Simulation of Spatiotemporal Dynamics of Water Fluxes In

HYDROLOGICAL PROCESSES Hydrol. Process. 24, 3289–3306 (2010) Published online 7 July 2010 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/hyp.7753

Simulation of spatiotemporal dynamics of water ﬂuxes in Germany under climate change

Shaochun Huang,* Valentina Krysanova, Hermann Osterle¨ and Fred F. Hattermann Potsdam Institute for Climate Impact Research, PO Box 601203, Telegrafenberg, 14412 Potsdam, Germany

Abstract: In most of Europe, an increase in average annual surface temperature of 0Ð8 °C is observed, and a further increase is projected. Precipitation tends to increase in northern Europe and decrease in southern Europe, with variable trends in central Europe. The climate scenarios for Germany suggest an increase in precipitation in western Germany and a decrease in eastern Germany, and a shift of precipitation from summer to winter. When investigating the effects of climate change, impacts on water resources are among the main concerns. In this study, the first German-wide impact assessment of water fluxes dynamics under climate change is presented in a spatially and temporally distributed manner using the state-of-the-art regional climate model, Statistical Regional (STAR) model and the semi-distributed process-based eco-hydrological model, soil and water integrated model (SWIM). All large river basins in Germany (lower Rhine, upper Danube, Elbe, Weser and Ems) are included. A special focus of the study was on data availability, homogeneity of data sets, related uncertainty propagation in the model results and scenario-related uncertainty. After the model calibration and validation (efficiency from 0Ð6to0Ð9 in 80% of cases) the water flow components were simulated at the hydrotope level, and the spatial distributions were compared with those in the Hydrological Atlas of Germany. The actual evapotransipration is likely to increase in most parts of Germany, while total runoff generation may decrease in south and east regions. The results for the second scenario period 2051–2060 show that water discharge in all six rivers would be 8–30% lower in summer and autumn compared with the reference period, and the strongest decline is expected for the Saale, Danube and Neckar. Higher winter flow is expected in all of these rivers, and the increase is most significant for the Ems (about 18%). However, the uncertainty of impacts, especially in winter and for high water flows, remains high. Copyright  2010 John Wiley & Sons, Ltd.

KEY WORDS water ﬂuxes; water discharge; statistical downscaling model STAR; eco-hydrological model SWIM; climate change impact; Germany Received 23 October 2009; Accepted 12 April 2010

INTRODUCTION showed a higher increasing trend (19%), while summer The Fourth Assessment Report (AR4) of the International precipitation did not change significantly (Schonwiese¨ Panel for Climate Change (IPCC, 2007) summarized the et al., 2006). The recent development of annual precip- knowledge of climate trends in the 20th century. In most itation in Germany shows a distinct spatial pattern: an of Europe, an increase in average annual surface tem- increasing trend in the western part, and reductions over perature is observed (0Ð8 °C over the continent on aver- large areas in eastern and southeastern areas (Menzel age), with stronger warming in winter than in summer. et al., 2006). Increasing flood trends are found in the most The precipitation tended to increase in northern Europe parts of Germany during 1951–2002 (Petrow and Merz, (10–40%) and decrease in southern Europe (up to 20% 2009), with more significant changes in winter than in in some parts). Other wide ranging impacts have been summer. Besides floods, other extreme events are also documented as well, such as retreating glaciers, longer observed recently. For example, the extremely hot sum- growing season, shifts of species distribution patterns and mer of 2003 in Germany is characterized by a return impacts on human health. period of about 455 years (Schonwiese¨ et al., 2004). Located in the central Europe, Germany has already Under the changing climate, it is necessary and impor- been affected by climate change. According to tant to study the impacts on water resources, as the water Schonwiese¨ et al. (2006), the annual average temperature cycle in river basins is sensitive to changes in climate has increased by ca 1 °C between 1901 and 2000, and the characteristics. The water balance components, such as winter months became especially warmer. The winters of evapotranspiration, runoff and groundwater recharge, that 1980s and 1990s were observed as the warmest in Ger- determine river discharge and the availability of water many during the 20th century (Schonwiese¨ et al., 2006). resources, will be inevitably affected. Generally, more In 1901–2000, annual precipitation in Germany exhib- evapotranspiration can be expected due to the increased ited a moderate increase (9%), and winter precipitation temperature. However, the actual evapotranspiration is constrained by the actual water availability in soil, and higher temperature could result in lower actual evapotran- * Correspondence to: Shaochun Huang, Potsdam Institute for Climate spiration if water availability is low. The changes in river Impact Research, PO Box 601203, Telegrafenberg, 14412 Potsdam, Germany. E-mail: [email protected] discharge are determined by changes in precipitation and

Copyright  2010 John Wiley & Sons, Ltd. 3290 S. HUANG ET AL. temperature and the regional environmental settings, such Germany. The statistical regional climate downscaling as land use. Groundwater recharge reacts very sensitively model, statistical regional (STAR) model (Orlowsky to even small changes in precipitation and temperature, et al., 2008) was applied in the study to produce cli- especially in lowland regions (Hattermann et al., 2004). mate change scenarios, because it has better reproduction Rising temperature will result in a longer vegetation of the historical climate conditions and hence more reli- growth period and higher potential evapotranspiration. As able scenarios than other regional climate models (RCMs) a result, lower groundwater recharge could be expected (Gerstengarbe et al., 2009). Besides, STAR generates as a consequence. In addition, the early coming of spring multiple realizations for each scenario condition, which in the future will shorten the groundwater recharge period allows the production of more robust and reliable results in winter. accounting for the inherent uncertainty of the climate Any changes in water balance components will influ- scenario. The eco-hydrological process-based model, soil ence the availability of regional water resources and and water integrated model (SWIM) (Krysanova et al., impact economic sectors, such as water management, 1998) was applied sequentially for the five river basins agriculture, forestry, tourism, hydropower production and (Ems, Weser, Elbe, upper Danube and lower Rhine) to river transport, as well as nature conservation and health. simulate water flow dynamics and the water flow com- For example, drier summer could lead to water deficit ponents. An important advantage of SWIM is that the in agriculture, and increase in annual river transportation model integrates hydrological cycle with vegetation pro- costs. More money may be needed for increasing the cesses and takes into account interactions between water design discharge and water level for river safety (Mid- fluxes and ecosystems. Besides, it allows changing sea- delkoop and Kwadijk, 2001). sonal timing of plant growth stages under warmer con- In order to indentify and investigate the effects of ditions. Therefore, it represents a more reliable tool for climate change on the water cycle, ecosystems and human climate impact assessment compared to the pure hydro- being, and then to develop coping strategies for the future, logical models. numerous studies focused on the influence of climate Application of a process-based river basin model for change on water resources have been carried out around such a large regional scale as Germany is a novelty and the world. The common approach is to use hydrological a challenge, because the data availability and heteroge- models driven by the projected climate scenarios for neous data sets (especially in the transboundary rivers the future. Many of these studies applied conceptual such as the Elbe, upper Danube and lower Rhine) cre- precipitation–runoff models with simple water balance ate problems and require non-standard solutions. Hence, components (e.g. Menzel and Burger,¨ 2002; Arnell, 2003; the experience of a large-scale model application and the Albek et al., 2004; Drogue et al., 2004; Thodsen, 2007), problems in simulating the international river basins are and process-based hydrological models (Muttiah and presented here, in addition. Wurbs, 2002; Krysanova et al., 2005; Hattermann et al., 2008) at the catchment scale. In Germany, several projects have been launched at STUDY AREA the catchment level aiming to develop strategies that can be applied in the future to reduce the vulnerabil- Germany is located in central Europe, bordering the ity or adapt to climate change. The projects GLOWA- North Sea and Baltic Sea, with the total area of Elbe (http://www.glowa-elbe.de) and GLOWA-Danube 357 021 km2. Generally, the German territory is divided (http://www.glowa-danube.de) are two examples of com- geographically into the North German Lowlands, the prehensive research on climate change impacts in the Central German Upland, the Southwest Rhine River Elbe and Danube river basins. Several papers were pub- Valley, the Alpine Foreland and the German Alps lished focusing on the impacts of climate change on water (Figure 1a). From the Northwest to the East and South- fluxes in different river basins of Germany (e.g. Menzel east, the maritime climate gradually changes into a more and Burger,¨ 2002; Krysanova et al., 2005; Hennegriff continental climate. The country’s average annual tem- et al., 2008; Mauser and Bach, 2009). However, these perature is about C9 °C, and the prevailing winds are catchment-scale projects do not provide an overview of westerly. The precipitation occurs in all seasons with sub- the climate impacts on the water sector for the whole stantial regional differences. In the North German Low- Germany. lands, annual rainfall varies between <500 (continental) The country-wide climate impact assessment using a to about 700 mm (maritime). The Upland mountainous process-based hydrological model is a challenge. It is areas receive from about 700 to >1500 mm of precipita- important because it has a potential to support an overall tion per year, and the Alps >2000 mm/year (Statistisches decision making at the country scale and climate change Bundesamt Deutschland, 2008). adaptation strategies in different sub-regions, such as The German territory is comprised of five large river continental, maritime and alpine climate regions. basins (the Elbe, upper Danube, Rhine, Weser and Ems), In this study, the main objectives were (1) to eval- three medium-scale basins in the coastal area (Elder, uate changes in the seasonal dynamics of water fluxes Schlei/Trave and Warnow/Peene) and small parts of the and (2) to assess spatial changes in water balance com- Oder and Maas basins (Figure 1b). Only the Ems and ponents under climate change for the whole territory of Weser basins lie entirely within the national German

Copyright  2010 John Wiley & Sons, Ltd. Hydrol. Process. 24, 3289–3306 (2010) SIMULATION OF WATER FLUXES IN GERMANY UNDER CLIMATE CHANGE 3291

Figure 1. Digital Elevation Map of Germany and German sub-regions (a), and the studied river basins with locations of the gauge stations used for the model calibration and validation (b)

Table I. Characteristics of ﬁve river basins chosen as case study areas

River Studied Large First gauge Other main Land use shares Number of basin drainage tributaries in on the border countries (%) climate stations area (km2) Germany as input data included Cropland Forest Grassland In Outside Germany Germany

Ems 13 000 Hase — No 66 10 15 63 — Weser 45 725 Werra, Fulda, — No 49 30 14 309 — Aller, Leine Danube 77 107 Inn, Naab, Isar, — Austria, 32 37 20 482 10a Salzach Switzerland Rhine 123 175 Neckar, Main, Rheinfelden France, 40 38 12 844 14a Moselle Luxemburg, Belgium Elbe 147 423 Havel, Saale, — Czech Republic 51 30 10 399 48 Schwarze Elster, Mulde a These climate data were from a gridded data set. territory, and these two basins are the smallest among Altogether, they cover about 90% of the whole German the five. The Rhine, upper Danube and Elbe drainage territory. basins have large parts outside Germany. The smallest river basin Ems (Figure 1b) is located in In our study, the five main river basins covering the the northwest of Germany. The entire basin has low relief territory of German (the Danube, Elbe, Ems, Rhine, terrain and the river flows through the North German and Weser) were selected as the major study areas Lowlands to the North Sea. This basin belongs to one of for the assessment of climate change impacts. The the most intensively used agricultural regions in Europe. characteristics of these five basins are listed in Table I. Arable land covers approximately 66% of the area. The

Copyright  2010 John Wiley & Sons, Ltd. Hydrol. Process. 24, 3289–3306 (2010) 3292 S. HUANG ET AL. other major land covers are forest (10%) and grassland in Germany, Rees (Figure 1b). For the Rheinfelden, (15%). discharge data from Switzerland were used as input. The The Weser basin (Figure 1b) is located in the north- water components were also calculated for the coastal western Germany. The Weser is the longest German river area and German parts of the Oder and Mass basins using whose basin is located fully within German territory. the parameterization of the neighbouring large basins. Formed by the rivers Fulda and Werra, it flows through the North German Lowlands, and reaches the North Sea. About half of the drainage basin area is used as arable MATERIALS AND METHODS land, 30% is covered by forest and 14% by grassland. The Elbe River (Figure 1b) originates in the Czech Statistical regional model Republic, drains across eastnorthern Germany and flows into the North Sea. About two thirds of the whole Elbe For projections of future changes in water flows for drainage basin (approximately 100 000 km2) are located the whole of Germany, robust regional climate change in Germany, one-third in the Czech Republic and the scenarios should be applied. As the General Circulation negligible parts belong to Austria and Poland. The Elbe Models (GCMs) cannot provide climate information with basin is also an intensive agricultural region with about sufficient spatial resolution for regional studies due to 50% of the total area used as arable land. About 30% their coarse horizontal resolution, several RCMs were of the drainage area is under forest and only 10% under developed in the recent years in Germany. There are so- grassland. called ‘dynamical downscaling models’, such as REMO The total Rhine River basin (Figure 1b) is distributed (Jacob, 2001) and CCLM (Bohm¨ et al., 2008). Besides, in nine countries. Beginning in the Swiss Alps, the there are also other approaches, such as a combination Rhine River flows through Germany to the Netherlands. of a statistical with an analogous downscaling approach, The basin includes small parts located in France, Aus- WettReg (Enke and Spekat, 1997) and a statistical tria, Italy, Liechtenstein, Luxembourg and Belgium. The downscaling technique, STAR (Orlowsky et al., 2008). rivers Main, Neckar and Moselle are the three main tribu- Gerstengarbe et al. (2009) compared these four models taries in Germany, and the Moselle receives drainage also for the historical period evaluating the agreement between from France, Luxembourg and Belgium. Two thirds of the the simulated and observed values. The evaluation results Rhine drainage basin area are situated in Germany, and show that each model has its strengths and weaknesses. the Alpine countries, of which Switzerland is the largest, The dynamical models REMO and CCLM generate form about 20% of the drainage area. The arable land largest deviations between the observed and simulated (40%) and forest (38%) are the two major land cover precipitation. Reproduction of precipitation is a common types in the German part of the Rhine basin. problem of all RCMs, partly caused by their relatively The upper part of the Danube basin upstream the low resolution and partly by their inherent difficulties gauge Achleiten (Figure 1b) is formed by Brigach and to reproduce precipitation dynamics. There are also Breg rivers located in southwest Germany, and several significant problems to reproduce trends with the models tributaries from the Alps and the Bavarian forest. The REMO, CCLM and WettReg. In contrast, the simulated main tributary Inn flows from the Swiss Alps through outputs from the model STAR have better agreement Austria to Germany and receives a large amount of cold with the observed statistics, especially for temperature melting water from snow and glaciers. About 73% of the and air pressure. The trends of climate variables are well drainage area above the gauge Achleiten is located within reproduced by STAR and the characteristics of events are the German territory. The main land cover types in this closer to the reality. basin are forest (37%) and arable land (32%). The upper The model STAR has also some weaknesses. For Danube has the highest grassland cover (20%) compared example, STAR fully relies on the historical climate data with other basins. and uses them for resampling, and it is principally not In order to assess the changes in river discharge able to reproduce extreme events (e.g. heavy precipita- and water flow components under climate changes’ tion) exceeding the already observed values. So, all the scenario, hydrological processes for the whole river projected extreme events will not exceed the extreme basins need to be simulated, and the calibration and events observed in the past. The second weakness of validation should be performed first. Therefore, the STAR is that this statistical model relies on large amount model setup should include the German areas and, in of observed historical data. Hence, it cannot be applied several cases, parts of other countries as well. In our in the poor-data regions. study, three of the five large basins: the Elbe, Ems and Currently, the climate scenarios generated by STAR Weser were simulated fully. Only the upper part of the are only available for the German territory due to lack Danube, namely, the area upstream of the last gauge in or non-availability of good historical data for other Germany Achleiten (Figure 1b), was considered in the neighbouring countries. So, taking into account the main study. Due to lack of land use data from Switzerland, the focus of this study and the characteristics of different simulation for the Rhine basin could only be performed climate models, the climate scenario produced by the for the area downstream of the gauge Rheinfelden statistical downscaling model STAR was applied for the located at the Swiss-German border until the last gauge assessment. In the future, other downscaling methods will

Copyright  2010 John Wiley & Sons, Ltd. Hydrol. Process. 24, 3289–3306 (2010) SIMULATION OF WATER FLUXES IN GERMANY UNDER CLIMATE CHANGE 3293 also be used for cross-comparison and extreme events basis of the models SWAT (Arnold et al., 1993) and analysis. MATSALU (Krysanova et al., 1989). Compared with other RCMs, which can project SWIM simulates hydrological cycle, vegetation growth 100 years or longer climate scenarios for Germany, and nutrient cycling with the daily time step by disaggre- STAR was developed for the medium-term (about gating a river basin to subbasins and hydrotopes. The 50–60 years) regional climate projections due to its sta- hydrotopes are sets of elementary units in a subbasin tistical analogue resampling technique. Analogue with homogeneous soil and land use types. Up to ten approaches such as STAR assume that observations of vertical soil layers can be considered for hydrotopes. It a given day from the training period can occur again or is assumed that a hydrotope behaves uniformly regarding in a similar way during the future period. Hence, simu- hydrological processes and nutrient cycling. The spatial lated series are constructed by resampling from segments disaggregation scheme in the model is flexible. In the of observation series, consisting of daily observations. regional studies, climate zones, grid cells of a certain Generating a future series can thus be seen as defin- size or other areal units can be used for disaggregating a ing a date-to-date-mapping, by which each date of the region instead of subbasins. future period is assigned a date and the concurrent mete- Water flows, nutrient cycling and plant growth are orological observations of the training period. No trend calculated for every hydrotope. Then lateral fluxes of elimination or any other modification is applied to the water and nutrients to the river network are simulated observational data prior to resampling. The advantage taking retention into account. After reaching the river of such resampling is that the physical consistency of system, water and nutrients are routed along the river both the spatial fields and the simultaneous combina- network to the outlet of the simulated basin. tions of different weather parameters is guaranteed. The The simulated hydrological system consists of four STAR resamples in blocks of 12 days, which ensures the control volumes: the soil surface, the root zone of soil, projected future time series with realistic persistence fea- the shallow aquifer and the deep aquifer. The soil root tures. zone is subdivided into several layers in accordance with One of the most important properties of STAR is that the soil database. The water balance for the soil surface the produced climate time series are forced only by the and soil column includes precipitation, surface runoff, linear temperature trend of the future period. Once the evapotranspiration, subsurface runoff and percolation. daily mean values of a long-term observed time series The water balance for the shallow aquifer includes are obtained, it is possible to impose the assumed trend groundwater recharge, capillary rise to the soil profile, onto the series and to create the simulated series com- lateral flow and percolation to the deep aquifer. plying with this trend. The scenario of future climate conditions used in this study was constructed by using Surface runoff is estimated as a nonlinear function of the trend of temperature derived from climate change precipitation and a retention coefficient, which depends scenario A1B produced by ECHAM 5 (Roeckner et al., on soil water content, land use and soil type (modifi- 2003). It was used to generate the corresponding modified cation of the Soil Conservation Service curve number temperature series. The time series of other climate vari- method, Arnold et al., 1990). Lateral subsurface flow (or ables, such as precipitation, radiation, humidity, etc. were interflow) is calculated simultaneously with percolation. generated by using the values recorded on the same day It appears when the storage in any soil layer exceeds as the temperature measurement. Therefore, the method field capacity after percolation and is especially important maintains the stability of the main statistical characteris- for soils having impermeable or less permeable layer(s) tics (variability, frequency distribution, annual cycle and below several permeable ones. Potential evapotranspira- persistence). For the spatially differentiated projections, tion is estimated using the method of Priestley–Taylor climatological sub-regions are identified (e.g. by meteo- (Priestley and Taylor, 1972), although the method of rological stations), and an individual temperature trend Penman–Monteith (Monteith, 1965) can also be used. is prescribed for each subregion, representing spatial pat- Actual evaporation from soil and actual transpiration by terns of future climate parameters. plants are calculated separately. In addition, STAR is much faster in computation time The module representing crops and natural vegeta- than the dynamical climate models, so it is able to tion is an important interface between hydrology and generate multiple climate projections by implementing nutrients. A simplified EPIC approach (Williams et al., a random process (Monte Carlo simulation). Therefore, 1984) is included in SWIM for simulating arable crops an ensemble of 100 realizations of the climate change (like wheat, barley, rye, maize and potatoes) and aggre- scenario was generated and applied in this study. This gated vegetation types (like pasture, evergreen forest and allows evaluating uncertainty of climate change impact mixed forest), using specific parameter values for each related to the climate scenario. crop/vegetation type. A number of plant-related parameters are specified for 74 crop/vegetation types in the Soil and water integrated model database attached to the model. Vegetation in the model The dynamic process-based eco-hydrological model affects the hydrological cycle by the cover-specific reten- SWIM (Krysanova et al., 1998) was developed for cli- tion coefficient, impacting surface runoff and influenc- mate and land use change impact assessment on the ing the amount of transpiration, which is simulated as

Copyright  2010 John Wiley & Sons, Ltd. Hydrol. Process. 24, 3289–3306 (2010) 3294 S. HUANG ET AL. a function of potential evapotranspiration and leaf area subbasin map for the Czech Republic (T.G.M. Water index (LAI). Research Institute, 2005) were available. On the basis of Interception of photosynthetic active radiation (PAR) is the DEM and the stream network, an average drainage estimated as a function of solar radiation and LAI. The area of 100 km2 was chosen as a threshold to discretize potential increase in biomass is the product of absorbed the areas in the Danube and Rhine basins outside Ger- PAR and a specific plant parameter for converting energy many into subbasins, because the standard subbasin map into biomass. The potential biomass is adjusted daily if for Germany had approximately the same discretization. one of the four plant stress factors [water, temperature, The land use map was obtained from the CORINE nitrogen (N) and phosphorus (P)] is <1Ð0, using the 2000 land cover data set of the European Environment product of a minimum stress factor and the potential Agency. Nine land cover types were considered in the biomass. The water stress factor is defined as the ratio study: water, urban areas, cropland, grassland, forest of actual to potential plant transpiration. The temperature coniferous, forest deciduous, forest mixed, wetland and stress factor is computed as a function of daily average, bare soil. No changes in land use patterns were assumed optimal and base temperatures for plant growth. The N for the reference and scenario periods in this study, and and P stress factors are based on the ratio of accumulated land use was considered to be ‘static’. This was done N and P to the optimal values. The LAI is simulated as a on purpose, in order to investigate the ‘pure’ impact function of a heat unit index (ranging from 0 at planting of climate change on water fluxes, without influence of to 1 at physiological maturity) and biomass. changing land use patterns. SWIM allows application of the complicated crop For the subbasins in Germany and Czech Republic, rotation schemes including several crops like wheat, climate data (temperature, precipitation, solar radiation barley, rye, maize, potatoes, etc., which could be made and air humidity) were interpolated to the centroids of specific for different sub-regions or federal states, and every subbasin by the inverse distance method using data differentiated for soil types. However, in this study, a from 2342 climate and precipitation stations (Figure 2). more robust although realistic enough crop rotation was However, climate stations with available climate data applied, as the main aim was to assess the impact of in the Czech Republic were much sparser than that in climate change. In the reference (current) conditions, Germany. winter crop like winter wheat or winter barley (dominant For other areas in the five considered river basins, that crops in Germany) is planted and harvested according to are located outside of Germany (France, Austria, Lux- the current practice schedule. There is usually a cover emburg, Figure 2), the available observed climate data crop growing between the harvest and next planting were even more poor: sparsely located virtual gridded of winter crop. In warmer (scenario) conditions, the ‘stations’ with daily temperature and precipitation data scheduling of agriculture crops is governed by harvest only. So, the daily temperature and precipitation data index, so the rotation scheme can be changed when the winter crop is harvested earlier than in the current condition. In this case, SWIM would allow earlier growth of cover crop right after the harvest of winter crop and let it grow until the next winter crop planting date.

Data preparation To derive the subbasin and hydrotope structure and the routing structure of the ﬁve basins, four spatial maps: the digital elevation model (DEM), the soil map, the land use map and the subbasin map were stored in a grid format with 250 m resolution. This resolution was proved to provide reliable results in previous studies for large river basins (Hattermann et al., 2007a). The DEM was provided by the NASA Shuttle Radar Topographic Mission (SRTM). The soil map of the study area was merged from the general soil map of the Federal Republic of Ger- many ‘BUK¨ 1000’ produced by the Federal Institute for Geosciences and Natural Resources (BGR), soil map of the Czech Republic (Koskova´ et al., 2007) and soil map from the European soil database (European Communities—DG Joint Research Centre). The data quality and their resolution were different, which could be reﬂected in the modelling results. Figure 2. Location of the climate and precipitation stations in Germany The standard subbasin map for Germany from the Fed- and Czech Republic (black), and the gridded climate data (grey cross) eral Environment Agency (Umweltbundesamt), and the which were available for the study

Copyright  2010 John Wiley & Sons, Ltd. Hydrol. Process. 24, 3289–3306 (2010) SIMULATION OF WATER FLUXES IN GERMANY UNDER CLIMATE CHANGE 3295 from the ‘Daily high-resolution gridded climate data set the simulation results, especially for the transboundary for Europe’ (www.ensembles-eu.org) were applied in this river basins with heterogeneous data sets. The quality of study for areas outside Germany and Czech Republic, data input will directly influence the simulation results, and other needed climate parameters required for SWIM and lead to difficulties in performing a sound evaluation (solar radiation and air humidity) were interpolated using of the model outputs. In our study, this problem was the records from the closest German climate stations. The solved by providing a part of results for the total German inverse distance method was used for the interpolation. territory, and restricting the study area by the solely Obviously, such interpolation is very uncertain and can national large representative subbasins for the climate produce large errors in the climate input data generated impact assessment. for France, Austria and Luxemburg, which will definitely propagate in the modelling results. Model calibration and validation procedure In general, the problems with climate data for areas outside Germany could be solved. Necessary data of The calibration procedure was carried out for five main observation were not available for this study, but most discharge gauges (Table II) for each of the five river probably they exist and could become available for basins in the period from 1981 to 1990. The parameter impact assessments at the scale of international river estimation, routine PEST (Doherty, 2004) was applied to basins in the future. calibrate the simulated discharge. The simulation period As the land use map for Switzerland is not included was then extended to 20 years from 1961 to 1980 to in the CORINE data base, and no observed climate data validate the simulation results at the same five gauges. In were available for Switzerland, the modelling setup for addition, 24 intermediate gauges (Table II and Figure 1b) the Swiss part was restrained, and therefore the upper at the main tributaries and the main rivers (mostly with 2 Rhine (upstream of the gauge Rheinfelden) had to be the drainage areas larger than 5000 km ), for which the excluded from the simulation. model was not calibrated, were included in the validation The climate scenario produced by STAR was avail- procedure in order to verify the spatial performance of able only for the meteorological stations in Germany (due SWIM for the period 1981–1990. After calibration and to lack of historical data with similar density and com- validation, the whole simulation period 1961–1990 was pleteness outside of Germany). Hence, the simulation of considered as the reference, and the model outputs in climate change impacts on water dynamics in the sce- the reference period were compared with those in two nario periods was only possible for the German territory. scenario periods: 2009–2018 and 2051–2060. Therefore, water flow components such as runoff, evap- For the basins Ems, Weser and Elbe, where the rivers otranspiration and groundwater recharge were analysed flow into the North Sea, the gauges Versen, Intschede for the whole German territory, whereas river discharge and Neu-Darchau (see location at Figure 1b) were used for the three largest basins: the Elbe, Rhine and Danube for the calibration, because these are the last gauges not was analysed in the reference and scenario periods for influenced by the tidal effect. the selected representative gauges, whose catchments are For the Danube basin, the gauge Hofkirchen was located fully in Germany. Namely, climate change sce- selected as the calibration gauge, instead of the last narios in the Elbe were analysed for the Saale subbasin discharge gauge Achleiten, because about 96% of the (one of the largest and most important subbasins in the catchment of the former is located in Germany, whereas German part of the Elbe basin), the Rhine basin was that of the latter includes also a large part of Austrian represented by the Main and Neckar subbasins, and the territory with quite poor available climate data (see Danube basin was represented by the intermediate gauge explanation in section on Data preparation). By that, the Hofkirchen, whose drainage area is located mostly in error caused by the poor climate data from Austria could Germany. be minimized. In summary, a good data base including all necessary For the Rhine basin, the gauge Frankfurt-Osthafen data for modelling with SWIM and scenario analysis is located at the Main River, which is one of the largest available for Germany, but the data availability outside tributaries of the Rhine and lies completely in Germany, of Germany is problematic. Some of the essential data was used for the calibration and validation, and the were not available so far for the international river basins, further scenario analysis. However, in the validation such as observed climate data and land use map for procedure the last gauge on the Rhine in Germany, Switzerland. For some other input data, the full data Rees, was also included. As the discharge at the gauge set was not homogeneous in terms of data quality. For Rees could not be correctly simulated without reasonable example, the soil map for Germany has finer classification input from Rheinfelden (border between Switzerland and for soil types than that from the European data base, Germany, Figure 1), the observed discharge data at the and the German soil parameterization is differentiated Rheinfelden were used as the inflow to the River Rhine by climate zones and land use types for each soil type, in the validation period. whereas the European soil data is not. The gridded As the last step in the simulations, the coastal areas climate data for river basin areas in France, Austria and and the small areas in the Maas and Oder basins were Luxemburg are much sparser and less complete than that simulated with the parameter sets from the nearest large in Germany. This of course will have implications on basins, the Rhine and the Elbe. The results for water

Table II. The gauge stations in the ﬁve river basins used for calibration and validation, and their corresponding drainage areasa

River basin Last gauge (used Drainage Intermediate River Drainage for calibration area (km2) gauge (used area (km2) and validation) for validation)

Greven Ems 2 842 Rheine Ems 3 740 Ems Versen 8 369 Dalum Ems 4 981 Letzter Heller Werra 5 487 Guntershausen Fulda 6 366 Schwarmstedt Leine 6 443 Marklendorf Aller 7 209 Weser Intschede 37 720 Vlotho Weser 17 618 Burghausen Salzach, Inn 6 649 Donauwoerth Danube 15 037 Passau Ingling Inn 26 084 Pfelling Danube 37 687 Danube Hofkirchen 47 496 Achleiten Danube 76 653 Schermbeck Lippe 4 783 Rockenau SKA Neckar 12 710 Trier UP Moselle 23 857 Maxau Rhine 50 196 Andernach Rhine 139 549 Rhine Frankfurt- 24 764 Rees Rhine 159 300 Osthafen (Main) Bad Dueben Mulde 6 171 Laucha Unstrut 6 218 Calbe-Grizehne Saale 23 719 Havelberg Havel 24 037 Elbe Neu-Darchau 131 950 Schona¨ Elbe 51 391 a The water discharge data is from The Global Runoff Data Centre (GRDC), 56 068 Koblenz, Germany, and the Ministry of the Environment and Conservation, Agriculture and Consumer Protection of the German State of North Rhine-Westphalia.

flow components in the reference and scenario periods Qobs and Qsim are the mean values of these parameters were used to complete the water components maps for for the whole simulation period. the whole Germany. The Nash and Sutcliffe efficiency can vary from minus No changes in land use patterns were assumed for the infinity to 1. A value of 1 denotes an absolute match of calibration and validation periods. In reality, land use in predicted and measured values, whereas value 0 for the Germany was definitely changing during 30 years, and deviation in balance means no difference between the taking into account land use patterns averaged at least for measured and simulated values. three decades could improve the results. However, such data are not available for Germany, and the validation had to be done with the ‘static’ land use data. RESULTS AND DISCUSSION In this study, the non-dimensional efficiency criterion of Nash and Sutcliffe (1970) (E) and the relative devia- Calibration and validation tion in water balance (B) were used to evaluate the quality The calibration and validation results in terms of of simulated daily water discharge. E is a measure to criteria of fit are presented in Table III. In the calibration describe the squared differences between the observed period, the Nash–Sutcliffe efficiency varies from 0Ð80 and simulated values using the following equation: to 0Ð90 for the five main gauges (Versen, Intschede, 2 Neu-Darchau, Frankfurt-Osthafen and Hofkirchen) and Qobs Qsim E D 1 1 the deviation in water balance is not more than 3%. 2 Qobs Qobs In the validation period (20 years), the Nash–Sutcliffe efficiency and the deviation for these five gauges are B describes the long-term differences of the observed within the ranges from 0Ð81 to 0Ð85 and from 8to values against the simulated ones in percent for the whole 6%, correspondingly. These results indicate that SWIM modelling period: can reproduce water discharge in large river basins quite well. Qsim Qobs B D ð 100 2 Besides, Table III includes the Nash–Sutcliffe efficien- Q obs cies and deviations in water balance for the 24 additional Here Qobs means the observed discharges, whereas intermediate gauges. In general, the discharge at most Qsim is the corresponding simulated value. The variables of these gauges can also be well reproduced by SWIM

Copyright  2010 John Wiley & Sons, Ltd. Hydrol. Process. 24, 3289–3306 (2010) SIMULATION OF WATER FLUXES IN GERMANY UNDER CLIMATE CHANGE 3297

Table III. The Nash and Sutcliffe efﬁciencies and deviations in water balance for the ﬁve main gauges in the calibration (1981–1990) and validation (1961–1980) periods, as well as for 24 selected intermediate gauges in the simulation period 1981–1990 (considered as spatial validation)

Gauge Drainage Percentage Calibration Validation area (km2)ofdrainage area inside Nash and Sutcliffe Deviation in Nash and Sutcliffe Deviation in Germany (%) efﬁciency water balance (%) efﬁciency water balance (%)

Versen 8 369 100 0Ð87 0 0Ð85 8 Greven 2 842 100 — — 0Ð87 4 Rheine 3 740 100 — — 0Ð78 1 Dalum 4 981 100 — — 0Ð83 3 Intschede 37 720 100 0Ð90 0 0Ð82 2 Letzter Heller 5 487 100 — — 0Ð71 4 Guntershausen 6 366 100 — — 0Ð57 5 Schwarmstedt 6 443 100 — — 0Ð68 3 Marklendorf 7 209 100 — — 0Ð76 9 Vlotho 17 618 100 — — 0Ð86 0 Hofkirchen 47 496 94 0Ð83 0 0Ð82 5 Burghausen 6 649 17 — — 0Ð56 3 Donauwoerth 15 037 100 — — 0Ð75 2 Passau Ingling 26 084 26 — — 0Ð61 12 Pfelling 37 687 95 — — 0Ð79 1 Achleiten 76 653 73 — — 0Ð74 6 Frankfurt-Osthafen 24 764 100 0Ð80 3 0Ð81 6 Schermbeck 4 783 100 — — 0Ð78 5 Rockenau SKA 12 710 100 — — 0Ð75 8 Trier UP 23 857 22 — — 0Ð21 57 Maxau 50 196 23 — — 0Ð84 4 Andernach 139 549 59 — — 0Ð81 5 Rees 159 300 64 — — 0Ð83 6 Neu-Darchau 131 950 61 0Ð86 2 0Ð84 3 Bad Dueben 6 171 91 — — 0Ð59 2 Laucha 6 218 100 — — 0Ð60 13 Calbe-Grizehne 23 719 99 — — 0Ð74 1 Havelberg 24 037 100 — — 0Ð43 31 Schona¨ 51 391 0 — — 0Ð60 30

with the efficiency above 0Ð6 and deviation within š10%, to the river discharge of several national subbasins and even without additional calibration. There are only a few country-wide water components. problematic gauges whose simulated results do not com- Figure 3 shows the comparison between the simulated ply with the observed values well enough, e.g. Trier UP, and observed river discharge at two selected gauges Havelberg and Schona.¨ This is mainly due to two reasons: Intschede (Weser) and Hofkirchen (Danube) as the daily poorer input data, or water regulation or management time series in the period 1984–1989 (Figure 3, left), which was not considered in the simulation. and as average daily dynamics for the whole calibration The discharge simulated for the gauge Havelberg at period 1981–1990 (Figure 3, right). As one can see, the River Havel in the Elbe basin is much lower than the in both cases the simulated river discharge is in a measurements. The underestimation of the river discharge satisfactory agreement with the observed one, and the is mainly due to the mining activities in the catchment, seasonal dynamics is also well reproduced. especially during the 1970s and 1980s. In conjunction The hydrograph pattern at the gauge Intschede (Figure with the mining activities, large amounts of water were 3a, right) is very typical for the most parts of Germany. extracted from underground and discharged into the rivers As precipitation is not concentrated in a ‘rainy season’, causing higher than natural water level in the Havel, but distributed in all four seasons (although not evenly), which cannot be reproduced by the model. there are usually higher water flows in winter due to The gauges such as Trier UP, Schona,¨ Passau Ingling low evapotranspiration, whereas in summer the abundant and Burghausen, which include drainage areas outside of vegetation and high temperature lead to high water losses Germany, do not show satisfactory results either. The to atmosphere via evapotranspiration and lower runoff. poor results at these gauges reflect the data problems In comparison with that, the seasonal water flow pattern mentioned earlier: poor available climate data outside at the gauge Hofkirchen (Figure 3b, right) is different, of Germany and heterogeneous soil data sets. As such with relatively high water flows in early summer. The problems could not be avoided in this study, the main snow melting from the Alps contributes to additional focus of the climate scenario evaluation was restricted water peaks in this period. The satisfactory reproduction

Figure 3. Simulated and observed water discharge (left) at the gauges Intschede (Weser) (a) and Hofkirchen (Danube) (b) in the period 1984–1989, and the corresponding average daily discharge (right) for the same stations for the whole calibration period (1981–1990) of both patterns by SWIM confirms the model suitability forests are the main drivers of the higher evapotranspira- for large regions with different hydrological conditions. tion in southern Germany. The distribution pattern of total runoff is closely Comparison of spatial patterns related to the geographic characteristics. The Black Forest The annual average evapotranspiration, total runoff (a region, the Alpine Foreland and the mountainous areas sum of surface runoff, interflow and groundwater flow) with high reliefs (as compared with the DEM map) and groundwater recharge simulated by SWIM for the produce larger amounts of total runoff (>700 mm). The reference period (1961–1990) were compared with those Elbe, the Warnow/Peene and the Oder basins have the estimated for the same time period and presented in the lowest runoff production (about 178, 150 and 90 mm on Hydrological Atlas of Germany (HAD, 2000). Obviously, average, respectively) among all the basins mainly due no comparison to real measurements is possible at to the smallest amount of precipitation. this scale. The mean annual actual evapotranspiration The patterns of annual average groundwater recharge depth presented in the Atlas was calculated based on are affected by the amount of seepage water and soil the grass reference evapotranspiration (Wendling, 1995) properties. Apart from the similarity of positive ground- and the BAGLUVA method developed by Glugla et al. water recharge distribution, the negative groundwater (2002). The runoff map in the Atlas was obtained from recharge was also simulated in SWIM in the wetland the simple water balance subtracting estimated actual areas and riparian zones, where groundwater is shallow evapotranspiration from precipitation. The groundwater and plants can satisfy their water needs also from water recharge was estimated from a regression model using flowing from upper parts of the catchments into wet- an empirical equation of Kille (1970) to determine the land areas. As a result, the total plant water uptake from baseflow index (a measure of the ratio of baseflow to the groundwater in spring and summer can be higher than the total runoff). In contrast to SWIM, estimates for three amount of groundwater recharge for the same plot dur- water flow components for the Atlas were not balanced ing the winter season, and the net groundwater recharge at the catchment scale. is negative. The spatial patterns of the water flow components In addition, Figure 4g–i shows the difference in the estimated for the Hydrological Atlas (Figure 4a–c) and absolute values between the Atlas and SWIM simula- simulated by SWIM (Figure 4d–f) are similar, especially tion results. The yellow colour shows the difference for the total runoff and groundwater recharge. The maps < š50 mm. The dark blue and dark green highlight demonstrate substantial regional differences in all three the hotspots of major differences. In southern and in water flow components in Germany. In most of the cen- some locations in western Germany, the evapotranspira- tral and northern German areas, evapotranspiration is tion simulated by SWIM is up to over 200 mm higher ranging from 400 to 600 mm, and it is lower than in than that in the Atlas, and the total runoff in these the Black Forests, Rhine Valley and in the Alpine areas regions is correspondingly lower. The highest difference (>600 mm). The higher temperature in summer, higher in groundwater recharge is in western Germany, where precipitation in mountainous areas and high density of consolidated rocks impede the seepage water to reach

Copyright  2010 John Wiley & Sons, Ltd. Hydrol. Process. 24, 3289–3306 (2010) SIMULATION OF WATER FLUXES IN GERMANY UNDER CLIMATE CHANGE 3299

Figure 4. The annual average water ﬂow components in the period 1961–1990 estimated for the Hydrological Atlas of Germany (a–c), simulated by SWIM for the same period (d–f) and the difference maps (SWIM - HAD) (g–i). Maps for actual evapotranspiration: (a, d, g); total runoff: (b, e, h) and groundwater recharge: (c, f, i) (units: mm/year) groundwater (Bogena et al., 2005) (feature not repre- the Atlas, as the input data for both methods is from sented in SWIM). the same data bank (DWD, National Meteorological Table IV allows further comparison of average annual Service of Germany). The deviation of 1–3% could be water components from both methods (and also with due to different interpolation methods and precipitation the actual measured runoff depth) in the ﬁve main correction functions. The actual evapotranspiration and basins under study. The precipitation depths used in groundwater recharge simulated by SWIM are higher SWIM have no more than 3% deviation to these in than those estimated for the Atlas, and the difference

Table IV. Comparison of the annual average precipitation and water ﬂow components (evapotranspiration, total runoff and groundwater recharge) simulated by SWIM in the reference period 1961–1990 with those estimated for the Hydrological Atlas of Germany (HAD, 2000), and the measured runoff

Catchment Precipitation Evapotranspiration Runoff Groundwater (mm/year) (mm/year) (mm/year) recharge (mm/year)

HAD SWIM HAD SWIM HAD SWIM Measured HAD SWIM

Ems (Versen) 842 845 521 547 321 298 307 152 163 Weser (Intschede) 832 831 530 568 302 263 277 125 157 Danube (Hofkirchen) 1025 1052 535 633 490 419 436 201 263 Main (Frankfurt-Osthafen) 851 844 552 588 299 256 243 107 158 Saale (Calbe-Grizehne) 687 691 516 526 171 165 167 67 86

Figure 5. Changes in annual precipitation (a) and average annual temperature (b) in Germany in the period 2051–2060 (projected by STAR, medium realization) compared with 1961–1990 (observed). (Source: National Meteorological Service of Germany; DWD; PIK, base scenarios, PIK, 2010) between them is especially substantial in the upper et al., 2008) of the 100 realizations generated by STAR, Danube basin. In contrast, the total runoff obtained from and compared with those for the reference period SWIM simulation is lower than that in the Atlas, but the 1961–1990 estimated from the observed records for the SWIM results are closer to the measured water discharge whole Germany (Figure 5). The annual precipitation is values (see columns 6 and 7 in Table IV) as they were expected to decrease in eastern and southeastern Ger- validated for the large catchments. many significantly, whereas in the northwestern and west- In general, the comparison of spatial patterns of water ern Germany an increase in precipitation is prevailed. flow components simulated by SWIM with that in the The annual temperature is expected to rise by 2 °Cto Hydrological Atlas of Germany shows a good agreement. >3 °C in the country. The central range and the Harz The runoff data have also a good agreement with the region will retain a cooler climate as compared with measured values. The results presented in sections on other parts of Germany. According to the scenario, in Calibration and validation and Comparison of spatial the southern areas, climate change will manifest in par- patterns confirm that the model SWIM is appropriate ticular by a substantial increase in temperature. Hence, it for conducting climate impact assessment for Germany, is an important hotspot to analyse the impact on water and evaluating changes in spatial variability of water resources. components. To demonstrate the performance of STAR in different climate regions, Figure 6 shows the observed and pro- Climate scenarios jected annual dynamics of precipitation, and differences Firstly, the average annual precipitation and temper- in average monthly values of precipitation between the ature for the scenario period 2051–2060 were calcu- ones generated by STAR and the ones observed in the lated from one selected realization which stands for reference period for the basins Ems (Figure 6a), Saale the medium climatic-water balance condition (Wechsung (Figure 6b) and the upper Danube (Figure 6c). These

Copyright  2010 John Wiley & Sons, Ltd. Hydrol. Process. 24, 3289–3306 (2010) SIMULATION OF WATER FLUXES IN GERMANY UNDER CLIMATE CHANGE 3301

Figure 6. Observed annual precipitation (1961–2006) and generated by STAR annual precipitation (2007–2060) in the Ems (a), Saale (b) and the upper Danube (c) basins (left); and the difference in monthly average precipitation between the projected realizations (2051–2060) and the historical data (1961–1990) for the same basins (right) three basins belong to different climatic zones: a maritime Climate impacts on seasonal river discharge. Climate climate (Ems), a drier and more continental climate change impact on river discharge was analysed for (Saale) and an Alpine climate (upper Danube). The the gauge stations: Versen (Ems), Intschede (Weser), annual precipitation observed and simulated by STAR Calbe-Grizehne (Saale), Hofkirchen (Danube), Frankfurt- and averaged over the basins is shown in Figure 6 (left). Osthafen (Main) and Rockenau SKA (Neckar), as The grey and dark grey boundaries include the simu- described above in section on Data preparation. Two sce- lated precipitation from 100 and 80 realizations, and nario periods were evaluated: 10 years from 2009 to 2018 represent the uncertainty of precipitation projections. The (to test reliability of the hydrological response using the dashed line represents the medium realization of the 100 STAR realizations), and the last 10 years (from 2051 to in terms of precipitation amount. It shows that STAR is 2060) as the main scenario period to evaluate the climate able to project the annual variability reasonably. In addi- impacts on water resources. tion, there are slight downward trends in the Saale and Figure 7 shows the simulated average seasonal water Danube basins. In Figure 6 (right), the seasonal changes discharge in two scenario periods together with the in precipitation are presented. For all the three rivers, an simulated average seasonal water discharge for the increase in winter precipitation (strongest for the Ems) reference period 1961–1990 for six selected gauges. and decrease in summer precipitation (strongest for the The light grey bounds include all simulated results Danube) are projected. from 100 realizations, and the dark grey bounds cover

Figure 7. Seasonal water discharge in two scenario periods (2009–2018 and 2051–2060) including 80 and 100 realization and the medium average daily discharge of 100 realizations compared with the simulated discharge for the reference period 1961–1990 for six basins: (a) the Ems basin (gauge Versen); (b) the Weser basin (gauge Intschede); (c) the Saale basin (gauge Calbe-Grizehne); (d) the Danube (gauge Hofkirchen); (e) the Main basin (gauge Frankfurt-Osthafen) and (f) the Neckar basin (gauge Rockenau SKA) the 80 percentile of 100 runs. The dashed line is the winter time of the second scenario period, river the medium average daily discharge simulated with discharge is likely to increase in all rivers, especially in the 100 realizations. And the solid lines represent the Ems. A robust trend seems to be that the recession the average daily water discharge during the refer- of the winter flow starts earlier in spring and lasts ence period. The upper graph of each sub-figure shows longer into late summer. In summer and autumn, all changes in the next 10 years (2009–2018), and the the rivers tend to have lower water discharge, especially lower graphs show changes in the mid of the century from July to September. The main reasons are higher 2051–2060. evapotranspiration due to higher temperature, and lower In all six cases in Figure 7, the water discharge or practically the same precipitation in summer. The simulated in the first scenario period 2009–2018 has a earlier harvest of winter crops and the following faster similar seasonal dynamics compared with the observed growth of cover crop aggravate the loss of soil water one; only in winter water discharge is higher practically and decrease of runoff in these months. Among the six in all basins. As the climate is changing gradually, the river basins, water discharge decreases in summer most hydrological dynamics is not likely to vary suddenly and dramatically in the Danube, Saale and Neckar basins significantly in a short term. In the second scenario period (Table V), where almost all 100 scenario realizations (2051–2060), the changes in seasonal water discharge show a lower level. It is also worth mentioning that become obvious and differentiated among the basins and in all rivers except the Danube the water discharge in seasons, and some of them are considerably strong. In summer is already very low in the current condition,

Copyright  2010 John Wiley & Sons, Ltd. Hydrol. Process. 24, 3289–3306 (2010) SIMULATION OF WATER FLUXES IN GERMANY UNDER CLIMATE CHANGE 3303

Table V. Changes in seasonal river dynamics for the six rivers (medium average daily discharge simulated in the scenario period minus average daily discharge in the reference period)

Catchment Winter Spring Summer Autumn (December to February) (March to May) (June to August) (September to November) (%) (%) (%) (%)

Ems (Versen) 17Ð911Ð3 8Ð2 8Ð2 Weser (Intschede) 13Ð1 0Ð8 19Ð1 17Ð4 Danube (Hofkirchen) 12Ð9 8Ð4 24Ð1 17Ð7 Main (Frankfurt-Osthafen) 15Ð7 8Ð0 14Ð7 19Ð0 Neckar (Rockenau SKA) 5Ð3 14Ð9 22Ð1 19Ð7 Saale (Calbe-Grizehne) 13Ð1 15Ð5 24Ð4 30Ð1 and the projected river flow in the Saale in 2051–2060 There are even some negative trends in some parts of the approaches zero in autumn. Elbe basin, probably due to low precipitation. The higher This means an increased risk for hydropower plants, evapotranspiration means high loss of water, which could navigation and water availability for cooling of thermal be critical, especially for areas with negative or minor power plants. The increasing risk of low flow conditions positive tendency in precipitation. in Germany simulated by STAR and SWIM also complies Figure 8b shows that, according to climate sce- with the results of other climate impact studies. For nario, runoff would be significantly reduced (up to example, Hennegriff et al. (2008) forecasted the impact 100 mm/year) in the southern part of the upper Danube of climate change on low water conditions in the German basin and the Rhine River valley. The Black Forest and State Baden-Wurttemberg,¨ where the river Neckar, a part the upper Elbe catchment (German part) are also likely of the Rhine and headwater of the Danube are located. to have significant reduction in water runoff in the future. They argued that in the months of July to September, In contrast, the northwest areas would have more avail- the monthly average low flow may decrease by 10–20% able water resources on average, mainly due to higher in the Neckar and Danube in the period 2021–2050 precipitation. compared with the reference period 1971–2000. Another The groundwater recharge is very sensitive to climate climate impact study for the Danube basin (Mauser et al., change. As one can see in Figure 8c, large areas in the 2008) has shown that the annual low flow (minimum Danube, Rhine and Elbe basins have lower groundwa- 7 days mean discharge) could be reduced to half of the ter recharge in the scenario period. In general, spatial reference value (1971–2003) by 2030, and to one-third patterns of changes in runoff and groundwater recharge by 2060 under the IPCC A1B scenario. are quite similar (Figure 8b,c). In warmer conditions, the In addition, Figure 7 also illustrates the hydrological higher temperatures can extend the vegetation period, response to the uncertainty of climate projections. The and more water will be taken up from underground. The maxima and minima of the water discharge define the groundwater recharge period could also be shorter due uncertainty boundaries. In general, the uncertainty of high to the shortening of the snow cover time. The higher water discharge is much larger than that of low flow, and the uncertainty in winter time is much larger than that water uptake by plants and shortage of the groundwater in summer. The uncertainty boundaries imply the high recharge time would negatively influence the quantity of potential of drier summers and more frequent high water groundwater and the water table level. levels in winter in the future. Figure 8d–f shows the standard deviation of the changes based on 100 realizations. Larger standard devi- Impacts on average annual water flow components. ation means higher inter-annual variations. The mean The differences in average actual evapotranspiration, total changes in actual evapotranspiration are relatively certain runoff and groundwater recharge between the scenario compared with runoff and groundwater recharge. The rea- period (2051–2060, averaged over 100 realizations) son is that the main driver of potential evapotranspiration and the reference period (1961–1990) in Germany are is temperature whose trend is consistent in all 100 realiza- illustrated in Figure 8. In general, there is an increase tions. Hence, the actual evapotranspiration is increasing of evapotranspiration in most areas of Germany. One as long as water supply (precipitation) is sufficient. In of the most important drivers is the higher temperature some regions, where the water resources are very vul- in the future. In this study, the increase in temperature nerable to the changing climate, the uncertainty is more by 2 °C on average by the mid of 21st century for the substantial. As one can see in Figure 8d, both the Elbe whole territory of Germany was accompanied by an basin and the upper Rhine Valley have high uncertainty, increase in actual evapotranspiration of about 25 mm on implying the high sensitivity of evapotranspiration to cli- average. In wetlands and in some mountainous areas, mate in these areas. In contrast, these drier areas have the increase can reach even more than 100 mm/year. In lower uncertainty in projected runoff and groundwater some drier areas where precipitation may be dropped recharge. The regions with high water productions have down in the future, the increasing trend is moderate. higher uncertainty in runoff and groundwater recharge.

Figure 8. Difference maps for the simulated water ﬂow components between the climate scenario period 2051–2060 (mean of 100 realizations) and the reference period 1961–1990 (a–c), and the standard deviations from the 100 realizations (d–f). Maps for actual evapotranspiration: (a, d); total runoff: (b, e) and groundwater recharge: (c, f) (units: mm/year)

CONCLUSIONS AND OUTLOOK and quality of available data on land use and soil were poor. This of course had implications on the simula- In this study, the first German-wide impact assessment tion results (poorer validation results for gauges corre- of water fluxes dynamics under climate change using a sponding to transboundary catchments). The quality of process-based river basin model is presented in a spatially data input directly influenced simulation results and lead and temporally distributed manner. A special focus of the to difficulties in performing a sound evaluation of the study was on data availability, homogeneity of data sets, model outputs. In our study, this problem was solved and related error and uncertainty propagation in the model by doing a maximum of possible, e.g. providing a part results. This is especially important for transboundary of results for the total German territory (spatial patterns river basins like the Rhine, upper Danube and Elbe. of water fluxes), and restricting the study area by the Besides, climate scenarios incorporate uncertainty, and solely national large representative subbasins for the cli- it is necessary to analyse the related uncertainty in the mate impact assessment. results of impact assessment. The inherent uncertainty related to data availability, We can summarize that a good data base including especially in the transboundary basins, and to climate all necessary data for modelling with SWIM and sce- scenarios has to be better explored in follow-up inves- nario analysis was available for Germany, but data avail- tigations. When better and homogeneous data input will ability outside of Germany regarding density of stations be available, the validation of the model results could be

Copyright  2010 John Wiley & Sons, Ltd. Hydrol. Process. 24, 3289–3306 (2010) SIMULATION OF WATER FLUXES IN GERMANY UNDER CLIMATE CHANGE 3305 improved, and the impact assessment study will cover The scenario results show that, under assumption that the whole large basins, and provide more sound sce- the applied climate scenario is realistic, the reduction of nario results. Cross-comparison of climate impact results water discharge in streams is likely to be a considerable driven by several climate downscaling methods could, problem in Germany by the middle of the century, on one hand, extend the uncertainty bounds, and, on the especially in southeastern part of Germany. Summer other hand, increase certainty of some trends. may become the most problematic season in the future The study was devoted to climate impact assessment, regarding water availability in streams, as all the six and no changes in land use patterns were considered. The rivers show a decline in water discharge (about 10–25%) same land use patterns were assumed for the reference in summer months. The projected low flow conditions and scenario periods in this study (‘static’ land use). also comply with the results of other regional studies This was done in order to investigate the ‘pure’ impact in Germany. The water conflicts among different water of climate change on water fluxes, which is reasonable users could become more severe in the most parts of at this stage. Only the crop scheduling was adjusted in Germany. In the winter time, all the rivers tend to have the warmer climate. However, the SWIM model is able more stream flow (about 5–18%), especially the Ems to simulate complex crop rotations and changes in land river basin in the northwest coastal region. Regarding the use patters, as demonstrated in several previous regional major water flow components, evapotranspiration would studies (Wechsung et al., 2000; Hattermann et al., 2007b; increase by about 25 mm on average in Germany, mainly Yu et al., 2009). Therefore, possible changes in land due to higher temperature. The changes in groundwater use patterns and crop rotation could be considered in recharge and runoff generation are spatially different. The the follow-up studies to combine them with changes southern Germany and large parts of the Elbe basin have a in climate and to explore how changes in land use negative tendency in both groundwater recharge and total could compensate for undesirable changes in water flow runoff, whereas more water is expected in northwestern dynamics. Germany. The climate impact assessment demonstrates The first German-wide impact assessment of water potential lower water resource availability in the Elbe, fluxes dynamics under climate change was performed upper Danube and upper part of the Rhine Valley. using the state-of-the-art statistical RCM, STAR and the The uncertainty bounds of river discharge, which semi-distributed process-based eco-hydrological model, are resulted from the uncertainty of the downscaling SWIM. STAR provides a robust climate data set for technique, show that the uncertainty of high water hydrological studies and impact assessment. It reproduces discharge is much larger than that of low flow, and not only trend in temperature, but also annual variabil- the uncertainty in winter time is much larger than that in summer. The uncertainty boundaries imply the high ity and trends in precipitation, which comply with other potential of drier summers and more frequent high climate models. In addition, STAR generates a number water levels in winter in the future according to the of realizations for one scenario, which include the uncer- scenario. This conclusion emphasizes the higher risk tainty of projected climate characteristics, especially pre- of dry summers in the future. For some areas, such cipitation. In future, the climate impact results driven by as the Elbe basin, even a small uncertainty of the other climate downscaling methods should be compared downward trend in runoff implies a high potential of with those obtained with STAR. dryer condition. Hence, the adaptation to climate change The eco-hydrological model SWIM has proven to in water management and other related sectors is very adequately reproduce the temporal and spatial water important, and should be highlighted for such regions. dynamics and river discharge in large river basins, and for such a large and heterogeneous region as Germany. The water flow components simulated by SWIM were ACKNOWLEDGEMENTS also proved to produce comparable distribution patterns with the Hydrological Atlas of Germany. However, the The authors are grateful to M. Wodinski and T. Vetter lack of historical climate data and scenario data outside for their help in data and graphs preparation, and to Prof. of Germany restricted the calibration and validation of A. Bronstert and three anonymous reviewers for their SWIM for the largest river basins (Rhine and upper comments, which helped to improve the presentation of Danube), and the projection of river discharge for some simulation results and discussion of associated problems. other main gauges. Besides, a particular problem was to analyse and project the flood events, especially for the Rhine and REFERENCES the Elbe rivers, where the large upstream drainage areas Albek M, Od¨ utveren¨ UB,¨ Albek E. 2004. Hydrological modeling of outside of Germany are important as source areas for Seydi Suyu watershed (Turkey) with HSPF. Journal of Hydrology 285: 260–271. the extreme events. This underlines the importance of Arnell NW. 2003. Relative effects of multi-decadal climatic variability availability of homogeneous data sets including climate, and changes in the mean and variability of climate due to global land use and soil parameters for Europe, which would warming: future streamflows in Britain. Journal of Hydrology 270: 195–213. enable improvement of the climate impact assessment for Arnold JG, Allen PM, Bernhardt G. 1993. A comprehensive surface- large regions and countries. groundwater flow model. Journal of Hydrology 142: 47–69.

Arnold JG, Williams JR, Nicks AD, Sammons NB. 1990. SWRRB—A Krysanova V, Muller-Wohlfeil¨ DI, Becker A. 1998. Development and Basin Scale Simulation Model for Soil and Water Resources test of a spatially distributed hydrological/water quality model for Management, Texas A&M University Press: College Station; 255. mesoscale watersheds. Ecological Modelling 106: 261–289. Bogena H, Kunkel R, Schobel¨ T, Schrey HP, Wendland F. 2005. Mauser W, Bach H. 2009. PROMET—large scale distributed hydrologi- Distributed modeling of groundwater recharge at the macroscale. cal modeling to study the impact of climate change on the water flows Ecological Modelling 187: 15–26. of mountain watersheds. Journal of Hydrology 376: 362–377. Bohm¨ U, Keuler K, Osterle¨ H, Kucken¨ M, Hauffe D. 2008. Quality Mauser W, Marke T, Stoeber S. 2008. Climate change and water of a climate reconstruction for the CADSES region. MetZ Special resources: scenarios of lowflow conditions in the Upper Danube River Issue Regional Climate Modeling with COSMO-CLM (CCLM) 17(8.): Basin. XXIVth Conference of the Danubian Countries, IOP Conf. 477–485 p. Series: Earth and Environmental Science 4. Doherty J. 2004. PEST: Model Independent Parameter Estimation,5th Menzel L, Burger¨ G. 2002. Climate change scenarios and runoff response edn, of user manual. Watermark Numerical Computing: Brisbane, in the Mulde catchment (Southern Elbe, Germany). Journal of Australia. Hydrology 267: 53–64. Drogue G, Pfister L, Leviandier T, El Idrissi A, Iffly J-F, Matgen P, Menzel L, Thieken AH, Schwandt D, Burger¨ G. 2006. Impact of climate Humbert J, Hoffmann L. 2004. Simulating the spatio-temporal change on the regional hydrology—scenario-based modelling studies variability of streamflow response to climate change scenarios in a in the German Rhine catchment. Natural Hazards 38(1–2): 45–61. mesoscale basin. Journal of Hydrology 293: 255–269. Middelkoop H, Kwadijk JCJ. 2001. Towards integrated assessment of Enke W, Spekat A. 1997. Downscaling climate model outputs into local the implications of global change for water management—the Rhine and regional weather elements by classification and regression. Climate experience. Physics and Chemistry of Earth (B) 26(7–8): 553–560. Research 8(3): 195. Monteith JL. 1965. Evaporation and the environment. Symposia of the Gerstengarbe F, Kucken¨ M, Werner P. 2009. Modellvergleich der Society for Experimental Biology 19: 205–234. regionalen Klimamodelle REMO, CCLM, WETTREG und STAR II .PIK Muttiah RS, Wurbs RA. 2002. Modeling the impacts of climate change (Potsdam Institute for Climate Impact Research) internal report. on water supply reliabilities. Water International, International Water Glugla G, Jankiewicz P, Rachimow C, Lojek K, Richter K, Furtig¨ G, Resources Association 27(3): 407–419. Krahe P. 2002. Wasserhaushaltsverfahren BAGLUVA zur Berechnung Nash JE, Sutcliffe JV. 1970. River flow forecasting through conceptual vieljahriger¨ Mittelwerte der tatsachlichen¨ Verdunstung und des models. Part I: a discussion of principles. Journal of Hydrology 10(3): Gesamtabfluß. BfG-Bericht 1342, Bundesanstalt f. Gewasserkunde,¨ 282–290. Berlin/Koblenz. Orlowsky B, Gerstengarbe F-W, Werner P. 2008. A resampling scheme HAD (Hydrologischer Atlas von Deutschland). 2000. Edited by for regional climate simulations and its performance compared to a Bundesministerium fur¨ Umwelt, Naturschutz und Reaktorsicherheit. dynamical RCM. Theoretical and Applied Climatology 92: 209–223. ISBN 3-00-005624-6. Petrow T, Merz B. 2009. Trends in flood magnitude, frequency and Hattermann FF, Conradt T, Bronstert A. 2007a. Berechnung großskaliger seasonality in Germany in the period 1951–2002. Journal of Hydrology Verdunstung unter den Bedingungen des globalen Wandels. In 371: 129–141. Verdunstung. Beitrage¨ zum Seminar Verdunstung am 10/11 Oktober Priestley CHB, Taylor RJ. 1972. On the assessment of surface heat flux 2007, Miegel Konrad, Kleeberg (Hrsg.): Hans-B (eds). Universitat¨ and evaporation using large scale parameters. Monthly Weather Review Potsdam. Forum fur¨ Hydrologie und Wasserbewirtschaftung: Heft; 100: 81–92. 231–245. ISBN: 978-3-940173-05-8. Roeckner E, Bauml¨ G, Bonaventura L, Brokopf R, Esch M, Gior- Hattermann FF, Gomann¨ H, Conradt T, Kaltofen M, Kreins P, Wech- getta M, Hagemann S, Kirchner I, Kornblueh L, Manzini E, Rhodin A, sung F. 2007b. Impacts of global change on water-related sectors Schlese U, Schulzweida U and Tompkins A. 2003. The atmospheric and society in a trans-boundary central European river basin—Part 1: general circulation model ECHAM5. Part I: model description. Max project framework and impacts on agriculture. Advances in Geoscience Planck Institute for Meteorology Report 349: 127. [Available from MPI 10: 1–8. 2007. for Meteorology, Bundesstr. 53, 20146 Hamburg, Germany]. Hattermann FF, Krysanova V, Wechsung F, Wattenbach M. 2004. Schonwiese¨ C-D, Staeger T, Tromel¨ S. 2004. The hot summer 2003 in Integrating groundwater dynamics in regional hydrological modelling. Germany. Some preliminary results of a statistical time series analysis. Environmental Modelling and Software 19(11): 1039–1051. Meteorologische Zeitschrift, N.F. 13: 323–327. Hattermann FF, Post J, Krysanova V, Conradt T, Wechsung F. 2008. Schonwiese¨ C-D, Staeger T, Tromel¨ S. 2006. Klimawandel und Assessment of water availability in a Central-European River Basin Extremereignisse in Deutschland, Klimastatusbericht 2005, Deutscher (Elbe) under climate change. Advances in Climate Change Research 4: Wetterdienst, Offenbach, S. 7–17. 42–50. Statistisches Bundesamt Deutschland. 2008. Statistical Yearbook 2008 . Hennegriff W, Ihringer J, Kolokotronis V. 2008. Prognose von Available in German Accessed August 16, 2009. Auswirkungen des Klimawandels auf die Niedrigwasser-verhaltnisse¨ in T.G.M. Water Research Institute. Hydroecological Information System. Baden-Wurttemberg.¨ Korrespondenz Wasserwirtschaft 6(1): 309–314. Accessed September 2005. IPCC. 2007. Climate change 2007: impacts, adaptation and Thodsen H. 2007. The influence of climate change on stream flow in vulnerability—summary for policymakers. Working Group II Contri- Danish rivers. Journal of Hydrology 333: 226–238. bution to the Fourth Assessment Report of the Intergovernmental Panel Wechsung F, Gerstengarbe FW, Lasch P, Luttger¨ A. 2008. Die on Climate Change. IPCC Secretariat. Ertragsfahigkeit¨ ostdeutscher Ackerflachen¨ unter Klimawandel. PIK Jacob D. 2001. Meteorology and Atmospheric Physics 77: 61–73. Report No. 112, page 62. Available in http://www.pik-potsdam.de/ Kille K. 1970. Das Verfahren MoMNQ, ein Beitrag zur Berechnung research/publications/pikreports. Accessed March 16, 2010. der mittleren langjahrigen¨ Grundwasserneubildung mit Hilfe der Wechsung F, Krysanova V, Flechsig M, Schaphoff S. 2000. May land monatlichen Niedrigwasserabflusse.¨ Zeitschrift der Deutschen Geolo- use change reduce the water deficiency problem caused by reduced gischen Gesellschaft, Sonderheft Hydrogeologie Hydrochemie: 89–95. brown coal mining in the state of Brandenburg? Landscape and Urban Koskova´ R, Nemeckova´ S, Hesse C. 2007. Using of the soil Planning 51(2–4): 105–117. parametrisation based on soil samples databases in rainfall–runoff Wendling U. 1995. Berechnung der Gras-Referenzverdunstung mit modelling. In Proceedings of the Adolf Patera Workshop ‘Extreme der FAO Penman–Monteith–Beziehung. Wasserwirtschaft 85(12): hydrological events in catchments’, 13 November 2007. Jakub´ıkovaA,´ 602–604. Broza V, Szolgay J (eds). Bratislava; 241–249. ISBN 978-80-01- Williams JR, Renard KG, Dyke PT. 1984. EPIC—a new model for 03960-1 [in czech]. assessing erosion’s effect on soil productivity. Journal of Soil and Krysanova V, Hattermann F, Habeck A. 2005. Expected changes in Water Conservation 38(5): 381–383. water resources availability and water quality with respect to climate Yu P, Krysanova V, Wang Y, Xiong W, Mo F, Shi Zh, Liu H, Vetter T, change in the Elbe River basin (Germany). Nordic Hydrology 36(4–5): Huang Sh. 2009. Quantitative estimate of water yield reduction caused 321–333. by forestation in a water-limited area in northwest China. Geophysical Krysanova V, Meiner A, Roosaare J, Vasilyev A. 1989. Simulation Research Letters 36: L02406. DOI: 10.1029/2008GL036744. modelling of the coastal waters pollution from agricultural watersheds. Ecological Modelling 49: 7–29.