Evaluation of an empirical orthogonal function-based correction method for precipitation nowcasting

Effectively reducing numerical model errors remains a major challenge in improving the accuracy of 0–2 h precipitation nowcasts. Statistical post-processing provides an important approach to improving short-term precipitation forecasts produced by numerical weather prediction (NWP) models. To improve the precipitation nowcasting accuracy of the China Meteorological Administration Mesoscale Model (CMA-MESO), this study develops a statistical correction method based on Empirical Orthogonal Function (EOF) analysis tailored to CMA-MESO precipitation forecasts using multisource merged precipitation observations as the reference. The method is further applied to ERA5 reanalysis precipitation data to evaluate its dependence on the input data source. The results show that: (1) The EOF-based correction method effectively captures the statistical relationships between the spatial modes and temporal coefficients of observed and forecast precipitation. By correcting the temporal coefficients of the forecast modes while incorporating the spatial structure information derived from observations, the method can effectively reconstruct the corrected precipitation field. (2) Both a representative case study and batch correction experiments demonstrate that the method effectively reduces errors in precipitation magnitude and spatial location of CMA-MESO precipitation forecasts. After correction, the mean spatial correlation coefficient between CMA-MESO forecasts and observations increases from 0.08 to 0.56. For precipitation thresholds of 0.1 and 30 mm·h⁻¹, the mean Equitable Threat Score (ETS) increases from 0.07 and 0.01 to 0.40 and 0.10, respectively, while the mean Fractions Skill Score (FSS) increases from 0.40 and 0.10 to approximately 0.80 and 0.50, respectively. (3) The method also substantially improves ERA5 reanalysis precipitation, with the mean spatial correlation coefficient increasing from 0.12 to 0.61, demonstrating its applicability to different precipitation data sources. (4) The correction method provides stable improvements for CMA-MESO precipitation forecasts within the first 3 h, whereas its effectiveness decreases rapidly at lead times beyond 3 h. Overall, the EOF-based correction method can effectively reduce precipitation nowcasting errors in CMA-MESO and significantly improve the accuracy of short-term precipitation forecasts, providing a useful reference for the operational application of statistical post-processing in precipitation nowcasting.