Comparison of a Semi-Distributed Empirical Model and a Distributed Physical Model in a Snow-Covered Mediterranean Catchment Under Climate Change Scenarios

The catchments of the Quéntar and Canales reservoirs are two adjoining valleys on the north-western side of the Sierra Nevada in Spain. Canales drains the northern slope of the river Genil, with 15 linear km of peaks above 3000 meters, culminating in the highest in the Iberian Peninsula, Mulhacen, at 3479 meters. With 83 km2 above 2000 m, this river exhibits a clear nival hydrological regime. In contrast, the Quéntar basin, which collects water from the Padules and Aguas Blancas rivers, drains a smaller area with a maximum altitude of 2336 m and only 7 km2 above 2000 m. Its regime is pluvio-nival, with a much more marginal influence of snow.

To understand and predict the hydrological behaviour of these catchments under climate change scenarios, we have calibrated two different hydrological models. These models will provide the predictive tools needed to calculate river temperature and substances, particularly nitrogen (N) and phosphorus (P). The first model, SWAT (Soil and Water Assessment Tool), is a well-known conceptual semi-distributed parametric model based on linear reservoir equations that simulates snow using a modified degree-day model. The second model, NIVAL, is a distributed model based on physical processes, featuring a specific snow module that relies on mass and energy balance, specifically designed for use in the Sierra Nevada.

The two models differ significantly in terms of preparation, calibration and performance. SWAT's advantages are those of any distributed model: fast computation, easy calibration (facilitated by automatic algorithms) and a reduced need for input data. These features make SWAT a practical choice for many applications. On the other hand, NIVAL offers a more detailed representation of the hydrological processes and greater robustness to changes in scenarios outside the calibration range. This makes NIVAL particularly valuable for studying individual processes and hypothetical future scenarios.

It was expected that the flow adjustment in SWAT would be less accurate than in NIVAL, especially in the Canales basin due to the significant snow influence. However, the calibration and validation of both models on daily flows for both basins yielded very similar results in the most common statistics. For instance, the Nash-Sutcliffe Efficiency (NSE) values were around 0.63/0.70, the Kling-Gupta Efficiency (KGE) was 0.70/0.74, and the Percent Bias (PBIAS) was 2.49/19.08 for the Canales and Quéntar cases. These results demonstrate that SWAT is a reliable option for calculating total flows in historical scenarios.

Nevertheless, NIVAL's detailed process representation makes it more reliable for studying individual processes or hypothetical future scenarios. The next step in this research is to compare these models against various climate change scenarios to assess the differences in their predictions. This will help us understand the strengths and limitations of each model and improve our ability to predict and manage water resources in snow-covered Mediterranean catchments under changing climate conditions.

Aknowledments: This research has been supported by Grant TED2021-130744B-C22 funded by MICIU/AEI /10.13039/501100011033 and by the European Union Next GenerationEU/ PRTR