Water Resources of the Black Sea Catchment under Future Climate and Landuse Change Projections

: As water resources become further stressed due to increasing levels of societal demand, understanding the effect of climate and landuse change on various components of the water cycle is of strategic importance. In this study we used a previously developed hydrologic model of the Black Sea Catchment (BSC) to assess the impact of potential climate and landuse changes on the fresh water availability. The BSC model was built, calibrated, and validated against observed daily river discharge for the period of 1973(cid:150)2006 using the Soil and Water Assessment Tool (SWAT) as the modeling tool. We employed the A2 and B2 scenarios of 2017(cid:150)2050 generated by the Danish Regional Climate Model (HIRHAM), and four potential future landuse scenarios based on the Intergovernmental Panel of Climate Change (IPCC)’s special report on emissions scenarios (SRES) storylines, to analyze the impact of climate change and landuse change on the water resources of the BSC. The detailed modeling and the ensemble of the scenarios showed that a substantial part of the catchment will likely experience a decrease in freshwater resources by 30 to 50%.


Introduction
Observational evidence from all continents and oceans shows that natural systems are affected by regional climate changes, particularly by increases in temperature [1]. Nearly all regions of the world are expected to experience a net negative impact of climate change on water resources and freshwater ecosystems [1]. A number of studies have shown that climate change will have significant effects on water availability, water stresses, and water demand [2,3]. Climate change will pose added challenges to managing high disaster-risk areas, as it is virtually certain that the frequency and magnitude of warm daily temperature extremes will increase, and cold extremes will decrease, in the 21st century at the global scale [1]. It is expected that the frequency of heavy precipitation will increase in Southern and Central Europe and the Mediterranean region, and that droughts will intensify because of reduced precipitation and/or increased evapotranspiration [1].
The focus of this study is on the Black Sea Catchment (BSC), which lies in a transition zone between the Mediterranean region in an arid climate of North Africa and the temperate and rainy climate of central Europe. It is affected by interactions between mid-latitude and tropical processes. Because of these features, even relatively minor modifications of the general circulation can lead to substantial changes in the Mediterranean climate [4]. This makes the BSC a potentially vulnerable region to climatic changes as induced, for example, by increasing concentrations of greenhouse gases [4]. Indeed,

Soil Water Assessment Tool (SWAT)
SWAT was used to simulate the hydrology of the BSC. SWAT is a process-based, semidistributed hydrologic model that is developed to quantify the impact of land management practices on water, sediment, and agricultural chemical yields in large complex watersheds with varying soils, landuses, and management conditions over long periods of time. SWAT has been used for the assessment of landuse and management impacts on water quantity and quality in many studies worldwide. The spatial heterogeneity of the watershed is preserved by topographically dividing the catchment into multiple sub-basins. The sub-basins are further subdivided into hydrologic response units (HRU). These are lumped areas within a sub-basin with a unique combination of slope, soil type, and landuse that enable the model to reflect differences in evapotranspiration for various crops and soils. A simulation of the hydrologic cycle is separated into a land phase and a water phase [23]. The simulation of the land phase is based on the water balance equation, which is calculated separately for each HRU. Runoff generated in the HRUs is summed up to calculate the amount of water reaching the main channel in each sub-basin [23]. The water phase of the hydrologic cycle describes the routing of runoff in the river channel, using the variable storage coefficient method by Williams [24]. A detailed description of the model can be obtained from Neitsch et al. [23]. Figure 2 depicts a schematic view of processes accounting for the land phase of the SWAT model.

Soil Water Assessment Tool (SWAT)
SWAT was used to simulate the hydrology of the BSC. SWAT is a process-based, semidistributed hydrologic model that is developed to quantify the impact of land management practices on water, sediment, and agricultural chemical yields in large complex watersheds with varying soils, landuses, and management conditions over long periods of time. SWAT has been used for the assessment of landuse and management impacts on water quantity and quality in many studies worldwide. The spatial heterogeneity of the watershed is preserved by topographically dividing the catchment into multiple sub-basins. The sub-basins are further subdivided into hydrologic response units (HRU). These are lumped areas within a sub-basin with a unique combination of slope, soil type, and landuse that enable the model to reflect differences in evapotranspiration for various crops and soils. A simulation of the hydrologic cycle is separated into a land phase and a water phase [23]. The simulation of the land phase is based on the water balance equation, which is calculated separately for each HRU. Runoff generated in the HRUs is summed up to calculate the amount of water reaching the main channel in each sub-basin [23]. The water phase of the hydrologic cycle describes the routing of runoff in the river channel, using the variable storage coefficient method by Williams [24]. A detailed description of the model can be obtained from Neitsch et al. [23]. Figure 2 depicts a schematic view of processes accounting for the land phase of the SWAT model.

Landuse Scenarios
The landuse scenarios that are used in the current study were developed within the European Union's seventh research framework through the enviroGRIDS project [25]. The developed scenarios comprise a number of plausible storylines based on a coherent set of assumptions, key relationships, and driving forces, to create a set of quantitative, internally consistent, and spatially explicit scenarios of future landuse covering the entire BSC ( Figure 3). A trend of landuse change for different scenarios is summarized in Table 1.

Landuse Scenarios
The landuse scenarios that are used in the current study were developed within the European Union's seventh research framework through the enviroGRIDS project [25]. The developed scenarios comprise a number of plausible storylines based on a coherent set of assumptions, key relationships, and driving forces, to create a set of quantitative, internally consistent, and spatially explicit scenarios of future landuse covering the entire BSC ( Figure 3). A trend of landuse change for different scenarios is summarized in Table 1.

Landuse Scenarios
The landuse scenarios that are used in the current study were developed within the European Union's seventh research framework through the enviroGRIDS project [25]. The developed scenarios comprise a number of plausible storylines based on a coherent set of assumptions, key relationships, and driving forces, to create a set of quantitative, internally consistent, and spatially explicit scenarios of future landuse covering the entire BSC ( Figure 3). A trend of landuse change for different scenarios is summarized in Table 1.  Communities (Eurostat) and the UN World Population and Urbanization Prospects [26]. The European regional projections (forest, grassland, urban and built up, and cropland) were disaggregated at the level of smaller administrative units (Nomenclature of Units for Territorial Statistics, level 2 (NUTS2)), and then used as input to the regional/local land allocation Metronamica model [27] for 214 regions in the BSC. The landuse change scenarios were quantified as yearly changes in landuses on 1 km × 1 km grid cells, in two time steps, 2025 and 2050, for four scenarios covering the entire BSC. The landuse scenarios were developed for cropland, grassland, forest, and urban areas for the BSC countries, while the input data were derived from the MODIS land cover datasets [22] for the years 2001 and 2008. A full description of the scenario developments and quantitative measures of changes in land cover classes is presented by Mancosu et al. [28].

Climate Change Scenarios
Two future climate scenarios, HS and HB, were simulated for the period of 2010 to 2050, representing the IPCC's A2 and B2 scenarios, respectively. In this study, we used precipitation and minimum and maximum temperature data from the PRUDENCE website. The data were downscaled and bias corrected using the Delta method [29] based on Climate Research Unit (CRU) [30] data . The CRU time series data for the control period are perturbed with changes, allowing increase as a function of time. These changes are based on the parted differences of the monthly probability distribution function (PDF). The PDF is partitioned into deciles, and observed time series are gradually perturbed. The assumption behind the Delta method is that future model bias for both mean and variability will be the same as for present-day simulations. Detail on downscaling techniques and a bias correction procedure is available in a study by Gago Da Silva et al. [31].

Model Setup, Calibration, and Validation
The 2.3 million km 2 area of the catchment is divided into 12,982 sub-basins. The sub-basins are further divided into unique combination of soil, landuse, and slope, and formed 89,202 Hydrological Response Units (HRUs). For a calibration and uncertainty analysis, we used the Sequential Uncertainty FItting program SUFI-2, which is a tool for sensitivity analysis, multi-site calibration, and uncertainty analysis. SUFI-2 is linked to SWAT in the SWAT-CUP software [32].
The inputs of the model are summarized in Table 2. The simulation period was 1970-2006, designating the first three years as a warm up period. We used the ArcSWAT2009 interface for the model setup and SWAT2009 Rev528 for a model run. Table 3 gives an overview of the relevant methods used in the model setup. More details of the hydrologic model's structure, setup, and performance in the BSC can be found in Rouholahnejad et al. [19]. For future simulations (2017-2050, including 3 years of warm up period), the two climate change scenarios (HB and HS) were paired with the four landuse change scenarios (BS ALONE, BS COOL, BS COOP, and BS HOT), leading to eight combinations of climate-landuse scenarios, and applied as inputs to the calibrated hydrologic model of the catchment in eight separate model setups. In addition, two model setups were designed to look at climate change only, with static landuse. The latter models used MODIS landuse data of the historical hydrologic model as the static landuse. In the other eight combinations of landuse-climate scenarios, a dynamically updating algorithm updated the landuses yearly up to the end of the simulation period. A total of 10 different scenarios were therefore built and analyzed.

Results and Discussion
The BSC hydrologic model was calibrated  and validated (1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006) at 144 river discharge stations at daily time scales. The most sensitive parameters to discharge are as shown in Table 4. An example of the simulated and observed stream flow along with the predictive uncertainty band for the two labeled river discharge stations in Figure 1 are presented in Figure 4. These are two examples of calibrated discharge stations, one at the border of Hungary and Romania and one in Romania. We ran the model and calibration simulations at a daily time scale. However, given the scale of the watershed, the outputs were extracted at monthly time scales.   Figure 1). Shown in the picture are observation time series, best simulation along with 95% prediction uncertainty band (green band). The P-factor is the percentage of measured data bracketed by the 95 PPU band. It ranges from 0 to 1, where 1 is ideal and means all of the measured data are within the uncertainty band. The R-factor is the average width of the band divided by the standard deviation of the measured variable. It ranges from 0 to 1, where 0 reflects a perfect match with the observation. Based on the experience, an R-factor of around 1 is usually desirable. NS and R 2 are Nash-Sutcliffe efficiency and coefficient of determination, respectively.

Temperature and Precipitation in a Changing Climate
The HB and HS long-term average temperature scenarios show a 1-2.4 °C temperature increase with a west to east gradient in the catchment. While the overall long-term pattern of increase in temperature is similar in the two scenarios, HS depicts a larger temperature increase and over larger areas (Figure 5g,h). The variation over time in future temperature scenarios (2020-2050) has almost the same pattern as the historic temporal variation  with some discrepancies in the Danube basin (Figure 5b,d,f).  Figure 1). Shown in the picture are observation time series, best simulation along with 95% prediction uncertainty band (green band). The P-factor is the percentage of measured data bracketed by the 95 PPU band. It ranges from 0 to 1, where 1 is ideal and means all of the measured data are within the uncertainty band. The R-factor is the average width of the band divided by the standard deviation of the measured variable. It ranges from 0 to 1, where 0 reflects a perfect match with the observation. Based on the experience, an R-factor of around 1 is usually desirable. NS and R 2 are Nash-Sutcliffe efficiency and coefficient of determination, respectively.

Temperature and Precipitation in a Changing Climate
The HB and HS long-term average temperature scenarios show a 1-2.4 • C temperature increase with a west to east gradient in the catchment. While the overall long-term pattern of increase in temperature is similar in the two scenarios, HS depicts a larger temperature increase and over larger areas (Figure 5g,h). The variation over time in future temperature scenarios (2020-2050) has almost the same pattern as the historic temporal variation  with some discrepancies in the Danube basin (Figure 5b,d,f). The historical distribution of average precipitation (Figure 6a), along with its coefficient of variation (CV) over time (Figure 6b), mark distinct precipitation distribution patterns with significant temporal variation (as indicated by the high CV values) in regions around the Sea, and in the Danube, The historical distribution of average precipitation (Figure 6a), along with its coefficient of variation (CV) over time (Figure 6b), mark distinct precipitation distribution patterns with significant temporal variation (as indicated by the high CV values) in regions around the Sea, and in the Danube, Dnieper, and Don River basins. Historically, the upper Western part of the catchment in the alpine region receives rain of about 1300 mm year −1 on average, which is much higher than precipitation rates in other parts of the BSC where Ukraine, Russia, and Turkey lie. The regions with a smaller historical precipitation rates tend to show a higher temporal variation in precipitation (Figure 6b). Dnieper, and Don River basins. Historically, the upper Western part of the catchment in the alpine region receives rain of about 1300 mm year −1 on average, which is much higher than precipitation rates in other parts of the BSC where Ukraine, Russia, and Turkey lie. The regions with a smaller historical precipitation rates tend to show a higher temporal variation in precipitation (Figure 6b). The two future precipitation scenarios are similar to the historic ones with regard to patterns of long-term annual averages (Figure 6a,c,e). In future scenarios, the areas with small precipitation expand as compared to the historic period (Figure 6a,c,e). The HS scenario, in general, suggests larger decreases in precipitation in the catchment.
The anomaly maps of precipitation (Figure 6g,h) depict percent deviation from historic precipitation for the entire catchment. The differences are calculated between the averages of 2020-2050 with those of 1973-2006. The HB scenario suggests a 5-15% decrease in precipitation in most regions of the catchment, while the HS suggests a 10-24% decrease in the precipitation of Danube Basin (West of the Black Sea) and a 4-10% decrease in precipitation in the rest of the catchment.

Fresh Water Resources under Changing Landuse and Climate
The term "blue water" [41] (Figure 7b). In other words, the less the blue water is available, the more its variability over time is. The anomaly map of blue water (Figure 7c) depicts the deviation of long-term average blue water resources for the period of 2020-2050 from long-term average historic records  under future landuse and climate change scenarios (ensemble of 10 scenarios).
The blue water resources calculated with the ensemble of 10 scenarios suggest a 10-50% decrease in blue water resources in most parts of the catchment (Figure 7c). According to the future scenario ensembles, blue water increases on average by 50% in the coastal areas of Georgia and Turkey, with historically small blue water resources. However, this does not bring a significant increase in terms of net blue water resources availability of the whole catchment, as the historical records in these regions are quite small. Historical variations of blue water indicate low reliability (higher variability over time) of blue water resources in Romania, parts of Ukraine, the Russian parts of the catchment, and Turkey ( Figure 7b). Our analysis shows that the poor conditions in these regions in terms of fresh water availability will be further intensified under climate change ( Figure 7c).
As soil moisture is an integral component of rainfed agriculture, the soil water distributions projected under future climate change and landuse change scenarios are of a strategic importance. The spatial variation of long-term average annual green water storage (soil moisture) (Figure 8a) shows that most of the catchment lies in the range of 70-250 mm of soil moisture. However, the variation over time shows a distinct East-West pattern, suggesting different levels of reliability for soil water in the region. The Southern parts of the catchment tend to have smaller soil water with more variation over time (Figure 8b), which makes the region less reliable in terms of green water storage resources.
The anomaly map of soil water (average of 10 scenarios) depicts the deviation of average future (2020-2050) soil moisture from the average historic  soil moisture (Figure 8c). The average of the 10 scenarios shows up to a 10-25% reduction in soil water in the Danube catchment and Northern BSC in the upstream of Dnieper and Don, in Ukraine and Russia. There are indications of soil moisture increase in Georgia, where both climate scenarios suggested an increase in precipitation. The coefficient of variation among the 10 scenarios indicates that there is a good agreement between the scenarios with less than 2% variation in the prediction of soil water in most parts of the catchment and 10-15% variation in the Danube Delta and areas surrounding the Black Sea (Figure 8d). The average of 10 scenarios suggests both an increase and a decrease in evapotranspiration across the BSC under climate and landuse change scenarios (Figure 9c). There is also a sharp increase in evapotranspiration of Georgia, which fits the increase in precipitation and temperature in this area. The variation among the 10 scenarios is the highest in the Danube delta and the Black Sea costal area. The average deviation from historic values (Figure 9c) suggests that evapotranspiration decreases by up to 12% in the Danube basin under future scenarios of change. The variation between model predictions of blue water, evapotranspiration, and soil moisture using four different future landuse scenarios shows limited impacts of landuse changes as compared to climate (the coefficient of variation is less than 1.3% among the four landuse scenarios) ( Figure 10). Hence, the climatic signature is more significant than landuse in this study. The variations between scenarios are more pronounced in the blue water predictions (Figure 10a,b). The average of 10 scenarios suggests both an increase and a decrease in evapotranspiration across the BSC under climate and landuse change scenarios (Figure 9c). There is also a sharp increase in evapotranspiration of Georgia, which fits the increase in precipitation and temperature in this area. The variation among the 10 scenarios is the highest in the Danube delta and the Black Sea costal area. The average deviation from historic values (Figure 9c) suggests that evapotranspiration decreases by up to 12% in the Danube basin under future scenarios of change. The variation between model predictions of blue water, evapotranspiration, and soil moisture using four different future landuse scenarios shows limited impacts of landuse changes as compared to climate (the coefficient of variation is less than 1.3% among the four landuse scenarios) ( Figure 10). Hence, the climatic signature is more significant than landuse in this study. The variations between scenarios are more pronounced in the blue water predictions (Figure 10a,b).

Extreme Events
We compared the frequency of occurrences of wet days with the precipitation thresholds of >2 and >10 mm d −1 in five selected sub-basins in different climatic regions across the BSC (Figure 11). In general, although the long-term average precipitation based on the two scenarios suggests a general decrease in precipitation in the catchment, the frequency of the wet days are slightly higher under future scenarios. In a sub-basin in the Eastern part of the BSC in Russia, the two climate scenarios

Extreme Events
We compared the frequency of occurrences of wet days with the precipitation thresholds of >2 and >10 mm d −1 in five selected sub-basins in different climatic regions across the BSC (Figure 11). In general, although the long-term average precipitation based on the two scenarios suggests a general decrease in precipitation in the catchment, the frequency of the wet days are slightly higher under future scenarios. In a sub-basin in the Eastern part of the BSC in Russia, the two climate scenarios predicted a slightly higher number of days with precipitation larger than 2 mm as compared to historical climate (Figure 11a,b). This is a region with low annual rainfall (350-450 mm year −1 ) where there are few rainfall events exceeding 10 mm d −1 throughout the year. In a sub-basin in Southern Ukraine with 450-500 mm year −1 average annual precipitation, the frequency of wet days at the threshold of >2 mm d −1 stays as large as the historic period (Figure 11c). The slight decreases in number of days with >10 mm d −1 rainfall events indicate a smaller groundwater recharge, hence a larger chance of receding groundwater.
Water 2017, 9,598 14 of 18 predicted a slightly higher number of days with precipitation larger than 2 mm as compared to historical climate (Figure 11a,b). This is a region with low annual rainfall (350-450 mm year −1 ) where there are few rainfall events exceeding 10 mm d −1 throughout the year. In a sub-basin in Southern Ukraine with 450-500 mm year −1 average annual precipitation, the frequency of wet days at the threshold of >2 mm d −1 stays as large as the historic period (Figure 11c). The slight decreases in number of days with >10 mm d −1 rainfall events indicate a smaller groundwater recharge, hence a larger chance of receding groundwater.  In the selected sub-basin in Austria with a rainfall rate of 1000-1350 mm year −1 , the frequencies of wet days with more than 2 mm d −1 rainfall increase as compared with the historical data (Figure 11e).
In a sub-basin in the Alpine region in the middle of the Danube Basin, historic records of precipitation range between 1000-1350 mm year −1 . Both the HB and HS scenarios predict an increase in the number of wet days with a threshold of >2 mm d −1 , while the increase in the winter months is more distinct. The HS scenario predicts larger wet-day frequencies with the threshold of >10 mm d −1 than the HB scenario. The increase in precipitation frequencies at the threshold of >10 mm d −1 may indicate more flood risks in this region. However, the increase in the frequencies of precipitation with a large threshold (>10 mm d −1 ) stay within the historic records in this sub-basin (Figure 11h).
Finally, in a sub-basins in Turkey, the climate models behave differently from the Alpine region, as the HB and HS climate scenarios predict a distinctly larger number of days with precipitation more than 2 mm d −1 than what the historic records show. The precipitation events at 10 mm d −1 threshold, however, are predicted to decrease in this selected sub-basin in Turkey (Figure 11i,j).

Summary
Combinations of two regional climate scenarios and four regional landuse scenarios were incorporated in the current study to explore the possible future impacts of climate and landuse changes on water resources of the Black Sea Catchment. The landuse scenarios were driven by the IPCC's special report on emissions scenarios (SRES) corresponding to four marker scenarios that represent different global socio-economic development pathways. The climate scenarios were generated from the Danish Regional Climate Model (RCM) (HIRHAM) for the IPCC's SRES A2 and B2 scenarios (HS and HB scenarios respectively). On average, the climate scenarios suggested a 5-15% decrease in future long-term average annual precipitation in most parts of the catchment. The decrease in precipitation is more pronounced in the HS scenario. According to the HS climate scenario, the Western part of the catchment (Danube Basin) will experience a decline in precipitation by 25%. As the historic precipitation records are large in this region, this is expected to have a large impact on the water resources of the entire region, and leaves the catchment with a significant net decrease of precipitation. Both scenarios suggest an increase in temperature by up to 2 • C with a west to east gradient. The extent of changes in temperature is more severe in the HS scenario as compared to the HB scenario.
We also quantified the impacts of combined climate and landuse changes on freshwater distribution in the BSC. As suggested by the ensemble of scenarios, on average, the catchment is expected to experience a decrease in its blue water and green water storage resources, while the green water flow (evapotranspiration) increases in some parts of the catchment and decreases in other parts (Figures 7-9). In addition, the decrease in fresh water resources in areas with high temporal variability in their water resources component (mainly in low lying countries around the Black Sea, such as Romania and Ukraine and the Russian part of the catchment) increases the vulnerability with regard to fresh water resources in these regions.
In our analysis, climate change had more pronounced effects on water resources, especially blue water, as opposed to the landuse change. To see the detailed effect of landuse change on the water resources component, it is beneficial to look at the water cycle at the HRU level, where the landuses are identical. This will give a true measure of landuse change impacts on water resources. The strength of the current work is the application of combined landuse and climate change scenarios. However, the study neglects the future changes in soil parameters over time, which accompanies changing landuses. Accounting for these changes will increase the confidence in the projected results, and needs to be further investigated.
An additional concern is the use of two regional climate scenarios (HS and HB) in model prediction while pursuing a thorough investigation based on the combined effect of many other Global Climate Models or Regional Climate Models would reflect climate model uncertainties, and hence is recommended. The study however, provides the basis to improve societal capabilities to anticipate and manage water resources both today and in the future climate change environment in the Black Sea Catchment.