基于省域尺度的农田土壤重金属高光谱预测  

Prediction of heavy metal content in soil of cultivated land:Hyperspectral technology at provincial scale

作　　者：夏芳[1] 彭杰[1,2] 王乾龙[1] 周炼清[1,3] 史舟[1,3] 

机构地区：[1]浙江大学环境与资源学院农业遥感与信息技术应用研究所,浙江杭州310058 [2]塔里木大学植物科学学院,新疆阿拉尔843300 [3]浙江大学唐仲英传感材料及应用研究中心,浙江杭州310058

出　　处：《红外与毫米波学报》2015年第5期593-598,605,共7页Journal of Infrared and Millimeter Waves

基　　金：国家863计划课题(2013AA10230105);国家自然科学基金(41271234)~~

摘　　要：利用浙江省36个县市的643个农田耕层土样的可见-近红外反射率数据以及重金属与有机质含量数据,分析了Ni、Cu、As、Hg、Zn、Cr、Cd、Pb含量与有机质含量的相关性,对比了不同重金属元素与有机质敏感波段的位置,并建立了各重金属元素含量的偏最小二乘回归(PLSR)模型.研究结果表明,Ni、Cr与有机质的相关性最优,As最差,相关系数分别为0.54、0.59、0.20,各重金属元素与有机质的相关系数与它在前三个主成份载荷图中与有机质的距离成反比;不同的重金属元素与有机质高光谱敏感波段的重叠度、回归系数的正负一致性具有明显差异,与有机质相关性越高的元素,其重叠度也越高、正负一致性也越好;在所有8种重金属元素的PLSR预测模型中,Ni、Cr的建模与预测效果较好,RPD值分别为1.94、1.80,模型具有一般的定量预测能力,其余6种重金属元素预测模型的RPD值均在1.00和1.40之间,模型只具备区别高值和低值的预测能力.该研究结果为大尺度区域土壤重金属污染的高光谱遥感监测提供了一定的理论依据与参考.A total of 643 farmland topsoil samples distributed in 36 counties and cities of Zhejiang Province w ere collected. The correlation betw een contents of Ni,Cu,As,Hg,Zn,Cr,Cd and Pb and that of organic matter w as probed by measuring the reflectance of soil samples in visible-near infrared light band. The characteristic w ave bands of heavy metal elements and organic matter w ere compared. The partial least squares regression（ PLSR） model for the content of each heavy metal element w as established. The results indicated that Ni and Cr have the best correlation w ith organic matter,w hile As has the w orst,w ith the correlation coefficients 0. 54,0. 59 and 0. 20,respectively. The distance betw een heavy metal elements and organic matter in the first three principal components loading diagram w as inversely proportional to their correlation coefficient. Degree of overlap betw een different heavy metal elements and organic matter at hyperspectral sensitive band and the positive and negative consistency of regression coefficients varied greatly,the greater the correlation w ith organic matter is,the higher degree of overlap is,and the better the positive and negative consistency. In PLSR models of heavy metals,models for Ni and Cr performed w ell in modeling and predicting w ith a good ability of quantificational prediction,w ith RPD values of 1. 94 and 1. 80 separately. The remaining models for other 6 heavy metals could only conduct distinguishing for high and low values w ith RPD values ranged from 1. 00 to 1. 4. The results of this study provide certain theoretical assistance and reference for hyperspectral remote sensing monitoring of soil contamination by heavy metals in large-scale areas.

关 键 词：农田土壤 重金属污染 高光谱 预测 

分 类 号：O211.67[理学—概率论与数理统计]