SSA rasters

[brentwilder]

**Updated 12/11/2023

General things

• Closed out the CHM bug. I also realized we were not saving .TIF to disk? So I have fixed this.
• We should check laz2dem and make sure we mean to have 1m as the IDW interp in PDAL.

SSA specific things

• Things work! Biggest bottleneck was running PDAL 26 times. We only run it once now by normalizing reflectance by incidence angle to start, and then using PDAL merge.
• This uses subprocesses call to access R script (independent of conda). So I believe the set up would be to just install base R and the 6 or 7 libraries on whichever machine this runs on.
• The SSA equation via asymptotic radiative transfer (ART) model boils down to one long equation that combines all of the various rasters. So it is relatively speedy.
• This requires the user to input the .LAS with extra bytes (this is not what is saved on SNOWDATA). However, Josh was kind enough to export the flightlines from Feb 9 with the extra bytes. So, there is a check now in the pipeline that iff this SSA commands are true, it must be the extra-bytes version of the LAS file (that contain "Intensity as Reflectance=TRUE")
• To calculate incidence angle, we take the approach of assessing the vector between the helicopter position and the ground position with respect to surface normal. In order to align the two temporally, we must match them in the datatable. Onboard IMU tracks location every 0.005sec. To make the best match, I use roll="nearest" (https://search.r-project.org/CRAN/refmans/data.table/html/data.table.html). This may result in temporal interpolation error that could result in errors on the order of tens of cm depending on the velocity and/or acceleration of the helicopter.
• As an argument, the ice-road-pipeline takes a known intrinsic reflectance. For our case, this is the known reflectance of the asphalt for a stable area in Eagle, ID I have defined. I have wrote this in a way that it can really be any stable target, as long as the supplied shp file falls within the calibration .LAS and the intrinsic material reflectance is known.
• SSA results are still unvalidated but I am working on this to try and wrap it together to a nice poster at Western Snow and upcoming paper!

[brentwilder]

@shadoneel @ZachHoppinen

Thank you Shad for the email questions this morning! To summarize,

What is the canopy height model bug?

• Put simply we were not writing the file to disk.

What about SSA - what does that stand for? I can weave together what you are working on, but want to make sure im tracking. Will this be default in any ice roads run that happens if we migrate to the new version on kelvin, or is it a flag? How intensive is the computation?

• Specific Surface Area (SSA) which is similar to the optical grain size of snow by the density of ice. In other words, it tells us about the grain size at the surface.
• And not to worry about previous workflow, this will only be ran if the flags for SSA are supplied. You can still run normal runs to just compute snow depth rasters.
• Computation is fairly intensive simply because PDAL IDW has to be called for each flight line individually (to preserve incidence angle and reflectance information that is flightline specific).

• Additionally, as a note to Zach: the SSA processing chain will spit out for example 13 SSA rasters in a new directory. I am not merging them into one raster at this time, simply because the accuracy of retrieval is a function of the incidence angle. And so mosaicking them just doesn't make sense. Much of my analysis will be studying this increase in uncertainty at the edge of flight lines..

**EDIT: I found a much faster solution to have 1 normalized reflectance raster, leading to 1 SSA raster. This preserves the incidence angle information and reduces computational load x10. I have already output all of my incidence angles for my analysis and so this is really just an extra thing.

[brentwilder]

The biggest bottleneck now is computing calibrated reflectance for each flightline (roughly 5 min per flight line). But still is not bad. The other bottleneck is merging and performing IDW on one LAS file (but this is unavoidable and arguably the fastest possible option).

[brentwilder]

I'm going to test it soon to make sure everything still works after the changes, but thanks so much for the comments. I added logging for the main R functions, and streamlined pipeline to call a function from las2grain.

I didn't add the R packages to the yml file at this time because I feel like it may not be as helpful for folks using it right now. However, I realized the README was a bit frightening with the amount of detail on grain size. I dialed it back a bit and so now it is there and people can look into it more if need be.