Predictive Understanding of the Transformation of Ferrihydrite with Machine Learning: Key Factors, Mechanisms, and Implications for the Immobilization of Pollutants and Soil Organic Carbon

Ferrihydrite (Fh), which plays a critical role in element cycling and contaminant sequestration in the environment, is susceptible to transformation. However, understanding and predicting its transformation process are challenging due to the multitude of influencing factors. Herein, we developed six machine learning (ML) models and selected the best performance model (Random Forest) to predict the transformation and fate of Fh. Key influencing factors for Fh transformation, including the C/Fe ratio and Fe(II)/Fe(III) ratio, were identified. ML further implied a distinct transformation route between biological-induced and chemical-induced Fh transformation; that is, the former one resulted primarily in lepidocrocite (Lp), while the latter tended to more readily yield goethite (Gt) or magnetite (Mt) via intermediate phases. Linking with the fixation capacity of different Fe minerals, our ML model indicated the risk of emerging pollutants (e.g., perfluoroalkyl and polyfluoroalkyl substances, PFAS) and global warming during the Fh transformation under different environmental conditions. The predicted release risk of PFAS and soil organic carbon could reach up to 67.8% in paddy soil and 181.6 Pg globally, respectively. Through a data-driven approach, this study provides new insights into the transformation, fate, and implications of poorly crystalline iron minerals, highlighting their roles in pollutant turnover and carbon cycling.