Influencing Factors of Selenium Spatial Distribution Based on Machine Learning: A Case Study of Surface Soil in Lianzhou City, Guangdong Province

Selenium is an essential trace element for human body, which plays an important role in preventing cancer and enhancing immunity. However, the distribution of selenium in soil is affected by many factors such as soil parent material, soil physicochemical properties and human activities. It is of great significance to accurately predict the spatial distribution of selenium in soil and study its influencing factors for rational development of selenium rich resources and protection of human health. In this paper, 3 009 topsoil samples were collected in Lianzhou City, Guangdong Province. The soil organic carbon and heavy metal contents were determined; and the topographic, geochemical, ecological and human activity factors were combined. Machine learning algorithm (RF, ANN, SVM and LightGBM) and traditional spatial interpolation method (OK) were used to predict the spatial distribution of selenium. The results show that the machine learning model is significantly superior to the traditional methods in terms of prediction accuracy, among which the LightGBM model has the highest accuracy (R2=0.707, RMSE=0.109), while the OK model has poor prediction effect (R2=0.374, RMSE=0.159), and the RF model is more prominent in the details of spatial prediction. Finally, through spatial autocorrelation analysis and LISA cluster analysis, the key factors affecting selenium distribution were revealed: Among the geochemical elements, Mo, Corg and pH are significantly correlated with selenium and their contribution value is relatively high. There was a positive correlation between elevation and Se content, and there were 82.75% HH cluster points at 200-800 m elevation. In terms of ecology, precipitation and vegetation coverage were positively correlated with Se content, and the Se content in different soil types was significantly different. Land use has significant effect on selenium distribution in human activities. In this study, the distribution rule and influencing factors of selenium in soil in Lianzhou City were effectively revealed by machine learning method, which provided a strong scientific basis for the rational development and utilization of selenium rich resources.