Prediction of Precipitation Using Swin-Unet and Meso-Scale NWP Analytical Data

Puntos clave

• The aim is to develop advanced techniques using deep learning for precise precipitation forecasting compared to conventional numerical methods.
• Used deep learning model Swin-Unet for prediction.
• Input mesoscale numerical weather prediction (NWP) analytical data from Japan Meteorological Agency.
• Conducted two experiments, one predicting MSM's precipitation forecast and another for analyzed rainfall.
• Achieved high precision in inferring precipitation forecasts from the MSM model.
• Outperformed MSM in predicting analyzed rainfall, evidenced by better RMSE and thresholds for capture rate, false alarm rate, and threat score.