What question did this study set out to answer?

The aim is to develop a deep learning framework for predicting high-intensity rainfall using satellite data.

March 16, 2026Open Access

HIR Nowcasting using ConvLSTM

Key Points

The aim is to develop a deep learning framework for predicting high-intensity rainfall using satellite data.
Employs ConvLSTM2D models for rainfall nowcasting
Utilizes satellite data from GEO-KOMPSAT-2A and GPM IMERG
Focuses on brightness temperature, cloud analysis, and rain rate products
Achieves critical success indices up to 50%
Records F1 scores exceeding 70%
Demonstrates potential for reliable HIR prediction in radar-scarce regions

Abstract

This repository presents a deep learning-based framework for high-intensity rainfall (HIR) nowcasting, leveraging geostationary satellite data from GEO-KOMPSAT-2A (GK2A) and Global Precipitation Measurement (GPM) IMERG. Three Convolutional Long Short-Term Memory 2D (ConvLSTM2D) models are developed using brightness temperature (BT), cloud analysis, and rain rate products. These models achieve critical success indices (CSI) up to 50% and F1 scores above 70%, demonstrating the potential for HIR prediction globally, especially in regions without ground-based radar coverage.

HIR Nowcasting using ConvLSTM

Key Points

Abstract

Cite This Study