Vegetation Health Classification in the Kurdistan Region Using AI and Sentinel-2A Satellite Data

Background: Monitoring climate science is imperative, as it provides the scientific information needed on current changes in the Earth's climate system and their environmental and ecological consequences. However, precisely and scalably classifying vegetation health in heterogeneous semi-arid regions is challenging, particularly with limited labeled data and computational resources. Aims: This study develops an effective vegetation health classification model using Sentinel-2 imagery and machine learning. This paper examines vegetation health assessment as an ecological indicator of environmental stress, rather than climate change prediction perse. Methodology: A Support Vector Machine (SVM) classifier was trained using vector histogram features extracted from the Sentinel-2A spectral bands (B2, B3, and B4), and its hyperparameters were tuned via Bayesian optimization to achieve optimal performance. The model was trained and evaluated on 379 images from the Kurdistan Region, belonging to three vegetation health classes (moderately healthy, unhealthy, and dead). Results: The proposed method achieved 90.79% test accuracy, outperforming Convolutional Neural Network (CNN), Random Forest (RF), and K-Nearest Neighbors (KNN) baselines while maintaining low computational complexity. Conclusion: These findings highlight the effectiveness of compact spectral representations for accurate and efficient vegetation health classification, supporting scalable monitoring in resource-constrained semi-arid regions.