Comparative analysis of machine learning algorithms and vegetation indices for mapping <i>Eucalyptus</i> tree woodlots using Sentinel‐2 imagery in the Akaki River catchments

Abstract Eucalyptus trees (ETs) cover approximately 20 million ha globally, with Brazil leading at 5.6 million ha. In Africa, ET plantations span around 1.5 million ha, primarily in South Africa and Ethiopia. In Ethiopia, ET plays a key role in agroforestry, covering 506,000 ha, which represents 90% of all planted trees in the form of woodlots. These woodlots, particularly in the Akaki River catchment, are essential for local livelihoods, providing timber, fuelwood, and construction materials. However, while the socioeconomic benefits and ecological effects of ET are well‐documented, the spatial distribution of ET across Ethiopia remains underexplored. Furthermore, previous studies have examined machine learning (ML) algorithms and vegetation indices (VIs) separately for identifying tree species, but limited research has compared these methods for mapping specific land features, such as ET distribution. This study aims to address this gap by comparing ML algorithms, including artificial neural networks (ANN), random forest (RF), and support vector machines (SVM), with VIs like the normalized difference vegetation index (NDVI), green optimized soil adjusted vegetation index, green chlorophyll index, and modified soil adjusted vegetation index to map the spatial distribution of Eucalyptus globulus Labill. woodlots using Sentinel‐2 imagery. Our results show that RF outperformed other ML techniques with 96.3% overall accuracy (OA) and a 0.93 kappa coefficient (K), while ANN and SVM attained 88.7% and 81.7% OA, respectively. Among the VIs, NDVI was the most reliable, with an OA of 90.7% and a K of 0.887. We conclude that ML algorithms provide a more robust method for mapping specific land features like ET distribution than VIs. Future research should investigate the effects of ET on ecosystem services, incorporating socioeconomic data and advanced ML techniques, such as deep learning, to enhance mapping accuracy.