两种统计降尺度模型在雅砻江流域未来气候预测中的应用  

Application of Two Statistical Downscaling Models for Future Climate Prediction in Yalongjiang River Basin

作　　者：秦兴隔 董晓华[1,2,3] 董立俊 杨晨 王雅琳 李璐[1,2,3] QIN Xing-ge;DONG Xiao-hua;DONG Li-jun;YANG Chen;WANG Ya-lin;LI Lu(College of Hydraulic and Environmental Engineering,China Three Gorges University,Yichang 443002,China;Engineering Research Center of Eco-environment in Three Ciorges Reservoir Region,Ministry of Education,Yichang 443002,China;Hubei Collaborative Innovation Center for Water Resoures Security,Wuhan 430072,China)

机构地区：[1]三峡大学水利与环境学院,湖北宜昌443002 [2]三峡库区生态环境教育部工程研究中心,湖北宜昌443002 [3]水资源安全保障湖北省协同创新中心,湖北武汉430072

出　　处：《水电能源科学》2022年第8期17-21,共5页Water Resources and Power

基　　金：欧洲空间局、中国国家遥感中心项目(58516);中国电建集团华东勘测设计研究院有限公司项目(DJ-ZDZX-2016-02-09)。

摘　　要：全球气候变化背景下,选择适宜的降尺度方法预估区域尺度的未来气候变化趋势,对雅砻江流域水资源规划利用具有重要意义。采用人工神经网络降尺度模型(ANN)和统计降尺度模型(SDSM),基于雅砻江流域内13个国家气象站1981~2005年数据和NCEP再分析资料构建两种模型的历史期日最高气温、最低气温、降水降尺度模型,将CNRM-CM5全球气候模式的3种排放情景(RCP2.6、RCP4.5、RCP8.5)数据输入模型预测未来2022~2100年最高气温、最低气温、降水变化,比较人工神经网络与SDSM降尺度模型在雅砻江流域模拟效果。结果表明,ANN模拟气温、验证期降水效果优于SDSM,确定性系数R^(2)分别高0.02~0.27、0.01~0.14,均方根误差R_(RMSE)分别小0.22~2.40、0.01~0.77,ANN模拟率定期降水效果不如SDSM,确定性系数R^(2)小0.02~0.07;在空间分布上,ANN模拟气温效果在整个流域优于SDSM,日最高气温和日最低气温模拟效果最好的区域为中游和上游,ANN模拟验证期日均降水在流域上、下游明显优于SDSM,在流域中游略优于SDSM,模拟率定期效果在流域大部分区域不如SDSM;未来三种RCP情景下,SDSM、ANN预测未来最高气温、最低气温、降水在中后期呈增加趋势,且在RCP8.5情景下后期增加最高。Under the background of global climate change,choosing the appropriate climate downscaling method in the Yalong River basin to predict the future climate change trend at the regional scale is of great significance to the rational utilization of regional water resources.The Yalong River Basin was selected as the research object,and the downscaling scale models of daily maximum temperature,minimum temperature and precipitation in historical periods were constructed by using artificial neural network downscaling scale model(ANN)and statistical downscaling scale model(SDSM)based on the 1981-2005 data of 13 national meteorological stations in the basin and NCEP reanalysis data.Three emission scenarios(RCP2.6,RCP4.5,and RCP8.5)of the CNRM-CM5 global climate model were input into the model to predict the maximum temperature,minimum temperature and precipitation changes during 2022-2100,and the simulation effects of artificial neural network and SDSM downscaling model in Yalongjiang River Basin were compared.The results show that the simulation results of ANN is better than SDSM in temperature and validation period precipitation,the deterministic coefficient R^(2) is 0.02-0.27 and 0.01-0.14 higher respectively,and the root mean square error R_(RMSE) is 0.22-2.40 and 0.01-0.77 lower respectively.The calibration period precipitation of ANN simulation is not as good as SDSM,and the deterministic coefficient R^(2) of ANN simulation is 0.02-0.07 smaller than SDSM.In spatial distribution,ANN is better than SDSM in simulating temperature in the whole basin.The best simulation results of the daily maximum temperature and the daily minimum temperature are in the middle and upper reaches of the basin.The validation period average daily precipitation of ANN simulation is significantly better than SDSM in the upper and lower basin,and slightly better than SDSM in the middle basin.The calibration period effect of ANN simulation is inferior to SDSN in most areas of the basin.The SDSM and ANN predict that the maximum temperature,minimu

关 键 词：SDSM 人工神经网络 CNRM-CM5 气候变化 雅砻江流域 

分 类 号：TV13[水利工程—水力学及河流动力学] P467[天文地球—大气科学及气象学]