基于人工神经网络的甘肃省年积温空间插值研究

中国西部地区气象站点空间分布不均匀,而且数量较少,导致气象要素空间插值的误差较大。以甘肃省为研究区,将其划分为东、中、西3个插值区域,利用人工神经网络方法,对1960~2009年气象站点地面观测的≥0℃年积温值进行拟合,将202个气象站点数据扩展到586个,插值结果表明:3个区域年积温平均绝对误差(MAE)为246.53℃、平均相对误差(MRE)为8.37%;东中西3个区域精度与原有气象站点数据量有明显关系。研究方法和结果可为进行甘肃省生态环境变化研究提供方法和数据参考,为进一步进行区域空间插值提供数据基础。

Study on annual accumulative temperature spatial interpolation of Gansu Province based on artificial neural network

There are uneven distribution of a few meteorological stations in western China,which result in more errors in spatial interpolation of meteorological elements.Selected Gansu as study area,dividing into eastern,middle and western region,the paper used artificial neural network to fit ≥0 ℃ annual accumulative temperature between 1 960 ～ 2009 and extended the meteorological data from 202 meteorological stations to 586 sites.The results of the interpolation showed that the mean absolute error （MAE）and mean relative error （MRE）of average annual accumulative temperature in three regions were 246.53 ℃ and 8.37%,respectively. The accuracy of data is related to the number of meteorological stations.The study provide the approach and references for the study of ecological environment in Gansu,and the basis for further studying of regional spatial interpolation.