Spatial variability of selected metals using auxiliary variables in agricultural soils

• Published in: Catena (Giessen) 2019-03, Vol.174, p.499-513
• Main Authors: Song, Ying-Qiang, Zhu, A-Xing, Cui, Xue-Sen, Liu, Yi-Lun, Hu, Yue-Ming, Li, Bo
• Format: Article
• Language: English
• Subjects: Agricultural soils; Artificial neural network-ordinary kriging; Geostatistics; Heavy metals; Hyperspectral remote sensing; Spatial variability
• ISSN: 0341-8162; 1872-6887
• DOI: 10.1016/j.catena.2018.11.030

Spatial estimation of the content of soil heavy metals in agriculture is critical for monitoring farmland contamination and ensuring sustainable eco-agriculture. In this study, a hyperspectral image (115 bands) and 17 environmental parameters were used as auxiliary variables to predict the concentrations of As, Cd, Cr, Pb and Zn in agricultural soils. The study also combined hybrid geostatistical methods, such as regression-ordinary kriging (ROK) and back propagation neural network-ordinary kriging (BPNNOK). To this end, a total of 100 topsoil (0–20 cm) samples were collected in Zengcheng in the Pearl River Delta (PRD), China. The results showed that the BPNNOK with input principal components (PCs) from auxiliary variables had the best prediction performance for As (RMSE = 5.02 mg kg−1 and R2 = 64.21%), Cd (RMSE = 0.014 mg kg−1 and R2 = 61.82%), Cr (RMSE = 3.78 mg kg−1 and R2 = 54.39%), Pb (RMSE = 4.09 mg kg−1 and R2 = 60.92%), and Zn (RMSE = 2.86 mg kg−1 and R2 = 59.31%) in comparison with simple kriging (SK), ordinary kriging (OK) and ROK. The effects of hyperspectral remote sensing data exhibited the best sensitivity and responses for all soil heavy metals by the multiple linear regression (MLR) and back propagation neural network (BPNN) models, which were obviously superior to environmental parameters. It was concluded that information from hyperspectral remote sensing data is promising, and that the efficient use of auxiliary variables for monitoring sustainable pollution prevention and control of soil heavy metals, which improve mapping quality of soil heavy metals, is the primary process in non-point source variations of soil heavy metals. [Display omitted] •Environmental and hyperspectral data are used to predict soil heavy metal contents.•The BPNNOK has the best prediction accuracy of content of soil heavy metals.•The spatial distribution of As, Cd, Cr, Pb and Zn concentrations is mapped.•Hyperspectral variables have the better responses for variability of soil heavy metals.