By Sreeja Krishnamari, Arizona State University
1. Introduction
A Canopy Height Model (CHM) represents the height of vegetation, buildings or other objects above the ground surface. CHM is often derived by subtracting the Digital Terrain Model (DTM) from the Digital Surface Model (DSM):
CHM = DSM – DTM
DSM represents the topography and elevation of the earth that includes all natural and man-made features, such as trees and buildings. DSM’s are typically created from the first lidar returns, corresponding to the first points the laser hits before returning to the sensor.
DTM represents the bare-earth model of the ground surface, without any vegetation or built structures. DTM’s are typically created using only the ground-classified points from the lidar data.
The area of interest in this analysis is situated in Coconino County, Arizona and is sourced from the AZ Coconino B1 2019 LiDAR dataset from the USGS 3DEP program. The data was downloaded via OpenTopography. The original horizontal coordinate reference system is WGS84 Web Mercator (EPSG:3857) and we project the coordinate reference system to UTM zone 12. The vertical coordinate reference system is NAVD88 – Geoid12B. In total, 2,702,058 points were retrieved for the specified bounding box. This area spans a canyon with ponderosa pine trees.
Why is CHM Important?
CHMs are essential in forestry, ecology, and remote sensing for:
- Estimating tree height
- Mapping forest structure
- Modeling wildlife habitat
- Assessing fuel loads for wildfire risk
- Quantifying biomass and carbon storage
2. How a Canopy Height Model (CHM) Is Calculated: Two Common Methods
There are two main methods to create a CHM.
2.1 Method 1: Subtracting DSM and DTM raster datasets
Steps:
- Create a Digital Surface Model (see Figures 1 and 2 below).
- Create a Digital Terrain Model (see Figure 3).
- Subtract the two to get Canopy Height Model, as indicated by the equation above
The height of the objects (trees, bushes, buildings, etc.) above the ground surface is determined by subtracting the ground elevation (DTM) from the surface elevation (DSM).
CHM = DSM – DTM
Figure 1: Topographic hillshade of the digital surface model plotted above the colored elevation. This topography shows the underlying canyon topography and vegetation.
Figure 2: 3D Visualization of the Digital Terrain Model (DTM). An oblique 3D perspective of the digital terrain model (DTM), revealing the ground surface topography of the underlying ground. Gradual gradients in elevation are apparent, giving an impression of terrain variability across the landscape.
Figure 3: Visualization of the Digital Terrain Model. Planar 2D color-coded height representation of the DTM. This visualization highlights terrain features such as the sloped terrain and canyon topography. Mapping the landscape is crucial for the derivation of accurate vegetation height in CHM models.
Figure 4: Canopy height model visualization. , Canopy Height Model is calculated by: CHM =DSM – DTM. and represents a model of the vegetation height.
2.2 Method 2: Normalized Point Cloud & Canopy Height Extraction
Step 1: What Is Normalization?
Lidar data are collected by measuring the time it takes for emitted laser pulses to hit a target surface and return back to the sensor. This time difference reflects the elevation of the ground and surface features and allows us to classify points as either ground or elevated objects like trees and buildings. Since the ground itself may already have varying elevation, normalization is used to remove the influence of the terrain. This process smooths the ground surface and adjusts the heights of all points relative to the ground surface, as shown in Figure 5.
Figure 5: Illustration of terrain normalization for canopy height modeling: Left is the original elevation of the trees about a terrain with variable elevation. Right shows the terrain flattened and the elevation of the trees above this flattened terrain.
Step 2: Extracting the Canopy Height Model
Once the point cloud is normalized, the height of each point is the height above the ground surface. To translate this 3D cloud into a 2D raster model (CHM), the study area is divided into grid cells with a user-defined resolution, usually 1m x 1m. Within each cell, the maximum height value from all points is taken as the representative canopy height. This grid-based maximum extraction is what creates the CHM, as shown in Figure 6.
Cells with no laser return, e.g., under closed canopy or within tree crowns, are blank or white. These are positions where no valid canopy surface was detected and are normally dealt with using interpolation or masking, as determined based on the goals of the project.
Figure 6: CHM is Generation: This grid-based CHM was created from a normalized point cloud, where each cell shows the tallest point measured above the ground within each cell. White cells indicate areas where no lidar returns were recorded within the cell often due to gaps in canopy coverage or occlusion.
Doing this across the entire area produces a Canopy Height Model – a 2D map where each pixel tells us how high the tallest vegetation is in that location.The results from this method are shown in Figure 7.
Figure 7: Canopy Height Model (CHM) created using a normalized point cloud, shown as a hillshade colored by the CHM. By removing terrain effects, this method highlights the actual height of vegetation above the ground, giving an accurate view of forest structure.
3. How the two CHM methods Differ: A Pixel-Wise Comparison
Figure 8: Comparison of canopy height histograms from the two CHM methods. Blue: raster-based subtraction. Purple: Normalized point cloud. The normalized point cloud method shows slightly higher canopy values, especially for taller trees.
The histogram in Figure 8 compares the canopy height distributions produced by the two CHM methods. The low canopy height values represent the ground surface with no trees. The point cloud normalization method tends to produce slightly taller canopy height values at higher ranges (>10 m). There are some negative CHM values, particularly for the raster-based approach, likely reflecting gridding errors in the DSM and/or DTM.
Figure 9: Map of height differences between two CHM methods: Method 1 (DSM – DTM) and Method 2 (Normalized Point Cloud). The diverging color scale shows where the methods disagree. Brown areas indicate that Method 1 estimated lower canopy heights than Method 2 (negative difference), while blue areas mean it estimated higher heights (positive difference). The dominance of brown suggests Method 1 consistently underestimates canopy height, particularly along tree edges and crown boundaries. These spatial differences are profiled across the red transect shown in the image.
Figure 10: Canopy height profiles taken along the red transect are shown in Figure 9. The red line shows height estimates based on method 1 (DSM – DTM), and the blue line represents method 2 (Normalized Point Cloud). The CHM from method 2 is often higher than from method 1.
The profile tool in Figure 14 shows a profile through both canopy height models, highlighting their differences. Method 1 estimates lower canopy heights compared to Method 2. This matches what we see in the map Method 2 tends to pick up taller trees, especially around the edges of the canopy or where there are sudden changes in height.
This suggests that the DSM-based method (Method 1) might smooth over or miss some tree heights near the edges, while the Normalized Point Cloud method (Method 2) captures more detailed and sharper canopy structures.
References:
Roussel, J.-R., Auty, D., Coops, N. C., Tompalski, P., Goodbody, T. R. H., Meador, A. S., et al. (2020). lidR: An R package for analysis of Airborne Laser Scanning (ALS) data. Remote Sensing of Environment, 251, 112061. https://doi.org/10.1016/j.rse.2020.112061
U.S. Geological Survey (2020). AZ Coconino B1 2019. Distributed by OpenTopography. https://portal.opentopography.org/usgsDataset?dsid=AZ_Coconino_B1_2019, last accessed 2025-05-19.