基于人工神经网络的统计降尺度模型研究  

Study of Statistical Downscaling Model Based on Artificial Neural Network

作　　者：王晓燕[1] 杨涛[1] 石晓燕[2] 李会会[1] 郝晓博[1] 

机构地区：[1]河海大学水文水资源与水利工程科学国家重点实验室,江苏南京210098 [2]河海大学土木与交通学院,江苏南京210098

出　　处：《水电能源科学》2012年第4期1-5,共5页Water Resources and Power

基　　金：国家自然科学基金资助项目(40901016;40830639);水文水资源与水利工程科学国家重点实验室自主探索课题基金资助项目(2009586612;2009585512);中央高校基本科研业务费基金资助项目(2010B00714)

摘　　要：基于BP神经网络模型对黄河源区的降水、温度进行了统计降尺度研究,探讨了统计降尺度模式中考虑预报量的敏感大气环流因子随季节变化时对降水的降尺度效果的影响。结果表明,人工神经网络降尺度模型能成功地捕捉黄河源区的日平均温度及气温极值的年际变化趋势,纳什效率系数均达0.95以上;比较CON模型及PIE模型对降水指标的模拟能力,发现两种模型对1961～2000年不同降水指标时间序列的模拟能力相当;从季节尺度看,在冬季PIE模型显示了更好的模拟能力,但在夏秋季节PIE模型对多数降水指标的模拟能力略不及CON模型。总之,CON模型对降水指标的模拟效果更好。Based on BP artificial neural network, the statistical downscaling study of temperature and precipitation in source region of the Yellow river basin is conducted. And the effect of the seasonality of predictors on precipitation down- scaling is discussed. The results show that the artificial neural network downscaling model can simulate annual variation trend of daily-average temperature and air temperature extreme value very well; Nash efficiency factor attains 0.95 or a- bovel for the downscaling model based on not considering the seasonality of predictors（CON model） and the other one considering the seasonality of predictors （PIE model）, the performance for downscating precipitation between 1961 and 2000 is similar; at seasonality, the PIE model performs better in winter while the CON model performs a litter better in summer and autumn; in a whole, the CON model is better for modeling precipitation indexes.

关 键 词：气候变化 人工神经网络 统计降尺度 降水 温度 

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]