Abstract
Heavy metal pollution in soil seriously threatens ecosystem and human health. However, traditional monitoring methods usually rely on intensive sampling, which is costly and difficult to be extended to large regional scales. Based on Orbita Hyperspectral Satellites (OHS) imagery and 175 sample sets out of 1589 samples, Multiple Linear Regression (MLR), Partial Least Squares Regression (PLSR), Support Vector Machine Regression (SVM), Back Propagation Neural Network (BPNN), and Convolutional Neural Network (CNN) models were constructed to predict eight elements (As, Cd, Cr, Cu, Hg, Ni, Pb, Zn). To explore the feasibility of using a small number of samples to invert the distribution trend of heavy metals in a large area. The results show among the above eight elements, the retrieval of Pb is the best, with the R(2) of BPNN and CNN reaches 0.80. BPNN and CNN achieves the optimal inversion of As, Cd and Pb. MLR and PLSR has the best accuracy in Cr and Cu, Hg, Ni and Zn. In addition, the distribution trends of 8 heavy metals retrieved from a small number of samples were basically consistent with the interpolation maps of 1589 samples, indicating that it is completely feasible to use a small number of samples to retrieve the distribution trends of heavy metals in large areas. This study provides important technical support for regional soil pollution prevention and control, and has significant application value and promotion potential.