Identification of critical effect factors for prediction of spatial and intra-annual variability of shallow groundwater nitrate in agricultural areas

Shallow groundwater nitrate nitrogen (NO3 --N) concentrations in agricultural areas usually show high spatial and intra-annual variability. It is hard to predict such concentrations due to the complexity of influencing factors (e.g., different forms of N in soil, vadose zone characteristics, and groundwater physiochemical conditions). Here, a large number of groundwater and soil samples were collected monthly over two years at 14 sites to analyze the soil and groundwater physiochemical properties and the stable isotopes of 815N and 818O of groundwater NO3 --N in agricultural areas. Based on field observations, a random forest (RF) model was used to predict the groundwater NO3 --N concentrations and reveal the importance of effect factors. The results show that there are large spatiotemporal variations in NO3 --N, 815N-NO3-, and 818O-NO3-in groundwater. NO3 --N is the major dominant specie of inorganic N in groundwater, and the groundwater NO3 --N concentration in 24 % of the samples failed to meet the drinking water standard of the WHO (10 mg L-1). The RF model satisfactorily predicted groundwater NO3 --N concentrations with R2 of 0.90-0.94, RMSE of 4.54-5.07, and MAE of 2.17-3.38. Groundwater nitrite and ammonium are the most important factors related to NO3 --N consumption and production, respectively, in groundwater. Denitrification and nitrification were further identified by the relationships among 815N-NO3-, 818O-NO3-, and NO3 --N, and by the ranges of 815N-NO3-, 818O-NO3-, temperature, pH, DO, and ORP in groundwater. Soil-soluble organic nitrogen (S-SON) and the depth of groundwater table were identified as vital factors related to N sourcing and leaching. Overall, as a first approach to adopting a RF model for high spatiotemporal-resolution prediction of groundwater NO3 --N variations, the findings of this study enable a better understanding of groundwater N pollution in agricultural areas. Optimizing man-agement of irrigation and N inputs is anticipated to reduce S-SON accumulation and mitigate the threat to groundwa-ter quality in agricultural areas.

Automated summary

Shallow groundwater nitrate nitrogen concentrations in agricultural areas exhibit high spatial and intra-annual variability. A random forest (RF) model was used to predict the concentrations based on field observations and identified groundwater nitrite and ammonium as important factors. The model showed satisfactory prediction accuracy and improved understanding of groundwater pollution.