Enhancing Radar-Based Precipitation Nowcasting Model with AI-Predicted Precipitation Intensity Change Rates

The importance of precipitation nowcasting is gradually expanding due to the increasing frequency and intensity of localized rainfall caused by climate change. The growth and decay processes of precipitation are critical factors influencing the accuracy of precipitation nowcasting, necessitating advanced modeling approaches. This study proposes a novel methodology that integrates artificial intelligence (AI) with high-resolution radar data to predict the growth and decay processes of precipitation, incorporating these predictions into a radar-based nowcasting model. In this study, AI was applied to predict radar-based precipitation intensity change rates up to two hours ahead, and these predictions were integrated into a precipitation nowcasting model. The AI effectively learned the spatiotemporal patterns of nonlinear precipitation evolution using the RainNet architecture. The AI was trained on three years (2021 – 2023) of radar-derived precipitation intensity change rates, with one year (2020) used for validation to evaluate its performance. The nowcasting model was developed using cross-correlation techniques to calculate motion vectors of the precipitation system at different spatial scales, and a semi-Lagrangian backward extrapolation method was employed for precipitation prediction. Integrating AI-predicted precipitation intensity change rates into the nowcasting model resulted in significant improvements in prediction performance. The results showed a 10% improvement in precipitation prediction accuracy compared to the baseline nowcasting model that did not incorporate AI-based precipitation intensity change rate predictions. The model effectively captured rapid changes in precipitation intensity, demonstrating the utility of AI-based predictions for short-term nowcasting. This study highlights the potential of combining traditional nowcasting models with AI techniques, presenting a promising approach for enhancing precipitation prediction accuracy.

This research was supported by the "Development of radar based severe weather nowcasting technology (KMA2021-03122)" of "Development of integrated application technology for Korea weather radar" project funded by the Weather Radar Center, Korea Meteorological Administration.