.26 −0.08 0.35 0.73 0.58 0.54 0.88 0.96 0.23 0.32 −0.55 0.19 857 Lobith 0.40 0.60 0.67 0.64 0.08 −0.04 613 Andernach 0.41 0.22 0.61 0.88 0.57 0.58 0.65 0.97 0.14 0.23 −0.07 0.10 613 Andernach 0.42 0.52 0.62 0.60 0.02 −0.09 592 Cochem 0.54 0.78 0.55 0.96 0.86 0.83 0.65 0.98 −0.31 0.17 −0.14 0.13 592 Cochem 0.55 0.48 0.64 0.62 −0.25 −0.29 586 Kalkofen 0.21 0.13 0.50 0.97 0.61 0.20 0.95 0.97 −0.49 −0.31 −0.52 −0.08 586 Kalkofen 0.34 0.41 0.53 0.54 −0.36 −0.31 497 Raunheim 0.58 0.32 0.68 0.91 0.82 0.77 0.89 0.98 0.43 0.62 −0.34 0.21 497 Raunheim 0.40 0.48 0.67 0.59 −0.03 −0.09 428 Rockenau 0.05 −0.09 0.31 0.79 0.70 0.30 0.85 0.93 0.50 0.28 0.06 0.12 428 Rockenau −0.86 0.32 0.64 0.60 0.22 −0.06 363 Maxau −1.39 −0.71 0.59 0.72 0.50 0.48 0.92 0.96 0.27 0.22 −0.41 0.09 363 Maxau −0.62 0.28 0.39 0.49 0.18 −0.01

accurate, however, for other peaks in the validation period these parameter settings were apparently less applicable. Concerning the low flows, VIC tends to underestimate the minimum values and HBV tends to overestimate the low flows under consideration. For the duration of the most extreme low flows below a threshold of 1300 m3 /s, both models underestimated the duration of the low flows significantly and showed variable performance on the less extreme low flow periods.

2.6

Discussion and conclusions

In the view of the utility of hydrological models in climate scenario studies, the goal of this Chapter was to compare the hydrological models HBV and VIC by testing their performance for simulating historical discharge in the Rhine basin. These models have different model structures and there is no consensus in research on rainfall-runoff modeling on what model structure is to be preferred. Some research suggest, however, that the VIC approach more accurately simulates the timing of peak discharges (Troy et al., 2007). Different meteorological data sets were used as model input and HBV

40

Chapter 2. Comparing model performance of two rainfall-runoff models

Table 2.6: Analysis of peak flows and low flows at the outlet of the basin (Lobith), showing observed maximum discharge (max. Qobs ), relative difference between observed and simulated maximum discharge (dmax. Qsim ), difference in peak timing (dT ), observed minimum discharge (min. Qobs ), relative difference between observed and simulated minimum discharge (dmax. Qsim ) and duration of the low flow period under a threshold of 1300 m3 /s (DUT).
Peak flows Max. Qobs (m3 /s) dmax. Qsim VIC (%) dmax. Qsim HBV (%) dT VIC (days) dT HBV (days) Low flows Min. Qobs (m3 /s) dmin. Qsim VIC (%) dmin. Qsim HBV (%) DUT Qobs (days) DUT VIC (days) DUT HBV (days) 31/01/1995 11 775 −26.7 −16.3 2 0 09/2003 788 −20.3 18.7 141 104 93 25/12/1993 11 034 1.5 0.9 2 0 11/1997 931 −10.7 6.5 68 22 33 04/11/1998 8410 −35.1 −34.9 6 0 08/1998 983 4.5 44.4 39 15 0 07/01/2003 9366 −15.5 −32.1 5 −1 09/1996 1077 −26.0 −15.8 37 60 54 28/03/2001 8666 −12.5 −20.6 4 −1 03/1993 1228 −38.5 0.5 13 68 11

and VIC were compared at both basin and sub-basin scale using various performance criteria. Furthermore, simulated peak flows and low flows were compared. We have seen that the performance of both models was less in upstream basins than at the basin outlet (gauging station Lobith), but that for all upstream basins HBV still performed better than VIC at a daily basis. We have seen that HBV was more robust when the performance of the calibration period (E=0.49, r2 =0.75 vs. E=0.47, r2 =0.64 at Lobith) and the validation period (E=0.62, r2 =0.65 vs. E=0.31, r2 =0.54 at Lobith) were compared. In addition, HBV forced with CHR data (E=0.85, r2 =0.97 for the calibration period at Lobith) performed much better than VIC forced with CHR (E=0.44, r2 =0.81) and than both VIC and HBV forced with ERA15. For the most extreme peak flows, HBV simulated maximum discharges best (dmax. Qsim HBV 1–17%, dmax. Qsim VIC 2–27%), while VIC performed better at the moderate peak flows (dmax. Qsim HBV 21–35%, dmax. Qsim VIC 13–35%). Besides simulating measured values of discharges, timing of peak flows was investigated. It appeared that VIC displayed several days delay in estimating timing of the peak discharge. Most low flows were underestimated by VIC, where HBV showed overestimation of the low flows. Also the performance of both models in reproducing duration of low flows was poor. Hence, the semi-distributed lumped conceptual HBV model performed much better than the distributed land surface model VIC. This deflects from the general idea that

2.6. Discussion and conclusions

41

more complex distributed modeling better represents observed discharges as compared to simple conceptual model approaches (Refsgaard, 1996; Reggiani and Schellekens, 2003). These results support the notion that even for a well documented river basin such as the Rhine, more complex modeling does not automatically lead to better results (Booij, 2003; Uhlenbrook, 2003). We are convinced, though, that VIC should be able to perform better than it has done so far in the Rhine basin, and thus model performance might be improved (Hurkmans et al., 2008). The performance of VIC might increase using a longer calibration period and further refining the spatial distribution of adjusted parameters. Furthermore, by solving both the water and the energy balance VIC holds the potential to better describe soil-atmosphere feedback processes, if the model scheme were to be combined with an atmospheric model. In the line of small-scale hydro-meteorology and modeling the effects of land use change, this is a conclusive reason for further development (Hurkmans et al., 2008). Moreover, VIC has performed well in the past, for example in studies by Liang et al. (1994) and Troy et al. (2007). But also the HBV model for the Rhine basin can be improved. Lake retention for example, is not implemented yet in both models. Especially concerning the Bodensee, a large upstream lake in the Rhine basin, this is a quite drastic simplification and an obvious potential for further improvement. We subscribe the recommendation of (Seibert, 1999), that model development and calibration is an undertaking that should not be carried out by a single researcher, but requires scientific dialog. The results also lead us to the conclusion that forcing data has a considerable influence on model performance, irrespectively to the type of model structure. It emphasizes the need for ground-based meteorological measurements and a suggestion might be to correct downscaled climate model re-analysis data such as ERA15, whenever measurements are available. It should be kept in mind that comparing mean values of precipitation and temperature provides little guide to the quality of the data during more extreme events that affect hydrological systems. Pitman and Perkins (2007), for example, propose a probability density function based assessment and a skill score that shows a climate model’s ability to simulate the 95th rainfall percentile. Comprehensive comparison and correction of downscaled climate model output is a challenging task for further research. The conclusion as to the application of hydrological models in climate scenario studies, then, is that for the Rhine basin HBV is preferred, since it has shown better overall performance and seems to be more robust than VIC. The extreme events were simulated best by HBV, which implies that HBV can provide the most reliable indication of possible future shifts in extreme events due to climate change. The more realistic representation of evaporation processes by VIC than HBV did not result in better performance even in the dry periods, when the evaporation volume is substantial in the water balance. The final advantage of HBV over VIC is that HBV has

42

Chapter 2. Comparing model performance of two rainfall-runoff models

short computation times, which makes it suitable for simulating long time series of the many available different climate scenarios. Acknowledgements This research was supported by the project ACER (A7) under the Dutch BSIK Climate Changes Spatial Planning program. We wish to thank H. Buiteveld from the Dutch Institute for Inland Water Management and Waste Water Treatment (RIZA) and the German Federal Institute of Hydrology (BfG) for both allowing the use of the HBV model, and providing meteorological and discharge measurement data. D. Jacob and E. Mazurkewitz from the Max Planck Institute for Meteorology are kindly acknowledged for providing meteorological data. Finally we thank two anonymous colleagues for occasional data editing and their constructive comments that helped to improve this Chapter.

Chapter

3

Simulating low probability peak discharges using resampled climate modeling data

Te Linde, A.H., Aerts, J.C.J.H., Bakker, A.M.R. and Kwadijk, J.C.J., 2010. Simulating low probability peak discharges for the Rhine basin using resampled climate modeling data. Water Resources Research, 46 (WR03512), doi: 10.1029/2009WR007707. Abstract Climate change will increase winter precipitation and in combination with earlier snowmelt it will cause a shift in peak discharge in the Rhine basin from spring to winter. This will probably lead to an increase in the frequency and magnitude of extreme floods. In this Chapter we aim to enhance the simulation of future low probability flood-peak events in the Rhine basin using different climate change scenarios and downscaling methods. We use the output of a regional climate model (RCM) and a weather generator to create long, resampled time series (1000 years) of climate change scenarios as input for hydrological (daily) and hydrodynamic (hourly) modeling. We applied this approach to three parallel modeling chains, where the transformation method from different resampled RCM outputs to the hydrological model varied (delta change approach, direct output, and bias-corrected output). On the basis of numerous 1000-year model simulations the results indicate a basin-wide increase in peak discharge in 2050 of 8% to 17% for probabilities between 1/10 and 43

44

Chapter 3. Simulating low probability peak discharges

1/1250 years. Furthermore, the results show that increasing the length of the climate data series using a weather generator reduced the statistical uncertainty when estimating low probability flood-peak events from 13% to 3%. We further conclude that bias-corrected direct RCM output is to be preferred over the delta change approach, because it provides insight in geographical differences in discharge projections under climate change. Also, bias-corrected RCM output can simulate changes in the variance of temperature and rainfall and in the number of precipitation days, as changes in temporal structure are expected under climate change. These added values are of major importance when identifying future problem areas due to climate change, and when planning potential adaptation measures.

3.1

Introduction

Recent research findings conclude that climate change will increase the risk of flooding around the world (Milly et al., 2002, 2005; IPCC, 2007a). In North-West Europe, both observations and models agree on a trend towards wetter winter conditions (EEA, 2008). In the Rhine basin, the peak discharge is likely to move from spring to winter in many areas due to earlier snowmelt in addition to increased precipitation in winter (e.g. Middelkoop et al., 2001; Kundzewicz et al., 2007). This will probably lead to an increase in flood frequency (Hooijer et al., 2004; Pinter et al., 2006). As extreme discharge events are more important in water management than mean values, estimates of changes in the frequency and magnitude of floods are desired for the planning of adaptation measures. For estimating future flood frequencies, climate change scenarios are used as input for hydrological models. For sub-continental basins like the Rhine, however, the spatial resolution of a general circulation model (GCM) is inadequate for forcing a hydrological model and downscaling is required. This can be done by means of perturbation of historical series with average changes of climate parameters, referred to as the delta change approach (Lenderink et al., 2007a; Te Linde, 2007), by means of statistical downscaling (Jacob et al., 2008), or by dynamic downscaling using regional climate models (RCMs) that are nested in a GCM (Kay et al., 2006). The latter is known as the direct forcing approach. This is an emerging field as climate models continue to improve, and different methods are available to downscale and transform climate change scenarios for hydrological modeling studies (Wilby et al., 2000; Menzel et al., 2006; Lenderink et al., 2007b). All methods result in future meteorological time series that are used as input in hydrological models to simulate the hydrological response to climate forcing. The resulting discharge series can then be used to estimate future flood-peak probabilities. The estimation of probabilities of very rare extreme events (> Q200 , i.e. a discharge with a return period longer than 200 years), however, is far from trivial. For the

3.1. Introduction

45

present conditions, frequency analysis is often applied on historical discharge series, which requires the extrapolation of fitted extreme value distributions (Kottegoda, 1980; Garrett and M¨ller, 2008). Safety levels along the Rhine are relatively high, u up to 1/1250 per year at Lobith, and with ∼ 110 years observed discharge data available, statistical extrapolation leads to very large uncertainties (Klemeˇ, 2000a,b). s More sophisticated approaches combine weather generators with hydrological models (Buishand and Brandsma, 2001; Orlowsky et al., 2007) to create such long discharge series that extrapolation is redundant. This approach is also not without debate, since it requires hydrological modeling of extreme events far beyond historic event size. Earlier hydrological studies on the effects of climate change on peak discharges in the Rhine basin revealed that mean winter discharges are expected to increase by 5–30% and mean summer discharge to decrease by 0–45% by 2050 (compared to the current climate), using a range of climate change scenarios, from different GCMs, RCMs, and hydrological modeling methods (Buishand and Lenderink, 2004; Hundecha and B´rdossy, 2005; Fujihara et al., 2008). As a consequence, the once in a 1250-year flood event at the gauging station of Lobith (which is the current design discharge at the border of Germany and the Netherlands) is estimated to increase from 16 000 m3 /s at present to somewhere between 16 500 and 19 500 m3 /s in 2050 (Kwadijk and Middelkoop, 1994; Grabs, 1997; Shabalova et al., 2003; Lenderink et al., 2007a). A shortcoming of these studies is that they used hydrological models running at a monthly or 10-day time step, and applied a linear interpolation between model output and observed daily values of extreme discharge to estimate the climate change impact on daily based peak flow events. For a more extensive review of recent climate change related studies in the Rhine basin, the reader is referred to Vellinga et al. (2008). No prior studies have been performed for the Rhine basin that combine and compare different downscaling methods and projections of climate change with the aim of estimating flood-peak probabilities. Methods targeted at simulating long time series of future discharges are still unexplored. Also, high quality hydrodynamic models simulating wave propagation through the main river channels and floodplains are hardly applied in climate change related hydrological modeling. In this Chapter, we aim to improve the estimation of future low probability floodpeak events in the Rhine basin. We will use an approach that enables the use of RCM output and a weather generator for creating long, resampled time series of climate change scenarios as input for hydrological (daily) and hydrodynamic (hourly) modeling for the Rhine basin. We will apply this approach on three parallel modeling chains, where the difference lies in the transformation method (delta change approach, direct, and bias-corrected direct) of different RCM outputs to hydrological model input. The results are compared and analyzed for the control climate period (1961– 1995) and the projection year 2050, both at basin-wide and sub-basin scale.

46

Chapter 3. Simulating low probability peak discharges

N
Netherlands
Lobith

100 Kilometers

Lower Rhine Lahn Belgium
Andernach Kaub

Germany

Luxembourg

Middle Rhine
Maxau

Main

Mosel

Neckar

France

Basel

Upper Rhine Switzerland

Austria

Figure 3.1: Map of the Rhine basin. The Chapter is structured as follows. Section 3.2 shortly describes the Rhine basin. In Section 3.3 the method is introduced, specifying the hydrological, statistical and climate models. Section 3.4 evaluates changes in precipitation, and Section 3.5 evaluates simulated discharge and flood-peak probabilities. Finally, the discussion and conclusions are presented in Sections 3.6 and 3.7.

3.2
3.2.1

The Rhine basin
Geographical Characteristics

The Rhine is a cross-boundary river located in North-West Europe and has a length of ca. 1320 km. It originates in the Swiss Alps, and flows through parts of Germany, France, and Luxembourg, before it enters the Netherlands at Lobith (Figure 3.1). The Rhine basin comprises an area of ca. 185 000 km2 , of which the major part (160 800 km2 ) lies upstream of Lobith. About 51% of the Rhine basin is used for agriculture, 39% is forested, 5% is built-up area, and the remaining 5% is surface water or bare rock and glaciers (Disse and Engel, 2001; Middelkoop et al., 2001).

3.2. The Rhine basin

47

Table 3.1: Basin and sub-basin discharge characteristics. Surface area (km2 ) is defined by the basin area upstream of the gauging station.
Basin Gauge Surface area (km2 ) 160 800 139 549 27 088 5304 27 142 12 710 50 624 Mean Q (m3 /s) 2302 2116 334 48 176 141 1297 Min Q (m3 /s) 665 618 10 0 44 3 379 Max Q (m3 /s) 11 885 10 406 4020 730 1991 2105 4430 Mean annual max Q (m3 /s) 6781 6494 2190 364 1043 1133 3191 Data period

Rhine Rhine Mosel Lahn Main Neckar Rhine

Lobith Andernach Cochem Kalkofen Raunheim Rockenau Maxau

1961–2004 1961–2004 1961–2004 1961–2004 1989–2005 1971–1990 1961–2004

The Rhine basin is densely populated and the river is intensively used for inland shipping (Jonkeren et al., 2007). The Rhine also provides water supply for the cooling of energy plants, and for industrial, domestic, and agricultural purposes (Grabs, 1997). In order to protect vital economic activities in the Rhine basin, dikes have been constructed in the lower Rhine delta since the 19th century and they have been widened and heightened since. To facilitate inland shipping, the main Rhine branch has been straightened and weirs have been constructed. As a result, the river Rhine is currently a completely regulated river system (Lammersen et al., 2002). Two extreme peak discharge events that caused flooding in the Lower Rhine and near flooding the Netherlands in 1993 and 1995 exemplified the vulnerability of the river basin to extreme flood peaks.

3.2.2

Hydrological Characteristics

A substantial amount of precipitation in the Alpine area is stored temporarily as snow in winter (DJF). Both rainfall and meltwater contribute to discharge generation, depending on the season (Pinter et al., 2006). Table 3.1 describes the basin and subbasin characteristics. The mean annual discharge at the Lobith gauging station is ca. 2300 m3 /s, and the maximum observed discharge is 12 600 m3 /s in 1926. Extreme flood events mainly occur during the winter period. During such events, meteorologic low-pressure systems cross the basin while releasing great amounts of rainfall that can last for several days. Research on flood genesis of historical floods in the Rhine basin shows that frozen soils can lead to more extreme runoff volumes, when they prevent rainfall to infiltrate (Engel, 1997). Sudden melting of snow and frozen soils can also contribute to extreme floods (Disse and Engel, 2001). As a result of the size and shape of the basin, both the volume and height of the discharge peak strongly depend on the direction, speed, and rainfall intensity associated with the low-pressure systems, and depend less on the intensities of individual storms. Therefore, different

48

Chapter 3. Simulating low probability peak discharges

flood events show quite different geneses. The variability of the 10-day precipitation volume over the whole basin, however, seems to correlate well with the variability in peak discharge in winter (DJF) at the Lobith gauging station (Beersma et al., 2008). Changes in the future discharge regime are therefore determined by (Beersma et al., 2008): 1. The variation and change in the amount of precipitation and evaporation during the different seasons. The change in evaporation affects mainly low and summer flows. 2. Change in temperature that will change the distribution between snow (delayed runoff) and rainfall (direct runoff) in the Alpine region, particularly in winter. Temperature thus determines the length of the snow season and affects the regime of the river. 3. The change in the (relative) variability of multi-day precipitation amounts for the Rhine, and especially the 10-day precipitation amounts. This affects in particular the magnitude of peak flows at Lobith. Increases of the 10-day precipitation variability tend to increase peak flows while decreased variability leads to decreased peak flows. 4. The timing of peak flows entering the main branch from the major sub-basins determines the magnitude of the flood peak further downstream. Changes in flood management strategies, together with meteorological changes, can affect this timing. Most floodplains are embanked, which narrows the riverbed leading to increased peak water levels. The safety levels of embankments vary between the different riparian countries of the Rhine basin. In Germany, safety levels vary between 1/200 to 1/500 per year, while in the Netherlands the once in 1250-year flood (now estimated at 16 000 m3 /s at Lobith) is the boundary condition for designing dike heights (Lammersen et al., 2002; Diermanse, 2004). Because safety levels in the Rhine basin are relatively high, the design discharges estimated by using extreme value distributions are uncertain since observed discharge records only exist for 110 years (Klemeˇ, s 2000a,b; Sprokkereef, 2001). In practical handbooks, it is recommended that at least 200 years of discharge data are needed to confidently estimate the once in 100 year event (e.g. Shaw, 2002).

3.3

Methodology

The research approach followed in this Chapter (see Figure 3.2) compares three different methods; each one using weather generators to obtain long time series of

3.3. Methodology

49

Method 1 Delta change approach Control climate Observed daily P and T 1961 - 1995

Method 2 Direct RCM output

Method 3 Bias-corrected RCM output

ECHAM 5-RACMO 1961 - 1995 direct

ECHAM5-RACMO 1961 - 1995 direct

Resampling P and T using Beersma et al. (2001)

bias correction

1000 yrs daily P and T reference Climate change scenarios KNMI'06 G,Gp,W,Wp, ECHAM 5-RACMO delta values

1000 yrs daily P and T reference

1000 yrs daily P and T reference

ECHAM 5-RACMO 1950 - 2100 direct

ECHAM 5-RACMO 1950 - 2100 direct

35 yrs P and T ca. 2050

35 yrs P and T ca. 2050

35 yrs P and T ca. 2050

Resampling P and T

bias correction

1000 yrs daily P and T for 2050 Hydrological models

1000 yrs daily P and T for 2050

1000 yrs daily P and T for 2050

HBV - SOBEK

HBV - SOBEK

HBV - SOBEK

1000 yearly max Q

1000 yearly max Q

1000 yearly max Q

Extreme value analysis Return periods max Q Return periods max Q Return periods max Q

Figure 3.2: Flowchart displaying all modeling steps.

50

Chapter 3. Simulating low probability peak discharges

Table 3.2: Delta values of climate change scenarios for the year 2050, based on decade values and displayed per season, compared to basin-wide, seasonal means of observed temperature and precipitation. RACMO refers to the RCM simulation ECHAM5RACMO2.1, and G, Gp, W and Wp are four KNMI’06 scenarios, based on a suite of GCM and RCM simulations. The four KNMI’06 scenarios are separated based on expected temperature rise in 2050 (moderate (G) and warm (W)), and on scenarios with (p) or without (-) a strong change in atmospheric circulation.
Temperature (◦ C) Observed 1961–1995 d T G (◦ C) d T Gp (◦ C) d T W (◦ C) d T Wp (◦ C) d T RACMO (◦ C) Precipitation (mm/day) Observed 1961–1995 d P G (%) d P Gp (%) d P W (%) d P Wp (%) d P RACMO (%) DJF 0.18 0.89 1.15 1.78 2.30 1.52 DJF 2.60 3.62 7.09 7.24 14.18 5.38 MAM 7.38 0.88 1.27 1.75 2.54 1.36 MAM 2.61 3.02 3.05 6.05 6.11 −0.42 JJA 15.92 0.86 1.40 1.71 2.80 1.66 JJA 3.21 2.77 −9.54 5.54 −19.07 −4.72 SON 8.4 0.87 1.28 1.74 2.56 1.50 SON 2.59 2.94 −3.00 5.89 −6.00 1.58 Year 7.97 0.87 1.27 1.75 2.55 1.51 Year 2.73 3.09 −0.60 6.18 −1.20 0.45

daily discharge under different climate change scenarios for 2050 (Dettinger, 2005; Orlowsky et al., 2007). All methods use the same nearest neighbor resampling technique to extend the relatively short time series (∼ 35 years) to 10 000 years (Beersma et al., 2001; Buishand and Brandsma, 2001; Leander and Buishand, 2007). Bakker and Van den Hurk (2009) showed that the statistical properties of the resampled time series are close to the statistical properties of the input data. We applied this resampling technique within three methods that either use historical time series or RCM outputs as baseline climate information (Figure 3.2). The first method uses historical climate data (1961–1995) (Sprokkereef, 2001) that are converted to a climate change scenario using the delta change approach. The second method directly uses the output from an RCM simulation (ECHAM5-RACMO2.1: 1950–2100). The third method uses the same RCM output but applies an additional bias correction. The three methods result in three time series of 10 000 years of precipitation and temperature. Periods of 1000 years of daily values were then used as input to a hydrological and a hydrodynamic model in order to calculate 1000 years of annual maximum discharges and their probabilities (estimated by extreme value analysis). All methods and modeling steps are shown in Figure 3.2 and are further described in detail in the following sections.

3.3. Methodology

51

3.3.1

Generating long meteorological time series for different climate change scenarios in 2050

Method 1: delta change approach In 2006, the Royal Netherlands Meteorological Institute (KNMI) presented four new climate scenarios for the Netherlands (Van den Hurk et al., 2006, 2007). The KNMI’06 climate scenarios are based on a suite of GCM and RCM simulations produced within the European PRUDENCE project (Christensen et al., 2007; Lenderink et al., 2007b). In PRUDENCE, dynamical downscaling was applied using 10 RCMs and 5 GCMs, all for two 30-years time slices: a control period 1960–1990, and a future period 2070– 2100. The projected changes for 2050 result from linear interpolation between these dates. The simulation results show variable changes in projected strength of westerly winds in the Rhine basin area. A strong change in atmospheric circulation is expected to result in milder and wetter winters due to more westerly winds (Gp and Wp), when compared to scenarios without atmospheric circulation change (G and W). Hence, both temperature and changes in the atmospheric circulation are used as steering parameters to discriminate four climate change scenarios. The KNMI’06 scenarios are summarized in Table 3.2; they all describe a plausible range of possible future climate conditions, based on best available knowledge, and no statement is made on the probability of occurrence of a given scenario. Because of large uncertainties at a spatial scale smaller than the Netherlands, the KNMI provided uniform scenarios for the Netherlands, which we uniformly applied for the whole Rhine basin. A historical data set from the International Commission for the Hydrology of the Rhine basin (CHR) was used to describe the control climate for the 134 sub-basins that are used in the HBV model (see Section 3.3.2). This data set contains daily values of precipitation and temperature for the years 1961 to 1995, which are based on 174 measurement stations throughout the basin (Sprokkereef, 2001). Decadal values of the KNMI’06 climate scenarios were used to perturb the resampled historical data set of daily precipitation and temperature for the period 1961–1995 (i.e. the delta change approach (Lenderink et al., 2007a; Te Linde, 2007)). For this purpose, a year was divided in 36 decades of days (3 decades per month). The first two decades of a month always consist of 10 days and the third decade covers the remaining days. For comparison reasons, we also applied this approach using delta values of the ECHAM5-RACMO2.1 scenario (see Method 2). Decadal basin averages were calculated for the control period (1961–1995) and the future period (2036–2065). Subsequently, these decadal values were smoothed using a moving average with a window of seven decades. The decadal delta values were obtained from these smoothed decadal

52

Chapter 3. Simulating low probability peak discharges

averages. The basin-wide seasonal delta values of RACMO precipitation and temperature are summarized in Table 3.2, where the KNMI’06 scenarios are also displayed. This simple delta approach is widely used in water management practices (e.g. Aerts and Droogers, 2004; Andr´asson et al., 2004) but the delta change approach has e limitations. First, as the KNMI’06 represent spatially uniform scenarios, applying these will not take into account possible geographical differences in the change of precipitation and temperature. Second, the present day variance of temperature and the coefficient of variation of precipitation are left unchanged, while changes in variance due to climate change are expected. Third, possible changes in the number of precipitation days are not considered. We evaluated the effect of the ECHAM5-RACMO2.1 scenarios and all four KNMI’06 scenarios on monthly mean discharge (G, Gp, W and Wp). To reduce the number of simulations, we only compared one KNMI’06 scenario (Wp) to ECHAM5-RACMO2.1 on yearly maximum discharges and flood-peak probabilities. Wp was chosen because it is the most extreme of the four scenarios. Method 2: direct output ECHAM5-RACMO2.1 To overcome the limitations of the delta change approach, we applied a second method that uses direct output from a regional climate model RACMO2.1 for the years 2036– 2065 (hereafter referred to as RACMO) (Lenderink et al., 2003; Meijgaard et al., 2008). This RCM is forced by the ECHAM5-GCM member 3 output forced by the SRES-A1B emission scenario. RACMO output consists of daily data at a spatial resolution of 25×25 km. Its results suggest a transformation of the dry and wet frequency, variance and spatial variability for a future climate in 2050. Earlier versions of RACMO performed well in simulating precipitation climatology in the Rhine area (Van den Hurk et al., 2005) and mean and interannual variability of summertime temperature (Lenderink et al., 2006). We used Thiessen polygons to transform the grid-based output of RACMO to 134 sub-basins that are necessary to run the HBV model of the Rhine (see Section 3.3.2). Method 3: Bias-corrected output ECHAM5-RACMO2.1 Finally, a third method was used that improves the direct RCM output by applying a bias correction (Wilby et al., 2000; Wood et al., 2004; Orlowsky et al., 2007). Bakker and Van den Hurk (2009) compared the ECHAM5-RACMO2.1 output to observed values for the control climate period (1961–1995) and several biases were identified. A bias correction method was developed that considerably improves the spatial and monthly variability and statistics of extreme values of large-scale multi-day rainfall. The monthly bias was determined for all 134 subcatchments separately. RACMO

3.3. Methodology

53

substantially overestimates the number of wet days (daily precipitation > 0.05 mm). A correction was applied that reduces the change of the probability distribution of precipitation amounts on wet days to a minimum (bias correction 1). Based on the distribution function, target amounts for the required removal of wet days are set. Next, days with values close to this target amounts are selected and given zero precipitation. Days are only selected for drying if four of the six surrounding days are dry as well. This selection criterion leaves the multiday variability and extremes in tact. After adjusting the number of wet days, the mean (bias correction 2) and coefficient of variation (bias correction 3) are corrected by applying a power-law function. The monthly coefficients of the power-law function are calculated for the 134 subcatchments separately.

3.3.2

Hydrological models

Simulation of rainfall-runoff We used the semi-distributed conceptual HBV model (Hydrologiska Byr˚ Vattenans balansavdelning) (Bergstr¨m, 1976; Lindstr¨m et al., 1997) to simulate discharge on o o a daily basis. HBV is a soil moisture accounting model. It uses daily values of precipitation and temperature, and monthly potential evapotranspiration as input. The model uses different routines in which snowmelt is computed by a degree-day relation, and groundwater recharge and actual evaporation are functions of the water storage in a soil box. Discharge formation is presented by three linear reservoir equations and the sub-basins are linked with a simplified Muskingum approach to simulate the routing processes (see Appendix A). The HBV model for the Rhine is applied to 134 sub-basins and was developed in 1999 (Eberle et al., 2005). Potential evaporation is transformed by temperature at a daily basis, and is based on monthly potential evapotranspiration values. Potential evapotranspiration values differ between land use classes and are based on several studies by the German Bundesanstalt f¨r u Gew¨sserkunde (Eberle et al., 2005). When implementing a climate change scenario a in HBV, monthly potential evapotranspiration values remain the same, but actual evaporation values are influenced by changes in temperature. The daily-based HBV model we used in the current study performs well for the Rhine (Nash&Sutcliffe=0.85, r2 =0.97, for 1993), as is further described in Eberle et al. (2005) and Te Linde et al. (2008a). Simulation of flood wave propagation in the main river The routing scheme in HBV is not capable of simulating flood wave propagation, backwater effects, and damping in low gradient river stretches where floodplain in-

54

Chapter 3. Simulating low probability peak discharges

undation plays an important role. Therefore, a 1D-hydrodynamic model (SOBEK) was used to simulate flood routing in the main Rhine branch at an hourly time step (Delft Hydraulics, 2005) (see Appendix B). SOBEK is an integrated numerical modeling package and is based on the 1D Saint Venant equations (Chow, 1959). This model allows the implementation of structural measures, such as dike heightening, dike relocation, weirs, and detention areas. Cross sections at every 500 m, and dike locations, dike heights, and detention areas as they currently exist in the Rhine are schematized in the model. The assumption was made, though, that under current circumstances no flooding occurs in the Rhine basin. To reduce computing time, SOBEK was only used to simulate discharge during a 30-day window around the yearly maximum peak discharge, and in those calculations HBV output for the subbasins were used as boundary conditions. The increased performance on estimating peak flows due to SOBEK is described in Eberle et al. (2004). Extreme flood peaks (>8000 m3 /s at Lobith) were overestimated by an average of 5% when only HBV was used, which reduced to an average of 2% when the routing scheme in HBV was replaced by SOBEK, but the results showed large variations (up to 20% overestimation by HBV).

3.3.3

Extreme value analysis

The current practice in river management in the Netherlands to estimate the Q1250 design discharge, is to fit the Gumbel, the log-normal and the Pearson-III distributions on the annual maximum discharges at the Lobith gauging station (Diermanse, 2004). The Gumbel fit is mostly used in visualizations of extreme value analysis at Lobith. We chose to extend the Gumbel or type I extreme value distribution by adding a shape parameter so it evolves into the Generalized Extreme Value distribution (GEV) (see Appendix C). Three types of extreme value distribution functions (F (x)) are combined into the GEV distribution: x−γ δ
1 β

F (x) = exp − 1 + β

where γ is the location parameter, δ is the scale parameter, and β is the shape parameter. When β = 0, the GEV corresponds to the type I (Gumbel) distribution; β < 0 corresponds to the type II (Fr´chet) distribution; and β > 0 corresponds to e the type III (Weibull) distribution, which has a finite upper limit. We plotted the GEV distribution on Gumbel paper, creating a concave curve instead of a straight line when β = 0. The maximum likelihood approach was used to estimate the distribution parameters. The exceedance probability denotes the probability that a certain discharge value will be exceeded in one year. In this Chapter, we mainly use the

3.3. Methodology

55

Table 3.3: Basin-wide, seasonal means of observed temperature and precipitation, compared to simulated historical temperature by RACMO (control climate experiment, 1961–1995).
Temperature (◦ C) 1961–1995 Observed RACMO (method 2) RACMO (method 3) Precipitation (mm/day) 1961–1995 Observed RACMO (method 2) RACMO (method 3)
100

DJF 0.18 2.02 0.60

MAM 7.38 8.19 7.95

JJA 15.92 16.00 16.49

SON 8.40 9.30 8.81

Year 7.97 8.90 8.50

DJF 2.60 3.32 2.87

MAM 2.61 3.11 2.65

JJA 3.21 2.92 3.03

SON 2.59 3.07 2.72

Year 2.73 3.10 2.80

50

Bias (%)

0

-50

-100

Jan

Feb

Jun Jul Aug Sep Oct Month RACMO RACMO resampled RACMO resampled corrected

Mar

Apr

May

Nov

Dec

Figure 3.3: Box-whisker plots of monthly biases of the 134 sub-basins in mean precipitation. The black line denotes the median, the solid box denotes the range from 25th to 75th percentiles of the sample, and the whiskers show the data range. The outliers (dots) are defined as values that are more than 1.5 times the interquartile range away from the top or bottom of the box. term return period, which denotes the mean interval between two events of the same intensity and is the inverse of the exceedance probability.

56

Chapter 3. Simulating low probability peak discharges

100

50

Bias (%)

0

-50

-100

Jan

Feb

Jun Jul Aug Sep Oct Month RACMO RACMO resampled RACMO resampled corrected

Mar

Apr

May

Nov

Dec

Figure 3.4: Box-whisker plots of monthly biases of the 134 sub-basins in the coefficient of variation (CV ) of the precipitation.

3.4

Evaluation of changes in precipitation and temperature
Comparing observed and RACMO data for the control climate (1961–1995)

3.4.1

Here we show the quality of the RACMO regional climate model in reproducing the observed precipitation and temperature, where we compare method 2 (no bias correction) and 3 (with bias correction). Figure 3.3 and 3.4 show the distribution of the monthly biases of the 134 sub-basins for the mean precipitation and for the coefficient of variation (CV ), respectively. In the winter months, the mean precipitation is substantially overestimated, up to biases of almost a factor two in some sub-basins in Switzerland. Conversely, the winter CV is substantially underestimated. RACMO (method 3) resembles summer characteristics of both the mean and the CV of observed precipitation much better. However, there is a large spatial spread in biases in both the mean and the CV over the Rhine basin for all months. Seasonal changes in basin-wide means of precipitation and temperature obtained by RACMO (method 2 and 3) are compared to observed values in Table 3.3. It displays that the seasonal basin-wide precipitation and temperature means are represented well by RACMO after bias correction (method 3).

3.4. Evaluation of changes in precipitation and temperature

57

100 10−day precipitation (mm) 80 60 40 20 0 Observed RACMO method 2 35 years RACMO method 2 resampled RACMO method 3 resampled

Figure 3.5: Box-whisker plots of basin-wide average 10-day precipitation for the control climate (1961–1995) The box-whisker plots of basin-wide average of 10-day precipitation for the control climate in Figure 3.5 show that the median is overestimated by RACMO (method 2) by 20%, and that the 1st and 3rd quartiles are overestimated by 7–30%. Figure 3.5 also shows that the resampling of the RACMO output (method 2) hardly changes the general statistical characteristics when compared to the 35 years of RACMO output for the control climate period. This demonstrates that errors due the use of the stochastic rainfall generator can be considered small compared to the errors resulting from GCM–RCM simulations. This was expected since the resampling is designed to preserve the general statistical descriptors while generating very long time series including extreme multiday precipitation events. In addition, Figure 3.5 shows that after the bias correction the resampled RACMO output resembles observed values very well. The median is now overestimated by 1% only. Bakker and Van den Hurk (2009) further explains that the yearly pattern in the bias in both the median (Figure 3.3) and the CV (Figure 3.4) largely disappeared and that the spatial distribution of the bias considerably decreased after bias correction. To evaluate the performance for extreme peak rainfall events, the yearly maxima of 10-day values are plotted against their return period in Figure 3.6. The figure shows the extreme value distribution of observed values of 1961–1995, 2500 years resampled observed values, 2500 years resampled RACMO output for the control climate (method 2), and 10 000 years of bias corrected RACMO output (method 3) for both the control climate and the future climate (2050). Increasing the number of data points from 35 to 2500 or 10 000 results in smoothing of the extreme value distributions. Contrary to the median and the 1st and 3rd quartiles (Figure 3.5), the Rhine basin average 10-day precipitation amounts obtained by method 2 are largely underestimated when compared to (resampled) observed values. The 10-day precipitation extremes poorly resemble the 10-day extremes as assessed by resampling

58

Chapter 3. Simulating low probability peak discharges

Return period (year) 190 Average 10−day precipitation (mm) 180 170 160 150 140 130 120 110 100 2 4 6 8 Standardized Gumbel variate 10 50 200 1250 10000

Resampled observed (method 1) 2500 yrs RACMO control climate (method 2) 2500 yrs RACMO control climate (method 3) 10 000 yrs RACMO 2050 (method 3) 10 000 yrs

Figure 3.6: Extreme value plot of basin wide, resampled 10-day precipitation values. Displayed are 2500 years of resampled observations of the control climate (1961–1995, black dots), 2500 years of RACMO method 2 for the control climate (red crosses), 10 000 years of RACMO method 3 for the control climate (green line), and 10 000 years of RACMO method 3 for 2050 (blue line).

3.5. Evaluation of simulated discharge and flood-peak probabilities

59

of the observed time series (method 1). After correction, the large extremes with return periods of 10 years and more fit the extremes of resampled observed time series very well (Figure 3.6). The large improvement of the RACMO output after bias correction emphasizes the need for bias correction of RCM output. For a more complete description of the bias correction of RACMO output and a more elaborate discussion of the resulting changes in precipitation and temperature for the control climate, see Bakker and Van den Hurk (2009).

3.4.2

Climate change scenarios: comparing ECHAM5-RACMO with the delta change approach

Projected changes in mean seasonal values of temperature and precipitation for the year 2050 are displayed in Table 3.2. According to the climate change projections, a rise in mean annual temperature of 0.9◦ C for the G scenario and 2.6◦ C for the Wp scenario can be expected by 2050. With a value of 1.5◦ C, the bias-corrected RACMO scenario lies just below the middle of those scenarios. Changes in projected precipitation vary considerably between scenarios and seasons. The G and W scenarios project an increase of precipitation by 3 and 6% in all seasons, while the Gp scenario is comparable to the W scenario in the winter, but shows a decrease in mean precipitation by 9.6% in the summer. The Wp scenario projects the most extreme increase of 14.2% during the winter months and a decrease in precipitation in summer and early autumn up to 19.1%. Like the projections for changes in temperature, the RACMO precipitation scenario falls in between all four KNMI’06 scenarios. It displays an increase of 5.4% in mean precipitation in the winter months and a decrease of 4.7% in the summer months. Van den Hurk et al. (2006) compared mean seasonal change in precipitation and temperature of the GCMs that are used to construct the KNMI’06 scenarios. This study showed that the ECHAM5 output used to force RACMO was positioned near the average values of those GCMs, which explains the observations above.

3.5

Evaluation of simulated discharge and flood-peak probabilities

The impact of climate change on the simulated change in discharge behavior is analyzed both on mean monthly discharge values and on daily flood peaks. In the following paragraphs, first the effects of methods 2 and 3 (RACMO runs with and without bias correction) on present-day simulated flood-peak probabilities are compared. Second, the performance of simulated discharge by method 3 at sub-basin scale is analyzed by comparing basic statistical descriptors for the control climate period

60

Chapter 3. Simulating low probability peak discharges

x 10 2

4

Discharge (m3/s)

1.5

1

0.5

0 Observed Method 1 Method 2 Method 3

Figure 3.7: Box-whisker plots of daily values of observed and simulated discharges for the control climate (1961–1995) at Lobith. Displayed are observed discharges (35 years) and simulation results from different resampled forcing datasets (1000 years): observed precipitation and temperature (CHR, method 1),; direct RACMO output (method 2); and bias-corrected RACMO output (method 3). with the observed discharge (1961–1995). Third, the projected impact of different climate change scenarios on mean monthly discharge is evaluated for all three methods. Finally, the impact of climate change on flood-peak probabilities is examined, again using each method.

3.5.1

Effect of meteorological bias correction on simulated discharge and flood-peak probabilities for the Control Climate (1961–1995)

Section 3.4.1 showed that the bias-corrected RACMO output well resembles both mean and extreme values of observed precipitation and temperature, in contrast to direct RACMO output. Here we discuss the effect of the bias-corrected data on simulated discharge at the Lobith gauging station. In order to evaluate which forcing data result in a good representation of current discharge behavior, we have compared discharge observations with simulated values resulting from HBV forced with observed meteorological data (method 1) and different RACMO outputs (method 2: direct, and method 3: bias-corrected, see Figure 3.2). The result of this comparison is displayed in a box-whisker plot (Figure 3.7). The many outliers of yearly maximum values in this plot show the skewed distribution of daily discharge. When compared to

3.5. Evaluation of simulated discharge and flood-peak probabilities

61

precipitation in Figure 3.5, the daily discharge is much more skewed than 10-day precipitation volumes. The use of method 1 slightly overestimates the median (4%) and other quartiles of discharge when compared to observed discharges. The many more outliers are a direct result of the sampling size (35 years vs. 1000 years). When RACMO output is directly used as forcing data (method 2), the median of simulated daily discharge is overestimated by 470 m3 /s (23%). All simulated discharges appear to be higher than observed and the interquartile range is wider than for observed values. Also, direct RACMO output (method 2) results in more extreme flood peaks (outliers) than method 1 for the control climate. The bias-corrected precipitation and temperature generated by RACMO (method 3), though, result in a simulated median discharge of 2177 m3 /s, which is an over estimation of only 5%. Comparable small over estimations are observed for the 25th and 75th percentiles. Also the tail of the discharge distribution resulting from bias-corrected RACMO output resembles discharges obtained from the measured climate series (method 1) more closely than results forced by direct output. Considering basic statistical descriptors, such as medians and percentiles, bias correction of RACMO forcing data results in simulated discharge that is more similar to observed discharge than in a model structure without bias-correction. Figure 3.10a shows yearly maximum observed and simulated discharge data (method 1) plotted against their return periods for the control climate period. A Gumbel distribution was fitted through the 100 years of observed yearly maxima at Lobith, and through the 1000 years of simulated yearly maximum discharges for the control climate period (Method 1, Figure 3.10a). Note that only yearly maxima above a threshold of 8000 m3 /s are displayed in Figure 3.10, but that all values were included when fitting the extreme value distributions. Also shown are the 95% confidence intervals that can be interpreted as the statistical uncertainty of the distribution fits. At a return period of 100 years, the estimated peak discharge is 13 000 ± 1750 m3 /s (13%) according to the width of the 95% confidence interval for observed values. The Gumbel fit through the 1000-year control climate period agrees very well with the observed data fit. The estimated peak discharge is also 13 000 m3 /s at a return period of 100 years, but the 95% confidence interval is much narrower with ± 450 m3 /s (3%). Thus, the confidence interval narrows to a quarter of its original size while increasing the sample size from 100 to 1000 years, reducing the statistical uncertainty that results from fitting extreme value distributions and extrapolation of those fits. The ten most extreme discharges of the 1000-year control climate period (method 1) lie around the lower 95% confidence interval, and the two most extreme peaks even lie outside the confidence intervals of the Gumbel fit. The Gumbel distribution seems a good fit for historical data, but apparently fails to describe the upper tail of the distribution of discharges according to the simulated discharge for the control climate, as is plotted in Figure 3.10a. Therefore, a shape parameter was introduced in the GEV distribution. The result for the control climate situation is shown as

62

Chapter 3. Simulating low probability peak discharges

the dashed line in Figure 3.10b, where it can be seen that the GEV fit bends down also describing the most extreme values of the sample. Figure 3.10b also displays the extreme discharges resulting from method 2 at Lobith for the control climate period. Contrary to our observation for extreme 10-day precipitation values in Section 3.4.1, yearly maximum discharges resulting from direct RACMO output (method 2) are overestimated when compared to observations. The overestimation varies between 8 and 25%, in the range of return periods of 50 to 1250 years, of which the latter corresponds with an overestimation of 4000 m3 /s. After bias correction (method 3) the yearly maximum discharges resemble the control climate (method 1) much better, but above 12 000 m3 /s (T =100 years) both data sets diverge and discharges resulting from RACMO (method 3) are still overestimated according to the simulation results. The difference increases up to 3500 m3 /s at a return period of 1250 years (See also Table 3.7). As explained in Section 3.2.2, the common assumption and observation is that yearly maximum values of 10-day precipitation and discharge are closely related. Our results on the effect of bias correction contradict this; the yearly maximum 10-day precipitation was underestimated before bias correction (Figure 3.6), whereas the yearly maximum discharge was overestimated. It appears that the yearly maximum 10-day precipitation can occur in all seasons, while the yearly maximum discharge mainly occurs in winter, which might explain our observation. The overestimation of yearly maximum discharges before bias correction is in line with the overestimation of daily values of both precipitation and discharge as displayed in Figure 3.5 and Figure 3.7, respectively. Overall, we have shown that an error in the meteorological forcing data propagates to an error in simulated peak discharges, leading to an incorrect representation when compared to observed values. The considerable improvement of simulated discharge as a result of bias correction stresses the need for bias correction of meteorological data from RCM output when using simulated discharges based on those data in climate change scenario studies.

3.5.2

Performance of simulated discharge by method 3 at subbasin scale

One of the major differences between methods 2 and 3 on the one hand, and method 1 on the other hand, is a spatially explicit climate scenario at the RCM grid scale of 25×25 km in method 2 and 3, instead of a geographically uniform application of relative changes in precipitation and temperature as method 1 does. We therefore tested the performance of the simulated discharge obtained by method 3 at the sub-basin scale for the control climate (1961–1995). Since the resampled data set is a synthetic time series, a day-to-day comparison to observations with performance indicators

3.5. Evaluation of simulated discharge and flood-peak probabilities

63

Table 3.4: Basin and sub-basin discharge characteristics for the control climate (1961– 1995). Displayed are observed values and discharges obtained by method 1 (CHR 35 years and CHR resampled (1000 years)) and method 3 (bias-corrected RACMO resampled). SD is the standard deviation from the mean.
Basin Gauge Rhine Lobith Observed CHR 35 (mtd 1) CHR (mtd 1) RACMO (mtd 3) Observed CHR 35 (mtd 1) CHR (mtd 1) RACMO (mtd 3) Observed CHR 35 (mtd 1) CHR (mtd 1) RACMO (mtd 3) Observed CHR 35 (mtd 1) CHR (mtd 1) RACMO (mtd 3) Observed CHR 35 (mtd 1) CHR (mtd 1) RACMO (mtd 3) Observed CHR 35 (mtd 1) CHR (mtd 1) RACMO (mtd 3) 10th percentile (m3 /s) 1156 1137 1202 1177 81 82 91 84 13 6 7 7 70 66 69 67 46 41 43 37 691 667 699 685 Mean (m3 /s) 2299 2355 2456 2525 335 340 352 371 49 46 50 48 164 197 209 212 140 141 148 153 1297 1272 1334 1385 90th percentile (m3 /s) 3767 3921 4275 3997 736 741 855 765 104 111 118 121 322 409 447 436 259 264 304 278 2034 2054 2280 2127 Mean ann. max. Q (m3 /s) 6691 7053 7405 7630 2147 2077 2134 2217 369 337 374 349 995 1019 1088 1151 1008 936 1028 1149 3207 3505 3639 3964 SD (m3 /s) 1204 1290 1317 1415 365 369 372 401 59 59 64 62 176 191 207 221 141 136 149 170 549 590 610 686 Skewness – 1.96 1.99 2.10 1.93 3.10 3.15 3.08 2.84 3.58 3.31 3.35 3.24 4.61 3.29 3.46 3.45 4.38 4.25 4.36 4.50 1.10 1.45 1.50 1.55

Mosel Cochem

Lahn Kalkofen

Main Raunheim

Neckar Rockenau

Rhine Maxau

that are common practice in hydrologic modeling (i.e. Nash & Sutcliffe, volume error or correlation) is not possible. We therefore compared statistical descriptors of the obtained data sets to test the performance of simulated discharge by method 3 at the basin and sub-basin scales. Table 3.4 displays the basic statistical descriptors of daily discharges for the control climate (1961–1995). We compared 35 years of observed discharge with simulated discharge by HBV that was forced with observed precipitation and temperature data (CHR data). We also used resampled CHR data (method 1) and resampled bias-corrected RACMO data (method 3) to create 1000 years of simulated discharge and compared those to the other data sets. In this way, we distinguished: 1. the difference between observed and simulated discharge by HBV resulting from observed meteorological data (CHR 35),

64

Chapter 3. Simulating low probability peak discharges

Table 3.5: Seasonal change in discharge at Lobith according to KNMI’06 scenarios and bias-corrected RACMO output, for 2050.
Discharge (m3 /s) Control climate (1961–1995) d Q G (mtd 1) (%) d Q Gp (mtd 1) (%) d Q W (mtd 1) (%) d Q Wp (mtd 1) (%) d Q RACMO (mtd 3) (%) DJF 2778 6.71 5.58 13.04 10.56 14.20 MAM 2518 2.03 3.00 4.47 6.50 −0.75 JJA 2314 −2.16 −17.40 −3.94 −31.68 −17.36 SON 1809 1.47 −17.63 3.02 −33.25 −12.63 Year 2355 2.01 −6.61 4.15 −11.97 −4.14

2. the difference between simulated discharge resulting from observed and from resampled observed meteorological data (CHR resampled, method 1), 3. the difference between simulated discharge resulting from resampled observed meteorological data (CHR resampled, method 1) and from resampled RACMO output (method 3). The distribution of the discharge at Lobith and from the Upper Rhine is less skewed than from the other sub-basins, as shown in Table 3.4. This can be explained by the snowmelt component in the discharge regime of the Alpine area in the Upper Rhine. Snowmelt results in higher base flow values and therefore displays a less skewed probability distribution than discharge from mainly rain fed sub-basins. For the 10th percentile, the mean, the 90th percentile, and the skewness, all three simulated discharge data sets agree very well with observed values for all sub-basins. The standard deviation of discharge data obtained by method 3, though, is somewhat higher for all basins, except for the Lahn. Together with the observation that the mean annual maximum is overestimated by method 3 for the same basins, it follows that this data set contains slightly higher peak discharges than observed. Overall, the simulated discharge obtained by method 3 resembles the observed spatial structure in the discharge regime very well.

3.5.3

Climate change impact on monthly mean discharge

Mean seasonal and monthly changes in discharge for Lobith were analyzed in detail, and for several sub-basins for the winter season only. The impact of different climate change scenarios on discharge is displayed for Lobith in Table 3.5 (seasonal) and Figure 3.8 (monthly). When applying the KNMI’06 scenarios (method 1, Figure 3.2), mean winter discharge is expected to increase between 6.7% (G) and 13.0% (W). However, the projected decrease in summer and autumn stands out, ranging between −2.2% (G) and −33.3% (Wp), and this range is even more emphasized in monthly values (Figure 3.8) where September shows a maximum decrease of −40%. The bias-

3.5. Evaluation of simulated discharge and flood-peak probabilities

65

3500 Change in mean discharge (%) Jul Aug Sep Oct Nov Dec Mean discharge (m3/s) 3000 2500 2000 1500 Reference G Gp W Wp RACMO

30 20 10 0 −10 −20 −30 −40 −50 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec

1000 Jan Feb Mar Apr May Jun

Figure 3.8: Monthly mean change in discharge at Lobith according to KNMI’06 scenarios and bias-corrected RACMO output, for 2050, absolute values (left) and relative values (right). corrected RACMO scenario (method 3) most resembles Gp, except for the months of January and February. In the winter months the results of RACMO (method 3) are 14.2% higher than the average observed values; this is more than can be observed in any of the four KNMI’06 scenarios. Moreover, when comparing Table 3.5 with Table 3.2, where the delta values of the seasonal change in precipitation and temperature are displayed, Wp has higher values of increase in precipitation than bias-corrected RACMO (method 3) in the winter season (14.2 vs. 5.4%). There is a non-linear relationship between the relative increase in precipitation and temperature on one hand, and resulting discharge in the winter months on the other, which is due to inherent complex hydrological and groundwater processes (Uhlenbrook, 2003; Bogaard et al., 2005). Further comparison of Table 3.2 and Table 3.5 for other seasons reveals, for example, that the relative decrease in discharge in summer and autumn is higher than the relative decrease in precipitation volume. It is likely that the projected temperature increase results in higher evaporation rates, and therefore lower discharges. Our results on projected monthly mean change in discharge are in agreement with values obtained by previous studies on climate change in the Rhine basin (Van Deursen, 2002; Kwadijk and Rotmans, 1995; Middelkoop and Kwadijk, 2001). Mean changes in discharge in winter months at the sub-basin scale are displayed in Figure 3.9, where only the results for the Wp (method 1) and RACMO (method 3) scenarios are shown. Both scenarios roughly display the same spatial structure, with the largest increase in discharge in the Main (19–30%), followed by the Neckar, the Lahn, and the Mosel (13–24%), and the smallest changes upstream of Maxau in the Alpine area (7–18%) and in the Middle and Lower Rhine (0–12%). Compared to the basin-wide changes (Table 3.5), these results imply that mean winter discharge increase in the Main can be almost twice the basin-wide increase. The RACMO scenario shows much more geographical differentiation in mean change in winter discharge between sub-basins than the Wp scenario, due to spatial differentiation in the RACMO output of precipitation and temperature for 2050.

66

Chapter 3. Simulating low probability peak discharges

RACMO method 3

Wp method 1

Lower Rhine Lahn Middle Rhine Mosel Neckar
(%)
0-6 7 - 12 13 - 18 19 - 24 25 - 30

Lower Rhine Lahn Main Mosel Neckar Middle Rhine
Main

Upper Rhine

Upper Rhine

Figure 3.9: Mean change in winter discharge for the Wp and bias-corrected RACMO scenarios for 2050. Table 3.6: Statistical parameters of yearly maximum discharges for the control climate (1961–1995) and two climate change scenarios for 2050. All four data sets contain 1000 simulated yearly maxima. SD is the standard deviation from the mean.
Lobith Control clim. CHR (mtd 1) Control clim. RACMO (mtd 3) 2050 Wp (mtd 1) 2050 RACMO (mtd 3) Mean 6327 5623 7301 6534 SD 2192 2498 2651 2941 Skewness 0.84 1.08 0.76 1.07 Min 2011 863 1757 1326 Max percentile 15 694 19 801 18 215 25 110 25% percentile 4702 3714 5347 4402 75% 7705 7104 8999 8303

The relatively modest increase that we project in the Alpine area in Figure 3.9 is probably the consequence of only displaying the winter season (DJF), while Alpine increase is very likely most profound in the spring. Stewart (2008) compared different studies of observed and projected changes in snow cover and snowmelt-derived streamflow for the European Alps, and summarizes that a general decrease in annual snowpack, and earlier spring snowmelt runoff have been observed. Under a warming climate, the Alpine mountain range will lose some of its functions as seasonal water storage. A detailed study by Bavay et al. (2008) on two Alpine headwater catchments project a much narrower snowmelt discharge peak in spring, and heavy precipitation events in the fall are expected to change from mainly snow-fed to mainly rain-fed.

3.5. Evaluation of simulated discharge and flood-peak probabilities

67

Return period (year) 2 1.8 1.6 1.4 1.2 Yearly maximum discharge (m3/s) 1 0.8 x 10 a) 1.8 1.6 1.4 1.2 1 0.8
4

50 100

500 1250

2

x 10 b)

4

50 100

500 1250

2 x 10 c)
4

4

6

8

2 x 10 d)
4

4

6

8

2 1.8 1.6 1.4 1.2 1 0.8

50 100

500 1250

2 1.8 1.6 1.4 1.2 1 0.8

50 100

500 1250

2

4

6

8

2

4

6

8

Standardized Gumbel variate Observed discharges (110 yrs) Gumbel t observed 95% conf.int. observed Control climate (method 1) Gumbel t control climate (1000 yrs) 95% conf.int. control climate GEV t control climate (method 1) Control climate RACMO (method 3) GEV t control climate RACMO (method 3) Control climate RACMO (method 2) GEV t control climate RACMO (method 2) 2050 RACMO (method 3) GEV t 2050 RACMO (method 3) 2050 RACMO (method 1) GEV t 2050 RACMO (method 1) 2050 KNMI Wp (method 1) GEV t 2050 KNMI Wp (method 1)

Figure 3.10: Extreme value distributions and GEV fits of yearly maximum discharge at Lobith for the reference situation and the year 2050. Displayed are (a) 100 years of observed data and 1000 years of resampled and modeled data for the control climate (method 1, 1961–1995) and Gumbel distribution fits projected as a straight line with 95% confidence intervals; (b) control climate, and RACMO (method 2 and 3) for the control climate; (c) control climate, and RACMO (method 3) for 2050; (d) control climate, RACMO (method 1), and Wp (method 1) for 2050. Notice the scale difference between 10c and the others. Note that only yearly maxima above a threshold of 8000 m3 /s are displayed, but that all values were included when fitting the extreme value distributions.

68

Chapter 3. Simulating low probability peak discharges

Table 3.7: Discharge in m3 /s at Lobith at different return periods (T) for the control climate (1961–1995) and under climate change. The numbers based on their GEV fits, displayed in Figure 3.10.
Scenario CHR (mtd 1) RACMO (mtd 3) Wp (mtd 1) RACMO (mtd 3) RACMO (mtd 1) T (yr) 10 9 234 8 896 10 824 10 041 10 056 50 11 802 12 392 13 809 14 350 12 844 100 12 821 13 933 14 970 16 050 13 935 200 13 801 15 508 16 070 17 763 14 972 500 15 038 17 647 17 442 20 052 16 271 1250 16 216 19 856 18 726 22 372 17 492

Control clim. Control clim. 2050 2050 2050

Table 3.8: Relative changes in discharge at Lobith at different return periods (T) under climate change. First displayed is the relative change in %, where RACMO (method 1) and Wp (method 1) are related to CHR (method 1) for the control climate and RACMO (method 3) for 2050 is related to RACMO (method 3) for the control climate. These changes in % are then used to calculate the absolute discharges of RACMO (method 3) that are displayed in the second part (‘relative’).
Discharge 2050 2050 2050 Discharge 2050 (%) RACMO (mtd 1) Wp (mtd 1) RACMO (mtd 3) (m3 /s) ‘Relative’ RACMO (mtd 3) T (yr) 10 8.9 17.2 16.9 10 796 50 8.8 17.0 15.8 13 666 100 8.7 16.8 15.2 14 769 200 8.5 16.5 14.5 15 807 500 8.2 16.0 13.6 17 088 1250 7.9 15.5 12.7 18 271

3.5.4

Climate change impact on flood-peak probabilities

Basin-wide results at the German–Dutch Border (Lobith) The probability and volume of extreme peak values is generally of higher interest than the projected changes in mean monthly discharge. Climate change impacts on flood peak events with return periods of 200 to 1250 years were further investigated, since these are the safety levels in the Rhine basin (see Section 3.2). Yearly maximum discharges of 1000 years for the control climate (1961–1995, method 1 and method 3, see Figure 3.2), the RACMO scenario with bias correction for 2050 (method 3) and the Wp scenario for 2050 (method 1) are depicted in Table 3.6. RACMO (method 2) is not displayed or discussed in this Section. Mean yearly maximum discharge is expected to increase by 981 m3 /s according to the Wp scenario (method 1), whereas RACMO (method 3) shows an increase of 911 m3 /s, when compared to RACMO for the control climate. RACMO (method 3) results in a higher skewness and standard deviation, indicating a wider probability distribution of the yearly maxima than values

3.5. Evaluation of simulated discharge and flood-peak probabilities

69

obtained by method 1. This results from the overestimation of discharge peaks above 12 000 m3 /s at Lobith by RACMO. Both by RACMO and Wp obtained maximum values under climate change in 2050 indicate a substantial increase compared to the maximum obtained discharge in the control climate period (Table 3.6). The probability of reaching discharges in this range is estimated by extreme value analysis. In Figure 3.10c the results of RACMO (method 3) are displayed for 2050 relative to the control climate, with GEV distributions fitted through the data. Notice the outlier of 25 110 m3 /s, which is plotted outside the scale on the y-axis. For comparison reasons a RACMO scenario with the delta change approach (method 1) was constructed as well (Section 3.3.1, Figure 3.2). In Figure 3.10d, RACMO (method 1) and Wp (method 1) are plotted relative to the control climate, where it can be seen that the Wp scenario results in higher extreme discharge levels than RACMO (method 1). When Figure 3.10c and 3.10d are compared, it can be seen that RACMO (method 3) projects the most extreme scenario of increase in flood-peak probabilities. Table 3.7 summarizes information that can be obtained from Figure 3.10 for several return periods at Lobith. Presented flood peaks and their probabilities are calculated based on their GEV fits, which are also shown in Figure 3.10. The RACMO (method 1) and Wp (method 1) scenarios indicate an increase of the 1250-year event from 16 216 m3 /s to 17 492 and 18 729 m3 /s, respectively. The RACMO (method 3) scenario projects a considerably larger increase to 22 372 m3 /s. From Figure 3.10 it is obvious, though, that the RACMO (method 3) GEV fit is highly influenced by the maximum obtained flood peak of 25 110 m3 /s, which seems extraordinarily large. Our hypothesis is that this is an extreme event that is less probable than once in 1000 years. In Section 3.5.2 we discussed the likely overestimation of events above 12 000 m3 /s at Lobith when using bias-corrected RACMO (method 3) for the control climate period (Figure 3.10b). In an attempt to correct for the overestimation of RACMO (method 3), we present relative values of projected changes in flood peaks in Table 3.8, calculated from the absolute numbers in Table 3.7. First displayed is the relative change in %, where RACMO and Wp (method 1) are compared to ‘Control climate CHR (method 1)’ and RACMO direct is compared to ‘Control climate RACMO’. In this way, we correct for the overestimation of RACMO for the control climate. These relative changes in % are then used to calculate the discharges displayed in the second part of Table 3.8. RACMO (method 1) and Wp (method 1) are thus unchanged, but RACMO direct (method 3) is changed. The RACMO values obtained in this way (‘relative’ RACMO (method 3)) resemble much more the Wp (method 1) scenario than the RACMO (method 1) scenario. ‘Relative’ RACMO bias-corrected and Wp delta indicate a peak discharge increase of 12.7–17.3% for all return periods, whereas RACMO delta indicates a shift of 7.9–8.9%.

70

Chapter 3. Simulating low probability peak discharges

The extreme value plots in Figure 3.10 and the calculated parameters of the GEV fits can also be used to calculate the projected relative shifts in probabilities of certain events. According to the ‘relative’ RACMO scenario (method 3) a discharge at a return period of 1250 years in the control climate situation would occur once in 460 years in 2050. This is equivalent to an increased probability with a factor 2.7. The RACMO (method 1) scenario projects that the 1250-year discharge will increase in frequency to once in 510 years (factor 2.5), and the Wp (method 1) scenario projects an increased frequency to once in 265 years (factor 4.7). Lenderink et al. (2007a) used 90 years of simulated discharge to estimate extreme peak flows at a return period of 100 years for the year 2100, and found an increase between 10 and 30%. If we assume linear interpolation to scale these numbers back to the year 2050, these results are comparable to our results, while Lenderink et al. (2007a) used different climate and hydrological models and different emission scenarios than we did. They compared bias-corrected direct RCM model output to force their hydrological model with the delta change approach, which is partly similar to our methodology. However, they applied the delta change approach to bias-corrected reference data in the control climate, instead of observed values as we did. Shabalova et al. (2003) and Buishand and Lenderink (2004) also obtained comparable results for the year 2100. These studies, like earlier work from Kwadijk and Middelkoop (1994) and Kwadijk and Rotmans (1995), gave basin-wide estimations on projected changes in flood-peak probabilities. Kleinn et al. (2005) used the RCM CHRM, and the hydrological model WaSiM, to derive flood-frequency curves at the sub-basin scale in the Rhine basin, but they did not simulate future climate change projections. Results at sub-basin scale In addition to basin-wide projections, we calculated flood peak probabilities for seven sub-basins and analyzed the spatial variation of the results. Table 3.9 shows that the relative change in peak discharge due to climate change increases with increasing return period in all sub-basins, except for the Neckar, where the RACMO (method 3) scenarios depicts a decrease with increasing return period. For the return period of 200 years, resulting values are shown in Figure 3.11. RACMO (method 1) and Wp (method 1) show less spatial variation than RACMO (method 3); this can be expected since the delta change approach (method 1) is uniformly applied to the river basin. The major differences arise in the Main, Mosel and Lahn sub-basins. In the Mosel, RACMO (method 3) projects an increase of almost 40% at a return period of 200 years, while RACMO (method 1) and Wp (method 1) show increases of 10 and 17%, respectively. Comparable differences are found for the other return periods. In the Main and the Lahn the differences between the scenarios vary between 10 and 40%, and 6 and 32%, respectively. In the RACMO (method 3) scenario, for example, the upstream part of the Rhine will not face a large increase in flood-peak probability

3.5. Evaluation of simulated discharge and flood-peak probabilities

71

RACMO method 3 Wp method 1

Lower Rhine 17.8 16.2 Lahn 9.0 27.3 14.0 10.0 Middle Rhine Moselle 38.9 17.4 10.0
8.1 16.0 5.9

RACMO method 1

Main 31.7 21.6 11.6 Neckar 5.8 18.4 7.6

Upper Rhine 8.3 23.2 12.2

Figure 3.11: Simulated change in discharge (%) at T =200 year at sub-basin scale for the bias-corrected RACMO scenario (method 3) and Wp and RACMO scenarios obtained by the delta change approach (method 1).

72

Chapter 3. Simulating low probability peak discharges

Table 3.9: Relative change in discharge at all sub-basins at different return periods (T) under climate change.
Basin Gauge Discharge Cochem (%) RACMO (mth Wp (mtd 1) RACMO (mtd RACMO (mtd Wp (mtd 1) RACMO (mtd RACMO (mtd Wp (mtd 1) RACMO (mtd RACMO (mtd Wp (mtd 1) RACMO (mtd RACMO (mtd Wp (mtd 1) RACMO (mtd 1) 3) 1) 3) 1) 3) 1) 3) 1) 3) T (yr) 10 7.8 15.1 28.7 7.1 14.3 20.1 10.0 22.3 20.8 7.5 18.2 10.5 7.0 16.0 7.1 50 9.0 16.3 34.3 6.5 14.1 23.9 10.8 22.0 26.1 7.6 18.4 8.1 10.0 20.3 8.0 100 9.5 16.9 36.6 6.3 14.0 25.6 11.2 21.8 28.8 7.6 18.4 7.0 11.1 21.8 8.2 200 10.0 17.4 38.9 6.2 14.0 27.3 11.6 21.6 31.7 7.6 18.4 5.8 12.2 23.2 8.3 500 10.7 18.1 41.9 6.0 13.9 29.6 12.1 21.3 35.9 7.6 18.3 4.1 13.7 24.8 8.2 1250 11.3 18.8 44.9 5.8 13.9 32.0 12.7 21.0 40.2 7.6 18.1 2.4 15.1 26.4 8.2

Mosel

Lahn

Kalkofen

Main

Raunheim

Neckar

Rockenau

Upper Rhine

Maxau

due to climate change, but in some areas of the Lower Rhine the increase will be higher than the basin-wide projections. The Wp (method 1) scenario, on the other hand, projects the major change in the upstream part of the Rhine basin and projects slightly less changes in the Middle and Lower Rhine.

3.6

Discussion

The aim of this Chapter was to enhance the simulation of future low probability floodpeak events in the Rhine basin. We used an approach that enabled us to use RCM data and a weather generator to create long, resampled time series of climate change scenarios as input for hydrological (daily) and hydro-dynamic (hourly) modeling for the Rhine basin. We applied this approach on three parallel modeling methods, each having a different transformation method of climate data (delta change (method 1), RACMO direct (method 2) and RACMO bias-corrected (method 3)) of different RCM outputs to hydrological model input. We used resampled meteorological time series up to 1000 years to create long series of daily discharge in order to estimate low probability flood-peak events, without extrapolating extreme value distributions. Extreme value analysis on these data resulted in a GEV distribution to describe the probability distribution instead of the commonly used Gumbel distribution at Lobith (Te Linde and Aerts, 2008). Increased sample

3.6. Discussion

73

size from 100 years of observed data to 1000 years of simulation results reduced the statistical uncertainty related to distribution fitting, and the extrapolation of those fits, from 13 to 3%. At a return period of 100 years, the estimated peak discharge at Lobith was 13 000 ± 1750 m3 /s using only 100 years of observed data, which we reduced to ± 440 m3 /s when using a resampled series of 1000 years of climate data. We further recommend using a combination of hydrological modeling and a hydrodynamic model in order to properly simulate flood wave propagation and to address planned measures and structures in the main channel of the Rhine as suggested by Lammersen (2004) and Te Linde and Aerts (2008). Evaluating simulated precipitation and temperature by direct ECHAM5 forced RACMO2.1 output for the control climate period (1961–1995) revealed a moderate performance by the RACMO RCM. After bias- correction the performance of the RACMO output increased considerably when compared to the observed values, both for seasonal means and extreme values. We saw that an error in the forcing data (direct RACMO output) of +20% (median) propagates to an overestimation of median simulated discharge of 23% compared to observed values, thus stressing the need for bias correction of meteorological data. With bias-corrected RACMO output we succeeded in obtaining realistic mean seasonal discharges in the present climate, both at the basin-wide and sub-basin scales. Other statistical descriptors, such as the standard deviation and skewness, were also well represented at both basin-wide and sub-basin scale. We identified an underestimation of the most extreme yearly maximum 10-day precipitation events before bias correction for the control climate period (1961–1995), whereas the yearly maximum discharge values were overestimated before bias correction for the same period. This contradicts the common assumption that yearly maximum values of 10-day precipitation and discharge are closely related. We explained this by the observation that the yearly maximum 10-day precipitation events occurred in all seasons, while the yearly maximum discharges mainly occurred in winter, but a thorough interpretation requires further work. The overestimation of yearly maximum discharges before bias correction was in line with the overestimation of daily values of both precipitation and discharge. However, after bias correction the overestimation of peak discharges generated by RACMO (method 3) was reduced. At flood peaks with very low probability (i.e. return periods more than 500 years), discharges obtained by RACMO (method 3) were overestimated by 8–25% for the control climate period. Hay et al. (2002) observed comparable differences in simulated and observed discharge when they used direct RCM output for hydrological simulation in three snowmelt-dominated basins in the United States, and concluded that their RCM, even after bias correction, did not contain the day-to-day variability present in observed precipitation that is necessary for basin-scale hydrological modeling. Lenderink et al. (2003) already recognized an overestimation in daily precipitation values in the winter season by an earlier version of RACMO. Future work

74

Chapter 3. Simulating low probability peak discharges

will continue to develop methods to remove biases, where more emphasis is needed on extreme events. When considering the effect of climate change on mean monthly discharge, the RACMO scenario showed an increase of 14.2% in winter and a decrease of 17.4% in the summer months. This was comparable to the range of the four KNMI’06 scenarios that showed a maximum projected increase in winter by 13.0% and a change in mean summer discharge of +3.0 to −33.3%. Comparing the KNMI Wp scenario and RACMO output at a sub-basin scale, RACMO showed more geographical differentiation in relative contribution between sub-basins than KNMI Wp. This is most likely a direct result from the geographical variation in direct RACMO output, compared to the uniformly distributed delta change approach that was applied to the KNMI Wp scenario. We have shown that RACMO (method 3) led to surprising results when absolute values of extreme discharges were considered under climate change. The GEV fit was highly influenced by the maximum obtained flood peak of 25 110 m3 /s. Our hypothesis is that this is an extreme event that is less probable than once in 1000 years. Estimates of the Q1250 in 2050 in earlier studies on climate change in the Rhine basin never reached more than 20 000 m3 /s at Lobith (Vellinga et al., 2008). We therefore continued to compare relative values of projected changes in flood peaks, to correct for the overestimation of RACMO (method 3). On the basis of 1000-year simulation results, RACMO direct (method 3) and Wp (method 1) indicated a peak discharge increase in 2050 of 12.7–17.3% for return periods between 10 and 1250 years, whereas RACMO delta (method 1) indicated a shift of 7.8–8.9%. Our results projected that a flood-peak event with a return period of 1250 years in the control climate situation will increase its frequency in 2050 according to a variety of scenarios: once in 460 years (factor 2.7, RACMO direct); once in 510 years (factor 2.5, RACMO delta); and once in 265 years (factor 4.7, Wp), in 2050. At the sub-basin scale, RACMO (method 1) and Wp (method 1) showed less spatial variation than RACMO direct in estimated flood-peak volumes under climate change. Again, this results from the geographical differentiation across the Rhine basin in direct RCM output. The major differences arose in the Main, Mosel, and Lahn subbasins. In the Mosel, bias-corrected RACMO (method 3) projected an increase of almost 40% at a return period of 200 years, while RACMO direct (method 1) and Wp (method 1) displayed increases of 10 and 17%, respectively. In the RACMO (method 3) scenario, for example, the upstream part of the Rhine will not face a large increase in flood-peak probability due to climate change, but in some areas of the Lower Rhine the increase will be higher than for the basin- wide projections. However, the Wp (method 1) approach projected the largest change in the upstream part of the Rhine basin, and slightly less changes in the Middle and Lower Rhine. These differences

3.7. Conclusions and further work

75

can be of major importance when implementing adaptation measures in relation to climate change, such as detention areas, bypasses, or dike relocations. A limitation of using RACMO is that the climate change scenario results from one GCM and one RCM, instead of an ensemble of climate models as in the KNMI’06 scenarios. However, as a result of projects such as ESSENCE (Sterl et al., 2007), PRUDENCE (Christensen et al., 2007), and the FP6 ENSEMBLES project (Jacob et al., 2008), more GCM–RCM combinations will become available that can be used to create more resampled and bias-corrected direct RCM climate change scenarios for the Rhine basin. In the current study, this was not feasible. The KNMI’06 scenarios can only be implemented by the delta change approach, but provide a valuable indication of the bandwidth of different climate change projections. This bandwidth can be seen as a delineation of the uncertainty in climate change projections. But KNMI, does not provide an indication on the probability of occurrence of the separate KNMI’06 scenarios (Van den Hurk et al., 2006).

3.7

Conclusions and further work

On the basis of numerous 1000-year model simulations the results indicate a basinwide increase in peak discharge of the Rhine in 2050 of 8% to 17% for probabilities between 1/10 and 1/1250 years. The results, furthermore, show that increasing the length of the climate data series using a weather generator reduced the statistical uncertainty when estimating low probability flood-peak events from 13% to 3%. When evaluating future climate, the delta change approach to create climate change scenarios is more transparent than using bias-corrected RCM output. It is a robust method that makes it possible to use output from climate models, even when these climate models do not represent the control climate accurately (Grabs, 1997). However, we are convinced that the use of bias-corrected RCM output is to be preferred in climate change analysis since it incorporates projections of geographical differentiation, and changes in the variance of temperature, the coefficient of variation of precipitation, and the number of precipitation days. Our results are in line with Lenderink et al. (2007a) who also found that direct RCM output was more suitable in a hydrological impact study than the delta change approach using the HadRM3H RCM. But the bias-correction can be improved, especially concerning extreme events. In addition, the use of a rainfall generator in combination with hydrological and hydrodynamic modeling is recommended when simulating low probability peak discharges under climate change. Lengthening the 1000-year model simulations to 2000 years or more, would further improve the simulation of extreme events with a probability of once in 1250 years.

76

Chapter 3. Simulating low probability peak discharges

Both scientists and policy makers recognize the inherent uncertainty associated with performing flood risk analysis within climate change scenario studies (Prudhomme et al., 2003; Menzel et al., 2006; Hingray et al., 2007; Pitman and Perkins, 2007), and policy makers struggle on decision-making related to climate change (e.g. Middelkoop et al., 2004). When identifying future problem areas and potential adaptation measures, it is of major importance to be able to estimate spatial differences in flood-peak probabilities. Our results demonstrate that bias-corrected direct RCM output, in combination with the use of a weather generator, can help to meet this requirement. Further work might focus on analyzing flood peaks in more detail, to gain insight on changes in timing of all sub-basins and how timing influences the ultimate flood peak, with a focus on snowmelt related peak discharges in the Alpine region (see also Stewart (2008)). Work is ongoing to extend the hydrological modeling to 10 000 years, in order to further improve extreme value analysis on peak discharges. Gathering (biascorrected) GCM and RCM output data of different models to create more climate change scenarios should create insight in the probability of future scenarios, and is already the aim of the international project RheinBlick (information is available at http://www.chr-khr.org/en/projects/rheinblick2050). In addition, low-flow periods deserve more attention in climate change impact studies, since drought risk might be of the same order as flood risk even in temperate climate zones. Finally, in the current situation in the Rhine basin, safety levels vary to such an extent that flooding is likely to occur upstream in Germany, while flood stages do not reach dike heights in the Lower Rhine and in the Netherlands. This effect probably influences discharge peaks and the impact of upstream flooding should be investigated in relation to projected changes in flood-peak probabilities due to climate change scenarios. Acknowledgements This research was supported by the project ACER (A7) under the Dutch BSIK Climate Changes Spatial Planning program. We wish to thank Hendrik Buiteveld and Rita Lammersen from the Ministry of Transport, Public Works and Water Management and the German Federal Institute of Hydrology (BfG) for both allowing the use of the HBV and SOBEK models, and providing meteorological (CHR) and discharge measurement data. Marcel Ververs and Simone Patzke from Deltares are kindly acknowledged for their support on the development of the hydrological modeling cascade (GRADE). We also thank Bart van den Hurk (KNMI) and colleagues from the IVM PhD peer group for their constructive comments that helped to improve this Chapter. Finally, we thank Dennis Lettenmaier and two anonymous reviewers for their comments on an earlier version of the Chapter.

Chapter

4

Future flood risk estimates along the river Rhine

Te Linde, A.H., Bubeck, P., Dekkers, J.E.C., De Moel, H. and Aerts, J.C.J.H., 2011. Future flood risk estimates along the river Rhine. Natural Hazards and Earth System Sciences, 11: 459–473, doi: 10.5194/nhess-11-459-2011. Abstract In Europe, water management is moving from flood defense to a risk management approach, which takes both the probability and the potential consequences of flooding into account. It is expected that climate change and socio-economic development will lead to an increase in flood risk in the Rhine basin. To optimize spatial planning and flood management measures, studies are needed that quantify future flood risks and estimate their uncertainties. In this Chapter, we estimated the current and future fluvial flood risk in 2030 for the entire Rhine basin in a scenario study. The change in value at risk is based on two land-use projections derived from a land-use model representing two different socio-economic scenarios. Potential damage was calculated by a damage model, and changes in flood probabilities were derived from two climate scenarios and hydrological modeling. We aggregated the results into seven sections along the Rhine. It was found that the annual expected damage in the Rhine basin may increase by between 54% and 230%, of which the major part (∼ three-quarters) can be accounted for by climate change. The highest current potential damage can

77

78

Chapter 4. Future flood risk estimates

be found in the Netherlands (e 110 billion), compared with the second (e 71 billion) and third (e 58 billion) highest values in two areas in Germany. Results further show that the area with the highest fluvial flood risk is located in the Lower Rhine in NordRhein-Westfalen in Germany, and not in the Netherlands, as is often perceived. This is mainly due to the higher flood protection standards in the Netherlands as compared to Germany.

4.1

Introduction

Over the last couple of decades Europe has witnessed a growth in the scale and frequency of extreme natural disasters. Storms and floods are the most frequent and costly extreme weather events occurring in Europe, representing 69% of the overall natural catastrophic losses. For example, flooding in the Elbe basin in 2002 caused approximately e 8 billion of economic damage in Germany, Austria and the Czech Republic collectively (CEA, 2007). Total damage of the summer floods in 2007 in the UK amount to e 4 billion (Environment Agency, 2007). In 2010, Poland suffered from major floodings, of which the total damages are yet unknown. When focusing on the Rhine basin in North-West Europe, flood events in 1993 and 1995 caused severe damage of e 1.4 billion and e 2.6 billion, respectively (Engel, 1997; Glaser and Stangl, 2003; Brakenridge and Anderson, 2008). The impact of flood events on societies and economies in the Rhine basin is likely to increase further as a result of two complementary trends. First of all, climate change is expected to increase the frequency and magnitude of flood peaks in the Rhine basin (Hooijer et al., 2004; Pinter et al., 2006). Annual maximum peak discharges are expected to increase by 3–19% in 2050 due to climate change (Kwadijk, 1993; Middelkoop et al., 2001; Vellinga et al., 2001). Te Linde et al. (2010) estimate an increase in the occurrence of an extreme 1/1250 per year flood event in the Lower Rhine delta by a factor of three to five in 2050. Secondly, the economic impact of natural catastrophes is increasing due to the growing number of people living in areas with a high flood exposure level, as well as the increased economic activity in these regions (e.g. Bouwer et al., 2007; Pielke Jr. et al., 2008). The International Commission for the Protection of the Rhine (ICPR) estimated an increase in potential damage in flood-prone areas in the Rhine basin of 23% between 1995 and 2005 (ICPR, 2005a). These projected trends have led to an increased interest in a risk-based approach in water management, addressing both the probability and potential consequences of flooding (Merz et al., 2004; B¨chele et al., 2006; De Bruijn and Klijn, 2009; Kreibich u and Thieken, 2009; Wheater and Evans, 2009). Such an approach, for example, is currently stimulated by the EU Flood Directive 2007/60/EC (EU, 2007a), obliging

4.1. Introduction

79

member states to create flood risk maps and basin-wide flood risk management plans (De Moel et al., 2009). Quite a lot of literature exists on how the discharge regime in the Rhine may alter due to climate change (e.g. Kwadijk, 1993; Middelkoop et al., 2001; Menzel et al., 2006; Te Linde et al., 2010). However, in terms of land-use change and flood-damage potential only a few studies exist. The ICPR uses the Rhine Atlas approach to estimate aggregated flood damage for the whole Rhine basin (ICPR, 2001, 2005b). The Rhine Atlas damage evaluation has some flaws, though, for two reasons. Firstly, it has been recognized that the Rhine Atlas yields rather low damage potential values for different land-use classes compared to other studies and probably underestimates potential flood damage (Thieken et al., 2008; De Moel and Aerts, 2010). Secondly, the Rhine Atlas differentiates between only six different land-use classes; it uses a single urban class, whereas differentiation between urban classes for flood damage estimates is essential (Apel et al., 2009). Research, however, on assessing current and future flood risk (addressing both flood probability and potential damage) is still in its early stages and a basin-wide assessment of flood risk is lacking. For the Rhine delta in the Netherlands two studies are available that calculate current and future flood risks (Aerts et al., 2008b; Bouwer et al., 2010). These authors use a method in which the results of flood depths and land-use information are combined within a flood damage model. In this method projected land-use simulations using a land-use model are combined with inundation information to derive potential flood damages using stage-damage curves (Merz et al., 2007). Flood risk (in terms of expected annual damage) is assessed by multiplying the potential damage with the probability associated with the inundation information. Climate change is taken into account by simulating future discharges and probabilities using climate change scenarios as input for hydrological models (e.g. Te Linde et al., 2010). In addition to a current and future perspective, De Moel et al. (2011) also assessed the historical trends in the 20th century for flood damage in the central part of the Netherlands. In order to conduct an assessment for trends in flood risk (in terms of flood probabilities and flood damage) for the Rhine basin we need to address the following two research issues. (1) A land-use model for the Rhine basin does not exist, and hence it is difficult to estimate future land use and potential flood damage. (2) Furthermore, despite existing research focusing on the (future) hydrology of the Rhine (e.g. Kwadijk, 1993; Middelkoop et al., 2001; Bronstert et al., 2002; Shabalova et al., 2003), few estimates exist for changes in future trends of low probability events. For the latter issue, climate impact simulations are required that allow for extreme value analysis techniques (Raff et al., 2009; Te Linde et al., 2010). The goal of this Chapter is, therefore, to estimate current and future flood risk for the entire Rhine basin in a scenario study. For this, we first assessed changes in

80

Chapter 4. Future flood risk estimates

flood probability at various locations along the Rhine using climate scenarios and hydrological models. Second, we developed a land-use simulation model for the Rhine basin to generate future changes in land use. Third, these future land-use maps were used to estimate potential flood damage in flood-prone areas using a damage model. Finally, we multiplied flood probabilities with flood damage to derive current and future flood risk for the Rhine basin. The remainder of this Chapter is organized as follows. Section 4.2 describes the case study area. Section 4.3 provides a description of the data and research method we used. Results are presented in Section 4.4 after which we discuss these results and provide conclusions in Section 4.5.

4.2

Case study area: The Rhine basin

The river Rhine originates in the Alps in Switzerland, forms part of the boundary between France and Germany and continues flowing through Germany before it enters the Netherlands at Lobith (Figure 4.1a). The Rhine is one of the most important industrial transport routes in the world and connects one of the largest sea harbors, the port of Rotterdam, to the inland European markets and its large industrial complexes (Jonkeren, 2009). About 58 million people inhabit the river basin, of which 10.5 million live in flood-prone areas (ICPR, 2001). The average discharge at Lobith in the Lower Rhine is 2200 m3 /s and the maximum observed discharge was 12 600 m3 /s in 1926 (Pinter et al., 2006). Water management has heavily influenced the characteristics of the Rhine. Prior to the 19th century, the Rhine was a multi-channel braided river system upstream of Worms and meandering from that point downstream. However in order to reduce flooding, the Upper Rhine was canalized between 1817 and 1890 (Blackbourn, 2006). Furthermore, to aid shipping, engineers further rectified and canalized the main branch until 1955, causing additional acceleration of flood wave propagation in the Rhine (Lammersen et al., 2002). The basin area is 185 000 km2 and in particular the flood-prone areas in the basin are densely populated (Figure 4.1b). Hence, flood management has predominantly focused on major dike reinforcements along the Rhine over the last 20–30 years. Safety levels vary from 1/200 to 1/500 per year in Germany to 1/1250 and 1/2000 per year in the Netherlands. The design discharge that is associated with a safety level of 1/1250 per year (the discharge used when designing flood defenses) is estimated at 16 000 m3 /s (Ministry of Transport, Public Works and Water Management, 2006b) Figure 4.1a). Due to lower safety levels in Germany, floods may occur at upstream sections in Germany while the Dutch dike system will still prevent huge areas from inundation downstream (Gudden, 2004; Apel et al., 2006).

4.2. Case study area: The Rhine basin

81

a)
G 1/1250 F 1/500 E 1/200 D 1/200 C 1/200 B 1/1000 A 1/200

b)

±

Residential high density Residential low density Commercial Infrastructure Construction Recreation Nature Agriculture Cultivation Pasture Water courses

Netherlands Lobith
!

Köln
!

Belgium
!

Germany Koblenz
!

Mainz

Luxembourg

Meter
0m - 1m 1m - 2.5m

France
!
Kilometers 100

2.5m - 4.5m 5m +

Basel Austria

0

25 50

Switzerland

Figure 4.1: Maps of the Rhine basin: a) (estimated) safety levels and b) land use in the reference situation. Figure 1b) also shows the potential inundated area due to fluvial flooding from the Rhine.

82

Chapter 4. Future flood risk estimates

4.3

Data and method

We followed the steps displayed in Figure 4.2 to estimate expected flood damage per year (risk) for the reference situation and different future scenarios for the year 2030. Economic value of land-use classes determines the potential flood damage in case of a flooding event. Current land-use information was based on the CORINE land cover data (Bossard et al., 2000). Future changes in flood damage was estimated using a land-use model, simulating future land use for two different socio-economic scenarios (see Section 4.3.1). Through combining existing basin-wide flood inundation depth maps (see Section 4.3.2) with land-use information , potential damage were calculated using a damage model (see Section 4.3.3). Flood risk was calculated by multiplying potential flood damage with the accompanying flood probability for different sections along the Rhine. Current flood probabilities were estimated using research by ICPR (2005b) and Silva and Van Velzen (2008) (Figure 4.1a). Changes in flood probabilities were calculated using a hydrological model and two climate change scenarios (Te Linde et al., 2010) (see Section 4.3.4). The flood damage calculations were performed at spatial grids of 100×100 m and aggregated into seven regions along the Rhine (see regions A through G in Figure 4.1a) and the entire basin to calculate expected damage per year. The steps used in this method, as well as the data and future scenarios, are described in detail below.

4.3.1

Current and future land use

Current land use is based on the CORINE land cover map for 2000 (Bossard et al., 2000) that has a resolution of 100×100 m. Future land-use projections from the EUruralis project exist for the whole of the European Union for the year 2030 (Verburg et al., 2008). However, these projections distinguish only a single urban land-use class and have a relatively low resolution of 1×1 km, while it is important to have an accurate representation of urban land use in flood damage simulations (Bouwer et al., 2010). This is illustrated by De Moel and Aerts (2010) who show that urban land use contributes the largest share of flood damage (∼ 80%) and because maximum damages differ substantially between different urban classes in their damage model (from e 0.3 million/hectare for recreational areas to e 9.1 million/hectare for high density residential areas), differentiation within urban land use is desirable for flood damage assessments. To address this issue, we have set up a new and more detailed land-use model (the Land Use Scanner) to downscale land-use projections from the EUruralis project, both spatially and thematically. The Land Use Scanner for the Rhine basin is based on the method described by Hilferink and Rietveld (1999).

4.3. Data and method

83

CORINE land cover map (Bossard et al., 2000)

EUruralis land use change projections (GE / RC) on 1x1 km grid (Verburg et al., 2009)

Land Use Scanner - downscale to 250 x 250 m resolution for Rhine basin - downscale urban land types - integrate relevant spatial information / drivers

Land use maps Scenarios: - 2000 reference - 2030 RC - 2030 GE Inundation map

Damage Scanner

Flood loss maps Scenarios: - 2000 reference - 2030 RC - 2030 GE Flood probabilties Scenarios: - 2000 reference - 2030 RACMO - 2030 Wp Flood risk maps Scenarios: - 2000 reference - 2030 RC - RACMO - 2030 GE - Wp

Figure 4.2: Flowchart of the method used for estimating future flood risk.

84

Chapter 4. Future flood risk estimates

Land Use Scanner The Land Use Scanner simulates future land use and is based on demand-supply interaction of land, whereby different sectors compete for allocation of land within land suitability and policy constraints (Loonen and Koomen, 2008). The model has previously been applied in a number of policy-related research projects in European countries (Wagtendonk et al., 2001; Hartje et al., 2005; Koomen et al., 2005; Dekkers and Koomen, 2007). It was recently applied in studies on the long-term development of flood risk in the Netherlands and the evaluation of the effectiveness of various adaptation strategies (MNP, 2007; Aerts et al., 2008b; Bouwer et al., 2010). The land-use model for the Rhine basin operates on a spatial resolution of 250×250 m grid cells and provides information on 13 different land-use classes, such as infrastructure, nature, agricultural land uses and water, including six different urban functions (Figure 4.1b). Scenarios and land-use claims To be able to simulate future land-use patterns with the Land Use Scanner, the expected increase or decrease of each land-use class (called claims) has to be defined. These claims were derived from the EUruralis project (Verburg et al., 2008; Verburg and Overmars, 2009). In this project land-use projections and their underlying claims have been developed for four socio-economic scenarios, in line with the four scenarios in the Special Report on Emissions Scenarios (SRES) by IPCC (2000). For the present study, two of these projections and their land-use claims were selected: the ‘Global Economy’ (GE) and the ‘Regional Community’ (RC) scenarios which can be regarded as the upper and lower boundaries of possible future urban land-use change. The ‘Global Economy’ scenario reflects a future with high economic and population growth, international economic integration as well as little environmental concern on behalf of governments, resulting in a large increase in urban land-use functions with no restrictions on urban sprawl. The Regional Communities scenario, on the other hand, represents a future with low economic and population growth, a regional focus and strict environmental regulations enforced by governments, resulting in a substantially lower increase in urban areas and restrictions on urban sprawl. We have used the NUTS3 level to derive land-use claims and as a starting point for our downscaling. NUTS3 corresponds to level 3 administrative units under the Nomenclature of Territorial Units for Statistics in Europe for which socio-economic data is available. These are mainly rural districts and cities with more than 100 000 inhabitants. The land-use claims for the two future scenarios were derived by assessing the decrease or increase of each land-use class between the scenario projections and the baseline situation in 2000.

4.3. Data and method

85

Downscaling We spatially scaled down the land-use change projections from the NUTS3 polygons to 250×250 m, which is the required level of detail needed for the Land Use Scanner. Furthermore, the single urban land-use class distinguished in the EUruralis projections was distributed into five urban land-use classes; residential land use, commercial, recreation, infrastructure and construction/mines. Using the CORINE 2000 land cover map the percentage of the five different urban land-use categories of total urban land use was calculated for each NUTS3 region within the study area. Subsequently, the total change in urban land use was assessed by comparing the EUruralis projections for 2030 to the start year, again at the NUTS3 regional level. The resulting change in total urban land use was then distributed over the five urban land-use classes according to the previously established divisions, taking into account the storylines for the two scenarios. On top of differentiating the EUruralis urban land-use class, an extra residential class representing high-density residential areas was defined using the LandScan population data base (Landscan, 2009). This was done because the CORINE 2000 land cover data makes very little distinction between high and low urban density residential areas. Suitability Maps The land-use claims provide information on the scale of future land-use change but give no indication as to where these claims might be realized. This allocation process is carried out by the Land Use Scanner on the basis of suitability maps. These maps give a definition for every location (grid cell) of its attractiveness for the different landuse types available, depending on its current land use, physical properties, operational policies and expected relations to nearby land-use functions (Ritsema van Eck and Koomen, 2008). For example, a location (grid cell) with a steep slope (physical property) that is situated in a nature protection area (operational policy) and far away from existing urban infrastructure (relation to nearby land use) is thus considered as highly unsuitable for the realization of a residential land-use claim. The suitability maps can also be used to further reflect the effect of socio-economic scenarios and thus the land use change simulations by integrating flood-risk specific information. For example, the regional communities scenario assumes a world with a regional focus and strict environmental regulations enforced by governments. To reflect this, the 1/100 per year flood zone, which is mainly embanked river foreland, is given a low suitability value for further urbanization, a policy that has already been adopted in Germany. In contrast, the global economy scenario assumes a world where governments have little environmental concern, resulting in a large increase in urban land-use functions with no restrictions on urban sprawl. We, therefore, simulated land use according to

86

Chapter 4. Future flood risk estimates

Table 4.1: Suitability maps used for the ‘Land Use Scanner’.
Category Physical properties Policy maps Suitability Map Peat bog, marsh, moor Slope Population density Nature 2000 sites Flood retention areas Flood zone (1/100 per year) Flood zone (extreme event) Distance to metropolitan areas Distance to long-distance train stations Distance to passenger railway stations Distance to motorway exits Distance to international airports Distance to road network Distance to major rivers Neighborhood statistics Extent Basin Basin Basin Basin Germany Basin Basin Basin Basin Basin Basin Basin Basin Basin Basin Source CORINE SRTM LandScan DG Environment ICPR / TU Dortmund ICPR Rhine Atlas ICPR Rhine Atlas ESRI TU Dortmund TU Dortmund TU Dortmund ESRI ESRI ESRI Own analysis

Relational maps

this scenario without limitations as far as the 1/100 per year flood zone is concerned. Moreover, the suitability of urban areas close to a river course is increased in the global economy scenario as it is assumed that more people would like to live near the water and are willing to pay for this location. This development has also been observed in the past during periods of economic growth (ICPR, 2005b). An overview of the suitability maps used for the Land Use Scanner for the Rhine basin is given in Table 4.1.

4.3.2

Inundation map

One of the inputs for the flood damage model is a map displaying the water depth of a possible flooding event in the entire Rhine basin. Such a map was developed in 2001 by the International Commission for the Protection of the Rhine (ICPR), known as the Rhine Atlas (ICPR, 2001). This atlas contains a collection of maps that displays the potential flooded area in the Rhine basin at different flooding probabilities (1/10 per year, 1/100 per year, and ‘extreme’, without a probability estimate). We used only the ‘extreme’ situation to indicate the inundated area in the case of flooding, since safety levels along the Rhine are all higher than 1/100 per year (Figure 4.1a). Based on the Rhine Atlas, we cannot predict how the flood extent will change in the future and therefore we have assumed that the inundated areas for the reference situation and in 2030 are the same. For the Netherlands, we have used

4.3. Data and method

87

flood risk maps made available by the Dutch government that are based on multiple inundation model runs (Van den Berg et al., 2010). We have only included inundated areas that are prone to flooding by the river Rhine and not areas that are influenced by storm surges from the sea.

4.3.3

Flood damage

Potential flood damage can be assessed in various ways, ranging from the use of very detailed, object-based data to the use of aggregated asset values per hectare (or square meter) for a certain land-use category (Messner et al., 2007). Given the spatial and temporal scale of the present study, which looks at the development of flood risk on a basin-wide level in the future, we used a simple damage model for land-use categories, the Damage Scanner (Aerts et al., 2008b; Bouwer et al., 2010; Klijn et al., 2007). This model is based on two input parameters: water depth and land use. Potential damage is calculated by the model using so-called damage functions that define for a land-use category the damage that can be expected when a respective inundation level occurs. The model applies damage functions for the 13 land-use classes distinguished by the Land Use Scanner and reflects predominantly direct tangible damage caused by physical contact between economic assets and flood water. Note that direct intangible losses such as loss of life are not reflected by the model. However, the Damage Scanner also implicitly comprises approximately 5% of indirect damages as a surcharge on direct damages. Indirect damages refer to a loss of turnover due to business interruption during a flood event and can make up a substantial share of total flood damages (RebelGroup, 2007).

4.3.4

Climate change scenarios for changes in flood probabilities

Figure 4.1a shows current safety levels for seven regions along the main Rhine branch. In the Netherlands, there is a legal standard for flood defense safety levels. In Germany, dike heights are often legally defined and the related safety levels are estimated and described by ICPR (2005b) and Silva and Van Velzen (2008). The differences in safety levels were used to distinguish the regions for which aggregated flood damage and flood risk can be calculated. The seven regions have different surface areas. The larger the surface area, the larger the aggregated damage and risk will be, since we assume that at the given probabilities the entire region will flood. Nevertheless, we made no corrections in our results for the different surface areas of the seven regions. For an individual region, aggregated damage and risk define the dimensions of the hazard and can be compared to other regions, in contrast to damage or risk per km2 .

88

Chapter 4. Future flood risk estimates

We assumed flooding occurs at probabilities corresponding to the safety levels in the reference situation. Hence, we did not simulate flood damage due to dike failures that may occur at lower probabilities and furthermore assumed that dike heights will not change in the future. The current policy in the Netherlands, however, foresees adaptation of the flood defenses (i.e. dike heightening or lowering of the flood plains) when flood probability increases in order to maintain current safety levels. We used two climate change scenarios (a moderate and an extreme scenario) to estimate future changes in flood probabilities along the main Rhine branch, which were taken from Te Linde et al. (2010). The first climate scenario (referred to from now on as ‘Wp’) represents an extreme climate change scenario, based on Van den Hurk et al. (2006) and describes the most extreme scenario out of four in terms of winter precipitation and resulting floods along the Rhine in 2050 (Te Linde, 2007). This climate scenario corresponds with a 2◦ Celsius increase in global temperature in 2050 with respect to 1990 and changes in atmospheric circulation resulting in drier summers and wetter winters. The second climate scenario (further referred to as ‘RACMO’) displays more moderate climate change effects and follows the output of the RACMO2.1 regional climate model (Lenderink et al., 2003; Meijgaard et al., 2008; Bakker and Van den Hurk, 2009). This scenario corresponds with the IPCC SRES-A1B scenario and projects more spatial variation in meteorological changes than the Wp scenario does. Both climate scenarios are available in time series of 35 years, and were resampled into time series of 1000 years of daily data. These resampled times series were subsequently used to drive the hydrological model HBV (Bergstr¨m, 1976) and to simulate river o discharges and related flood peak probabilities (Te Linde et al., 2010). By comparing current flood probabilities with future flood probabilities, changes in flood-peak probability were derived for the seven regions along the Rhine (see Table 4.2). Te Linde et al. (2010) evaluated changes in flood probabilities between 1990 and 2050. Since the reference year in this study is 2000, and the scenario year 2030, we divided the projected changes in flood probabilities by Te Linde et al. (2010) by two in order to take the shorter timescale into account.

4.4
4.4.1

Simulation results
Discharges and probabilities

Figure 4.3 shows an extreme value plot for annual maximum discharges at Lobith, for the year 1990 and two climate change scenarios for 2050. The results represent 1000-year runs for the reference and each climate change scenario (Wp and RACMO). From the simulation results it can be derived that the discharge corresponding to a

4.4. Simulation results

89

Table 4.2: Climate change scenarios for increased flooding probabilities in 2030. Flooding probabilities (per year) for the reference situation are estimated based on literature. The probability (p) increase is displayed as a factor (1 Estimate, based on (Silva and Van Velzen, 2008) and on the Evaluation of the Action Plan on Floods (ICPR, 2005a)).
Region Alpine A Upper Rhine B Upper Rhine C Middle Rhine D Lower Rhine E Lower Rhine F Delta G Reference1 p 1/200 (0.0050) 1/1000 (0.0010) 1/200 (0.0050) 1/200 (0.0050) 1/200 (0.0050) 1/500 (0.0020) 1/1250 (0.0008) 1/2000 (0.0005) RACMO p 1/139 (0.0072) 1/691 (0.0014) 1/160 (0.0062) 1/159 (0.0063) 1/134 (0.0075) 1/327 (0.0031) 1/673 (0.0015) 1/1080 (0.0015) Wp p 1/64 (0.0157) 1/261 (0.0038) 1/77 (0.0129) 1/80 (0.0125) 1/80 (0.0125) 1/162 (0.0062) 1/437 (0.0023) 1/702 (0.0023) RACMO p incr. 1.4 1.5 1.3 1.3 1.5 1.5 1.9 Wp p incr. 3.1 3.9 2.6 2.5 2.5 3.1 2.9

probability of 1/1250 per year at Lobith increases by 16% for the Wp scenario and by 13% for the RACMO scenario. The discharge currently corresponding to the 1/1250 event (about 16 000 m3 /s) will increase to 1/460 per year for the RACMO scenario and 1/265 per year for the Wp scenario, meaning the probability increases by a factor of 2.7 to 4.7 respectively (Te Linde et al., 2010). Similar projected changes in flood probabilities are available for several locations along the Rhine branch, representing the regions A through G in Figure 4.1a with different safety levels. The projected increases in flood probabilities for 2030 range from a factor of 1.3 to 3.8, depending on region and climate change scenario (Table 4.2).

4.4.2

Land-use change

Table 4.3 shows surface percentages of land-use classes in the flood prone area of the Rhine, according to the CORINE land cover map. Agriculture, cultivation and pasture have the largest combined share of 71% in the reference situation. High and low density residential and commercial areas comprise 17% of the total basin area. The RC scenario for 2030 displays by far the largest increase in nature (110%), whereas residential and commercial areas each increase on average by 19%. In the GE scenario, the residential and commercial areas each increase on average by 44%. Both scenarios project a decrease in agricultural area (∼ −15%). Cultivated area and pasture remain fairly stable in both scenarios (less than 6% change). These trends are also illustrated in Figure 4.4, showing output maps of the land-use simulations. The map shows a clear increase in urbanized areas close to the river in the GE scenario, whereas the increase in nature is the most apparent change in the

90

Chapter 4. Future flood risk estimates

Return period (year) 2 Yearly maximum discharge (m3/s) 1.8 1.6 1.4 1.2 1 0.8 x 10
4

50 100

500 1250

Wp RACMO Reference GEV fits

2

4

6

8

Standardized Gumbel variate

Figure 4.3: Extreme value distributions of annual maximum discharge at Lobith, and Generalized Extreme Value (GEV, Appendix C) fits (lines) for the reference situation, and the RACMO and Wp climate change scenarios for the year 2050 (adapted from Te Linde et al., 2010).

Table 4.3: Surface percentages of different land-use classes in the flood prone area of the Rhine basin, for the reference situation in 2000, and the RC and GE scenarios in 2030. Percentages and Euros in Tables 4.3 through 4.7 are rounded to two significant digits.
land-use class Residential High Density Residential Low Density Commercial Infrastructure Construction / mines Recreation Nature Agriculture Cultivation Pasture Reference Area (%) 3.7 9.0 3.7 1.0 0.69 1.7 9.4 23 10 37 RC Area (%) 4.3 11 4.3 1.1 0.77 2.0 19 19 10 36 GE Area (%) 5.4 14 5.2 1.1 0.77 2.2 11 20 10 40 RC Change (%) 16 23 18 14 12 14 110 −16 2.3 −2.7 GE Change (%) 45 47 42 13 11 28 21 −15 1.7 5.9

4.4. Simulation results

91

Reference 2000
Netherlands

024 Emmerich

±

8 Kilometers

Residential - High Density Residential - Low Density Commercial Infrastructure Construction / Mines Recreation Nature Agriculture Cultivation Pasture Inland Water

Wesel

Oberhausen Germany Duisburg

RC 2030
Netherlands Netherlands

GE 2030

Emmerich

Emmerich

Wesel

Wesel

Oberhausen Germany Duisburg Germany

Oberhausen

Duisburg

Figure 4.4: Land-use maps for the reference year 2000, and for 2030 under the RC and GE socio-economic scenarios. The image is zoomed on the Lower Rhine near the border between Germany and the Netherlands, and shows only land-use types in flood-prone areas. RC scenario. These results obviously correspond to the scenario descriptions that were used in the simulations (see Section 4.3.1).

4.4.3

Flood damage

Table 4.4 and Figure 4.5a display the expected damage aggregated for the seven regions along the Rhine. For the reference year (2000), we estimated the total potential damage for the whole basin to be e 300 billion. This is substantially more than the

92

Chapter 4. Future flood risk estimates

Table 4.4: Expected damage for different regions in 2000 and 2030 (at 2000 prices) (1 The estimate of the ICPR (2005a)).
Region Alpine A Upper Rhine B Upper Rhine C Middle Rhine D Lower Rhine E Lower Rhine F Delta G Total Reference e billion 0.46 21 58 15 71 25 110 300 ICPR 20051 0.0 1.6 11 1.5 22 170
1

200 1995: 160

Scen RC e billion 0.39 21 62 12 80 30 120 320

Scen GE e billion 0.50 26 73 18 90 37 140 380

Scen RC Change (%) −0.18 1.8 5.9 −23 11 18 7.6 7.5

Scen GE Change (%) 8.2 18 20 15 21 34 20 21

ICPR estimate of e 200 billion. The ICPR damage estimates are, however, recognized to be rather low compared to other methods and studies. Several land-use types such as residential and commercial areas or agriculture have substantially lower maximum damage values compared to the damage model applied in our study (for more details see ICPR, 2001; Thieken et al., 2008; De Moel and Aerts, 2010). his can be explained, amongst others, by the observation that the results of the Damage Scanner also comprise a share of, on average, 5% indirect damages, which is not included in the Rhine Atlas estimates. The expected damage gradually increases downstream. The delta in the Netherlands (region G) is the largest and most densely populated region, and has therefore the highest potential damage, both in the reference situation as well as in the future projections of both socio-economic scenarios. Between the two scenarios, the RC scenario yields the lowest increase in potential damage: 7.5% over the entire basin. In most regions potential damage changes little, with the exception of the Lower Rhine region (F) (+18%). In some areas, such as the Middle Rhine, the RC scenario even projects a decrease in potential damage. The GE scenario gives an overall larger increase in potential damage (21%). Moreover, expected damage seems to increase substantially in almost all regions, often by more than 15% and ranging up to 34%. Results of expected damage per land-use class for the entire Rhine basin are presented in Table 4.5. The potential damage of residential and commercial areas in the Rhine basin is e 200 billion, which comprises 63% of the total damage, and is projected to increase to e 260 billion (RC) and e 320 billion (GE) (Table 4.5). Agriculture, cultivation and pasture comprise e 93 billion damage (29% of the total damage), which is projected to decrease to e 61 billion (RC) and e 63 billion (GE).

4.4. Simulation results

93

a)

Delta G
0 25

50

±

b)

Delta G
Kilometers 100 0 25

50

±

Kilometers 100

Lower Rhine F

Lower Rhine F

Lower Rhine E

Lower Rhine E

Middle Rhine D

Middle Rhine D

Upper Rhine C

Upper Rhine C

2000 Potential damage
Expected losses (€ billion)
A 0.46 D 15 B 21 F 25 C 58 E 71 G 109 Alpine A Upper Rhine B

2000 Risk
Annual expected losses (€ million/y)
A 2.3 B 21 F 49 D 77 G 87 C 290 E 350 Alpine A Upper Rhine B

Figure 4.5: Potential damage (a) and flood risk (b), aggregated to seven regions along the Rhine.

Table 4.5: Expected damage for different land-use categories in 2000 and 2030 (at 2000 prices).
Land-use class Residential H D Residential L D Commercial Infrastructure Constr / mines Recreation Nature Agriculture Cultivation Pasture Reference e billion 73 85 42 7.0 2.5 4.7 9.1 31 32 30 RC e billion 86 120 53 6.3 2.9 2.0 14 15 30 16 GE e billion 110 150 66 6.2 2.9 2.3 7.9 16 30 17 Reference (%) 23 27 13 2.2 0.8 1.5 2.9 10 10 10 RC (%) 25 34 16 1.8 0.9 0.6 4.1 4.5 8.8 4.6 GE (%) 27 36 17 1.5 0.7 0.6 2.0 4.0 7.5 4.3 RC Chng (%) 18 39 28 −11 19 −57 56 −49 −5.4 −48 GE Chng (%) 46 72 59 −12 17 −51 −13 −47 −6.1 −43

94

Chapter 4. Future flood risk estimates

Table 4.6: Basin-wide annual expected damage (risk) in e million per year. The factor of change is displayed in brackets. The reference year is 2000 and the scenarios represent 2030.
Socio-economic scenario Reference RC GE 880 950 (7.5%) 1100 (27%) 1300 (43%) 1400 (54%) 1600 (81%) 2300 (160%) 2500 (180%) 2900 (230%)

Climate scenario

Reference RACMO Wp

4.4.4

Flood risk

Figure 4.5b shows estimates of expected annual flood damage in the reference year (2000) for the seven regions along the Rhine. In contrast with potential damage (Figure 4.5a), the highest flood risk estimates are not found in the Dutch Delta (G), but rather in the Lower Rhine (E) in the German state NordRhein-Westfalen and in the Upper Rhine (C). This is the result of the substantially higher flood protection levels in the Delta region G, which obviously determines and lowers the flood-risk estimates to a large extent. This also implies that uncertainties of flood probabilities heavily affect the reliability of (future) flood-risk estimates in this region. For the future risk projections, the RACMO climate scenario is combined with the RC socio-economic scenario and Wp with the GE scenario. The combination RACMO-RC can be considered as the lower estimate and Wp-GE as the upper estimate in the risk simulations. Basin-wide results are displayed in Table 4.6. The flood risk estimates of the scenarios show a large variation. In the reference situation, we estimate the basin-wide expected annual flood damage to be e 880 million on average per year. The RACMO-RC scenario projects the risk to increase to e 1400 million per year, an increase of 54%. The Wp-GE scenario projects a much larger increase in flood risk, tripling it to e 2900 million per year (an increase of 230%). The contribution made by climate change is considerably larger than socio-economic change in both scenario combinations. Due to climate change, basin-wide flood risk increases by 43–160%, whereas land-use change results in an increase of 6.5–27% (Table 4.6). In order to illustrate the relative increase of annual expected damage due to each of the driving forces, we displayed the basin-wide flood risk scenarios in a bar chart (Figure 4.6). The bar chart displays the contributions to change in annual expected damage, from a) climate change only, b) socio-economic change only, and c) the combination of both impacts. Climate change accounts for ∼ three-quarters (6/8) of the increase, whereas socio-economic change only results in ∼ 1/8 of the total increase in annual expected damage. The combination of impacts adds the remaining ∼ 1/8 to both scenarios respectively.

4.4. Simulation results

95

3500 Expected loss (€ million/year) Socio-economic + climate 3000 2500 2000 1500 1000 500 0 2000 RACMO - RC Wp - GE Climate Socio-economic Reference risk

Figure 4.6: Basin-wide annual expected flood damage (risk) for 2030, compared to the reference situation. Table 4.7: Annual expected damage (risk) in e million per year for different regions in 2000 and 2030 (at 2000 prices).
Region Alpine A Upper Rhine B Upper Rhine C Middle Rhine D Lower Rhine E Lower Rhine F Delta G Total Reference situation p e bln e mln/yr 0.0050 0.46 2.3 0.0010 21 21 0.0050 58 290 0.0050 15 77 0.0050 71 350 0.0020 25 49 0.00080 110 87 0.00050 300 880 RACMO and RC p e bln e mln/yr 0.0072 0.39 2.8 0.0014 21 31 0.0062 62 390 0.0063 12 78 0.0075 80 590 0.0031 30 91 0.0015 120 180 0.00093 320 1400 p 0.0157 0.0038 0.0129 0.0125 0.0125 0.0062 0.0023 0.0014 Wp and GE e bln e mln/yr 0.50 7.9 26 98 73 940 18 220 90 1100 37 230 140 380 310 2900

As this is the first assessment of basin-wide future flood risk, it is interesting to compare different sections along the Rhine and to evaluate if differences with regard to the drivers of future flood risk can be observed. To assess differences between regions along the Rhine, Table 4.7 shows annual expected damage for seven regions. Bar charts similar to Figure 4.6 are shown in Figure 4.7, but now disaggregated to seven regions along the Rhine. The bar charts show large variations in base risk and flood risk projections between regions, and, like the basin-wide projections, the dominant contribution of climate change to increased flood risk. The Alpine area (A) and the Upper Rhine (B) display hardly any annual expected damage at the vertical scale they are presented (less than e 110 million per year). Just as we have seen in Figure 4.5b for the reference flood risk, projections for annual expected damage in 2030 are the highest in the Upper Rhine (C) (up to e 940 million per year in the Wp-GE scenario, an increase of 220%) and the Lower Rhine (E) (up to e 1100 million per year, an increase of 210%). The Middle Rhine (D), the Lower

96

Chapter 4. Future flood risk estimates

Delta G
Expected damage (€ million /year) 1000 800 600 400 200 0 2000 RACMO-RC Wp-GE

±
0 25 50
Lower Rhine E
2000 RACMO-RC Wp-GE

Kilometers 100

Lower Rhine F
Expected damage (€ million /year) 1000 800 600 400 200 0 2000 RACMO-RC Wp-GE 0 Expected damage (€ million /year) 1000 800 600 400 200

Middle Rhine D
Expected damage (€ million /year) 1000 800 600 400 200 0 2000 RACMO-RC Wp-GE

Upper Rhine C
Expected damage (€ million /year) 1000 800 600 400 200 0

Upper Rhine B
Expected damage (€ million /year) 1000 800 600 400 200 0 2000 RACMO-RC Wp-GE

2000

RACMO-RC

Wp-GE

Socio-economic + climate Climate Socio-economic Reference
Expected damage (€ million /year) 1000 800 600 400 200 0 2000

Alpine A

RACMO-RC

Wp-GE

Figure 4.7: Annual expected flood damage, for the reference situation and projections for 2030, aggregated into seven regions along the Rhine.

4.5. Discussion and conclusions

97

Rhine up to the Netherlands (F), and the Dutch Delta (G) show risk projections of between e 220 and e 310 million per year in the Wp-GE scenario. For the different regions, the relative contribution of climate change to increased flood risk varies between 5/8 and 7/8, whereas socio-economic change results in zero to 2/8 of the total increase in annual expected damage.

4.5

Discussion and conclusions

The aim of this Chapter was to estimate future flood risk in 2030 for the entire Rhine basin. We took the year 2000 as a reference and used scenarios in a model simulation to assess changes in flood probability due to climate change, and to assess changes in potential damage due to land-use change. The combined simulations provided different projections for future flood risk. It was found that, in absolute terms, potential flood damage is highest in the Dutch Delta region (G), namely e 110 billion, compared to e 71 billion of the second highest value in the Lower Rhine region (E). Flood risk (damage × probability) is, on the other hand, much higher in other regions, most notably in the Lower Rhine region E (e 350 million per year) and the Upper Rhine C (e 290 million per year), whereas the Dutch Delta region (G) only reaches e 87 million per year. Our research further projected that flood risk in the Rhine basin will not be stationary and might considerably increase over a period of several decades. Expected annual damage in the entire Rhine basin may increase by between 54% and 230%, due to socio-economic and climate change. The results display large variations in current risk and flood-risk projections between regions along the Rhine. The increase in flood risk can mainly be attributed to increasing probabilities of flood peaks due to climate change (43–160%, which is ∼ 6/8 of the total risk increase), whereas socio-economic change accounts for 7.5–27% increase, which is ∼ 1/8 of the total risk increase. This is in contrast with the findings of Bouwer et al. (2010), who found, for a Dutch polder, that the effects of socio-economic change and climate change are similar in magnitude (climate change: 46–201% increase; socio-economic change: 35–172% increase, which resulted in an estimated total increase of between 96 and 719%). However, they used 2040 as scenario year, while we addressed 2030. Also, Bouwer et al. (2010) included projections for increasing capital value in their socio-economic scenarios, in addition to projections for land-use change. This accounts for the major part of their estimate of the contribution from socio-economic change to total flood risk. When wealth increase is not included in Bouwer et al. (2010), the relative change in flood risk is much more similar (socio-economic change inflicts an increase of 3–44%). We omitted wealth increase projections for the Rhine basin due to lack of reliable future projections for the entire basin.

98

Chapter 4. Future flood risk estimates

Our method provides a more comprehensive assessment of basin-wide flood risk in the Rhine than was previously possible as existing studies either assessed flood risk in the Netherlands or in upstream areas in Germany (Apel et al., 2004; Klijn et al., 2007; Aerts et al., 2008b; Bouwer et al., 2010). Furthermore, our method enables basinwide scenario projections for future land use and potential damage, by integrating a land-use model with a damage model at a high spatial resolution. We have shown that expected annual damage depends heavily on estimated floodprobabilities. Further work might focus on acquiring actual safety levels along the Rhine in more detail, by analyzing dike heights and water levels. In reality, there are no jumps in dike height or thus in safety levels along the Rhine between countries or Bundesl¨nder, as we assumed here, but instead the shift is gradual. In addition, due a to dike failure processes such as piping, the actual flood-probability might be much higher than the probabilities of flood events dikes are designed to cope with (Ministry of Transport, Public Works and Water Management, 2006a). On the other hand, due to over dimensioning of dikes, flood probabilities can also be much lower than currently perceived. Understanding this requires more research, which is ongoing in detail in the Netherlands (Ministry of Transport, Public Works and Water Management, 2006a), but, to our knowledge, not on a large scale in Germany and France. Flood damage estimates contain uncertainties related to the choice of the damage model and the simulation of inundation depth. The uncertainty in absolute damage estimates and increases can be considerable when applying damage models (Apel et al., 2009; Merz and Thieken, 2009; De Moel and Aerts, 2010). However, the relative damage increase (increase as a percentage compared to the reference) is much more robust De Moel and Aerts (2010). For this reason, the absolute values of flood risk increase presented in this Chapter should also be interpreted with care. In our approach, we assumed for the Netherlands that all areas (‘dike rings’) will inundate during a flooding event, while they might only partly flood in reality. Therefore, both basin-wide potential damage, as well as expected annual damage, do not provide information on the damage of a single event. For the part of the Rhine basin upstream of the Netherlands, we used inundation maps from the Rhine Atlas (ICPR, 2001) that are to date the best available. The Rhine Atlas assumes flood prone areas to inundate completely. However, several 2D hydrodynamic inundation simulations for the Lower Rhine by Lammersen (2004) showed that the flood-prone areas do not always entirely inundate, depending on breach locations and flood wave characteristics. We therefore recommend more inundation calculations upstream of the Netherlands which are currently only incidentally available, in order to aid further flood risk assessments. Finally, the implementation of flood defense measures, such as retention basins and dike heightening, might prevent the increase in flood probability due to climate change, and thus the contribution of climate change to flood risk. This requires a thorough analysis of the effectiveness of flood management measures under different

4.5. Discussion and conclusions

99

climate change scenarios. Spatial planning policies and damage mitigation measures and risk transfer mechanisms, such as flood proofing of buildings and insurance, might further reduce flood risk. Such flood risk decisions may have implications for several decades. Therefore, flood risk management needs to deal with expected climate and socio-economic changes (Merz et al., 2010). Acknowledgements We would like to thank Christian Lindner and the Institut f¨r Raumplanung Uniu versit¨t Dortmund for providing spatial data for the land-use model. The comments a made by our colleague Peter Verburg are also gratefully acknowledged. Finally, we thank Walter Pfl¨gner and an anonymous reviewer for their valuable comments. This u research project was carried out in the framework of the Dutch National Research Programme Knowledge for Climate (www.knowledgeforclimate.org), which is co-financed by the Ministry of Housing, Spatial Planning and the Environment (VROM). Furthermore, the project was financed by Deltares and the Dutch Ministry of Transport, Public Works and Water Management (V&W).

100

Chapter 4. Future flood risk estimates

Chapter

5

Effectiveness of flood management measures on peak discharges in the Rhine basin under climate change

Te Linde, A.H., Aerts, J.C.J.H. and Kwadijk, J.C.J., 2010. Effectiveness of flood management strategies on peak discharges in the Rhine basin. Journal of Flood Risk Management, 3: 248–269, doi: 10.1111/j.1753-318X.2010.01076.x. Abstract Climate change increases flood probabilities in the Rhine river basin, which complicates long-term flood management planning. This Chapter explores a method to evaluate the effectiveness of flood management measures for the river Rhine assuming a relatively extreme climate change scenario for the year 2050. Considered are planned measures described in the Rhine Action Plan on Floods and several additional measures, which include the restoration of abandoned meanders, a bypass around Cologne, the implementation of additional retention polders and land-use change to forest. The method includes resampling of meteorological data and a hydrological model to simulate long discharge series (10 000 years), and can be considered as a process-based approach to estimate peak discharges of low-probability flood events. It is found that upstream flooding in Germany has a profound decreasing effect on the simulated peak water levels and discharges along the main Rhine branch and downstream in the Netherlands. Currently implemented and proposed measures in the Action Plan 101

102

Chapter 5. Effectiveness of flood management measures

on Floods, as well as most additional measures, seem inadequate to cope with the increased flood probabilities that are expected in the future climate change scenario.

5.1

Introduction

Over the last decades, the number of fatalities and economic damage caused by river floods worldwide has increased considerably (e.g. Kron, 2009; Munich Re AG, 2008) and it is expected that flood risks will continue to increase due to climate change and the growth of economic wealth (Milly et al., 2005; Kundzewicz et al., 2007). A similar trend can be observed for the river Rhine in North-West Europe, where it is expected that climate change will have major implications for its discharge regime (Kwadijk, 1993; Middelkoop et al., 2001). Studies show that mean winter discharges are expected to increase by 5–30% and mean summer discharge to decrease by 0–45% by 2050 (compared to the current climate), using a range of climate change scenarios and hydrological modeling methods (Buishand and Lenderink, 2004; Hundecha and B´rdossy, 2005; Fujihara et al., 2008). As a consequence, the 1/1250 per year flood a event for which dikes are designed in the lower parts of Rhine is estimated to increase from 16 000 m3 /s at present to between 16 500 and 19 500 m3 /s in 2050 (Kwadijk and Middelkoop, 1994; Te Linde et al., 2010). Various flood management measures in the Rhine basin have already been developed according to the Action Plan on Floods (APF) that was initiated in the 1990s (ICPR, 2005a). Implemented and planned measures include dike relocation, the allocation of retention basins and land-use change to store water in head watersheds. The APF is scheduled to be completely implemented by 2020. However, an evaluation of the APF in 2005 revealed that the targets for water level and risk reduction set out in the plan will not be met given current climate conditions (ICPR, 2005a). Moreover, the plan does not address the impact of climate change on peak discharges and questions exist as to whether the plan is effective in the long term, especially when focusing on managing extreme flood events. Two methodological challenges exist, however, to evaluate the effectiveness of flood management measures targeted at managing extreme flood events. The first difficulties relate to the high safety standards in the Rhine basin (varying from 1/200 in Germany to 1/1250 per year in the Netherlands). These flood peaks have not been observed to date and current research extrapolates historical data to derive low-probability flood peaks (Lammersen et al., 2002; ICPR, 2005a; Bronstert et al., 2007). However, only a relatively short period of measured discharge data exists for the Rhine (∼ 110 years) and extrapolation of these data may introduce large uncertainties (Klemeˇ, 2000a,b; s Shaw, 2002). Also, statistical extrapolation assumes stationarity of the observed data record. However, in the last 110 years, both meteorological conditions and the river

5.2. Rhine basin

103

basin have changed, and the principle of stationarity does not hold, which probably adds to the uncertainty associated with the statistical extrapolation (Milly et al., 2008). Hence, recent research suggests the use of resampling methods to create long time series (> 1000 years) of meteorological data (both current data and future climate scenarios) and use these data as input for hydrological models to create long discharge time series (Leander, 2009; Te Linde et al., 2010). In this way, meteorological and hydrological processes are simulated for extreme flood events and statistical extrapolation can be avoided. A second challenge relates to the effect of upstream flooding in the Rhine basin. It appears that existing flood management evaluation studies for the Rhine did not incorporate the effect of upstream flooding, while it is known that upstream flooding does occur at extreme peak events and has a substantial reducing effect on peak discharges downstream in the Rhine delta (Te Linde et al., 2010). This Chapter will explore a method to evaluate the effectiveness of flood management measures for different locations along the Rhine, assuming a climate change projection for 2050. To overcome the two methodological challenges mentioned above, our approach includes the resampling of meteorological data and a hydrological model to simulate long discharge series, including the effect of upstream flooding. Furthermore, using the long time series of possible discharges, an ensemble of different flood waves that belong to the same return period is selected, for different locations along the Rhine. A hydrodynamic model will then be used to evaluate measures that are proposed in the APF, as well as additional flood management measures, on their ability to reduce peak water levels and the probability of flooding. The remainder of the Chapter is organized as follows. Section 5.2 describes the Rhine basin and briefly reviews its long history of flooding and flood management practice. In Section 5.3 the method and models are explained. Section 5.4 summarizes the results of the evaluation of flood management measures. Finally, Section 5.5 contains the discussion and conclusions and the outlook for further research.

5.2
5.2.1

Rhine basin
General description

The Rhine is a cross-boundary river located in North-West Europe and has a length of ca. 1320 km. It originates in the Swiss Alps and flows through parts of Germany, France, and Luxembourg, before it enters the Netherlands at Lobith (Figure 5.1). The Rhine basin comprises an area of ca. 185 000 km2 . Approximately 50% of the Rhine

104

Chapter 5. Effectiveness of flood management measures

75

Netherlands
Lobith

§

Bypass Cologne

Kilometers

!

Lower Rhine ! Belgium
Cologne

! ! ! Leverkusen ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !! ! ! ! ! ! ! ! ! ! ! ! ! ! !! ! !! !! ! ! Cologne ! ! ! ! ! ! ! ! ! ! ! ! !! ! !! !! !! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! ! !!!!!! !!!! !!! ! !! !! ! !!! !!! !! ! !

Restored abandoned meanders

Mainz ! !! Oppenheim !! ! ! ! ! ! ! ! ! ! ! ! !! ! !! ! ! !! !!! ! !! ! ! ! ! ! !! ! !! ! ! ! ! ! !! !! !!! ! !! ! !! ! ! Stockstadt am Rhein !! ! !! ! ! ! ! ! ! ! ! ! !! ! !! !!! ! !!! ! ! Gimbelsheim! ! !! ! ! !! ! ! ! !! ! !

Germany Lahn

Bonn !

Andernach

! !

Koblenz
Worms

Kaub !

!

Mainz

Main

Luxembourg

Middle Rhine Worms !

Lampertheim !!!!! ! !! ! ! !! ! ! ! ! ! !! ! !! ! ! ! !! ! ! ! ! ! !!! !! !!!!! ! ! !!!! ! ! ! ! ! ! ! ! Bobenheim! ! !! ! !! ! !! !! ! ! ! ! ! ! ! Mannheim

Mosel ! Maxau Neckar

Upper Rhine

Basel

France

! Austria

Switzerland

!!! !!! ! !!!! ! ! ! !!!! !! !!! ! ! !! ! ! ! ! Waldsee ! ! ! ! !! ! !! !! ! !! !! ! ! !! !Ketsch ! !! ! ! ! !!! ! !! ! !! ! !!! ! !!!! !!!!! Otterstad! !! ! !! ! !!! ! !! ! ! !! ! ! !!! Speyer !!!!!! ! ! Hockenheim !!! ! ! !! ! !! ! !! ! !!! ! !!!!!! ! ! !!! ! ! !! !! !! ! ! ! ! ! !! Roemerberg!! ! !! !! !! !! !! !! Lingenfeld! !!! ! !! ! ! ! ! ! !!! !!!! ! ! !! ! ! !!!! !!! ! !! Philipsburg !! ! ! ! ! ! ! !! !! !! ! ! ! !!! !! ! ! !! !!!!!! !! !!!! ! ! Hoerdt!!!! !!! ! !! ! !! ! !! ! Linkenheim !! !!!! !! ! ! ! !! ! ! ! ! !! ! !! ! !! ! !! ! !! ! !! !! ! !! !! ! ! !!!!!!! !!!!!! !! ! ! !! !! !! ! Neupotz!!!!!!! ! ! ! ! !! ! !! ! !!!!! !! !! ! ! ! ! ! ! ! ! !! !! !!!! ! ! Karlsruhe Maxau

Figure 5.1: Rhine basin. The flood prone areas of the Rhine are displayed in light blue and were derived from the Rhine Atlas (ICPR, 2001). The measures ‘Cologne bypass’ and ‘Restored abandoned meanders Upper Rhine’ are enlarged. basin is used for agriculture, 33% is forested, 11% is built-up, and the remaining 6% is surface water (Disse and Engel, 2001; Middelkoop et al., 2001). It connects one of the worlds largest sea harbors, the Port of Rotterdam, to the inland European markets and their large industrial complexes (Jonkeren, 2009). Approximately 58 million people inhabit the river basin and 10.5 million of these live in flood prone areas (ICPR, 2005c).

5.3. Methods

105

5.2.2

Flood management in the Rhine basin

Extreme flood events in the Rhine basin downstream of Maxau mainly occur during the winter and early spring (Beersma et al., 2008). Evaporation rates are low and soils are often saturated and sometimes frozen in winter, which can lead to increased runoff (Disse and Engel, 2001). Two more recent extreme flood events in the Lower Rhine and The Netherlands in 1993 and 1995 exemplified the vulnerability of the river basin to flood events. Human activity has influenced the channel characteristics of the Rhine since the Roman era (Lammersen et al., 2002; Blackbourn, 2006). Prior to the 19th century, the Rhine was a multi-channel braided river system upstream of Maxau and meandering from that point downwards. In order to force an incision of the main Rhine branch with the aim of reducing flooding, the Upper Rhine was straightened between 1817 and 1890 (Blackbourn, 2006). Furthermore, to aid shipping, engineers further straightened and canalized the main branch up until 1955 and constructed weirs and dikes between 1955 and 1977. These activities caused an acceleration in flood wave propagation in the Rhine (Lammersen et al., 2002) and hence increased flood risk downstream. This effect is illustrated in Figure 5.2, which displays the form of two flood waves originating from comparable rainfall volumes. One is before and the other is after the canalization of the main Rhine branch in the Upper Rhine. However, differences in the spatial distribution of the rainfall volumes might also add to the alteration of the discharge wave, since in the 1955 event the Neckar basin received a larger fraction of the precipitation than in 1882. Moreover, land-use change and urbanization directly along the main Rhine branch significantly contributes to increased flood risk, since urbanization in flood prone areas increase the potential economic losses due to floods (Hooijer et al., 2004). These trends have led to major dike reinforcements along the Lower Rhine over de the last 20 years. Safety levels vary from 1/200 to 1/500 per year in Germany, while in the Netherlands the 1/1250 per year flood peak is the basis for the design discharge of 16 000 m3 /s (the maximum discharges for which flood protection measured are designed) (Silva, 2003). Due to lower safety levels in Germany, flooding may be occurring at upstream parts in Germany while the Dutch dike system is still protecting huge areas from inundation (Gudden, 2004; Apel et al., 2006).

5.3

Methods

Let us assume that T equals 1/p, where p represents probability and T the return period. The most correct way to describe a low-probability peak event is to denote the probability of occurrence per year (i.e. 1/1250 (p = 0.0008) per year), rather than

106

Chapter 5. Effectiveness of flood management measures

Discharge (m3 /s)

8000

1955
7000

6000 5000

1882/1883
4000

3000

2000

1000

0 0 2 4 6 8 10 12

Days

Figure 5.2: Discharge wave at Worms before and after canalization of the Upper Rhine. Both waves result from heavy rains in the southern part of the basin and have comparable volumes. Adapted from Silva et al. (2001). to claim an event will occur once every x years (i.e. a return period of 1250 years). However, in frequency analysis of flood-peak probabilities it is common to discuss return periods, in order to prevent the use of fractions and very small numbers. In this Chapter, we will therefore use return periods when discussing low-probability peak events. The different steps of the research approach in this Chapter are displayed in Figure 5.3. Steps 1, 2 and 3 focus on simulating long discharge series using resampled meteorological data and the hydrological model HBV (Section 5.3.1). Due to computation time limits, it was not feasible to test the effectiveness of measures on all the annual maximums of the long discharge series, or on a large number of peaks above certain thresholds. Therefore, within steps 4 and 5, flood waves were selected for use in the evaluation of measures. The selection takes into account the fact that no unique flood wave exists that belongs to a specific return period. Instead, many different flood waves (in terms of duration and shape) exist that belong to, for example, the 1/1250 per year event. Hence, a specific measure that is evaluated using a single 1/1250 per year flood wave may perform differently when another 1/1250 per year flood wave is used (Lammersen, 2004; Te Linde et al., 2008b). Also, a 1/1250 per year flood wave at location A is not a 1/1250 per year flood wave at location B, due to differences in river geometry and inflow from side branches. Therefore, we used an ensemble of flood waves in order to create representative flood waves at different return periods and locations. We selected four flood waves with different

5.3. Methods

107

Resampled time series and climate scenarios (1) 10,000 yrs daily P and T Control climate (2) Run HBV 10,000 yrs daily P and T Wp scenario 2050

(3) 10,000 yrs daily Q at 4 locations: Lobith, Andernach, Kaub and Maxau

(6)
T = 1250 yrs

Measures

(4)

T = 500 yrs 4000

Wave selection

T = 200 yrs Selection process ood waves 0 at 4 locations and 4 return T = 2000 800010 yrs periods:

T = 10 T = 200 T = 500 T = 1250

Q (m3 /s)

4000 0

0

5

10 15 20 25 30 35 T (days)

1) APF2002 2) AFP2020 3) Land use change to forest 4) Extra retention polders 5) Bypass Cologne 6) Increased friction by reforestation of the ood plains 7) Restored abandoned meanders Upper Rhine 8) Increased dike height

(5) 16 ood waves

(7) Run SOBEK Model runs and analysis (8) Evaluation of results

Figure 5.3: Flowchart describing all steps of the method. APF2002 are existing measures implemented in the framework of the Action Plan on Floods (ICPR, 2005a). APF2020 are planned measures for 2020.

108

Chapter 5. Effectiveness of flood management measures

return periods, at four locations, resulting in sixteen flood waves that can be used for the evaluation of flood measures. Finally, steps 6, 7 and 8 involved the evaluation of seven flood management measures in terms of their effect on peak water levels and discharges using the selected flood waves and the models SOBEK and HBV (Section 5.3.1). The simulations included the effects of potential floods in the Upper and Lower Rhine. Results were analyzed at the gauging stations at Lobith, Cologne, Andernach, Kaub, Worms and Maxau.

5.3.1

Hydrological modeling

The semi-distributed conceptual HBV model (Bergstr¨m, 1976; Lindstr¨m et al., o o 1997) was developed for the Rhine in 1999 and since then recalibrated several times for the period 1961–1995 (Eberle et al., 2005). The model performs well for the Rhine basin (e.g. Nash&Sutcliffe=0.85; r2 =0.97, for daily discharge in 1993, Te Linde et al. (2008a)), but this paper does not further consider hydrological modeling uncertainty. The Rhine basin is represented by 134 sub-basins in HBV, and the model simulates snow accumulation, snowmelt, actual evaporation, soil moisture storage, groundwater depth and runoff. The model requires daily values of precipitation, temperature and potential evaporation as input. It uses different routines in which snowmelt is computed by a day-degree relation, and groundwater recharge and actual evaporation are functions of the water storage in a soil box. In this study, HBV is used to simulate daily discharges and to simulate the effect of the measure land-use change on discharges. SOBEK is an integrated numerical modeling package that is based on the 1D St. Venant equations (Chow, 1959) (Appendix B) and runs at an hourly time step (Delft Hydraulics, 2005). SOBEK -contrary to HBV- is capable of simulating flood wave propagation, backwater effects, and damping in low gradient river stretches where floodplain inundation plays an important role. In this study, SOBEK is used for simulating the effect of upstream flooding and for the implementation of structural measures, such as dike heightening, dike relocation, weirs and retention polders.

5.3.2

Simulating long discharge series (Steps 1, 2 and 3)

We used a weather generator to create 10 000 years of daily precipitation and temperature data (Figure 5.3, Step 1) for 134 sub-basins of the Rhine basin. This so-called nearest-neighbour resampling technique was developed by Buishand and Brandsma (2001) and uses a historical meteorological data set that contains precipitation and temperature for the period 1961–1995 (Sprokkereef, 2001), which is further referred to as the control climate period. The method produces resampled time series of

5.3. Methods

109

10 000 years that have the same statistical properties as the original input data and has been thoroughly described by Beersma et al. (2001), Leander and Buishand (2007) and Te Linde et al. (2010). We also applied the resampling technique on 35 years of meteorological data representing the year 2050, in order to simulate 10 000 years of climate change data. We applied the W-plus scenario from Van den Hurk et al. (2006) on the Rhine region following Te Linde et al. (2010). W-plus is an extreme climate scenario, based on combined global climate model (GCM) and regional climate model (RCM) outputs, with an annual mean temperature increase of 2.5◦ C, an increase in mean monthly winter rainfall of 14%, and a decrease in mean monthly summer rainfall of −19% (Van den Hurk et al., 2006). The W-plus scenario of projected climate in 2050 was obtained by the delta change approach. In this approach, the outputs from RCMs describing both the reference situation and the projected climate in 2050 are compared, in order to derive average changes of climate parameters. These average changes of precipitation and temperature are used to perturb observed meteorological series. These data were used as input for the hydrological model HBV to simulate daily discharges series of 10 000 years, for both the control climate and changed climate in 2050 (Figure 5.3, Step 2). For more information on the application of the delta change approach on the resampled time series, see Te Linde et al. (2010) where the authors compared climate projections obtained by the delta change approach to direct RCM output. They concluded that bias-corrected direct RCM output is to be preferred over the delta change approach because it provides insight in geographical differences and can simulate change in the number of precipitation days. However, they also observed that the delta approach is more transparent and more robust than using bias-corrected RCM output. Therefore, the delta approach has been used in several studies in the Rhine basin (Beersma et al., 2008), and we chose to do so as well in this paper to make our results comparable.

5.3.3

Selection of 16 flood waves (Steps 4 and 5)

From the 10 000 year series, we selected flood waves at four locations (Maxau, Kaub, Andernach and Lobith) Figure 5.1 and at four return periods (T =10, 200, 500 and 1250), resulting in 16 flood waves. These return periods are relevant for flood management policies in the Rhine basin. A flood wave at location X culminates from unique meteorological conditions and discharge contributions from the sub-basins, and we define those conditions as a unique flood event for location X. The selection was done both for the control climate and changed climate in 2050. To obtain results for the additional locations Worms and Cologne we used flood events that were set for relatively nearby locations. These locations showed comparable peak flow characteristics when we analyzed discharge data at multiple locations. The flood event

110

Chapter 5. Effectiveness of flood management measures

conditions at Maxau were also applied to Worms, and the flood events at Andernach were also applied to Cologne. The selection process at each location consists of the following. First, we constructed probability plots of 10 000 yearly maxima that were extracted from the simulated daily discharges by HBV. Second, we fitted extreme value distributions through these points, and from these relations we derived the peak discharge at each return period of interest. Third, at each return period, we extracted an ensemble of five flood waves from the daily discharge series of 10 000 years that reach peak discharges closest to the derived peak discharge by extreme value analysis. These flood waves differ in shape and duration, but their peak discharges are considerably similar and thus belong to the same return period (the grey lines in the four Q − t plots in Figure 5.3, Step 4). Finally, we created one representative flood wave per location and return period as being the mean of the ensemble (the thick black lines in the Q − t plots in Figure 5.3, Step 4). More details on the selection process are available in Te Linde (2009).

5.3.4

Hydraulic modeling and description of measures (Steps 6 and 7)

We used the 16 selected flood waves as boundary conditions for the hydrodynamic model SOBEK (Figure 5.3, Step 7). The SOBEK schematization contains the main Rhine branch downstream from Maxau, including the branches in the Netherlands. The main tributaries (i.e. Mosel, Main, Neckar) are also schematized for several kilometers upstream of their mouth to the Rhine. The model contains geometry of the cross sections at every 500 m, includes retention polders as they currently exist in the Rhine, and is calibrated by tuning bed friction values (Van der Veen, 2007). Upstream flooding in Germany is schematized as large retention polders with regulated inlet and outlet structures (see for details: Van der Veen et al., 2004; Te Linde and Aerts, 2008). The total potential volume available for flooding is 3892 Mio m3 (Eberle et al., 2004). In addition to implementing measures from the APF (nos. 1 and 2 in step 6, Figure 5.3), we developed six flood management measures (nos. 3 to 8 in step 6, Figure 5.3). The measure land-use change to forest was simulated using HBV and the remaining measures were schematized in SOBEK. SOBEK was also used for simulating upstream flooding. Measure no. 8 simulates the effect of increased dike height to such an extent that upstream flooding cannot occur. See Te Linde (2009) for a more detailed description of the measures.

5.3. Methods

111

Existing APF measures (APF2002) and planned APF measures for 2020 (APF2020) The APF measures are listed in Table 5.1; they can be divided into measures realized in 2002 (APF2002) and planned for 2020 (APF2020). The majority of the measures are controlled flood retention polders, while at some locations, dikes are relocated. The total retention volume in the reference situation in 2002 is 121×106 m3 and in 2020 it is planned to have increased this to 294×106 m3 . APF2002 is the SOBEK model schematization that contains all flood protection measures of the APF that were implemented in 2002 (Table 5.1), and is used as the reference situation in this study. All measures under consideration are implemented upstream of Lobith (see also: Lammersen, 2004; ICPR, 2005a; Bronstert et al., 2007). Additional retention polders We implemented several additional retention polders to the APF measures to temporarily store parts of the peak discharge volume. Their location and size are displayed in Table 5.1, and are based on Raadgever et al. (2008). The total additional volume is 140 Mio m3 . The operational rules for these retention polders to start inundating at a defined water level or discharge were derived from surrounding retention polders as schematized in APF2020. Land-use change to forest Reforestation is often perceived as an efficient measure to reduce flooding. In theory, higher interception, evaporation and infiltration rates result in reduced runoff volumes (Hundecha and B´rdossy, 2004). The effectiveness, however, of reforestation seems to a depend on scale and reduces with increasing basin size (FAO, 2005; Bronstert et al., 2007) and the discussion continues on the links between deforestation, reforestation and floods (e.g. FAO, 2005; Bruijnzeel, 2008). HBV distinguishes four land-use classes (forest, non-forest, lakes and glaciers) and these have different interception values, potential evaporation rates and soil structure. The more diversified land-use classes in the HBV schematization of the Rhine basin, such as arable land and built-up areas (Table 5.2), are represented by different factors for infiltration rates and potential evaporation (Eberle et al., 2000). In the reforestation measure, we replaced all non-forest classes in HBV with mixed forest, which results in a 96.2% cover of forest in the whole basin, 2.4% rock and glacier and 1.4% lake.

112

Chapter 5. Effectiveness of flood management measures

Table 5.1: Measures along the Rhine where RP is retention polder and DR is dike relocation. Displayed are the location and volume of implemented measures in 2002 in the Action Plan on Floods (APF2002) and planned measures for 2020 (APF2020), and the measure ‘additional retention polders’ (* is used to highlight the additional retention compared to APF2020).
Rhine km Name Type Volume (106 m3 ) APF 2002 Measure APF Additional 2020 retention v v v v v v v v v v v v v v v v v v v v v v v v v v v v v v v v v v * v v v v v v v v * v v

160 235 246 253.5 275 276 280 308 321 330 359 360 368 381.3–383.0 384 388.4 390.4 392.6 409.9 411.5 436 438 438 440 488 490 517 546 614 660 668.5–673.5 688 705.5–708.5 707.5–713.5 723.5–727.5 750 750.5–754.5 760.5–769.5 797.5–803.5 818.5–823.5 832.5–833.5 837.5–847.5 882

Basel Breisach/Burkheim Wyhl/Weiswel Mouth of the Elz Erstein Ichenheim/Meisenheim Altenheim Freistett S¨llingen/Grefferen o Moder Daxlander Au Maxau Neupotzh/W¨rth o Elisabethenw¨rth o near Germersheim Mechtersheim Rheinschanzinsel Flotsgr¨n u Kollerinsel Waldsee/Altrip/Neuhofen Petersau/Bannen Worms Worms B¨rgerweide u Mittlerer Busch near Darmstadt Bodenheim/Laubenheim Ingelheim Kaub Andernach Upstream Cologne Cologne-Langel Cologne Worringer Bruch Monheim Itter-Himmelgeist Downstream Cologne Ilvericher Bruch M¨ndelheim u Orsoy Land Bislicher Insel Lohrwardt Griether Busch Lobith

RP RP RP RP RP RP RP RP RP RP DR + RP RP RP RP RP RP RP DR + RP DR DR DR RP RP RP

6.5 7.7 5.3 7.8 5.8 17.6 9.0 12.0 5.6 5.1 16.2 11.9 40.0 7.4 6.2 5.0 6.1 9.1 1.4 3.4 2.3 40.0 6.4 3.8

v v

v v

v

RP RP RP DR DR DR RP DR DR DR RP RP

50.0 4.5 29.5 8.0 2.0 60.0 8.4 3.0 10.0 17.4 25.0 25.0

v v

v v v v v v v v v v

* v v v v * v v v v v v

v v v v

5.3. Methods

113

Table 5.2: Current land use and the measure land-use change to forest in the Rhine basin.
Land-use class Forest Arable land Grassland Built-up areas Rock and glacier Lakes Other Total Current Area (km2 ) Area (%) 62 194 38.7 36 304 22.6 45 294 28.2 7 930 5.0 3 797 2.4 2 284 1.4 2 996 1.9 160 800 100 Land-use change Area (km2 ) Area (%) 154 719 96.2 0 0 0 0 0 0 3 797 2.4 2 284 1.4 0 0 160 800 100

Cologne bypass Cologne suffered from flooding in the years 1993 and 1995. The economic damage in 1993 was estimated at e 75 million. However, in spite of improved disaster management after the 1993 and 1995 floods, the city remains at high risk, since a flooding event at T =200 years (the current flood protection standard around Cologne) will result in water depths around 3 m in large parts of the city (Gocht and Vogt, 2008). To lower the water levels around Cologne during extreme floods, a bypass is proposed. A potential location for this bypass is shown in Figure 5.1. Since a hilly region borders the region east of Cologne, we chose a western route with relatively little urbanization. The bypass has a length of 72 km, starting at Rhine kilometer (rkm) 664, and ending at rkm 712, with Cologne being located at rkm 688). The cross section of the bypass has a depth of 10 m and a width of 120 m, whereas the main channel through Cologne measures around 350 m in width. Identical cross sections were placed at every 500 m and the bottom level was linearly interpolated between rkm 664 and 712. Friction values of the bypass are assumed to be the same as in the main Rhine branch. Increased friction by reforestation of the floodplains Increasing hydraulic friction in floodplains by reforestation or small dams is controversial as current policies in the Rhine are aimed at removing obstacles to increase flow velocity (Ministry of Transport, Public Works and Water Management, 2006c). Our hypothesis, though, is that reducing flow velocity in an upstream part of the river might be beneficial downstream. We tested the effectiveness of the measure by assuming an emergent vegetation cover in all the flood plains in the Upper as well as in the Lower Rhine. We assumed the height of a soft wood production forest as more than 3 m. According to the literature

114

Chapter 5. Effectiveness of flood management measures

Table 5.3: Properties of the measure restored abandoned meanders. The location names are displayed in Fig. 1. Maxau is located at Rhine kilometer (rkm) 360 and Worms at rkm 438.
Location Neupotz Linkenheim Hoerdt Lingenfeld Philipsburg Roermerberg Hockenheim Otterstad Ketsch Waldsee Bobenheim Lampertheim Gimbelsheim Stockstadt am Rhein Oppenheim Total length Length (km) 7.5 3.9 13.2 14.8 7.0 7.6 17.8 4.2 5.6 18.1 11.3 8.6 11.3 16.9 17.8 200.7 km Width (m) 400 300 200 200 200 200 200 200 150 150 150 150 400 300 300 rkm 367.5 367.5 373 385 389 391 399 403 406 409 436 438 466 467.5 476

the roughness coefficient (Manning) is 0.10 (Chow, 1959; Straatsma and Baptist, 2008). As only the floodplains will be forested, the value has been implemented only in the floodplain sections of the SOBEK schematization. Restored abandoned meanders of the Upper Rhine Increasing the river length substantially by restoring abandoned meanders is currently not implemented on a large scale in water management practices in the Rhine basin. However many abandoned meanders are still visible in the landscape as small depressions and sometimes as oxbow lakes. They are often not densely built upon, but are used for agriculture, grassland and nature. To simulate the restoration of these abandoned meanders, we schematized many additional branches in the Upper Rhine in the SOBEK model, based on Google Earth satellite images. We maintained the main channel, though, to allow for shipping. All additional branches are listed in Table 5.3 and displayed in Figure 5.1. Increased dike height We increased the dike height in our SOBEK model along all the branches to such an extent (several meters) that water levels would never reach the crest level. The

5.4. Results

115

possibility of dike failure is ignored. In this way, we created a situation where upstream flooding cannot occur.

5.4

Results

Figures 5.4 and 5.5 summarize the modeling results for the change in peak discharge and water levels, respectively. In the left panels the APF2002 simulation with control climate boundary conditions is used as a reference. The crosses in the same panels display the effect of the APF2020 measures compared to the reference situation. The circles in the left panels indicate the effect of climate change in 2050 in combination with the APF2020 measures on maximum water levels, also compared to the reference situation. In the right panels the APF2020 simulation with the climate change boundary conditions is used as a reference. All additional measures are evaluated against this reference situation and are indicated with different symbols (see the legend). Tables 5.4 through 5.7 contain all the numbers that are shown in Figures 5.4 and 5.5. The effect of dike heightening is relatively large (about a factor 4 larger than the effects of other measures) and hence does not fit well on the vertical scale in the right panels of Figures 5.4 and 5.5 (see Section 5.4.1). Hence, the effect of (extreme) dike heightening to an extent that no flooding occurs is not displayed in Figures 5.4 and 5.5. Furthermore, it should be noted that in the right panels of Figure 5.4, no effect of the measures at the location Maxau can be observed, except for land-use change to forest. This can be explained by the way the SOBEK model operates. SOBEK uses an imposed discharge series as upstream boundary conditions, which we derive from the HBV simulations. The model only calculates water levels belonging to the discharge series at this node. Thus, discharge remains the same, while water levels vary between different simulations due to adjustments in the river geometry or friction values downstream. In our model setup, Maxau is the upstream boundary node, and therefore, no effect on the discharge is observed at Maxau (Figure 5.4). The water level, on the other hand, does change between simulations at Maxau (Figure 5.5). Land-use change to forest is the only exception, since it influences the discharge generation by HBV through changed soil characteristics. As a result, the boundary discharge imposed at Maxau changes, which can be observed in Figure 5.4. The water level due to land-use change also changes (Figure 5.5).

116

Chapter 5. Effectiveness of flood management measures

T = 10
2000

T = 10 dQ (m3/s)
500 0 −500 −1000 L A K W M

dQ (m3/s)

1000 0 L 2000 A K W M

T = 200 dQ (m3/s) dQ (m3/s)
500 0 −500 −1000 L A

T = 200

1000 0 L 2000 A K W M

K

W

M

T = 500 dQ (m3/s) dQ (m3/s)
500 0 −500 −1000 L A

T = 500

1000 0 L 2000 A K W M

K

W

M

T = 1250 dQ (m3/s) dQ (m3/s)
500 0 −500 −1000 L A

T = 1250

1000 0 L A K W Location M

K W Location

M

Wp climate scenario 2050 + APF2020 Planned measures for 2020 (APF2020) Land use change to forest Extra retention polders Increased friction by reforestation of the flood plains Restored abandoned meanders Upper Rhine

Figure 5.4: Effect of strategies and the Wplus climate change scenario on maximum discharge (m3 /s). In the left panels the APF2002 schematization with control climate boundary conditions is used as reference. In the right panels the APF2020 schematization with the climate change boundary conditions is used as reference (W plus scenario for 2050). All values are displayed in Table 5.4 and 5.6. L is Lobith, A is Andernach, K is Kaub, W is Worms and M is Maxau.

5.4. Results

117

T = 10
1 0.5

T = 10

dh (m)

dh (m)
A K W M

0.5 0 L

0 −0.5 L A K W M

T = 200
1 0.5

T = 200

dh (m)

dh (m)
A K W M

0.5 0 L

0 −0.5 L A K W M

T = 500
1 0.5

T = 500

dh (m)

dh (m)
A K W M

0.5 0 L

0 −0.5 L A K W M

T = 1250
1 0.5

T = 1250

dh (m)

dh (m)
A K Location W M

0.5 0 L

0 −0.5 L A K Location W M

Wp climate scenario 2050 + APF2020 Planned measures for 2020 (APF2020) Land use change to forest Extra retention polders Increased friction by reforestation of the flood plains Restored abandoned meanders Upper Rhine

Figure 5.5: Effect of strategies and the W-plus climate change scenario on maximum water level. In the left panels the APF2002 schematization with control climate boundary conditions is used as reference. In the right panels the APF2020 schematization with the climate change boundary conditions is used as reference (W-plus scenario for 2050). All values are displayed in Table 5.5 and 5.7. L is Lobith, A is Andernach, K is Kaub, W is Worms and M is Maxau.

118

Chapter 5. Effectiveness of flood management measures

Table 5.4: Effect of planned measures for 2020 and W-plus climate change scenario on peak discharge (m3 /s). The numbers are plotted in Figure 5.5, left panels.
Climate Land use Dike height Measures Return period Lobith 10 200 500 1250 Andernach 10 200 500 1250 Kaub 10 200 500 1250 Worms 10 200 500 1250 Maxau 10 200 500 1250 Control Current Current APF2002 Q 10 491 12 022 12 317 12 827 9 017 12 158 12 724 13 415 6442 7737 7692 8139 5313 6546 6692 6692 4729 7160 7710 8421 Control Current Current APF2020 dQ −81 33 70 37 −27 −22 −2 −27 −64 −5 −4 −25 −78 −44 −25 −22 0 0 0 0 Wp 2050 Current Current APF2020 dQ 9 110 517 512 1227 33 755 1638 1047 1702 232 818 236 425 178 541 802 1431 1727 1438

5.4. Results

119

Table 5.5: Effect of planned measures for 2020 and W-plus climate change scenario on peak water level (m). The numbers are plotted in Figure 5.6, left panels.
Climate Land use Dike height Measures Return period Lobith 10 200 500 1250 Andernach 10 200 500 1250 Kaub 10 200 500 1250 Worms 10 200 500 1250 Maxau 10 200 500 1250 Control Current Current APF2002 h 15.99 16.57 16.68 16.87 60.84 62.84 63.17 63.55 75.39 76.44 76.40 76.75 91.26 92.09 92.15 92.16 106.79 108.11 108.33 108.61 Control Current Current APF2020 dh −0.04 0.01 0.02 0.01 −0.02 −0.01 0.00 −0.01 −0.05 0.00 −0.01 −0.02 −0.02 −0.02 −0.01 −0.05 0.00 0.00 0.00 0.00 Wp 2050 Current Current APF2020 dh 0.00 0.05 0.19 0.18 0.77 0.03 0.42 0.87 0.84 1.37 1.24 0.67 0.10 0.16 0.06 0.06 0.55 0.57 0.65 1.41

120

Chapter 5. Effectiveness of flood management measures

Table 5.6: Effect of additional measures on maximum discharge (m3 /s). The numbers are plotted in Figure 5.5, right panels.
Climate Land use Dike height Measures Wp 2050 Current Current APF2020 Wp 2050 Forest Current APF2020 Wp 2050 Current Current Additional retention dQ −8 −134 −321 −271 0 −3 −158 −50 −152 −6 −308 −41 −32 −15 −22 −17 0 0 0 0 Wp 2050 Current Current Increased friction dQ −590 −806 −973 −811 −275 −608 −792 −849 −551 −800 −848 −764 −298 −613 −622 −602 0 0 0 0 Wp 2050 Current Current Restored meanders dQ −4 330 224 439 2 226 636 349 116 280 220 400 145 642 666 701 0 0 0 0 Wp 2050 Current No flooding APF2020

Return period Lobith 10 200 500 1250 Andernach 10 200 500 1250 Kaub 10 200 500 1250 Worms 10 200 500 1250 Maxau 10 200 500 1250

Q 10 420 12 166 12 903 13 376 10 217 12 169 13 477 15 025 7425 9433 7920 8933 5000 6926 6845 6757 5532 8591 9437 9859

dQ −918 −362 −552 −533 −864 −415 −958 −1012 −571 −247 −309 −308 −407 −137 −120 −114 −498 −773 −849 −887

dQ −7 3234 1238 4378 0 3937 3991 2977 232 2691 4209 3848 21 2950 3664 1459 0 0 0 0

5.4. Results

121

Table 5.7: Effect of additional measures on peak water level (m). The numbers are plotted in Figure 5.6, right panels
Climate Land use Dike height Measures Wp 2050 Current Current APF2020 Wp 2050 Forest Current APF2020 Wp 2050 Current Current Additional retention dh 0.00 −0.06 −0.12 −0.09 0.00 0.00 −0.09 −0.03 −0.13 −0.01 −0.02 −0.04 −0.03 −0.01 −0.01 −0.01 0.00 0.00 0.00 0.00 Wp 2050 Current Current Increased friction dh −0.19 −0.32 −0.37 −0.29 0.31 0.21 0.18 0.24 −0.35 −0.43 −0.42 −0.46 −0.07 0.01 0.02 0.02 0.47 0.64 0.69 0.76 Wp 2050 Current Current Restored meanders dh 0.00 0.13 0.08 0.13 0.00 0.13 0.34 0.18 0.08 0.21 0.24 0.31 −0.72 −0.40 −0.39 −0.40 −0.48 −0.22 −0.20 −0.19 Wp 2050 Current No flooding APF2020

Return period Lobith 10 200 500 1250 Andernach 10 200 500 1250 Kaub 10 200 500 1250 Worms 10 200 500 1250 Maxau 10 200 500 1250

h 15.99 16.62 16.89 17.06 61.60 62.86 63.59 64.41 76.18 77.80 76.34 77.40 90.89 92.23 92.20 92.17 107.34 108.68 108.98 109.13

dh −0.30 −0.15 −0.20 −0.19 −0.48 −0.24 −0.53 −0.53 −0.47 −0.21 −0.04 −0.26 −0.31 −0.05 −0.05 −0.05 −0.33 −0.30 −0.31 −0.31

dh 0.00 1.02 0.29 1.29 0.00 2.12 2.05 1.48 0.18 1.88 3.25 2.70 0.02 1.81 2.17 2.45 0.00 0.00 0.00 0.00

122

Chapter 5. Effectiveness of flood management measures

5.4.1

Basin-wide effects

Climate change The mean increase in peak discharge due to climate change in 2050, when APF2020 measures are implemented, is 770 m3 /s, but the increase shows a large variation among locations and return periods (Table 5.4, Figure 5.4, left panels). For example, at Andernach at T =500, the discharge increase is 755 m3 /s, while at Maxau at T =500, the increase is 1727 m3 /s. The mean increase in peak water level is 50 cm, but varies between several centimeters and 137 cm (Table 5.5, Figure 5.5, left panels). Increased dike height From Tables 5.6 and 5.7 we can read that at T =10 in the W-plus climate change scenario, raising dikes has hardly any effect on water levels and discharges. Apparently, under climate change, flooding does not occur at any of the five locations along the Rhine at current dike heights, and thus raising them has no effect. To be more precise we should explain that Kaub is located at the Middle Rhine, where the river flows through a narrow valley, and is currently not embanked by dikes. In the model setup with increased dike height, we also embanked the Rhine at Kaub. At higher return periods, though, peak water levels rise when dike heights are increased. This implies that flooding will occur at these return periods under climate change, when dike heights are left at their current levels. The maximum water level increase at Lobith is 129 cm (T =1250), at Andernach 212 cm (T =200), at Kaub 325 cm (T =500) and at Worms 245 cm (T =1250) (Table 5.7). At Maxau, no flooding occurs in our model, and thus increasing dike height has no effect. This relates again to Maxau being the upper boundary node of the SOBEK model. From Figure 5.1 we can learn, though, that Maxau lies within the flood prone area of the Rhine, and the flood prone area extends even further upstream up to Basel. Peak discharge at return periods above T =10 increase between 1238 m3 /s at Lobith (T =500) and 4378 m3 /s at Lobith (T =1250). Our results imply that a dike raise is needed along the Rhine that varies between 1.29 m and 3.25 m, depending on location, to prevent these areas from flooding. Land-use change to forest Land-use change to forest seems an efficient measure to reduce peak discharges and peak water levels at a basin-wide scale. All simulated annual discharge maximums decrease by 114 m3 /s up to 1012 m3 /s as a result of land-use change to 96% forest,

5.4. Results

123

and water levels decrease between 4 cm and 53 cm (Tables 5.6 and 5.7, Figures 5.4 and 5.5, right panels). Reforestation in the Rhine basin results in lower peak discharges and water levels at all simulated return periods, i.e. even at the most extreme peak events. Reduced flood discharge from the tributaries is probably due to both higher evaporation and infiltration rates, causing a decrease in the baseflow.

5.4.2

Local effects of flood management measures

APF2020 The APF2020 measures result in peak discharge reductions of only −80 m3 /s and water level reductions of −5 cm both for current climate conditions and the increased discharges under climate change (Figures 5.4 and 5.5, left panels). The minor effectiveness of the APF2020 can be explained by two phenomena that relate to (1) the way the retention polders are operated and (2) on whether upstream flooding occurs, both illustrated in Figure 5.6. (1) Retention polders are designed to reduce the maximum water level of a flood peak by inundating the retention area when the water level in the main river branch reaches a critical level. Retention basins in Germany become operational at T =50 to 100 (IKSR 2005). Their volumes are designed for peak events of corresponding size. At higher peak events, with return periods larger than 100 years, retention basins will be full when the maximum peak discharge arrives. In Figure 5.6a the APF2020 measures have some effect at T =10 at Lobith (indicated by the dashed line). Apparently, upstream of Lobith this particular flood wave reaches threshold levels (at T =50 or more) of several retention polders along the Rhine. Figure 5.6b shows that retention polders become operational at the same discharge as in Figure 5.6a (∼10 500 m3 /s), but that the flood peak reaches a level where retention polders fill up completely, and the efficiency of the retention polders declines. This explains partly the ineffectiveness of the retention measures in the APF2020 at extreme discharges of more than 12 000 m3 /s at Lobith at T =200 and more. (2) Upstream flooding acts as a major retention basin, therefore blurring the effect of the actual planned retention polders. At T =10 and T =100 (Figure 5.6a and b), no upstream flooding occurs. At T =200, though, flooding does occur (Figure 5.6c). When the effectiveness of the retention polder is tested in a simulation run without flooding (black line), the retention polders become operational at ∼10 500 m3 /s, as in Figure 5.6a and b, and manage to decrease the maximum peak discharge of 13 000 m3 /s by ∼100 m3 /s (the dark gray dotted line). When flooding is simulated, the maximum peak discharge reaches only 12 000 m3 /s (the grey line in Figure 5.6c). Flooding apparently occurs at 11 000 m3 /s at Lobith

124

Chapter 5. Effectiveness of flood management measures

1.4 1.3 Q (m3/s)

x 10 a)

4

T = 10 1.4 1.3 Q (m3/s) 1.2 1.1 1 0.9

x 10 b)

4

T = 100

1.2 1.1 1 0.9

50

100 150 t (hr) T = 200

200

50

100 150 t (hr)

200

1.4 1.3

4 x 10

c)

2002 incr. dike height (no flooding) Q (m3/s) 1.2 1.1 1 0.9 2020 incr. dike height (no flooding) 2002 current dike height 2020 current dike height

50

100 150 t (hr)

200

Figure 5.6: Effect of APF2020 retention measures on flood peaks with different return periods at Lobith, with and without the simulation of flooding. Flooding does not occur at T =10 and 100.

5.4. Results

125

and retains a substantial volume, illustrated as the area between the black and gray lines. The relatively small volume of the APF2020 measures scarcely inflicts extra retention (the dotted line). This further explains why in Figures 5.4 and 5.5 the effects of APF2020 measures are scarcely visible. Additional retention polders The additional retention volume results in a mean decrease of flood peak water levels of 3 cm, and varies between 0 and 13 cm, depending on the location (Table 5.7). The operational rules of retention polders, i.e. at which water level or discharge they become operational, are the same as schematized for retention volumes planned for 2020 that are closest to the additional retention polders. This means that these additional retention volumes are also most effective at T =50 to 100. Furthermore, flooding obscures the dampening effect of these retention measures, in the same way as described above in Section 5.4.2 for the APF2020 measures. Increased friction by reforestation of the floodplains Increasing friction by reforestation of all floodplains is very effective in lowering the discharge at all locations and return periods by 280 m3 /s down to 970 m3 /s (Figure 5.4 and Table 5.6). However, lower flow velocities result in a storing effect and hence higher water levels, comparable to the effect of a bottleneck. This storing effect is visible both at the Upper Rhine (Kaub, Worms and Maxau) and the Lower Rhine (Lobith and Andernach) (Figure 5.5). At Maxau, the water level increases by a mean value of 64 cm, at Worms there is no effect and at Kaub, the water level decreases by 41 cm. At Andernach, the water level increases by a mean value of 21 cm before it decreases by 29 cm at Lobith (Table 5.7). In short, increased friction might be beneficial on a local scale, but can easily have an opposite effect in the upstream direction. Restored abandoned meanders in the Upper Rhine Based on historical descriptions of the Rhine and its discharge behavior (Blackbourn, 2006), it was expected that increasing the flow path of the Upper Rhine by restoring abandoned meanders in our model would reduce flow velocity and broaden and attenuate flood waves basin-wide. However the results contradict this theory. The restored meanders result in increased discharges at all locations and over all return periods (Figure 5.4). It appears that since the canalized channel remains in use in this schematization (to aid shipping), the restored meanders are merely additional branches, creating an increased flow capacity and therefore increasing discharge. The

126

Chapter 5. Effectiveness of flood management measures

Table 5.8: Effect of the bypass around Cologne on peak discharge (m3 /s) and peak water level (m).
Climate Land use Dike height Measures Wp2050 Current Current APF2020 Wp2050 Current Current APF2020 and Cologne bypass dQ −1939 −2240 −2320 −3542 dh −0.96 −0.99 −0.97 −0.85

Return period Cologne 10 200 500 1250

Q 10 542 12 258 12 786 13 560

h 45.45 46.40 46.70 47.03

water levels therefore decrease in the Upper Rhine (−19 cm to −72 cm, Table 5.7), which is a positive result. In the Lower Rhine, on the other hand, water levels increase (up to 34 cm, Table 5.7) as a result of increased discharge, where there is no change in the flow capacity. Cologne bypass The bypass results in a substantial lowering of the peak water level of almost 1 m at Cologne (Table 5.8) and thus is very effective, but the effect is only local. Twenty percent of the total discharge runs through the bypass in the case of a flood peak event, and the probability of flooding decreases significantly for the city of Cologne (a Q10 event will become a Q1250 event, Table 5.8).

5.4.3

Longitudinal profiles of peak water levels

Longitudinal profiles of peak water levels can help to determine in more detail the location and the extent of effectiveness of flood management measures along the Rhine branch. As an example, we plotted the effect on peak water levels of different measures but for only one particular flood wave (one of the 16 available flood waves) in longitudinal profiles in Figure 5.7. The displayed flood wave has a return period of 200 years at Andernach. The water levels are plotted at every cross section at 500 m intervals. In Figure 5.7a, the effect of flooding is illustrated, under current climate conditions (light blue), and under the W-plus scenario in 2050 (dark blue). Because the discharge is expected to increase as a result of climate change, the lowering of peak water levels

5.4. Results

127

a) 0 dh (cm) −100 −200 850 b) dh (cm) 200 100 0 850 c) dh (cm) 50 0 −50 −100 850 800 750 700 650 600 rkm 550 500 450 400 800 750 700 650 600 rkm 550 500 450 400 800 750 700 650 600 rkm 550 500 450 400

ne

ub

bi

ac

m

Ka

og

W or

Lo

rn

ol

C

Figure 5.7: Longitudinal profile of the change in peak water level (dh). The boundary conditions are T =200 at Andernach (see Section 5.3.3). Displayed are, a) the effect of flooding, under control climate conditions (light blue), and under the W-plus scenario in 2050 (dark blue); b) the effect of climate change, in a schematization with infinite dike heights (no upstream flooding) (magenta), and at current dike height (flooding occurs) (purple); c) the effect of Cologne bypass (grey) and 96% forest (black).

An

de

M

ax

au

th

h

s

128

Chapter 5. Effectiveness of flood management measures

under the W-plus scenario is higher than under current conditions. The range of the decrease is 50 to 150 cm under current climate conditions and 100 to 250 cm under climate change in 2050. In Figure 5.7b, the effect of climate change is displayed, in a schematization with infinite dike height (no flooding) (magenta), and in a schematization with current dike height (flooding can occur) (purple). The water level increases up to 200 cm as a result of climate change, when upstream flooding is not simulated. When upstream flooding is simulated, from rkm 600 and downward, there is no increase in water level. It seems that the increase in peak water levels from climate change is compensated completely by the effect of flooding. Nevertheless, on the Upper Rhine an increase of peak water levels due to climate change remains, even when flooding is simulated. Figure 5.7c displays the local effect of the bypass around Cologne (light grey). The basin-wide decrease in maximum water level as a result of reforestation is also shown (black).

5.4.4

Flood-peak probability at Lobith

So far, we have evaluated the effect of climate change and flood management measures on reducing flood-peak discharges and water levels, at different return periods. These results, however, do not provide information on how extreme value distributions of yearly maximum flood-peaks, and thus flood-peak probabilities, might change. Therefore, in this section, we will analyze the effectiveness of a few flood management measures, the effect of climate change and the effect of upstream flooding on floodpeak probabilities. We limit the analysis of flood management measures to APF2020 and land-use change to forest, and to the gauging station Lobith (at the border of the Netherlands and Germany). To do this, we made several extra runs of 1000 years using the same model setup with HBV and SOBEK. The reference situation contains all the flood protection measures of the APF that were implemented in 2002. Simulation results of 1000 years of yearly maximum peak discharges for different scenarios are shown in extreme value plots in Figure 5.8 and Table 5.9. According to Figure 5.8a and 5.8b, flood probability will increase by a factor of 4 as a result of the Wp climate change. However, both in the control climate and when using a climate change scenario for 2050, the effects of flooding can be seen at discharges above 12 500 m3 /s. Without flooding, the extreme values describe more or less a straight line in the extreme value plots in Figure 5.8 (indicated by circles). As flooding tops off the highest peaks (crosses), a breakpoint can be observed in extreme value plots. The increase in peak discharge as a result of climate change ranges from 16–19% when upstream flooding does not occur, and from 8–11% when upstream flooding is taken

5.4. Results

129

a)

Return period (year) 1.8 x 10 20
4

b)

Return period (year) x 10 20
4

50

200 500 1250

1.8 Yearly max. Q (m3/s) 1.6 1.4 1.2 1 0.8

50

200 500 1250

Yearly max. Q (m3/s)

1.6 1.4 1.2 1 0.8

Contr cl no fl Contr cl with fl 2 4 6 Gumbel variate (−) Return period (year)
4

Clim ch (Wp) no fl Clim ch (Wp) with fl 2 4 6 Gumbel variate (−) Return period (year)
4

8

8

c)

d)

1.8 Yearly max. Q (m3/s) 1.6 1.4 1.2 1 0.8
e)

x 10 20

50

200 500 1250

1.8 Yearly max. Q (m3/s) 1.6 1.4 1.2 1 0.8

x 10 20

50

200 500 1250

Contr cl no fl Contr cl no fl APF2020 2 4 6 Gumbel variate (−) Return period (year)
4

Contr cl with fl Contr cl with fl APF2020 2 4 6 Gumbel variate (−) 8

8

1.8 Yearly max. Q (m3/s) 1.6 1.4 1.2 1 0.8

x 10 20

50

200 500 1250 Breakpoint due to flooding

Clim ch (Wp) no fl Clim ch (Wp) no fl landuse forest 2 4 6 Gumbel variate (−) 8

Figure 5.8: Extreme value plots for the yearly discharge maxima at Lobith for different climate conditions and measures. Contr cl is control climate, fl is flooding, clim ch (Wp) is the Wp climate change scenario for 2050.

130

Chapter 5. Effectiveness of flood management measures

Table 5.9: Estimated return periods obtained by ranking the 1000 years of simulated peak discharges at Lobith according to size, and linking return periods to the ranks.
Rank Return period Control climate Without upstream flooding m3 /s 15 694 15 047 14 321 12 880 11 938 W plus (2050) Without upstream flooding m3 /s (%) 18 215 (16.1) 17 696 (17.6) 16 704 (16.6) 15 186 (17.9) 14 147 (18.5) Control climate With upstream flooding m3 /s 13 918 13 719 13 052 12 554 11 807 Wp (2050) With upstream flooding m3 /s (%) 15 445 (11.0) 14 809 (7.9) 14 457 (10.8) 13 576 (8.1) 13 025 (10.3)

1 2 5 10 20

year 1000 500 200 100 50

into account (Figure 5.8a and 5.8b). The breakpoint due to flooding is at T =100 (∼12 500 m3 /s) in the reference situation, while under climate change it corresponds to T =25 (i.e. a factor of 4 increase in probability of flooding). Flooding in areas upstream of Lobith significantly lowers discharge peaks at Lobith, with 2–13% in the control climate and 10–19% in the W-plus climate change scenario. The curve in Figure 5.8a and 5.8b indicates that there is a physical maximum of the peak discharge that can reach Lobith, due to upstream flooding. Figure 5.8c and 5.8d visualize the effect of the APF2020 measures in a control climate situation without and with flooding. It shows that the retention polders become operational between T =20 and 200, with a minimal decreasing effect on peak discharge. Considering Figures 5.8a through 5.8d, it is obvious that the APF2020 cannot restrain the impact of climate change. Finally, Figure 5.8e shows the effect of land-use change in the Wp climate change scenario. All flood peaks are lowered by ∼ 1000 m3 /s as a result of reforestation to 96% forest in the Rhine basin.

5.5
5.5.1

Discussion and conclusions
Methods

The aim of this Chapter was to explore a method to evaluate the effectiveness of flood management measures that are planned in the Action Plan on Floods (ICPR, 2005a) and additional measures along the river Rhine, assuming a climate change scenario for 2050. Our approach addressed two methodological challenges needed to evaluate the effectiveness of flood management measures in the Rhine basin. First, we explained the issue of high safety standards in the Rhine basin (up to 1/1250 per year) that requires

5.5. Discussion and conclusions

131

extrapolation of historical time series to reach peak flows at such low probabilities of occurrence. In addition, extrapolation assumes stationarity of the data record, while both meteorological conditions in and physical conditions of the river basin have changed. Both issues introduce uncertainty. In the traditional way of estimating dimensions of low-probability flood peaks based on 100 years of observations, the statistical uncertainty is 13% more or less discharge volume at the 1/1250 per year event (Silva et al., 2001; Te Linde et al., 2010). In order to tackle this uncertainty, we applied a weather generator to create long time series of meteorological data (10 000 years) at multiple locations that were used as input for hydrological models in order to simulate long discharge series. These long time series have proven to be useful in reducing the statistical uncertainty from 13% to 3% at Lobith when estimating dimensions of low-probability flood peaks, because extrapolation of extreme value distribution fits is no longer necessary (e.g. Te Linde et al., 2010). Also, in our simulation approach, the impacts of climate change and the alterations in river geometry and land use due to human influence are parameterized in models, in an attempt to physically describe extreme situations and the consequences of changed conditions. In doing so, we reject the basic assumption of stationarity of water systems, following Pielke Sr. et al. (2009). Such a simulation approach is referred to as ‘process-based’ by McMillan and Brasington (2008), and is advocated by Sivapalan and Samuel (2009) and Raff et al. (2009). The method allows us to simulate flood waves at very low probabilities to test the effectiveness of measures upon extreme events. However, the use of a resampling technique to generate extreme events inherits an unknown uncertainty, since it has been trained using a relatively short reference period (35 years) which does not necessarily include such extremes. The second methodological issue was the necessity to include hydrodynamic modeling to allow for the simulation of upstream flooding. Existing flood management evaluation studies for the Rhine did not incorporate upstream flooding ICPR (2005a); Bronstert et al. (2007), while upstream flooding does occur at extreme peak events and has a substantial reducing effect on discharges downstream in the Rhine delta (Lammersen, 2004). We found that flooding in the Upper and Lower Rhine in Germany, upstream of Lobith, has a profound decreasing effect on discharge peaks at Lobith. The decrease varied between 2–13% in control climate conditions and 10–19% in the W-plus climate change scenario. The curve in Figure 5.8a and 5.8b indicate that there is a physical maximum of the peak discharge that can reach Lobith, due to upstream flooding. Hence, upstream floods in Germany are favorable for reducing flood risk in the downstream areas of the Netherlands. However, it is possible that future flood policies in Germany will aim at raising their dikes, especially in a scenario with increased flood probabilities due to climate change. This may increase peak discharges and water levels downstream (in the Netherlands).

132

Chapter 5. Effectiveness of flood management measures

5.5.2

Effectiveness of measures

The mean increase in peak water level due to climate change in 2050 is 50 cm, but varies between several centimeters and 137 cm (Table 5.5, Figure 5.5, left panels). Currently implemented and proposed measures in the Action Plan on Floods, as well as most additional measures we evaluated, seem inadequate to cope with increased flood probabilities that are expected in a future climate change scenario. According to our results, the only measure that can prevent the Rhine from flooding is drastic dike heightening of between 1.29 and 3.25 m, depending on location, on the assumption that these dikes cannot fail. The APF2020 measures, as well as additional retention polders, reduce peak water levels by 5 cm to 13 cm over medium return periods (between 50 and 100 years) for the control climate. At T =200 and more, they have no effect at all. The minor effectiveness of the APF2020 can be explained firstly by the way the retention polders are operated. We have shown that retention polders as outlined in the APF2020 become operational between T =20 and 200, and require well-defined control rules and excellent flood forecasting in order to operate optimally, which is also explained by (Lammersen, 2004). At higher flood peaks with longer return periods, such as in our simulations, they are not effective. Second, upstream flooding acts as a major retention basin, therefore blurring the effect of the actual retention polders. Flooding retains a substantial volume, illustrated as the area between the black and gray lines in Figure 5.6, and the relatively small volume of the APF2020 measures and additional retention polders hardly imposes extra retention. Increased friction by reforestation of the flood plains showed to be beneficial at a local scale by lowering the water level several decimeters. However, higher friction values resulted in a storing effect that caused increased water levels in the upstream direction. Swiatek and Kubrak (2007) explain in an experimental study how vegetation causes the reduction of an active area of a cross-section, increases flow resistance and finally generates rising water levels. The bypass around the city of Cologne reduced water levels at a local scale. Restoration of abandoned measures in the Upper Rhine was also very effective in reducing water levels locally, but resulted in increased water levels in the Lower Rhine. Land-use change to forest decreased maximum water levels between 4 cm and 53 cm. The range in percentage of decrease in discharge is between 2% and 9%. However, modeling land-use change with conceptual models, such as HBV and RhineFlow (Van Deursen and Middelkoop, 2001), has limitations, since these models cannot represent all the processes influenced by land-use change. For example, simulations of the HBV model are very sensitive to changes in the maximum field capacity (Seibert, 1999).

5.5. Discussion and conclusions

133

The basin-wide model might perform well, while field capacity parameters might be over- or underestimated for different land-use classes. Also, HBV assumes higher evaporation rates in forests, which results in smaller saturated areas in case of heavy rainfall. A change in land use can then inflict a large shift in the simulated discharges by HBV, but these can be model artefacts. Physically-based rainfall-runoff models describing soil-surface processes in more detail than conceptual models should be able to perform better, but are very demanding in terms of data requirement, parameter estimation and computation time. Hurkmans et al. (2009) therefore used the land surface model VIC in an effort to simulate relative changes in peak discharge due to land-use change by describing several processes in more detail than HBV does, such as infiltration and evaporation rates. As a result, VIC requires more input data and calibration parameters Te Linde et al. (2008a). The authors used a scenario of 80% forest, 11% grass and 4.6% urbanized area. However, in their results, even at a Q1000 event, this scenario hardly inflicted any change in peak discharge (< 1%) at Lobith, nor in sub-basins such as the Mosel or the Neckar. Moreover, Pfister et al. (2004) points out that no clear evidence exists from 20th century historical time series that land-use changes influenced flood probability and magnitude in the main channel of the Rhine. Bronstert et al. (2007) reveal that landuse change to forest might be beneficial on a sub-basin scale in the Rhine basin, but the effect diminishes with increasing scale. In short, our results on the effectiveness of land-use change must be questioned, when compared to earlier work. The only exception is a study by Hundecha and B´rdossy a (2004) who used the HBV model in a similar simulation setup for the Rhine basin. They found reduced peak discharges of 10% to 19% for several historical events, resulting from reforestation of the entire basin. Therefore, the simulated land-use effect might be a limitation of HBV. Based on the fact that a 96% forested area (the remaining 4% being bare rock and surface water) is a very unlikely scenario in the Rhine basin, we do not consider land-use change to forest a profitable or efficient option to reduce flood probability on a basin-wide scale in the Rhine basin.

5.5.3

Further work

In this Chapter, we only evaluated measures to lower peak water levels and the probability of flooding. If a continuous dike raise is undesirable, new research could also focus on flood damage reduction measures. Based on our results (expected increase in flood probability due to climate change, and flood management which seem to be not very effective in water level reduction), we are inclined to support the conclusions made by Hooijer et al. (2004) and the Action Plan on Floods (ICPR, 2005a) that more could be gained from damage reduction and spatial planning than from flood defense measures to lower flood risk in the Rhine basin. This assumption, however, is not yet

134

Chapter 5. Effectiveness of flood management measures

confirmed by research. This would also require an upgrade from 1D to 2D inundation modeling to perform a process-based flood risk assessment, which needs 2D inundation information to meet the needs of advanced flood mitigation measures (McMillan and Brasington, 2008). Also, for simulating the effects of upstream flooding and restoration of abandoned meanders a 2D model application is advised. We made an improvement in modeling techniques towards assessing low-probability flood events under climate change. Several uncertainties remain of which some might be tackled in further research. We chose an extreme climate change scenario to test the robustness of measures, but to address the uncertainty of climate change impact models, a stochastic approach with multiple scenarios might be considered. In order to improve the simulation of the effects of flood management measures and upstream flooding above Maxau, the hydrodynamic model we used can be extended in the upstream direction (up to Basel). Finally, we did not take into account changes in morphology, while the Rhine currently incises with an average rate of 2 cm a year at Lobith (Silva, 2003) and deposits its sediments in other areas. If the current erosion continues, this would significantly influence peak water levels in 20 or 50 years time. Acknowledgements The meteorological dataset was kindly provided by the International Commission for the Hydrology of the Rhine basin (CHR) and we thank the Dutch Ministry of Transport, Public Works and Water Management and the Bundesanstalt f¨r Gewasserkunde u (BfG) for providing discharge observations. We also thank the BfG for sharing landuse data and for the use of HBV. Deltares and the Dutch Ministry of Transport, Public Works and Water Management are kindly acknowledged for sharing their SOBEK model. Special thanks to Rolf van der Veen and Rita Lammersen for their assistance in obtaining and improving the SOBEK schematizations. We are also grateful to Marcel Ververs and Albrecht Weerts for their work on the implementation of the simulation models in GRADE. This research project was carried out in the framework of the Dutch National Research Programme Climate changes Spatial Planning (www.klimaatvoorruimte.nl).

Chapter

6

Synthesis and recommendations

6.1

Summary of results

The overall objective of this thesis was to assess the effect of climate and socioeconomic changes on flood risk in the Rhine basin and where needed to improve simulation methods. The focus was on the estimation of the probability of flooding and the development of cross-boundary flood management measures to cope with expected changes. This section first describes the improvements made to the simulation method. Next, results are summarized for the projected changes in flood-peak probability and flood risk, and the effectiveness of flood management measures. The additional sections of this Chapter subsequently respond in more detail to the research questions addressed in the thesis and provide implications, recommendations and remaining challenges. Previously available methods for simulating the effect of climate change on floodpeak probabilities in the Rhine basin contained various shortcomings. First, the prevalent method for estimating future low-probability peak discharge events provides results only with considerable uncertainties, because it extrapolates observed discharge records (∼ 100 year) to estimate the peak discharges, with probabilities of 1/200 to 1/1250 per year. The observation records are too short for that purpose. In addition, such an approach assumes stationarity of the system; the idea that a natu-

135

136

Chapter 6. Synthesis and recommendations

ral system fluctuates within a fixed range of variability (Milly et al., 2008). However, both human interventions and climate change effects imply that stationarity may not apply to the Rhine basin. Second, in many cases of transforming climate model output to discharge, either inadequate rainfall-runoff models or too simple climate scenario transformation methods have been used, or a combination of both. Thus, improvements are necessary. For accomplishing this, we coupled the available simulation tools to interpret the climate change scenarios, rainfall-runoff modeling, and extreme flood-peak estimation. The methods used in this thesis address the need to define the most relevant uncertainties and processes in the cause-effect chain of climate change and integrate these into a process-based simulation approach. In addition, a flood risk model that can interpret the scenarios for both climate and socio-economic change was previously not available for the Rhine basin, and is newly developed in this thesis (Chapter 4). The improvement of the simulation method for low probability flood events in the Rhine basin is not primarily found in a more detailed description of hydrological processes (Chapter 2), but in carefully coupling different methods and models needed to simulate low-probability flood events. On the basis of this research, it is recommended to improve the projections of the intensity and duration of future extreme meteorological events, and a better description of routing processes in flood peak generation. The use of a weather generator results in long series of resampled meteorological data (1000 years of daily values), which considerably decreases the statistical uncertainty within the extreme value analysis (Chapter 3). The weather generator was implemented both for the present situation, and for different climate scenarios in 2050. In order to explore the uncertainties of the climate change scenarios, we used different climate models and downscaling methods. A hydrodynamic model was introduced to simulate flood wave propagation, backwater effects, flood plain inundation and even upstream flooding, as these processes have been largely ignored in existing climate impact studies to date in the Rhine basin. The hydrodynamic model was also used to simulate the impact of flood management measures (Chapter 5). The impacts of climate change and the alterations in river geometry and land use due to human influence were parameterized in models, in an attempt to improve the simulations of extreme flood events and its consequences. In doing so, we avoided the basic assumption of stationarity of water systems (e.g. Pielke Sr. et al., 2009). This simulation approach is referred to as ‘process-based modeling’ by McMillan and Brasington (2008), and is advocated by Sivapalan and Samuel (2009) and Raff et al. (2009). The method allows us to simulate flood waves at very low probabilities to test the effectiveness of measures upon extreme events and can be applied to other river basins. The improved process-based modeling method involved: 1. Selection of an adequate hydrological model.

6.1. Summary of results

137

2. Downscaling of the GCM output to meteorological forcing data for 134 subbasins in the Rhine basin, performed for different climate change scenarios. Downscaling can be done by a RCM or by the delta method. 3. Resampling of meteorological forcing data to 1000 (when possible 10 000) years of daily precipitation and temperature values in 134 sub-basins. 4. Generation of a series of 1000 (when possible 10 000) years of daily discharge using a semi-distributed conceptual hydrological model. Long time series decrease the statistical uncertainty when simulating low-probability peak events. 5. Implementation of a hydrodynamic model to simulate routing processes of flood peaks, upstream flooding, and the impact of flood management measures. In Chapter 2 we found that conceptual model HBV performs better than the more complex land-surface model VIC for the hydrological simulation of climate change scenarios in the Rhine basin, since it models the peak discharges well. In addition, HBV has very short calculation times compared to VIC, making it suitable for the generation of long time series of daily discharge. In Chapter 3 we concluded that for downscaling climate data, bias-corrected direct RCM output performs better than the delta method, since this method can describe spatial and temporal variation in meteorological events that are relevant for the planning of flood management measures. The HBV model, forced by RCM output, transforms the changed characteristics and extremes in the atmospheric system into physically plausible responses of extreme discharges. The results in Chapter 3 further indicate that there is a substantial increase in floodpeak probability in 2050 throughout the Rhine basin, when compared to the control climate period of 1961–1995, due to increased predicted precipitation under climate change. The increase in yearly maximum peak discharge varied considerably throughout the basin and between RCM transformation methods, ranging from 6% to 40%. At Lobith, the probability of flooding (currently 1/1250 per year) is expected to be three to five times as high in 2050 as it is today. Increasing the length of the climate data series to 1000 years reduced the statistical uncertainty when estimating low-probability (1/200 per year, or less) flood peak discharges, from ± 13% to ± 3%. The statistical uncertainty is illustrated here by the 95% confidence intervals around an extreme value distribution (GEV, see Appendix C. Chapter 4 provided estimates of future flood risk, which is defined here as the product of a probability and a consequence. We estimated that the annual expected loss in the Rhine basin may increase between 54% and 230% in 2030 compared to 2000, of which ∼ three-quarters of the increase can be attributed to climate change, and the remaining part to socio-economic developments. Results further showed that the area

138

Chapter 6. Synthesis and recommendations

with the highest flood risk numbers is located in the Lower Rhine in North RhineWestphalia in Germany, and not in the Netherlands as previously believed (ICPR, 2005b). Chapter 5 evaluated the effectiveness of flood management measures for their ability to reduce peak water levels and the probability of flooding. In the current situation, safety levels vary throughout the Rhine basin, and decrease in the upstream direction. Consequently, flooding is more likely upstream in Germany and France, while dikes prevent the Rhine from flooding in the Netherlands. It was found that upstream flooding in Germany has a profound decreasing effect on the simulated peak water levels and discharges along the main Rhine branch and downstream in the Netherlands. For example, an event that results in 16 000 m3 /s at Lobith when simulated without flooding, will decrease to ∼ 14 000 m3 /s when simulated with upstream flooding. The mean increase in peak water level along the Rhine due to climate change in 2050 is 50 cm, but varies between several centimeters and 140 cm. The currently implemented and proposed measures in the Action Plan on Floods, as well as most additional measures we evaluated, seem inadequate to cope with increased flood probabilities that are expected in a future climate change scenario. According to my results, the only measure that can prevent the Rhine from flooding is drastic, basin-wide, dike heightening of between 1.30 m and 3.30 m, depending on location, on the assumption that these dikes cannot fail.

6.2
6.2.1

Research questions
Comparing model performance

Does a physically-based land-surface model perform better than a conceptual rainfall-runoff model when simulating the effect of climate change on the discharge regime? The land-surface model VIC and the conceptual rainfall-runoff model HBV were compared for their performance in predicting climate change effects. Both models were forced with different meteorological data sets and their results compared at both a basin and sub-basin scale using various performance criteria (Appendix A). The comparison was made for daily results over the period 1993–2003. Both models performed less well in the upstream basins but improved at the Rhine basin outlet at Lobith. For all sub basins, HBV performed better than VIC. At Lobith, the HBV model forced with ERA15 performed moderately well (E=0.62, r2 =0.65), whereas the VIC model performed badly (E=0.31, r2 =0.54). For the calibration year 1993 with the model forced with observed CHR data, the HBV performance increased up to a coefficient

6.2. Research questions

139

of efficiency (E) of 0.85 and a coefficient of determination (r2 ) of 0.97. However, the performance of VIC remained poor when forced with CHR (E=0.44, r2 =0.81). HBV performed more consistently between the calibration and validation period than did VIC, which indicates that HBV is more robust (Figure 2.6). Overall, these results show that not only does HBV performs better than VIC at the sub-basin scale at the basin outlet, but also that the forcing data to a large extent influences the performance values of both models. For the most extreme peak flows (above 10 000 m3 /s at Lobith), HBV performed best simulating the maximum discharge (dmax. Qsim HBV 1–17%, dmax. Qsim VIC 2– 27%). HBV was also able to accurately predict the timing of flood peaks, whereas VIC tended to delay the flood peaks, some of them up to six days. Overall, the semi-distributed conceptual HBV model performed much better than the distributed land surface model VIC. This is in contrast to the general idea that more complex distributed models are better at predicting observed discharges than simple conceptual models (Reggiani and Schellekens, 2003; Refsgaard, 1996). Even for a well documented river basin such as the Rhine, more complex modeling does not automatically lead to better results (Booij, 2003; Uhlenbrook, 2003). For the Rhine basin thus, the HBV model is the preferred hydrological model for use in climate scenario studies, since it has better overall performance and appears to be more robust than the VIC model. Also, HBV better simulated extreme events, which may occur more in the future due to climate change. The more realistic representation of evaporation processes within VIC than HBV did not improve VICs performance even in the dry periods, when evaporation is a substantial component of the water balance. The final advantage of HBV over VIC is that HBV has short computational times, making it suitable for simulating long time series for many different climate scenarios.

6.2.2

Simulating climate change

How can we optimally use climate change scenarios of precipitation and temperature to estimate expected changes in low-probability flood peak events? To answer this question, the approach adopted in this thesis used RCM data and a weather generator to create long, resampled time series of climate change scenarios for input to hydrological (daily) and hydrodynamic (hourly) modeling for the Rhine basin. This approach was applied to three parallel modeling methods, each using a different method to transform the climate data of different RCM outputs to the hydrological model input. The three modeling methods were: delta change (method 1), RACMO direct (method 2), and RACMO bias-corrected (method 3). RACMO

140

Chapter 6. Synthesis and recommendations

refers to a climate scenario that is the output of a one RCM (RACMO2.1), which is forced by the ECHAM5-GCM member 3 output that was forced by the SRES-A1B emission scenario (Lenderink et al., 2003; Meijgaard et al., 2008). The delta change method was applied to the KNMI’06 W-plus scenario. The KNMI’06 scenarios are based on a suit of GCM and RCM simulations, forced by several SRES emission scenarios (Van den Hurk et al., 2006, 2007). On the basis of numerous 1000-year model simulations for the Rhine, the results suggest a basin-wide increase in peak discharge in 2050 of 8 to 17%, for probabilities between 1/10 and 1/1250 per year. For the 1/1250 per year discharge of 16 000 m3 /s at Lobith, this implies an increase to 17 280 m3 /s or 18 725 m3 /s. Furthermore, the results show that the statistical uncertainty when estimating low probability flood-peak events can be reduced from ± 13% (at 16 000 m3 /s this corresponds to ± 2100 m3 /s and ± 70 cm) to ± 3% (at 16 000 m3 /s this corresponds to ± 500 m3 /s and ± 15 cm) by increasing the length of the climate data series using a weather generator. The flood frequency is expected to increase as a result of climate change (Figure 3.10). The calculated parameters from the Generalized Extreme Value (GEV) distribution through the yearly maximum discharges were used to calculate the projected shifts in the probability of extreme events. According to the bias-corrected RACMO scenario (method 3), in 2050, a discharge with a return period of 1250 years in the control climate would occur once every 460 years. This is equivalent to an increased probability by a factor of 2.7. The RACMO (method 2) scenario projects that the 1250-year discharge will increase in frequency to once every 510 years (a factor of 2.5), and the Wp (method 1) scenario projects an increased frequency of once every 265 years (a factor of 4.7). When creating future climate scenarios, the delta change approach is more transparent than the bias-corrected RCM output. The delta change approach is also robust, allowing the use of climate model output even if they inaccurately represent the control climate (Graham et al., 2007). However, it can be concluded that the use of biascorrected RCM output is preferred in a climate change analysis, since that method incorporates projections of geographical differentiation, and includes changes to the variance of temperature, the coefficient of variation of precipitation, and the number of precipitation days. When identifying future problem areas and potential adaptation measures, it is of paramount importance to estimate spatial differences in flood-peak probabilities. However, the bias correction method has still room for improvement, in particular for extreme events. In addition, the use of a weather generator in combination with hydrological and hydrodynamic modeling is recommended when simulating low probability peak discharges under climate change.

6.2. Research questions

141

6.2.3

Future flood risk

What is the present basin-wide flood risk, how will it change up to 2030, and what is the relative contribution of climate and socio-economic change to this risk? Current flood risk in the Rhine basin (taking 2000 as reference year) was calculated by estimating both the flood probability and consequence of a flood event, expressed in potential direct damage. Future flood risk was simulated using scenarios to assess changes in flood probability due to climate change and by using land-use change scenarios for calculating changes in potential damage. These data combined provided different projections for future flood risk in 2030 for the entire Rhine basin. The projected change in land-use value put at risk by flooding is based on two different socio-economic scenarios derived from a land-use model. Potential damage is calculated by a damage model, and changes in flood probabilities were derived from two climate scenarios and hydrological modeling. It was found that, in absolute terms, potential flood losses are highest in the Dutch Delta region (G), namely e 110 billion, compared to e 80 billion of the second highest value in the Lower Rhine region (E) (for locations of the regions, see Figure 4.5). Flood risk (damage × probability) is, on the other hand, much higher in other regions, most notably in the Lower Rhine region (E) (e 350 million per year) and the Upper Rhine (C) (e 290 million per year), whereas the Dutch Delta region only reaches e 87 million per year. This thesis further projected that flood risk in the Rhine basin will not be stationary and might considerably increase over a period of several decades. Expected annual losses in the entire Rhine basin may increase by between 54% and 230%, due to socio-economic and climate change. The results displayed large variations in current risk and flood-risk projections between regions along the Rhine. The increase in flood risk can mainly be attributed to increasing probabilities of flood peaks due to climate change (43–160%, which is ∼ 6/8 of the total risk increase), whereas socioeconomic change accounts for 7.5–27% increase, which is ∼ 1/8 of the total risk increase (Figure 4.6). The method in this thesis provides a more comprehensive assessment of basin-wide flood risk than was previously possible for the Rhine, and is based on 13 land-use classes and sophisticated damage functions derived from the Damage Scanner (Klijn et al., 2007; Aerts et al., 2008b; Bouwer et al., 2010). Furthermore, the method integrates a land-use model with a damage model at a spatial resolution of 250 m, which enables basin-wide scenario projections for future land use and potential loss. However, both the land-use model and the damage scanner contain uncertainties that need further quantification.

142

Chapter 6. Synthesis and recommendations

6.2.4

Effectiveness of flood management measures

Which flood management measures are most effective in reducing flood stages and flood frequency on a basin-wide scale? As explained above, climate change may increase flood probabilities, which may strongly influence the expected annual losses. Therefore, it is important to test whether the implementation of flood defense measures, such as retention basins and dike heightening, might prevent the increase of flood probabilities due to climate change, and thus the flood risk. We evaluated the effectiveness of flood management measures included in the Action Plan on Floods (ICPR, 2005a), and additional measures at different return periods along the Rhine, assuming the relatively extreme W-plus climate change scenario for 2050 (Van den Hurk et al., 2006). These alternative measures include reforestation, restoration of abandoned meanders, implementation of extra retention polders, and a bypass around Cologne. All measures are aimed at reducing flood-peak probability and flood stages at a basin-wide scale. It was found that upstream flooding in the Upper and Lower Rhine in Germany has a profound diminishing effect on the peak water levels and peak discharges along the main Rhine branch and downstream in the Netherlands. The decrease varied between 2–13% in control climate conditions and 10–19% in the W-plus climate change scenario. Hence, upstream floods in Germany are favorable for reducing flood risk in the downstream areas of the Netherlands. However, it is possible that future flood policies in Germany will aim at raising their dikes, especially in a scenario with increased flood probabilities due to climate change. This may increase peak discharges and water levels downstream (e.g. in the Netherlands). The mean increase in peak water level along the Rhine due to climate change in 2050 is 50 cm, but varies between several centimeters and 137 cm. The currently implemented and proposed measures in the Action Plan on Floods, as well as most additional measures we evaluated reduce peak water levels by 5 cm to 13 cm over medium return periods (between 50 and 100 years) for the control climate. These simulations included the effects of potential upstream floods. As a result, these measures may be inadequate to cope with the increased flood probabilities that are expected in future climate scenarios. According to my results, the only measure that can prevent the Rhine from flooding in 2050 is drastic dike, basin-wide, heightening of between 1.30 m and 3.30 m, depending on location, with the assumption that these dikes cannot fail.

6.3. Implications and recommendations

143

6.3

Implications and recommendations

Below I will reflect on the implications of these results, mainly focused on the Netherlands. Since the geography and dike system of the Lower Rhine in Germany is very much alike, although maintained under a different law system, many remarks may also be relevant in that area. Recommendations are made for further work and for potential applications of the modeling approach. I discuss the consequences of ignoring the uncertainties in available data and methods on flood defense management decision making, and plead for a shift toward flood risk management that include uncertainties.

6.3.1

The consequences of ignoring the uncertainty of the design discharge

Although we have been able to decrease the statistical uncertainty in estimations of low-probability flood peak events, a residual uncertainty remains. For example, the impact of upstream flooding and climate change add to the lack of predictability of extremes. Estimating extreme events at high safety levels (i.e. at probabilities lower than 1/200 per year) requires either statistical or hydrological modeling far outside the calibrated range of those models. This introduces an unknown error and thus will always entail uncertainty. Increasing safety levels to reduce the probability of flooding will therefore inevitably go together with increased uncertainty in the estimation of accompanying peak discharge, no matter how far models have improved. In the Netherlands, flood management measures are designed to prevent coastal and fluvial flooding up to flood stages with a very low probability of occurrence. Safety standards vary between 1/10 000 per year in the low-lying coastal zones, to 1/1250 per year in the higher areas which are only threatened by fluvial flooding. These norms were established by the first Delta Committee in 1960 (Van Dantzig, 1956). However, the actual probability of flooding does not necessarily agree with these norms; it might be higher with higher failure probabilities, or lower if the dikes are taller than design level and do not fail. The design discharge associated with a safety standard of 1/1250 per year at Lobith is estimated to be 16 000 m3 /s. The related maximum water level determines the dike heights along the major rivers in the Netherlands (Ministry of Transport, Public Works and Water Management, 2006b). This discharge volume, however, is very uncertain, for three reasons. First, current methods to estimate extreme discharges only use a statistical extrapolation of 100 years of historical data for an event that happens once every 1250 years. Second, climate change and altered river geometry contribute to the non-stationarity of the system. Third, the impact of upstream

144

Chapter 6. Synthesis and recommendations

flooding is not considered in this extrapolation. These uncertainties are largely ignored in the Netherlands’ current flood management strategy and they are explained in more detail below. 1. Statistical extrapolation of historical data Substantial uncertainties are introduced when the design discharge is estimated only with historical data. The 95% confidence interval spans 13% more or less than the estimated design discharge (Figure 3.10), i.e., a 70 cm higher or lower maximum water level at Lobith. Although this uncertainty band is recognized and has been published previously (Silva et al., 2001, 1.8), the value of 16 000 m3 /s is chosen as the best estimate and therefore as the design discharge. Dike heights are derived from design water levels and 50 cm is added to account for wave run-up on dikes. In a guidance on dike design (not a law), the Ministry of Transport, Public Works and Water Management (2007) advices to add a robustness level of 30 cm to the dike height, to account for future changes and uncertainties. 2. Changing conditions In conjunction with the fact that 100 years of observation data is too brief to estimate the properties of a 1/1250 per year event, fitting an extreme value distribution through a historical data set implies that these data are stationary, i.e., that the physical conditions of the river system did not change. Obviously, over 100 years the geometry of the Rhine has changed and very likely also the meteorological conditions, and the principle of stationarity does not hold. Even after an attempt to ‘normalize’ the data, this adds to the uncertainty. Moreover, discharge behavior is likely to change in the future, as was established in this thesis by the assessment of climate change impact on flood peaks. At Lobith, the 1/1250 per year peak discharge of 16 000 m3 /s is projected to increase between 17 280 m3 /s (+ ∼ 40 cm) or 18 725 m3 /s (+ ∼ 90 cm) in 2050. 3. Upstream flooding One of the most remarkable results of this research is the influence of upstream flooding, adding to the uncertainty of the design discharge. Flooding in Germany and in the Upper Rhine in France lowers the flood stage up to 130 cm at the German-Dutch border, and a peak discharge of 16 000 m3 /s tops off to 14 000 m3 /s (Figure 5.8a). This effect is completely ignored in estimates of the design discharge at the current safety standard of 1/1250 per year (Ministry of Transport, Public Works and Water Management, 2006b), probably leading to a considerable overestimation. The Second Delta Committee (Deltacommissie, 2008), however, advised to apply a physical maximum discharge at Lobith when considering climate projections, of 18 000 m3 /s. The estimates of the physical maximum were provided by Vellinga et al. (2008) that contained some of my

6.3. Implications and recommendations

145

results of Chapter 5, although the Delta Committee reduced the effect upstream flooding to some extent. A reason to ignore the effect of upstream flooding in Dutch safety policy is that dikes in Germany might be raised in the future (Deltacommissie, 2008). The assumption that Germany will raise their dikes after a flood event is realistic, as was observed in the Elbe basin where flood protection was implemented after the floods in 2002 (Petrow et al., 2006). However, it does not seem realistic that safety standards will be raised from current levels of 1/200 to 1/500 per year up to Dutch standards (1/1250 per year). A costly flood of the Rhine in Germany does not fail the country as a whole, which is the risk in the Netherlands. For France, the flood prone area of the Rhine is even less important. Reasons for this are threefold. First, Germany governs many more large river basins, such as the Weser (substantial flooding in 2003), the Danube (large floods in 1999, 2002, and 2006, although most of the damage occurred in Slovakia, Bulgaria, and Romania), and the Elbe (large floods in 2002, with e 2 billion insured damage in Germany) (RMS, 2006). Second, in countries other than the Netherlands, the flood prone areas are surrounded by hills or mountains, so that the maximum extent of even the most extreme flooding is known and will not change in the future. Third, both in Germany and France homeowners and companies can insure their properties for flood damage (Kron, 2009). Therefore, raising the safety levels of the whole Rhine basin up to Dutch standards would probably be a huge over-investment in Germany and France. If the uncertainty around the estimation of the design discharge is neglected, instead of considering an unusually wet or dry year as part of a natural variation, the design discharge might change in response. This has occurred in the past, for instance shortly after the flood events in 1993 and 1995, the extreme value distribution was ‘lifted’ causing the design discharge to change from 15 000 to 16 000 m3 /s at Lobith. This had large implications for the Dutch policy on river management that continues through the present, which I will explain briefly below. If instead uncertainty bands had been taken into account, it would have been recognized that the flood peaks of 1993 and 1995 were well within those bands and one could argue (and maybe engineers and policymakers would have argued) that these two flood peaks do not justify a sudden adjustment of the design discharge and related safety standards for flood defenses and spatial measures. The 1993 and 1995 events also revealed that 650 km of dike length did not meet the safety standards (of 15 000 m3 /s at Lobith) and needed immediate reinforcement. The ‘Deltaplan’ was quickly enacted (Ministry of Transport, Public Works and Water Management, 1995) and has been completed in 2000. Furthermore, it was decided to compensate for the increase of 1000 m3 /s in the design discharge by creating extra room for the river wherever possible, avoiding the need for dike heightening (Ministry of Transport, Public Works and Water Management, 2006c). Although I would cri-

146

Chapter 6. Synthesis and recommendations

ticize the decision for the sudden shift in design discharge, the ‘Room for the River’ concept is a very interesting and daring diversion from the former policies of continuing dike enforcement and heightening. It has been recognized that increased dike heights leads to increased maximum water levels and thus to a potentially larger flood hazard. Therefore, the objective for making room for the river is to lower peak water levels. The measures under consideration include deepening of low flow channels and flood plains, removing hydraulic structures, dike replacements, and lowering of groins. The impact on the maximum water level of these measures has been studied at a very detailed level, on the order of millimeters (Ministry of Transport, Public Works and Water Management, 2006c). However, it seems contradictory to plan and implement measures accurately to the millimeter based on a fixed design discharge and water level, and to ignore the huge uncertainty in the design water level of ± 70 cm due to statistical uncertainty. On top of that, there are the uncertainties related to changing conditions (non-stationarity) and upstream flooding. In my view, the design discharge and accompanying design water levels are wrongly communicated as certain values, fixed for as long as no extreme event occurs, while these two preconditions are in fact very uncertain. As a result, enormous efforts are invested in maintaining and designing the Dutch dike system and landscape according to these uncertain preconditions, while different design values would probably lead to very different decisions.

6.3.2

How to implement uncertainty in the calculations of the design discharge

Acknowledging the uncertainties, and to use them in a sensitivity analysis of the effects and costs of proposed measures and alternative strategies is important. The potential explosion of the number of required simulations by the already very detailed models may however be a constraint. A stochastic modeling setup that considers the uncertainties of various modeling steps, including the effects of upstream flooding, along with a more flexible interpretation of the results, might bring some relief. Instead of assuming that there will be no upstream flooding, it may be better to gain a better understanding of the hydraulic behavior of potential flooding events in the Lower Rhine. Flooded water in the German Lower Rhine might enter the Netherlands via other routes, for example, from Emmerich to Doetinchem, or from Kleve to Nijmegen (Lammersen, 2004; Silva et al., 2006). Due to lower design levels along the German Rhine branch, the probability of flooding via this ‘back door’ seems higher than a riverine flood event in the Netherlands only. More detailed research using well-founded inundation models is necessary to explore ways to cope with such a flood event. Experience from the Dutch compartment study (Asselman et al., 2008) might be beneficially applied to the border area between Germany and the Netherlands.

6.3. Implications and recommendations

147

With cross-boundary cooperation, already existent in the Arbeitsgruppe Hochwasser (Raadgever et al., 2008) and the International Commission for the Protection of the Rhine (ICPR), the development of 2D hydrodynamic models, flooding scenarios and measures can be taken up. I argued in this thesis that process-based modeling, which considers spatial and temporal meteorological variation and the physical properties of the river basin, is a better method for estimating extreme peak discharges than is the extrapolation of historical data. Resampling and process-based modeling can handle the large variety of flood wave shapes for extreme events, and the modeling results include a description of uncertainty. In addition, the method can simulate upstream flooding, and enables testing and visualization of the effectiveness of measures using a probabilistic approach. Based on my research the applied modeling method has been partly implemented for the Meuse in a tool named GRADE (De Wit and Buishand, 2007), with the goal of testing an alternative method for estimating design discharges. It also uses long resampled meteorological time series as input for the hydrological model, and a hydrodynamic model to simulate all yearly flood peaks. However, it does not incorporate upstream flooding, the impact of flood management measures, or climate change scenarios.

6.3.3

Lowering peak water levels effective at local scale

Another imperative result of this thesis is that, with the Rhine already heavily embanked and canalized, adding new retention basins will not substantially lower maximum water levels downstream. This is especially true since safety levels that determine dike height and operational rules for retention areas are different throughout the basin. It is therefore not possible for downstream areas, like the Netherlands and the German federal state North Rhine-Westphalia, to derive much benefit from extra retention areas located more than ∼ 50 km upstream, let alone in the Upper Rhine. Historical flood events exemplify this observation (Volkskrant, 1995, Becker, personal communication), when the decisions to inundate retention basins were solemnly diverted to local needs in the Upper Rhine. However, flood management measures, including extra retention basins or land-use changes to detain water in small tributaries, can be beneficial at a local scale. This thesis demonstrated that a bypass around a valuable area, such as a city, can also be very effective (Figure 5.7). In practice, this would mean differentiation at the local scale of flood protection levels. Rural areas will flood, while the city is protected. This relates to the controlled flooding of retention areas. In the Netherlands, the implementation of retention polders Ooijpolder and Rijnstrangen (Commissie Noodoverloopgebieden,

148

Chapter 6. Synthesis and recommendations

2002) was prevented by the public (Roth et al., 2006), while in Germany, people have been less opposed to controlled flooding (Ouwendijk et al., 2001). Protecting a city using a large bypass might be perceived differently by policy makers and the public than controlled flooding.

6.3.4

Risk-based approach can be beneficial basin-wide

According to the calculations in this thesis, the only measure that can prevent the Rhine from flooding from a 1/1250 per year event in 2050 is drastic, basin-wide, dike heightening of between 1.30 m and 3.30 m, depending on location, with the assumption that these dikes cannot fail. However, there are several reasons why such continuous basin-wide dike raise as the solution against potential flooding from the Rhine may not be feasible. First, assuming that a considerable dike raise is not blocked from a landscape value point of view, in many places along the German Rhine it is simply physically impossible (Lammersen, personal communication), since dike heightening implies substantial dike widening. Second, the risk of a flood hazard can never be ruled out, even when very high safety levels are maintained. A residual risk maintains that an extreme event of a proportion that has not been considered to date might still occur. The historical description of the very extreme and devastating Rhine flood in 1374 exemplifies this, with estimates of peak discharge volume at Cologne ranging from 18 000 to 24 000 m3 /s, based on historical descriptions and flood marks on buildings (Sprong, 2009; Herget and Meurs, 2010). This would imply even higher discharges at Lobith, while in this thesis, the most extreme estimated peak discharge for the 1/1250 per year event under 2050 climate conditions reaches 18 725 m3 /s at Lobith. Third, the assumption that dikes, as they are currently designed along the Rhine, cannot fail seems unrealistic, as the FLORIS study demonstrated for the Netherlands (Ministry of Transport, Public Works and Water Management, 2006a). The FLORIS study estimated a substantial number of dikes in the Netherlands that do not meet their design safety standards. To estimate actual risks it is important that dikes are at least able to withstand the water levels they are designed for. A quick scan on ‘unbreakable dikes’, or dikes that can flood, but will not fail due to their substantial width and robust surface material, displayed that the implementation of such dikes seems physically and economically feasible in the Netherlands (Silva and Van Velzen, 2008). The ability of a river system to recover from floods using controlled flooding and damage reduction is referred to as resilience of the system (Klijn et al., 2004). Instead of indiscriminately raising dikes, they could be designed as ‘unbreachable’ and maintained at levels determined using a flood risk approach that considers the conse-

6.3. Implications and recommendations

149

quences of flooding, changing physical properties of the river basin, natural variability and (future) probabilities of meteorological extremes. The advantages of resilience in flood risk management for the Netherlands are explained by De Bruijn (2005) and Vis et al. (2003). Several studies have examined the impacts of flooding, flood risk, and the positive effect of hazard reduction measures in the Netherlands (Ministry of Transport, Public Works and Water Management, 2006a; Aerts et al., 2008b). Nevertheless, recent findings of the (second) Delta Committee (Deltacommissie, 2008) in the Netherlands have focused mainly on protection against flooding. These findings are formalized into a ‘delta law’, and will form the basis for a new water safety policy. Due to an increase in economic development and potential fatalities since 1960, the Committee has advised an increase in safety standards. Current safety standards are outdated according to the Committee, but as explained in this thesis, increasing safety norms has the effect of increasing uncertainties of the properties of the then lowerprobability design events. A combination of flood defense measures, flood hazard mapping and risk zoning, damage reduction, spatial planning, and early warning would be more appropriate. It has been argued before that an effective climate change adaptation policy in the Netherlands should not only address the reduction of flood probabilities with flood defense measures, but should also consider a wide range of other adaptation options (Aerts et al., 2008b). Damage reduction can be achieved through controlled flooding and measures that reduce the vulnerability of buildings to flooding. In addition, financial arrangements such as insurance can compensate for losses and heighten flood risk awareness (Botzen and Van den Bergh, 2008; Botzen, 2010; Bouwer, 2010). The optimum combination of measures can be evaluated using portfolio theory, as is common in financial services (Aerts et al., 2008a). The EU Flood Directive (EU, 2007a) (see Chapter 1) might help to force a transition in all member states from a policy of flood control to the acceptance that flood risks, and thus consequences of flooding, should be managed. The considerable potential for damage in flood prone areas in the Rhine basin, coupled with projections for substantial future growth, as calculated in this thesis (Figure 4.7), might help to concentrate more on this aspect of flood risk management, especially in the Lower Rhine and the Netherlands. Available data on direct and indirect damages are literally more down-to-earth than estimates of flood stages and flood probabilities at extremely high protection levels; however, studies of models simulating inundation and damage scenarios reveal that there is still considerable uncertainty associated with flood hazard and risk estimates (e.g. Apel et al., 2004; Merz and Thieken, 2009; De Moel and Aerts, 2010). These uncertainties in flood inundation and damage projections are receiving increasing attention in research (e.g. Merz et al., 2004; Messner et al., 2007; Merz et al., 2010).

150

Chapter 6. Synthesis and recommendations

6.3.5

Toward communicating uncertainty and adaptive management

Policymakers are positively capable of acting in the face of uncertainty (Pappenberger and Beven, 2006). Unfortunately however, large policy shifts are often reactive (Huitema and Meijerink, 2010), as multiple examples for both natural hazards and human failures show. It seems without doubt that uncertainty will continue to complicate decision-making and policy-making processes in cross-boundary river basin management. Although uncertainty in predictions might appear unsettling, it is an integral component of our understanding, and it aids our advancement to more holistic knowledge (Ivanovi´ and Freer, 2009). c Therefore, known uncertainty should not be hidden by engineers, but communicated clearly and truthfully to policymakers and the general public (Webster, 2003). One option for exploring an uncertain future is scenario testing of management decisions that allows for participatory modeling; engages stakeholders, policy makers, and experts; and improves the dialog between them (Evans et al., 2004; Schanze et al., 2008). It has been recognized that major floods often trigger policy changes due to public indignation (Samuels et al., 2006). This reactive behavior should be transformed into proactive behavior and strategies of adaptation. Pahl-Wost (2008) explains how iterative learning circles incorporated into overall management are an essential element of adaptive management. An example is the EU Flood Directive, which requires a review, and if necessary an update, of the flood hazard and risk assessment every six years (EU, 2007a). The future contains inherent uncertainty due to unknown economic growth, political stability, population growth, climate change, etc. The process of decision-making with respect to the uncertain predictions of these aspects is complicated and heavily politicized (Ivanovi´ and Freer, 2009). But since policymakers are capable of acting c on uncertainty, they need only start taking action to adapt to projected changes. In adaptive water management, inherent uncertainties are embraced, certainties and risks emphasized, and new management measures continuously reformulated after learning from previous experiences.

6.4

Remaining challenges

This thesis has provided answers to the research questions posed and the described approach is a useful tool for estimating the effects of climate change, socio-economic development, and measures on extreme flood peaks and flood risk in the Rhine basin.

6.4. Remaining challenges

151

However, during the research process there emerged possible areas for future improvement. These remaining challenges are discussed in this final section of the thesis. Recent findings show that Alpine temperatures during the past century have increased more than twice the global average (EEA, 2009). This makes Alpine mountains especially vulnerable to changes in the hydrological cycle, and decreases in snow and glacier cover are already occurring. Future research should focus on the timing of flood peaks from the Alps, which seems to advance toward the beginning of the year, in relation to the timing of flood peaks from other tributaries. If these peaks coincide in the future, flood-peak probabilities in the Lower Rhine will further increase. During the summer, 90% of the river discharge originates from the Alps (Pinter et al., 2006). Drought and low flows deserve more attention in the Rhine basin. The apparent vulnerability of the Alps to climate change underlines these potential problems (EEA, 2009). Low flows can have consequences for shipping, availability of cooling water for power plants, agriculture, and drinking water, and the yearly expected loss can be considerable (Van Beek et al., 2008) when compared to flood risk (Te Linde et al., 2011). Land-surface and atmospheric interactions are more relevant for climate impact modeling on droughts than on floods (Seneviratne et al., 2006). Land-surface models such as applied by Hurkmans (2009) should therefore be further developed to aid research on low flows in the Rhine basin. To improve the assessment of the impact of flooding, multiple two-dimensional hydrodynamic simulations are required in the Rhine basin. In Lammersen (2004), several 2D inundation runs are described for the Lower Rhine; however, to our knowledge, no 2D inundation model exists for the flood prone areas of the Upper Rhine. Also, future projections of morphological changes in the main Rhine channel and its flood plains require more research, since morphology has a profound effect on flood stages (Van Vuren, 2005).

152

Chapter 6. Synthesis and recommendations

Chapter

7

Summaries

7.1

English summary

In the past decades, the number of fatalities and the economic loss caused by floods have worldwide increased considerably. The river Rhine in North-West Europe also has a long history of floods that have caused casualties and severe damage. Despite flood defenses improved over time, it is expected that flood risk in the Rhine basin will continue to increase, which can be attributed to both socio-economic developments in flood prone areas, as well as increased flood frequency as a result of climate change. Climate change projections for the Rhine basin show that flood peaks may increase due to increased precipitation and earlier snow melt, and peak discharges are likely to shift from spring to winter and to increase in volume and frequency. Socio-economic projections showing trends in population growth and wealth indicate that the exposure to floods in the Rhine will increase as well. The combination of both trends will increase the flood risk. Assessing the effect of climate change on flood risk, and maintaining and planning flood management measures are urgent issues for the riparian countries within the Rhine basin. Since safety levels along the Rhine are relatively high (1/200 to 1/1250 per year), the estimation of the size and duration of flood peaks at these low probabilities is very relevant to water managers. However, simulating those low-probability

153

154

Chapter 7. Summaries

floods requires extrapolation of only ∼110 years of measured data, which introduces large uncertainties. Many methodological challenges remain on simulating the effect of climate change and flood management measures on runoff, particularly when estimating the probability of occurrence of extreme flood peaks. In addition, no future basin-wide flood damage estimates exists to reflect socio-economic and land-use change. This thesis investigates the combined effect of climate change and socio-economic projections on flood risk in the Rhine basin using improved simulation methods. The focus is on the estimation of low-probability flood events and anticipating to the impacts of these events through the development of cross-boundary flood management measures. Results show that improvements of the simulation method for the Rhine basin are not as much found in a more detailed description of hydrological processes, but in carefully coupling different methods and models needed to simulate low-probability flood events. For example, it is recommended to improve the estimates of the intensity and duration of extreme meteorological events by applying a weather generator to generate long time series of climate data, and to implement a better description of routing processes in flood peak generation by using a hydraulic model. Furthermore, the impacts of climate change and the alterations in river geometry and land use due to human interventions are parameterized in models, in an attempt to physically describe extreme situations and the consequences of changed conditions. In doing so, I reject the often applied assumption of stationarity of water systems and apply an approach which can be referred to as ‘process-based modeling’. The thesis compares two hydrological models, and the results show the conceptual model HBV simulates peak discharges best and is therefore preferred over the more complex land-surface model VIC for the hydrological simulation of climate change scenarios in the Rhine basin. As input for the HBV model, a weather generator is used to create long series of resampled meteorological data (up to 10 000 years of daily values), which considerably decreases the statistical uncertainty within extreme value analysis. As a result, for low-probability (1/200 per year, or less) flood peak discharges, the statistical uncertainty (illustrated by the 95% confidence intervals) decreased from ± 13% to ± 3%. To estimate the impact of climate change on river discharge, the output of Global Climate Models (GCMs) is used as scenario input for HBV. Different downscaling methods were applied to convert data from GCMs to the required regional scale for hydrological modeling in the Rhine basin. From this comparison, it appears that bias-corrected direct Regional Climate Model (RCM) data is preferred over the delta approach, since the first method better describes spatial and temporal variation in the meteorological events, which are relevant for the planning of flood management measures. When using the downscaled data in HBV, the simulation results indicate

7.1. English summary

155

a substantial increase in flood-peak probability in 2050 throughout the Rhine basin, when compared to the control climate period of 1961–1995. At gauging station Lobith, at the German-Dutch border, the probability of flooding (currently the safety level is set at 1/1250 per year) is expected to be three to five times as high in 2050 as it is today. Furthermore, a basin-wide flood damage model was developed to estimate current and future potential flood losses. These damage estimates can be combined with flood probability to estimate the annual expected loss (i.e. flood risk) in the Rhine basin. It appears the annual expected loss may increase between 54% and 230% in 2030 compared to 2000, of which ∼ three-quarters of the increase can be attributed to climate change, and the remaining part to socio-economic developments. Results show that the area with the highest flood risk is located in the Lower Rhine in NordrheinWestfalen in Germany, and not in the Netherlands as previously believed. This is mainly related to the high safety standards in the Netherlands. In the current situation, safety levels of dikes and other flood defense structures decrease in upstream direction along the Rhine. The differences result from different flood management policies between countries and are roughly related to economic value in the protected flood prone areas. Consequently, flooding is more likely upstream in Germany and France (where safety levels vary between 1/200 to 1/500 per year), compared to the Netherlands (which has a safety level of 1/1250 per year). It has been shown that flooding in the Upper and Lower Rhine in Germany, upstream of Lobith, has a profound decreasing effect on discharge peaks at Lobith. The decrease varied between 2–13% in control climate conditions and 10–19% in the W-plus climate change scenario, at return periods of 50 years and more. The mean increase in peak water level along the Rhine due to climate change in 2050 is 50 cm, but varies between several centimeters and 140 cm. These simulations included the effects of potential floods in the Upper and Lower Rhine. Various flood management measures in the Rhine basin have already been developed according to the Action Plan on Floods (APF) that was initiated in the 1990s by the International Commission for the Protection of the Rhine (ICPR). The currently implemented and proposed measures in the APF, as well as most additional measures that were evaluated, such as extra retention areas, reduce peak water levels by only 5 cm to 13 cm over medium return periods (between 50 and 100 years). As a result, these measures seem inadequate to cope with the increased flood levels that are expected in future climate scenarios. According to the results of this thesis, the only measure that can prevent the Rhine from flooding at current safety levels in 2050, is drastic, basin-wide, dike heightening of between 1.30 m and 3.30 m, depending on location, on the assumption that these dikes cannot fail. Although this research has been able to decrease the statistical uncertainty in estimations of low-probability flood peak events, a residual uncertainty remains. Due to the

156

Chapter 7. Summaries

too short time series of available discharge measurements (∼110 years), estimating extreme events at high safety levels (1/200 per year or less) requires either statistical or hydrological modeling far outside the calibrated range of those models. This introduces an unknown error and thus will always entail uncertainty. In addition, the impact of upstream flooding and climate change adds to the unpredictability of extremes. Increasing safety levels to reduce the probability of flooding, which is currently considered in the Netherlands, will force engineers to even further stretch their models. This will thus increase uncertainty in the estimation of the accompanying peak discharge, no matter how far models have improved. Water managers should be more aware of this fact that increasing safety levels will entail increased uncertainty of the related, simulated design discharge. An effective climate change adaptation policy on flood risk management should embrace inherent uncertainties, and not only address flood defense measures, but also consider a wide range of other adaptation options on damage reduction.

7.2

Nederlandse samenvatting

Wereldwijd is de economische schade en het aantal slachtoffers door overstromingen de laatste decennia aanzienlijk gestegen. Ook de Rijn heeft een lange geschiedenis van overstromingen met veel schade en slachtoffers. Bewoners langs de rivier beschermden zich tegen het water door de rivier te kanaliseren en dijken te bouwen, die vaak na een nieuwe overstroming verbeterd werden. De verwachting is dat het overstromingsrisico in het stroomgebied van de Rijn, dat wil zeggen de kans op een overstroming vermenigvuldigd met het gevolg, zal blijven toenemen. Dit kan worden toegeschreven aan zowel sociaal-economische ontwikkelingen in kwetsbare gebieden, als de toenemende overstromingsfrequentie door klimaatverandering. Klimaatscenario’s voor het stroomgebied van de Rijn laten zien dat de neerslag in de winter zal toenemen, met een hogere afvoer tot gevolg. Door globale opwarming zullen afvoerpieken als gevolg van sneeuwsmelt eerder in het jaar plaats vinden dan dat nu het geval is. Bij elkaar genomen resulteert dit in een hogere frequentie van extreem hoge afvoeren. Daarnaast vertonen sociaal-economische scenario’s voor het Rijnstroomgebied stijgende trends in bevolkingsdichtheid en welvaart, waardoor de kwetsbaarheid van de maatschappij voor overstromingen verder toeneemt. Voor het ontwikkelen van hoogwaterbeschermende maatregelen is het noodzakelijk om het het effect van klimaatverandering op de afvoer van de Rijn goed te kennen. Doordat de veiligheidsniveau’s langs de Rijn relatief hoog zijn (1/200 tot 1/1250 per jaar) is kennis over de grootte en duur van hoogwaterpieken met een dergelijk lage kans van voorkomen relevant voor waterbeheerders. Echter, het schatten van de eigenschappen van deze hoogwaterpieken gebeurt nu door extrapolatie van slechts

7.2. Nederlandse samenvatting

157

∼110 jaar aan beschikbare meetgegevens, wat resulteert in grote onzekerheden. Er zijn nog meer methodologische obstakels wanneer we het effect van klimaatverandering en hoogwaterbeschermende maatregelen op de afvoer willen bepalen, in het bijzonder wanneer het gaat om extreem hoge afvoeren met een zeer lage kans op voorkomen. Zo bestaan er nog geen stroomgebiedsbrede schattingen in het Rijnstroomgebied over hoe de potenti¨le schade zal toenemen door sociaal-economische ontwikkelingen. e Dit proefschrift onderzoekt het gecombineerde effect van klimaatverandering en sociaaleconomische ontwikkelingen op het overstromingsrisico in het stroomgebied van de Rijn met de nadruk op bovenstroomse gebieden, waarbij beschikbare simulatiemethoden waar nodig verbeterd worden. De focus ligt op het simuleren van extreem hoge afvoeren en op het ontwikkelen van grensoverschrijdende, hoogwaterbeschermende maatregelen. De resultaten laten zien dat het verbeteren van de simulatiemethode niet zozeer zit in een meer gedetailleerde beschrijving van hydrologische processen, maar in het slim combineren van deels nieuwe en bestaande methoden en modellen. Ten eerste wordt geadviseerd om de intensiteit en duur van zeldzame meteorologische situaties, bijvoorbeeld extreem hoge en langdurige neerslag, beter te simuleren. Dit kan met behulp van een zogenaamde weergenerator die op basis van statistische parameters lange synthetische meteorologische reeksen genereert. Ten tweede is het van belang de routing processen van afvoergolven beter te simuleren door een hydraulisch model toe te passen. Een vaak gebruikte aanname in studies naar extreme afvoeren is dat riviersystemen stationair zijn. Deze aanname wijs ik af en ik pas een methode toe die beschreven kan worden als ‘proces-geori¨nteerd modelleren’. In deze verbeterde simulatiemethode e wordt zowel de invloed van klimaatverandering als de door de mens veroorzaakte veranderingen in het landgebruik en de geometrie van de hoofdgeul, geparameteriseerd in modellen. Op deze manier kan het effect van deze veranderingen op het afvoergedrag van de Rijn door de tijd heen gesimuleerd worden. Een modelvergelijking laat zien dat het conceptuele hydrologische model HBV piekafvoeren van de Rijn beter beschrijft dan het meer complexe model VIC, en daarom is HBV gebruikt als hydrologisch model in dit proefschrift. De lange meteorologische tijdreeksen uit de weergenerator (tot 10 000 jaar dagwaarden) worden in het HBV model ingevoerd, dat vervolgens 10 000 jaar afvoerdata simuleert. Als gevolg hiervan wordt de extreme waarden analyse nauwkeuriger dan wanneer alleen de 110 jaar aan historische gegevens gebruikt wordt. Dat de statistische onzekerheid (aangegeven door het 95% betrouwbaarheidsinterval) bij het schatten van extreme waarden met een lage kans van voorkomen (1/200 per jaar of lager) wordt hierdoor verlaagd van ± 13% naar ± 3%.

158

Chapter 7. Summaries

Om het effect van klimaatverandering op de afvoer te berekenen, is uitvoer van globale klimaatmodellen (GCMs) gebruikt als invoer voor HBV. Om de meteorologische gegevens van GCMs terug te brengen naar de vereiste regionale schaal, prefereer ik het gebruik van directe uitvoer na bias-correctie van regionale klimaatmodellen (RCMs) boven de zogenaamde ‘delta methode’. RCM uitvoer geeft namelijk beter inzicht in ruimtelijke en temporele variatie van het klimaatsignaal, wat relevant is voor het plannen van hoogwaterbeschermende maatregelen. De simulatie resultaten wijzen op een aanzienlijke toename van de overstromingskans in 2050 in het gehele stroomgebied van de Rijn, vergeleken met de referentieperiode van 1961–1995. Op de Duits-Nederlandse grens, bij Lobith, is de huidige overschrijdingsnorm 1/1250 per jaar. Deze kans zal naar verwachting in 2050 drie tot vijf keer zo hoog zijn. Voorts wordt geraamd dat de jaarlijks verwachte schade (oftewel het overstromingsrisico) in het stroomgebied van de Rijn tussen de 54% en 230% toeneemt in 2030 ten opzichte van 2000. Hiervan kan ongeveer driekwart van de toename toegeschreven worden aan het effect van klimaatverandering (toenemende overstromingskans), en de rest is het gevolg van sociaal-economische ontwikkelingen, zoals verstedelijking. Resultaten tonen aan dat het gebied met het hoogste overstromingsrisico gelegen is in de Benedenrijn in Nordrhein-Westfalen in Duitsland, en niet in Nederland zoals vaak wordt gedacht. Dit is vooral het gevolg van verschillende veiligheidsnormen van de waterkeringen in beide gebieden: in Nederland is deze norm relatief hoog waardoor het risico laag is. In de huidige situatie neemt het veiligheidsniveau van de dijken en andere kunstwerken af langs de Rijn in stroomopwaartse richting. De verschillen vloeien voort uit verschillen in waterbeheers tussen verschillende landen en Duitse Bundesl¨nder, en zijn a grofweg gerelateerd aan de economische waarde in overstroombare gebieden achter de dijken. Het resultaat is dat de kans op overstromen in Duitsland en Frankrijk (waar de veiligheidsniveaus geschat worden op 1/200 tot 1/500 per jaar) hoger is dan in Nederland (waar de veiligheidsnorm in het bovenrivierengebied 1/1250 per jaar is). Als gevolg hiervan vinden er overstromingen plaats in Frankrijk en Duitsland, bij een afvoer die lager is dan de ontwerpafvoer in Nederland. Dit heeft een vergaand verlagend effect op de piekafvoeren bij Lobith. De afname van de piekafvoer varieert tussen 2–13% in de referentie situatie en tussen 10–19% in het W-plus klimaatscenario in 2050, bij herhalingstijden van 50 jaar en hoger. De gemiddelde verhoging van de maximale waterstand tijdens piekafvoeren langs de Rijn in 2050 is 50 cm als gevolg van klimaatverandering, maar varieert tussen enkele centimeters en 137 cm. In deze simulaties is het effect van bovenstroomse overstromingen in de Boven- en Benedenrijn in Frankrijk en Duitsland meegenomen. Het in de jaren ’90 door de ICBR (Internationale Commissie ter Bescherming van de Rijn) ge¨ ınitieerde Actieplan Hoogwater bevat verschillende hoogwaterbeschermende maatregelen die momenteel worden uitgevoerd en gepland zijn tot aan 2020. De

7.2. Nederlandse samenvatting

159

resultaten in dit proefschrift laten zien dat zowel de maatregelen in het Actieplan Hoogwater als verscheidene extra maatregelen, zoals meer retentiegebieden, de maximale waterstanden met slechts 5 cm tot 13 cm verlagen voor hoogwatersituaties met middelgrote herhalingstijden tussen de 50 en 100 jaar. De conclusie is dat deze maatregelen ontoereikend lijken om te voorkomen dat de kans op overstromen als gevolg van klimaatverandering toeneemt. Volgens de resultaten in dit proefschrift is de enige maatregel die kan voorkomen dat de Rijn vaker overstroomt in de toekomst een drastische dijkverhoging tussen de 1,30 m en 3,30 m, afhankelijk van de locatie, in de veronderstelling dat deze dijken niet kunnen falen. Hoewel door dit onderzoek de statistische onzekerheid verlaagd is bij het schatten van dergelijke piekafvoeren, zal er altijd een restonzekerheid door overige factoren zijn. Doordat de afvoer meetreeksen te kort zijn (∼110 jaar) vereist het schatten van extreme gebeurtenissen (1/200 per jaar of lager) hetzij statistische, hetzij hydrologische modellering ver buiten het gekalibreerde bereik van deze modellen. Dit introduceert een onbekende fout en zal dus altijd een bepaalde onzekerheid bevatten. Daarnaast draagt het effect van bovenstroomse overstromingen en klimaatverandering bij aan de onvoorspelbaarheid van extreem hoge afvoeren. Het verhogen van de veiligheidsnormen om de overstromingskans te verlagen, zoals nu wordt overwogen in Nederland, zal ingenieurs dwingen om hun modellen nog verder op te rekken. Dit verhoogt dus de onzekerheid omtrent het schatten van de bijbehorende ontwerpafvoer, ongeacht hoe goed de gebruikte modellen zijn. Waterbeheerders moeten zich meer bewust zijn van dit feit dat het verhogen van de veiligheidsnormen inhoudt dat de onzekerheid van de gesimuleerde maatgevende omstandigheden toeneemt. Een effectief klimaatadaptief beleid omtrent overstromingsrisicobeheer zal de inherente onzekerheden moeten omarmen, en niet alleen hoogwaterbeschermende maatregelen moeten overwegen, maar ook een breed scala aan adaptatiemaatregelen die de potenti¨le schade van overstromingen beperken. e

160

Chapter 7. Summaries

Appendix

A

The HBV model

The HBV model is a conceptual semi-distributed model, which is developed by Bergstr¨m in the early 70s (Bergstr¨m, 1976). HBV describes the most important o o runoff generation processes in a simple and robust way. The model requires precipitation, temperature and potential evaporation as input data, and consists of three routines: • Snow routine Initial precipitation is divided into rainfall and snowfall, depending on temperature. This process is ruled by a threshold temperature (parameter tt) below which precipitation is supposed to be snow. The transition from rain to snow can be realised continuously over a temperature interval (parameter tti ). Snow melt computations are based on a day-degree relation (snow melt factor cfmax ). The snow distribution is computed separately for different elevation and vegetation zones in the basin. • Soil routine The soil routine controls how much water evaporates and which part of the precipitation is direct runoff, or is stored in the soil. The runoff coefficient depends on the ratio of actual soil moisture, the maximum water storage capacity of the soil (parameter fc), and an exponent representing drainage dynamics (parameter beta). The parameter lp defines the water storage in the soil at which

161

162

Appendix A. The HBV model

actual evaporation becomes equal to potential evaporation. There is a special correction factor for evaporation in forest areas (cevpfo). Interception in forest areas and open land are also simulated (parameters icfo and icfi ). • Runoff generation routine This routine is the response function which transforms excess water from the soil routine to runoff. The routine consists of one upper, non-linear reservoir (parameters khg, hq and alpha) and one lower, linear reservoir (recession coefficient k4 ). The upper one represents direct runoff. The lower reservoir represents the base flow which is fed by groundwater. Groundwater recharge is ruled by a maximum amount of water that is able to penetrate from the soil to groundwater (parameter perc). Timing and distribution of the resulting runoff is further modified in a transformation function by means of a retention parameter (maxbas). This routine is a filter technique with a triangular distribution of the weights as shown in Figure A.1 at the bottom on the right (Bergstr¨m, o 1992).

163

Pinput
pcorr

Ssnow
snowmelt = cfmax · (T - tt)

tt, tti, T
tt, cfmax, T

Pyield
EW

icfo, icfi

SIc

EIc

Pyield = Pinput ⋅ pcorr − ( ∆S snow + ∆S Ic ) / ∆t

Pyield-EW

EW =

( (
S sm fc

beta

fc, Ssm, beta

⋅ Pyield

Suz
R

cflux

Ssm

lp, fc, Ssm, Eref

Eact

E act =
perc

E act

S sm ⋅ E ref when Ssm < lp · fc lp ⋅ fc = E ref ≥ when Ssm = lp · fc

(Suz)
khq, hq, alpha

Slz
k4

R = perc R = S uz / ∆t

when Suz ≥ perc · t = when Suz < perc · t

Q fast = k ⋅ (S uz )

( 1+ alpha )

Qfast

Qslow
Qfast + Qslow weight

Qslow = Slz ⋅ k 4
Qout

k = F (khq, hq, alpha)

maxbas
time time time

MAXBAS

Qout
Input Data: Pinput = recorded precipitation data T = air temperature Eref = reference evapotranspiration Pyield Ssnow SIc EIc Ssm Eact Suz = available water yield = snow storage = interception storage = interception evaporation = soil moisture storage = actual evapotranspiration = storage of upper zone (Suz) Slz Qfast Qslow Qout EW R = remaining Suz = storage of lower zone = fast runoff component = slow runoff component = computed runoff = excess water = groundwater recharge

model parameters in italics

Figure A.1: Illustration of discharge formation in the HBV model.

164

Appendix A. The HBV model

Appendix

B

The SOBEK model

SOBEK is a one-dimensional dynamic numerical modeling system for open-channel networks. It can be used to simulate unsteady and steady flow, salt intrusion, sediment transport, morphology and water quality. The model can be applied for flood forecasting, optimization of drainage systems, control of irrigation systems, sewer overflow design, ground-water level control, river morphology, salt intrusion and surface water quality. SOBEK is developed by WL | Delft Hydraulics (now Deltares) and the Institute for Inland Water Management and Waste Water Treatment (RIZA) (now Waterdienst), in the Netherlands (Delft Hydraulics, 2005). The water flow is computed by solving the 1-D cross-sectionally integrated shallowwater equations (St. Venant equations), representing the conservation of cross-sectionally integrated mass and momentum. These equations are: ∂A ∂Q + = qlateral ∂t ∂x ∂Q ∂ + ∂t ∂x Q2 Af ∂ (h + zb ) τb + Af =0 ∂x ρR (B.1)

αb

+ gAf

(B.2)

where A is the total cross-sectional area (m2 ), Af the cross-sectional flow area (m2 ), Q the discharge (m3 /s), qlateral the discharge added to the river per unit length (m2 /s),

165

166

Appendix B. The SOBEK model

αb the Boussinesq constant (-), g the acceleration due to gravity (m/s2 ), h the water depth (m), zb the bed level [m], ρ the mass density of water (kg/m3 ), τb the bed-shear stress (kg/m/s2 ) and R the hydraulic radius (m). Time and space are represented by t and x, respectively. The bed-shear stress τb is expressed by the Ch´zy formula: e τb gQ |Q| = 2 2 ρ C Af where C the Ch´zy coefficient (m1/2 /s). e (B.3)

Appendix

C

The GEV distribution

The generalized extreme value (GEV) distribution is the distribution of normalized maxima of a sequence of independent and identically distributed random variables. Because of this, the GEV distribution can be used as an approximation to model the maxima of long sequences of random variables. Three types of extreme value distribution functions (F (x)) are combined into the GEV distribution: x−γ F (x) = exp − 1 + β δ
1 β

(C.1)

where γ is the location parameter, δ is the scale parameter, and β is the shape parameter. The shape parameter governs the tail behavior of the distribution. When β = 0, the GEV corresponds to the type I (Gumbel) distribution; β < 0 corresponds to the type II (Fr´chet) distribution; and β > 0 corresponds to the type III (Weibull) e distribution.

167

168

Appendix C. The GEV distribution

Appendix

D

Performance indicators

Coefficient of efficiency (E)
The efficiency E proposed by Nash and Sutcliffe (1970) is defined as one minus the sum of the absolute squared differences between the predicted and observed values, normalized by the variance of the observed values during the period under investigation. It is calculated as:
n

E =1−

i=1 n i=1

(Oi − Pi ) Oi − O

2

2

(D.1)

with O observed and P predicted values. The range of E lies between 1 (perfect fit) and -∞. An efficiency of lower than zero indicates that the mean value of the observed time series would have been a better predictor than the model.

169

170

Appendix D. Performance indicators

Coefficient of determination (r2 )
The coefficient of determination (r2 ) is defined as the squared value of the coefficient of correlation. It is calculated as:    r =  
2 n i=1 n i=1

2 Oi − O
2

Pi − P
n i=1

Oi − O

Pi − P

    2

(D.2)

The range of r2 lies between 0 and 1 which describes how much of the observed dispersion is explained by the prediction. A value of zero means no correlation at all whereas a value of 1 means that the dispersion of the prediction is equal to that of the observation. The fact that only the dispersion is quantified is one of the major drawbacks of r2 if it is considered alone. A model which systematically over- or underpredicts all the time will still result in good r2 values close to 1.0 even if all predictions were wrong.

Volume error (V E)
The relative volume error (V E) is calculated as:
n

VE =

i=1

(Oi − Pi )
n i=1

(D.3) Oi

Bibliography

Abbott, M. B., Bathurst, J. C., Cunge, J. A., O’Connell, P. E., and Rasmussen, J. (1986). Introduction to the european hydrological system - systeme hydrologique europeen, ‘SHE’: history and philosophy of a physically based distributed modelling system. Journal of Hydrology, 87:45–49. Aerts, J. C. J. H., Botzen, W. J. W., Van der Veen, A., Krywkow, J., and Werners, S. (2008a). Dealing with uncertainty in flood management through diversification. Ecology and Society, 13(1). Aerts, J. C. J. H. and Droogers, P., editors (2004). Climate change in contrasting river basins. Adaptation strategies for water, food and environment. Cabi publishers, Wallingford, UK. Aerts, J. C. J. H., Sprong, T., and Bannnik, B. A. (2008b). Aandacht voor Veiligheid. Technical report, Leven met Water, Klimaat voor Ruimte, DG Water, Amsterdam, the Netherlands (in Dutch). 196 pp. Andr´asson, J., Bergstr¨m, S., Carlsson, B., Graham, L. P., and Lindstr¨m, G. (2004). e o o Hydrological change: Climate change impact simulations for Sweden. AMBIO: A Journal of the Human Environment, 33(4–5):228–234. Apel, H., Aronica, G. T., Kreibich, H., and Thieken, A. H. (2009). Flood risk analyses-how detailed do we need to be? Natural Hazards, 49(1):79–98, doi:10.1007/s11069-008-9277-8. Apel, H., Thieken, A. H., Merz, B., and Bl¨schl, G. (2004). Flood risk assessment and o associated uncertainty. Natural Hazards and Earth System Sciences, 4:295–308. Apel, H., Thieken, A. H., Merz, B., and Bl¨schl, G. (2006). A probabilistic modelling system o for assessing flood risks. Natural Hazards, 38:79–100, doi:10.1007/s11069-005-8603-7.

171

Asselman, N., Klijn, F., and Van der Most, H. (2008). Verkenning van nadere compartimentering van dijkringgebieden: hoofdrapport compartimenteringstudie. Technical report, Deltares, Delft, the Netherlands. Bakker, A. M. R. and Van den Hurk, B. J. J. M. (2009). Bias correction and resampling of RACMO output for the hydrological modelling of the Rhine. Technical Report TR-307, KNMI, De Bilt, the Netherlands. Barredo, J. I. (2009). Normalised flood losses in Europe: 1970–2006. Natural Hazards and Earth System Sciences, 9(1):97–104, doi:10.5194/nhess-9-97-2009. Bavay, M., Lehning, M., Jonas, T., and L¨we, H. (2008). Simulations of future snow cover o and discharge in Alpine headwater catchments. Hydrological Processes, 23(1):95–108, doi:10.1002/hyp.7195. Becker, G., Huitema, D., and Aerts, J. (2007). Transboundary flood management in the Rhine basin: Challenges for improved cooperation. Water Science and Technology, 56(4):125–135. Beersma, J. J., Buishand, T. A., and W´jcik, R. (2001). Rainfall generator for the Rhine o basin; multi-site simulation of daily weather variables by nearest-neighbour resampling, volume CHR-Report no. I-20, pages 69–77. International Commission for the Hydrology of the Rhine basin (CHR), Lelystad, the Netherlands. Beersma, J. J., Kwadijk, J. C. J., and Lammersen, R. (2008). Effects of climate change on the Rhine discharges, a review. In Vellinga, P., Katsman, C., Sterl, A., and Beersma, J., editors, Scenarios for flood protection. An international assessment. Background report for the Delta Committee 2008. KNMI, Deltares, Waterdienst, De Bilt, the Netherlands. Belz, J., Brahmer, G., Buiteveld, H., Engel, H., Grabher, R., Hodel, H., Krahe, P., Lammersen, R., Larina, M., Mendel, H.-G., Meuser, A., M¨ ller, G., Plonka, B., Pfisu ter, L., and van Vuuren, W. (2007). Das abflussregime des Rheins und seiner nebenfl¨sse u im 20. Jahrhundert. analyse, ver¨nderungen, trends. bericht nr. i-22. Technical report, a International Commission for the Hydrology of the Rhine Basin (CHR) (in German). Bergstr¨m, S. (1976). Development and application of a conceptual runoff model for Scano dinavian catchments. PhD thesis, SMHI Reports RHO 7, Swedish Meteorological and Hydrological Institute (SMHI), Nork¨ping, Sweden. o Bergstr¨m, S. (1992). The HBV model – its structure and applications. SMHI Reports RH 4, o Swedish Meteorological and Hydrological Institute (SMHI), Norrk¨ping, Sweden. o Bergstr¨m, S., Lindstr¨m, G., and Pettersson, A. (2002). Multi-variable parameter estimao o tion to increase confidence in hydrological modelling. Hydrological Processes, 16:413–421, doi:10.1002/hyp.332. Beven, K. (2001). How far can we go in distributed hydrological modelling? Hydrology and Earth System Sciences, 5(1):1–12. Beven, K. (2002a). Towards a coherent philosophy for modelling the environment. Proceedings of the Royal Society London A, 458:2465–2484, doi:10.1098/rspa.2002.0986.

172

Beven, K. (2002b). Towards an alternative blueprint for a physically based digitally simulated hydrologic response modelling system. Hydrological Processes, 16:189–206. Beven, K. (2006a). On undermining the science? Hydrological Processes, 20(14):3141–3146, doi:10.1002/hyp.6396. Beven, K. J. (2006b). Streamflow Generation Processes, volume 1 of Benchmark Papers in Hydrology. IAHS Press, Wallingford, UK. Bezuyen, M. J. and van Duin, M. (1998). Flood management in the Netherlands. Australian Journal of Emergency Management, 13(2):44–49. Bitesize, B. H. (2010). Hydrographs. http://www.bbc.co.uk/scotland/learning/bitesize/ higher/geography/physical. Accessed: 8 July 2010. Blackbourn, D. (2006). The conquest of nature: water, landscape, and the making of modern Germany. Norton, London, UK. 608 pp. Bogaard, T. A., Luxemburg, W. M. J., M. De Wit, M., Douben, N., and Savenije, H. H. G. (2005). Some hydrological challenges in understanding discharge generation processes in the Rhine and Meuse Basins. Physics and Chemistry of the Earth, 30:262–266, doi:10.1016/j.pce.2005.04.002. Booij, M. J. (2003). Determination and integration of appropriate spatial scales for river basin modelling. Hydrological Processes, 17:2581–2598, doi:10.1002/hyp.1268. Bossard, M., Feranec, J., and Otahel, J. (2000). CORINE land cover technical guide Addendum 2000. Technical report 40, European Environment Agency, Copenhagen, Denmark. Botzen, W. J. W. (2010). Economics of insurance against climate change. PhD thesis, VU University, Amsterdam, the Netherlands. 339 pp. Botzen, W. J. W. and Van den Bergh, J. C. J. M. (2008). Insurance against climate change and flooding in the Netherlands: Present, future, and comparison with other countries. Risk Analysis, 28(2):413–426, doi:10.1111/j.1539-6924.2008.01035.x. Bouwer, L. M. (2010). Disasters and climate change: analsyis and methods for projecting future risks. PhD thesis, VU University, Amsterdam, the Netherlands. 129 pp. Bouwer, L. M., Bubeck, P., and Aerts, J. C. J. H. (2010). Changes in future flood risk due to climate and development in a Dutch polder area. Global Environmental Change, 20(3):463–471, doi:doi:10.1016/j.gloenvcha.2010.04.002. Bouwer, L. M., Crompton, R. P., Faust, E., Hoppe, P., and Pielke Jr., R. A. (2007). Confronting disaster losses. Science, 318:753, doi:10.1126/science.1149628. Brakenridge, G. R. and Anderson, E. (2008). Darthmouth flood observatory. http://www.dartmouth.edu/ floods/, accessed: 8 March 2008.

173

Bronstert, A., B´rdossy, A., Bismuth, C., Buiteveld, H., Disse, M., Engel, H., Fritsch, U., a Hundecha, Y., Lammersen, R., Niehoff, D., and Ritter, N. (2007). Multi-scale modelling of land-use change and river training effects on floods in the Rhine Basin. River Research and Applications, 23:1102–1125, doi:10.1002/rra.1036. Bronstert, A., Niehoff, D., and B¨rger, G. (2002). Effects of climate and land-use change u on storm runoff generation: present knowledge and modelling capabilities. Hydrological Processes, 16(2):509–529, doi:10.1002/hyp.326. Bruijnzeel, L. A. (2008). Door de bomen de bodem niet meer zien. Oration given on 18 December 2008. VU University, Amsterdam, the Netherlands (in Dutch). B¨chele, B., Kreibich, H., Kron, A., Thieken, A., Ihringer, J., Oberle, P., Merz, B., and u Nestmann, F. (2006). Flood-risk mapping: contributions towards an enhanced assessment of extreme events and associated risks. Natural Hazards and Earth System Sciences, 6(4):485–503. Buishand, T. A. and Brandsma, T. (2001). Multi-site simulation of daily precipitation and temperature in the rhine basin by nearest-neighbor resampling. Water Resources Research, 37(11):2761–2776, doi:10.1029/2001WR000291. Buishand, T. A. and Lenderink, G. (2004). Estimation of future discharges of the river Rhine in the SWURVE project. Technical Report TR-273, KNMI, De Bilt, Netherlands. Caspary, H. J. (1996). Recent winter floods in Germany caused by changes in the atmospheric circulation across Europe. Physics and Chemistry of the Earth, 20(5-6):459–462, doi:10.1016/S0079-1946(96)00006-7. CCNR (2009). Market observation for inland navigation in Europe. Technical report, Central Commission for Navigation on the Rhine and European Commission Directorate-General for Energy and Transport, Strasbourg, France. CEA (2007). Reducing the social and economic impact of climate change and natural catastrophes insurance solutions and public-private partnerships. Technical report, Brussels, Belgium. 48 pp. Chow, V. T. (1959). Open channel hydraulics. McGraw-Hill Civil Engineering, London, UK. Christensen, J. H., Carter, T. R., Rummukainen, M., and Amanatidis, G. (2007). Evaluating the performance and utility of climate models: the PRUDENCE project. Climatic Change, 81:1–6, doi:10.1007/s10584-006-9211-6. Christensen, J. H. and Christensen, O. B. (2007). A summary of the PRUDENCE model projections of changes in European climate by the end of this century. Climatic Change, 81:7–30, doi:10.1007/s10584-006-9210-7. Commissie Noodoverloopgebieden (2002). Gecontroleerd overstromen (in Dutch). Covey, C., AchatuRao, K. M., Cubasch, U., Jones, P., Lambert, J. J., Mann, M. E., Philips, T. J., and Taylor, K. E. (2003). An overview of results from the Coupled Model Intercomparison Project. GlobalandPlanetaryChange, 37:103–133, doi:10.1016/S092108181(02)00193-5.

174

Daniel, V. E., Florax, R. J. G. M., and Rietveld, P. (2009). Flooding risk and housing values: An economic assessment of environmental hazard. Ecological Economics, 69(2):355–365, doi:10.1016/j.ecolecon.2009.08.018. Dankers, R. and Feyen, L. (2008). Climate change impact on flood hazard in Europe: An assessment based on high resolution climate simulations. Journal of Geophysical Research, 113, D19105, 17 pp, doi:10.1029/2007JD009719. De Bruijn, K. (2005). Resilience and flood risk management. A systems approach applied to lowland rivers. PhD thesis, TU Delft, Delft, Netherlands. 210 pp. De Bruijn, K. and Klijn, F. (2009). Risky places in the Netherlands: a first approximation for floods. Journal of Flood Risk Management, 2(1):58–67, doi:10.1111/j.1753318X.2009.01022.x. De Moel, H. and Aerts, J. C. J. H. (2010). Effect of uncertainty in land use, damage curves and inundation depth on flood damage estimates. Natural Hazards, doi:10.1007/s11069010-9675-6. De Moel, H., Aerts, J. C. J. H. ., and Koomen, E. (2011). Development of flood exposure in the Netherlands during the 20th and 21st century. Global Environmental Change, in press, doi:10.1016/j.gloenvcha.2010.12.005. De Moel, H., Van Alphen, J., and Aerts, J. (2009). Flood maps in Europe availability and use. Natural Hazards and Earth System Sciences, 9:289–301. methods,

De Roo, A. P. J., Wesseling, C. G., and Van Deursen, W. P. A. (1998). Physically based river modelling within a GIS. the LISFLOOD model. In Abrahart, R. J., editor, Proceedings of the 3rd International Conference on Geo-Computation. De Roo, A. P. J., Wesseling, C. G., and Van Deursen, W. P. A. (2000). Physically based river modelling within a GIS: The LISFLOOD model. Hydrological Processes, 14(11):1981–1992, doi:10.1002/1099-1085. De Wit, M. J. M. and Buishand, T. A. (2007). Generator of rainfall and discharge extremes (GRADE) for the Rhine and Meuse basins. Technical Report 218, KNMI, De Bilt, the Netherlands. Dekkers, J. E. C. and Koomen, E. (2007). Land-use simulation for water management: application of the Land Use Scanner model in two large-scale scenario-studies. In Koomen, E., Stillwell, J., Bakema, A., and Scholten, H., editors, Modelling land-use change; progress and applications, pages 355–373. Springer, Dordrecht, the Netherlands. Delft Hydraulics (2005). SOBEK-RE 1DFLOW technical reference 2.52.005. Technical report, Institute for Inland Water Management and Waste Water Treatment (RWS-RIZA), WL | Delft Hydraulics, Delft, the Netherlands. Deltacommissie (2008). Working together with water. A living land builds for its future. Findings of the Deltacommissie 2008. Technical report, the Netherlands. Available at www.deltacommissie.com.

175

Dettinger, M. D. (2005). From climate-change spaghetti to climate-change distributions for 21st century California. San Francisco Estuary and Watershed Science, 3(1). Diermanse, F. L. M. (2004). HR2006 recalculation of the design discharge Rhine. Technical report, WL | Delft Hydraulics, Delft, the Netherlands (in Dutch). Disse, M. and Engel, H. (2001). Flood events in the Rhine basin: Genesis, influences and mitigation. Natural Hazards, 23:271–290, doi:10.1023/A:1011142402374. Early, A. (2001). The Water Page - The Rhine River. http://www.thewaterpage.com/ rhine.htm. Accessed: 2 April 2007. Eberle, M., Buiteveld, H., Wilke, K., and Krahe, P. (2005). Hydrological modelling in the river Rhine basin part III - daily HBV model for the Rhine basin. Technical Report BfG1451, Institute for Inland Water Management and Wase Water Treatment (RIZA) and Bundesanstalt f¨ r Gew¨sserkunde (BfG), Koblenz, Germany. u a Eberle, M., Hammer, M., B¨ sch, N., Engel, H., Krahe, P., and Wilke, K. (2004). Crossu boundary effects of extreme peak discharges at the Lower Rhine. Section extreme discharges in the Rhine Basin. Technical report, Bundesanstalt f¨r Gew¨sserkunde (BfG), u a Koblenz, Germany (in Dutch). Eberle, M., Sprokkereef, E., Wilke, K., and Krahe, P. (2000). Hydrological modelling in the River Rhine Basin, part II - Report on hourly modelling. Technical Report BfG1338, Institute for Inland Water Management and Wase Water Treatment (RIZA) and Bundesanstalt f¨ r Gew¨sserkunde (BfG), Koblenz, Germany. u a EEA (2008). Impacts of Europe’s changing climate - 2008 indicator-based assessment. Technical Report doi: 10.2800/48117, European Environment Agency (EEA), Kopenhagen, Denmark. EEA (2009). Regional climate change and adaptation. The Alps facing the challenge of changing water resources. Technical Report 9/2009, European Environment Agency (EEA), Kopenhagen, Denmark. Engel, H. (1997). The flood events of 1993/1994 and 1995 in the Rhine river basin. In Leavesly, G. H., Lins, H. F., Nobilis, F., Parker, R. S., Schneider, V. R., and van de Ven, F. H. M., editors, Destructive Water: Water-Caused Natural Disasters, their Abatement and Control, IAHS Publication no. 239. IAHS Press, Wallingford, UK. Environment Agency (2007). Review of 2007 summer floods. GEHO1107BNMI-E-P, Almondsbury, Bristol, UK. Technical Report

EU (2007a). Directive 2007/60/EC of the European Parliament and of the Council of 23 october 2007 on the assessment and management of flood risks. Technical report, Commission of the European communities (EU), Brussels, Belgium. EU (2007b). Green paper. The European research area: New perspectives. Technical Report Com(2007) 161 final, Commission of the European communities (EU), Brussels, Belgium.

176

Evans, E. P., Ashley, R., Hall, J., Penning-Rowsell, E., Saul, A., Sayers, P., Thorne, C., and Watkinson, A. (2004). Foresight future flooding. Scientific summary: Volume I, Future risks and their drivers. Technical report, Office of Science and Technology, London, UK. FAO (2005). Forest and floods. Drowning in fiction or thriving on facts? Technical report, The Food and Agriculture Organization of the United Nations, Rome, Italy. Fujihara, Y., Tanaka, K., Watanabe, T., Nagano, T., and Kojiri, T. (2008). Assessing the impacts of climate change on the water resources of the Seyhan river basin in Turkey: Use of dynamically downscaled data for hydrologic simulations. Journal of Hydrology, 353:33–48, doi:10.1016/j.jhydrol.2008.01.024. Garrett, C. and M¨ller, P. (2008). Supplement to extreme events. Bulletin of the American u Meteorological Society, 89:ES45–ES56, doi:10.1175/2008BAMS2566.2. Glaser, R. and Stangl, H. (2003). Historical floods in the Dutch Rhine delta. Natural Hazards and Earth System Sciences, 3:1–9. Gocht, M. and Vogt, R. (2008). Flood forecasting and flood defence in Cologne. http://www.hrwallingford.co.uk/Mitch/Workshop2/Papers/GochtVogt.pdf. Accessed: 23 July 2008. Grabs, W. (1997). Impact of climate change on hydrological regimes and water resource management in the Rhine basin. Technical Report I-16, CHR, Lelystad, the Netherlands. Graham, L. P., Andr´asson, J., and Carlsson, B. (2007). Assessing climate change impacts on e hydrology from an ensemble of regional climate models, model scales and linking methods – a case study on the Lule River Basin. Climatic Change, 81:293–307, doi:10.1007/s10584006-9215-2. Gudden, J. (2004). Grensoverschrijdende effecten van extreem hoog water op de Niederrhein. Deelrapport overstromingen in Nordrhein-Westfalen en Gelderland. Technical report, Ministerium f¨ r Umwelt und Naturschutz, Landwirtschaft und Verbraucherschutz des Landes u Nordrhein-Westfalen and Provincie Gelderland and Minsterie van Verkeer en Waterstaat, Arnhem, the Netherlands (in Dutch). Hartje, V., Klaphake, A., Grossman, M., Mutafoglu, K., Borgwardt, J., Blazejczak, J., Gornig, M., Ansman, T., Koomen, E., and Dekkers, J. E. C. (2005). Regional projection of water demand and nutrient emissions - The GLOWA-Elbe approach. Poster presented at the Statuskonferenz Glowa-Elbe II, Cologne, Germany. Hay, L. E., Clark, M. P., Wilby, R. L., Gutowski, W. J., Leavesley, G. H., Pan, Z., Arritt, R. W., and Takle, E. S. (2002). Use of regional climate model output for hydrologic simulations. Journal of Hydrometeorology, 3(5):571–590. Herget, J. and Meurs, H. (2010). Reconstructing peak discharges for historic flood levels in the city of Cologne, Germany. GlobalandPlanetaryChange, 70(1–4):108–116, doi:10.1016/j.gloplacha.2009.11.011.

177

Hilferink, M. and Rietveld, P. (1999). Land Use Scanner: An integrated GIS based model for long term projections of land use in urban and rural areas. Journal of Geographical Systems, 1(2):155–177. Hingray, B., Mezghani, A., and Buishand, A. (2007). Development of probability distributions for regional climate change from uncertain global mean warming and an uncertain scaling relationship. Hydrology and Earth System Sciences, 11(3):1097–1114. Hooijer, A., Klijn, F., Pedroli, G. B. M., and Van Os, A. G. (2004). Towards sustainable flood risk management in the Rhine and Meuse basins: synopsis of the findings of IRMASPONGE. River Research and Applications, 20:343–357, doi:10.1002/rra.781. Horton, P., Schaefli, B., Mezghani, A., Hingray, B., and Musy, A. (2006). Assessment of climate-change impacts on alpine discharge regimes with climate model uncertainty. Hydrological Processes, 20(10):2091–2109, doi:10.1002/hyp.6197. Huitema, D. and Meijerink, S. (2010). Realizing water transitions: the role of policy entrepreneurs in water policy change. Ecology and Society, 15(2):26. Hundecha, Y. and B´rdossy, A. (2004). Modeling of the effect of land use changes on the a runoff generation of a river basin through parameter regionalization of a watershed model. Journal of Hydrology, 292(1–4):281–295. Hundecha, Y. and B´rdossy, A. (2005). Construction of scenarios of extreme precipitaa tion and temperature in western Germany for the end of the 21st century. In Geophys. Res. Abstracts. Vol. 7, Vienna, Austria. European Geosciences Union. General Assembly, Copernicus. Hurkmans, R. T. W. L. (2009). Effects of climate variability and land use change on the water budget of large river basins. PhD thesis, Wageningen University, the Netherlands. 174 pp. Hurkmans, R. T. W. L., Moel, H. D., Aerts, J. C. J. H., and Troch, P. A. (2008). Water balance versus land surface model in the simulation of rhine river discharges. Water Resources Research, 44(W01418), doi:10-1029/2007WR006168. Hurkmans, R. T. W. L., Terink, W., Uijlenhoet, R., Moors, E. J., Troch, P. A., and Verburg, P. H. (2009). Effects of land use changes on streamflow generation in the Rhine basin. Water Resources Research, 45(W06405), doi:10.1029/2008WR007574. ICPR (1998). Action Plan onf Flood Defence. Technical report, Intenational Commission for the Protection of the Rhine, Koblenz, Germany. ICPR (2001). Atlas 2001. Atlas of flood danger and potential damage due to extreme floods of the Rhine. Technical report, International Commission on the Protection of the Rhine. Koblenz, Germany. ICPR (2005a). Action Plan on Floods. Kurzfassung Bericht Nr. 156. Technical report, International Commission on the Protection of the Rhine, Koblenz, Germany.

178

ICPR (2005b). Nachweisinstrumente f¨r die reduzierung von schadensrisiken. Technical u Report 157, International Commission for the Protection of the Rhine, Koblenz, Germany (in German). 29 pp. ICPR (2005c). Rhine 2020. Programme on sustainable development. Balance 2000–2005. Technical report, International Commission for the Protection of the Rhine, Koblenz, Germany. IPCC (2000). Special Report on Emission Scenarios - a special report of working group III of the Intergovernmental Panel on Climate Change. IPCC (2007a). The 4th assessment report by IPCC. Intergovernmental Panel on Climate Change. IPCC (2007b). Climate Change 2007: The physical science basis. In Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K. B., Tignor, M., and Miller, H. L., editors, Summary for Policymakers. Contribution of Working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, pages 1–18. Cambridge University Press, Cambridge, UK. ISFD4 (2008). Statement of the 4th International Symposium on Flood Defence (ISDF4), 6–8 May 2008, Toronto, Canada. Ivanovi´, R. F. and Freer, J. E. (2009). Science versus politics: truth and uncertainty in c predictive modelling. Hydrological Processes, 23(17):2549–2554, doi:10.1002/hyp.7406. Jacob, C., Christensen, O. B., Doblas-Reyes, F. J., Goodess, C., Klein Tank, A., Lorenz, P., and Roeckner, E. (2008). Information on observations, global and regional modelling data availability and statistical downscaling. Technical report, EU FP6 Ensembles project, Brussels, Belgium. Jacob, D. (2001). A note to the simulation of the annual and inter-annual variability of the water budget over the Baltic Sea drainage basin. Meterology and Atmospheric Physics, 77:61–73, doi:10.1007/s007030170017. Jonkeren, O., Rietveld, P., and van Ommeren, J. (2007). Climate change and inland waterway transport - Welfare effects of low water levels on the River Rhine. Journal of Tranpsort Economics and Policy, 41:387–411. Jonkeren, O. E. (2009). Adaptation to climate change in inland waterway transport. PhD thesis, VU University, Amsterdam, the Netherlands. 153 pp. Jonkman, S. (2007). Loss of life estimation in flood risk assessment. Theory and applications. PhD thesis, Delft University, Delft, the Netherlands. 354 pp. Jothityangkoon, M., Sivalapan, M., and Farmer, D. L. (2001). Process controls of water balance variability in a large semi-arid catchment: downward approach to hydrological model development. Journal of Hydrology, 254:174–198, doi:10.1016/S0022-1694(01)00496-6. Katz, R. W. (2002). Techniques for estimating uncertainty in climate change scenarios and impact studies. Climatic Research, 20:167–185.

179

Kay, A. L., Reynard, N. S., and Jones, R. G. (2006). RCM rainfall for UK flood frequency estimation. I. Method and validation. Journal of Hydrology, 318:151–162, doi:10.1016/j.jhydrol.2005.06.012. Kleinn, J., Frei, C., Gurtz, J., L¨ thi, D., Vidale, P. L., and Ch¨r, C. (2005). Hydrologic u a simulations in the Rhine Basin driven by a regional climate model. Journal of Geophysical Research, 110, doi:10.1029/2004JD005143. Klemeˇ, V. (2000a). Tall tales about tails of hydrological distribution. I. Journal of Hydrologic s Engineering, 5(3):227–231, doi:10.1061/(ASCE)1084-0699(2000)5:3(227). Klemeˇ, V. (2000b). Tall tales about tails of hydrological distribution. II. Journal of Hydros logic Engineering, 5(3):232–239, doi:10.1061/(ASCE)1084-0699(2000)5:3(232). Klijn, F., Baan, P., De Bruijn, K., and Kwadijk, J. (2007). Overstromingsrisico’s in Nederland in een veranderend klimaat. Verwachtingen, schattingen en berekeningen voor het project ‘Nederland Later’. Technical report, Milieu en Natuurplan bureau (MNP), WL | Delft Hydraulics, Bilthoven, the Netherlands. 165 pp (in Dutch). Klijn, F., Van Buuren, M., and Van Booij, S. (2004). Flood-risk management strategies for an uncertaint future: Living with Rhine river floods in the Netherlands? AMBIO: A Journal of the Human Environment, 3:141–147, doi:10.1579/0044-7447-33.3.141. Koomen, E., Kuhlman, T., Groen, J., and Bouwman, A. (2005). Simulating the future of agricultural land use in the Netherlands. Tijdschrift voor Economische en Sociale Geografie, 96(2):218–224 (in Dutch). Koster, R. D., Suarez, M. J., Ducharne, A., Stieglitz, M., and Kumar, P. (2000). A catchment-based approach to modeling land surface processes in a general circulation model 1. model structure. Journal of Geophysical Research-Atmospheres, 105(D20):24809– 24822. Kottegoda, N. T. (1980). Stochastic Water Resources Technology. The MacMillan Press Ltd., London, UK. Krause, P., Boyle, D. P., and Base, F. (2005). Comparison of different efficiency criteria for hydrological model assessment. Advances in Geosciences, 5:89–97. Kreibich, H. and Thieken, A. (2009). Coping with floods in the city of Dresden, Germany. Natural Hazards, 51(3):423–436, doi:10.1007/s11069-007-9200-8. Kron, W. (2009). Flood insurance: from clients to global financial markets. Journal of Flood Risk Management, 2(1):68–75, doi:10.1111/j.1753-318X.2008.01015.x. Kundzewicz, Z. W., Mata, L. J., Arnell, N. W., D¨ll, P., Kabat, P., Jim´nez, B., Miller, o e K. A., Oke, T., Sen, Z., and Shiklomanov, I. A. (2007). Freshwater resources and their management. In Parry, M. L., Conziani, O. F., Palutikof, J. P., van der Linden, P. J., and Hanson, C. E., editors, Climate Change 2007: Impacts, Adaptation and Vulnerability. Contribution of Working Group II tot the Fourth Assessment Report of the Intergovernmental Panel on Climate Change, pages 173–210. Cambridge University Press, Cambridge, UK.

180

Kwadijk, J. C. J. (1993). The impact of climate change on the discharge of the River Rhine. PhD thesis, Department of Physical Geography, University of Utrecht, the Netherlands. Kwadijk, J. C. J. and Middelkoop, H. (1994). Estimation of impact of climate change on the peak discharge probability of the River Rhine. Climatic Change, 27:199–224, doi:10.1007/BF01093591. Kwadijk, J. C. J. and Rotmans, J. (1995). The impact of climate change on the River Rhine. A scenario study. Climatic Change, 30:397–425, doi:10.1007/BF01093854. Lammersen, R. (2004). Cross-boundary effects of extreme peak discharges at the Lower Rhine. Endreport. Technical report, Ministerium f¨r Umwelt und Naturschutz, Landu wirtschaft und Verbraucherschutz des Landes Nordrhein-Westfalen and Provincie Gelderland and Minsterie van Verkeer en Waterstaat, Arnhem, the Netherlands (in Dutch). Lammersen, R., Engel, H., Van den Langenheem, W., and Buiteveld, H. (2002). Impact of river training and retention measures on flood peaks along the Rhine. Journal of Hydrology, 267:115–124, doi:10.1016/S0022-1694(02)00144-0. Landscan (2009). LandScanTM Global Population Database. sci/landscan, acessed: 9 May 2009. http://www.ornl.gov/

Leander, R. (2009). Simulation of precipitation and discharge extremes of the river Meuse in current and future climate. PhD thesis, Utrecht University, Utrecht, the Netherlands. 124 pp. Leander, R., Buishand, A. T., Van den Hurk, B. J. J. M., and De Wit, M. J. M. (2008). Estimated changes in flood quantiles of the River Meuse from resampling of regional climate model output. Journal of Hydrology, 351(3–4):331–343, doi:10.1016/j.jhydrol.2007.12.020. Leander, R. and Buishand, T. A. (2007). Resampling of regional climate model output for the simulation of extreme river flows. Journal of Hydrology, 332:487–496, doi:10.1016/j.jhydrol.2006.08.006. Lenderink, B., Van Ulden, A., Van den Hurk, B., and Van Meijgaard, E. (2006). Summertime inter-annual temperature variability in an ensemble of regional model simulations: analysis of the surface energy budget. Technical Report 2006–04, KNMI, De Bilt, The Netherlands. Lenderink, G., Buishand, A., and Van Deursen, W. (2007a). Estimates of future discharges of the river Rhine using two scenario methodologies: direct versus delta approach. Hydrology and Earth System Sciences, 11(3):1145–1159. Lenderink, G., Van den Hurk, B., Van Meijgaard, E., Van Ulden, A. P., and Cuijpers, J. (2003). Simulation of present-day climate in RACMO2: First results and model developments. Technical Report 252, KNMI, De Bilt, the Netherlands. Lenderink, G., Van Ulden, A., Van den Hurk, B., and Keller, F. (2007b). A study on combining global and regional climate model results for generating climate scenarios of temperature and precipitation for the Netherlands. Climate Dynamics, 29:157–176, doi:10.1007/s00382-007-0227-z.

181

Liang, X., Lettenmaier, D. P., Wood, E. F., and Burges, S. J. (1994). A simple hydrologically based model of land surface water and energy fluxes for GCMs. Journal of Geophysical Research, 99(D7):14415–14428. Liang, X. and Zhenghui, X. (2001). A new surface runoff parameterization with subgrid-scale soil heterogeneity for land surface models. Advances in Water Resources, 24:1173–1193, doi:10.1016/S0309-1708(01)00032-X. Lindstr¨m, G., Johansson, B., Persson, M., Gardelin, M., and Bergstr¨m, S. (1997). Deo o velopment and test of the distributed HBV-96 hydrological model. Journal of Hydrology, 201:272–288, doi:10.1016/S0022-1694(97)00041-3. Liu, Z. and Todini, E. (2002). Towards a comprehensive physically-based rainfall-runoff model. Hydrology and Earth System Sciences, 6:859–881. Lohmann, D., Nolte-Holube, R., and Raschke, E. (1996). A large-scale horizontal routing model to be coupled to land surface parametrization schemes. Tellus, 48A(5):708–721. Loonen, W. and Koomen, E. (2008). Calibration and validation of the Land Use Scanner allocation algorithms. Technical report, Milieu- en Natuurplanburea, Bilthoven, the Netherlands. Ludwig, K. and Bremicker, L. (2006). The water balance model LARSIM - design, content and applications. Technical Report 22, Freiburger Schriften zur Hydrologie, Freiburg, Germany. McMillan, H. K. and Brasington, J. (2008). End-to-end flood risk assessment: A coupled model cascade with uncertainty estimation. Water Resources Research, 44(W03419), doi:10.1029/2007wr005995. Meijgaard, E., Van Ulft, L. H., Van den Berg, W. J., Bosveld, F. C., Van den Hurk, B. J. J. M., Lenderink, G., and Siebesma, A. P. (2008). The KNMI regional atmospheric climate model RACMO, version 2.1. Technical Report 302, KNMI, De Bilt, the Netherlands. Menzel, L., Thieken, A. H., Schwandt, D., and B¨rger, G. (2006). Impact of climate change u on the regional hydrology - scenario-based modelling in the German Rhine catchment. Natural Hazards, 38:45–61, doi:10.1007/s11069-005-8599-z. Merz, B., Hall, J., Disse, M., and Schumann, A. (2010). Fluvial flood risk management in a changing world. Natural Hazards and Earth System Sciences, 10(3):509–527. Merz, B., Kreibich, H., Thieken, A., and Schmidtke, R. (2004). Estimation uncertainty of direct monetary flood damage to buildings. Natural Hazards and Earth System Sciences, 4:153–163. Merz, B. and Thieken, A. (2009). Flood risk curves and uncertainty bounds. Natural Hazards, 51(3):437–458, doi:10.1007/s11069-009-9452-6. Merz, B., Thieken, A. H., and Gocht, M. (2007). Flood risk mapping at the local scale: concepts and challenges. In Begum, S., Stive, M., and Hall, J., editors, Flood Risk Management in Europe - innovation in policy and practice, pages 231–251. Springer, Dordrecht, the Netherlands.

182

Messner, F., Penning Roswell, E. C., Green, C., Meyer, V., Tunstall, S. M., and Van der Veen, A. (2007). Evaluating flood damages: guidance and recommendations on principles and practices. Technical Report T09-06-01, FLOODsite report. 178 pp. Middelkoop, H., Daamen, K., Gellens, D., Grabs, W., Kwadijk, J. C. J., Lang, H., Parmet, B. W. A. H., Sch¨dler, B., Schulla, J., and Wilke, K. (2001). Impact of climate change a on hydrological regimes and water resources management in the Rhine basin. Climatic Change, 49:105–128, doi:10.1023/A:1010784727448. Middelkoop, H. and Kwadijk, J. C. J. (2001). Towards integrated assessment of the implications of global change for water management - the Rhine experience. Physics and Chemistry of the Earth, 26(7–8):553–560, doi:10.1016/S1464-1909(01)00049-1. Middelkoop, H., Van Asselt, B. A., Van’t Klooster, S. A., Van Deursen, W. P. A., Kwadijk, J. C. J., and Buiteveld, H. (2004). Perspectives on flood management in the Rhine and Meuse rivers. River Research and Applications, 20:327–342, doi:10.1002/rra.782. Milly, P. C. D., Betancourt, J., Falkenmark, M., Robert, M., Hirsch, R. M., Kundzewicz, Z. W., Lettenmaier, D. P., and Stouffer, R. J. (2008). Stationarity is dead: Whither water management? Science, 319:573–574, doi:10.1126/science.1151915. Milly, P. C. D., Dunne, K. A., and Vecchia, A. V. (2005). Global pattern of trends in streamflow and water availability in a changing climate. Nature, 438:347–350. Milly, P. C. D., Wetherald, R. T., Dunne, K. A., and Delworth, T. L. (2002). Increasing risk of great floods in a changing climate. Nature, 415:514–517, doi:10.1038/415514a. Ministry of Transport, Public Works and Water Management (1995). Deltaplan grote rivieren. Technical Report Kamerstukken 1994-1995, 18 106, nrs. 54 en 57, The Hague, the Netherlands (in Dutch). Ministry of Transport, Public Works and Water Management (2006a). Flood risks and safety in the Netherlands (Floris) - full report. Technical Report DWW-2006-014, The Hague, the Netherlands. 134 pp. Ministry of Transport, Public Works and Water Management (2006b). Hydraulische Randvoorwaarden 2006 voor het toetsen van primaire waterkeringen voor de derde toetsronde 2006–2011. Technical report, The Hague, the Netherlands (in Dutch). Ministry of Transport, Public Works and Water Management (2006c). Spatial Planning Key Decision Room for the River. Investing in the safety and vitality of the Dutch river basin region. Technical report, The Hague, the Netherlands. Ministry of Transport, Public Works and Water Management (2007). Leidraad rivieren. Technical report, The Hague, the Netherlands. 322 pp (in Dutch). Mitchell, T. D. and Jones, P. D. (2005). An improved method of constructing a database of monthly climate observations and associated high-resolution grids. International Journal of Climatology, 25:693–712, doi:10.1002/joc.1181.

183

MNP (2007). Nederland Later; tweede duurzaamheidsverkenning deel fysieke leefomgeving Nederland. Technical Report 500127001/2007, Milieu- en Natuurplanbureau, Bilthoven, the Netherlands. M¨lders, R., Parmet, B., and Wilke, K. (1999). Hydrological modelling in the river Rhine u basin. Final report. Technical Report BfG-125, Federal Institute of Hydrology (BfG), Koblenz, Germany. Munich Re AG (2008). Natural catastrophes 2008. Analyses, assessments, positions. Technical report, M¨nicher R¨ckversicherungs-Gesellshaft, Munich, Germany. u u Nash, J. E. and Sutcliffe, J. V. (1970). River flow forecasting through conceptual models part I – A discussion of principles. Journal of Hydrology, 10:282–290. Nicholls, R. J., Hanson, S., Herweijer, C., Patmore, N., Hallegatte, S., Corfee-Morlot, J., Chˆteau, J., and Muir-Wood, R. (2008). Ranking port cities with high exposure and a vulnerability to climate extremes. Exposure estimates. Technical report, OECD, Paris, France. Orlowsky, B., Gerstengarbe, F., and Werner, P. C. (2007). A resampling scheme for regional climate simulations and its performance compared to a dynamical RCM. Theoretical and Applied Climatology, doi:10.1007/s00704-007-0352-y. Ouwendijk, E., Heiland, P., Haupter, B., and M¨llenkamp, S. (2001). Detention reservoirs o and spatial planning instruments as used for flood prevention in Germany and the Netherlands. Technical Report R3294.65, WL | Delft Hydraulics, TU Darmstadt, University of Bonn, Delft, the Netherlands. Pahl-Wost, C. (2008). Requirements for adaptive water management. In Pahl-Wost, C., Kabat, P., and M¨ltgen, J., editors, Coping with Complexity and Uncertainty, pages 1–22. o Pappenberger, F. and Beven, K. (2006). Ignorance is bliss: Or seven reasons not to use uncertainty analysis. Water Resources Research, 42:W05302, doi:10.1029/2005WR004820. Passchier, R. and Stone, K. (2003). Analysis of the instrumentation for computing design discharges. Technical Report Q2705, WL | Delft Hydraulics, Delft, the Netherlands. Perrin, C., Michel, C., and Andr´assian, V. (2000). Does a large number of parameters e enhance model performance? Comparative assessments of common catchment model structures on 429 catchments. Journal of Hydrology, 242:275–301, doi:10.1016/S00221694(00)00393-0. Petrow, T., Thieken, A. H., Kreibich, H., Merz, B., and Heinrich Bahlburg, C. (2006). Improvements on flood alleviation in Germany: Lessons learned from the Elbe flood in August 2002. Environmental Management, 38(5):717–732, doi:10.1007/s00267-005-6291-4. Pfister, L., Kwadijk, J. C. J., Musy, A., Bronstert, A., and Hoffman, L. (2004). Climate change, land use change and runoff prediction in the Rhine–Meuse basins. River Research and Applications, 20:229–241, doi:10.1002/rra.775.

184

Pielke Jr., R. A., Gratz, J., Landsea, C. W., Collins, D., Saunders, M., and Musulin, R. (2008). Normalized hurricane damages in the United States: 1900-2005. Natural Hazards Review, 9(1):29–42, doi:10.1061/(ASCE)1527-6988(2008)9:1(29). Pielke Sr., R., Beven, K., Brasseur, G., Calvert, J., Chahine, M., Dickerson, R. R., Entekhabi, D., Foufoula-Georgiou, E., Gupta, H., Gupta, V., Krajeweski, W., Krider, E. P., Lau, W. K. M., McDonnel, J., Rossow, W., Schaake, J., Smith, J., Sorooshian, S., and Wood, E. (2009). Climate change: The need to consider human forcings besides greenhouse gases. EOS, 90(45):413. Pinter, N., van der Ploeg, R. R., Schweigert, P., and Hoefer, G. (2006). Flood magnification on the River Rhine. Hydrological Processes, 20:147–164, doi:10.1002/hyp.5908. Pitman, A. J. and Perkins, S. E. (2007). Reducing uncertainty in selecting climate models for hydrological impact assessments. In Boegh, E., Kunstmann, H., Wagener, T., Hall, A., L.Bastidas, Franks, S., Gupta, H., Rosbjerg, D., and Schaake, J., editors, Quantification and Reduction of Predictive Uncertainty for Sustainable Water Resource Management, Symposium HS2004 at IUGG2007, Perugia, Italy, 3–15 July 2007, pages 3–15. IAHS. Plate, E. (2002). Flood risk and flood management. Journal of Hydrology, 267(1-2):2–11, doi:10.1016/S0022-1694(02)00135-X. Preusser, F. (2008). Characterisation and evolution of the river Rhine system. Netherlands Journal of Geosciences, 51(1):7–19. Prudhomme, C. (2006). GCM and downscaling uncertainty in modelling of current river flow; why is it important for future impacts? In Proceedings 5th FRIEND World Conference, November 2006, Havana, Cuba. IAHS, Wallingford, UK. Prudhomme, C., Jakob, D., and Svensson, C. (2003). Uncertainty and climate change impact on the flood regime of small UK catchments. Journal of Hydrology, 227:1–23, doi:10.1016/S0022-1694(03)00065-9. Raadgever, T., Bubeck, P., Rotter, S., Interwies, E., Becker, G., Aerts, J., and Te Linde, A. (2008). ACER Protokol Workshop 3. Scenarios and measures for successful water management (of floods) in the Rhine: a German - Dutch workshop. Technical report, Institute for Environmental Studies, Seecon and TU Delft, Amsterdam, the Netherlands (in German). 30 pp. Raff, D., Pruitt, T., and Brekke, L. (2009). A framework for assessing flood frequency based on climate projection information. Hydrology and Earth System Sciences, 13(11):2119– 2136. RebelGroup (2007). Indirecte effecten overstromingen in het kader van HIS-SSM en de KBA WV21. Technical report, RebelGroup Advisory Belgium, Antwerp (in Dutch). Refsgaard, J. C. (1996). Terminology, modelling protocol and classification of hydrological model codes. In Abbott, M. B. and Refsgaard, J. C., editors, Distributed Hydrological Modelling, pages 17–39. Kluwer Academic, the Netherlands.

185

Refsgaard, J. C. (1997). Parameterisation, calibration and validation of distributed hydrological models. Journal of Hydrology, 198:69–97, doi:10.1016/S0022-1694(96)03329-X. Reggiani, P., Hassanizadeh, S. M., Sivapalan, M., and Gray, W. G. (1998). A unifying framework for watershed thermodynamics: Balance equiations for mass, momentum, energy and entropy, and the second law of thermodynamics. Advances in Water Resources, 22:367–398. Reggiani, P., Hassanizadeh, S. M., Sivapalan, M., and Gray, W. G. (1999). A unifying framework for watershed thermodynamics: Constitutive relationships. Advances in Water Resources, 23:15–39. Reggiani, P. and Schellekens, J. (2003). Modelling of hydrological responses: The representative elementary watershed approach as an alternative blueprint for watershed modelling. Hydrological Processes, 17:3785–3789, doi:10.1002/hyp.5167. Rientjes, T. and Zaadnoordijk, W. J. (2000). Hoogwatervoorspelling: Fysisch gebaseerde regen-afvoermodellering. Dilemma of d´j` vu? Stromingen, 6:33–44. ea Ritsema van Eck, J. and Koomen, E. (2008). Characterising urban concentration and landuse diversity in simulations of future land use. Annals of Regional Science, 42(1):123–140. RMS (2006). Germany river flood. Technical report, Risk Management Solutions, London, UK. Roth, D., Warner, J., and Winnubst, M. (2006). Een noodverband tegen hoog water. Waterkennis, beleid en politiek rond noodoverloopgebieden. Technical report, Wageningen University, Wageningen, the Netherlands. Samuels, P., Klijn, F., and Dijkman, J. (2006). An analysis of the current practice of policies on river flood risk management in different countries. Irrigation and Drainage, 55:141–150, doi:10.1002/ird.257. Savenije, H. G. (2001). Equifinality, a blessing in disguise? Hydrological Processes, 15:2835– 2838, doi:10.1002/hyp.494. Schanze, J., Schwarze, R., Cartensen, D., and Deilmann, C. (2008). Analyzing and managing uncertain futures of large-scale fluvial flood risk systems. In Simonomic, S. P., Bourget, P. G., and Blanchard, S. F., editors, Managing Flood Risk, Reliability and Vulnerability. Proceedings of the 4th International Symposium on Flood Defence. 6–8 May 2008, Toronto, Canada. Schanze, J., Zeman, E., and Marsalek, J. (2006). Flood Risk Management: Hazards, Vulnerability and Mitigation Measures. NATO Science Series: IV: Earth and Environmental Sciences. Springer, Dordrecht, the Netherlands. 319 pp. Sch¨r, C., Luthi, D., and und E. Heise, U. B. (1998). The soil-precipitation feeda back: a process study with a regional climate model. Journal of Climate, 12:722–741, doi:10.1175/1520-0442(1999)012.

186

Schulla, J. and Kaspar, K. (2006). Model description WaSIM-ETH. Technical report, IAC, ETH, Zurich, Switzerland. Seibert, J. (1999). Conceptual runoff models - fiction or representation of reality. PhD thesis, Acta University, Uppsala, Sweden. Seneviratne, S. I., Luthi, D., Litschi, M., and Sch¨r, C. (2006). Land-atmosphere coupling a and climate change in Europe. Nature, 443:205–209, doi:10.1038/nature05095. Shabalova, M. V., Van Deursen, W. P. A., and Buishand, T. A. (2003). Assessing future discharge of the River Rhine using regional climate model integrations and a hydrological model. Climatic Research, 23:233–246. Shaw, E. M. (2002). Hydrology in Practice. Nelson Thornes Ltd., Cheltenham, UK. Silva, W. (2003). Hoeveel (hoog)water kan ons land binnenkomen via de Rijn bij Lobith, nu en in de toekomst. Technical Report 2003.015, Ministry of Transport, Public Works and Water Management, Lelystad, the Netherlands (in Dutch). Silva, W., Gudden, J., Stijnen, J., and Verhoef, F. (2006). Rampenbeheersingsstrategie. overstromingen Rijn en Maas. Quick scan overstromings(kans) en compartimentering grensoverschrijdende dijkringgebieden 42 en 48. Technical report, Ministry of Transport, Public Works and Water Management, The Hague, the Netherlands (in Dutch). Silva, W., Klijn, F., and Dijkman, J. P. M. (2001). Room for the Rhine branches in the Netherlands. What the research has taught us. Technical Report 2001.031, WL | Delft Hydraulics, Ministry of Transport, Public Works and Water Management, Delft, Lelystad, the Netherlands. 162 pp. Silva, W. and Van Velzen, E. (2008). De dijk van de toekomst? Quick scan Doorbraakvrije dijken. Technical Report 2008.052, Ministry of Transport, Public Works and Water Management, The Hague, the Netherlands (in Dutch). 83 pp. Sivakumar, B. (2005). Hydrologic modeling and forecasting: Role of thresholds. Environmental Modelling and Software, 20:515–519, doi:10.1016/j.envsoft.2004.08.006. Sivakumar, B. (2008). Undermining the science or undermining Nature? Processes, 22(6):893–897, doi:10.1002/hyp.7004. Hydrological

Sivapalan, M. and Samuel, J. M. (2009). Transcending limitations of stationarity and the return period: process-based approach to flood estimation and risk assessment. Hydrological Processes, 23(11):1671–1675, doi:10.1002/hyp.7292. Smith, K. (2001). Environmental Hazards: Assessing Risk and Reducing Disaster. Routledge, London, UK, third edition. Sprokkereef, E. (2001). A hydrological database for the Rhine basin. Technical report, International Commission for the Hydrology of the Rhine Basin (CHR), Arhnem, the Netherlands (in German).

187

Sprong, T. (2009). Extreme flood peaks on the Rhine. Discharges of 16000 m3 /s and higher: myth or reality (in Dutch). Sterl, A., Severijns, C., Van Oldenborgh, G. J., Dijkstra, H., Hazeleger, W., Van den Broeke, M., Burgers, G., Van den Hurk, B. J. J. M., Van Leeuwen, P. J., and Van Velthoven, P. (2007). The ESSENCE project - signal to noise ratio in climate projections. Technical report, KNMI, De Bilt, the Netherlands. Stewart, I. T. (2008). Changes in snowpack and snowmelt runoff for key mountain regions. Hydrological Processes, 23(1):78–94, doi:10.1002/hyp.7128. Straatsma, M. and Baptist, M. (2008). Floodplain roughness parameterization using airborne laser scanning and spectral remote sensing. Remote Sensing of Environment, 112(3):1062– 1080, doi:10.1016/j.rse.2007.07.012. Swiatek, D. and Kubrak, J. (2007). The vegetation influence on friction factors of a lowland river - a case study of the Lower Biebrza River. In Okruszko, T., Maltby, E., Szatylowicz, J., Swiatek, D., and Kotowski, W., editors, Wetlands: Monitoring, Modelling and Management. Proceedings and Monographs in Engineering, Water and Earth Sciences, pages 9–13. Taylor & Francis Ltd, London, UK. Te Linde, A. H. (2007). Effect of climate change on the discharge of the rivers Rhine and Meuse. Applying the KNMI 2006 scenarios using the HBV model. Technical report, WL | Delft Hydraulics, Delft, the Netherlands. Te Linde, A. H. (2009). Modeling the effect of flood management strategies in the Rhine basin. ACER working paper, Institute for Environmental Studies, VU University, Amsterdam. 27 pp. Te Linde, A. H. and Aerts, J. C. J. H. (2008). Simulating flood-peak probability in the Rhine Basin and the effect of climate change. In Samuels, P., Huntington, T. G., Allsop, W., and Harrop, J., editors, Flood Risk Management. Research and Practice. CRC Press, Oxford, UK. Te Linde, A. H., Aerts, J. C. J. H., Bakker, A. M. R., and Kwadijk, J. C. J. (2010). Simulating low probability peak discharges for the Rhine basin using resampled climate modeling data. Water Resources Research, 46(WR03512), doi:10.1029/2009WR007707. Te Linde, A. H., Aerts, J. C. J. H., Hurkmans, R. T. W. L., and Eberle, M. (2008a). Comparing model performance of two rainfall-runoff models in the Rhine Basin using different atmospheric forcing data sets. Hydrology and Earth System Sciences, 12:943– 957. Te Linde, A. H., Aerts, J. C. J. H., and Van den Hurk, B. J. J. M. (2008b). Effects of flood control measures and climate change in the Rhine basin. In Simonomic, S. P., Bourget, P. G., and Blanchard, S. F., editors, Managing Flood Risk, Reliability and Vulnerability. Proceedings of the 4th International Symposium on Flood Defence. 6–8 May 2008, Toronto, Canada.

188

Te Linde, A. H., De Moel, H., and Aerts, J. C. J. H. (2011). Towards risk-based flood management to accommodate future uncertainties, and an outlook to the assessment of drought risk. In Brils, J. and Vermaat, J., editors, Towards Risk-Based Management of River basins. Springer. Forthcoming. Ten Brinke, W. B. M., Saeijs, G. E. M., Helsloot, I., and Van Alphen, J. (2008). Safety chain approach in flood risk management. Proceedings of the Institution of Civil EngineersMunicipal Engineer, 161(2):93–102, doi:10.1680/muen.2008.161.2.93. Thieken, A. H., Muller, M., Kreibich, H., and Merz, B. (2005). Flood damage and influencing factors: New insights from the August 2002 flood in Germany. Water Resources Research, 41(W12430), doi:10.1029/2005WR004177. Thieken, A. H., Olschewski, A., Kreibich, H., Kobsch, S., and Merz, B. (2008). Development and evaluation of FLEMOps - a new Flood Loss Estimation MOdel for the private sector. In Proverbs, D., Brebbia, C. A., and Penning-Roswell, E., editors, Flood Recovery, Innovation and Response I. WIT, Ashurst, UK. 800 pp. Troy, T., Sheffield, J., and Wood, E. (2007). Temporal and spatial scales in hydrological model calibration. In Geophysical Research Abstracts. European Geosciences Union General Assembly 2007, Vienna, Austria. Uehrlinger, U., Wantzen, K., Leuven, R., and Arndt, H. (2009). The Rhine river basin. In Tockner, K., Robinson, C., and Uehlinger, U., editors, Rivers of Europe, pages 199–247. Elsevier, London, UK. Uhlenbrook, S. (2003). An emperical approach for delineating spatial units with the same dominating runoff generation processes. Physics and Chemistry of the Earth, 28:297–303, doi:10.1016/S1474-7065(03)00041-X. Uhlenbrook, S., Seibert, J., Leibundgut, C., and Rodhe, A. (1999). Prediction uncertainty of conceptual rainfall-runoff models caused by problems to identify model parameters and structure. Hydrological Sciences Journal, 44(5):779–798. UN-HABITAT (2008). State of the world’s cities 2008/2009. Harmonious cities. Technical report, United Nations Human Settlements Programme, Nairobi, Kenya. Van Beek, E., Haasnoot, M., Meijer, K. M., Delsman, J. R., Snepvangers, J. J. J. C., Baarse, G., Van Ek, R., Prinsen, G. F., Kwadijk, J. C. J., and Van Zetten, J. W. (2008). Verkenning kosteneffectiviteit van grootschalige maatregelen tegen droogteschade als gevolg van de G+ en W+ klimaatscenario’s. Technical report, Deltares, the Netherlands (in Dutch). Van Dantzig, D. (1956). Economic decision problems for flood prevention. Econometrica, 24(3):276–287. Van den Berg, A., Alberts, F., and Ligthart, N. (2010). Overstromingsrisicokaarten. http:// www.risicokaart.nl. Accessed 4 April 2010 (in Dutch).

189

Van den Hurk, B., Hirschi, M., Sch¨r, C., Lenderink, G., van Meijgaard, E., van Ulden, A., a Rockel, B., Hagemann, S., Graham, P., Kjellstr¨m, E., and Jones, R. (2005). Soil control o on runoff response to climate change in Regional Climate Model simulations. Journal of Climate, 18:3536–3551. Van den Hurk, B., Klein Tank, A., Lenderink, G., Van Ulden, A., Van Oldenborgh, G., Katsman, C., Van den Brink, H., Keller, F., Bessembinder, J., Burgers, G., Komen, G., Hazeleger, W., and Drijfhout, S. (2006). KNMI climate change scenarios 2006 for the Netherlands. Technical Report WR 2006-01, KNMI, De Bilt, the Netherlands. Van den Hurk, B., Klein Tank, A., Lenderink, G., van Ulden, A., van Oldenborgh, G., Katsman, C., van den Brink, H., Keller, F., Bessembinder, J., Burgers, G., Komen, G., Hazeleger, W., and Drijfhout, S. (2007). New climate change scenarios for the Netherlands. Water Science and Technology, 56(4):27–33. Van der Veen, R. (2007). Creation of SOBEK-model FEWS Rhine 3.01 and 3.02 (in Dutch). Technical report, Ministry of Transport, Public Works and Water Management, Lelystad, the Netherlands. Van der Veen, R., Lammersen, R., Kroekenstoel, D. F., and Brinkmann, M. (2004). Input data of BOS-IR Lower Rhine - Rhine branches. Cross-boundary effects of extreme peak discharges at the lower rhine (available in dutch and german), Ministerium f¨r Umwelt und u Naturschutz, Landwirtschaft und Verbraucherschutz des Landes Nordrhein-Westfalen and Provincie Gelderland and Minsterie van Verkeer en Waterstaat, Arnhem, the Netherlands. Van Deursen, W. P. A. (2002). Climate change in the Rhine and Meuse River Basins. Technical report, Carthago Consultancy, Rotterdam, the Netherlands (in Dutch). Van Deursen, W. P. A. (2006). Report on Rhineflow / Meuseflow. New KNMI scenarios May 2006. Technical report, Carthago Consultancy, Rotterdam, the Netherlands (in Dutch). Van Deursen, W. P. A. and Kwadijk, J. C. J. (1993). RHINEFLOW: an integrated GIS water balance model for the river Rhine. In Kovar, K. and Nachtnebel, H. P., editors, HydroGIS 93: Application of Geographic Information Systems in Hydrology and Water Resources., pages 507–518. IAHS, Wallingford, UK. Van Deursen, W. P. A. and Middelkoop, H. (2001). Development of flood management strategies for the Rhine and Meuse basins in the context of integrated river management. IRMA-SPONGE Project 2. Technical report, the Netherlands Centre for River Studies (NCR). Van Vuren, B. (2005). Stochastic modelling of river morphodynamics. PhD thesis, Delft University, Delft, the Netherlands. 275 pp. Vellinga, P., Katsman, C., Sterl, A., and Beersma, J. (2008). Exploring high-end climate change scenarios for flood protection of the Netherlands. Background report for the Delta Committee 2008. Technical report, WUR and KNMI, Wageningen, the Netherlands.

190

Vellinga, P., Mills, E., Berz, G., Bouwer, L. M., Huq, S., Kozak, L. A., Palutikof, J., Schanzenbacher, B., and Soler, G. (2001). Insurance and other financial services. In McCarthy, J., Canziani, O., Leary, N., Dokken, D., and White, K., editors, Climate Change 2001: Impacts, Adaptation, and Vulnerability. Contribution of Working Group II to the Third Assessment Report of the Intergovernmental Panel on Climate Change, pages 417–450 pp. Cambridge University Press, Cambridge, UK. Verburg, P. H., Eickhout, B., and Van Meijl, H. (2008). A multi-scale, multi-model approach for analyzing the future dynamics of European land use. Annals of Regional Science, 42(1):57–77, doi:10.1007/s00168-007-0136-4. Verburg, P. H. and Overmars, K. P. (2009). Combining top-down and bottom-up dynamics in land use modeling: exploring the future of abandoned farmlands in Europe with the Dyna-CLUE model. Landscape Ecology, 24(9):1167–1181, doi:10.1007/s10980-009-9355-7. Vis, M., Bruijn, K. M. D., and Klijn, F. (2003). Resilience strategies for flood risk management in the Netherlands. International Journal of River Basin Management, 1:33–40. Vivoni, E. R. (2003). Hydrologic Modeling using Triangulated Irregular Networks: Terrain Representation, Flood Forecasting and Catchment Response. PhD thesis, Cambridge Massaschusetts Institute of Technology, USA. Volkskrant, D. (1995). Elk jaar stand van de eeuw is Keulen wat te veel. De Volkskrant. 28 January 1995, in Dutch. Wagener, T., McIntyre, N., Lees, M. J., Wheater, H. S., and Gupta, H. V. (2003). Towards reduced uncertainty in conceptual rainfall-runoff modelling: Dynamic identifiability analysis. Hydrological Processes, 17:455–476, doi:10.1002/hyp.1135. Wagtendonk, A., Juli˜o, R., and Schotten, C. (2001). A regional planning application of a Euroscanner in Portugal. In Stillwell, J. and Scholten, H., editors, Land Use Simulation for Europe. Kluwer Academic Publishers, Amsterdam, the Netherlands. 293 pp. Ward, R. C. and Robinson, M. (1990). Principles of Hydrology. McGraw-Hill, Maidenhead, UK, second edition. Webster, M. (2003). Communicating climate change uncertainty to policymakers and the public - an editorial comment. Climatic Change, 61(1–2):1–8, doi:10.1023/A:1026351131038. Weerts, A. H. (2003). Assessing and quantifying the combined effect of model parameter and boundary uncertainties in model based flood forecasting. In Geophysical Research Abstracts. EGU Joint Assembly, April 2003, Nice, France. Weerts, A. H. and Van der Klis, H. (2004). FEWS-Rhine version 1.02. Improvements and adjustments. Technical Report Q3618, WL | Delft Hydraulics, Delft, the Netherlands. Wessel, J. (1995). Flood management of the transnational Rhine river. In Impacts, and Management of Extreme Floods. U.S.– Italy Research Workshop on the Hydrometeorology, 13–17 November 1995, Perugia, Italy.

191

Wheater, H. and Evans, E. (2009). Land use, water management and future flood risk. Land Use Policy, 26(Supplement 1):S251–S264, doi:10.1016/j.landusepol.2009.08.019. Wilby, R. L., Hay, L. E., Gutowski Jr., W. J., Arritt, R. W., Takle, E. S., Pan, Z., Leavesley, G. H., and Clark, M. P. (2000). Hydrological responses to dynamically and statistically downscaled model output. Geophysical Research Letters, 27(8):1199–1202. 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(1–3):189–216, doi:10.1023/B:CLIM.0000013685.99609.9e.

192

Dankwoord

Veel mensen hebben op enige manier direct of indirect bijgedragen aan mijn promotieonderzoek waar dit proefschrift het resultaat van is. Ik kan onmogelijk iedereen persoonlijk bedanken. Als ik terug kijk, besef ik me dat niet alleen mijn omgeving belangrijk was in de jaren waarin ik daadwerkelijk werkte aan mijn onderzoek, maar ook een aantal personen in de periode daarvoor. Dit proefschrift is opgedragen Jaap Griede. Na een warm onthaal op de VU “Welkom vanaf de overkant van de schitterende IJssel!”, heeft Jaap in de jaren die volgden als studieco¨rdinator, docent en veldwerkbegeleider een passie voor aardwetenschappen o overgebracht waarvoor ik hem dankbaar ben. Zoveel docenten brachten ons kennis bij, maar hij leerde mij om in het veld enorm te genieten door ontzag te hebben voor de processen die het landschap gevormd hebben. En ´´n van die processen? Water. ee Jaap schreef op het bord: ‘ZELF !’. Ik zie het steeds voor me als ik mezelf aan het werk moet zetten. Ik hoop dat Jaap rust in vrede. Mijn specialisatie werd geografische hydrologie. Ik voerde twee boeiende onderzoeken uit onder begeleiding van Sampurno Bruijnzeel, met veldwerken in Puerto Rico en Sloveni¨. Mijn ervaring met het meten van neerslag, wind, afvoer en andere hydroloe gische processen is zeer waardevol. Het heeft inzicht gegeven in het belang van –en de problemen bij– data verzamelen, en in de gerelateerde onzekerheden. Met Sampurno heb ik een eerste stap gemaakt in het wetenschappelijk publiceren, bedankt hiervoor! Daarna volgden een paar jaar advies- en onderzoekswerk bij WL|Delft Hydraulics, waarbij het gehate wekelijkse tijdschrijven inzicht gaf in mijn eigen effectiviteit. Noodzakelijke competenties als budgetteren en plannen werden in deze tijd diep verankerd, en hebben zeker bijgedragen aan de succesvolle afronding van dit proefschrift. Tijdens mijn promotie ging het WL op in Deltares, dat zich ontwikkelde tot een prachtig onderzoeksinstituut. Er zijn veel WL en Deltares collega’s waarvan ik werk heb gebruikt of die mij hebben ge¨ ınspireerd, maar ik wil een aantal mensen in het bijzonder be193

194

Dankwoord

danken, waaronder mijn voormalige kamergenoten Eelco Verschelling en Paul Baan. Eelco, bedankt voor je hulp bij het modelleren met SOBEK, en het wegwijs worden in de wereld van de civiel ingenieurs. Paul, bedankt voor onze discussies en jouw kennis over waterbeheer, economie, risicoperceptie en organisatiemanagement. Albrecht Weerts, Marcel Ververs en Simone Patzke, bedankt voor jullie hulp met HBV en GRADE. Marjolijn Haasnoot, bedankt voor je luisterend oor en adviezen. Tot slot wil ik mijn voormalige en huidige afdelingshoofd bedanken voor het faciliteren van mijn onderzoek, Rinus Vis en Cees van de Guchte. Pier Vellinga introduceerde me bij Jeroen Aerts op het IVM, wat resulteerde in een promotietraject en de fantastische combinatie en luxe van twee werkgevers. Jeroen, enorm bedankt voor het gestelde vertrouwen en de ruimte die je bood om mijzelf te ontwikkelingen. Ik had me geen betere promotor kunnen wensen, we hebben goed samengewerkt en ik heb enorm veel van je geleerd. Het was een verademing om zoveel op reis te kunnen voor internationaal congresbezoek, wat mij veel kennis, creativiteit en inspiratie opleverde, en een groot netwerk. Je energie en enthousiasme werken aanstekelijk. We hebben echter een verschillend inzicht wanneer het gaat om klassieke auto’s: Mercedes Benz vs. Saab, ik denk dat we agree to disagree. Mijn twee co-promotoren die ik uiteraard dankbaar ben, zijn Han Dolman, hoofd vakgroep hydrologie en geo-milieuwetenschappen (VU - FALW) en Jaap Kwadijk, unitmanager zoetwatersystemen Deltares. Han, ik had graag meer contact met de vakgroep willen hebben, maar met twee werkgevers en drie werkplekken was daar te weinig ruimte voor. Met name in het begin heb je geholpen mijn promotieonderzoek op de rails te zetten, en later was je een belangrijke wijze en objectieve spiegel. Je dacht daarbij alleen aan mijn belang en het wetenschappelijke werk, en niet aan de vaak meer praktische belangen van Deltares, IVM of KvR. Dit heeft mij geholpen om on track te blijven. Jaap, jouw rol werd juist groter naar het einde toe van mijn promotie. Ik bewonder je vermogen om inhoud en praktijk te verbinden, en je komt altijd met een origineel inzicht of idee. Als het niet over klimaatverandering en water gaat, dan wel over de beste toerski- of klimroutes. De combinatie van verschillende disciplines en vaak jonge en enthousiaste collega’s maakt de sfeer op het IVM bijzonder prettig. Er heerst nieuwsgierigheid en grote maatschappelijke betrokkenheid. Met name mijn SPACE collega’s, maar eigenlijk iedereen, bedankt hiervoor. Mijn meest directe mede promovendus in het ACER project, en kamergenoot op het IVM, is Gert Becker. Gert bleek al een keer gepromoveerd, en gepensioneerd na een imposante internationale carri´re in de chemische e industrie. Wat bracht dat een onverwachte schat aan levenservaring en kennis over Duitsland! Gert, bedankt voor onze samenwerking en discussies, en voor jouw belangrijke rol in de workshops in Duitsland. Hans de Moel en Philip Bubeck bedankt voor jullie uitgebreide inhoudelijke bijdragen aan dit proefschrift. Philip Ward, bedankt voor je reviews. Afdelingshoofden Ron Janssen en Peter Verburg, bedankt voor jullie ondersteuning.

195

Dan zijn er nog verschillende mensen bij andere instituten aan wie ik dank verschuldigd ben. Bij de Waterdienst Rita Lammersen en Hendrik Buiteveld voor het beschikbaar stellen van data, rapporten en modellen. Bij het KNMI Alexander Bakker voor het bewerken van meteorologische data. Jasper Dekkers (VU - FEWEB) voor het werk aan de Land Use Scanner. Mailin Eberle, Bundesanstalt f¨r Gew¨sserkunde, u a thank you for sharing HBV Rhine insights and co-authoring. I am also grateful to the members of the reading committee. Thank you for your time, remarks and compliments during the evaluation of my thesis. Al dit werk zou ik niet kunnen doen zonder mijn vrienden. Jullie helpen relativeren, en maken dat ik extra geniet van het leven. Een aantal mensen in het bijzonder bedankt voor jullie hulp op moeilijke momenten het afgelopen jaar, voor jullie interesse, en voor het medeleven van diegenen die hun boekje al enige tijd zien verstoffen in de kast: Aard-Jan, Bas, Chantal, Cor, Eva, Gideon, Janneke, Joost, Jos, Malika, Marike, Nanja en Saskia. En mijn paranimfen, Laurens en Anouk, wil ik in het bijzonder bedanken. Laurens, jij hebt mede bepaald dat ik bij het IVM kon en wilde werken, en ik heb veel kunnen leren van jouw publicaties en schrijfstijl. Ik heb ontzag voor de resultaten die je behaalt en je manier van werken en denken, die ik ook ken uit onze bestuurstijd. Bedachtzaam en nauwkeurig. Anouk, jouw ambitie en doorzettingsvermogen werkte en werkt nog altijd aanstekelijk, en heeft al tijdens onze studie mijn werkhouding ten goede veranderd. Of we nou aan het studeren waren, of GeoBlont barkrukken fabriceerden, jij gaat door tot het af is. Ik zou al blij zijn met 10% van jouw energie! De tijd heeft geleerd dat onze vriendschap niet zomaar kapot kan, en al vind ik het aantal kilometers water en aarde tussen onze woonplaatsen veel te veel, het levert leuke en boeiende vakantieadressen op. Veel heb ik te danken aan mijn ouders. Papa en mama, jullie zijn fantastische en lieve ouders en hebben altijd alles voor Arno en mij over gehad. Jullie hebben mij altijd gesteund, en zijn zo enthousiast over mijn studie en werk, dat jullie zelfs met Jaap het veld in zijn geweest. Mijn schoonfamilie wil ik bedanken voor hun warme ontvangst en interesse in mijn werk. Als laatste om te bedanken, maar de eerste in alle andere opzichten, is mijn lieve Sophie. Jij bracht rust en ruimte in mijn hoofd, en liefde in mijn hart. Sinds wij samen zijn kreeg mijn leven meer kleur en minder kant en klaar maaltijden. Ik leer nu over de witbalans, het puberend brein en het doorbreken van de vierde wand. Jij roept vol overtuiging: “Dolomiet!”, herkent middenmorenes in Chili en hangende dalen in IJsland. Ik kan nog een boek schrijven, maar dan alleen over jou. Het resultaat zal echter nooit zo mooi zijn als de songtekst van Frans Halsema, gezongen door Claudia: ‘Zij is meer dan deze woorden zeggen. In mijn lichaam heeft ze plaats gemaakt voor twee. Maar wie weet een wonder uit te leggen. En een wonder draag ik met me mee.’

196

Dankwoord

Publications related to this thesis

Peer-reviewed articles: Te Linde, A.H., Bubeck, P., Dekkers, J.E.C., De Moel, H. and Aerts, J.C.J.H., 2011. Future flood risk estimates along the river Rhine. Natural Hazards and Earth System Sciences, 11: 459–473, doi: 10.5194/nhess-11-459-2011. Te Linde, A.H., Aerts, J.C.J.H. and Kwadijk, J.C.J., 2010. Effectiveness of flood management strategies on peak discharges in the Rhine basin. Journal of Flood Risk Management, 3: 248–269, doi: 10.1111/j.1753-318X.2010.01076.x. Te Linde, A.H., Aerts, J.C.J.H., Bakker, A.M.R. and Kwadijk, J.C.J., 2010. Simulating low probability peak discharges for the Rhine basin using resampled climate modeling data. Water Resources Research, 46 (WR03512), doi: 10.1029/2009WR007707. Te Linde, A.H., Aerts, J.C.J.H., Bakker, A.M.R. and Kwadijk, J.C.J., 2010. Simuleren van extreme piekafvoeren op de Rijn voor verschillende klimaatscenario’s. Stromingen, 16(2): 1–16 (in Dutch). Te Linde, A.H., Aerts, J.C.J.H., Hurkmans, R.T.W. and Eberle, M., 2008. Comparing model performance of two rainfall-runoff models in the Rhine basin using different atmospheric forcing data sets. Hydrololgy and Earth System Sciences, 12: 943–957. Conference proceedings: Te Linde, A.H., De Moel, H. and Bubeck, P., 2010. Future flood risk in the Rhine basin. In: Proceedings of HydroPredict 2010, Prague, Czech Republic, 20–23 September, 2010. Te Linde, A.H. and Aerts J.C.J.H, 2008. Climate change and adaptive capacity to extreme events in the Rhine basin. In: Van Os, A.G. and Erdbrink C.D. (Ed), NCRdays 2008: 10 years NCR, Dalfsen, the Netherlands, 20–21 November 2008.

197

198

Publications related to this thesis

Te Linde, A.H. and Aerts J.C.J.H, 2008. Simulating flood-peak probability in the Rhine basin and the effect of climate change. In: Samuels, P., Huntington, T.G., Allsop, W. and Harrop, J. (Ed), Flood Risk Management. Research and Practice. CRC Press, Oxford, UK, 30 September–2 October, 2008. Te Linde, A.H., Aerts, J.C.J.H. and Van den Hurk B., 2008. Effects of flood control measures and climate change in the Rhine basin, In: Simonovic, S.P. et al. (Ed), Managing flood Risk, Reliability and Vulnerability. Proceedings of the 4th International Symposium on Flood Defence, Toronto, Ontario, Canada, 6–8 May, 2008. Te Linde, A.H. Effects of climate change on discharge behaviour of the river Rhine. In: Heinonen, M. (Ed), Proceedings of the Third International Conference on Climate and Water, Helsinki, Finland, 3–6 September, 2007. Poster presentations: Te Linde, A.H. and Bubeck, P., 2009. Future flood risk estimation in the Rhine basin. Presented at AGU Fall Meeting 2009, San Francisco, California, USA, 14-18 December 2009. Te Linde, A.H., Becker, G., Bubeck, P. and Aerts, J.C.J.H., 2008. A7 ACER: developing Adaptive Capacity to Extreme events in the Rhine basin. Presented at International Conference Climate Changes Spatial Planning, The Hague, the Netherlands, 12–13 September, 2008. Te Linde, A.H., Hurkmans, R., Aerts, J.C.J.H. and Dolman, A.J., 2007. Comparing model performance of the HBV and VIC models for the Rhine basin. Presented at EGU 2007, Vienna, Austria, 15-20 April 2007. Book chapters: Te Linde, A.H., De Moel, H. and Aerts, J.C.J.H., forthcoming. Towards risk-based flood management to accommodate future uncertainties, and an outlook to the assessment of drought risk. In Brils, J. and Vermaat, J. (Ed), Towards Risk-Based Management of River basins. Springer. Te Linde, A.H., Hurkmans, R., Aerts, J.C.J.H. and Dolman, A.J., 2007. Comparing model performance of the HBV and VIC models in the Rhine basin. In: Boegh E. et al. (Ed), Quantification and Reduction of Predictive Uncertainty for Sustainable Water Resource Management. IAHS Publication 313, 278–285, Perugia, Italy, 2–13 July 2007.

199

Technical reports: Te Linde, A.H., 2009. Modeling the effect of flood management strategies in the Rhine basin. Institute for Environmental Studies, VU University, Amsterdam, the Netherlands. Van Kreveld, A., Van Winden, A., Te Linde, A., Zwolsman, G., Jacobs, P., Souwerbren, P. and Gilbert, A., 2009. Het Benedenrivierengebied in tijden van klimaatverandering. KvR 014/2009. Klimaat voor Ruimte, Amsterdam, the Netherlands (in Dutch). Te Linde, A.H., 2007. Effect of climate change on the discharge of the rivers Rhine and Meuse. Applying the KNMI 2006 scenarios using the HBV model. Q4286, Delft Hydraulics, Delft, the Netherlands. Other publications: Te Linde, A.H., 2009. Future flood risk in the Rhine basin. In: Heinen, M. (Ed), Climate Research Netherlands - Research Highlights. A Publication of the Research Programmes Climate Changes Spatial Planning and Knowledge for Climate. Wageningen, the Netherlands, pp. 42–44. Bakker, A.M.R. and Te Linde, A.H., 2008. Scenario voor de piekafvoeren van de Rijn bij Lobith. Geografie, november/december 2008, 12–14 (in Dutch). Aerts, J., Te Linde, A. and Moors, E., 2007. Lange termijn scenario’s voor het stroomgebied van de Rijn: de relatie Nederland-Duitsland. H2O, 22: 40 (in Dutch).

200

Publications related to this thesis

Curriculum Vitae

Aline te Linde was born on 21 April 1977 in Deventer, the Netherlands. She attended high school at the Baudartius College in Zutphen and received her VWO diploma in 1995. From 1995–2001 she studied Earth Sciences at the VU University in Amsterdam. Her studies contained numerous field works abroad and to graduate as a Geographical Hydrologist, she wrote a dissertation on Montane Cloud Forests in Puerto Rico (2000). Aline completed her master degree with a M.Sc. Thesis entitled: ‘Modelling Rainfall Interception by Two Deciduous Forests of Contrasting Stature in the Upper Dragonja Catchment, Slovenia’ (2001). After graduation, she joined WL | Delft Hydraulics in 2002 where she worked as a hydrologist on various scales. Projects ranged from water management assignments for Dutch waterboards and the National Government, to international projects on crossboundary river basin management. In October 2005, Aline started her PhD-research on climate change and adaptative capacity to extreme events in the Rhine basin, parttime, at the Institute for Environmental Studies (IVM), VU University Amsterdam. She arranged a co-operation between WL | Delft Hydraulics (now Deltares) and the IVM, in which both institutes supported her research and which allowed Aline to combine her research with project management and advisory work at Deltares. This resulted in the present thesis. 201

Sign up to vote on this title
UsefulNot useful

Master Your Semester with Scribd & The New York Times

Special offer: Get 4 months of Scribd and The New York Times for just $1.87 per week!

Master Your Semester with a Special Offer from Scribd & The New York Times