Author(s): Sergio Ricardo Lopez-Chacon; Fernando Salazar; Ernest Blade
Linked Author(s): Ernest Bladé I Castellet, Fernando Salazar González
Keywords: No Keywords
Abstract: Fully representing the complexity of the rainfall-runoff processes occurring in a catchment for a variety of precipitation events is a challenging task for the physically based hydrological models: the calibration procedure is usually difficult and it might not accurately capture the streamflow response of the catchment for unseen scenarios. In order to achieve a model with enhanced capabilities, this study proposes a methodology to create a hybrid model combining a surrogate model based on the physical model Iber and a machine learning model to compensate for the possible residuals. Since the proposed methodology is designed to be used in early warning system applications, the evaluation is mainly focused on the high streamflow values. The results show that the hybrid model provides higher accuracy than the physically based model and closer approximations to the peak values than a single machine learning model. This results in an increased utility for the intended application. Moreover, it might represent an alternative to the calibration procedure for physically based models.
Year: 2024