Abstract Estimating water yield in ungauged catchments is often essential for water resources planning, infrastructure design, and the development of climate-resilient projects. In many developing countries, including India, sparse hydrological networks limit the direct application of process-based models. This study develops a comparative regionalization framework combining three approaches: (i) the GR2M conceptual rainfall–runoff model with parameter transfer through inverse distance weighting (IDW), (ii) regression-based rainfall–runoff relationships for monsoon months, and (iii) machine learning models for scalable yield prediction. In the analysis, climatic and hydrological data of 33 gauging sites across the Betwa, Chambal, Ken, Sindh, and Tons River Basins of the Ganga River System were used for calibration, validation, and cross-basin transfer testing.The GR2M–IDW method achieved transfer efficiencies of up to 93.4% in the Chambal basin. Still, it declined to 58% in the Sindh basin, demonstrating strengths in hydrologically homogeneous regions and limitations in heterogeneous terrains. Regression models provided coefficients of determination ranging from 0.59 in June to 0.82 in October, ensuring transparent and regulator-friendly estimates for statutory project clearance. Machine learning algorithms, particularly Gradient Boosting and Huber Regressors, consistently outperformed others, achieving accuracies above 88% and offering robustness in diverse catchments. The findings suggest a tiered regionalization framework: GR2M–IDW ensures process-based parameter transfer, regression supports regulatory requirements, and machine learning enhances predictive accuracy and scalability. Together, these approaches provide a technically sound pathway for water yield estimation in ungauged basins, advancing climate-resilient water resource planning.
Jaiswal et al. (Mon,) studied this question.