Estimating soil zinc concentrations using reflectance spectroscopy

Abstract

Soil contamination by heavy metals has been an increasingly severe threat to nature environment and human health. Efficiently investigation of contamination status is essential to soil protection and remediation. Visible and near-infrared reflectance spectroscopy (VNIRS) has been regarded as an alternative for monitoring soil contamination by heavy metals. Generally, the entire VNIR spectral bands are employed to estimate heavy metal concentration, which lacks interpretability and requires much calculation. In this study, 74 soil samples were collected from Hunan Province, China and their reflectance spectra were used to estimate zinc (Zn) concentration in soil. Organic matter and clay minerals have strong adsorption for Zn in soil. Spectral bands associated with organic matter and clay minerals were used for estimation with genetic algorithm based partial least square regression (GA-PLSR). The entire VNIR spectral bands, the bands associated with organic matter and the bands associated with clay minerals were incorporated as comparisons. Root mean square error of prediction, residual prediction deviation, and coefficient of determination (R2) for the model developed using combined bands of organic matter and clay minerals were 329.65mgkg−1, 1.96 and 0.73, which is better than 341.88mgkg−1, 1.89 and 0.71 for the entire VNIR spectral bands, 492.65mgkg−1, 1.31 and 0.40 for the organic matter, and 430.26mgkg−1, 1.50 and 0.54 for the clay minerals. Additionally, in consideration of atmospheric water vapor absorption in field spectra measurement, combined bands of organic matter and absorption around 2200nm were used for estimation and achieved high prediction accuracy with R2 reached 0.640. The results indicate huge potential of soil reflectance spectroscopy in estimating Zn concentrations in soil.