Exploring Deep Learning Techniques To Nowcast Satellite Precipitation Retrievals

Heavy rainfall events in urban areas pose significant challenges, leading to flooding, traffic congestion, and property damage. Accurate and timely precipitation nowcasting is crucial for mitigating risks associated with heavy rainfall. This research proposes a deep learning-based approach utilizing the U-Net architecture to forecast precipitation for the next half hour, enabling timely warnings to the community. The proposed nowcasting model utilizes the Global Precipitation Measurement (GPM) IMERG Final Precipitation L3 Half Hourly dataset covering the North Indian region from 1st July 2023 to 15th September 2023 as the input. The proposed U-Net model achieved a Root Mean Squared Error (RMSE) of $0.8 \mathrm{~mm} / \mathrm{hr}$ and a Pearson Correlation Coefficient (PCC) of 0.78, demonstrating its ability to accurately predict precipitation patterns within the next half hour.