Leveraging Sentinel-2 and Machine Learning for Precise Identification of Plantation Forests in the Upper Blue Nile

Precise identification of plantation forests is essential for sustainable land management, ecological monitoring, and climate change mitigation. This study evaluates the performance of five machine learning algorithms: Artificial Neural Network (ANN), k-Nearest Neighbors (KNN), Random Forest (RF), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGB) in identifying plantation forests using Sentinel-2 multispectral data. Key classification metrics, including overall accuracy (OA), precision, recall, F1-score, and area under the curve (AUC), were used to assess model performance. The results revealed that ANN achieved the highest OA (0.966) and recall (0.99), demonstrating its effectiveness in identifying plantation forests, while XGB and RF exhibited the strongest discriminative ability (AUC = 0.994). SVM showed balanced performance, whereas KNN had the lowest precision (0.883), indicating a higher false positive rate. Spatial distribution analysis indicated that SVM classified the largest proportion of plantation forests (18.32%), followed closely by ANN (18.2%), while KNN estimated the lowest coverage (12.66%). Variable importance analysis highlighted the Normalized Difference Red Edge Index (NDRE) as the most significant predictor in both RF and XGB models, followed by the red-edge (B05) and green (B03) bands, reinforcing their utility in vegetation discrimination. This study provides critical insights into optimal algorithm selection and feature importance for plantation forest monitoring, supporting future applications in remote sensing-based landscape monitoring. Further research should integrate multi-sensor data, including LiDAR and radar, with deep learning approaches to enhance classification robustness across diverse landscapes.