A Hybrid Physiology–Climate–Machine Learning Framework for High-Resolution Livestock Water Use Estimation

Accurate high-resolution livestock water use (LWU) estimation is vital for sustainable water management amid rising agricultural demands. Traditional static water consumption coefficients (WCC) ignore climatic and physiological variability, leading to spatial-temporal inaccuracies. This study introduces a hybrid framework: livestock-specific multiple linear regression (MLR) equations first estimate county-level WCC from physiological factors (age, body weight, dry matter intake, lactation; R² = 0.80–0.95). These are downscaled to county-level via machine learning linking to local climate (training MSE = 0.24). Separately, ML-guided ratio-based disaggregation downscales state-level consumption-to-withdrawal ratios to county-level using climatic gradients (precipitation, temperature, humidity). The approach enables robust annual county-level LWU predictions (consumption and withdrawal) for 1985–2022 (training MSE = 0.13, testing MSE = 0.57; residuals MAE = -0.07–0.27). Validated against USGS data, it pinpoints LWU hotspots, shows climate explains >80% of LWU variability and physiology >20% of WCC heterogeneity, and cuts estimation uncertainty by up to 40% versus static methods. This supports scalable planning; future work should address water sources, climate resilience, and productivity-environmental trade-offs.