Short-time Strong Rainfall Forecast Based on Ensemble Learning in Fujian Province
In order to improve short-time strong rainfall forecast, the LightGBM algorithm is applied to build the hourly precipitation forecast model based on the precipitation observation data and CMA-GD-3km model forecast products of Fujian Province from April to September in 2019 and 2020. Correction models are optimized by features processing, Bagging (Bootstrap Aggregating) and hyperparameters search. Combined with AUC, AUPR and traditional classification indices, A series of experiments are designed to evaluate different modeling schemes and verify the applicability for short-time strong rainfall forecast. The results show that: (1) All modeling schemes improve the original numerical model forecast representing the high POD and FAR in varying degrees. Bagging can enhance the stability of model prediction, and the slightly unbalanced sub-training set contributes to better performance by reducing the FAR, the best TS score of validation set is about 17.5%. (2) The largest contribution to the classification information gain is K index, followed by 500hPa dew point and time parameters. (3) The ranking of experiment indices from good to bad: random cross-validation, random hourly cross-validation and operational simulation test, indicating that the validity of correction models mainly result from the sample information at the same or adjacent moments. (4) The dynamic fusion scheme of heterogeneous models based on logistic regression increases indices of static homogeneous models, which decreases at least 540,000 false alarm samples with approximately 50% POD.