Multi-Mode Integrated Forecast of Surface Temperature based on Random Forest

Due to the influence of global climate change, extreme weather events such as high temperature and severe cold occur frequently in recent years. Traditional numerical prediction has been unable to meet the requirement of accuracy of temperature prediction. To improve the accuracy of forecast, single deterministic forecast has been gradually changing into multi-mode integrated forecast. Random forest has strong adaptability to data sets, without standardized training but fast training speed. It can be applied to large-scale data sets and is suitable for solving nonlinear data regression analysis problems such as meteorological factors. In this paper, the multi-mode integrated forecast of 2m surface temperature based on random forest algorithm is studied by using the forecast results of ECMWF, JMA and NCEP models in TIGGE data set and compared with the multi-mode integrated forecast of CART algorithm. The results show that the prediction performance of multi-mode integrated forecasting is significantly improved compared with the single mode forecast of each center, meanwhile the prediction performance of multimode forecast based on random forest is significantly improved compared with the existing CART multi-mode forecast.