Handle bias adjustment factors outside of hard-coded threshold

[btobers]

Handle gcm bias adjustment factors which fall outside of hard-coded thresholds in an alternative way. As currently coded, there's an assertion that bias adjustment factors be 0.5<x<2. Perhaps instead of an assertion, a warning should be raised, and the gcm bias adjustment factors can be returned to run_simulation.py so that they can be recorded during runs.

Originally posted by @Ruitangtang in #116

[drounce]

I think this is a sensible way to proceed. I would keep an error for maybe some larger values (e.g., >10) just to ensure the model remains physical, but clearly these bounds are a bit too strict for Ruitang's application.

[ehultee]

If you're going to update that code, just a suggestion: the assertion will be a bit more flexible if you compare the bias-corrected precip versus the raw GCM precip, rather than bias-corrected GCM versus historical baseline. Especially true if you intend to run far-future (multi-century) simulations where the climatological baseline may have changed a lot. 🙂

that is:

gcm_prec_biasadj_frac = gcm_prec_biasadj_subset.sum(axis=1) / gcm_prec_nospinup.sum(
        axis=1
    )

replacing

gcm_prec_biasadj_frac = gcm_prec_biasadj_subset.sum(axis=1) / ref_prec_nospinup.sum(
        axis=1
    )

[drounce]

Great point! @btobers what do you think?

[btobers]

Great point! @btobers what do you think?

Yes, I think that's a great idea. I'll work on an update to resolve this. Thanks for the suggestion @ehultee.

[btobers]

I've drafted a PR to address this issue, doing the following:

1. As @ehultee suggested,

   gcm_prec_biasadj_frac = gcm_prec_biasadj_subset.sum(axis=1) / ref_prec_nospinup.sum(
           axis=1
       )
   

   was replaced with

   gcm_prec_biasadj_frac = gcm_prec_biasadj_subset.sum(axis=1) / gcm_prec_nospinup.sum(
           axis=1
       )
   

2. The gcm_prec_biasadj_frac is now returned to run_simulation.py, where an warning is raised (does not cause the run to stop) if the factor is less than 0.5 or greater than 2.

I'll note that at least for one test case, this calculation method is a bit more sensitive. For instance, in the PyGEM-notebooks simple_test.ipynb, running Khumbu for CESM2 ssp245 gave a precipitation bias adjustment factor of 0.98 with the previous approach, and now returns 0.42, therefore raising the implemented warning:

/Users/btober/Documents/tmp/PyGEM/pygem/bin/run/run_simulation.py:648: Warning: GCM precipitation bias adustment factor for 15.03733 greater than 2 (0.42)
  warnings.warn(
Total processing time: 0.9481937885284424 s

I'll leave this PR as a draft for now. Happy to hear any thoughts on these modifications.

[btobers]

This issue was addressed in PR #119, commit #4be2186

The check on bias-adjusted GCM precipitation was left such that the bias-corrected GCM precipitation was compared with the reference climate precipitation for the same time period. However, the previous assertions were removed so that if the bias adjusted precipitation differs from the reference climate precip by >2x a warning will be raised and the simulation will continue.