Lightning Prediction and Environment Factor Analysis using Random Forest Algorithm in Shandong Province, China

Using random forest algorithm, through under-sampling and selecting probability threshold, the 0–6 hour by hour lightning forecast models in Shandong Province Based on ERA5 data are established. After the independent sample test, the accuracy of the model is 0.827-0.864, and the hit rate is 0.677-0.851 for 0–6 hour lightning prediction in 0.25° × 0.25°grid. The performance of the hourly prediction model gets better with the approach of forecast time. The results demonstrate that all the six models focus on atmospheric stability and thermal factors, especially the temperature and humidity stratification conditions in the middle and lower layers, and reveal that the generation of lightning in Shandong province was closely related to the high temperature and high humidity atmospheric environment conditions under the background of summer maritime climate. The SWISS index ranks the top 5 among the predictive factors of the models. It is a comprehensive index commonly used in thunderstorm prediction, which indicates that the factors selected in the model have obvious physical significance and are consistent with subjective prediction.