Data-Driven Insights into Nanomaterial Impacts on Soil Nitrogen Processes

The widespread application of nanomaterials (NMs) in the environment has raised growing concerns about their potential effects on soil nitrogen cycling, particularly on microbially mediated nitrification and denitrification processes. Given the complexity of NM interactions with soil components and microbial communities, in this study we employed meta-analysis and machine learning (ML) to first establish the overall effect of NMs on the nitrification and denitrification processes, and subsequently integrate NM-induced effects on nitrification and denitrification, considering the physicochemical properties of NMs, soil characteristics, and environment metrics. Three algorithms (XGBoost, SVM, and RF) were trained and evaluated to predict NM-induced changes in nitrification and denitrification rates. The XGBoost model demonstrates optimal performance in predicting the effects of NMs on nitrification and denitrification rates (nitrification: <i>R</i>² = 0.7416; denitrification: <i>R</i>² = 0.7316). Furthermore, to ensure interpretability and mechanistic insight, different model interpreters at global and individual levels were employed. It suggested that NM concentration is the domain factor for both processes, while secondary factors differed between processes: soil pH exerted stronger control over nitrification, whereas NM size had greater influence on denitrification. The Partial Dependence Plots (PDP) analysis revealed significant interactive effects among factors such as concentration, exposure time, and soil texture, which collectively determine the nitrification and denitrification rates. Principal Component Analysis (PCA) indicated that concentration, total nitrogen, and pH explained the majority of variance in nitrification, while concentration, size, and exposure duration did so for denitrification. The cumulative variance explained by the first three principal components exceeded 70% for both processes. Furthermore, local interpretation with varying carbon concentrations and exposure durations accurately predicted their trend changes. This study established a predictable and interpretable ML framework, identified key drivers of NM impacts on soil nitrogen cycling, and provides theoretical basis for ecological risk assessment of NMs.