Advanced monitoring of almond orchard water status using machine learning and remote sensing

• A random forest model predicted stem water potential in almond orchards with an NSE of 0.91. • Integrated satellite imagery, meteorological, and soil data for improved SWP monitoring. • Identified air temperature, evaporative fraction, and wind speed as key predictors of SWP. • Generated spatial maps of SWP to support optimized precision irrigation management in orchards. Water management is essential for optimizing yields in almond orchards, particularly in regions with limited water resources. This work integrates different sources of information (satellite imagery, meteorological data, soil data, and machine learning techniques) to monitor and predict the spatial variability of stem water potential (SWP) in an almond orchard. A random forest (RF) model was developed using weekly SWP measurements collected between 2019 and 2021 at 24 monitoring locations. Predictor variables included meteorological variables (air temperature, solar radiation, wind speed, relative humidity, vapor pressure deficit), soil parameters (soil bulk density and soil water content), and vegetation indices derived from satellite imagery (normalized difference vegetation index) and the evaporation fraction were used. The RF model demonstrated high accuracy in predicting SWP, with a Nash-Sutcliffe Efficiency (NSE) of 0.91 and a root mean square error (RMSE) of 0.17 MPa, as evidenced by the goodness-of-fit evaluation. The cumulative probability plot indicated that 78.2 % of the NSE values fall within the "very good" range and 21.8 % within the "good" range, underscoring the model's reliability. It was found that the key variables for SWP prediction are air temperature, evaporation fraction, wind speed, and solar radiation. Spatial maps generated with high-resolution aerial imagery by the model revealed significant within-field variability, particularly during critical growth stages such as the hull split period in July. During this period, the orchard was managed using deficit irrigation strategies, a common practice to mitigate hull rot and optimize water use, which resulted in lower (more negative) SWP values indicating water stress. These maps highlighted areas of the orchard experiencing greater water stress, guiding more precise irrigation management. This study highlights the importance of integrating high-resolution remote sensing data with machine learning algorithms to enhance water management practices in almond orchards. The findings suggest that the spatial and temporal predictions of SWP using the RF model can support precise irrigation scheduling, leading to improved water use efficiency and sustainability in almond production.