Update

[amdasilva]

Background

When implementing the AOP (Aerosol Optical Properties) module in pyobs it became evident that existing ordering of the array dimensions in the netcdf Mie LUT tables was somewhat inefficient. This prompted the adoption of a revision of these tables.

The versions of the LUT does not refer to the revision of the values in each table, but rather to the structure of the netCDF files containing these tables

v0.0.0: original file structure, e.g.

qext(radius, rh, lambda) 
pmom(nPol, nMom, radius, rh, lambda) 

v1.0.0: dimension ordering has been updated for cache optimization (use case: radiation and RT calculations).
The following dimensions have been renamed: lambda --> wavelength, radius --> bin.

qext(bin,wavelength, rh)
pmom(bin,wavelength, rh, p, m)

This utility can be used to convert v0.0.0 files to v1.0.0:

https://github.com/GEOS-ESM/GMAOpyobs/blob/develop/src/utils/aop_0to1

New Tables on Discover

Optical tables v1.0.0 can be found on discover at:

/discover/nobackup/projects/gmao/share/dasilva/fvInput/ExtData/chemistry/AerosolOptics/v1.0.0

Update GOCART to work with MieTables v1.0.0

The Fortran code in Process_Library/GOCART2G_MieMod.F90 needs to be updated to work with the dimension ordering of tables v1.0.0.

Testing

This is a zero-diff mod. Start with GOCART under develop, which reads v.0.0 tables, make a short one day run, saving aer_Nx files. Modify the code to work with tables v1.0.0 and make sure the aer_Nx files are zero-diff.

[weiyuan-jiang]

I cannot find the variable qext in the develop branch. @amdasilva

[weiyuan-jiang]

@amdasilva , Is it just pmom ? I think all the variables should change.

[weiyuan-jiang]

In this data structure block, all the variables' dimension reflecting the order in the file. Should we change the pmom too ?pmom

I suggest pmom ( m, p, r, c, b) @amdasilva

[amdasilva]

@weiyuan-jiang Just now I saw your questions questions. I typically filter out all GitHub email. Answering your questions:

• I cannot find the variable qext in the develop branch
  • Answer: this variable is in the file, it could have been an omission
• Is it just pmom ? I think all the variables should change.
  • Answer: Yes, ALL variables must be changed to conform to the changes in the dimension ordering
• I suggest pmom ( m, p, r, c, b)
  • Answer: This is the correct order in FORTRAN. As you know, NetCDF uses C conventions for dimension ordering. The code should follow the same data layout as in the file. Of course, going from C to Fortran the dimension ordering is reversed.

[weiyuan-jiang]

@amdasilva , I think in C's order, the two variables' dimension order in v1.0.0 should be changed to
pmom( m, p, bin, wavelength, rh)
pback( p, bin, wavelength, rh)

In C, the last dimension is continuous and faster to access  .