Enhancing IMERG precipitation estimates using machine learning and bias correction across elevation zones in India

Accurate precipitation estimation is crucial for hydrological and climate applications, yet satellite datasets like IMERG often suffer from misclassification and bias. This study improves IMERG estimates over India by integrating ground-based observations with a two-step approach: (1) precipitation/non-precipitation (P/NP) machine learning classification models, and (2) category-specific bias correction. To address spatial variability, India was stratified into four elevation zones (>200, 200–600, 600–1200, <1200 m) and analysed across four seasons. Raw IMERG showed high false precipitation rates: 45%, 53%, 46%, and 67% decreased to 20%, 21%, 25%, and 42% after P/NP classification. Bias correction further lowered RMSE from 15.95–17.01 mm to 7.11–8.86 mm, with the greatest gains during monsoon and summer. The critical success index rose from <0.4 to >0.8, especially in winter and post-monsoon when detecting light rainfall. These refinements enhance IMERG’s suitability for hydrological modelling, extreme weather forecasting, and climate impact assessments.