Multi-Temporal Tree Species Classification using Sentinel-2 Imagery and Machine Learning: A Case Study from Bolu, Aladağ Forests

Abstract. This study investigates the potential of multi-temporal Sentinel-2 imagery combined with machine learning techniques for tree species classification in the Aladağ Forests of Bolu province, Türkiye. Five dominant species—Black Pine, Scots Pine, Nordmann Fir, Beech, and Sessile Oak—were classified using a comprehensive feature set comprising spectral bands, spectral band indices, topographic attributes, and seasonal indicators. Sentinel-2 images acquired in April, August, and November were employed to capture phenological variations influencing classification accuracy. Field data from 112 sample plots supported the training and validation of four machine learning models: Random Forest (RF), Support Vector Machines (SVM), Artificial Neural Networks (ANN) and Ensemble Learning (EL). On the independent test dataset, the overall classification accuracies were 99.22% for SVM with polynomial kernel, 99.03% for SVM with Radial Basis Function (RBF) kernel, 98.61% for RF, 97.91% for EL, and 96.66% for ANN, respectively. Seasonal analysis showed that August imagery provided the best classification performance, benefiting from peak canopy contrast, while accuracy decreased in November, particularly for deciduous species. The results underline the effectiveness of RF and emphasize the importance of integrating multi-seasonal satellite observations with machine learning for improved forest species mapping.