Combining AI and numerical models for improving short-medium range precipitation prediction

Abstract

Short-to-medium-range precipitation forecasts, critical for water resource management and flood risk mitigation, remain challenging for both conventional numerical weather models (NWMs) and rapidly-advancing artificial intelligence (AI)-based models. We present an AI-NWM hybrid approach to improve this task over the Chinese mainland. The approach develops an AI-based model that predicts 24-hour total precipitation using meteorological fields, bypassing uncertainties in NWM cloud parameterizations, and then drives it with NWM-predicted meteorological fields to produce multi-lead daily precipitation forecasts, leveraging the physical consistency of NWM forecasts. Using meteorological forecasts from the National Centers for Environmental Prediction Global Forecast System (GFS), the approach significantly enhances GFS precipitation prediction accuracy. For forecasts with leads of 1–16 days, the root-mean-square error drops by 16.8%, and the pattern correlation coefficient rises by 22.1%, extending the reliable horizon by nearly 2 days. Higher or comparable equitable threat scores across 1–100 mm day−1 thresholds confirm its reliability.