Re: Recommendations and best practices for DEM & DSM production using LAStools

[Martin Isenburg]

Hello Matt,

> I'm hoping to get some insight to the processes and best practices involved
> in creating high quality Digital Elevation Models (bare earth) and Digital
> Surface Models ("first return" or including trees, buildings, etc...) from
> LiDAR data using LAStools.

Thank you for raising this question. The lack of a proper "tutorial
document" makes it sometimes a bit tricky to find this information. It
is there but spread out between the README files, this user forum [1],
the facebook page [2], the twitter feed [3], the linkedin group [4],
and the USGS CLICK bulletin board [5] which unfortunately was taken
down by the authorities for - as far as I understand - matters of
national homeland security ... (-;

[1] http://groups.google.com/group/lastools
[2] http://facebook.com/lastools
[3] http://twitter.com/lastools
[4] http://www.linkedin.com/groups/LAStools-4408378
[5] http://lidarbb.cr.usgs.gov

> The las files I am working are supposed to
> conform to the current USGS-NGP LiDAR Guidlines and Base Specification.

I am not entirely familiar with this specification but Karl Heidemann
who has written it and is listening in on this group could summarize
the key points for us.

> For DEM:
> Can I just use all model keypoints (class 8) to create a DEM and expect
> acceptable and accurate results?
> Would I get anything out of including points classified as bare earth
> (class 2)?

For the LAS/LAZ files I am working with I use the ground and - if
available - the key points (also called mass points) to create the
bare-earth DTM. You can do so with las2dem (up to 30 million points)
or with blast2dem (up to 2 billion points). Note that lasgrid is
usually not the right tool as it does not interpolate across larger
voids that you tend to have in ground classified LiDAR (unless its all
nature and there are no buildings or similar that were filtered out).
Here are the typical examples commands with LAStools

las2dem -i lidar.las -keep_class 2 8 -extra_pass -o dtm.asc -step 1
blast2dem -i lidar.las -keep_class 2 8 -o dtm.asc -step 1

> Do I need to worry about overlap, return number, intensity values, and/or
> scan angles?

The only thing you may have to worry about is the overlap. In case you
have data from multiple flight lines that were ground classified
separately and are not perfectly aligned you may get what I call
"spargel-felder" (white asparagus fields) in your DTM because the
ground points may jump up and down in the overlap areas. In this case
you can either run lasground on the unified data set or - probably
better - mark the "overage points" and leave them out of the DTM
generation. More info on that in my recent post on the new lasoverage
tool:

http://groups.google.com/group/lastools/browse_thread/thread/47bb2ed40225c3d8

> Are there any special considerations if I want to include points classified
> as water (class 9)?

I see no harm in doing so if you have "good" water returns and no
other information about your waterbodies at hand. But often people use
hydro breaklines around lakes and other waterbodies to "flatten" these
areas with a known elevation. How to do this with las2dem is described
here:

http://groups.google.com/group/lastools/browse_thread/thread/a5acd87f03696f83/95a603209ceb7b2b

> For DSM:
> Again, are there recommendations for the inclusion/exclusion of points
> based on flightline overlap, return number, intensity value, and/or scan
> angle?

For DSM production it seems most people prefer to use only the first
returns to lessen the affects of aliasing. If the point density of
your data is much higher than the resolution of the DSM you want to
generate you can also consider to "thin" the LiDAR points. Here an
example:

las2dem -i lidar.las -first_only -extra_pass -o dtm.asc -step 1
las2dem -i lidar.las -first_only -thin_with_grid 1.0 -extra_pass -o
dtm.asc -step 2

> I appreciate any comments, suggestions, and links to appropriate
> resources.  Thanks.

Hope to have answered your (and a few other people's) questions.

Cheers,

Martin @rapidlasso