Find the mechanism before the formula
A three-dimensional canopy grid is not measured evenly. Laser paths enter some voxels, stop in others and never reach the deepest cells. Plant-area density therefore depends on which rays actually traversed each voxel, how far they travelled inside it and how much unseen space remains behind earlier interception.
We will trace rays through voxels, calculate density from a gap fraction, transfer the result to an oblique path, distinguish geometric exposure from ecological absence, and set limits on inverse-probability occlusion corrections.
This guide belongs to the Gold Standard of Remote Sensing, the Gold Standard of Education and the How X Works hub.
Choose a reading route
Start with the question closest to the decision you need to make. Every route returns to the same worked evidence and transfer checks.
| Route | Chapters | Use it when |
|---|---|---|
| 1 | 1–3 | You need the voxel, path and gap-probability model. |
| 2 | 4–5 | You want the worked density and oblique-path calculations. |
| 3 | 6–8 | You need occlusion correction and waveform transfer. |
| 4 | 9–10 | You must validate assumptions and uncertainty. |
| 5 | 11–12 | You want FAQs and a reproducible canopy checklist. |
Chapter contents
Open the group you need, then use the chapter links. Teaching sections remain open on the page.
Build the model
Test the mechanism
1. A voxel is a sampling cell, not a measured leaf cube
Divide the canopy into cells with known horizontal and vertical dimensions. A LiDAR ray may miss a cell, cross part of it, cross the full diagonal or terminate after an interaction. Return counts alone ignore these unequal opportunities to observe plant material.
For each pulse, retain the entry and exit points of every traversed voxel. A fast grid-traversal algorithm advances to the next x, y or z boundary and accumulates the segment length inside each cell. The result is an exposure table: which rays reached a voxel and how much path each contributed.
- Use one coordinate reference and height datum.
- Record beam origin and direction.
- Retain per-voxel path length.
- Keep no-return and quality flags.
2. Occlusion is missing opportunity, not zero vegetation
Upper foliage intercepts energy and reduces the number of pulses available to sample deeper canopy. A deep voxel with no returns can mean empty space, weak reflectance or complete shadow. Treating all three as zero plant area biases vertical profiles toward the top.
The relevant denominator is the number or weighted energy of rays that actually reached the voxel. Rays blocked above it provide information about the upper canopy, not direct evidence about the hidden cell.
3. Directional gap probability links transmission to plant area
For a horizontally homogeneous layer, a Beer–Lambert form writes Pgap(θ)=exp[−G(θ)Ω(θ)PAI/cosθ]. G is the projection coefficient, Ω a clumping factor and PAI the plant area index including leaves and woody material. A voxel form replaces layer PAI with plant-area density times ray path length.
For a voxel crossed by path length L, a simple model is Pgap=exp(−GΩ·PAD·L), so PAD=−ln(Pgap)/(GΩL). The formula is only as good as its exposure, clumping and directional assumptions.
| Quantity | Meaning | Common failure |
|---|---|---|
| Pgap | fraction transmitted without interception | uses pulses that never reached the cell |
| L | path inside the voxel | replaced by voxel height |
| G(θ) | projection coefficient | assumed 0.5 without sensitivity test |
| Ω | clumping correction | silently set to one |
| PAD | plant area per volume | confused with leaf area only |
4. Worked example: a one-metre voxel path
Suppose 100 usable pulses enter a voxel and 80 leave without a classified plant interception. The directional gap estimate is 0.80. Let G=0.50, Ω=1 and L=1.0 m. Then PAD=−ln(0.80)/(0.50×1×1.0)=0.446 m²/m³.
The units follow because plant area per volume is inverse length. The estimate does not say that 20 percent of the voxel volume is vegetation. It says the observed transmission is consistent with an optical plant-area density of about 0.446 m² per cubic metre under the model.
5. Transfer check: lengthen the path without changing transmission
Now let the same gap fraction arise along an oblique segment of L=1.6 m, with the same G and Ω. PAD becomes 0.2231/(0.8)=0.279 m²/m³. Longer exposure needs less plant area per volume to produce the same attenuation.
If an analyst substitutes the one-metre voxel height for the actual 1.6 m path, the estimate remains 0.446 and is about 60 percent too high. Exact traversal length matters most for oblique scans, irregular voxel intersections and sloping terrain.
6. Model the probability that a ray reaches each depth
Let Sv be the estimated probability that a pulse survives the path above voxel v. A Horvitz–Thompson style correction can weight an observed deep traversal by 1/Sv. This restores representation only when survival probabilities are well estimated and the missingness model is defensible.
When Sv is small, weights explode and variance becomes unstable. Cap or regularise weights, publish the cap, and mark cells below a minimum exposure threshold as unresolved rather than filling them with confident density values.
7. Discrete returns and waveforms need different observation models
Discrete-return systems classify a few echo locations per pulse. Full-waveform systems record energy over range bins, while photon-counting systems record event times with background and dead-time effects. Each requires a different likelihood for interception and transmission.
A waveform ratio can estimate cumulative gap probability, but detector response, ground energy, multiple scattering and reflectance differences must be handled. Do not combine return counts and waveform energy as if they were the same sampling unit.
8. Integrate PAD only after voxel support is aligned
A vertical plant-area profile integrates horizontal voxel estimates at each height, weighting by sampled ground area and exposure. Integrating PAD over height yields a plant area index when definitions and units align. Gaps in deep support should remain uncertainty, not be silently set to zero.
Footprint-scale missions such as GEDI retrieve directional gap probability and plant-area-volume-density profiles from calibrated waveforms. Small-footprint point clouds can resolve lateral variation, but their scan-angle and occlusion patterns require more explicit voxel exposure accounting.
9. Validate geometry, energy and ecology separately
First test ray traversal with synthetic lines through known grids. Then test radiometry or return classification on calibration targets. Finally compare PAD or PAI with independent field measurements such as destructive sampling, litter traps or carefully designed optical instruments.
Agreement in total PAI can hide a wrong vertical profile. Compare cumulative gap probability, height-resolved density and exposure maps. Report how results change with voxel size, angle bin, G, Ω and minimum-ray threshold.
10. Limits and misconceptions
- No return is not automatically an empty voxel.
- A pulse blocked above a voxel is not part of that voxel’s exposure denominator.
- Voxel height is not always ray path length.
- G=0.5 assumes a leaf-angle model and is not universal.
- Inverse-probability correction can amplify noise catastrophically.
- PAD from LiDAR usually includes woody area unless the model separates it.
Beer–Lambert inversion also assumes a tractable scattering and clumping model. Dense tropical canopies, multiple scattering and weak ground signals may need richer radiative-transfer treatment.
11. Frequently asked questions
Should I use smaller voxels for more detail?
Only while enough rays traverse each cell. Smaller voxels improve nominal resolution but can create sparse, angle-dependent estimates with large occlusion variance.
Can several flight lines reduce occlusion?
Yes. Diverse view directions expose different canopy gaps. Merge them with per-ray geometry and calibration retained, not by simply pooling counts.
Is PAVD the same as leaf area density?
Not necessarily. Plant-area-volume density can include leaves, branches and stems. The product definition and retrieval assumptions determine what is represented.
12. A reproducible voxel-density checklist
| Stage | Evidence to retain |
|---|---|
| Geometry | beam origin, direction and datum |
| Traversal | entry, exit and path length per voxel |
| Exposure | rays that actually reached each cell |
| Optics | gap, G and clumping definitions |
| Correction | survival weights and thresholds |
| Validation | synthetic, field and sensitivity checks |
The gold standard is to show one ray-length calculation, preserve unresolved shadow and test a changed scan angle. A canopy map becomes credible when empty, unseen and weakly observed voxels are not treated as the same thing.
Reading routes · Previous chapter · Contents · Continue reading
Continue through the Gold Standard ecosystem
Return to the Gold Standard of Remote Sensing, the Gold Standard of Education parent, or the How X Works hub.
Read next in this subject
- LiDAR Waveform Decomposition, Footprint Mixing and Canopy–Ground Separation
- LiDAR Multiple Scattering, Canopy Attenuation and Waveform Energy Closure
- Photon-Counting LiDAR, Dead Time and Background-Noise Bias