So I've been working on creating digital terrain models for the Costa Rica data that was collected in the winter. One of my goals is to ensure that the DTM's we create are of the best possible quality.

I am slightly frustrated as the suite of tools we use for filtering ground points from non ground points is not perfect. I can get a pretty good filter that captures mostly all ground points, but there are some big patches of canopy that do not get filtered out. This would noticeably affect the final result of the DTM generation, and to me represents an issue of data quality.

My thought was to manually remove those patches of canopy, but that can obviously lead to all sorts of debates on subjective vs objective treatment of the data. But then again there is no perfect algorithm to do this sort of task. There are trade-offs between each methodology, but I'm not sure if there is a clear answer.

Views: 250

Reply to This

Replies to This Discussion

I think you are right Andrew, there is no clear answer.  

I have looked at and tested a few of the other DTM filtering algorithms, including the one you are using, ALDPAT that includes multiple algorithms, MCC-LIDAR, and the terrain filtering tools in LAStools for ArcGIS.  All of these algorithms always require some tweaking and tuning by the operator.  It is often the case that even with tweaking and tuning of parameters, some points just don't get removed, no algorithm is perfect for this.  I have read in several journal articles that even after tweaking, tuning, or 'optimizing' the filtering parameters, manual editing may still be required to get the best results.  Commercial LIDAR contractors typically use several proprietary algorithms and then have teams of analysts remove points that weren't removed by the filters.

We use several statistical and DTM filtering stages in our Ecosynth pipeline and have not in the past used manual filtering, even though some 'good' tree points are either incorrectly removed as noise or classified as ground and retained. 

Perhaps an approach would be to produce a DTM filtered point cloud as an L1 product and then produce an L2 product with manual filtering applied.

Do you have any screen caps of the work so far?

This is the most evident example I can give of the issue. The area highlighted in red is one patch of forest, as well as the area in yellow. You can see in the red AOI there is a large swath of points that aren't filtering out via AldPat.

There is a bit more ambiguity in the yellow AOI, and I wouldn't be as confident to manually remove points. But certainly in the red patch it is very obvious that it is all forest, and a lot of those points are not filtered out.



© 2019   Created by Erle Ellis. Content is made available under CC BY 4.0.   Powered by

Badges  |  Report an Issue  |  Terms of Service