Author(s): Federico Vilaseca; Christian Chreties; Alberto Castro; Angela Gorgoglione
Linked Author(s): Christian Chreties
Keywords: No Keywords
Abstract: This paper proposes a methodology based on data augmentation to improve the performance of data-driven rainfall-runoff models on high flows. Problems in the representation of such range of discharges by data-driven models were presented in previous research, which the authors of this work attribute to class imbalance of discharge data, where high flows are underrepresented. This ends up biasing the learning process towards the representation of low flows. The proposed methodology was tested for two incremental watersheds of the Santa Lucia Chico basin in Uruguay, showing an increase in performance of 17 % for Nash-Sutcliffe efficiency and 38 % for peak-flow Nash-Sutcliffe efficiency. Results demonstrate that class imbalance is a relevant issue affecting the performance of data-driven rainfall-runoff models and that the proposed methodology allows to tackle said issue and improve model performance for high flows.
Year: 2024