Deciphering Heavy Metals Adsorption on Soil by Physicochemical Property Diversity Using Machine Learning Method

Abstract

The soil physicochemical properties play a pivotal role in the geochemical processes of heavy metals (HMs) within the soil environment. However, the precise soil property that most significantly impacts HMs binding capacity remains elusive. To address this knowledge gap, we collected 22 types of soil samples from diverse regions across China. Utilizing these samples, we developed 12 machine learning models to predict the adsorption of HMs by soil, leveraging 2112 experimental data points. Using the most effective model, we identified the master control factors of soil properties that dictate the soil adsorption capacity of HMs, including Cr, Pb, Cd, and As. Our findings indicate that reactive metal oxides and soil organic matter contents play a critical role in determining the adsorption capacity of oxyanions like Cr and As. Meanwhile, soil pH, soil particle size, and soil organic matter contents account for the majority of differences in cation adsorption capacity, specifically for Pb and Cd. Notably, soil bulk density and exchangeable potassium content emerged as crucial factors in predicting the relative adsorption capacity of HMs, an area that has been understudied. This study deciphering light on a complex adsorption process that has significant implications for environmental health and pollution control efforts.

Keywords: Soil, Heavy metals (HMs), Machine learning (ML), Feature analysis, adsorption