Water Quantity and Quality under Future Climate and Societal Scenarios: A Basin-Wide Approach Applied to the Sorraia River, Portugal

: Water resources are impacted by several stressors like over-population and over consumption that compromises their availability. These stressors are expected to progressively intensify due to climate change in most regions of the world, with direct impact on watersheds and river systems. This study investigates the effect of different watershed pressure scenarios due to climate change in the hydrological regime of the Sorraia River basin, Portugal. This catchment includes one of the largest irrigated areas in the country, thus being strongly inﬂuenced by anthropogenic activities, associated to hydrological (irrigation, ﬂow regulation, damming) and nutrient stressors. The Soil Water Assessment Tool has been used to simulate water ﬂow and nutrient dynamics in the watershed while considering inputs from two climate models and three societal scenarios. Results have shown that the predicted rainfall reductions will have a signiﬁcant impact on river ﬂow and nutrient concentrations when compared to baseline conditions. River ﬂow will expectably decrease by 75%, while nitrogen and phosphorus concentrations in river water will expectably increase by 500% and 200%, respectively. These differences are more evident for storylines that consider increasing pressures such as population growth and agricultural expansion marked with unsustainable practices and increased reliance on technology. The results of this study indicate a possible future outcome and provide effective guidelines for the formulation of water management policies to counter the impacts of climate change and corresponding environmental pressures in the Sorraia River basin.


Introduction
Agriculture is one of the main factors responsible for variation in the landscape of a region [1]. Unsustainable agricultural practices and excessive urban expansion have drastically affected the hydro morphological characteristics of river systems in the wake of climate change. The ever-increasing world population and globalisation of food products is one of the major reasons for the expansion of agroecosystems. High demands from the global food market have turned small and large-scale farmers towards excessive mechanized farming, the overuse of fertilizers and pesticides, as well as the

Study Area
This study was carried out in the Sorraia River (southern Portugal), the tributary of the Tagus River with the largest basin area, with ~7730 km 2 (Lat: 38.59° to 39.50°; Long: −8.99° to −7.24°), and with a longitudinal length of ~155 km (Figure 1). The climate in the region is dry sub-humid, with dry and hot summers, and mild and wet winters. Records from 14 meteorological stations [25] for a 20-year period (1996 to 2015) showed that the annual precipitation in the region varied from 200 to 900 mm, with the average of ~500 mm. The average monthly precipitation was ~50 mm, fluctuating up to 25 mm from April to September and 70 mm between October and March. The average annual surface air temperature was ~15 °C, varying from ~9 to ~22 °C. The reference evapotranspiration estimated according to Allen et al. [26] reached ~900 mm. The dominant soil types found in the region are Cambisols, Luvisols, and Regosols [27]. Fluvisols are also found at higher concentrations in the downstream irrigated areas. Two major reservoirs, Montargil and Maranhão, were built in the watershed during the second half of the twentieth century as a part of the Sorraia Valley Irrigation Implementation Plan. Currently, the Sorraia Valley is one of the largest irrigation areas in Portugal, totalizing 16,000 ha, in which corn (Zea mays L.), rice (Oryza sativa L.) and tomato (Solanum lycopersicum L.) predominate. The land use in the remaining area of the watershed is characterized by holm oak forest, rainfed cereals and pasture.
In terms of human population, the Sorraia watershed has a total of 153,100 inhabitants, with a density of 20 inhabitants km −2 [28]. The population is mainly concentrated in three core cities: Ponte de Sôr (16,700 inhabitants), Samora Correia (17,123), and Coruche (19,950). According to the River Basin Management Plan (RBMP [29]), hydro-morphological changes, diffuse pollution, municipal discharges, flow regulation, and water abstraction are the main pressures in the basin. Two major reservoirs, Montargil and Maranhão, were built in the watershed during the second half of the twentieth century as a part of the Sorraia Valley Irrigation Implementation Plan. Currently, the Sorraia Valley is one of the largest irrigation areas in Portugal, totalizing 16,000 ha, in which corn (Zea mays L.), rice (Oryza sativa L.) and tomato (Solanum lycopersicum L.) predominate. The land use in the remaining area of the watershed is characterized by holm oak forest, rainfed cereals and pasture.
In terms of human population, the Sorraia watershed has a total of 153,100 inhabitants, with a density of 20 inhabitants km −2 [28]. The population is mainly concentrated in three core cities: Ponte de Sôr (16,700 inhabitants), Samora Correia (17,123), and Coruche (19,950). According to the River Basin Management Plan (RBMP [29]), hydro-morphological changes, diffuse pollution, municipal discharges, flow regulation, and water abstraction are the main pressures in the basin.

The Soil and Water Assessment Tool (SWAT) Model
The SWAT model [24] is widely used to simulate watershed processes [18,19,[30][31][32][33]. SWAT is a semi-distributed watershed model focused on land management at a basin scale. The model splits the watershed into sub-basins that are assumed to be homogeneous in their Hydrologic Response Units (HRU), i.e., in terms of land use, soil and topographic characteristics. The relative straightforward formulation used in SWAT allows the model to run more demanding simulations within a reasonable time. The hydrology of the model is based on the daily water balance equation, as follows: (1) where SW t is the final soil water content (mm), SW 0 is the soil water content at the initial time step (mm), R day is the daily precipitation (mm), Q surf is the surface runoff (mm), E a is the actual evapotranspiration (mm), W seep is the percolated water (mm), and Q gw is the return flow (mm), all referring to day i, which varies from 1 to the number of simulated days (n). In this study, the potential evapotranspiration rates were estimated using the Penman-Monteith method [26], with E a being then dependent on soil water availability. Surface runoff was computed from daily precipitation using a modification of the Soil Conservation Service Curve Number (SCS-CN) method [34]. Groundwater recharge was estimated by combining a storage routing technique and a crack-flow model. The lateral flow was simulated using a kinematic storage method [24]. The SWAT model can further simulate the nitrogen (N) and phosphorus (P) cycles. The N present in the soil is represented by five different pools, considering mineral and organic forms. The mineral N is divided into two pools: ammonia (NH 4 + ) and nitrate (NO 3 − ). The organic N is divided into three pools: active, stable (associated to the humic substances) and fresh pool (associated to the crop residue). Molecular nitrogen is added naturally by biological and atmospheric nitrogen fixation. Anthropogenic activities and agricultural practices such as the use of fertilizers also act as sources of N in the environment. After fixation, N is converted to NH 4 + in the soil and then consumed by plants in the form of NO 3 − . Mineralisation is considered by the fresh organic pool associated with crop residues and the active pool associated with soil humus. At harvest, the remaining fraction of crop residues is incorporated into the first soil layer. N is removed from the soil through volatilisation, denitrification, erosion, and leaching (in the NO 3 − form). The total amount of NH 4 + lost by volatilisation or nitrification is calculated considering the NH 4 + amount and environmental factors. The SWAT code calculates the denitrification process according to the soil carbon and nitrate source, a rate coefficient, and different environmental factors such as temperature and soil water content. The denitrification process occurs in anaerobic conditions. In the case of the phosphorus cycle, the model includes three inorganic P pools (solution, active, and stable) and three organic P pools (active, stable and fresh). The fresh organic pool is associated with crop residue and microbial biomass, while the active and stable organic pools are associated with the soil humus. The sum of the six pools represents total soil P. P is simulated considering the supply and demand during plant growth. Soluble and organic forms can be removed from the soil via mass flow (runoff). The amount of soluble P removed in runoff is predicted using solution P concentration in the top 10 mm of the soil profile, the runoff volume, and a partitioning factor. Sediment transport of P is then simulated with a loading function [24]. In this study, the SWAT model was applied to the Sorraia basin using the ArcGIS extension from ESRI (Redlands, CA). The model application relied on available Geographical Information System maps for topography (SRTM), land use maps from Earth Observation (GSE Land M2.1) and soil maps and data from Cardoso et al. [35] (SROA). Daily discharge data provided by the reservoirs' manager (ARBVS-Farmers Association from the Sorraia Valley) were used in the model from 1996 to 2015. Meteorological time series were downloaded for the basin area from the National Water Resources Institute website (SNIRH). The input data considered for the present conditions (hereafter referred to as the baseline) are summarized in Table 1. In order to stabilize the model conditions, the period between 1996 and 2000 was considered as the warm-up period, and the baseline simulation was defined for the period 2001-2015.

Calibration and Validation
The SWAT model was calibrated by manually modifying one parameter at a time, considering the most sensitive parameters that determined the best results for simulating daily/monthly river flows. The hydrograms were analysed, and the parameters that affected flow peaks and baseflow were selected and modified until deviations between the model outputs and measured flow data were minimised ( Table 2). The previously calibrated parameters were then validated by comparing the results of the simulations with an independent dataset. Model calibration was thus performed for the period between 2001 and 2006, while the validation exercise was carried out from 2006 to 2015.  [25]. However, its location and the monitoring interval of measured values (usually 15 days to 1 month) hindered the representativeness of the monitoring record.
The goodness-of-fit indicators adopted for comparing model outputs with measured flow data were the coefficient of determination (R 2 ), the root mean square error (RMSE), the Nash-Sutcliffe model efficiency coefficient (NSE [37]), and the Model Bias (Bias). An R 2 value close to 1 indicates that the model explains the variance of observations well. RMSE values close to zero indicate small estimation errors and good model predictions. Bias was defined as the average difference between the estimator and the true value. Bias values close to zero indicate no under or over-estimation of the measured results. NSE values close to 1 indicate a perfect match of modelled discharge to the observed data, hence, indicating that the model predictions are good. On the contrary, when NSE is very close to 0 or negative, there is no gain in using the model.
For N and P, comparison was focused on the magnitude of the simulated and observed values due to data limitation, considering only R 2 and Bias indicators.

Storylines
The storylines defined in this study followed the framework developed within the Managing Aquatic Ecosystems and Water Resources Under Multiple Stress Project-MARS [3,38]. Here, the storylines are defined as a combination of societal and climate scenarios. Three storylines were considered by combining the work of O'Neill et al. [39] and Riahi et al. [14]. These authors defined Shared Socioeconomic Pathways (SSP's) as reference scenarios describing plausible alternative trends in the evolution of society and ecosystems over a century timescale in the absence of climate change or climate policies. Storylines also considered Moss et al. [40] who developed Representative Concentration Pathways (RCPs) for greenhouse emissions. RCP 4.5 assume that greenhouse gas emissions will peak around 2040 followed by a decline, while RCP 8.5 considers that emissions will increase throughout the 21st century.
Climate data considered in this study, such as the surface air temperature and precipitation, was extracted from the Inter-Sectoral Impact Model Intercomparison (ISI-MIP) project as it provided the best temporal and spatial resolution for the study area. In the ISI-MIP project, the bias-corrected time-series of surface air temperature and precipitation were downscaled at a 0.5 • resolution [41]. Following the IPCC's Fifth Assessment Report, the ISI-MIP project run specific climate models to obtain data [42], which included five of the Coupled Model Intercomparison Project Phase 5 (CMIP5) Global circulation models (GFDL-ESM2M, HadGEM2-ES, IPSL-CM5A-LR, MIROC-ESM-CHEM and NorESM1-M).
As established by Faneca et al. [43] and Birk et al. [44], the storylines considered in this study are the following (Figure 2): Storyline 1 (STL1): Techno World. It represents fast global economic growth, characterised by a rapid technological development but with high energy demands and no real drive to enhance or ignore the health of natural ecosystems. This world is based on a combination of SSP-5 which consider a conventional development and low population [39] and climate scenario RCP 8.5 [40]; Storyline 2 (STL2): Consensus world. It is a world in which the actual policies continue after 2020. The growth of economy keeps to the same pace as now, but with awareness for environment preservation. This world is based on a combination of SSP-2 which is considered as the intermediate stage [39] and the climate scenario RCP 4.5 [40]; Storyline 3 (STL3): Survival of the fittest. It represents a fragmented world, driven by the individual interest of countries, with fast economic growth in NW Europe but with recessions in other regions; with minimal or no investment and effort in environmental protection, conservation and restoration. This world is based on a combination of SSP-3, which consider a rapid technology for fossils, high demand and high economic growth [39] and climate scenario RCP 8.5 [40].
As established by Faneca et al. [43] and Birk et al. [44], the storylines considered in this study are the following ( Figure 2): Storyline 1 (STL1): Techno World. It represents fast global economic growth, characterised by a rapid technological development but with high energy demands and no real drive to enhance or ignore the health of natural ecosystems. This world is based on a combination of SSP-5 which consider a conventional development and low population [39] and climate scenario RCP 8.5 [40]; Storyline 2 (STL2): Consensus world. It is a world in which the actual policies continue after 2020. The growth of economy keeps to the same pace as now, but with awareness for environment The two climate models adopted for each climate scenario ( Figure 2) were: GFDL-ESM2M [45,46]: RCP 4.5 was used in Storyline 2 (hereafter referred as STL2 GFDL), and RCP8.5 was used in Storyline 1 (STL1 GFDL) and Storyline 3 (STL3 GFDL); IPSL-CM5A-LR [47]: RCP 4.5 was used in Storyline 2 (STL2 IPSL), and RCP8.5 was used in Storyline 1 (STL1 IPSL) and Storyline 3 (STL3 IPSL).
GFDL and IPSL were adopted in this study as they give results close to the ISI-MIP median for the Western Europe region [43]. These two models differ in terms of the atmospheric prognostic state and of the spatial resolution of their atmospheric grid (2.5 • lon. by 2.0 • lat. for GFDL, and 2.5 • lon. by 3.75 • lat. for IPSL). Further conceptual differences between the GFDL and IPSL models, and the downscaling of variables at the basin-scale acquired from the outputs of these two models are given in Warszawski et al. [41] and references therein. Bias-corrected time-series of air temperature and precipitation downscaled at a 0.5 • resolution [48] were considered. Additional bias correction for the Sorraia basin was applied on precipitation ( Figure 3) and surface air temperature ( Figure 4) values following Shrestha [49] and Shrestha et al. [50] and considering the measured data from 2006-2015. This is a statistical downscaling method known as Linear scanning bias correction based on the average difference between monthly observed and historical time series for the same period.
Two distinct temporal intervals were set up to run the simulations: 2030 (defined as a 10-year average from 2025 to 2034) and 2060 (defined as a 10-year average from 2055 to 2064). Thus, the term "climate change" refers here only to decadal changes. The period 1996-2015 was selected as a reference for the baseline simulation (present condition). preservation. This world is based on a combination of SSP-2 which is considered as the intermediate stage [39] and the climate scenario RCP 4.5 [40]; Storyline 3 (STL3): Survival of the fittest. It represents a fragmented world, driven by the individual interest of countries, with fast economic growth in NW Europe but with recessions in other regions; with minimal or no investment and effort in environmental protection, conservation and restoration. This world is based on a combination of SSP-3, which consider a rapid technology for fossils, high demand and high economic growth [39] and climate scenario RCP 8.5 [40].
GFDL and IPSL were adopted in this study as they give results close to the ISI-MIP median for the Western Europe region [43]. These two models differ in terms of the atmospheric prognostic state and of the spatial resolution of their atmospheric grid (2.5° lon. by 2.0° lat. for GFDL, and 2.5° lon. by 3.75° lat. for IPSL). Further conceptual differences between the GFDL and IPSL models, and the downscaling of variables at the basin-scale acquired from the outputs of these two models are given in Warszawski et al. [41] and references therein. Bias-corrected time-series of air temperature and precipitation downscaled at a 0.5° resolution [48] were considered. Additional bias correction for the Sorraia basin was applied on precipitation ( Figure 3) and surface air temperature ( Figure 4) values following Shrestha [49] and Shrestha et al. [50] and considering the measured data from 2006-2015. This is a statistical downscaling method known as Linear scanning bias correction based on the average difference between monthly observed and historical time series for the same period.    Two distinct temporal intervals were set up to run the simulations: 2030 (defined as a 10-year average from 2025 to 2034) and 2060 (defined as a 10-year average from 2055 to 2064). Thus, the term "climate change" refers here only to decadal changes. The period 1996-2015 was selected as a reference for the baseline simulation (present condition).
The downscaling of the socio-economic factors and foreseen management practices change in the Sorraia catchment was performed with the help of local water board stakeholders (Associação de Regantes e Beneficiários do Vale do Sorraia). Therefore, each storyline was naturally translated into quantitative data assuming that the Mediterranean climate imposes additional stress on the agriculture [43]. This additional stress was defined in terms of management practice changes, namely on the amount of fertilizer and irrigation applied to crops (Table 3). That assumption considered that in a changing environment, with temperatures increasing, higher irrigation needs are required to fulfil crop requirements. Additionally, higher temperatures lead to higher mineralisation rates and, thus, higher use of fertilizers. It also considered that higher usage of fertilizers and irrigation lead to larger diffuse pollutions from agricultural fields and greater environmental risk. Model inputs were thus related to the level of agriculture intensification and the environmental protection awareness considered in each Storyline. Fertilizers were applied to prevent nutrient stress by plants. The percentage variation (increase or decrease) defined in Table 3 for each storyline and timeline was applied on the baseline values and used as an input in the model.  The downscaling of the socio-economic factors and foreseen management practices change in the Sorraia catchment was performed with the help of local water board stakeholders (Associação de Regantes e Beneficiários do Vale do Sorraia). Therefore, each storyline was naturally translated into quantitative data assuming that the Mediterranean climate imposes additional stress on the agriculture [43]. This additional stress was defined in terms of management practice changes, namely on the amount of fertilizer and irrigation applied to crops (Table 3). That assumption considered that in a changing environment, with temperatures increasing, higher irrigation needs are required to fulfil crop requirements. Additionally, higher temperatures lead to higher mineralisation rates and, thus, higher use of fertilizers. It also considered that higher usage of fertilizers and irrigation lead to larger diffuse pollutions from agricultural fields and greater environmental risk. Model inputs were thus related to the level of agriculture intensification and the environmental protection awareness considered in each Storyline. Fertilizers were applied to prevent nutrient stress by plants. The percentage variation (increase or decrease) defined in Table 3 for each storyline and timeline was applied on the baseline values and used as an input in the model.

Model Calibration/Validation
Model calibration was carried out considering the period between 2001 and 2006, and validation was performed considering the period between 2006 and 2015. The statistical indicators obtained after comparing the daily and monthly simulated and measured flow values at the selected monitored stations are presented in Table 4. At Moinho Novo, the R 2 value of 0.71 for monthly data shows that a considerable proportion of variability of the observed data was explained by the model (Table 4 and Figure 5). The RMSE value of 6 m 3 /month indicates a small error of model estimates, while the NSE value of 0.71 indicates that the residual variance results were much smaller than the measured data variance (Table 4 and Figure 5). The comparison of daily values produced, as expected, worse results, with errors being mostly minimized during the monthly analysis due to data aggregation. For the validation period, the indicators were found to be similar, indicating a reasonable calibration of model parameters when considering all the uncertainties related to measurements.

Model Calibration/Validation
Model calibration was carried out considering the period between 2001 and 2006, and validation was performed considering the period between 2006 and 2015. The statistical indicators obtained after comparing the daily and monthly simulated and measured flow values at the selected monitored stations are presented in Table 4. At Moinho Novo, the R 2 value of 0.71 for monthly data shows that a considerable proportion of variability of the observed data was explained by the model ( Table 4 and Figure 5). The RMSE value of 6 m 3 /month indicates a small error of model estimates, while the NSE value of 0.71 indicates that the residual variance results were much smaller than the measured data variance (Table 4 and Figure 5). The comparison of daily values produced, as expected, worse results, with errors being mostly minimized during the monthly analysis due to data aggregation. For the validation period, the indicators were found to be similar, indicating a reasonable calibration of model parameters when considering all the uncertainties related to measurements.   At Ponte Vila Formosa, the statistical indicators found for the calibration and validation periods were similar to those obtained at Moinho Novo. The exception was the negative NSE value computed for daily values, which suggests that the average of observed values has a larger predictive power than the model results. Nonetheless, the NSE statistics computed from monthly values are positive and the average monthly flows have a behaviour similar to the measured data ( Figure 5). For N and P, no calibration was carried out due to the data limitation with the default parameters of the SWAT model being considered. The evaluation of model outputs focused on its ability to reproduce the order of magnitude and the peaks of the measured values, which proved that the model is able to offer reasonable results. For the Total N at Ponte de Coruche, the following statistical indicators from the comparison between simulated and measured concentrations were obtained: R 2 = 0.59; Bias = 0.22 mgN L −1 . For the Total P, the following statistical indicators were also obtained: R 2 = 0.14; Bias = −0.067 mgP L −1 .

Water Quantity
After calibration and validation, the SWAT model was used to evaluate the developed storylines. Climate models showed a decrease in precipitation from 400 mm year −1 to an average of 264 mm year −1 on both timelines ( Figure 6). As a result, the simulated scenarios showed a substantial reduction of monthly flows for the 2030 and 2060 timelines (Figure 7). Flow reduction evidenced a non-linear relationship between precipitation and monthly river flow averages. This resulted from the uncertainty associated with climate models, particularly in the forecast of extremes events ( Figure 6). The monthly average flow for the baseline simulation was 42 m 3 s −1 (Table 5), while the IPSL model showed a decrease by half of this value in the timeline 2030 (to about 19 m 3 s −1 ), and a further decrease in the timeline 2060 (to 10 m 3 s −1 ). More severe results were obtained with the GFDL model for the same scenarios, predicting a decrease in the precipitation average to half of the present value, which resulted in monthly flow averages of 4.5 m 3 s −1 and 2.5 m 3 s −1 for the timelines 2030 and 2060, respectively (Table 5). Moreover, the increasing temperature trend predicted by both climate models combined with decreasing precipitation further led to more hazardous agricultural practices for the environment, with irrigation and fertilizer requirements increasing as considered in STL1 and STL3. Even for the scenario with more sustainable agricultural practices (STL 2), the changes in the climate variables had the same severe effect on river flow ( Figure 5). For both climate scenarios, the decrease in irrigation in STL2 resulted in small differences in the monthly flow average when compared with the previous results (Table 5). At Ponte Vila Formosa, the statistical indicators found for the calibration and validation periods were similar to those obtained at Moinho Novo. The exception was the negative NSE value computed for daily values, which suggests that the average of observed values has a larger predictive power than the model results. Nonetheless, the NSE statistics computed from monthly values are positive and the average monthly flows have a behaviour similar to the measured data ( Figure 5). For N and P, no calibration was carried out due to the data limitation with the default parameters of the SWAT model being considered. The evaluation of model outputs focused on its ability to reproduce the order of magnitude and the peaks of the measured values, which proved that the model is able to offer reasonable results. For the Total N at Ponte de Coruche, the following statistical indicators from the comparison between simulated and measured concentrations were obtained: R 2 = 0.59; Bias = 0.22 mgN L −1 . For the Total P, the following statistical indicators were also obtained: R 2 = 0.14; Bias = −0.067 mgP L −1 .

Water Quantity
After calibration and validation, the SWAT model was used to evaluate the developed storylines. Climate models showed a decrease in precipitation from 400 mm year −1 to an average of 264 mm year −1 on both timelines ( Figure 6). As a result, the simulated scenarios showed a substantial reduction of monthly flows for the 2030 and 2060 timelines (Figure 7). Flow reduction evidenced a non-linear relationship between precipitation and monthly river flow averages. This resulted from the uncertainty associated with climate models, particularly in the forecast of extremes events ( Figure  6). The monthly average flow for the baseline simulation was 42 m 3 s −1 (Table 5), while the IPSL model showed a decrease by half of this value in the timeline 2030 (to about 19 m 3 s −1 ), and a further decrease in the timeline 2060 (to 10 m 3 s −1 ). More severe results were obtained with the GFDL model for the same scenarios, predicting a decrease in the precipitation average to half of the present value, which resulted in monthly flow averages of 4.5 m 3 s −1 and 2.5 m 3 s −1 for the timelines 2030 and 2060, respectively (Table 5). Moreover, the increasing temperature trend predicted by both climate models combined with decreasing precipitation further led to more hazardous agricultural practices for the environment, with irrigation and fertilizer requirements increasing as considered in STL1 and STL3. Even for the scenario with more sustainable agricultural practices (STL 2), the changes in the climate variables had the same severe effect on river flow ( Figure 5). For both climate scenarios, the decrease in irrigation in STL2 resulted in small differences in the monthly flow average when compared with the previous results (Table 5).     The simulated storylines focused on distinct management practices in the watershed. Changes were associated with climate change models IPSL and GFDL developed for this purpose. All simulated scenarios showed a significant decrease in water quantity, clearly visible in the reduction of river flow (Figure 7). This outcome was a direct consequence of the significant decrease in precipitation generally estimated by all climate models, especially for Mediterranean countries, due to the increased anticyclonic circulation that yields increasingly stable conditions, and to a northward shift of the Atlantic storm track [11,51,52]. Similarly, Bucak et al. [53] predicted flow variations between +18 and −59% for the Beyşehir watershed (Turkey), depending on the climate scenario considered. Pascual et al. [54] reported the largest reductions (34%) in mean streamflows (for 2076-2100) to be expected in the headwaters of two humid catchments in Catalonia (Spain), while lesser variations (25% of mean value for 2076-2100) were to be expected in a drier area. Additionally, in all three catchments, the most notable projected decreases in streamflow were observed in autumn (50%) and summer (30%). The largest reductions in the Sorraia river flow were associated with agriculture activities, namely irrigation, which, combined with climatic change, augmented the problem. Therefore, while water availability in several Mediterranean basins is mostly conditioned by precipitation, the results of this study in the Sorraia River basin show that other processes related to agricultural practices also contribute to water scarcity, such as evapotranspiration and irrigation.

Water Quality
In the Sorraia basin, irrigated crops are traditionally sown during mid-April/May and harvested in mid-September/October [55,56]. This leads to an increase of Total N concentration in the Sorraia River during those periods as noticed in all storylines, especially in STL3 (Table 5 and Figure 8), which results from the water reduction in the basin and increased use of fertilizers (STL1 and STL3). Total The simulated storylines focused on distinct management practices in the watershed. Changes were associated with climate change models IPSL and GFDL developed for this purpose. All simulated scenarios showed a significant decrease in water quantity, clearly visible in the reduction of river flow (Figure 7). This outcome was a direct consequence of the significant decrease in precipitation generally estimated by all climate models, especially for Mediterranean countries, due to the increased anticyclonic circulation that yields increasingly stable conditions, and to a northward shift of the Atlantic storm track [11,51,52]. Similarly, Bucak et al. [53] predicted flow variations between +18 and −59% for the Beyşehir watershed (Turkey), depending on the climate scenario considered. Pascual et al. [54] reported the largest reductions (34%) in mean streamflows (for 2076-2100) to be expected in the headwaters of two humid catchments in Catalonia (Spain), while lesser variations (25% of mean value for 2076-2100) were to be expected in a drier area. Additionally, in all three catchments, the most notable projected decreases in streamflow were observed in autumn (50%) and summer (30%). The largest reductions in the Sorraia river flow were associated with agriculture activities, namely irrigation, which, combined with climatic change, augmented the problem. Therefore, while water availability in several Mediterranean basins is mostly conditioned by precipitation, the results of this study in the Sorraia River basin show that other processes related to agricultural practices also contribute to water scarcity, such as evapotranspiration and irrigation.

Water Quality
In the Sorraia basin, irrigated crops are traditionally sown during mid-April/May and harvested in mid-September/October [55,56]. This leads to an increase of Total N concentration in the Sorraia River during those periods as noticed in all storylines, especially in STL3 (Table 5 and Figure 8), which results from the water reduction in the basin and increased use of fertilizers (STL1 and STL3). Total N concentrations were perceptibly higher when using the GFDL climate model due to the lower precipitation amount predicted. An increase of the total N concentration in the river during the crop growth periods was also predicted due to nutrient runoff and leaching. The most marked increase was observed immediately after the harvest season (Figure 8), coinciding with the mineralisation of crop residues and the beginning of the rainy seasons. The increase of soil water content further promoted nitrate leaching, as well as losses by the lateral flow. Therefore, precipitation reduction played a fundamental role in the future projections of catchment dynamics. precipitation amount predicted. An increase of the total N concentration in the river during the crop growth periods was also predicted due to nutrient runoff and leaching. The most marked increase was observed immediately after the harvest season (Figure 8), coinciding with the mineralisation of crop residues and the beginning of the rainy seasons. The increase of soil water content further promoted nitrate leaching, as well as losses by the lateral flow. Therefore, precipitation reduction played a fundamental role in the future projections of catchment dynamics.  (Table 5). For the GFDL scenarios, despite the agriculture practices outlined in STL2, there was a predicted increase of Total N concentration in the Sorraia River up to 2.2 mg N L −1 in the 2030 timeline, and up to 2.6 mg N L −1 in the 2060 timeline. For the STL1 and STL3, an average of 3.6 mg N L −1 was computed for the timeline 2030, and an average of 5.9 mg N L −1 was computed for the timeline 2060. For the IPSL model scenarios, the increase was not so outstanding, especially for STL2, where the monthly averages in the timelines of 2030 and 2060 were predicted to be 0.9 and 1.2 mg N L −1 , respectively. For this model, an average of 1.2 mg N L −1 was observed for the STL1 and STL3 in the timeline 2030, and an average of 1.4 mg N L −1 was noted for the timeline 2060 ( Table 5).
The projected increase of N concentration in the river resulted from the progressive use of this nutrient as fertilizer in most scenarios, but also as a consequence of natural processes occurring in the soils. This outcome is also visible in the more optimistic scenario (STL2), where the negative impact of water flow reductions on water quality was also high. Nitrate was the most abundant form of N simulated in the river (Figure 9), also contributing significantly towards this outcome were nitrate high solubility and leaching susceptibility [57,58], mostly during the periods of higher precipitation, and the type of fertilizers used by farmers [59]. The increase of Total N concentration in all storylines was drastic when compared with the present conditions (0.7 mg N L −1 ) ( Table 5). For the GFDL scenarios, despite the agriculture practices outlined in STL2, there was a predicted increase of Total N concentration in the Sorraia River up to 2.2 mg N L −1 in the 2030 timeline, and up to 2.6 mg N L −1 in the 2060 timeline. For the STL1 and STL3, an average of 3.6 mg N L −1 was computed for the timeline 2030, and an average of 5.9 mg N L −1 was computed for the timeline 2060. For the IPSL model scenarios, the increase was not so outstanding, especially for STL2, where the monthly averages in the timelines of 2030 and 2060 were predicted to be 0.9 and 1.2 mg N L −1 , respectively. For this model, an average of 1.2 mg N L −1 was observed for the STL1 and STL3 in the timeline 2030, and an average of 1.4 mg N L −1 was noted for the timeline 2060 ( Table 5).
The projected increase of N concentration in the river resulted from the progressive use of this nutrient as fertilizer in most scenarios, but also as a consequence of natural processes occurring in the soils. This outcome is also visible in the more optimistic scenario (STL2), where the negative impact of water flow reductions on water quality was also high. Nitrate was the most abundant form of N simulated in the river (Figure 9), also contributing significantly towards this outcome were nitrate high solubility and leaching susceptibility [57,58], mostly during the periods of higher precipitation, and the type of fertilizers used by farmers [59].
The most pronounced increase in the nutrients concentration was found to occur during the autumn/winter period, after the harvesting of corn in irrigated areas (September/October). The high temperature and low soil moisture values predicted during this time of the year enhanced the mineralization of crop residues as observed by the increase of simulated organic N in the river (Figure 9). Nitrite was the least abundant form of N as its oxidation process occurred very fast. The significant increase in N and P may enhance the eutrophication process, thus contributing towards the degradation of water quality [22]. Water 2018, 10, x FOR PEER REVIEW 13 of 18 The most pronounced increase in the nutrients concentration was found to occur during the autumn/winter period, after the harvesting of corn in irrigated areas (September/October). The high temperature and low soil moisture values predicted during this time of the year enhanced the mineralization of crop residues as observed by the increase of simulated organic N in the river ( Figure  9). Nitrite was the least abundant form of N as its oxidation process occurred very fast. The significant increase in N and P may enhance the eutrophication process, thus contributing towards the degradation of water quality [22]. Table 5 and Figure 10 further describe the monthly behaviour of the Total P concentrations in the baseline simulation and future scenarios while considering the different Storylines and climate models. On average, most of the scenarios led to similar monthly concentrations when compared with the present (0.2 mg P L −1 ). In the 2030 timeline, a decrease of the Total P concentration in the Sorraia River to 0.1 mg P L −1 was predicted for all storylines when using the GFDL model, mainly due to the significant decrease in the precipitation amount which reduced runoff and soil erosion, thus decreasing the transport capacity of P in the basin. In the 2060 timeline, an increase of the Total P concentration to 0.3 mg P L −1 in the STL 1 and STL2, and a double-fold increase (0.4 mg P L −1 ) in the STL3 were predicted, in line with the literature [60]. Serpa et al. [60] found a similar decrease of water quality in the Vale do Gaio River (Portugal), located a few kilometres south of the Sorraia River basin, with P concentrations increasing from +29% to +93% depending on the storylines adopted. All climate models and Storylines showed the same concentration averages for the 2060 timeline when comparing with the present monthly average (0.2 mg P L −1 ). However, the major differences observed between the results from GFDL and IPSL climate models may result from extreme precipitation events as shown above (Figure 7).  Table 5 and Figure 10 further describe the monthly behaviour of the Total P concentrations in the baseline simulation and future scenarios while considering the different Storylines and climate models. On average, most of the scenarios led to similar monthly concentrations when compared with the present (0.2 mg P L −1 ). In the 2030 timeline, a decrease of the Total P concentration in the Sorraia River to 0.1 mg P L −1 was predicted for all storylines when using the GFDL model, mainly due to the significant decrease in the precipitation amount which reduced runoff and soil erosion, thus decreasing the transport capacity of P in the basin. In the 2060 timeline, an increase of the Total P concentration to 0.3 mg P L −1 in the STL 1 and STL2, and a double-fold increase (0.4 mg P L −1 ) in the STL3 were predicted, in line with the literature [60]. Serpa et al. [60] found a similar decrease of water quality in the Vale do Gaio River (Portugal), located a few kilometres south of the Sorraia River basin, with P concentrations increasing from +29% to +93% depending on the storylines adopted. All climate models and Storylines showed the same concentration averages for the 2060 timeline when comparing with the present monthly average (0.2 mg P L −1 ). However, the major differences observed between the results from GFDL and IPSL climate models may result from extreme precipitation events as shown above (Figure 7).
Interestingly, the mineral P increase resulting from fertilizer use in corn fields was not obvious in terms of water quality results. This outcome can be explained by the fact that P presents low mobility, adsorbing to sediments and depending on rainfall and runoff events to be removed from agricultural fields. As shown in Figures 10 and 11, P concentrations are expected to exhibit a different behaviour from N in all storylines. However, the fact that P concentrations show a slight increase despite the decrease in precipitation suggests that its continued application will have a significant impact on water quality. Interestingly, the mineral P increase resulting from fertilizer use in corn fields was not obvious in terms of water quality results. This outcome can be explained by the fact that P presents low mobility, adsorbing to sediments and depending on rainfall and runoff events to be removed from agricultural fields. As shown in Figures 10 and 11, P concentrations are expected to exhibit a different behaviour from N in all storylines. However, the fact that P concentrations show a slight increase despite the decrease in precipitation suggests that its continued application will have a significant impact on water quality.  Interestingly, the mineral P increase resulting from fertilizer use in corn fields was not obvious in terms of water quality results. This outcome can be explained by the fact that P presents low mobility, adsorbing to sediments and depending on rainfall and runoff events to be removed from agricultural fields. As shown in Figures 10 and 11, P concentrations are expected to exhibit a different behaviour from N in all storylines. However, the fact that P concentrations show a slight increase despite the decrease in precipitation suggests that its continued application will have a significant impact on water quality. Results have also shown that the decrease of water in the basin had a significant influence on the quality of water in the river. Results for Total N and P concentrations (Figures 8 and 10) further suggest a significant deterioration of water quality in the Sorraia River, particularly with respect to the Total N.

Conclusions and Future Research
The modelling approach developed in this work highlighted possible cumulative impacts of future climatic changes in the Sorraia River basin by considering the expected changes in precipitation, temperature, and human activities. A significant decrease in precipitation is expected over the watershed for the near future (between 25% and 50%). It is likely that this decrease will lead to an increase in irrigation and fertilization needs over this basin by 35%. The water quantity is predicted to fall approximately by up to 75%, while water quality shows an unbalanced deterioration, with nutrient concentrations predicted to increase up to 200% for P and up to 500% for N. Such a rise in nutrient concentration is observed to be a consequence of the increased use of fertilizers and decrease in water availability in rivers.
The results show how societal and especially climatic changes can affect river water quantity and quality in the study basin and can be considered as a starting point for defining appropriate management plans to counteract such negative impacts. The SWAT model can be further explored to test the effects of management practices and the degree of stress they introduce on the environment, thereby contributing to a cost-effective adaptive management practice. For instance, the use of winter crops with minimum water requirements, the implementation of no-till practices to improve soil structure and soil infiltration rates, the use of cover crops for reducing soil water evaporation or the establishment of more resilient cropping systems to cope with water scarcity may allow future impacts to be minimised at the basin level.