Machine Learning based Forest Cover Change Detection in the Hyrcanian Forests of Iran

Abstract

I investigated forest and non-forest classification and temporal forest change detection in the part of the Hyrcanian forests of northern Iran. Landsat 8 and 9 Imagery (surface reflectance) for the years 2016, 2019, 2022, and 2025 were used. Four machine learning algorithms, namely Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), and MARS, were implemented. Model performance was evaluated using repeated sampling (20 iterations of 60 percent of test data) and multiple accuracy metrics, including Accuracy, F1-score, MCC, and IoU. The results indicated that the SVM and RF classifiers achieved the highest performance, respectively, with a total change of approximately 6324 and 6059 (ha), based on SVM and RF, between 2016 and 2025. Showing the 6.48 and 6.36 percent of forest (SVM, RF) area has changed to other categories. Make it 0.648 and 0.636 percent annually based on SVM and RF respectively.