How does one characterize habitats using lidar data (LAS files, LAS datasets, xyz files)?

3664
9
Jump to solution
09-12-2014 10:24 AM
Highlighted
New Contributor II

Hi there,

 

I have a project in which I have been assigned to characterize different plant habitats (fern, grass, forest, and barrents) on a small island, using LiDAR data. The data I have consists of LAS and XYZ files, in addition to TIF and contour shape files I am already in possession of.

 

Does anyone know how I got from LAS datasets (or any derivative of the files I have) to start to characterize habitats?

 

Any and all help is greatly appreciated,

 

Donald Pirie-Hay

1 Solution

Accepted Solutions
Highlighted
Frequent Contributor

The classes are built into las files, but not ever company has a good algorithm for putting the points into classes, so its not surprising that you don't have many points in classes or that a large number of classes are empty, but I figured it was worth asking.

As far as your plan, thats what I would likely do in your scenario, generate the DEM from all ground class returns(almost every company has an algorithm for classifying points as ground, since LIDAR is so widely used to generate DEMs, and you want only ground points to be used to make DEMS), and the DSM from all first returns, then extract the height information from subtracting the two(Giving your a height model of features).   You will are a bit on your own defining what classifies as low/medium/high vegetation.  In a study area I work in, I have it at 0-3ft/3-9 ft/>9 ft for low/medium/high vegetation, but depending on your area and the flora, you might want to make your own ranges, or use your ground truth to help determine which ranges work best.

I'm guessing there are other features on the island that you might need to seperate vegetation from.  The intensity values could help you distinguish grasses from bare earth or sand, or vegetation from man-made materials if they are present.

Where exactly is your study area(I'm curious)?

View solution in original post

9 Replies
Highlighted
Frequent Contributor

Hi Donald,

There has been significant research in using LIDAR to aid feature classification in conjunction with other forms of remote sensing (IfSAR, Aerial Imagery, Hyperspectral imagery, etc.), but not a great deal for feature classification using only LIDAR data except in urban settings(generally for extraction of building footprints), or in some cases in forested areas for road feature extraction.

This is somewhat similar to a project I am working on, do you have much experience working with Lidar data, and what sort of data do you have in your tifs?

Reply
0 Kudos
Highlighted
Frequent Contributor

Sorry my earlier reply was not quite as helpful as I meant it to be.

This is a good paper for feature classification from LIDAR datasets, it might be a good starting point for you. 

www.researchgate.net/publication/251601091_Object-oriented_land_cover_classification_of_lidar-derived_surfaces/file/72e7e52150bc74a30a.pdf

I'd review some more literature on the subject and it might give you an idea on where you want to go with it.

Reply
0 Kudos
Highlighted
New Contributor II

Thanks Ian,

I have very little experience working with Lidar data, and as such, I'm really just learning it as I go here. I have a basic grasp of it so far.

I will look into this paper, and other related ones, thank you very much for your help so far. It is greatly appreciated.

Should I have further questions or more specific ones, I will keep them in this thread.

Thanks again.

Reply
0 Kudos
Highlighted
Frequent Contributor

I didn't think about this earlier, but I just got back from a conference and talked with a bunch of guys from the US Geological Survey and they reminded me that alot of .las files now have points that have been classified into different classes, such as low vegetation, medium vegetation, high vegetation, buildings, ground, etc.  If your las files already have these classifications made, then it would be extremely easy to do what you are asking.  Older Las files usually only have points classified as ground or non-ground, but depending on how recent your data is, the points in your las my already be classified. 

What is the source of your LiDAR data?

Reply
0 Kudos
Highlighted
New Contributor II

The LAS files are indeed classified, but as specific as I am looking to do, nor are they 100% correct. The LAS files have classes of low, medium, and high vegetation, but there are only points of low vegetation, and zero returns that are classified as medium or high vegetation. I have been on the island and ground truthed many areas, and there are distinct areas of grass, and others of forested (high trees). Based on this, there should at least be some points classified as medium or high vegetation. The area I am investigating is a small island. The water around the island is also incorrectly classified as ground.

I am not sure if it is even possible to edit or correct the inherent classifications.

At this point, to distinguish habitats, I plan to create a DEM and a DSM, for the exact same area, and the difference between them should give different vegetation levels. Does this make sense?

In creation of a DEM from an LAS dataset, I need to use a filter, but I do not know what points to filter.

Oh, and I am still waiting on the metadata for the LiDAR data, when I receive it I will communicate its sources. Leading Edge Geomatics provided the data.

Reply
0 Kudos
Highlighted
Frequent Contributor

The classes are built into las files, but not ever company has a good algorithm for putting the points into classes, so its not surprising that you don't have many points in classes or that a large number of classes are empty, but I figured it was worth asking.

As far as your plan, thats what I would likely do in your scenario, generate the DEM from all ground class returns(almost every company has an algorithm for classifying points as ground, since LIDAR is so widely used to generate DEMs, and you want only ground points to be used to make DEMS), and the DSM from all first returns, then extract the height information from subtracting the two(Giving your a height model of features).   You will are a bit on your own defining what classifies as low/medium/high vegetation.  In a study area I work in, I have it at 0-3ft/3-9 ft/>9 ft for low/medium/high vegetation, but depending on your area and the flora, you might want to make your own ranges, or use your ground truth to help determine which ranges work best.

I'm guessing there are other features on the island that you might need to seperate vegetation from.  The intensity values could help you distinguish grasses from bare earth or sand, or vegetation from man-made materials if they are present.

Where exactly is your study area(I'm curious)?

View solution in original post

Highlighted
New Contributor II

The study is Baccalieu Island Ecological Seabird Reserve, Newfoundland, Canada. It has the highest concentration of breeding Leach`s Storm-Petrels in the world, and as such, is an extremely important island in various ecological facets.

The goal is to characterize habitats and determine their extent and area to better estimate the amount of Storm-Petrels that breed on the island.

As for the habitat differentiation, ideally it would be divided into Forest, Fern, Grass, and Barren. Whether Fern is consistently high enough to distinguish from grass remains to be seen. There are also a couple of lakes and small rocky areas, as well as the remains of an old lighthouse, but other wise the island consists of the first four habitat types listed.

Thanks again for your advice.

Reply
0 Kudos
Highlighted
Frequent Contributor

Another thought on seperating out grass from ferns, since ferns are more broad-leafed then grasses, they might be more likely to get multiple returns than from grass(usually only get single returns from grasses), so that could possibly help differentiate between the two.  Without actually having your datasets I'm just theorizing, though if you wanted to share you data, I'd have no problem taking a look at it and seeing if we could come up with something. 

Highlighted
New Contributor II

I will take a look at the multiple point returns, good idea.

Thanks for the offer but I am not able to share the data as I am under agreements of its use with my employer.

The LiDAR data was provided by Leading Edge Geomatics.

Reply
0 Kudos