The capture and use of high density LiDAR data across forest plantations requires a good understanding of the final output. Sometimes this output is only a digital terrain model and or other times it is to quantify forest attributes, such as mean top height, DBH, basal area, volume and stocking.
A poorly designed workflow can lead to inaccurate outputs. The aim of this article is to determine how the returns classified as noise inside the canopy can affect the sub-canopy definition. The number of misclassified returns over the stems can affect the final outputs.
The following images show the impact of having returns classified as noise inside the canopy when a vegetation classification would be more appropriate. Inside the canopy it is easy for returns to be classified as class 7 (noise), as algorithms are passed over the data looking for lone returns with little or no neighbouring occupied data. However tree stems in an open forest is a good example of when this approach can cause loss of data for those of us interested in finding and using this data.
The original dataset looks like the image below with part of the stem classified as noise (pink colour) (Figure 1).
Figure 1: LiDAR profile by class
To determine the area inside the canopy, a digital terrain model (DTM) in blue and a digital surface model (DSM) in green have been derived (Figure 2).
Figure 2: Point cloud with DTM and DSM
Once this area has been defined, the noise inside the canopy can be reclassified as class 19 (black colour). See figure 3.
Figure 3: Noise reclassify as class 19
Figure 4 shows a single tree with reclassified noise inside the canopy as class 19 (orange colour)
If the misclassified returns are not found during the quality control / quality assurance process, the final deliverables can impact on the value of the forest attributes.