Toward practical irrigation decision support: Predicting grapevine stem water potential from soil moisture and atmospheric demand

Water scarcity and recurrent drought in California continue to challenge vineyard productivity and fruit quality, increasing the need for irrigation scheduling tools that improve water use efficiency while maintaining vine performance. This proof-of-concept study evaluated whether grapevine stem water potential can be estimated from a practical combination of soil moisture sensing and local weather station observations to support irrigation decisions in a hot viticultural region. The experiment was conducted in a Barbera (Vitis vinifera L.) vineyard in Fresno, California, over three growing seasons (2023-2025) under four irrigation regimes (50%, 75%, and 100% crop evapotranspiration (ETc) replacement, and an over-irrigated treatment). Soil moisture was monitored using spatial frequency domain transmissometry (SFDT) and time-domain reflectometry (TDR) sensors installed at 20, 37.5, and 75 cm, with an additional 150 cm TDR sensor installed to characterize deeper profile dynamics. Weather variables were obtained from the nearest California Irrigation Management Information System (CIMIS) station. Midday stem water potential (Ψstem) was measured using a pressure chamber and used to train and evaluate predictive models intended to provide continuous, sensor-based inference of vine water status. Permutation feature importance (PFI) analyses across algorithms consistently identified deep soil moisture (75 cm) and air temperature as dominant predictors, with relative humidity providing additional explanatory value. Among candidate algorithms, partial least squares (PLS) regression produced the most stable performance when including three growing seasons’ data, achieving R2 = 0.629 using a two-variable input set (soil moisture at 75 cm + air temperature) and R2 = 0.639 using a three-variable set (soil moisture at 75 cm + air temperature + relative humidity) for Ψstem prediction. Overall, the results demonstrated the feasibility of estimating grapevine Ψstem from a minimal, adoption-oriented sensor package, supporting scalable irrigation decision for vineyards in hot and water-limited regions.