Analyzing DBH estimation accuracy and ecological influences across size classes using SLAM LiDAR and RANSAC algorithm in Korean Forests

Simultaneous Localization and Mapping (SLAM) Light Detection and Ranging (LiDAR) technology is a powerful option for forest inventory, particularly for estimating tree Diameter at Breast Height (DBH). However, the variability in DBH‐estimation accuracy across different size classes and ecological conditions remains underexplored. We therefore assessed a handheld SLAM LiDAR workflow in three ecologically distinct Korean forests – Quercus mongolica on Namsan, Pinus densiflora on Jeombongsan, and a mixed Quercus stand on the same mountain – selected to span gradients in elevation, canopy density, and species composition. Point clouds acquired with an Ouster OS1-32 sensor (650 k pts s−1, 30 mm accuracy, 120 m range) were geo-referenced with ground-control points, denoised, and processed with a Random Sample Consensus (RANSAC) cylinder-fitting algorithm to derive DBH. LiDAR-derived DBH corresponded closely to field measurements (R2 = 0.98, RMSE = 1.84 cm, MAE = 1.45 cm, maximum error = 4.35 cm). Measurement reliability increased sharply above a threshold of 23.5 cm, whereas trees with DBH under 10 cm exhibited the largest deviations (MAPE = 30.76%; rRMSE = 33.71%). Accuracy peaked for larger trees, with R2 reaching 0.96 in the 40 cm class. Shade-tolerant species such as Acer pseudosieboldianum and Styrax japonicus, which often have curved stems, showed greater variability, whereas canopy-dominant Quercus mongolica and Pinus densiflora displayed uniform growth and lower error. These discrepancies in small-diameter trees are likely due to the limited number of LiDAR points captured and the stronger influence of competitive growth, as non-upright stems can introduce errors in both LiDAR and manual measurements.