Seasonal streamflow modelling in brazilian semiarid using artificial intelligence and gridded climate data

Abstract

The integration of gridded climate data into hydrological modelling has become an effective alternative in regions with a sparse network of hydrometeorological monitoring stations, as is the case in much of the Brazilian semi-arid region. This approach can enhance water resource management, providing greater security in planning for multiple water uses. In this context, data-driven models (DDMs) have gained prominence for bypassing the structural complexity of physically based models, relying instead on numerical algorithms to capture patterns from historical data. However, selecting appropriate input variables remains a challenge, particularly due to the spatial and temporal heterogeneity of hydrological processes. This study aims to evaluate the application of different DDM approaches, such as multiple linear regression (MLR), artificial neural networks (ANN), and the k-nearest neighbours algorithm (KNN), for modelling the rainfall-runoff relationship in the Vaza-Barris River watershed, using spatial climate data as explanatory variables. The results indicate that, despite limitations in accurately simulating peak flows, the models were able to satisfactorily represent minimum flows and the watershed’s hydrological seasonality. Overall performance was classified as only moderate, based on the Nash-Sutcliffe Efficiency (NSE) coefficient, with values varying across the tested algorithms. These findings highlight the need for improvements in variable selection and model parameterization.