Predicting reservoir inflows with an advanced SWAT+ model calibration in the Tagus River headwaters (Spain)

The Tagus River is the longest in the Iberian Peninsula and its basin is highly managed: it is the most populated and subject to the Tagus-Segura water transfer, which sends an annual average of 330 hm³ to southeast Spain. Its starting point is a hyper-reservoirs' system (Entrepeñas-Buendía-Bolarque) located in the basin´s headwaters sector. The total inflow to this reservoir system has decreased by 50% in the last decades, mostly as a consequence of the already noticeable impacts of climate change, and this situation will be further aggravated in the future. Thus, both gaining knowledge about the hydrological behaviour of the Tagus River headwaters and developing reliable tools to predict inflows to this reservoirs' system are highly relevant tasks to aid for a sustainable water resources management in coming years.

In this work, we set up a highly detailed catchment model with SWAT+ in the Tagus River headwaters. Before calibration, two additional tasks were addressed: 1) HRUs were grouped into three classes with varying lithology and permeability (carbonate, detrital of high and medium permeability, detrital of low permeability), and 2) two hydrological indices, the runoff rate and the baseflow index, were obtained for eight subbasins that have streamflow records. Three sets of parameters were designed, one for each HRU geological class, and then a complex calibration procedure was addressed. A soft calibration, narrowing parameters´ ranges to reproduce the hydrological indices realistically, was followed by a multi-site hard calibration of the streamflow in eight subbasins. During hard calibration, the streamflow simulation performance and the accuracy of the model reproducing the runoff coefficient and the groundwater contribution were considered. Afterwards, a best simulation was chosen and tested with an initial validation of the daily streamflow produced in each reservoir watershed, obtaining both statistically and graphically satisfactory results. After some final modifications in the model, a second and final validation on the monthly inflows into the hyper-reservoirs system was done, successfully reproducing the observed records, with NSE, R² and PBIAS values of 0.86, 0.88, 2.5% in Entrepeñas and 0.89, 0.91 and -8.5% in Buendía.

The SWAT+ calibration approach followed in this work is novel because it takes into account the heterogeneity in the hydrogeological properties of a catchment to parameterize it, optimizing the parameters values at a geological class level and evaluating the model performance at subbasin level, which implies a higher spatial calibration resolution that the usual one in SWAT studies. As a result, the model not only simulates accurately subbasins with uniform geological properties,  but the entire Tagus headwaters,  including those subbasins  with  mixed  geology, thus resulting  in a model that accurately simulates reservoir inflows. By performing a multi-site calibration on smaller subbasins, results are more accurate, and the model represents more realistically local hydrological conditions. For that reason, this methodology helps also to understand how specific environmental conditions might affect all hydrological model process, thus also helping in water management decision making.