Characterization and modeling of exogenous selenite aging in soils using machine learning and traditional data analysis

Abstract

Understanding and predicting the aging process of exogenous selenium (Se) in soil is crucial for Se biofortification. However, the long-term aging of selenite in various soils has rarely been reported, and the key factors influencing this aging process remain unclear. Our study involved nineteen typical Chinese soils with varying physiochemical properties, all spiked with potassium selenite (1.0 mg kg⁻¹ Se) and incubated for 180 days. Soil available Se extracted using a 0.1 M K₂HPO₄-KH₂PO₄ solution was measured through the whole aging process. The average available Se% (the percentage of available Se in aged soils to total added Se) of all soils decreased from 55.4 % on the day 1 to 32.6 % on day 60, remaining stable thereafter. Pseudo-second-order equation provided the optimal fit (R² > 0.989, P < 0.01) for characterizing the dynamic process of selenite aging in soil, indicating that chemisorption, rather than internal diffusion, controlled the main rate-limiting step in the selenite aging process. Both machine learning and traditional correlation analysis indicated aging time was the most critical feature and the key soil property that contributed to available Se was pH. Empirical models incorporating soil properties and aging time were developed to predict changes of available Se in soil during aging under aerobic conditions. The reliability of the prediction model was further validated using data collected from previous studies. The developed aging model could potentially be used to scale biofortification data of Se generated from different soils under different aging times.