Explainable machine learning links organic-matter stress to microbial controls of sedimentary ammonium accumulation in marsh-type shallow lakes

Shallow lakes with high primary productivity and macrophyte dominance often accumulate organic matter (OM) in sediments, which can promote ammonium (NH₄⁺-N) accumulation and increase the risk of free-ammonia toxicity. However, under high-OM conditions, the microbial functional changes associated with NH₄⁺-N accumulation, as well as their key environmental drivers and threshold windows, remain poorly understood. Here, we investigated Baiyangdian Lake, a representative macrophyte-dominated shallow lake, by integrating high-throughput sequencing, qPCR quantification, and interpretable machine learning (XGBoost–SHAP). The results showed a clear functional divergence under high OM relative to low OM, characterized by enhanced mineralization but suppressed nitrification: ureC increased by 40.63 %, whereas archaeal amoA and bacterial amoA decreased by 94.88 % and 94.30 %, respectively. SHAP further indicated that OM is a core driver of variations in these three functional genes and exhibits threshold-like nonlinear effects: both amoA genes shifted to suppression when OM exceeded 19.62 %/22.27 %, while ureC shifted to promotion when OM exceeded 12.71 % and approached saturation at OM ≈ 17 %. Together, this study reveals distinct threshold regimes in sediment nitrogen functioning associated with NH₄⁺-N buildup, offering quantitative cues to delineate sensitive intervals of internal nitrogen risk and inform targeted management. • OM is a key factor influencing ammonium accumulation. • High OM conditions promoted the expression of the mineralization-related gene ureC . • High OM inhibited nitrification gene expression: archaeal amoA and bacterial amoA. • OM exhibited nonlinear threshold effects on functional genes: ureC and amoA .