Evaluation And Calibration Of High-resolution Simulation Of Extreme Temperature And Precipitation In China Based On CWRF Model

Using the daily temperature and accumulated precipitation data of 12 experiments of the CWRF regional climate prediction system and the temperature and precipitation data of 2419 national stations from the National Meteorological Information Center,the simulation accuracy of the CWRF model on the temporal and spatial distribution characteristics of temperature and precipitation is analyzed.Four experiments are selected to evaluate the simulation results of each extreme temperature and precipitation indexes.Finally,RF algorithm is used to revise the simulation results of CWRF temperature,precipitation,maximum temperature and minimum temperature.The conclusions are as follows:(1)The results of daily temperature during 1980 to 2016 returned from each experiment have a high precision,which can simulate the spatial and temporal distribution characteristics of 37 years average temperature well.July to October is the period with the smallest biases in a year.Through the analysis of the statistics such as standard deviation and relative root mean square error of temperature,it is found that the evaluation ability of Cu-KPeta,Cu-Emanuel,RA-CCCMA and PBL-Boulac experiment is relatively good.(2)The four experiments also show certain simulation ability in the spatial and temporal distribution of extreme high temperature index and low temperature index.The percentage of extreme low temperature is generally high in Western China,low slightly in Northeast China and North China.The percentage of extreme high temperature is opposite to the percentage of extreme low temperature.Further analysis found the spatial distribution characteristics of the percentage are closely related to the terrain height.The temporal distribution characteristics of the percentage of extreme temperature are mainly high from June to August.Among all,the simulation accuracy of the lowest temperature is the highest.(3)CWRF experiments simulate the spatial distribution and annual trend of the daily accumulated precipitation well,with the biases between-0.5 mm/d and 0.5mm/d.But the bias of the precipitation in Guangxi is large.Based on the comprehensive analysis,four experiments with better simulation ability are selected for further study,including control experiment,MPThompson,MP-Morrison-Aero and PBL-Boulac.(4)The result of extreme precipitation index calculated by precipitation data shows that CWRF model can simulate extreme precipitation events to some extent.Compared with the observation,the annual average ratio of heavy rain station number and extreme precipitation threshold of each experiment has smaller bias in winter and spring,with larger biases in rainy season.The model can capture the increasing distribution of heavy rain day number ratio and extreme precipitation threshold accurately in China from northwest to Southeast.It can be seen that the amplitude of the contribution rate oscillated along with time in rainy season is significantly larger than that in observation,according to the result of extreme precipitation contribution rate calculated by the 90% threshold value of observed daily precipitation.The spatial distribution figure shows that the simulation results of each experiment are mostly low,however,they are more accurate in the northwest and northeast China.The bias of the three indexes in Yunnan-Guizhou Plateau is equally large,which needs to be improved.But the simulation results in Northwest and Qinghai Tibet Plateau are more accurate.(5)The RF(Random forest)machine learning algorithm greatly improves the negative biases of temperature and maximum(minimum)temperature;it also improves the positive deviation of precipitation.The revised score of each temperature value is generally above 0.93.The score of precipitation calibration is about 0.54,because of the lack of large value in the revised result of monthly average precipitation.