HIR Nowcasting using ConvLSTM

This repository presents a deep learning–based framework for short-term prediction of High-Intensity Rainfall (HIR) using GEO-KOMPSAT-2A (GK2A) and Global Precipitation Measurement (GPM) IMERG satellite data. Three ConvLSTM2D (Convolutional Long Short-Term Memory 2D) models were developed utilizing brightness temperature (BT), cloud analysis, and precipitation data. The models achieve up to 50% Critical Success Index (CSI) and over 70% F1 score, demonstrating strong potential for global HIR prediction, particularly in regions where ground-based radar observations are unavailable.