Canopy gap profiles¶
Eleven TERN plots scanned with a RIEGL VZ-2000i, 36–138 upright scan
positions each. canopy.GapProfile reads the raw RXPs: every 4th pulse with
all its echoes, fired pulses rebuilt from the downward zenith lines, and a
ground plane fitted per scan to its downward echoes. It is compared with:
- pylidar-tls-canopy's Jupp et al. (2009) profiles of the same scans (hinge PAI, pooled per plot and per scan);
- ray-traced 0.5 m voxel PAI (raycloudtools
rayvoxel, free-path-length attenuation, occlusion traced; not comparable in kind: voxels count all plant area, not the upward view); - optical LAI from digital cover (DCP) and hemispherical (DHP) photographs, the nearest date to the scan (dates from the RXP names).
| plot | scans | Sylva hinge | pylidar hinge | per-scan r | voxel | clumping | canopy height (m) | photo (days apart) |
|---|---|---|---|---|---|---|---|---|
| ROBSON_2023 | 116 | saturated | 4.65 | – | 9.11 | 0.64 | 29.0 | DHP 2.54 (132) |
| WARRA_2023 | 85 | 3.04 | 3.07 | 0.97 | 6.36 | 0.61 | 55.5 | – |
| WOMBAT_2023 | 84 | 3.16 | 3.19 | 0.99 | 5.74 | 0.71 | 29.5 | – |
| CUP_2022 | 51 | 2.55 | 2.45 | 0.99 | 3.25 | 0.56 | 24.5 | DCP 1.67 (85) |
| TUMBA_2022 | 138 | 1.69 | 1.63 | 1.00 | 1.86 | 0.55 | 43.0 | DCP 1.14 (111) |
| BOYAGIN_2021 | 57 | 0.88 | 0.87 | 1.00 | 1.63 | 0.50 | 13.0 | DHP 0.99 (36) |
| ALICE_2021 | 40 | 0.84 | 0.82 | 1.00 | 1.62 | 0.41 | 7.5 | DHP 0.76 (1600) |
| LITCH_2021 | 36 | 0.87 | 0.86 | 1.00 | 1.44 | 0.67 | 20.0 | DCP 1.04 (29) |
| FLETCH_2022 | 36 | 0.24 | 0.23 | 1.00 | 0.85 | 0.42 | 12.5 | – |
| GWW_2021 | 36 | 0.17 | 0.17 | 1.00 | 0.70 | 0.41 | 18.5 | DCP 0.51 (1) |
| CALP_2021 | 36 | 0.01 | 0.01 | 1.00 | 0.67 | 0.65 | 18.0 | DCP 0.78 (826) |
- Against pylidar, on the ten plots that are not saturated: r = 0.999, median ratio 1.01, and every plot within 4 %. Scan by scan (599 scans), r is 0.97–1.00 per plot.
- Robson Creek lets less than 0.1 % of pulses through at 57.5°. The
hinge PAI there is bounded by the number of pulses, not measured, and the
report flags it (
saturated). - Against the photographs taken within 120 days (5 plots), Sylva's effective hinge PAI gives r = 0.96 at 0.89 × the photo value. Correcting for clumping (Lang & Xiang 1986 over scan-sector segments) overshoots it (1.76 ×), consistent with the photo values being effective too. The voxel PAI is 1.63 × the photos.
- Low woodland (Calperum mallee, Alice mulga) is mostly at or below the 1.5–2 m tripod. Upward-looking gap fraction does not see it: 0.01 at Calperum where the photos find 0.78 and the voxels 0.67.
- pylidar's Jupp linear estimate (2.7–4.4 on every plot, including ones whose hinge PAI is 0.01) is not usable. Sylva's linear fit is close to its hinge value in forest and 0.3–0.4 above it in open woodland. There the near-horizontal rings cross stems and shrubs close to the ground, which the linear model reads as vertical foliage.
Getting the dense plots right took one fix. RiVLib's stream leaves out the pulses that returned nothing, so fired pulses have to be reconstructed. The first version took a percentile of pulses per zenith line, which runs about 7 % high: the mirror's angles do not sit on the nominal lines, so some line bins catch extra pulses. Where nearly every pulse returns, 7 % too many fired pulses reads as 7 % gap and capped the hinge PAI near 3 (Robson, Warra, Wombat were 25–40 % low). Counting pulses per line on the downward lines, where every pulse hits the ground, fixed it.