Heavy metals pollution in agricultural soils has been an important problem to human health, mapping large-scale spatial distribution of soil heavy metals is urgently needed. Instead of traditional methods, time-consuming and destructive, soil properties predicted by remote sensing technology shows a lot of advantages, which makes large area of real-time dynamic monitoring as possible. However, before achieving prediction using spectra data, the first thing to do is that finding the spectral characteristics of soil heavy metals. In this paper, taking Cr and Cu for example, the correlations between soil heavy metals content and laboratory-measured reflectance is studied using partial least squares regression (PLSR), which is an adaptive method to examine linear between spectrum and concentration. First of all, using the raw spectra, remove outliers of heavy metals concentration by PLSR modeling. Next, though comparing RMSEC and RMSEV against PLSR components, and cumulative explanatory of spectral components to metal content using different pre-precessing methods, find the right pre-pcocessing is CR and optimum number of components to Cr and Cu are 3 and 2 respectively. Simultaneously, with the meaning of PLSR models regression coefficients, we analysis the spectral characteristics of Cr and Cu, although can not to realize the prediction only take use of these spectra, which is still essential to achieve simulating spatial distribution of soil heavy metal by remote sensing.
Modeling the Spatial Distribution of Soil Heavy Metals Using Random Forest Model—A Case Study of Nairobi and Thirirka Rivers’ Confluence