Search

Establishing the corresponding forecasting models based on numerical model that fails to capture gale events is helpful to improve the ability to forecast gales in coastal areas of China. First, we screened for gale events occurred in coastal areas of Qingdao from 2016 to 2019, and established the dataset of gale events that are missed by the forecast with ECMWF (European Center for Medium-range Weather Forecast) high resolution model (hereinafter referred to as EC model). Second, based on support vector machine (SVM), artificial neural network (ANN) and long short-term memory network (LSTM) algorithms, we established the forecast models for gales along the coast of Qingdao, and used these models to revise the wind speed forecasted by the EC model. Finally, after comparative analysis, a model suitable for forecasting gale along the coast of Qingdao, i.e. the forecast model SVM_2 based on SVM algorithm, was selected and its operational forecasting results were ex⁃ amined. Results show that the SVM_2 model has the smallest forecast error for wind speed compared to other models. In order to evaluate the prediction performance of SVM_2 model in terms of the gale events, two gale events occurred in the coast of Qingdao and influenced by differ⁃ ent weather systems are selected to further examine the prediction error of SVM_2 model and EC model for the gale events. The results show that the error of the maximum wind speed forecasted by the SVM_2 model against the observations is significantly smaller than that of the EC model, and the SVM_2 model also has some improvement on the weak gale along the coast of Qingdao compared to the forecast by EC model.