A Novel Technique for Rain/ No-Rain Classification Using Outgoing Longwave Radiation Data from INSAT-3D

This study presents a novel approach for rainfall detection over the Indian subcontinent using INSAT-3D Outgoing Longwave Radiation (OLR) data and Random Forest classification. We employed the Random Forest classifier, experimenting with various hyperparameters to optimize performance. We utilized data from June 2021, covering the region $0 \mathrm{~N}-40 \mathrm{~N}$ and $60 \mathrm{E}-100 \mathrm{E}$, with Global Satellite Mapping of Precipitation (GSMaP) as reference. Our model, trained on 80% of the data and validated through five-fold cross-validation, outperforms operational methods in Critical Success Index and Equitable Threat Score. Notably, our approach demonstrated superior performance compared to two existing models, the Hydro-Estimator method (HE) and the INSAT Multispectral Rainfall Algorithm (IMSRA). Our proposed method achieved a Critical Success Index(CSI) value of 0.41 and Equitable Threat Score(ETS) value of 0.39 which represent an improvement over the corresponding values for $\mathrm{HE}(0.25$ and 0.23) and IMSRA (0.31 and 0.30).