Hyperspectral estimation of mercury content of soil in Oasis city in arid zones of China

Abstract

Mercury (Hg) is one of the most toxic heavy metals to the human body. Conventional methods for measuring Hg content in soil are time-consuming and expensive. In order to select a high-effective method for estimating soil Hg content based on hyperspectral remote sensing techniques, a total of 85 soil samples were collected from the Urumqi city, northwest China, to obtain the Hg contents and related hyperspectral data. A total of 12 spectral transformation methods were used to the original spectral data for selecting significant wavebands. The partial least squares regression (PLSR), random forest regression (RFR) and support vector machine regression (SVMR) were used to establish hyperspectral inversion models for soil Hg content using selected significant wavebands. The results showed that the Hg content of soil was significantly higher than its corresponding background value, which obviously enriched in soil in the study area. The spectral transformation of the original wavebands can effectively reduce the interference of the background noise and can improve the correlations between the spectral data and the soil Hg content. The RFR model based on logarithmic first-order differential (LTFD–RFR) or on reciprocal logarithmic first-order differential (ATFD–RFR) had the best inversion effects, with the highest prediction ability (R 2 = 0.856, RMSE = 0.002 and MAE = 0.072). The LTFD–RFR or ATFD–RFR methods can be used as a means of inversion of Hg content of soil in oasis cities. The novel contribution of this work is to construct hyperspectral inversion model which can accurately estimate the Hg content of urban soils in arid zones. Results of this study can provide a technical support for hyperspectral estimation of soil Hg content.