**Background:** Apiculture is critically dependent on the availability of specific forage plants, yet non-apicultural invasive species can displace these plants, particularly after disturbances such as wildfires. On Greek Aegean Islands, Thymus capitatus is the most valuable apicultural plant, while Sarcopoterium spinosum is an aggressive, fast-growing shrub that colonizes land traditionally covered by Thymus, negatively impacting honey production. Remote sensing, particularly UAV-based multispectral imaging combined with machine learning, offers a potential tool for detecting and distinguishing these morphologically similar species, but applications in apiculture remain limited.
**Methods:** The study was conducted in three experimental sites on Lemnos Island, Greece (SA1: 25°19′10″E, 39°48′43″N; SA2: 25°18′39″E, 39°48′24″N; SA3: 25°19′5″E, 39°48′50″N). A DJI P4 multispectral drone equipped with five monochrome sensors (BLUE 450 nm, GREEN 560 nm, RED 650 nm, RED EDGE 730 nm, NIR 840 nm) and an RGB sensor was flown at 20 m altitude with 70% front and 60% side overlap, acquiring 3,138 images. Three orthophotos with 0.8 cm ground sampling resolution were created. Ground reference samples were collected and classified into three categories: Thyme, Sarcopoterium, and Others (bare soil, shadows, dry and other vegetation). Six vegetation indices were calculated and added as additional bands. Five supervised classification algorithms—Random Forest (RF), Classification and Regression Trees (CART), Gradient Tree Boost (GTB), Mahalanobis Minimum Distance (MMD), and Support Vector Machine (SVM)—were implemented on the Google Earth Engine (GEE) platform. Training used 70% of samples; 30% was reserved for validation. Accuracy was assessed using Producer's Accuracy (PA), User's Accuracy (UA), F-score, Overall Accuracy (OA), and Kappa coefficient.
**Key Results:** For SA1, RF and GTB classifiers with the combinations ALL BANDS + VEGETATION INDICES, VIS + NIR + RE, and VIS + NIR achieved the highest accuracy (Kappa = 0.97, OA = 0.98). The lowest accuracy was from SVM with VIS (Kappa = 0.42, OA = 0.61). For SA2, RF with VIS + NIR + RE, VIS + NIR, and ALL BANDS + VEGETATION INDICES, and GTB with ALL BANDS + VEGETATION INDICES achieved Kappa = 0.90 and OA = 0.93. SVM with VIS was lowest (Kappa = 0.47, OA = 0.65). For SA3, GTB and RF with ALL BANDS + VEGETATION INDICES, VIS + NIR + RE, and VIS + NIR achieved Kappa = 0.92 and OA = 0.95. SVM with VIS was lowest (Kappa = 0.58, OA = 0.72). Across all study areas, average performance was: GTB and RF (Kappa = 0.91, OA = 0.94), CART (Kappa = 0.88, OA = 0.92), MMD (Kappa = 0.78, OA = 0.83), and SVM (Kappa = 0.70, OA = 0.80). The ALL BANDS + VEGETATION INDICES combination scored highest on average (Kappa = 0.90, OA = 0.93). Area estimates using GTB were: SA1 (Thyme = 0.21 ha, Sarcopoterium = 0.10 ha, Other = 0.07 ha); SA2 (Thyme = 0.13 ha, Sarcopoterium = 0.04 ha, Other = 0.02 ha); SA3 (Thyme = 0.52 ha, Sarcopoterium = 0.26 ha, Other = 0.15 ha).
**Clinical Implications:** This study demonstrates that low-cost UAV multispectral imaging combined with cloud-based machine learning on GEE can accurately detect and distinguish apicultural from non-apicultural plants, achieving near-perfect classification accuracy. For beekeepers, this technology enables mapping of valuable Thymus capitatus forage areas, monitoring encroachment by Sarcopoterium spinosum—especially after wildfires—and making informed decisions about hive placement. For conservation authorities, it provides a tool to monitor and protect Thymus habitats that are legally protected but often destroyed due to misidentification. The approach is particularly valuable in Mediterranean ecosystems where climate change and wildfires intensify competition between species, threatening honeybee populations and the production of high-value thyme honey.