The combination of ground hyperspectral and satellite multispectral with LSO-RF algorithm for enhanced inversion accuracy of heavy metal content in soil in coal mining areas

Abstract

Although multispectral remote sensing technology has been widely used in soil heavy metal monitoring worldwide, the limitation of spectral resolution often affects the monitoring accuracy. In contrast, ground hyperspectral data provide more detailed information in identifying and quantifying heavy metals in soil due to its narrow-band characteristics, but its small coverage area and high requirements for related equipment and technology limit the application of large-scale monitoring. Therefore, based on the ground hyperspectral data, this study establishes a relationship between the spectral response function and the satellite multispectral remote sensing data, so that the multispectral band contains the weighted average information of multiple bands in the hyperspectral data. In this way, the detailed spectral information of the hyperspectral data can be retained, and the satellite multispectral data processing tools and technologies can be utilized. A regional high precision soil heavy metal inversion framework was formed. By sampling two mining areas in China, one for experiment and the other for verification, an improved Random forest model (LSO-RF) based on spectral optimizer was established, and the inversion model was applied to multi-spectral satellite remote sensing images to predict the spatial distribution of heavy metals in the whole region. This method significantly improves the inversion accuracy on the hyperspectral improved multispectral data, and the prediction accuracy of the LSO-RF model is significantly better than that of the traditional model. Taking the improved Landsat 8 data as an example, the R² of the LSO-RF model is 0.95 in the prediction of Zn in the Shenmu research area. Compared with SVR, PLSR and traditional RF models, the results are 0.32, 0.30 and 0.10 respectively, and the other three heavy metal elements are also significantly improved. In the Ejin Horuo Banner research area, the LSO-RF model also showed significant improvement in the prediction of heavy metal elements, and the R² could reach 0.97. This study not only confirms the feasibility of improving multispectral remote sensing data based on hyperspectral ground data, but also provides a high-precision inversion framework for the accurate monitoring of soil heavy metals.