Multisite and multivariate calibration of the SWAT+ model in Galicia-Costa, NW Spain

Calibration and validation are fundamental steps in hydrological modeling, ensuring the accuracy and reliability of model predictions for effective management of freshwater resources. Traditional calibration approaches rely solely on streamflow, although other hydrological variables (soil moisture, evapotranspiration) have also proved useful. One of the most widely used hydrological models is the Soil and Water Assessment Tool (SWAT), which simulates spatial and temporal variations in watershed processes such as the water balance, streamflow routing, and the transport of nutrients and sediments. In this study, SWAT+, a revised version of the SWAT model, was applied to the Ulla River basin within the Galicia-Costa Hydrographic Demarcation in Spain to evaluate the performance of three calibration strategies: (1) single-variable calibration using streamflow (SC-Q), (2) single-variable calibration using evapotranspiration (SC-ET), and (3) multivariate calibration, integrating both streamflow and evapotranspiration (MC-QET). Multi-site calibration and validation were performed using the Sequential Uncertainty Fitting Algorithm (SUFI-2), with the Nash-Sutcliffe efficiency (NSE) index as the objective function and NSE ≥ 0.60 defined as the behavioral threshold. Observed streamflow data was obtained from three river gauging stations distributed along the river network (one downstream and two upstream). Ground-truth evapotranspiration (ET) data were estimated via triple collocation analysis combining three independent datasets (remote sensing-based, land surface model output, and reanalysis product). Results revealed that for streamflow, the MC-QET calibration scheme yielded the best performance at the downstream validation site (NSE = 0.82, PBIAS = -4.51), whereas SC-Q achieved superior results at the upstream stations (NSE = 0.82-0.86 and PBIAS = +6.35 – +12.72). Meanwhile, SC-ET performed the worst for streamflow overall, although model performance was still acceptable (NSE = 0.70 – 0.75). For evapotranspiration, both SC-ET (NSE = 0.89, PBIAS = +6.56) and MC-QET (NSE = 0.90, PBIAS = +6.24) clearly outperformed SC-Q (NSE = 0.66, PBIAS = +22.97). These findings suggest that while streamflow- or ET-only calibration can optimize the targeted variable, incorporating multiple hydrological variables during model calibration improves the overall representation of watershed processes and the water balance. However, the acceptable performance of the ET-only calibration highlights that this calibration scheme can till serve as a valid alternative in data-scarce regions where streamflow observations are limited or inconsistent. Furthermore, this study demonstrates the reliability of triple collocation analysis in improving ET estimates by reducing uncertainty among independent data sources. In conclusion, integrating multivariate calibration strategies in SWAT+ significantly enhances spatial transferability, ensures physically realistic model outputs, and improves overall prediction reliability. At the same time, ET-based calibration alone remains a practical and defensible option for data-limited watersheds, demonstrating the growing potential of remote sensing–driven hydrological modeling for comprehensive and resilient water resource assessment.