HYPERSPECTRAL INVERSION OF SOIL ORGANIC MATTER UNDER DIFFERENTIAL VEGETATION COVER SCENARIO BASED ON CARS-RF

The development of coal resources has caused serious environmental problems such as soil degradation and crop damage even as promoting economic development.Accurate and efficient determination of soil organic matter content is essential to contribute to controlling soil degradation and improving land fertility in mine-grain mixed zones.Therefore, a typical ore-grain composite area was selected as the research site.The collected soil spectral data were filtered and denoised, and the secondorder differential transformations were used to enhance the spectral characteristics.The competitive adaptive reweighted sampling (CARS) algorithm was used to screen the sensitive bands, and the inversion model of soil organic matter content was constructed based on partial least squares regression, random forest and XGBoost algorithms.The results show that the correlation between soil spectra and organic matter content can be improved by using SG denoising combined with second-order differential transformation to extract effective information of soil spectra.The CARS algorithm extracts important features, removes redundant spectral information, and improves modeling accuracy.In terms of inversion, the RF model built based on CARS feature extraction had the highest accuracy, and the prediction accuracy (R) reached 0.95 in the low vegetation coverage area.The second was XGBoost, PLSR model inversion accuracy is the lowest, in high and low vegetation coverage area R is 0.79 and 0.78, respectively.The results fully demonstrate the effectiveness and feasibility of machine learning methods for retrieving hyperspectral soil organic matter content.This research can provide theoretical and scientific basis for the rapid monitoring of large-scale soil organic matter.