Underestimation of DSM for canopies using the OT Notebook

[Link-dev]

Thanks for providing so wonderful tools for processing USGS 3DEP lidar data.

But I recently found that the CHM produced by the tutorial underestimated the canopy height significantly. The sites I tested were spatially distributed in the continental US to cover most vegetation types. The conclusion is that the median and the top 5% of the CHM data produced by the tutorial are only at ~4.5 m and ~14.5 m. Compared to other global canopy height products covering the same areas (median: ~16.0 m, top 5%: 31.0 m, https://code.earthengine.google.com/?scriptPath=users/sat-io/awesome-gee-catalog-examples:agriculture-vegetation-forestry/GLOBAL-1m-CANOPY-HEIGHT), these values are extremely low.

I set the point cloud resolution, dsm resolution, dtm resolution to 2.0 m, 10 m, 10 m, and other parameters remained the same as the default.

I wonder if there is anything else that need to be adjusted carefully before applying this tutorial to generate CHM products to address this underestimation issue?

[dshean]

I believe this is related to how the DSM is prepared. The PDAL pipeline uses idw of all returns, which will bias the DSM surface low over vegetation and other surfaces with multiple returns. The DSM settings should be limited to 'first,only' returns. This was an issue for others in our group using the OT notebooks, but I'm not sure it ever made it upstream. If necessary, we can open a separate issue, as I don't think it's specific to CHM, but the default DSM PDAL pipeline.

[ShashankBice]

Yes, David is correct. I am attaching a modified DSM pipeline, which you can adapt for your test site. Specially note the filters.return and group filter for first,only. We would be interested to know what you find for your CHM values with the updated pipeline.

pipeline_dsm_0000.json

I agree, we should make a separate issue and PR to update the upstream OT notebooks for DSM creation!

Cheers,
Shashank

[Link-dev]

Thanks so much, David and Shashank! It does work for me. I have changed the title to relate this issue to the DSM to avoid confusion. If neccessary, we can open a seperate issue.

A follow-up question is that when I use the "{"type": "filters.returns", "groups": "first,only"}" to filter first only returns, I found the pipeline with this operation cannot be excuted in streaming mode. So, I replaced it with "{"type": "filters.range", "limits": "ReturnNumber[1:1]"}". I wonder if these two operations output the same DSM or not?

I did try it at one test site, they will output similar CHM but not exactly identical.

[nunderwood6]

Is adjusting the DSM PDAL pipeline with ""{"type": "filters.range", "limits": "ReturnNumber[1:1]"}" the recommended approach now? Because I'm working with a large study area executing the pipeline in streaming mode is a must for me.