Identifying the driving factors of soil nitrate nitrogen via MLs-GIS framework in an intensive plain agricultural area, China

Abstract

Excessive accumulation of nitrate nitrogen (SNN) in soil is a global ecological issue, with adverse effects on terrestrial and aquatic ecosystems. Here, a total of 251 soil samples were collected from a typical agricultural planting area on the North China Plain, China. On the basis of the analysis of multisource data (N fertilizer input, physicochemical properties of soil, distance factors, remote sensing factors, etc.), a framework combining machine learning models with spatial analysis technology in GIS (MLs-GIS) was built to clarify the fit of the spatial distribution and identify key factors influencing SNN. The results revealed that the content of SNN at different sites varied significantly (from 0.02 to 58.85 mg/kg). According to the performance of the ML models, namely, the random forest (RF), gradient boosting decision tree (GBDT), light gradient boosting machine (LightGBM) and extreme gradient boosting (XGBoost) methods, soil pH and chemical N fertilizer input were the main factors influencing SNN. The RF model exhibited the best performance, with the highest prediction accuracy (R² value of 0.64), and warrants increased attention in the management of agricultural nitrogen pollution. In summary, the MLs-GIS framework is efficient at identifying the key influencing factors and the spatial distribution of SNN.