Super-Resolution Downscaling of Severe Precipitation Ensemble Forecasts Using a Self-Attention-Enhanced CNN

The Ensemble Prediction System (EPS) provides reliable precipitation forecasts. However, constrained by computational resources, its relatively coarse spatial resolution directly limits its capability to predict high-impact severe rainfall events. Given that downscaling to super-resolution offers a computationally efficient and highly practical solution to enhance forecast resolution, this study develops a Self-Attention-Enhanced Convolutional Neural Network (SAECNN) for downscaling coarse ensemble forecasts of precipitation over North China, an area that has frequently experienced severe rainfall in recent years. The SAECNN integrates a self-attention mechanism and inception-style module. It is trained through a two-step process using pairs of high-resolution (HR) and low-resolution (LR) precipitation data under a composite loss function. The model is trained by using 3-h accumulated summer precipitation data from 2010 to 2019 obtained from the ECMWF ERA5-Land reanalysis dataset. Subsequently, taking the Global Ensemble Prediction System of ECMWF (ECMWF-GEPS) as an example, the SAECNN is applied to the LR GEPS to generate HR precipitation ensemble forecasts. Ablation experiments demonstrate that the combination of Huber loss and mean absolute error with minimized missed rate, along with the two-step training strategy, effectively reduces forecast errors. Independent validation against bilinear-interpolated forecasts of the ECMWF-GEPS during 2020–2021 demonstrates that SAECNN yields realistic and detailed precipitation forecasts, reducing the probabilistic forecast bias (ranked probability score) by 8%–10%. A case study of the Zhengzhou extreme rainfall in 2021 further reveals that the model can improve the accuracy of percentile forecasts and neighborhood probability forecasts. Overall, the SAECNN provides a computationally efficient approach for downscaling and improving coarse precipitation forecasts of EPS, while enhancing probabilistic forecasting skills.