Predicting the Gray Water Footprint and Water Use Efficiency in Farms Using ML Models (Case Study: Lorestan Province)

This study aims to (1) evaluate the Crop Water Productivity (CWP) and gray Water Footprint (WFGray) for key agricultural systems in Lorestan province, Iran, to identify hotspots of inefficiency and pollution, and (2) develop and compare Machine Learning (ML) models for predicting these metrics to aid in management and forecasting. We calculated CWP and WFGray for major crops (including forage corn, wheat, beans, potatoes and vegetables) across multiple meteorological stations in Lorestan province. Furthermore, we employed two ML algorithms including Random Forest (RF) and Support Vector Machine (SVM) to model and predict these indices. Model performance was evaluated using the Mean Absolute Error (MAE). The assessment revealed significant regional and crop-specific disparities. Forage corn was the most efficient and sustainable system (CWP: 2.173 kg/m³, WFGray: 0.05 m³/kg), whereas bean production was the least efficient (CWP: 0.064 kg/m³). Spatially, stations like Azna (potato) demonstrated best practices, while Kuhdasht was identified as a critical area of concern due to low efficiency and high fertilizer pollution. In modeling, the optimal algorithm was target-dependent: RF was superior for predicting CWP (MAE: 0.236), while SVM performed relatively better for the more complex WFGray. This study concludes that addressing water security and agricultural pollution in the region requires tailored, crop-specific interventions and improved farm management practices. Furthermore, while ML model (particularly RF) proves to be a powerful tool for forecasting water productivity, accurately modeling the environmental impact (WFGray) remains a challenge, highlighting the need for more robust data and further research in this domain.