Development of a Mesoscale Atmospheric and Precipitation Prediction Model Using Afno With Consideration of Precipitable Water Vapor

Key Points

• The aim is to develop a mesoscale atmospheric and precipitation prediction model using AFNO, focusing on rainfall forecasting accuracy.
• Applied AFNO to mesoscale weather data
• Incorporated rainfall advection and precipitable water vapor into the input
• Utilized Fourier Amplitude and Correlation Loss as the loss function
• Evaluated prediction accuracy for heavy precipitation events (≥30mm/h) over 2-3 hour forecasts
• Improved accuracy in predicting rainfall during linear precipitation events
• Validated AFNO's effectiveness with respect to heavy rainfall forecasts
• Achieved significant advancements in prediction of precipitation events with precipitation advection considered