Using LiDAR to Identify Tree Species

The city of Seattle just had a plane fly over it, emitting lasers to gather a 3-D point-cloud image of the entire city. This is called LiDAR (it’s also a technology that’s being used in “self-driving cars”.  During a meeting with the company supplying us the data, it was mentioned that tree species could be identified using the data they pulled. I became a bit curious as to how tree species are identified via LiDAR.

One document I found was a doctorate dissertation by Sooyoung Kim titled “Individual tree species identification using LiDAR- derived crown structures and intensity data”. Although it was written in 2007, very old for the tech field, I believe the fundamentals will still be relevant. I picked this one because it was the only paper I found that included multiple methodologies for tree species identification (and perhaps I have a soft spot for other UW students). As a bonus, the data for this study came from the Washington Park Arboretum in Seattle – so it is likely that the methodology that worked in this study will also work in the rest of Seattle due to the fact that there will likely be many of the same species.

So, let’s jump into it! The paper outlines 3 different methods for identifying vegetation:

  1. Individual tree species identification using LiDAR intensity data
    1. Light intensity data – that is the intensity of the light that bounces back to the LiDAR device – is typically the quickest way to distinguish conifer from deciduous. Intensity of conifers hovers around 30% whereas broadleaved deciduous trees often have intensities of closer to 60%. This approach works best for leaf-off datasets, however… We have leaf-on data.
    2. Therefore, LiDAR intensity data has stronger potential for species classification when augmented with 3-D structural data of the tree. Having both leaf-on AND leaf-off data can improve the accuracy of this methodology.
    3. There are methods for isolating individual trees. Once these are used, each individual tree can be tested for height, crown diameter, stand level estimates, biomass, and stand volumes. To be honest, I don’t understand every word in this, but they are using the LiDAR data to automatically read the shape of the tree. This will be covered more thoroughly in section 2, below.
    4. This study got better results in March than in August. Also, they ended up using light intensity data to classify trees because they realized they could – even though it did not appear to be in previous literature mentioned in the article. So that’s cool!
  2. Individual tree species identification using LiDAR-derived structure measurements
    • Crown Structure Measurements
      • Height and crown diameters
    • Vertical Distributions of Laser Returns
      • Picture a cone. Picture passing LiDAR over a cone. You’re going to get many more laser point returns from the base of the cone than the top of the cone. So, the “Vertical Distribution of Laser Returns” assigns a height to each return and summarizes the tree by the number of returns per height range. I hope that makes sense… the explanation on page 68 confused the heck out of me until I read it several times.
    • Upper Crown Shape
      • This methodology is required for trees that are touching or close together… The portions of the tree that are touching another tree are ignored and only the upper crown is considered for measurement.
    • CONCLUSION: “The relative height percentiles could not separate broad-leaved and coniferous species although they could explain some structural characteristics for specific species. The length to width ratio indicated significant differences between broad-leaved and coniferous species.” Leaf-on data provided better classification accuracy.
  1. LiDAR-based species classification using multivariate cluster analysis
    • Supervised vs Unsupervised Classification
      • Supervised: the species are grouped in advance, and are used to classify future observations
      • Unsupervised: clustering is used to find and group similar data (trees)
    • This portion of the study was done using intensity data. The author recommends that people use both a leaf-on and a leaf-off dataset.

The data for the flyover we received was taken in late summer, with leaf-on conditions.  The company supplying us the data claimed to be very confident that they could supply us with accurate species identification. It will be interesting to ask them what may have changed since 2007 to ensure that we get the right information with only a leaf-on dataset (no leaf-off dataset). I am inclined to trust them, believing that their confidence was not a marketing ploy. Time will tell!

Leave a Reply

Your email address will not be published. Required fields are marked *