Advancing Mediterranean Biodiversity Monitoring in South Africa Through Machine Learning and Cost‐Effective UAS Imagery

Abstract The Greater Cape Floristic Region (GCFR) of South Africa is globally recognized for its exceptional plant diversity and endemism, yet faces mounting threats from habitat loss, altered fire regimes, and invasive species. Fire is a key ecological driver in the Fynbos (shrubland) biome of the GCFR, influencing vegetation structure, composition, and nutrient cycling. Understanding the dynamics of Fynbos sites in terms of the time since last burn is a key aspect of understanding its ecology, as it helps reveal post‐fire succession stages and vegetation recovery patterns. Although satellite remote sensing supports biodiversity monitoring, its relatively coarse resolution limits its utility in capturing fine‐scale vegetation dynamics. To address this, we employed high‐resolution unmanned aerial system (UAS) multispectral imagery to classify vegetation of different post‐fire ages and map species diversity in the Fynbos biome. Our methodology, grounded in the Spectral Variation Hypothesis (SVH), leverages optimal spectral and textural features derived from UAS imagery to distinguish between vegetation of different post‐fire ages and estimate alpha diversity at fine scales within Fynbos. We used sequential feature selection (SFS) to identify key predictors, achieving high classification performance with a support vector machine (SVM) classifier (overall accuracy: 97%; F1 score: 97.47%). We employed a similarity metric, Euclidean distance to map alpha diversity across vegetation of different post‐fire ages within the Fynbos biome, by utilizing optimal features and the Shannon diversity index from ground truth samples. This study highlights the role of advanced remote sensing and ecological research in supporting biodiversity monitoring in regions like the GCFR.