Streamflow forecasting in Tocantins river basins using machine learning

Understanding the behavior of the river regime in watersheds is fundamental for water resources planning and management. Empirical hydrological models are powerful tools for this purpose, with the selection of input variables as one of the main steps of the modeling. Therefore, the objectives of this study were to select the best input variables using the genetic, recursive feature elimination, and vsurf algorithms, and to evaluate the performance of the random forest, artificial neural networks, support vector regression, and M5 model tree models in forecasting daily streamflow in Sono (SRB), Manuel Alves da Natividade (MRB), and Palma (PRB) River basins. Based on several performance indexes, the best model in all basins was the M5 model tree, which showed the best performances in SRB and PRB using the variables selected by the recursive feature elimination algorithm. The good performance of the evaluated models allows them to be used to assist different demands faced by the water resources management in the studied river basins, especially the M5 model tree model using streamflow lags, average rainfall, and evapotranspiration as inputs. HIGHLIGHTS The Recursive Feature Elimination was the best input feature selection algorithm.; The machine learning models were efficient in the daily streamflow forecasting.; The performance of the M5 model tree was better than the other models.; The models are potential tools to assist water resources management.;