Girth-Based Anchor Matching for Handheld SLAM LiDAR Forest Inventory Under Closed Tropical Canopies

초록

Per-tree geolocation in closed tropical canopies has typical uncertainties of 5-15 m with GNSS receivers, preventing automated linking of field inventories to point-cloud stem data. We propose an anchor-based matching framework that does not require per-tree GNSS. A handheld SLAM LiDAR scanner maps stems and girths within approximate to 40 min; field crews record species, girth, and serial numbers without physical markers or tools. Dataset linkage uses a small subset of reflective-tape anchor trees (35 and 43 per hectare, roughly one per 400-500 m(2)) with approximate GNSS locations. Species identity is transferred using median-based GNSS bias correction and quadrant-partitioned Hungarian matching with global deduplication; accuracy is validated by leave-one-anchor-out (LOAO) tests and exact binomial statistics. Tested in two 1-ha plots of open Dry Dipterocarp Forest (DDF; 280 trees/ha) and dense Dry Evergreen Forest (DEF; similar to 1054 trees/ha) at the Sakaerat Biosphere Reserve, Thailand, SLAM girth matched tape data with R-2 = 0.997, RMSE = 1.82 cm in DDF; in DEF, after correcting a 6.38 m GNSS bias, R-2 = 0.986 and RMSE = 7.01 cm, with >= 99% detection for stems >= 30 cm girth (99.2% DDF; 100% DEF). LOAO accuracy was 35/35 in DDF and 40/43 in DEF. Retroreflective-tape anchors were additionally detected automatically from the SLAM intensity channel in 71.4% of DDF anchors (95% CI 53.7-85.4) and 76.7% of DEF anchors (95% CI 61.4-88.2) at intensity >= 150 DN, with up to 59-fold enrichment over matched non-anchor controls at I >= 250 in DEF (Fisher's exact p < 1 x 10(-15)), enabling a fully automated anchor-detection pipeline.

키워드