Research on correction methods for low cloud cover forecasts in Hunan province

Utilizing the low-level cloud cover data from the Three-Dimensional Cloud Merge Analysis(3DCloudA) product for the Hunan region from June 1, 2023, to May 31, 2024, combined with CMA-WSP2.0 meteorological forecast data(NWP), this study investigates correction methods for low cloud cover forecasts in Hunan using the Probability Density Function (PDF) matching method and the LightGBM (LGB) machine learning approach. The results indicate that the NWP low cloud cover forecasts are consistently overestimated across all four seasons, with more pronounced overestimations in autumn and winter. Both the PDF and LGB methods demonstrate effective corrections to the numerical model, with the LGB method showing relatively better performance in spring, autumn, and winter, with mean biases of -1.6%, 1.5%, and -1.3%, respectively. The Root Mean Square Error (RMSE) shows reduction after corrections by both methods, with the LGB method exhibiting more significant reduction, reducing the RMSE by over 10% in most seasons. The distribution of RMSE for the LGB correction model is unbiased. The correlation coefficients improved across all seasons after LGB method corrections, with a notable increase in summer, where the average correlation coefficient rose from 0.26 to 0.32. In terms of the standard deviation ratio, the LGB method's ratio is less than 1 across all seasons, whereas the PDF method's ratio is closer to 1 compared to the LGB method. Case studies are conducted for both clear-sky and rainy conditions to assess model performance in extreme cloud cover events. Under clear-sky conditions, both methods produced satisfactory forecasts; whereas during rainy conditions, the extreme-event-optimized LGB method exhibited superior forecasting capability for extreme low cloud cover cases.