WeatherFusionNet: Predicting Precipitation from Satellite Data

2022/11/30 by Pihrt, Jiří, Raevskiy, Rudolf, Šimánek, Petr +1 · 3 citations
#Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Machine Learning (cs.LG)

paper · doi:10.48550/arxiv.2211.16824

Abstract

The short-term prediction of precipitation is critical in many areas of life. Recently, a large body of work was devoted to forecasting radar reflectivity images. The radar images are available only in areas with ground weather radars. Thus, we aim to predict high-resolution precipitation from lower-resolution satellite radiance images. A neural network called WeatherFusionNet is employed to predict severe rain up to eight hours in advance. WeatherFusionNet is a U-Net architecture that fuses three different ways to process the satellite data; predicting future satellite frames, extracting rain information from the current frames, and using the input sequence directly. Using the presented method, we achieved 1st place in the NeurIPS 2022 Weather4Cast Core challenge. The code and trained parameters are available at \urlhttps://github.com/Datalab-FIT-CTU/weather4cast-2022.

Cited by

• PIANO: Physics-informed Dual Neural Operator for Precipitation Nowcasting
• A Space-Time Transformer for Precipitation Nowcasting
• Enhancing Heavy Rain Nowcasting with Multimodal Data: Integrating Radar and Satellite Observations

Related

• PAUNet: Precipitation Attention-based U-Net for rain prediction from satellite radiance data
• Data-driven Precipitation Nowcasting Using Satellite Imagery
• PIANO: Physics-informed Dual Neural Operator for Precipitation Nowcasting
• Super-resolution Probabilistic Rain Prediction from Satellite Data Using 3D U-Nets and EarthFormers
• MetNet: A Neural Weather Model for Precipitation Forecasting