Akademik Veri Yönetim Sistemi
JOURNAL OF THE INDIAN SOCIETY OF REMOTE SENSING, 2026 (SCI-Expanded, Scopus)
The society's demand for forest resources is constantly increasing and diversifying. Forest management plans are prepared and expired plans are periodically updated to maintain planned operations and sustainability in forest ecosystems. The aim of this study is to obtain a high-accuracy stand map using machine learning algorithms. Aerial LiDAR and WorldView-3 imagery were used to create a highly accurate stand map. To eliminate the intensive labor, cost and subjectivity of stand maps produced by visual interpretation, object-based classification studies were carried out with the help of Support Vector Machines (SVM) and Random Forest (RF) using spectral and textural features of optical remote sensing data and LiDAR. Classifications were performed for three different criteria (canopy cover, development stage and species composition) that constitute stand identification, and sufficient accuracies were obtained. The SVM approach achieved the highest overall accuracy of 98% and 92% kappa coefficient for the canopy cover layer, while the RF approach achieved 87% overall accuracy and 80% kappa coefficient for the development stage layer and 80% overall accuracy and 78% kappa coefficient for the species composition layer. Thematic layers created separately for canopy cover, development stage and species composition were combined using Geographic Information Systems (GIS), and a detailed stand map was obtained. Our study demonstrated that the gradual classification approach can produce stand maps with high accuracy (overall accuracy of 75% and 74% kappa coefficient) in heterogeneous forest ecosystems. Moreover, it also showed that machine learning classifiers provide adequate results for the practical use of stand maps.