PAUNet: Precipitation Attention-based U-Net for rain prediction from satellite radiance data

TL;DR

PAUNet is a novel deep learning model combining U-Net and Res-Net architectures with attention mechanisms, designed to improve precipitation prediction accuracy from satellite radiance data, especially for medium and heavy rain.

Contribution

This paper introduces PAUNet, a new deep learning architecture that effectively captures large-scale contextual information for satellite-based precipitation forecasting, outperforming baseline models.

Findings

PAUNet achieves higher CSI scores than baseline models.

The use of exponential Focal Precipitation Loss improves rain category importance.

PAUNet demonstrates robust performance across European regions.

Abstract

This paper introduces Precipitation Attention-based U-Net (PAUNet), a deep learning architecture for predicting precipitation from satellite radiance data, addressing the challenges of the Weather4cast 2023 competition. PAUNet is a variant of U-Net and Res-Net, designed to effectively capture the large-scale contextual information of multi-band satellite images in visible, water vapor, and infrared bands through encoder convolutional layers with center cropping and attention mechanisms. We built upon the Focal Precipitation Loss including an exponential component (e-FPL), which further enhanced the importance across different precipitation categories, particularly medium and heavy rain. Trained on a substantial dataset from various European regions, PAUNet demonstrates notable accuracy with a higher Critical Success Index (CSI) score than the baseline model in predicting rainfall over multiple time slots. PAUNet's architecture and training methodology showcase improvements in precipitation forecasting, crucial for sectors like emergency services and retail and supply chain management.