Integration of Optical and Microwave Data for Forest Mapping in Mongolia

This study aims to evaluate and compare the effectiveness of different classification techniques for extracting forest and other land cover classes using integrated optical and synthetic aperture radar (SAR) datasets. The analysis was conducted using multitemporal Sentinel-2 multispectral imagery and Sentinel-1 SAR data, together with derived additional features. Three classification algorithms such as artificial neural network (ANN), support vector machine (SVM), and maximum likelihood classifier (MLC) were applied and compared. Among the tested methods, SVM demonstrated the most stable performance, achieving a maximum accuracy of 94.98% with a 35-feature dataset. The ANN classifier achieved the highest overall accuracy (95.87%) using a 22-feature combination. However, ANN performance decreased when additional derived features were included, highlighting the importance of optimal feature selection. In contrast, the MLC classifier produced lower and less stable accuracies, reflecting the limitations of traditional statistical classifiers when applied to complex multisource datasets. With respect to the discrimination of forest classes, the ANN method achieved the highest accuracy, followed by SVM, while MLC showed comparatively lower performance. Overall, the research demonstrates that machine learning classifiers combined with multisource satellite data significantly improve the classification accuracy of forest and other land cover classes.