Sustainable Modeling Framework for the Prediction in Ungauged Catchments: Leveraging Large-Sample Hydrology Datasets with Machine Learning

Predicting hydrological behavior in ungauged basins presents a major challenge for sustainable water resource management in regions with limited data, such as the Indian subcontinent. Accurate prediction of water availability typically requires calibrating hydrological models using historical discharge data, which is often unavailable for many Indian catchments. In this context, comparative hydrology remains a valuable approach, as comparing diverse catchments has proven effective in understanding and forecasting hydrological processes, particularly in ungauged regions.This study focuses on the critical role of catchment characteristics in improving the prediction of hydrological responses in ungauged basins across 4 countries. By employing large-sample hydrology methods, we analyze comprehensive datasets from gauged basins worldwide and compare them to Indian ungauged basins to identify patterns that can be applied for predictive purposes. For this analysis, we use the CAMELS (Catchment Attributes and Meteorology for Large-sample Studies) datasets from the USA, Australia, and Chile for physio-climatic data. For India, we compile data from 216 catchments, categorize it into three groups based on attributes, and evaluate clustering using Normalized Mutual Information. We also correlate catchment features-such as soil properties and vegetation cover-with hydrological responses.Our findings demonstrate how attribute-based clustering of catchments can be leveraged to identify similar catchments over large spatial scales. Ungauged basins that share significant similarities with gauged basins can benefit from the extrapolation of data, enabling more accurate predictions. Additionally, incorporating climatic indices with catchment attributes enhances the accuracy of hydrological forecasts under different climate conditions.