ABSTRACT Heavy metal contamination in agricultural soils threatens ecosystem stability and food safety. Rapid and accurate estimation of arsenic (As), cadmium (Cd), and lead (Pb) is therefore essential for environmental protection and soil remediation. Near‐field sensing technologies provide a fast and cost‐efficient alternative to laboratory analysis, yet single‐spectrum approaches often suffer from limited information coverage and reduced prediction accuracy. This study investigates the capability of Ultraviolet (UV), Visible–Near Infrared (Vis–NIR), and portable X‐ray fluorescence (pXRF) spectral data—individually and in combination—for predicting soil heavy metal concentrations using 110 farmland samples from Chabuqiaer Xibo Autonomous County. Seven preprocessing methods were employed to optimize spectral data quality, in conjunction with principal component analysis (PCA) for feature dimension reduction. Three fusion strategies and four machine learning models were employed for modeling and prediction. The results showed that the Parallel Concatenation Fusion of Multi‐Sensor Data Based on Self‐Attention Mechanism (PCFMS‐SAM) fusion strategy performed best in modeling the three heavy metal elements, with the best model for As being Random Forests (RF) ( R 2 = 0.92), for Cd being VPPSO‐SVM ( R 2 = 0.79), and for Pb being RF‐XGB ( R 2 = 0.84). The consistency correlation coefficients (LCCC) of all optimal models were above 0.8, reflecting a strong alignment between model outputs and observed values. The integration of multi‐source spectral data resulted in a considerable improvement in prediction accuracy over single‐sensor models, underscoring its potential for rapid assessment of soil heavy metals.
Gao et al. (Sun,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: