Charlie Zender

Responses interleaved: > You explained that the quantization does not reduce dataset size. So do we envision for CMIP the following? > > a data provider bitgrooms an array using...

Oops forgot to attach poster mentioned above... [pst_cf_agu_202412.pdf](https://github.com/user-attachments/files/18188658/pst_cf_agu_202412.pdf)

@mauzey1 Your understanding is correct. The CF Convention requires that there be a quantization variable that contains the algorithm name and its implementation, regardless of whether the quantization levels differ...

Good question. The only algorithm for which we know a simple analytic formula for `MRE` is `bitround`. The other algorithms (`bitgroom`, `granular_bitround`) have analytic formulae for Maximum Absolute Error (MAE)....

NCO implements the MRE field for `bitround` and does nothing for the other quantization methods because it's unclear if/how to summarize the MAE results in a single, scalar value.

@mauzey1 I'm not that familiar with the Python-language netCDF library wrapper `netcdf4-python` so I had to do some reading. The wrapper includes a method named `quantization` with this prototype: `def...

This is a more exhaustive list of where the SB constant would need to be changed to ensure consistency with RRTMG(P) and/or CODATA: ``` zender@spectral:~/e3sm/components$ grep -rn '5.67e-8' . ./eam/src/physics/crm/samxx/test/shr_const_mod.F90:21:...

@proteanplanet good catch with the search for `567.0e-10`. My search included all the submodules that I know about, to wit: ``` zender@spectral:~/e3sm/components$ ls cice/ CMakeLists.txt eam/ elm/ mosart/ mpas-framework/ mpas-seaice/...

No, it's the exact value. We ran a short test with the exact value and it reduced the bias by a factor of ~30x, from 0.02 W/m2 to 0.0008 W/m2...

@cbegeman Thanks for leading this effort. The 100yr LR and 5 yr HR simulation seem sufficient to me. I do not know of any other tests that would need to...