HRRR model levels

[bbuzz31]

Use the native model levels rather than the pressure levels in order to cover a more complete height of the troposphere.

Closes #568

[dbekaert]

Thanks Brett, this looks already much better based on your figure. One of the test seems to have an issue with the definition of the layers.

[jlmaurer]

LGTM. I updated the NotImplementedError message on pressure levels to help me remember in the future why we aren't using them.

[bbuzz31]

@jlmaurer here we transpose the geo_hgt, which then propagates to the z and p

[bbuzz31]

if you don't specify height levels, then all the model layers will be used. the top of the model layer (ht) is by definition = high_ht at the top. So nothing is calculated and then the stacking crashes. It's a very tiny contribution to the integral.

[bbuzz31]

@cmarshak any idea why the GUNW test could be failing?

It writes the first delay successfully but then seems to kill itself during the next datetime:
The end of stdout is as follows:

Starting to run the weather model calculation
Date: 20200124
Beginning weather model pre-processing
Weather model HRRR is available from 2016-07-15 to Present
Weather model HRRR is available from 2016-07-15 to Present

It should continue with, e.g., ✅ Found ┊ model=hrrr ┊ product=nat ┊ 2020-Jan-30 13:00 UTC F00 ┊ GRIB2 @ aws ┊ IDX @

But instead just stops and returns:

subprocess.CalledProcessError: Command '['raider.py', '++process', 'calcDelaysGUNW', '-f', '/root/project/test/GUNW/S1-GUNW-D-R-071-tops-20200130_20200124-135156-34956N_32979N-PP-913f-v2_0_4.nc', '-m', 'HRRR', '-o', '/root/project/test/GUNW']' died with <Signals.SIGKILL: 9>

Maybe a memory issue on circleci? I can't reproduce it locally, and a main change here was to use the product=nat which is bigger

[cmarshak]

I would run the test on a linux server via pytest test/test_GUNW.py and re-install the environment (because I have seen that issue with env issues). I ran the test a few days ago (specifically the GUNW) and it ran to completion (still failed on Circle CI). I don't know what that sig kill means in this context. I have seen that when gdal is not installed correctly or there is an issue with the environment. However, don't know. The integration tests for HRRR take a significant time to run and maybe require lots of memory.

[bbuzz31]

Thanks yeah I think it has to be something weird with the memory. If I just run with HRRR (and not GMAO) it works.