Abstract: This paper takes Daxigou mining area in Zhashui county, Shaanxi province as the research area. By collecting soil samples on the spot, combining with radiation luminance value and spectral derivative index extracted from Landsat 8 multi-spectral remote sensing images, and three topographic factors extracted from ASTER GDEM, modeling factors are determined through correlation analysis. The random forest regression model of arsenic, copper and lead is established by K-fold cross validation. The experimental results show that the prediction accuracy of the established model is better than that of the multiple linear regression model and the CART model, which verifies that the random forest regression model is suitable for the heavy metal content inversion in the case of small samples. Through the field investigation, the spatial inversion results are in good agreement with the actual situation, which proves the accuracy of the random forest regression model based on multi-spectral remote sensing in the soil heavy metal inversion of mining area.

Key words: soil heavy metal inversion, multispectral remote sensing, K-fold cross validation, random forest regression model