Hyperspectral inversion of heavy metal content in farmland soil under conservation tillage of black soils

Globally, heavy metal (HM) soil pollution is becoming an increasingly serious concern. Heavy metals in soils pose significant environmental and health risks due to their persistence, toxicity, and potential for bioaccumulation. These metals often originate from anthropogenic activities such as industrial emissions, agricultural practices, and improper waste disposal. Once introduced into the soil, they can bind to soil particles, making them difficult to remove, while potentially entering the food chain through plant uptake or water contamination. Rapid access to reliable data on HM viscosity in soils is necessary to efficiently monitor remediated soils. Visible and near-infrared reflectance spectroscopy (350–2500 nm) is an economical and zero-pollution method that can evaluate multiple HM concentrations in soil simultaneously. Black soil is a valuable agricultural resource that helps guarantee food security worldwide and can serve as a soil carbon reservoir, but its protection faces several challenges. Due to long-term high-intensity development and utilization and the severe over-exploitation of groundwater, the arable land in China's black soil area has been degraded. Using hyperspectral inversion of heavy metal content in soil can reduce the destructive sample collection and chemical pollution of soil, better protect black land resources, and steadily restore and improve the basic fertility of black land. Focusing on the black area region of Jilin Province, this study explored the correlation between three HMs, namely copper, zinc, and cadmium, and organic substances, clay minerals, and ferromanganese oxides through an in-depth analysis of soil samples using soil reflectance spectrometry. The spectra were transformed using first-and second-order derivatives, multiple scattering corrections, autoscales, and Savitzky–Golay smoothing. The successive projection algorithm was used to screen characteristic bands (Table S1) to establish the link between HM content in soil and soil spectra. By employing the support vector machine (SVM), random forest (RF), and partial least squares (PLS) models, feature band-based soil HM inversion modeling was established. Moreover, the optimal combinations of spectral transforms and inversion models were also examined. The findings indicate that the RF model (R2 > 0.8, RPIQ > 0) outperformed the SVM and PLS models in anticipating the three soil HMs, thus demonstrating superior accuracy. Understanding the behavior of heavy metals in soils and developing effective management strategies are essential for ensuring sustainable land use and protecting public health. This study contributes to the development of large-scale monitoring systems for the HM content of soil and assessments of HM contamination.