Machine learning prediction and driver analysis of nitrate leaching in field and vegetable cropping systems of Northwest China

Intensive agriculture in arid and semi-arid Northwest China leads to severe nitrate leaching, posing a serious threat to groundwater quality and ecosystem health. To investigate its driving factors, we applied four machine learning algorithms—Support Vector Machine (SVM), Random Forest (RF), Extreme Gradient Boosting (XGBoost), and Convolutional Neural Network (CNN)—to a synthesis of 43 published studies from the region. Results revealed that vegetable systems (44.75 kg ha −1 season −1 ) leached 67% more nitrate than field crop systems (26.76 kg ha −1 season −1 ). Among the models, XGBoost achieved the best performance (R 2 ≥ 0.75). SHAP analysis further identified irrigation and nitrogen input as the primary drivers, while soil organic matter played a key modulating role, particularly in vegetable systems. Nitrate leaching increased significantly when irrigation rates exceeded 300 mm or when nitrogen inputs surpassed 200 kg ha −1 for crops and 680 kg ha −1 for vegetables. These findings provide an empirical, data-driven foundation for agroecological risk assessment and support the design of precision nitrogen management practices in water-limited farming systems. • Vegetable systems leach more nitrate than crops due to higher N and irrigation inputs. • XGBoost performed the best for crops, while CNN and SVM outperformed in vegetables. • Nitrate Leaching rises sharply above 300 mm irrigation or 200/680 kg N ha −1 . • SOM is the dominant regulator of nitrate leaching in vegetable systems. • Thresholds guide leaching reduction via N/irrigation limits and SOM enhancement.