Canopy Height Model (CHM) is a difference between Digital Surface Model (DSM), being developed from first returns and Digital Terrain Model (DTM) created from the ground returns.

chm_example
Figure 1 – CHM image with trees between 0 and 65 meters of height. The lowest heights are in blue colour and the highest points are in red.   With some trees reaching 65m this well above the norm for radiata pine growth plantations in NZ.

There are a couple of practical tips we need to understand when we speak about CHM in forestry and using this for the estimation of tree heights or assessments of stand characteristics.

Leaning Trees or Adverse Slopes / Cliffs

The first is that it is possible that our tallest tree is not actually “the tallest tree” in a traditional tree stem height sense.   Sometimes the trees are leaning over a cliff or down slope.   In figure 1 this has resulted in heights being dislayed well above normal expectation.   In this situation the CHM offers a false tree height compared to how it would be measured in the field using a hypsometer.   An example is shown in the elevation profile: the tree shown is reported in the CHM as 69m, but more realistic is an assessment is only 44 meters in tree stem length.    This is important as our yield modelling techniques develop log product yield from tree stem length, not a direct height above ground assessment.   That being there is not an additional 25m (69-44m) of stem length to manufacture into logs for this tree!

tallest_tree2

Figure 2 Showing an example cross section profile of LiDAR data where the resulting CHM would percieve the tree height as 69m due to the sloping ground, but this is more like a tree stem length of 44m

lidar_chmoffcliff

Figure 3 An example of a plantation where all trees are the same age and most are ~40m in height, but 2 trees show heights of ~70m in the CHM (visualised as the normalised point cloud).  Looking in the image to the right this shows the point cloudn in 3D and you can see the trees are simply leaning off a cliff.

There is good work happening in this space and research is evaluating methods which “may” help in adjusting for slope in CHM models.   Interpine will continue to follow this and see if this can be adapted into our work flows in the future.   For now however something to be aware of.

Understand Limitations and Work Effort in Classifying Point Clouds

The second point is that when we are working with a CHM we need to think;

How will we query and use the CHM for information and what will be our area of interest when we look at the CHM?

This is beacuse there might be more than trees in our CHM!  Depending on the extent in which point cloud classification has been carried out (ICSM Level 1 through 4) we are likely to find vehicles, bridges, power lines, buidlings, water tanks, and even people in the CHM.   Normally this would be easily filtered out if we conducted Level ICSM 4 standard classification, but this is costly.

non_trees
Figure 4 – Example of powerlines, a vehicle and a bridge classified and extracted from a LiDAR dataset.

According most forest projects will be using Level 1 and 2 ICSM classfication standard and not getting a human to review the point cloud automated classification to ensure that all these features are removed from the vegetation class. The end result means our CHM will include these features !

chm_objects
Figure 5 – water tanks and a logging truck inside the vegetation classifised point cloud (resulting 1m CHM on left, and 3D view of point cloud on right.   Note the powerline in this case was picked up during automated Level 1 ICSM classification.

So does it matter?

This depends on the area of interest (AOI).  Most of the time when calcuating yields or using CHM to derive statistics on our area of interest, this will exclude roads or we often buffer powerlines and cable ways and remain focused on the net stocked area of plantation.  Therefore the extra work effort to delete or clean all the non-tree data in the full data set is often not justifised.   Why would we worry about trucks classifised in the vegetation cloud on a roadway when we will never likely query the CHM or vegetation mertics in meaningful way on a roadway!

Figure 4 shows another example of a powerline which is classified as vegetation by Level 1 ICSM classification workflow, but is clearly outside of the of the net stocked area.

powerline_chm2Figure 6 – a powerline appearing in the CHM as if vegetation, but clearly buffered out from the surrounding net stocked area.