modelling Level 4

Terrestrial LiDAR Point Clouds

How do we create millimeter-accurate 3D models of terrain and structures? Light Detection and Ranging (LiDAR) uses laser pulses to measure distances, generating dense point clouds with billions of 3D coordinates. This model derives ranging equations, implements point cloud classification, demonstrates DEM generation, and shows applications from forest inventory to archaeological site mapping.

Prerequisites: ranging equations, point cloud processing, surface reconstruction, classification

27 Feb 2026 Updated 12 Mar 2026 12 min read

1. The Question

How tall is this forest canopy, and how much biomass does it contain?

LiDAR (Light Detection and Ranging):

Active sensor that measures distance by timing laser pulse return.

Principle:

\[R = \frac{c \times \Delta t}{2}\]

Where:

$R$ = range (distance)
$c$ = speed of light (3 × 10⁸ m/s)
$\Delta t$ = round-trip travel time
Factor 2: Light travels to target and back

Example:

$\Delta t = 100$ nanoseconds = 100 × 10⁻⁹ s

\[R = \frac{3 \times 10^8 \times 100 \times 10^{-9}}{2} = \frac{30}{2} = 15 \text{ m}\]

Point cloud:

Million to billions of 3D points, each with (x, y, z) coordinates.

Platforms:

Airborne (aircraft, drone)
Terrestrial (tripod-mounted scanner)
Mobile (vehicle-mounted)
Spaceborne (GEDI, ICESat-2)

Applications:

High-resolution DEMs (0.5-1 m vertical accuracy)
Forest structure (canopy height, biomass)
Urban 3D modelling (building extraction)
Power line monitoring (vegetation encroachment)
Archaeological surveys (detect subtle features)
Coastal change (erosion, dune migration)

2. The Conceptual Model

Pulse Return Characteristics

Single return:

Hard surface (ground, building)
All energy reflected from single surface
One distance measurement per pulse

Multiple returns:

Vegetation canopy (partial transmission)
First return: Canopy top
Intermediate returns: Mid-canopy
Last return: Ground

Full waveform:

Record entire return signal shape
Extract multiple peaks
Provides vertical structure detail

Typical: 1-5 returns per pulse

Point Density

Airborne typical:

1-10 points/m² (standard)
10-50 points/m² (high-density)
50+ points/m² (ultra-high)

Terrestrial:

1000-10,000+ points/m² (very dense)
Sub-centimeter spacing

Trade-offs:

Higher density → better detail, larger file size, slower processing
Lower density → faster, cheaper, less detail

Flying height effect:

Point spacing $\approx$ altitude × scan angle / pulse rate

Point Cloud Attributes

Essential:

X, Y, Z coordinates (location)
Intensity (return strength)
Return number (first, last, intermediate)
Classification (ground, vegetation, building, etc.)

Optional:

RGB color (if camera integrated)
GPS time
Scan angle
NIR intensity

File formats:

LAS/LAZ (standard, compressed)
ASCII XYZ (simple, large)
E57 (terrestrial scans)

3. Building the Mathematical Model

Ranging Equation

Received power:

\[P_r = \frac{P_t \eta_{\text{sys}} \eta_{\text{atm}} \beta_{\text{target}} A_r}{R^2}\]

Where:

$P_r$ = received power (W)
$P_t$ = transmitted power (W)
$\eta_{\text{sys}}$ = system efficiency
$\eta_{\text{atm}}$ = atmospheric transmission
$\beta_{\text{target}}$ = target reflectance (0-1)
$A_r$ = receiver aperture area (m²)
$R$ = range (m)

Key: $R^{-2}$ dependence → weaker signal at greater distance

Minimum detectable:

\[P_r > P_{\text{threshold}}\]

Maximum range:

\[R_{\max} = \sqrt{\frac{P_t \eta_{\text{sys}} \eta_{\text{atm}} \beta_{\text{target}} A_r}{P_{\text{threshold}}}}\]

Typical airborne: $R_{\max} \approx$ 500-3000 m

Ground Point Classification

Progressive TIN Densification algorithm:

Steps:

Seed points: Lowest points in grid cells (likely ground)
Build initial TIN (Triangulated Irregular Network)
Test remaining points:
- Calculate distance to TIN surface
- If distance < threshold AND angle < threshold → add to ground
Rebuild TIN with new points
Iterate until no new points added

Distance threshold: Typically 0.3-1.5 m

Angle threshold: Typically 8-15°

Robustness: Filters vegetation, buildings, bridges

Canopy Height Model

Digital Surface Model (DSM):

Elevation of first returns (top surface).

Digital Terrain Model (DTM):

Elevation of ground points only.

Canopy Height Model (CHM):

\[\text{CHM} = \text{DSM} - \text{DTM}\]

Interpretation:

CHM = 0: Bare ground
CHM = 20m: 20-meter tall vegetation/structure

Forest applications:

Tree height extraction
Crown delineation
Biomass estimation

Biomass Estimation

Allometric relationship:

\[B = a \times H^b\]

Where:

$B$ = aboveground biomass (kg or Mg)
$H$ = height (m) from LiDAR CHM
$a$, $b$ = species/region-specific constants

Typical: $b \approx 2-3$

Example (temperate forest):

$a = 0.15$, $b = 2.5$

Tree height $H = 25$ m:

\[B = 0.15 \times 25^{2.5} = 0.15 \times 1953 = 293 \text{ kg}\]

Scale to stand:

Sum all trees, convert to Mg/ha.

4. Worked Example by Hand

Problem: Calculate range from timing and classify point.

Laser pulse measurements:

Point A:

Travel time: 200 ns
Intensity: 850
Elevation above geoid: 1535 m

Point B (nearby):

Travel time: 180 ns
Intensity: 920
Elevation: 1520 m

Ground elevation in area: ~1520 m

Classify Point A (ground or vegetation).

Solution

Step 1: Calculate ranges

\[R_A = \frac{3 \times 10^8 \times 200 \times 10^{-9}}{2} = \frac{60}{2} = 30 \text{ m}\] \[R_B = \frac{3 \times 10^8 \times 180 \times 10^{-9}}{2} = \frac{54}{2} = 27 \text{ m}\]

Step 2: Height above ground

Point A: $1535 - 1520 = 15$ m above ground

Point B: $1520 - 1520 = 0$ m (on ground)

Step 3: Classification

Point A: 15 m above ground → Vegetation (likely tree canopy)

Point B: 0 m above ground → Ground

Step 4: Intensity interpretation

Point B higher intensity (920 vs 850):

Ground typically higher reflectance than vegetation
Consistent with classification

Step 5: Tree height

If Point A is first return from tree top:

Tree height = 15 m

Step 6: Biomass estimate (using allometric equation)

\[B = 0.15 \times 15^{2.5} = 0.15 \times 435 = 65 \text{ kg}\]

Single tree biomass ≈ 65 kg (or 0.065 Mg)

5. Computational Implementation

Below is an interactive LiDAR point cloud simulator.

Terrain type: Vegetation coverage (%): 60 Point density (pts/m²): 5 Show classification:

Total points: --

Ground points: --

Vegetation points: --

Mean canopy height: -- m

Observations:

Ground points (brown) form continuous surface
Vegetation points (green) distributed above ground
Complex terrain shows elevation variations
Higher point density reveals finer structure
Classification separates ground from vegetation returns
Mean canopy height derived from vegetation points

Key insights:

Multiple returns capture vertical structure
Point cloud density affects detail level
Classification algorithms identify ground surface
Canopy height directly measurable from point cloud

6. Interpretation

High-Resolution DEMs

LiDAR advantages over photogrammetry:

Penetration: Sees ground through vegetation canopy

Accuracy: ±5-15 cm vertical (vs ±30-50 cm stereo photos)

Automation: Less manual editing required

Examples:

USGS 3DEP (3D Elevation Program):

National LiDAR coverage (USA)
1-m resolution DEMs
Quality level 2 (8 points/m²)

Applications:

Flood modelling (precise elevations critical)
Infrastructure planning
Archaeological features (subtle earthworks)

Forest Inventory

Metrics from LiDAR:

Individual tree detection:

Local maxima in CHM = tree tops
Watershed segmentation = crown boundaries
Height, crown diameter, position

Stand-level:

Mean height, height percentiles
Canopy cover fraction
Vertical structure (understory presence)

Biomass:

Plot-level calibration:

Field measure biomass (destructive sampling or allometry)
Correlate with LiDAR metrics
Apply regression across landscape

Accuracy: ±15-25% biomass estimation (vs ±30-50% optical)

Carbon accounting:

Forest biomass × 0.5 = carbon stock

Critical for:

REDD+ programs
Carbon offset verification
Climate change mitigation

Archaeological Applications

Bare-earth DEM reveals:

Ancient roads/paths
Building foundations
Earthworks/fortifications
Agricultural terraces

Example - Angkor Wat, Cambodia:

LiDAR through jungle canopy
Revealed extensive urban grid
Hydraulic infrastructure
Changed understanding of city extent

Example - Mayan cities, Central America:

Discovered hidden structures
Population estimates revised upward
Settlement patterns clarified

7. What Could Go Wrong?

Ground Classification Errors

Type I (commission):

Vegetation misclassified as ground.

Cause: Low vegetation, dense understory

Impact: DTM too high → CHM underestimate

Type II (omission):

Ground misclassified as vegetation.

Cause: Bridges, large boulders, low point density

Impact: DTM too low → CHM overestimate

Solution:

Visual inspection
Cross-validation with field data
Iterative editing

Point Density Insufficient

Sparse data misses features.

Critical density:

Buildings: 4-8 pts/m²
Forest: 2-5 pts/m²
Bare earth: 1-2 pts/m²

Example:

1 pt/m² → Miss narrow roads, small buildings

Solution: Increase flight density or lower altitude

Co-Registration Errors

GPS/IMU accuracy:

Determines absolute positioning.

Typical: ±5-10 cm horizontal, ±10-15 cm vertical

Differential GPS improves to ±2-5 cm

Multi-temporal comparison:

Requires precise registration or errors appear as change.

Solution:

Ground control points
Strip adjustment
ICP (Iterative Closest Point) alignment

Vegetation Penetration Limits

Dense canopy:

Few pulses reach ground.

Rainforest: <5% ground returns

Conifers: 10-30% ground returns

Deciduous (leaf-off): 60-90% ground returns

Solution:

Leaf-off acquisition (temperate forests)
Higher point density
Full-waveform LiDAR (better penetration)

8. Extension: Full-Waveform LiDAR

Discrete return LiDAR:

Records 1-5 return distances per pulse.

Full-waveform:

Records entire reflected signal (continuous function).

Advantages:

More returns: Extract 10+ returns per pulse

Intensity variation: Within canopy structure

Ground detection: Better in dense vegetation

Calibrated intensity: Physical reflectance

Applications:

Forestry: Leaf area index, understory structure

Bathymetry: Water column returns

Snow depth: Snow surface + ground returns

Urban: Power line detection, facade detail

Processing:

Gaussian decomposition:

Fit multiple Gaussian peaks to waveform
Each peak = one return
Width = target extent

9. Math Refresher: Speed of Light Constant

Fundamental Constant

\[c = 299,792,458 \text{ m/s}\]

Exactly defined (SI definition of meter).

In vacuum: Truly constant

In atmosphere: Slightly slower

\[c_{\text{air}} = \frac{c}{n}\]

Where $n$ = refractive index ≈ 1.0003

Effect on ranging:

\[\Delta R = R \times (n - 1) \approx R \times 0.0003\]

For 1000 m range: Error = 30 cm

Correction applied in precision LiDAR.

Nanosecond Timing

1 nanosecond = 10⁻⁹ seconds

Light travels in 1 ns:

\[d = c \times 1 \times 10^{-9} = 3 \times 10^8 \times 10^{-9} = 0.3 \text{ m} = 30 \text{ cm}\]

Round trip: 15 cm range resolution

LiDAR timing precision: Sub-nanosecond

Achieves: <5 cm range precision

Summary

LiDAR measures distance by laser pulse timing enabling precise 3D point clouds
Range equation: R = c·Δt/2 where timing precision determines accuracy
Point clouds contain millions to billions of x,y,z coordinates with intensity and classification
Ground point classification via progressive TIN densification filters non-ground returns
Canopy Height Model derived as DSM minus DTM reveals vegetation structure
Applications span high-resolution DEMs, forest biomass, urban 3D modelling, archaeology
Accuracy typically ±5-15 cm vertical, ±5-10 cm horizontal with differential GPS
Challenges include ground classification errors, vegetation penetration limits, co-registration
Full-waveform LiDAR captures complete return signal for enhanced information extraction
Critical tool for topographic mapping, vegetation analysis, and change detection