ABSTRACT
Hyperspectral analysis is useful in monitoring and evaluating soil heavy metal pollution. The difficulty in estimation of heavy metals by hyperspectral analysis is to extract the spectral features related to soil heavy metal concentration which may be interfered by signals resulting from other soil components. In this study, we established a comprehensive framework for prediction of Cr and Ni from hyperspectral data combined with soil Al-Fe minerals using advanced machine learning methods. The Al-Fe minerals were measured in laboratory and predicted from the spectral data, respectively. The results show that (1) from the original hyperspectral data, the model performances for Cr (R2 = 0.61, RMSE = 16.96) and Ni (R2 = 0.34, RMSE = 6.92) are not satisfactory. (2) With the incorporation of the measured Al-Fe minerals as the predictors, the optimal machine learning method for prediction of Cr and Ni is XgBoost, and the model performances are improved obviously (Cr: R2 = 0.85, RMSE = 10.67; Ni: R2 = 0.71, RMSE = 4.55). (3) With the incorporation of the Al-Fe minerals predicted from the spectral data, the optimal machine learning method for prediction of Cr and Ni is GdBoost, and the model performances are acceptable (Cr: R2 = 0.74, RMSE = 13.93; Ni: R2 = 0.68, RMSE = 4.79). This study provides a preliminary analysis for mapping soil heavy metal content and monitoring heavy metal pollution over a large spatial extent.
Acknowledgements
This research was supported by the Key-Area Research and Development Program of Guangdong Province, China (2020B1111370001), National Key Research and Development Program of China (2022YFF1303104). The authors also thank the chemical pollution control and ecological restoration teams for their support.
Data availability statement
The participants in this study did not give written consent for their data to be shared publicly, so due to the sensitive nature of the research supporting data are not available.