关键词: 距离反比法, 梯度距离反比法, 降雨, 温度

Abstract:

As no single interpolation method is optimal for all regions and data, it is important to compare the results obtained using alternative methods applied to each data set In this study, two methods for spatial interpolation of climatic data from sparse weather station networks were compared Forty year monthly mean temperature and precipitation data from regions in Gansu province were interpolated using a deterministic estimation method termed "Inverse distance weighted" (IDS) and a statistical method termed "Gradient plus Inverse Distance Squared"(GIDS) By design, their power parameters were optimized on the basis of minimum root mean square error (RMSE) Corresponding cross validation tests show that optimal inverse distance had consistently better results than usual: As for temperature, the value of MAE decreases by 6 77% and 9 95% in the method of IDW and GIDW respectively; as for precipitation, the value of MAE decreases by 28 19% and 6 25% in the two methods correspondingly Summary statistics were used to determine if one method was significantly better than the other on the basis of mean absolute error (MAE), mean relative error (MRE) and root mean squared error(RMSE) Based on the mean absolute errors from cross validation tests, the methods were ranked GIDS>IDW for interpolating monthly precipitation and temperature, being average by 0 73℃ for monthly temperature and 4 90 mm for monthly precipitation Based on the mean relative errors from cross validation tests, the methods were also ranked GIDW>IDW for interpolating monthly precipitation and temperature, being averagely 9 02% for monthly temperature and 17 82% for monthly precipitation In addition, GIDW yields more accurate predictions than IDW when the correlation between rainfall and elevation is high (less than 0 80 in the case study) For Gansu province, except January and December, the remaining months in the year see the correlation between elevation and rainfall higher than 0 80,while these two months are not very important for us to study soil water balance, so we prefer GIDW method to our study Furthermore, before the interpolation of precipitation, we try to analyse the data, and the results show it is very important to analyse the character and distribution of data before interpolation In our study area, a cubic transformation improves the accuracy well

Key words: Inverse distance weighted, Gradient plus Inverse-Distance-Squared, temperature, precipitation