Prediction of metal nanoparticle interactions with soil properties: Machine learning insights into soil health dynamics

Metal nanoparticles (MNPs) offer great potential to enable precision and sustainable agriculture. However, a comprehensive understanding of the interactions between multiple MNPs and soil properties, including impacts on overall soil health, remains elusive. Here, 4 different interpretable machine learning models were employed to systematically analyze the interactive effects of 7 soil physicochemical properties, 3 MNP properties, and 3 external factors on soil health indicators including pH, soil microbial biomass, Shannon index, and enzyme activities. We identified soil cation exchange capacity (SCEC), soil organic matter (SOM), soil clay (SC), and exposure duration (ED) as pivotal factors influencing the effects of MNPs on soil health. Notably, the adsorption and fixation of metal ions by soil significantly modulate MNP toxicity over time, underscoring the importance of long-term exposure in soil health research. This study predicts the impact of MNPs on soil health indicators across 12 United States Department of Agriculture (USDA)-classified soil orders from a global perspective. The impact of MNPs on soil health is highly dependent on soil properties, with effects varying across different soil types globally. Entisols, the most abundant and widespread soil, characterized by low water holding capacity and SOM content, showed heightened sensitivity in pH to MNP exposure, with pH changes ranging from 2.6% to 11.5%. In contrast, changes in pH for other soil types were between −2% and 4%. Mollisols and Inceptisols, which represent important cultivated lands in Europe, the United States, Canada, and China, exhibited significant sensitivity in microbial biomass and diversity to long-term MNPs exposure. Over a 365 day exposure period, the microbial biomass changes of Mollisols and Inceptisols ranged from −268% to −60%, while for a 30 day exposure period, changes in microbial biomass were between −20% and −4%. By conducting global predictions across diverse soil types, this research enabled an assessment of potential soil health risks at a global scale, highlighting high-risk regions and countries for targeted analysis to support more science-based, data-driven environmental management and sustainable agricultural practices.