利用Cokriging提高估算土壤盐离子浓度分布的精度——以黄河三角洲为例

Abstract

估算土壤中化学物质的含量与空间分布是了解多孔介质中水盐运移规律并进而因地制宜地提出盐渍土改良措施的关键。大面积的实地采样分析费时费力且耗资巨大。通过地统计分析,使用有限的采样数据可获得土壤溶质的准确变异。本文探讨和比较了Ordinarykriging(OK)与Cokriging(COK)这两种内插方法。结果显示一半的采样点数据的COK较之全部采样点数据的OK精度更高,相对均方根误差降幅为130.83%;采用同样的协同变量(239个全盐量数据),一半的采样点数据的COK较之全部采样点数据的COK精度更高,相对均方根误差降幅为20.10%。协同变量与主变量的相关度决定了COK的预测精度,当相关系数由77%升高为99%时,相对均方根误差降低了48.30%。

Estimation of the quantity and distribution of soil chemicals is a major component in the study of chemical transportation in the vadose zone and groundwater system. Such information is also essential in undertaking any proper measures to meliorate soil salinization. However, it is a time-consuming, laborious, and expensive process to carry out detailed sampling in the field, especially when it is large. Accurate variability of soil solute can be determined from a limited number of the available samples through geostatistical analysis. In this study two interpolation methods (ordinary kriging and cokriging) were compared with each other in terms of their accuracy. It is found that cokriging of half of the observations (239) resulted in more accurate results than ordinary kriging of all the samples. Cokriging is able to reduce relative root mean square error (RMSE) by 130.83% in comparison with ordinary kriging. Using the same number of samples (239) for the secondary variable (total salt), cokriging attained a higher accuracy with half of the samples than it did with all the samples, the relative reduction of RMSE being 20.10%. Furthermore, the relationship between the secondary and the primary variables governs the estimation accuracy. As the correlation coefficient between them increases from 77% to 99%, the relative RMSE of estimation is reduced by 48.30%.