Grapevine stem water potential estimation based on sensor fusion

•Multiple sensors were used to measure vines with various water stress levels.•Thirty-two daily scale factors were extracted based on the sensor measurements.•Seven factors were selected based on their relations to stem water potential (SWP).•A machine learning model for SWP estimation showed high performance levels.•Daily water stress estimation using sensor fusion approach was verified. Estimating vine water status is essential for achieving the desired balance between wine grape quality and yield in viticulture management and irrigation planning. The growing demands of smart farming for extracting meaningful information from field data to support irrigation decisions may be facilitated using field monitoring technology along with advanced modeling algorithms. This research took place in a Vitis Vinifera cv. “Sauvignon Blanc” experimental plot within a commercial vineyard. The main objective was to generate a water stress estimation model for wine grapevines based on data fusion from multiple sensors. Data were collected using sensors from five monitored grapevines, each treated with a different water stress regime. The sensors provided measurements of trunk and fruit growth, leaf temperature, and soil water content (SWC) at depths of 20 and 40 cm. Additionally, meteorological station in the vineyard recorded the temperature, relative humidity, wind speed, and solar radiation. A daily-scale multivariate time series was composed using the data derived from the sensors. Stem water potential (SWP) values were measured for the monitored vines and analyzed against the multivariate time series of factors to define their interrelations and interactions. Finally, a prediction model for SWP estimation was defined using boosted regression trees (BRT) algorithm, with an optimized set of factors. Validation was conducted using comparative statistics between a randomly selected test set and a predicted set of SWP values achieved using the trained BRT model. After processing the data and eliminating most of the factors due to multicollinearity, the model was composed of daily maximum leaf temperature (LT), SWC at noon at 40 cm, tree water deficit (TWD), SWC daily amplitude at 20 cm, water input, vapor pressure deficit, and phenological stages. The highest contributor to the BRT model was maximum LT (44.5%), followed by SWC at noon at 40 cm (16.9%) and TWD (16%). The model performance was estimated using various measures and resulted in a correlation of 0.9 between the