A machine learning framework for California vineyard water status monitoring using sUAS Imagery and short-term meteorological data

Efficient irrigation management is essential for sustainable crop production under increasing temperatures and tightening water supplies. In vineyards, water status significantly influences vine growth, yield, and fruit quality, and deficit irrigation is often used to impose controlled stress while avoiding damaging levels of water limitation. This creates a practical need for routine, field-scale monitoring of vine water status. In this study, we developed an operational machine-learning framework to estimate grapevine leaf water potential (Ψ_leaf) by integrating daytime sUAS thermal imagery with short-term local meteorological information. When all candidate predictors were included, the trained eXtreme Gradient Boosting (XGB) model achieved R² = 0.71, RMSE = 0.14 MPa, and bias = − 0.06 MPa on the independent test dataset. A simplified XGB model using a compact predictor set–maximum air temperature in the 24 h prior to flight, air temperature at flight time, their difference, and canopy temperature – achieved R² = 0.63, RMSE = 0.16 MPa, and bias = − 0.06 MPa, with performance not significantly different from the full model at α = 0.05. This reduced-feature formulation supports vineyard-scale Ψ_leaf estimation and mapping while maintaining strong predictive skill and low computational burden. Our research highlights the potential for broader applicability, particularly for monitoring rapidly developing hot and dry conditions and supporting adaptive water resource management.