Abstract: The contamination of heavy metals in agricultural soils caused by natural or anthropogenic factors strengthens the importance of soil environmental quality survey and managements. Nevertheless, the conventional methods of sampling and analysis are time-consuming and costly. Visible and near-infrared (Vis-NIR) reflectance spectroscopy is a rapid and inexpensive alternative to measure soil physical and chemical parameters. This study explored the capacity of Vis-NIR reflectance spectroscopy models for soil heavy metal contamination risk classification. We collected 390 agricultural soils from a typical anthropogenic contaminated area (Wenling, Zhejiang) and a typical high geological background area (Hengxian, Guangxi), examining the concentrations of eight soil heavy metals (As, Cd, Cr, Cu, Hg, Ni, Pb and Zn), soil pH values, as well as soil Vis- NIR reflectance spectroscopy. Partial least square regression (PLSR) and support vector machine (SVM) algorithms were used to calibrate regression models for predicting the concentrations of soil heavy metals and pH values. Based on the predicted values of spectral models, soil heavy metal contamination risk types were classified. Results showed that, the regression prediction deviation (RPD) values of spectral models for Cd and Cu, the major pollutant elements in Wenling soils, were 1.23 and 1.19, respectively. The spectral prediction mechanism is the correlation between Cd and Cu with organic matter. The two major pollutant elements in Hengxian soils are As and Cd, which had the RPD values of 1.98 and 1.93, respectively. The spectral prediction mechanism is the correlation between As and Cd with iron oxides and clay minerals. In the high geological background soils at Hengxian, there were generally strong positive correlations between soil heavy metals and iron oxides, where spectral models gave more accurate predictions for heavy metals. The RPD values of spectral models for soil pH values were 1.76 at Wenling and 1.68 at Hengxian. Spectral classification of soil contamination risk has reliable overall accuracy (75.0%~100% at Wenling, 80.0%~100% at Hengxian). Therefore, the spectral method combining with remote sensing technology is helpful for rapid classification of heavy metal contamination risk in agricultural soils.

Key words: soil heavy metals, contamination risk classification, visible and near infrared (Vis-NIR) reflectance spectroscopy; pH values, hyperspectral remote sensing