基于光谱信息辅助的污灌区农田土壤镉协同克里格分析

摘要

为获取农田土壤重金属精确空间分布信息,以某污灌区52个土壤全量镉、有效镉含量为目标变量,以土壤一阶微分光谱为辅助协同变量,采用协同克里格法,进行空间变异及插值研究.结果表明,土壤反射光谱相对有机质、氧化铁等单一土壤环境变量,能反映更多土壤表面属性信息,与土壤镉含量表现出更高显著相关性；选择其一阶微分光谱作为协同变量,进行Cokriging插值,与普通Kriging和以有机质、氧化铁等协同变量的Cokriging插值结果相比较,估测精度明显提高.以土壤光谱作为辅助变量,能大大提高土壤重金属插值精度,获取更精确空间分布信息,而且相对常用协同变量,具有测定简单、省时、无损等优点,是提高土壤重金属空间插值的理想辅助因子.%To acquire the accuracy distribution information of soil heavy metal,improving interpolation precision is very important for agricultural safety production and soil environment protection.In the present study,the spatial variation and Cokriging interpolation of soil Cd was studied in a sewage irrigation area.Fifty two soil samples were collected to measure the contents of soil total Cd (TCd),available Cd (ACd),pH,organic matter (OM),iron oxide (Fe2O3) and soil reflection spectrum.Through correlation analysis,it was found that TCd and ACd had a significant correlation with soil first-order differential spectrum (-0.585** at 759 nm and-0.551 * * at 719 nm,respectively),which were much higher than the correlation coefficients between soil Cd contents and other environmental variables (pH,OM and Fe2 O3).The spatial patterns of soil Cd were predicted by Cokriging which used soil first-order differential spectrum as covariate.Compared with the Kriging,the root-mean-square error decreased by 8.22％ for TCd and 20.09％ for ACd,respectively; the correlation coefficients between the predicted values and measured values increased by 27.45％ for TCd and by 53.13％ for ACd,respectively.Meanwhile,the prediction accuracy improved by Cokriging with soil spectrum as covariate was still higher than by Cokriging with soil environment variables (OM and Fe2 O3).Therefore,it was found that Cokriging was a more accurate interpolation method which could provide more precise distribution information of soil heavy metal.At the same time,soil reflection spectrum was shown to be more economic,time-saving and easier to acquire than these usual environment variables,which indicated that soil spectrum information is more suited as a covariate used in Cokriging.