三种多站点随机天气发生器日降水模拟对比研究——以嫩江流域为例  

Comparison of precipitation simulation by three multisite daily scale weather generators

作　　者：范聪慧 殷水清 李志[2] FAN Conghui;YIN Shuiqing;LI Zhi(State Key Laboratory of Earth Surface Processes and Resource Ecology,Faculty of Geographical Science,Beijing Normal University,Beijing 100875,China;College of Natural Resources and Environment,Northwest A&F University,Yangling 712100,Shaanxi,China)

机构地区：[1]北京师范大学地理科学学部,地表过程与资源生态国家重点实验室,北京100875 [2]西北农林科技大学资源环境学院,杨凌712100

出　　处：《地理研究》2023年第5期1425-1440,共16页Geographical Research

基　　金：国家自然科学基金项目(41877068)。

摘　　要：天气发生器是对气候模式低分辨率输出数据进行统计降尺度的一种重要工具,可为气候变化对生态、水文、作物生长和土壤侵蚀等地表过程的影响评价提供高时空分辨率气候输入数据。特别是针对流域或区域尺度的模拟,为保留降水等变量的时空相关性,需使用多站点天气发生器。当前多站点天气发生器众多,但其适用性缺乏系统评估。本文利用嫩江流域及其周边区域75个气象站点1961—2020年日降水数据,对比了三种主流的多站点日尺度天气发生器,分别为MulGETS模型(Multi-site weather GEneraTor)、TSWG模型(Two-Stage Weather Generator)和EHS(EOFA+HHT+SS)模型,在日、月和年降水量均值和极值,日降水空间相关性和时间相关性等方面的模拟精度。结果表明:①三个模型中EHS模型表现最佳,除了年降水偏态系数以外,其余参与评估指标的平均绝对百分比误差均在15%以下;TSWG模型表现其次,除了年降水偏态系数、月降水偏态系数和一阶自相关系数以外,其余参与评估指标的平均绝对百分比误差在30%以下。②TSWG模型对95%分位数日降水量、最大连续降水日数、最大和平均连续非降水日数四个指标的模拟最优;其余指标均为EHS模型最优。③三个模型在降水极值、连续降水日数和连续非降水日数方面的模拟能力均有待进一步提高。研究可为多站点日降水发生器的优选和进一步发展提供参考。Weather generators are one of important statistical downscaling methods that can provide high spatio-temporal resolution random simulated climate data for assessing the impact of climate change on earth surface processes,such as crop production,ecological,hydrological and soil erosion processes.In recent years,multi-site weather generators based on spatial correlation have been developed greatly.Among many climatic variables,precipitation is an important and difficult one to simulate.Based on the daily precipitation data of 75 meteorological stations over the Nenjiang River Basin and its surrounding areas from 1961 to 2020,we compared the simulation accuracy of the MulGETS(Multi-site weather GEneraTor)model,TSWG(Two-Stage Weather Generator)model and EHS(EOFA+HHT+SS)model in terms of daily,monthly and annual precipitation,daily precipitation extremes,spatial correlation and temporal correlation of daily precipitation.The MulGETS model simulated the spatial relationship through spatially correlated random numbers which were similar to the spatial relationship between observations.The TSWG model wsa a two-stage model,which generated singlesite precipitation in the first stage and realized spatial correlation with shuffle method in the second stage.The EHS model decomposed the spatio-temporal series into spatial modes and time series,operated random simulation on the time series,and retained the spatial mode characteristics.The results showed that:(1)EHS model is the best of the three models,and the average absolute percentage error of the other indexes is less than 15%except for the annual precipitation skewness coefficient.TSWG model performed second,except for the annual precipitation skewness coefficient,monthly precipitation skewness coefficient and firstorder autocorrelation coefficient,the average absolute percentage errors of other indexes involved in the evaluation were less than 30%.(2)The accuracy of MulGETS model in simulating spatial correlation was the best of the three models.TSWG model is the best to

关 键 词：天气发生器 降水 MulGETS TSWG EHS 

分 类 号：P426.6[天文地球—大气科学及气象学] P333[天文地球—水文科学]