海涂土壤剖面电导率的协同克立格法估值及不同取样数目的比较研究

ESTIMATES OF ELECTRICAL CONDUCTIVITY FOR COASTAL SALINE SOIL PROFILE USING COKRIGING UNDER DIFFERENT SAMPLING DENSITY

应用地统计学的半方差函数和普通克立格法 ,分析了 2 0m× 2 0m的采样间隔上 ,同一剖面0～ 2 0、2 0～ 4 0、4 0～ 6 0、6 0～ 10 0cm四个层次上海涂土壤电导率的空间相关性和空间分布。并在对协同区域化变量进行交互半方差分析的基础上 ,用协同克立格法 ,将 0～ 2 0cm层上的土壤样品作为辅助变量 ,其他三层上初始及减少后的土壤样品作为主变量 ,来对 2 0～ 4 0、4 0～ 6 0、6 0～ 10 0cm土层上的电导率进行估值。同普通克立格法相比 ,协同克立格法在这三层估值所产生的均方根误差 (RMSE)减少了 9%到 2 5 1% ,平均标准误差(ASE)减少了 9 4 %到 2 1 9% ,预测值与实测值之间的相关系数协同克立格法比普通克立格法提高了 4 6 %到85 %。数据分析的结果表明 ,利用协同克立格法 ,深层的样品数目在被减少至 6 0 %的情况下仍能满足精度要求 ,没有明显地导致空间信息的丢失。因此 ,协同克立格作为一种更为精确和经济的方法来对样品少的或难以测试的变量进行估值不但可以大大提高估测精度 ,而且可以减少采样数目 ,节省采样的成本。

Estimating spatial variability of saline soil electrical conductivity (EC) which was intensely correlative with salinity content is an important issue in costal region. In this study, semivariogram was firstly used to explore the spatial correlation of EC for four depths of 0～20 cm layer, 20～40 cm layer, 40～60 cm layer and 60～100 cm layer on a 20 m by 20 m sampling grid, EC distribution at the four layers were then estimated using ordinary kriging, and then cross-semivariogram was analysed to explore coregionalized relationship between top layer and deeper layers. Finally cokriging was used to improve saline soil EC estimation at deeper layers by treating the samples of 0～20 cm depth as auxiliary variable and the original and reduced samples of the other three layers as primary variable. Compared with the ordinary kriging, the root-mean-square error produced by cokriging decreased by 9% to 25.1%, the average standard error decreased by 9.4% to 21.9%, correlation coefficient between the predicted value and the observed value increased by 46% to 85%. The analysis results indicated the reduction of the original sampling density of 40% would have resulted in no evident loss of spatial information. Cokriging was shown to be a more accurate and economic method, and it allowed easily obtained variables to be used to estimate the difficultly obtained variable based on the correlation on between two variables, and not only can greatly improve prediction accuracy, but also can save the sampling cost.