Climate change-driven droughts increasingly threaten Mediterranean forests. Early detection is crucial for mitigating long-term impacts; yet, conventional methods are limited in spatial and temporal coverage. Remote sensing offers a large-scale solution, but its application at the individual-tree level remains limited, particularly in mixed-species forests. We combined a controlled rainfall exclusion experiment with drone-based hyperspectral imaging and machine learning to classify drought stress at the individual-tree level in a semi-arid Mediterranean forest (Yishi Forest, Israel). Six 0.05-ha plots with five co-occurring tree species were monitored over two hydrological years. Hyperspectral data (274 bands, 400-1000 nm) were used as is and after synthetically simulating Planet, VEN mu S, and Sentinel-2 bands in three machine learning classification models. Results show that rainfall was reduced by nearly half in treated plots. Standard physiological metrics-leaf water potential, carbon assimilation, and transpiration-showed limited treatment sensitivity across most species and seasons, whereas hyperspectral-driven machine learning classification models accurately distinguished between drought-treated and control stands. Logistic Regression (LR) outperformed Support Vector Machines (SVM) and Random Forest (RF), reaching an accuracy of 0.85, a recall of 0.94, and an F1 score of 0.83 in classifying treated stands on a held-out test set. High performance persisted after reducing input to 21 bands. Simulated satellite spectral data showed that SVM performed best using VEN mu S bands (accuracy = 0.74, F1 = 0.73). When applied to real VEN mu S imagery from three independent forest sites, the model identified areas of high drought risk one to two years before visible canopy decline. The presented approach offers a scalable and transferable tool for real-time forest drought monitoring, supporting early warning systems amid growing climate pressures.