Machine Learning Based Prediction of Riverine Nitrate Flux for Indian Rivers from Agricultural Landscapes

India is predominantly an agrarian country, with ~46% of its population dependent on agriculture. Meeting rising food demand often relies on intensive synthetic fertiliser use, which boosts crop yields but causes environmental impacts. Nutrient runoff from fields is a significant source of riverine nitrate pollution, contributing to the gradual degradation of water quality. To explore the link between agricultural intensification and riverine nitrate, this study applies multiple machine learning models, including Linear, Polynomial, Decision Tree, Random Forest, Support Vector, XGBoost, Neural Network, and Multi-layer Perceptron, to predict nitrate concentrations using hydrological, socioeconomic, and agricultural nitrogen input variables across major river basins and identify key controlling factors. The best-performing model achieved an R² of 0.57. Results show significant spatial and temporal variation of riverine nitrate flux across major river basins between 1966 and 2017. At the national scale, the average nitrate flux declined from 535.6 kg/km²/year to 443.8 kg/km²/year, reflecting ~17.1% an overall reduction. The decline in nitrate is primarily attributed to reduced precipitation and an increase in consecutive dry days, as shown by the overall trend analysis. Analysis suggests that lower rainfall reduces surface runoff, thereby limiting the transport of nutrients to rivers. Despite an overall decline in nitrate, larger basins such as the Brahmaputra and the Ganga maintained high concentrations due to their high discharge, greater catchment area, and intensive agriculture. Basin-wise correlation analysis further shows a positive correlation between precipitation, discharge, and nitrate export, confirming that these hydrological variables are the dominant controls, as they enhance runoff. This increased runoff strengthens hydrological connectivity between agricultural fields and river channels, thereby mobilising nitrogen from soils and fertilisers into surface water. Furthermore, our Pearson correlation analysis indicates that net anthropogenic nitrogen inputs contribute more strongly to soil nitrogen build-up, groundwater contamination, and atmospheric emissions, rather than directly influencing riverine nitrate through runoff pathways. Overall, the major drivers of riverine nitrate dynamics across Indian basins are precipitation and discharge, with agricultural practices and basin hydrology acting as secondary influences.

 

Keywords: Machine Learning, Riverine Nitrate, Agricultural Landscapes, Environmental Impact