LiDAR Canopy Structure & Foliage Height Diversity

A drone flies a LiDAR over a forest plot: every pulse is a Beer-Lambert Monte Carlo draw through Blender-modelled crowns, and the foliage profile, leaf area index and diversity are recovered from the first returns.

Switch to 2D

Presets

Playback

Parameters

Camera

Layers

Quality

Model check

The science

What airborne LiDAR measures. A laser scanner on a drone or aircraft fires pulses downwards and times the echoes. A pulse that meets leaves is partly reflected, so the height of the echo is a height of foliage; a pulse that passes through all the gaps returns from the ground. From many pulses one can recover how much leaf area sits at each height without any field survey, which is how missions such as NASA GEDI map forest structure.

The forest and the pulses. The 60 m by 60 m plot holds crowns modelled as ellipsoids (radius R, vertical semi-axis a) with uniform leaf area density LAD (m2 of leaf per m3 of crown). Along a path of length l inside a crown a pulse is intercepted at rate k = G LAD, with G = 0.5 for randomly oriented leaves, so the probability of passing unhit is exp(-G LAD l), the Beer-Lambert law. Each pulse draws tau* = -ln(U) and the simulation walks down the ray, summing G LAD l over the crowns it crosses (overlapping crowns add), until the optical depth reaches tau*: that is the height of the first return; if it never does, the pulse reaches the ground. Pulses are fired from flight lines at 70 m with a scan angle limited by the line spacing.

MacArthur-Horn profile and diversity. In a layer of thickness dz the plant area density follows from the number of pulses entering and leaving it, PAD = ln(N_in / N_out) cos(theta) / (G dz), where N_out counts pulses that reach the bottom of the layer and N_in those that reach its top. The foliage height diversity is the Shannon index of the vertical distribution, FHD = -sum p_k ln p_k with p_k = PAD_k / sum PAD, the index MacArthur and MacArthur (1961) related to bird species diversity; here it is only reported as a structural index, no bird numbers are predicted.

What is simplified, and a real bias. Crowns are smooth ellipsoids of uniform density (the Blender trees are only a lumpy picture of them); only the first return is modelled, with no beam divergence, intensity, multiple returns, trunks or branches; the ground is flat for ranging although the picture undulates. Real canopies are clumped, and the Horn estimator, like every gap-fraction method, assumes random foliage inside each layer, so it underestimates the true leaf area index: the readout shows the true LAI, the Horn LAI and the (lower) value from the overall gap fraction, and thinner layers recover more of the truth. Exact agreement is only expected for a uniform layer, which is checked numerically. This is an educational model, not a processing chain for real point clouds.

Numerical checks (run on load): the Monte Carlo ray marcher against Beer-Lambert for a uniform layer, the Horn inversion on that layer, the ray-ellipsoid chord against its formula, the ordering gap-fraction LAI, Horn LAI and true LAI in a test forest, and the crown volume measured from the Blender mesh.

Loading model...
Loading Blender model...
Plant area profile: true against MacArthur-Horn
Share of pulses below a horizontal plane
First-return heights
Made with 🟠 Blender · Three.js