Precipitation Estimation from Himawari-8 Multiple Spectral Channels Using U-Net

Precipitation profoundly impacts Earth's ecosystems, from soil and plants to animals and humans. Accurate rainfall estimation is vital for various purposes, including weather and flood forecasting, enhancing agriculture, ensuring traffic safety, and minimizing natural disaster damage while safeguarding lives. Currently, rainfall estimation methods are mainly based on three main sources of information: satellite images, radar images and rain gauges data. However, these methods have limitations such as spatial coverage, limited range, capturing precipitation dynamics in complex terrainand, differentiating between precipitation types and accurately estimating intensity. To address these challenges, recent research explores deep learning-based methods that combine these data sources for accurate and comprehensive precipitation estimatation. This paper outlines directions for developing such methods using the U-Net deep learning model, incorporating data from Himawari-8 satellite imagery, rain radar data, meteorological data, and topographic information. This approach yields improved results in predicting rainfall rates and generating radar maps.