[FEATURE REQUEST] Output satellite diagnostic as NetCDF

[eamarais]

I've updated GEOS-Chem version 13.0.2 to output the satellite diagnostics (ND51) in NetCDF format by piggybacking on existing code infrastructure in diagnostics_mod.F90, history_mod.F90, history_netcdf_mod.F90, and state_diag_mod.F90. Im finding though that there are small differences between output from my updated code and output from ND51.

I've attached images I've generated that compare the old (ND51) and new (NetCDF) output on 20180102 during 12h00-15h00 local time in the lowest model layer for CH2O, relatively humidity (RH), and the number of data points. The latter was a sanity check to ensure that accumulation of model values was the same in the two diagnostics. It is. Is it possible the difference is due to differences in the timing of accumulating/updating the ND51 versus the History diagnostic routines?

[yantosca]

Thanks @eamarais for doing this. How did you get the local time diagnostics? In the past I would try to keep a counter field and then divide by that in post-processing. I tried doing that for the JValueLocalNoon collection a couple of years back but there were always artifacts that I couldn't remove. If I can find them I'll post a couple of PPTs (but they may be on my work PC back in the office so it may be a while before I get them.)

[eamarais]

@yantosca I can share the code updates as a patch if you'd like. I essentially created a new hour range (hrrange) entry in HISTORY.rc, wrote code in diagnostics_mod.F90 to identify when the relevant fields coincided with the local time range of interest using GET_LOCALTIME and included an IF statement in history_netcdf_mod.F90 to divide the satellite diagnostic arrays by a 3D or 2D array depending on the size of the item. It's probably not eloquent coding, but I think it works (though I'd feel more confident about the code if I could resolve the difference between my updated diagnostic and ND51).

[yantosca]

Thanks @eamarais. The only thing is that we in the past have avoided bringing in any of the GEOS-Chem Classic specific time routines in order to make the HISTORY diagnostics usable by GCHP, GEOS, WRF-GC, etc.

[yantosca]

Here are a couple of presentations I made to our Model Development Group from 2018 about trying to implement the local noon J-value diagnostics in HISTORY (which is similar to the satellite diagnostic problem):

1. MDS 2018-01-05 Yantosca JValueDiags.pdf

2. MDS_23Jan2018_BMY.pdf

[eamarais]

Thanks @yantosca for the slides. When you say "GEOS-Chem Classic specific time routines", do you mean anything in time_mod.F90? I ask because there are routines from time_mod.F90 that are already used in a few HISTORY routines. Or is there something specific about GET_LOCALTIME that would make it incompatible with GCHP etc.?

[yantosca]

@eamarais, I think I mean the local time. Just that GET_LOCALTIME isn't available in e.g. GCHP or GEOS since that uses the MAPL time routines.

[yantosca]

@eamarais, I just took a quick look at the History code. The only time routines used are the ones in julday_mod.F90 and the routine YMD_EXTRACT, which just peels out Y, M, D from YYYYMMDD. History doesn't use any of the time computing functions.

[eamarais]

@yantosca would it help if I rewrote get_localtime in the History code?

[yantosca]

I think it's a bit more complicated than that. The GCClassic get_localtime function assumes a Cartesian grid. But if you go to other models like WRF-GC that might not be the case. Then you'd have to also bring in awareness of where you are in the grid and it gets hairier from there.

Also tagging @sdeastham @LiamBindle who have had some experience w/ 1-D diagnostics.

[sdeastham]

Given that what we want is really a sort of local solar time (as that's what dictates whether or not a satellite is overhead), would it be safer to ask for the UTC time and then just subtract longitude (in deg)/15? That data should be available from every wrapper. Sorry if I'm totally missing the point here!

[yantosca]

Thanks @sdeastham. I think it goes to whether or not we want to bring time & grid awareness into the HISTORY diagnostics, or to handle that in post. Our philosophy had been to do it in post, I think.

[eamarais]

@yantosca I only use get_localtime in GeosCore/diagnostics_mod.F90 in the update. I read the local time values defined my the user in history_mod.F90

[yantosca]

@eamarais Oh OK. Then that would be OK if its outside of History.

[sdeastham]

Will this approach work in GCHP, do you know (by which I mean - will this diagnostic be accessible from GCHP)? We were discussing this exact problem (satellite diagnostics) with the MAPL team just Friday.

[eamarais]

@sdeastham, I can't answer that, as I'm not familiar with the different approaches of the two models that might cause issues. I can share the code as a patch if you'd like to see whether it would work.

[sdeastham]

@eamarais that sounds great! Unless @yantosca you happen to know offhand if this would/would not require modification?

[yantosca]

I would say send us the patch @eamarais and we can take a look. (But probably not at least till next week after the holiday here.)

[eamarais]

@sdeastham, @yantosca, I'll share the code as a patch (give me a few minutes to figure out how to do this...). I've uploaded the HISTORY.rc file with the satellite output (SatDiagn) collection. The file is uploaded as a text file to accommodate limited file format options. The SatDiagn diagnostics aren't a complete list of what's in ND51. For example, it doesn't include AOD, but, if my modifications are appropriate for general use across all models, the framework is there for others to add more entries.

HISTORY_rc_with_SatDiagn.txt .

[eamarais]

@yantosca, @sdeastham, patch attached. The parent commit is in the attached file. Again, converted to text file so that it's compatible. Odd oversight that Git doesn't support this file format, unless I'm missing something.

The changes to diag51_mod.F90 aren't relevant. I was playing around with array types to try resolve the differences in the plot I shared earlier.

netcdf-sat-diagn-diff.txt

[eamarais]

@yantosca, @sdeastham, I found the cause for the difference between the NetCDF and ND51 diagnostics. It's a minor update. I was subtracting the dynamic timestep from the local time, as in ND51, but this isn't necessary for the NetCDF satellite diagnostic, presumably because of when it's called in the code. Updated patch attached. This merges both my commits.

netcdf-sat-diagn-v2-diff.txt

[msulprizio]

The pull request for this feature has now been merged into dev/14.1.0 and will be included in that version release.