Streamflow estimation in the Indus River Basin using a water balance framework and artificial neural networks with satellite- and model-derived global hydro-climatic data sources

The Indus River Basin (IRB) and the Upper Indus Basin (UIB). Uncertainties in key water balance components, such as precipitation (P) and evapotranspiration (E), present a critical challenge for streamflow estimation, particularly in large, data-scarce basins like the IRB, where significant spatiotemporal climate variability exists. To address these uncertainties and enhance streamflow estimation, we applied water balance-based methods. First, the water balance framework was employed to select the best-performing datasets for P and E from various sources, including satellite observations, reanalysis products, and combined satellite-gauge data. These selected datasets, together with terrestrial water storage (TWS) from the Gravity Recovery and Climate Experiment (GRACE), were then used in separate applications of the water balance equation and artificial neural networks (ANN) for streamflow estimation. Our findings indicate that MERRA2 reanalysis products for P and E provided the least closure error for the IRB, while ERA5 P performed better for the UIB. The streamflow estimation using the ANN models demonstrated high accuracy for both basins. The water balance equation also showed good performance in streamflow prediction for the UIB, suggesting that this method could be effectively used without the need for model training or auxiliary datasets. Our study demonstrates the potential and limitations of water balance concept to improve streamflow estimation in large, data-scarce river basins. • Water balance closure is applied to select the best P and E datasets for streamflow. • Bias-corrected water balance components with auxiliary variables improved streamflow prediction by ANN models. • A water balance equation with bias-corrected components reliably estimates streamflow in UIB. • Inclusion of snowmelt and temperature as auxiliary variables in the ANN model significantly improved streamflow prediction. • This framework enables accurate streamflow prediction in data-scarce river basins.