Quantifying the contributions of natural and anthropogenic factors to nitrogen exceedance in shallow groundwater in agricultural areas using a hybrid machine learning approach.

Groundwater nitrogen pollution in intensively farmed regions threatens water safety. Prevailing studies often attribute exceedances solely to anthropogenic sources, overlooking natural background levels. Utilizing data from 706 sites (2011-2020) in China's Sanjiang Plain and a hybrid machine learning model, this study quantifies the predictive influence of natural versus anthropogenic factors. Results reveal a fundamental divergence: nitrate exceedances show strong association with human activities (60-72% of SHAP-derived relative influence, mainly fertilizer), whereas ammonium exceedances are largely linked to natural geological background (67-80% of the model-derived contribution). These findings underscore the necessity of considering geogenic sources. They provide a scientific basis for implementing targeted pollution risk zoning and advancing sustainable agricultural management.