Stem Modeling for Savanna Tree Characterization with Close-Range Photogrammetry: Comparison of the Performance of Automatic Algorithms

Abstract

Sudanian savannas remain underexplored in terms of utilizing close-range photogrammetry (CRP) for assessing tree characteristics, leaving a gap in ecological research. This study evaluates the performance of automatic stem modeling techniques using CRP-generated point clouds for 30 trees from five savanna species. Two labeling methods, a machine learning-based approach (StemML) and a flatness/vertical structure-based method (StemFlat), were used to extract stem points. We applied three diameter estimation techniques: convex-hull line fitting (CHM), least squares circle fitting (LSM), and Ransac circle fitting (RANSAC), comparing their results against field measurements using root mean square error (RMSE), bias and the coefficient of determination R2. The combination of StemML and CHM yielded the best performance, with an RMSE of 2.1 cm (5.8%), R2 of 0.983 and a bias of −0.30 cm, accurately identifying 93% of stem segments. Diameter estimation accuracy varied with height, with optimal alignment between CRP-derived profiles and manual measurements occurring between 0.5 and 2.5 m. These findings demonstrate CRP’s potential for modeling savanna tree stems and highlight the importance of method selection in ensuring reliable measurements.

Code and data availability

公開論文であることは確認できましたが、現在の公式API・許可済み取得経路では本文を自動取得できませんでした。

No evidence-backed public reproduction asset is currently recorded.