DELIVERABLE 1.2 DEFINITION OF CLIMATE CHANGE SCENARIOS

Transcription

1 DELIVERABLE 1.2 DEFINITION OF CLIMATE CHANGE SCENARIOS PP RESTRICTED TO OTHER PROGRAMME PARTICIPANTS Report for: European Commission Directorate-General for Energy and Transport 1049 Brussels December 10, 2010 Authors: Bastian Klein Boglárka Gnandt Imke Lingemann Tanya Prozny Gabriella Szépszó

2 ECCONET - Effects of climate change on the inland waterway transport network contract number FP7 2 Table of contents TABLE OF CONTENTS...2 LIST OF TABLES...3 LIST OF FIGURES DEFINITION OF CLIMATE CHANGE SCENARIOS INTRODUCTION, MOTIVATION BACKGROUND OF THE MODEL SELECTION AND SELECTION ASPECTS DETERMINATION OF EXTREME MODEL CHAINS FOR THE UPPER DANUBE Introduction Data Geographical representations of the Rhine and Upper Danube Catchments Methodology Test sensitivity of low-flow discharges to climate variables Define relationships between low-flow discharges and climate variables for the Rhine to verify KLIWAS model choices Assume Rhine relationships also hold for the Upper Danube Upper Danube observations Determine four extreme model chains for the Upper Danube THE SELECTED RCM EXTREME MODEL CHAINS AND THEIR SIMULATIONS Introduction Characteristics of the selected RCMs and their simulations EVALUATION AND POST-PROCESSING OF THE METEOROLOGICAL DATA VALIDATION Validation method Results Summary BIAS CORRECTION METHOD Introduction Linear Scaling method Application SPATIAL INTERPOLATION OF THE METEOROLOGICAL DATA SUMMARY, CONCLUSIONS...56 ACKNOWLEDGEMENTS...59 REFERENCES...60

3 ECCONET - Effects of climate change on the inland waterway transport network contract number FP7 3 List of tables Table 1.3.1: Overview of climate model projections including the regional climate models (RCMs; top) and global climate models (GCMs; bottom). Table 1.3.2: Pearson s Correlation Coefficient, R, between observations of precipitation sum over the true Upper Danube Catchment during different time periods and observations of the lowest seven-day mean discharge (NM7Q in m 3 /s) in hydrological winter (November through April) for observed data at gauging Achleiten for the years Seasons are indicated in that the letters correspond to their respective months; for example, SONDJ indicates September, October, November, December, January. Table 2.1.1: Main characteristics of the chosen model experiments. Table 2.1.2: Annual, seasonal (MAM: March April May, JJA: June July August, SON: September October November, DJF: December January February) and monthly mean temperature difference between the results of five regional climate models and ECA&D reference dataset, and the multi-model mean of these departures (last column) for the period of for the Rhine (left panel of every column) and Danube (right panels) catchments. Table 2.1.3: Annual, seasonal (MAM: March April May, JJA: June July August, SON: September October November, DJF: December January February) and monthly mean relative precipitation difference between the results of five regional climate models and ECA&D reference dataset, and the multi-model mean of these departures (last column) for the period of for the Rhine (left panel of every column).

4 ECCONET - Effects of climate change on the inland waterway transport network contract number FP7 4 List of figures Figure 1.1.1: The fraction of total variance in decadal predictions explained by the three (orange: internal variability, green: emission scenario, blue: model) components of total uncertainty for mean temperature (left) and precipitation (right) over Europe (Hawkins and Sutton, 2009, 2010). Figure 1.2.2: (a) Lowest seven-day mean discharge (NM7Q in m 3 /s) at major gauging stations along the River Rhine for the time horizons (red) and (purple). Fat lines indicate representative projections (Source: KLIWAS, 2009); (b) Historical time series at the gauge station Kaub for NM7Q during both hydrological summer (May through October) and hydrological winter (November through April). Figure 1.3.1: Flow diagram outlining the simple approach of determining the four extreme model chains for the Upper Danube which will provide an estimate of the range of future low-flows for the river. Hatched boxes indicate work that has been done in a previous project (top; green box) or work that will be completed further along in ECCONET (bottom; yellow box). Figure 1.3.2: (a): The chosen domain for the River Rhine superimposed on the true Rhine River Catchment (Shabalova et al., 2003). The coordinates of the blue box are: 46.5 to 50.5 N; 7 to 10 E; (b): The chosen domain for the Upper Danube superimposed on the true Upper Danube River Catchment. Source: Via Donau. The coordinates of the purple box are: 47 to 50 N; 9 to 16 E. Figure 1.3.3: Detailed flow diagram outlining the methodology of determining four extreme model chains for the Upper Danube which will provide an estimate of the range of future low-flows for the river. Hatched boxes indicate work that has been done in a previous project (top; green box) or work that will be completed further along in ECCONET (bottom; yellow box). Inside the solid boxes are the titles and numbers of each subsection to follow; green arrows point toward the specific contents. Figure 1.3.4: Percent difference in lowest seven-day mean discharge (NM7Q in m 3 /s) during hydrological summer (May through October) at station Kaub for the River Rhine versus percent difference in precipitation sum over the Rhine Catchment during various seasons for the near-future (a) and far-future (b) compared to the normal period ( ). Seasons are indicated in the legend where the letters correspond to their respective months; for example, MJJASO indicates May, June, July, August, September, October. The Pearson s Correlation Coefficient, R, for each season is also listed in the legend. Individual points represent the simulated discharge and precipitation sum for each individual RCM listed in Table Figure 1.3.5: Same as Figure except that the absolute difference, rather than percent difference, is used to indicate future change and the climate variable is two-metre temperature in C rather than precipitation sum. Figure 1.3.6: Percent difference in lowest seven-day mean discharge (NM7Q in m 3 /s) during hydrological summer (May through October) at station Kaub for the River Rhine versus percent difference in precipitation sum over the Rhine Catchment during the same season for the near-future (a) and far-future (b) compared to the normal period ( ). The linear fit line and its corresponding equation and Pearson s Correlation Coefficient, R, is shown on the figure. Individual points represent the simulated discharge and precipitation sum for each individual RCM listed in Table

5 ECCONET - Effects of climate change on the inland waterway transport network contract number FP7 5 Figure 1.3.7: Same as Figure except that the absolute difference, rather than percent difference, is used to indicate future change and the climate variable is two-metre temperature in C rather than precipitation sum. Figure 1.3.8: Percent difference in lowest seven-day mean discharge (NM7Q in m 3 /s) during hydrological summer (May through October) at station Kaub for the River Rhine versus percent difference in precipitation sum over both the Rhine Catchment (diamonds) and the Upper Danube Catchment (squares) during the same season for the near-future (a) and far-future (b) compared to the normal period ( ). The Rhine Catchment data (diamonds) and corresponding linear fit line are a replication of Figure and are shown here for comparison. The drop lines point to possible changes in Upper Danube discharges that have been calculated using the linear fit line equation for the Rhine with area-averaged precipitation sum data for the Upper Danube Catchment. The individual points represent each individual RCM listed in Table 1.3.1; some points overlap. The KLIWAS low and high extreme model chains are labelled in the figure in a distinct way. Figure 1.3.9: Same as Figure except that the absolute difference, rather than percent difference, is used to indicate future change and the climate variable is two-metre temperature in C rather than precipitation sum. Figure : Percent difference in lowest seven-day mean discharge (NM7Q in m 3 /s) during hydrological summer (May through October) at station Kaub for the River Rhine versus percent difference in precipitation sum over the Rhine Catchment (large diamonds; linear fit line) and the Upper Danube Catchment, as simulated by RCMs (squares) and additional GCMs (triangles), during the same season for the near-future (a) and far-future (b) compared to the normal period ( ). The drop lines point to possible changes in Upper Danube discharges that have been calculated using the linear fit line equation for the Rhine with GCM simulated area-averaged precipitation sums for the Upper Danube Catchment. The individual points represent each model listed Table 1.3.1; some points overlap. The GCMs as well as the KLIWAS low and high extreme model chains are labelled in the figure in a distinct way. Figure : Same as Figure except that the absolute difference, rather than percent difference, is used to indicate future change and the climate variable is two-metre temperature in C rather than precipitation sum. Figure a: Historical time series for the Upper Danube at gauging station Achleiten of the lowest seven-day mean discharge (NM7Q in m 3 /s) during hydrological summer and hydrological winter from 1900 to present. Figure b: Dependence between observations of precipitation sum over the true Upper Danube Catchment for various time periods and observations of lowest seven-day mean discharge (NM7Q in m 3 /s) during hydrological winter at the gauging station Achleiten for the years The graphs here correspond to the seasons with the highest correlation coefficients in Table Figure a: Near-future percent difference in lowest seven-day mean discharge (NM7Q in m 3 /s) during hydrological summer (May through October) at station Kaub for the River Rhine versus percent difference in precipitation sum over both the Rhine Catchment (diamonds) and the Upper Danube Catchment (squares). (Right) Same as Figure 1.3.8a except that precipitation sum over the Upper Danube Catchment has been determined for the season September through January rather than hydrological summer, May through October. (Left) An exact replica of Figure 1.3.8a for comparison.

6 ECCONET - Effects of climate change on the inland waterway transport network contract number FP7 6 Figure b: Same as Figure a except for the far-future. On the left is an exact replica of Figure 1.3.8b for comparison. Figure : Percent difference in precipitation sum for the Upper Danube Catchment versus percent difference in precipitation sum for the Rhine Catchment during both hydrological summer (MJJASO, May through October; blue diamonds) and SONDJ (September through January; orange squares) for the nearfuture period (a) and the far-future period (b) compared to the normal period ( ). Figure : Dates of the yearly lowest seven-day discharge (NM7Q in m 3 /s) and the seasonality vector for different time periods at the gauging station Achleiten on the German Austrian border. Figure : RCM projected precipitation changes for the Rhine Catchment during hydrological summer (left column) and the Upper Danube Catchment during both September through January (middle column) and hydrological summer MJJASO (right column) for the near-future (left) and far-future (right) compared to the normal period. Figure 1.4.1a: Absolute difference in near-future ( ) versus normal period ( ) longterm annual precipitation sum in mm/day (top) and two-metre temperature in C (bottom) for the KLIWAS high (SMHI BCM) and low (Had3Q0) models for seven-day lowest discharge (NM7Q in m 3 /s) over the Rhine Catchment. Note that seven-day lowest discharge has a direct relationship with precipitation sum and an inverse relationship with two-metre temperature. Figure 1.4.1b: Absolute difference in far-future ( ) versus normal period ( ) long-term annual precipitation sum in mm/day (top) and two-metre temperature in C (bottom) for the KLIWAS high (KNMI) and low (CCLM) models for seven-day lowest discharge (NM7Q in m 3 /s) over the Rhine catchment. Note that seven-day lowest discharge has a direct relationship with precipitation sum and an inverse relationship with two-metre temperature. Figure 1.4.2a: Absolute difference in near-future ( ) versus normal period ( ) longterm annual precipitation sum in mm/day (right) and two-metre temperature in C (left) for the high (SMHI ECHAM) and low (Had3Q0) models for seven-day lowest discharge (NM7Q in m 3 /s) over the Upper Danube catchment. Note that seven-day lowest discharge has a direct relationship with precipitation sum and an indirect relationship with two-metre temperature. Figure 1.4.2b: Absolute difference in far-future ( ) versus normal period ( ) long-term annual precipitation sum in mm/day (right) and two-metre temperature in C (left) for the high (KNMI) and low (CCLM) models for seven-day lowest discharge (NM7Q in m 3 /s) over the Upper Danube catchment. Note that seven-day lowest discharge has a direct relationship with precipitation sum and an inverse relationship with two-metre temperature. Figure 2.1.1: Annual and seasonal (MAM: March April May, JJA: June July August, SON: September October November, DJF: December January February) mean temperature difference between the results of five regional climate model and ECA&D observations for Figure 2.1.2: Seasonal cycle of the mean temperature for the period of for 2 different target regions: Rhine catchment (left) and Danube catchment (right) based on the results of five regional climate models and ECA&D reference dataset (red curve).

7 ECCONET - Effects of climate change on the inland waterway transport network contract number FP7 7 Figure 2.1.3: Normalized Taylor diagram of the mean temperature for the period of for 2 different target regions: Rhine catchment (left) and Danube catchment (right) based on the results of five regional climate models and ECA&D reference dataset (red symbol). Figure 2.1.4: Annual and seasonal (MAM: March April May, JJA: June July August, SON: September October November, DJF: December January February) relative mean precipitation difference between the results of five regional climate model and ECA&D observations for Figure 2.1.5: Seasonal cycle of the mean precipitation (mm/day) for the period of for 2 different target regions: Rhine catchment (left) and Danube catchment (right) based on the results of five regional climate models and ECA&D reference dataset (red curve). Figure 2.1.6: Normalized Taylor diagram of the mean precipitation for the period of for 2 different target regions: Rhine catchment (left) and Danube catchment (right) based on the results of five regional climate models and ECA&D reference dataset (red symbol). Figure 2.3.1: Grid resolutions obtained during the downscaling procedure. Figure 2.3.2: Schematic representation of the grid distance bisection. Figure 2.3.3: Example of an original temperature field (left) and the corresponding downscaled field (right).

8 ECCONET - Effects of climate change on the inland waterway transport network contract number FP Definition of climate change scenarios 1.1 Introduction, motivation The main objective of the ECCONET project is to assess the effect of climate change on the inland waterway transport network (with special emphasis on the Upper Danube and Rhine catchments) based on consolidation and analysis of earlier and existing research work as well as application of existing climate change and hydrological assessment tools. The project works in two parallel phases: (1) analysis of various effects of climate change on inland waterway transport and related sectors and (2) analysis of adaptation strategies and their impacts as well as development of recommendations and a strategic framework for the further development of the inland waterway model. A key point of view at planning the investigations in the project was that all the impact studies conducted within ECCONET should be comparable with each other. This can be guaranteed solely by the method, that every ECCONET study is based on the same basis, i.e., on a common meteorological basis. At the same time, the climate model simulations, which are the most physics- and process-oriented tools for description of the future climate evolution, include several uncertainties and another important aspect in the project is to take these uncertainties into account in the impact studies. In the climate model simulations there are three sources of uncertainty: 1. Internal variability. The internal variability is a natural characteristic of the climate system that also exists in the absence of any external radiative forcing. The impact of this feature can be observed in the different trends of shorter time periods (e.g., the temporary cooling in the so-called little ice age or during an even shorter period, for instance ). 2. Emission scenario uncertainty. An important and uncertain element of the future climate is the human activity. In order to quantify this, all the anthropogenic factors and socio economic aspects (population, energy consumption, industrial and agricultural structural changes, etc.) influencing the climate system are taken into account and for the climate change simulations their carbon dioxide emission and concentration equivalents are computed. The uncertainty is due to the estimation of the future evolution of the anthropogenic activity. Consequently, in order to assess these uncertainties, several scenarios are constructed for future emission tendencies, which include optimistic, pessimistic and medium versions, as well. 3. Model uncertainty. The climate models, which aim to estimate the response of the Earth system to climate change, often describe the processes of the climate system in different ways; there is not only one correct description of the climate system, consequently, there is not one climate model that is perfect in its every detail. Therefore, it is accepted to use several climate models, i.e., an ensemble of climate models, to estimate the climate change, since the different models simulate somewhat different changes as response even to the same (anthropogenic) radiative forcing. The relative contribution of the three components to the total uncertainty of the climate simulations varies depending on the investigated time period, region and meteorological parameter as indicated by Hawkins and Sutton (2009, 2010). Concentrating on the European region for the 21st century (Figure 1.1.1), the importance of the natural variability decreases in time and almost disappears by the end of the century. During the investigated period the weight of the uncertainty due to the emission scenarios increases at the expense of the other two sources of uncertainties. However, it has to be noticed, that scenario uncertainty becomes important only on the multi-decadal time scale. This is especially valid in the case of precipitation, where its importance is almost negligible with respect to the model uncertainty even at the end of the 21st century.

9 ECCONET - Effects of climate change on the inland waterway transport network contract number FP7 9 Figure 1.1.1: The fraction of total variance in decadal predictions explained by the three (orange: internal variability, green: emission scenario, blue: model) components of total uncertainty for mean temperature (left) and precipitation (right) over Europe (Hawkins and Sutton, 2009, 2010). Thus, since the time horizon of ECCONET is basically and the hydrological investigations are strongly based on precipitation-related variables, in agreement with the conclusions of Figure 1.1.1, the choice of the emission scenario did not play role at all at the selection of the model simulations for construction of the aforementioned common meteorological basis (i.e., only one emission scenario is investigated in the framework of the project). In ECCONET, an attempt is made to address the entire spectrum of uncertainty by determining representative projections or extreme model chains for the Rhine and Upper Danube catchments. These extreme model chains attempt to represent both the lower and upper signals of hydrological parameters for two future time horizons, and In the KLIWAS project, the whole spectrum of uncertainty, including the uncertainties due to the different global and regional climate models (GCMs and RCMs, respectively), was taken into account by utilizing hydrological impact models and an ensemble of regional climate model simulations based on various RCMs with different resolution. The raw climate model outputs, before using them as inputs for the hydrological impact model, were bias-corrected using the observation dataset for the past, and the same correction was applied also for the future. Then, a methodology was generated in order to select the representative projections from the ensemble of hydrological model output and four extreme model chains were chosen that describe the range of possible discharges of the Rhine in the future. For the Upper Danube, similar extreme model chains will be derived as part of ECCONET, but due to the large amount of data and calculation times needed for the high-resolution hydrological impact models, the four extreme model chains for the Upper Danube are found using a simple correlation analysis, between discharge and climate variable projections, as a proxy. The raw climate model outputs are calibrated to the observation dataset both for the past and future with application of the linear scaling method. Furthermore, the error-corrected 25 km resolution results are going to be downscaled for 10 km resolution with a simple interpolation technique in order to satisfy the requirements of the hydrological models concerning the horizontal resolution of the input data. The deliverable is structured in two main chapters: Chapter 1 provides an overview about the main aspects of the method used for selection of the extreme model chains for the Rhine (Sections 1.2 and 1.4) and Danube (Sections 1.3 and 1.4) providing the common meteorological foundations for the impact studies in the project. Chapter 2 begins with some insights into the main characteristics of model behaviour over the Danube and Rhine catchments (Section 2.1). The following sections are dedicated to briefly introduce the method needed to post-process the meteorological data, namely the error correction (Section 2.2) and spatial interpolation methods (Section 2.3). Finally, the deliverable is closed by the main conclusions drawn based on the presented results and gives some information about the future research works planned in the frame of the work package 1 in the ECCONET project.

10 ECCONET - Effects of climate change on the inland waterway transport network contract number FP Background of the model selection and selection aspects In recent years, global and regional climate modelling has achieved significant improvements, such that the results of the models are getting more and more reliable. Due to low resolution, simplification of processes, and assumptions about initial and boundary conditions, the model results still have large bias and uncertainties which can be grouped into three categories (for a more detailed analysis of the different sources of uncertainty, see Krahe et al., 2009): Internal variability: Uncertainties inherent in the system (aleatoric), for example deterministic chaotic behaviour of the climate system. Emissions scenario uncertainty: Uncertainties related to incomplete knowledge about the future of human activity. Model uncertainty: Uncertainties related to incomplete knowledge about the system (epistemic), for example measuring errors, simplification of the system (bias). Due to these uncertainties, there is not a single true climate model run; most likely there never will be. Hence, an ensemble, as generated via a multi-model approach using different global and regional climate models, and different hydrological models, should be used in order to account for these uncertainties and thus provide a range of possible futures. However, it is important to note that reality may actually lie outside this range. Figure shows the changes in the multi-annual (30 years) mean seasonal precipitation sums relative to the normal period of 18 global climate model projections for the Upper Danube Catchment. All simulations are based on the same AR4 SRES emission scenario A1B as developed by the IPCC (Nakicenovic et al., 2000). The span between model results is due to the uncertainties described above and in the introduction. There is a large difference between the projected precipitation changes ranging from approximately 0% to possible decreases/increases, depending on the season, of approximately ±20%. These results confirm the importance of the multi-model approach in view of an uncertain future. It can be seen in Figure that the model uncertainties in the near-future (2050) are larger than the climate change signal. For the far-future (2100), however, significant changes in climate depending on the selected greenhouse-gas emission scenario, which is associated with assumptions of future socio economic development, are more significant in comparison to the model uncertainties. Thus, in ECCONET, the focus will remain on results for the near-future time horizon, but analysis of the time horizon is also of interest when gauging the severity of climate change impacts since the climate signal is more dominant in the far-future.

11 ECCONET - Effects of climate change on the inland waterway transport network contract number FP7 11 Figure 1.2.1: Span of the seasonal (summer, winter) precipitation changes in the German part of the Upper Danube Catchment between 1950 and 2100, as simulated by 18 global climate models from the World Climate Research Programme's (WCRP's) Coupled Model Intercomparison Project phase 3 (CMIP3) multi-model dataset using the SRES emission scenario A1B. Shown are the changes of the multi-annual (30 years) mean precipitation sums relative to the normal period Source: Klein et al., The implementation of adaptation strategies in river-catchment management should be based on a range of possible changes in the system in order to take all available information and the existence of uncertainty into account. Ideally, the whole ensemble of available projections should be considered in the impact models in order to evaluate possible changes and adaptation options. However, in ECCONET this is not feasible due to the large amount of data and calculation times needed for the high-resolution hydrological impact models. Thus, the climate projections which generate the lower and upper boundaries of hydrological parameters are selected as representative projections and will serve as input for the hydrological impact models in order to provide a possible range of futures. But how should these representative projections be selected? The selection must correspond to the aim of the investigation, and the selected projections should cover the range of the possible effects due to climate change. The ECCONET project is focused on establishing policy guidelines and development plans for inland water transportation to react to changes in water regimes and the dependencies of navigation. In KLIWAS (2009) and Rheinblick2050 (Görgen et al., 2010), a methodology was generated to select the representative projections based on an ensemble of hydrological model output. Figure 1.2.2a is an example which demonstrates this with focus on hydrological summer low-flows at selected gauging stations on the River Rhine as measured by the seven-day lowest mean discharge (NM7Q). The graph is based on a climate projection ensemble consisting of 20 members for the time horizon All

12 ECCONET - Effects of climate change on the inland waterway transport network contract number FP7 12 members are bias-corrected (Lenderink et al., 2007) and taken as input for the hydrological model HBV134 (Görgen et al., 2010; te Linde et al., 2008). The representative projections are selected subjectively on the outer rim of the "scenario horizons" as indicated by fat lines. "Scenario horizons" are zones in the ensemble of discharge projections which are consistently simulated by most of the ensemble members and cover all ensemble members whose results are within a higher probability density. Clear outliers are subjectively omitted. Figure 1.2.2: (a) Lowest seven-day mean discharge (NM7Q in m 3 /s) at major gauging stations along the River Rhine for the time horizons (red) and (purple). Fat lines indicate representative projections (Source: KLIWAS, 2009); (b) Historical time series at the gauge station Kaub for NM7Q during both hydrological summer (May through October) and hydrological winter (November through April). The results obtained from the KLIWAS project will be utilized to provide an overview of which climate model projections generate extreme flow conditions that mainly affect inland waterway transport for the Rhine. Here, the parameter lowest seven-day mean discharge (NM7Q), as determined by a hydrological impact model, will describe extreme low-flows. In ECCONET, station Kaub has been selected as the critical station for the Rhine, because for the large majority of ships, the water depth at Kaub is the bottleneck and determines its maximum load factor (Jonkeren et al., 2007). As shown in Figure 1.2.2b, there has been a slight tendency in recent years toward increasingly lower water depths during hydrological summer at station Kaub, a tendency which may extend into the future. Thus, in order to describe a range of possible extreme low-flow conditions for the River Rhine, KLIWAS has chosen, for the hydrological summer NM7Q at Kaub, the following climate model projections as representative projections (using KLIWAS nomenclature): 2050: Low: A1B_HADCM3Q0_HADRM3Q0_25 (HAD3Q0) High: A1B_BCM_RCA_25 (SMHI BCM)

13 ECCONET - Effects of climate change on the inland waterway transport network contract number FP : Low: A1B_EH5r1_CCLM_20 (CCLM) High: A1B_EH5r3_RACMO_25 (KNMI) The representative projections are also referred to as extreme model chains in that there is one model to represent both the lower and upper boundaries of low-flows for each time horizon and respectively. Thus, there are four extreme model chains in total. Additionally within ECCONET, it is required to define climate models which potentially lead to extreme low-flow conditions of the Upper Danube River as well. However, at the time of writing this report, there are still no area-wide, ensemble hydrological modelling results available for the Upper Danube. Thus, in order to obtain similar extreme model chains for the Upper Danube within the timeframe of this project, they must be found using another method than that used by KLIWAS for the Rhine. Thus, within ECCONET, the initial selection of the representative projections or extreme model chains for the Upper Danube is determined using a simple correlation analysis, between discharge and climate variable projections, as a proxy. Main low-flow characteristics for the Upper Danube as described by the historical time series will be taken into account as well. This simple approach and its subsequent results will be discussed in the following sections.

14 ECCONET - Effects of climate change on the inland waterway transport network contract number FP Determination of extreme model chains for the Upper Danube Introduction The ECCONET project is focussed on determining the future conditions of inland waterway transport for both the Rhine and Upper Danube rivers. As discussed in the previous section, the KLIWAS project has performed ensemble hydrological modelling in order to determine a possible range of future low-flow discharges for the River Rhine. The climate projections which generate the lower and upper boundaries of future low-flows are then selected as representative projections or extreme model chains for the Rhine. For each time horizon and , there is a low and high model, meaning there are in total, four extreme model chains for the Rhine. For the Upper Danube, however, there are no area-wide, ensemble of hydrological modelling results available at the time of writing this report, and due to time and resource constraints, doing so is not feasible as part of ECCONET. Thus, in order to determine a possible range of changes in future low-flows for the Upper Danube while working within the timeframe of this project, a simple correlation analysis between discharge and climate variables is used as a proxy that will allow for the identification of the four extreme model chains for the Upper Danube which will later serve as input for the hydrological impact models in later stages of ECCONET. The simple approach implemented in this section is outlined in a flow diagram in Figure Figure 1.3.1: Flow diagram outlining the simple approach of determining the four extreme model chains for the Upper Danube which will provide an estimate of the range of changes in future low-flows for the river. Hatched boxes indicate work that has been done in a previous project (top; green box) or work that will be completed further along in ECCONET (bottom; yellow box). The initial stage involves testing the sensitivity of simulated future low-flow discharges of the Rhine to simulated projections of climate variables for the Rhine Catchment by investigating the linear relationships between them for a variety of seasons. Following that, relationships between simulated future low-flow discharges and climate variables will be determined in order to verify the four chosen extreme model chains by KLIWAS. Next, the Rhine relationships are assumed to hold for the Upper

15 ECCONET - Effects of climate change on the inland waterway transport network contract number FP7 15 Danube as well and the appropriateness of the four extreme model chains for the Rhine is evaluated for the Upper Danube Catchment. Finally, the four extreme model chains for the Upper Danube are subjectively determined by examining future projections of precipitation over the catchment. It must be emphasized that this is a simple approach which is used to gain some initial knowledge into the possible range of future changes in low-flows for the Upper Danube. The assumption that the Rhine relationships will also hold for the Upper Danube may not be true; they are different rivers with different behaviours, after all. But, as stated above, the resources do not exist to perform hydrological simulations of an entire ensemble of model projections, which would be the case in an ideal world. Instead, one must use timely and currently available information to gain insight into possible future changes in low-flows of the river; here, this is achieved through determining a mini-ensemble of four extreme model chains in order to get a lower and upper estimate of low-flow extremes for the Upper Danube Catchment for both future time horizons and Though the approach is simple, there are some distinct advantages in that many RCM simulations can be utilized in order to provide the widest possible range of future low-flow discharges for the Upper Danube. Additional GCMs than those used in KLIWAS will also be utilized in order to place the results in the context of an even larger ensemble Data As mentioned above, climate simulations of both regional climate models and global climate models are utilized within this section. The regional climate model projections are available through the ENSEMBLES project (Hewitt and Griggs, 2004; van der Linden et al., 2009), save one RCM run that was obtained through the CERA (Climate and Environmental Retrieval and Archive) database, and the global climate model projections are part of the World Climate Research Programme's (WCRP's) Coupled Model Intercomparison Project phase 3 (CMIP3) multi-model dataset (Meehl et al., 2007). The monthly RCM and GCM data is used to obtain area-averaged values of climate variables over a range of months, i.e., seasons, for three time horizons. For the RCMs, these time horizons are: a normal period from 1961 to 1990, a near-future period from 2021 to 2050, and a far-future period from 2071 to For the available GCM simulations, the monthly values are averaged from 2046 to 2065 for the near-future (meaning the target time horizon for ECCONET, 2050, lies in the middle of the time span) and for the far future. An ensemble of twelve (12) regional climate models was used for the near-future time horizon ( ) and nine (9) RCMs for the far-future time horizon ( ); far-future simulations are unavailable for some models. The RCMs in this ensemble are driven by only three different GCMs. Thus, four (4) additional GCMs were added to the ensemble in order to investigate the range of possible futures more thoroughly, creating a total of seven (7) GCMs. The names and institutions of the RCMs, their driving GCMs, and the additional GCMs are shown in Table All RCM and GCM model runs are simulated using the SRES A1B emissions scenario as developed by the IPCC (Nakicenovic et al., 2000). In addition to the simulated climate data, model-based discharge simulations for the Rhine are utilized as well. These data have been provided by the KLIWAS project and consist of the lowest seven-day mean discharge (NM7Q in m 3 /s) at gauging station Kaub during hydrological summer (May through October), a season which has exhibited a tendency toward increasingly lower water depths in recent years (see Figure 1.2.2b).

16 ECCONET - Effects of climate change on the inland waterway transport network contract number FP7 16 Table 1.3.1: Overview of climate model projections including the RCMs (top) and GCMs (bottom). Institution RCM Driving GCM Acronym* MPI REMO ECHAM5 A1B_EH5r3_REMO_25 Yes Yes ICTP REGCM ECHAM5 A1B_EH5r3_REGCM_25 Yes Yes KNMI RACMO2 ECHAM5 A1B_EH5r3_RACMO_25 Yes Yes CNRM ALADIN45 ARPEGE A1B_ARP_ALADIN45_25 Yes No METNO HIRHAM HADCM3 A1B_HADCM3Q0_HIRHAM_25 Yes No METNO HIRHAM BCM A1B_BCM_HIRHAM_25 Yes No DMI HIRHAM5 ECHAM5 A1B_EH5r3_HIRHAM_25 Yes Yes SMHI RCA BCM A1B_BCM_RCA_25 Yes Yes SMHI RCA ECHAM5 A1B_EH5r3_RCA_25 Yes Yes UKMO Had3Q3 HadCM3 A1B_HADCM3Q3_HADRM3Q3_25 Yes Yes UKMO Had3Q0 HadCM3 A1B_HADCM3Q0_HADRM3Q0_25 Yes Yes MPI, GKSS CCLM ECHAM5 A1B_EH5r1_CCLM_20 Yes Yes *Nomenclature defined by the KLIWAS project (see Section 1.2). Institution GCM CCCMA CCCma Yes Yes NOAA GFDL 2.1 Yes Yes NIES MIROC HiRes Yes Yes UKMO HadGEM Yes Yes Geographical representations of the Rhine and Upper Danube Catchments In this analysis, both the Rhine and Upper Danube Catchments are defined via boxes composed of latitude and longitude coordinates in order to facilitate ease in comparison of model grids. Due to time constraints, the exact grid is not calculated, i.e., the chosen boxes are not exactly equal to the true river catchments, but rather they attempt to represent the geographical area that includes the portions of the catchment furthest upstream. The box definitions are not exceptionally strict. In the beginning stages of the analysis, the sensitivity of the climate signal was tested using boxes of varying coordinates that were shifted by at least one degree in any direction and it was found that the signal is robust across the various box definitions. Figure shows the chosen domains for this analysis superimposed upon the true river catchments for the Rhine (Figure 1.3.2a) and the Upper Danube (Figure 1.3.2b).

17 ECCONET - Effects of climate change on the inland waterway transport network contract number FP7 17 Figure 1.3.2: (a): The chosen domain for the River Rhine superimposed on the true Rhine River Catchment (Shabalova et al., 2003). The coordinates of the blue box are: 46.5 to 50.5 N; 7 to 10 E; (b): The chosen domain for the Upper Danube superimposed on the true Upper Danube River Catchment. Source: Via Donau. The coordinates of the purple box are: 47 to 50 N; 9 to 16 E Methodology The methodology followed in this section is illustrated in another flow diagram in Figure It is essentially a duplication of Figure 1.3.1, but contains more details related to the contents of the subsections to come. The hatched boxes in the figure indicate work that has already been completed in a previous project (top; green box) or work that will be completed further along in ECCONET (bottom; yellow box). In the solid boxes of the diagram are the titles and numbers of each subsection to follow, and next to the boxes are green arrows which point toward the specific subsection contents. The flow diagram will be discussed in more detail below. As mentioned above, the past work of KLIWAS (green hatched box; top) has determined four extreme model chains which are the climate model simulations that generate the lower and upper boundaries of future low-flows for the River Rhine at station Kaub; for each time horizon and , there is a low and a high model. For clarity, the Rhine extreme model chains are again: 2050: Low: A1B_HADCM3Q0_HADRM3Q0_25 (HAD3Q0) High: A1B_BCM_RCA_25 (SMHI BCM) 2100: Low: A1B_EH5r1_CCLM_20 (CCLM) High: A1B_EH5r3_RACMO_25 (KNMI) The next box in the flow diagram (blue) describes the first stage of the analysis in this study, namely testing the sensitivity of Rhine low-flow discharges to climate variables. Linear relationships are determined between model-based seven-day lowest mean discharge (NM7Q in m 3 /s) for the Rhine during hydrological summer (May through October) and future simulations of two climate variables: precipitation sum (mm/day) and two-metre temperature ( C). The linear relationships are derived for a variety of seasons and a simple correlation analysis is used in order to determine for which season the hydrological summer discharge projections relate strongest to the climate variable projections. The season of strongest

18 ECCONET - Effects of climate change on the inland waterway transport network contract number FP7 18 correlation will then be chosen to link both precipitation and two-metre temperature to discharge for the remainder of the study. Figure 1.3.3: Detailed flow diagram outlining the methodology of determining four extreme model chains for the Upper Danube which will provide an estimate of the range of future low-flows for the river. Hatched boxes indicate work that has been done in a previous project (top; green box) or work that will be completed further along in ECCONET (bottom; yellow box). Inside the solid boxes are the titles and numbers of each subsection to follow; green arrows point toward the specific contents. The next stage of the analysis (purple box) involves further investigating the linear relationships between simulated future low-flows of the Rhine and climate variables in order to verify the four Rhine extreme model chains as chosen by KLIWAS. The verification is performed using a simple correlation analysis and the procedure is completed for both of the climate variables precipitation sum and two-metre temperature. The linear relationships for the Rhine are then assumed to hold for describe the characteristics of changes in low-flows for the Upper Danube as well (orange box). First, the appropriateness of the four Rhine extreme model chains is evaluated for the Upper Danube for both precipitation sum and two-metre temperature. Then, additional GCMs than those used in KLIWAS are employed in order to place the results in the context of an even larger ensemble. Following this, Upper Danube observations in the form of historical time series are consulted in order to (possibly) refine the results. Again, the assumption that the relationships for the Rhine also hold for the Upper Danube is made in order to obtain a tentative, initial assessment of possible future low-flows for the Upper Danube since ensemble hydrological modelling results are not available. The last stage in this analysis utilizes the future Upper Danube precipitation regime, as projected by the RCMs in Table 1.3.1, in order to subjectively determine four extreme model chains for the Upper Danube. These extreme model chains describe the lower and upper boundaries of precipitation sum for each of the future time horizons. During further stages of ECCONET (yellow, hatched box; bottom) the

19 ECCONET - Effects of climate change on the inland waterway transport network contract number FP7 19 extreme model chains will be bias/error corrected and fed into hydrological impact models by VITUKI and BfG in order to estimate the possible range of future low-flows for the river Test sensitivity of low-flow discharges to climate variables As discussed and illustrated in the detailed flow diagram above (Figure 1.3.3, blue box), the initial stage of this study consists of testing the sensitivity of Rhine low-flow discharges to two climate variables, namely precipitation sum and two-metre temperature, for two future time horizons: and This is achieved through deriving linear relationships for a variety of seasons in order to determine for which season the hydrological discharge projections relate strongest to the climate variable projections by using a simple correlation analysis as a proxy. The season which exhibits the strongest correlation will then be chosen to link the climate variables to discharge for the remainder of the study. The hydrological discharge data consists of the lowest seven-day mean discharge (NM7Q) from May through October at gauging station Kaub as simulated by a hydrological impact model using climatological input from regional climate models, from which the area-averaged values for precipitation sum and two-metre temperature are also derived. See Subsection for more details. Precipitation The following figure, Figure 1.3.4, depicts the linear relationship between percent difference in hydrological summer (May through October) low-flow discharges (NM7Q) for the Rhine at station Kaub and percent difference in precipitation sum over the Rhine Catchment during a variety of seasons. The percent differences are determined by comparing simulated results for the normal period ( ) with simulated results for the two future time periods and The results are shown below for the time horizon , hereafter termed the near-future, in Figure 1.3.4a and the time horizon , hereafter termed the far-future, in Figure 1.3.4b. Each individual point in the figure represents the simulated discharge and precipitation sum changes for each individual RCM listed in Table 1.3.1; there are twelve (12) RCM simulations for the near-future and nine (9) for the far-future. Figure 1.3.4: Percent difference in lowest seven-day mean discharge (NM7Q in m 3 /s) during hydrological summer (May through October) at station Kaub for the River Rhine versus percent difference in precipitation sum over the Rhine Catchment during various seasons for the near-future (a) and far-future (b) compared to the normal period ( ). Seasons are indicated in the legend where the letters correspond to their respective months; for example, MJJASO indicates May, June, July, August, September, October. The Pearson s Correlation Coefficient, R, for each season is also listed in the legend. Individual points represent the simulated discharge and precipitation sum for each individual RCM listed in Table

20 ECCONET - Effects of climate change on the inland waterway transport network contract number FP7 20 In examining the linear relationships in the above figure, it is clear that there is a positive, or direct, relationship between low-flow discharge and precipitation sum; that is, while one variable increases (decreases), the other variable also increases (decreases). It is also evident from the figure that May through October (hydrological summer; MJJASO) is the season with the highest Pearson s Correlation Coefficient, R, between discharge and precipitation sum for both time horizons; in the near-future, R=0.48 and in the far-future, R=0.82. Thus, the season of lowest discharge at Kaub correlates strongest with precipitation sum over the Rhine Catchment during the same season, namely hydrological summer from May through October. For the first half of hydrological summer, May, June and July (MJJ), the correlation coefficient is of the same magnitude but slightly weaker for both time horizons; the same is true for the November through October (annual) correlation. In contrast, the precipitation sums during the more wintery periods of September through January (SONDJ), November through January (NDJ) and November through April (NDJFMA; hydrological winter) yield a correlation coefficient of nearly zero. Thus, there appears to be no dependence of hydrological summer low-flows at Kaub on precipitation sum during these seasons. In comparing the two time horizons, it appears that the RCMs project a larger drying signal for the farfuture compared to the near-future. In the near-future, the minimum percent difference in precipitation sum is approximately 10%, whereas for the far-future, the minimum is roughly 20%. The stronger signal is likely due to the larger signal-to-noise ratio in the far-future, making it easier to distinguish the climate signal from noise. The total uncertainty, however, is larger for the far-future than for the nearfuture, though the relative contribution from model uncertainty is less than the uncertainty due to emissions in the far-future. For both time horizons, the value of the Pearson s Correlation Coefficient, R, indicates that low-flow discharge at Kaub during hydrological summer, MJJASO, most strongly depends on precipitation sum over the Rhine Catchment during the same period. Thus, for the remainder of the study, the season MJJASO will be used when linking precipitation sum to low-flow discharge for both time horizons. Two-metre temperature The sensitivity of low-flow discharge to two-metre temperature is also tested for the same seasons and time horizons as above. The linear relationship between the absolute difference in hydrological summer (May through October) low-flows (NM7Q in m 3 /s) for the Rhine at station Kaub and the absolute difference in two-metre temperature in o C over the Rhine Catchment are displayed in Figure Note that for the model projections of two-metre temperature, the absolute difference between the two future time horizons and the normal period is used, rather than the percent difference which is used for precipitation sum. The results for two-metre temperature are shown below for the near-future period in Figure 1.3.5a and the far-future period in Figure 1.3.5b. Again, each individual point represents the simulated discharge and two-metre temperature changes generated by each individual RCM listed in Table