Assessment of Mean Daily Streamflow in Catchments with Hydrometeorological Records Using Machine Learning and Deep Learning Approaches

The application of artificial intelligence (AI) to rainfall–runoff modelling in catchments has attracted increasing attention in recent years because of its ability to represent complex hydrological relationships with relatively low computational requirements. This study presents a surrogate modelling framework for estimating mean daily streamflow in gauged catchments using machine learning (ML) and deep learning (DL) approaches. The methodology was applied to the Mocoa River basin in Putumayo, Colombia. Several ML algorithms were trained, validated, and tested to assess their predictive performance. In addition, DL models were implemented and evaluated for comparison with the ML approaches. The results indicated that the DL models provided better predictive performance than the ML models considered in this study. In particular, the Gated Recurrent Unit (GRU) model achieved the lowest RMSE, with a value of 18.36 m3/s, whereas the best-performing ML model, the Interactions Linear Model, achieved an RMSE of 40.00 m3/s. The models were developed using daily rainfall records from two rain-gauge stations and mean daily streamflow records from one stream-gauging station. The proposed methodology provides a computationally efficient approach for streamflow estimation in catchments with available hydrometeorological records. The surrogate modelling framework may support applications in hydrological assessment, water-resource management, and flood-risk analysis by providing timely estimates of mean daily streamflow. It may also offer a practical modelling approach for environmental agencies, water-resource managers, and other stakeholders requiring streamflow information to support hydrological decision-making.