Lidar Workflow for Classification Needed

[GeorgeDurkee1]

Hi All.

What's the best workflow for classifying lidar point clouds? When I decompress to LAS and, in ArcPro 3, create an LASD then run statistics, I get Ground, unassigned and, maybe, towers or something else but not vegetation heights or buildings. Should I do building footprints first, then classify the rest by height? Is the Ground tool better than or different from how Ground is created when calculating statistics and should I just reclassify everything to unassigned and then just go one classification type tool at a time?

I'll eventually want to estimate density of veg for different height classes and export to a raster. How do I eliminate multiple sampling from adjoining flightlines? I know there's a tool for that and it requires tiling. Where would that fit into a workflow. So:

LASD ==> Eliminate overlapping flightlines/oversampling (??) ==> Classify Ground ==> Classify building footprints (??)  ==> classify buildings (??) ==> classify by height for vegetation.

Or can footprints & buildings be done after veg.

I have successfully derived DEM, DSM, and nDSM. How would those best fit into the above for analysis and/or classification if at all.

Is LASTools better for this?

OK. Hope that makes some sense. I've looked for a beginner's workflow on the web but only finding bits and pieces and, of course, in such demos, everything always works -- which is usually not my experience without a lot of flailing. I appreciate any guidance folks can give me!

Thanks!

George

[CodyBenkelman · Accepted solution]

George

There are a lot of questions and assumptions in your post, so this will be more than a "one answer and done" response.

1. for a very general response, please review ArcGIS Pro Help system, and also note that the Imagery Workflows documentation site has advice on best practices (expanding on the core help topics).  You'll find resources there to provide a good background for questions such as this.
2. specific to your question, I'd advise you classify ground first, then buildings, then vegetation but there are many footnotes. 
   1. if your data already has ground points classified, you should consider carefully before altering those class codes (depending on the source).  If you feel the current Ground class is significantly in error, then yes you can "reclassify everything to unassigned and then just go one classification type tool at a time"
   2. You ask "is the Classify Ground tool better than or different from how Ground is created when calculating statistics?" but the latter does not perform classification - it's simply summarizing existing classifications
3. for your question re: vegetation density, one method (perhaps not the only way to do it) would be to generate rasters for POINT density and also PULSE density, then use the latter to normalize the former.  (Where you have extra points in the overlap, the PULSE density will quantify the extra sampling).  See discussion of those two QC rasters here (from the workflows site)
4. regarding building footprints, that is not a separate classification code in the lidar - you'd use the building classified points to generate polygons.  See this blog.

I'll forward your questions to some of my colleagues and they may have further advice for you.

Cody B. 

[GeorgeDurkee1]

Cody: This is great. Many thanks! Among other things in your answer, it's good to know some lidar sets come with ground classified. That seems to be true of the one I'm using (2021 of a county) so, as you suggest, probably no need to reclassify. The DSM I created seems to match well with the terrain.

Anyway, if your friends have any other suggestions, I'd really appreciate them. Fortunately this isn't part of my job, but lidar gives some great ability to visualize and analyze terrain. Other than the learning curve, it's kinda fun... .

Thanks again!

George