Integration of UAV photography, field data and machine learning algorithms for stem volume estimation

Abstract

In this study, the diameter and height of Pinus brutia Ten. trees were measured using orthomosaic data obtained from unmanned aerial vehicle (UAV) imagery, and the stem volumes were estimated using machine learning (ML) techniques. The research was conducted in southwestern Türkiye within the brutian pine stands managed by the Isparta Regional Directorate of Forestry. A total of 175 trees were measured for height and diameter at breast height (d1.3), and these measurements used to estimate volume. The accuracy of these estimations predictions was examined, with volume estimation values serving as dependent variables in various ML algorithms. The performance of nine ML algorithms - AdaBoost Regression, Artificial Neural Network, Deep Neural Network, Decision Tree Regression, Gradient Boosting Regression, Linear Regression, Random Forest Regression, Support Vector Regression, and eXtreme Gradient Boosting Regression - were compared. The results indicated that using only the diameter values (max. correlation 0.984) produced better results than using only the height values (max. correlation 0.932), while combining diameter and height variables (max. correlation 0.987) produced the most accurate results. Among the all algorithms, Random Forest Regression achieved the highest average correlation (0.968), whereas Decision Tree Regression had the lowest (0.906). All algorithms produced correlations exceeding 0.90. These findings demonstrate that ML models can effectively estimate stem volume from UAV-derived diameter and height data under field conditions similar to those in southwestern Türkiye. The integration of remote sensing and ML may therefore offer a viable approach for stem volume estimation in structurally comparable forest environments.

Plant phenotyping relevance

UAV画像から樹木の直径・樹高を取得し、機械学習で幹材積を推定する手法を比較評価しており、植物形質の取得・推定が研究の中心である。

Code and data availability

The paper reports UAV-derived diameter/height measurements and ML stem-volume estimation for 175 Pinus brutia trees, but no public dataset, imagery, code, or model repository is deposited. The underlying data come from a master's thesis (project 2021-YL1-0137) with no public availability statement; only generic tooling

No evidence-backed public reproduction asset is currently recorded.