Machine learning revealed geochemical drivers of cadmium availability and methane emissions in hydrologically...

Periodic hydrological fluctuations in paddy soil-water environments induce dynamic redox changes that alter cadmium (Cd) availability and generate substantial methane (CH₄) emissions. However, accurately predicting these two processes and identifying the key soil properties that simultaneously regulate them remains challenging. Here, we collected paddy soil samples from 11 provinces across China to simulate the flooding-drainage of paddy fields. Machine learning (ML) models are then introduced to predict soil available Cd and CH₄ emissions and to identify their key controlling factors. The ML models demonstrated the high predictive performance for soil available Cd (R² = 0.83) and CH₄ emissions (R² = 0.82), respectively. Total Cd (Cd_T), pH, and crystalline iron oxides (Fe_c) were identified as the dominant soil properties controlling available Cd, with a contribution ratio of 89.8%. Meanwhile, total organic carbon (TOC), pH, and Cd_T were found to govern CH₄ emissions, with a contribution ratio of 84.8%. Importantly, the integrated analysis indicated that paddy soils characterized by higher pH and Fe_c coupled with lower Cd_T and TOC are more favorable for the simultaneous mitigation of available Cd and CH₄ emissions. Furthermore, the spatial distribution of soil available Cd and CH₄ emission potential was predicted across a broad geographic region. Taken together, this study provides valuable insights into the prediction and simultaneous mitigation of Cd availability and CH₄ emissions in hydrologically fluctuating paddy soil-water environments.