Quantifying the Impact of Factors on Soil Available Arsenic Using Machine Learning

Abstract

Clarifying the contributions of various factors to available arsenic in farmland soils holds significant research value for understanding soil arsenic activity. This study utilized 442 datasets encompassing the total arsenic, available arsenic, and physicochemical properties of farmland (paddy and dryland) soils. To quantify the contributions of different factors to soil available arsenic, random forest, gradient-boosted machine, and multiple linear regression models were employed. The results showed that random forest had the strongest predictive ability for available arsenic in farmland soil. In dryland soils, the factors influencing soil available arsenic, in order of importance, were as follows: total soil arsenic content > cation exchange capacity > pH > organic matter. In contrast, in paddy field soils, the order of importance was total soil arsenic content > cation exchange capacity > organic matter > pH. Therefore, there were differences in the importance of pH and organic matter between paddy fields and drylands. In arsenic contamination management practices, it is crucial to prioritize the regulation of total soil arsenic and cation exchange capacity to reduce arsenic activity in farmland soils. The findings offer valuable guidance for treating arsenic-contaminated farmland soils using passivation or activation measures.

Keywords: Arsenic availability, Farmland soil, Soil property, Machine learning