Future scenarios for water resources in the Mediterranean region foresee changes in reservoir inputs, including lower available discharges from dams to meet the water demand from irrigated and urban areas [1
]. Water ecosystems are vulnerable to various forms of pressure caused by human activities and they are particularly vulnerable to climate change, all of which may affect water resources and demand. Therefore, hydrological studies are necessary to implement basin management plans to protect water resources. Contemporary catchment models have become very useful tools and the Soil and Water Assessment Tool (SWAT) model is one of the most widely used of the many catchment models available (e.g., [2
]). Since it has been continuously improved over the last 30 years, it has gained international acceptance as an interdisciplinary tool suitable for application at different scales of catchments with varying degrees of biophysical and climatic settings (e.g., [3
]). However, despite the increasing use of the model, to date the majority of SWAT applications have been carried out in the U.S. and other European regions, which are generally considered to be data-rich compared to other countries (e.g., snow, soil or/and crop data are not available in many countries).
Since spatial and temporal climate variability are key characteristics in hydrologic modelling [4
], distributed hydrological models like SWAT require a minimum of daily distributed meteorological data to adequately simulate the hydrological cycle. The drawback is that long-term continuous data are not available in many parts of the world. Given the fact that climate data is a major driver of hydrological and other processes simulated by the model, studies evaluating the impact of climatic input data on SWAT simulations are gaining increased attention. Gassman et al. [3
] indicate that poor results can be attributed to inadequate representation of rainfall inputs, either due to a lack of adequate rain gauges in the simulated catchment or to subcatchment configurations that are too coarse to capture the spatial detail of rainfall inputs [5
]. The impact of rain gauge density, the distribution of stations in the catchment, and the data sources for generating climate inputs have also been studied [6
Modelling hydrology is a difficult task to perform accurately in terms of time, space and volumes due to its great complexity and the many factors involved [8
]. In mountainous terrains, complex topography introduces diverse snowmelt patterns (drifting and blowing) and large elevation gradients can produce complicated precipitation distributions which require extrapolation of temperature and precipitation data. Those who work with radar or satellite precipitation input data sources find that input of areal rainfall measured by radar performs better than other rainfall data sources. Yu et al. [9
] explored different approaches to precipitation input in an alpine area of large topographic relief and concluded that radar, combined with rain gauge data, performed better than gauge-only data. Consequently, these studies have highlighted the need for additional research to improve precipitation input to SWAT simulations. Fontaine et al. [10
] first applied SWAT to mountain environments by adding elevation bands and improving the snowmelt routine, which significantly enhanced the ability to simulate volumes of water stored in the snowpack and its subsequent release during spring and summer snowmelt. Since then, SWAT has been implemented to estimate river discharge volumes and timing from mountainous catchments in several countries (e.g., [5
]); however, it has not been evaluated in the steep extreme elevation-climate gradient conditions of a mountain area where climate input data are scarce.
Mountains constitute the main source of water for many of the world’s rivers [14
], particularly in Mediterranean sub-humid and semi-arid environments as the lowland region of the Ebro Basin. Therefore, many Pyrenean rivers have been dammed to provide irrigation water to the lowland areas. The rugged topography, the regime of the rivers with frequent floods, and the changes in land use over the past few decades have unfortunately triggered soil erosion and, consequently, the siltation of these reservoirs [15
]. The strong dependence of mountain water resources on climate fluctuations and land-cover characteristics means that environmental change directly affects the total amount and temporal distribution of stream flow, and hence has major implications for water management to ensure water supplies [1
]. Environmental concerns about these fragile systems illustrate the need to have detailed understanding of their hydrological processes. Since it is difficult to obtain continuous direct measurements with sufficient spatial coverage in mountain regions, a robust computational hydrologic model like SWAT is an effective means of studying land-surface dynamics [4
]. Thus, SWAT is considered to be an appropriate tool to be used in the regional assessment of catchments within the Pyrenean region.
A snow-dominated mountain catchment such as the large alpine-prealpine drainage area of the Barasona reservoir (southern Pyrenees) presents a considerable challenge for spatially distributed modelling due to its highly heterogeneous climatic drivers with elevation-climate gradients and complex topography. Its elevation gradients and inaccessibility contribute to poor data resolution. Findings from a previous calibration of the catchment [19
] put into question the representativeness of the available climatic configuration, the scarcity of weather data in the studied area and its distributions in the valley bottoms endorse this issue. The hypothesis is that generated synthetic climate data, which could better define the climate for the alpine headwater together with the parameterization of elevation bands for the orographic effects, should enhance its climatic characterization and its hydrological simulation.
This research examines the effects of four different precipitation characterization scenarios. The main objective is to enhance the simulated discharges of the catchment in SWAT evaluating different strategies of spatial precipitation characterization for the highly variable climatic gradients of the Barasona catchment to better simulate the hydrological processes. The findings from this study provide insight into the relevance of taking into account the spatial distribution and source of weather data when using SWAT to estimate river discharges in mountainous catchments.
3. Results and Discussion
The SWAT model response for the Barasona catchment, using four different configurations for climate characterization, was compared by means of statistical measures of the simulated river discharges with that observed and the visual interpretation of the simulated hydrographs (Table 4
and Figure 3
). The model performed predictions of subcatchment river discharges, with a variety of NSE, Dv and RMSE values for the different scenarios, gauge stations and periods (Table 4
Differences between outputs for scenarios A1, B1, A2 and B2 have been found for each subcatchment and for the calibration and validation periods. Scenario A1 corresponded to the result of the calibration process and could be considered as the initial conditions in which the assessed climatic strategies were added. This scenario performed a general underestimation of the river discharges, particularly for the snowmelt period, for all of the gauge stations and periods, yielding statistical measures that were poor for the Linsoles and Graus gauge stations, and acceptable for Capella. As the inclusion of the synthetic station only affected the simulated climatic conditions of the Ésera headwater, the Isábena River subcatchment (Capella gauge) was not influenced by its introduction and its simulated discharge was only affected by the parameterization of the elevation bands for the estimated climate altitudinal gradient. Therefore, the statistical measures for Capella gauge station were equal between A1 and B1 and between A2 and B2. The scenarios which yielded the best statistical measures for the Capella gauge station were A2 and B2, both with the parameterization of the elevation bands. Graus and Linsoles gauge stations for the Ésera River yielded the greatest differences between subcatchments, scenarios and periods. The Ésera headwater at Linsoles gauge station showed the greatest differences because introducing the synthetic station directly affected its subcatchment climate characterization and simulated runoff (scenarios B1 and B2). This synthetic station developed high flows, which compared to scenario A1 improved runoff for Graus and Linsoles gauge stations in B1, but not after adding the elevation bands for scenario B2. Introducing the elevation bands increased the simulated available water and snowfall retention in the catchment, improving the simulated river discharges, and the greatest values of NSE and RMSE for all gauge stations and periods being those of scenario A2. Scenario B2 for Graus and Linsoles gauge station with the synthetic station and the parameterization of the elevation bands resulted in an excess of available water which worsened discharge simulation, especially for Linsoles gauge station. Apart from the high values of Dv observed for Linsoles and Graus gauge stations for the validation period, the statistical measures of the simulated river discharges for all the scenarios, gauge stations and periods indicated that scenario A2 is the strategy that best parameterizes the highly variable climate of the Barasona catchment.
Other authors have reported similar calibration statistics for SWAT model performance in mountain catchments. For example, Fontaine et al. [10
] described a monthly NSE of 0.86 with a Dv of −9.8%, Zhang et al. [11
] reported a maximum monthly NSE value of 0.85 and Flynn and Van Liew [12
] noted a monthly NSE of 0.86.
The dammed characteristics of the Ésera River influence simulation results and detailed daily outflow in Paso Nuevo and Linsoles reservoirs might contribute to a significant improvement in the simulation of Linsoles and Graus gauge stations. The inexistence of outflow data for Paso Nuevo reservoir and the limited information on Linsoles reservoir suggests that the most realistic approach to simulating both reservoirs is their parameterization via the SWAT target volume option. Discordances between simulated and observed high flows of snowmelt river discharge for the Ésera River (Linsoles and Graus gauge stations) could be due to this (annual/monthly homogenized) option of simulation for the reservoirs. For the Ésera headwater, the retention characteristics of the reservoirs were insufficient to accumulate the excess of available water generated in scenarios B1 and B2 (with the synthetic station
) to delay the simulated snowmelt high flows (Figure 3
The simulated karst system performed differently with average monthly discharges from 0.61 to 1.64 m3
according to the simulated available water of each scenario for the headwater. Scenario A1 predicted the lowest and B2 the largest amount of discharge with 13%–21% of discharge to the Garonne River for the Linsoles gauge station. Results are in the range of the observations for the Jueu Karst System [22
A detailed subcatchment subdivision and parameterization of the elevation bands is recommended to perform a detailed headwater simulation, given that altitudinal climate gradients in SWAT are defined at subcatchment level. Due to the rugged topography of the headwater the estimated optimal horizontal limit of 2000 m a.s.l., at which the estimated precipitation altitudinal gradient decreases to almost half, would seem to be difficult to achieve. Hence, for areas with complex topography and variable climatic elevation gradients, the elevation band level should have better parameterized the temperature and precipitation altitudinal gradients.
Scarce and poorly distributed climatic input data and highly variable climate in mountain regions constitute the main challenges to performing good characterization of the spatial patterns of climatic variables in hydrological models such as SWAT. It was necessary to establish and evaluate strategies for generating synthetic climate data and reproducing the climatic altitudinal gradients in order to overcome these challenges and to better characterize the climatic variables and the hydrological simulations of the Barasona catchment. In this study, the simulated four scenarios with different climatic inputs served to evaluate the strategies by river discharge performance and to identify that the parameterization of elevation bands with the estimated altitudinal gradients and without generated climatic data was the best scenario (A2). Variability in snow accumulation and melt processes and the impact on water balance emphasize the important role played by the parameterization of altitudinal climatic gradients in elevation band strategies in catchments with low-density precipitation and temperature datasets and strong altitudinal climatic gradients. In addition, simulated discharges allowed the influence of the headwater reservoirs and karst system to be evaluated. Other weather data sources, such as radar or satellite, would improve climatic characterization since these data sources could provide better definitions of the spatial climatic variability.
The increase in the water demand for agriculture and the predicted changes in hydrological trends facing water shortages, especially in the Mediterranean region, strengthen the need for assessment, prediction and planning of water resources from mountain headwater catchments that are the suppliers to the agricultural lowlands. These scenarios reinforce the necessity for modelling the availability of water to anticipate water storage in view of drought periods and discharge changes. Therefore, as the Spanish Pyrenean region is a main water source and holds several reservoirs that store water for irrigation of the drier agricultural lowlands, environmental management plans at a regional scale using models such as SWAT afford a perspective that cannot be obtained with other tools and serve as a chief instrument for assisting in the regulation of headwaters such as the Barasona catchment. The success of SWAT evaluation in computing the hydrologic component in this mountainous catchment, as illustrated in this study, provides an opportunity to extend the use of the model to other catchments in the Pyrenean region.