Airborne LiDAR flagged tens of thousands of possible giants across Taiwan, most of them ordinary trees perched on cliffs. Volunteers discarded nine in ten. On the giants that survived the filter, the model’s remaining error mostly ran the other way.
Researchers at Taiwan’s Forestry Research Institute and National Cheng Kung University built an island-wide canopy height model from airborne laser scanning flown between 2010 and 2016, subtracting a bare-earth elevation surface from the canopy surface. A 19-by-19-metre local-maximum filter then picked out every point reading above 65 metres.1
That produced far more leads than anyone could check. An expert analyst needs roughly 20 minutes per candidate to rotate the point cloud into a view that shows both treetop and base.1
Most leads were not tall trees. Steep ground defeats the subtraction, because the laser struggles to sample the true ground surface beneath a cliff edge. The paper’s worked example is a tree under 25 metres growing on top of a cliff, which the model recorded at 90 metres.1
So in 2022 the profile images went online. Volunteers marked treetop and base on each one, three people to an image, working independently. The 57,065 profiles came down to 4,736 candidates — 91.7 per cent discarded — and expert review of the full 3D point cloud confirmed 531 of those. Added to two earlier expert-only surveys, the island’s map now holds 941 trees above 65 metres, most of them on slopes averaging 40 degrees.1
The filter decided which trees were worth walking to. It did not make the model’s heights right. Across the 22 trees the team has climbed and measured since 2014, the root-mean-square error between modelled and field height is 4.55 metres.1
On the ten trees confirmed above 70 metres, the error runs opposite to the cliff artefact. The model read low on six and high on four, averaging 1.57 metres below the measured height. The Heaven Sword is the plainest case: 84.1 metres on the tape, 80.0 in the model. NanKengSi is the sharper one, at 72.9 metres in the field against 66.5 in the model — 1.5 metres above the 65-metre cut that would have deleted it before anyone saw it.1
One reading is that thin laser sampling under-reads a canopy apex, which on a tall conifer is the narrowest and hardest part to hit; the study notes that pulse density is a cost trade-off and that low densities risk under-sampling the apex.1 It does not test that mechanism, and the ten-tree pattern is consistent with several others.
The denominator is not clean. The abstract calls 57,065 the island-wide count of local maxima, but the methods say experts inspected 14,692 profiles in 2019 and 2020, described as about 20 per cent of Taiwan’s candidates — which implies an island-wide pool nearer 72,000, with 57,065 as the remainder handed to volunteers. We use 57,065 only for what volunteers screened, which both passages support.1
Three of the ten heights above 70 metres come from drone photogrammetry rather than tape drop. The Heaven Sword’s modelled height appears as 80.0 metres in the paper’s table and 79.5 in its text; we use the table. The 4.55-metre error figure rests on 22 climbed trees, and there is no independent height check on the other 919 on the map.1
The claim that this is East Asia’s tallest tree depends on where the region’s boundary is drawn. A 102.3-metre Cupressus austrotibetica in the Yarlung Tsangpo canyon in Tibet, China, was measured by drone LiDAR in 2023 and described by its authors as the tallest in Asia — a paper the Taiwan team cites.2 Taiwan’s tree is the tallest confirmed by direct tape drop.