中文摘要：

采用不同空间插值方法（普通克里格法OK、反距离权重法IDW和径向基函数法RBF）,对黄土高原地区85个气象站点1957—2013年逐日降水数据进行年均降水量空间插值分析,采用平均绝对误差（MAE）、均方根误差（RMSE）、Pearson相关系数（R）以及准确度（AC）对插值结果进行对比,研究不同方法对该地区年均降水量插值结果的影响.结果表明：研究期间,黄土高原地区年均降水量的Moran I值为0.67,各站点年均降水量存在较强的空间自相关性.63个预测站点和22个检验站点的验证结果显示,3种方法的实测值和预测值均显著相关,但反距离权重法和径向基函数法的平均绝对误差（MAEIDW=38.98,MAERBF=34.61）和均方根误差（RMSEIDW=51.49,RMSERBF=43.79）均高于普通克里格法.通过普通克里格法的不同半变异函数模型（环形模型、球体模型、指数模型、高斯模型）对比发现,环形模型的平均绝对误差最小（MAE=32.34）,准确度最高（AC=0.976）,指数模型平均绝对误差最大（MAE=33.24）.综合对比不同插值方法实测值与估算值的验证结果表明,采用半变异函数为环形模型的普通克里格法是进行黄土高原年均降水量插值较好的方法.

英文摘要：

Based on 57 years （1957-2013） daily precipitation datasets of the 85 meteorological stations in the Loess Plateau region, different spatial interpolation methods, including ordinary krig- ing （OK） , inverse distance weighting （IDW） and radial-based function （RBF） , were conducted to analyze the spatial variation of annual average precipitation regionally. Meanwhile, the mean ab- solute error （MAE）, the root mean square error （RMSE）, the accuracy （AC） and the Pearson correlation coefficient （R） were compared among the interpolation results in order to quantify the effects of different interpolation methods on spatial variation of the annual average precipitation. The results showed that the Moran＇ s I index was 0.67 for the 57 years annual average precipitation in the Loess Plateau region. Meteorological stations exhibited strong spatial correlation. The validation results of the 63 training stations and 22 test stations indicated that there were significant correla- tions between the training and test values among different interpolation methods. However, the RMSE （IDW=51.49, RBF=43.79） and MAE （IDW=38.98, RBF=34.61） of the IDW and the RBF showed higher values than the OK. In addition, the comparison of the four semi-variagram models （Circular, Spherical, Exponential and Gaussian ） for the OK indicated that the circular model had the lowest MAE （32.34） and the highest accuracy （0.976）, while the MAE of the expo- nential model was the highest （33.24）. In conclusion, comparing the validation between the train- ing data and test results of the different spatial interpolation methods, the circular model of the OK method was the best one for obtaining accurate spatial interpolation of annual average precipitation in the Loess Plateau region.