Quantitative estimation models of relationship between aeolian sand and soil geochemical composition based on machine learning algorithm

Abstract

Quantitative estimation models are key to studying the relationship between aeolian sand and soil geochemical composition, as well as for assessing the extent of their influence. However, few studies have developed quantitative models to evaluate how aeolian sand interferes with soil chemical element content, and current approaches primarily rely on provenance analysis through isotopes or major elements. This study combines machine learning regression models with curve regression models to explore the effectiveness of different approaches in constructing quantitative estimation model for aeolian sand interference on soil chemical composition. Our study demonstrated that, of the five different models applied in this study, the Support Vector Regression (SVR) and Random Forest Regression (RFR) models produced the most reliable results. The overall dataset of desert and soil samples conformed to a third-order polynomial fitting model in mathematical relationships, yielding a relatively stable explicit estimation model that quantitatively characterizes the extent of aeolian sand interference on soil chemical composition. This indicates that the Environmental Kuznets Curve (EKC) theory can be applied to geochemical data, offering significant scientific and practical value for exploring the driving mechanisms of surface elements in desert and surrounding areas.