The Application of Clustering Algorithms for Partitioning Forest Trees into Representative Sections

Accurate forest surveying is critical for sustainable resource management, but is often limited by time and cost constraints. This study investigates the application of multiple clustering algorithms-K-means, K-means with local DBSCAN refinement, hierarchical agglomerative clustering, hybrid clustering algorithms, DBSCAN clustering, and agglomerative clustering with KDE-based outlier detection-to partition tree data into representative sections. The goal is to enable efficient sampling that preserves the forest's structural patterns. Each method was evaluated using internal clustering metrics and visual comparison to a heatmap of tree diameter distribution. Results indicate significant differences in cluster shape accuracy, kernel density estimation, noise detection, and interpretability. Among the tested approaches, agglomerative clustering with KDE-based outlier detection most closely reproduced the forest's spatial structure, offering robust cluster delineation and principled noise handling. This method provides a reliable framework for representative sampling and informed forest management.

کلیدواژه ها:

Forest surveying ، Representative sampling ، Agglomerative clustering ، Kernel density estimation ، Outlier detection