GEOBIA with LiDAR and Multispectral Data

Why Combine LiDAR and Multispectral Data

Spectral and elevation data provide complementary information:

• Spectral data: Surface material properties, vegetation health, moisture content
• LiDAR data: Height, structure, 3D geometry, beneath-canopy information

For example, distinguishing grass from shrubs using optical imagery alone is often unreliable—both are green. Adding height information from LiDAR makes this discrimination straightforward.

LiDAR-Derived Products for GEOBIA

Raw LiDAR point clouds are typically processed into raster products for integration with imagery:

Terrain Models

• DEM/DTM: Ground surface elevation
• DSM: Top-of-surface elevation (including vegetation and buildings)
• nDSM (normalized): Height above ground (DSM - DTM)

Intensity

LiDAR intensity represents return signal strength, providing pseudo-spectral information in the near-infrared range. It can complement optical bands, though calibration challenges limit its reliability.

Point Density and Return Metrics

• Point density: Points per unit area (varies with surface type)
• Return ratio: First vs multiple returns (indicates vegetation penetration)
• Canopy height metrics: Percentile heights, mean, maximum

Segmentation Strategies

Two primary approaches exist for integrating LiDAR in segmentation:

Layer Stacking

Include LiDAR derivatives as additional layers in multi-band segmentation:

• Combine nDSM with multispectral bands
• Weight layers based on their discrimination importance
• Single segmentation incorporates both data types

This works well when spectral and height boundaries align (buildings, isolated trees). However, height discontinuities may fragment spectrally homogeneous objects.

Sequential Segmentation

Perform separate segmentations and combine results:

• Segment imagery for spectral boundaries
• Segment LiDAR for structural boundaries
• Intersect or hierarchically relate the results

This preserves the optimal segmentation for each data type but increases complexity.

Feature Extraction

LiDAR-derived features significantly expand the feature space:

Height Features

• Mean, maximum, minimum height above ground
• Height variance (indicates structural complexity)
• Height percentiles (canopy structure)

Structural Features

• Canopy cover (proportion of first returns above threshold)
• Canopy roughness (height texture)
• Return ratio (vegetation penetration indicator)
• Gap fraction (openness)

Terrain Features

• Slope and aspect
• Terrain ruggedness
• Topographic position index
• Curvature (profile, plan)

Application Examples

Urban Mapping

LiDAR-enhanced GEOBIA excels at detailed urban classification:

• Building detection and footprint extraction
• Building height and roof type classification
• Vegetation stratification (grass, shrub, tree)
• Impervious surface mapping

Forestry Applications

• Tree species classification (structure + spectral)
• Stand delineation and characterization
• Biomass estimation at object level
• Understory vegetation mapping

Wetland Mapping

• Micro-topography and hydrology modeling
• Vegetation type and structure
• Flood extent and frequency zonation

Practical Considerations

Data Registration

Accurate co-registration between LiDAR and imagery is critical. Misalignment creates edge artifacts and incorrect feature values. Verify registration by comparing visible features (building corners, road edges).

Temporal Matching

LiDAR and imagery may be acquired at different times. For dynamic features (vegetation, construction), temporal mismatch creates inconsistencies. Prioritize contemporaneous data when possible.

Resolution Differences

LiDAR raster products may have different resolution than optical imagery. Resampling should be done thoughtfully—bilinear or cubic interpolation for continuous surfaces (elevation), nearest neighbor for discrete values (classified returns).

Layer Weighting

When combining layers in multi-resolution segmentation, layer weights determine relative influence. Height data may need down-weighting if it dominates spectrally important boundaries, or up-weighting for applications where structure is primary.

Emerging Developments

• Multispectral LiDAR: Single-sensor fusion with multiple wavelength returns
• Photogrammetric DSM: Structure from motion as LiDAR alternative
• UAV platforms: High-resolution LiDAR + camera systems
• GEDI integration: Spaceborne LiDAR for vegetation structure

These technologies continue to expand opportunities for multi-sensor GEOBIA workflows.