Effect of Center Determination Method, Point Cloud Filtering and Cross-Section Type on the Accuracy of Tree Stem...

Static laser scanning is an advanced approach to timber volume estimation whose accuracy now matches or exceeds that of traditional forestry methods for volume estimation, enabling its practical application in woodworking management. This study evaluates how stem volume estimation accuracy from point clouds is affected by stem surface condition (with and without bark), cross-sectional area calculation methods, point cloud filtering, and stem center determination. Results show that the stem surface condition fundamentally controls the direction and magnitude of systematic errors. For stems with bark, the cross-sectional area derived from the squared diameter caused systematic volume underestimation, which increased after filtering, whereas the quadratic mean diameter led to overestimation that was substantially reduced by filtering and approached reference values obtained from computed tomography. For debarked stems, the squared diameter method produced systematic overestimation that was effectively mitigated by filtering, while the quadratic mean diameter resulted in persistent overestimation. The volumetric method itself had only a secondary influence on accuracy. Volume estimation accuracy was primarily governed by cross-sectional area formulation and stem surface condition. For logs with bark, the squared arithmetic mean radius produced systematic volume underestimation that increased by approximately 1% after filtering, whereas the quadratic mean radius overestimated volume by about 2% after filtering. For debarked logs, volume overestimation decreased from 4.05-4.72% to 1.57-1.80% after filtering when using the squared arithmetic mean radius, while overestimation remained high (6.55-6.92%) when the quadratic mean radius was applied. In contrast, the choice of volumetric method and center determination affected volume estimates by less than 1%. These findings confirm the high potential of static laser scanning and emphasize the need for point cloud-adapted computational models.