Low agricultural water productivity, together with the growing impacts of climate change, has increased pressure on water resources. Continuous monitoring of plant water stress at field and orchard scales can help alleviate this pressure. In this study, Machine Learning (ML) models were combined with remote sensing data to directly estimate the Water Deficit Index (WDI). Two scenarios were developed for WDI estimation in pistachio orchards of the Rafsanjan Plain. In the first scenario, Sentinel-2 spectral bands were used as input data, while WDI images derived from the METRIC algorithm were used as the target output. In the second scenario, the input data consisted of a combination of Sentinel-2 spectral bands and the thermal band of Landsat-8, while the output remained the same as in the first scenario. The novelty of this study lies in the development of a methodological framework for high-spatial-resolution mapping of WDI through the integration of ML models with Sentinel-2 and Landsat-8 satellite imagery. The results showed that, in the second scenario, incorporating thermal information improved the accuracy of WDI estimation across different plant growth stages. Among the evaluated models, the Artificial Neural Network (ANN) achieved the best performance, with DC and KGE values of 0.718 and 0.818, respectively, outperforming the SVR and RF models.