A method for heavy metal estimation in mining regions based on SMA-PCC-RF and reflectance spectroscopy.

• SMA was used for the first time for the selection of hyperspectral wavebands. • A novel algorithm of SMA-PCC-RF is proposed to invert heavy metal concentrations. • The algorithm was applied to inversion of heavy metal concentrations in mining area. • Spectral transformations can enhance heavy metals and spectral curves correlation. • SMA-PCC-RF algorithms offers the better inversion accuracy for all heavy metals. Heavy metal contamination has long been a concern of intense research within the field of environmental protection. The exacerbation of heavy metal pollution due to mineral resource over-exploitation and the implementation of agricultural modernization has further emphasized the need for rapid monitoring. Although traditional geochemical survey methods have been deemed reliable, they lack the ability to conduct non-destructive and rapid surveys of large areas due to their high cost, low real-time capability, and cumbersome operations. We collected 120 soil samples in the field using Xin'an County, Henan Province as the research region, and obtained hyperspectral curves and contents of the samples using the spectrometer and chemical analysis. After preprocessing the spectral curves, we used the slime mold algorithm (SMA) to preselect feature wavebands, and then the mathematically transformed method was used to improve their correlation. Next, we calculated the correlation coefficients of these wavebands with six heavy metals, and the final modeled wavebands were obtained through precise feature selection based on the criterion that the correlation coefficient's absolute value exceeded 0.298. The inversion model was also established by using adaptive boosting (AdaBoost), gradient boosted decision tree (GBDT), random forest (RF), and partial least squares (PLS). The results indicated that SMA-PCC can effectively downscale the high-dimensional hyperspectral data and obtain the feature wavebands with a high contribution to the modeling. We also observed that the mathematical transformation method improved the relevance between heavy metal elements and spectra. Among the four models, RF showed better overall inversion accuracy for all heavy metals (i.e., Zn: R AdaBoost 2 = 0.85 , R GBDT 2 = 0.56 , R RF 2 = 0.89 , R PLS 2 = 0.57). This paper showcases that the algorithm presented can provide meaningful technical guidance for the large-scale investigation and assessment of heavy metal levels in soil. [ABSTRACT FROM AUTHOR]