Cost-effective monitoring of a rare Pannonian steppe plant using drone imagery

Efficient monitoring of rare and protected plant species is essential for biodiversity conservation. This study evaluates a cost-effective and non-invasive UAV-based method using RGB vegetation indices derived from the red, green and blue image channels to enhance vegetation colour contrast to detect Sedum urvillei subsp. hillebrandtii , a protected Pannonian subendemic species. A fixed-wing drone equipped with an RGB camera surveyed an 8.5-ha open sandy grassland within the Kiskunság National Park, Hungary, during the species' peak flowering period. Three RGB-based vegetation indices – Triangular Greenness Index (TGI), Colour Index of Vegetation Extraction (CIVE), and Red-Green Ratio (RGR) – were applied to an orthophoto mosaic to classify flowering individuals. Negative masking using RGR significantly improved the classification accuracy of both TGI and CIVE. Statistical analysis revealed high overall accuracy (up to 99.3%) and specificity (>99%) for the CIVE–RGR and TGI–RGR combinations, while sensitivity increased from 79.1% to 89.0% after applying RGR-based masking. Automated thresholding using the Otsu method increased sensitivity but resulted in reduced specificity and overall classification stability due to increased false positive detections. These results demonstrate that spectral differentiation based on floral coloration can enable reliable species-level detection using low-cost RGB imagery under favorable flowering and illumination. Compared to complex multispectral and hyperspectral approaches, this method offers a reproducible, rapid, and scalable solution for conservation monitoring. However, its applicability remains restricted to the flowering period. The methodology may be transferable to other visually distinctive flowering species, provided that species-specific threshold validation is performed, and may support habitat monitoring and pollinator-related ecological studies in Pannonian steppe habitats. • Low-cost drone imagery detects a rare flowering steppe plant. • Flower colour contrast enables reliable species detection. • Classification accuracy exceeded 99% under optimal conditions. • Negative masking reduces false positives and improves accuracy. • Method is limited to the flowering period of the target species.