【摘要】 为比较和探讨中高纬度山区多种气温空间插值方法的精度及适用性,本文利用大兴安岭山脉北段及其周边区域气象站点实测气温数据,以平均绝对误差(MAE)和均方根误差(RMSE)作为评价指标对六种气温空间插值方法进行精度比较。研究结果表明:(1)反距离权重插值法(IDW)、普通克里金插值法(OK)、样条函数插值法(Spline)三种传统的气温插值方法只能粗略反映气温要素的空间分布状况,不适合气象站点稀少而地形起伏较大的区域。(2)BP神经网络(MAE:0.62℃～1.43℃,RMSE:0.84℃～2.02℃)和线性回归+残差内插的空间插值算法(MAE:0.61℃～1.55℃,RMSE:0.82℃～2.30℃)优于常规的插值方法,且BP神经网络算法能较好地反映研究区地形的高低变化以及山脉的走向。(3)在一天中的12:00—22:00时间段内,六种气温空间插值方法的插值精度与插值效果都不理想。对比六种气温空间插值方法表明,BP神经网络算法对研究区气温空间模拟效果最好,且插值效果与训练样本数量成正比。本文可为中高纬度山区气温空间化研究提供参考。更多还原

【Abstract】 To compare the accuracy and applicability of six temperature spatial interpolation methods in mid-high latitudes, the measured temperature data from meteorological stations were used in this study. The accuracy of six temperature spatial interpolation methods was compared by MAE and RMSE as evaluation indexes. The research result showed that:(1) The three traditional temperature interpolation methods including IDW, OK, and Spline, could roughly reflect the spatial distribution of temperature factors, which were not suitable for areas with rare meteorological stations and undulations terrain.(2) Error Back Propagation(MAE: 0.62 ℃～1.43 ℃, RMSE: 0.84 ℃～2.02 ℃) and MLR+RI(MAE: 0.61 ℃～1.55 ℃, RMSE: 0.82 ℃～2.30 ℃) were superior to the conventional interpolation methods, and it could well reflect the changes of terrain and the extension of the mountains range in the study area.(3)In the period of 12:00 to 22:00 of a day, the interpolation accuracy and effect of the six temperature interpolation methods were not ideal. The results showed that the Error Back Propagation had the best simulation effect on spatial of temperature in the study area, and the interpolation effect was proportional to the number of training samples. The study provides a reference for the research of temperature spatialization in mid-high latitude mountainous areas.更多还原