Fix bugs/issues in metallicity.py

[hzovaro]

There are currently a few (relatively minor) issues with the implementation of strong-line gas-phase metallicity diagnostics:

• [x] There appears to be a bug in the implementation of the Rcal/Scal diagnostics such that spaxels are being incorrectly assigned to upper and lower branches.
• [x] In a number of diagnostics there are finite metallicity measurements when the associated line ratios are outside the range of acceptable values.
• [x] The code needs general double-checking to ensure the correct functional forms, line ratios, etc. are being used.
• [x] The R23_C17 diagnostic has not been implemented.
• [x] In R23_KK04, there are points with R23 ratios lying far above the polynomial curves near the upper/lower branch division.
• [x] Repeatability for metallicity measurements calculated using the MC approach needs to be verified.
• [ ] Fix the KD02 and R23_KK04 diagnostics to use only the OII3726 line in the denominator rather than the doublet flux. Can test this by plotting the R23 diagnostic from K19 (which we know uses the doublet) against the KD02/KK04 ones.
• [x] Clarify in the wiki which definition of [OII] we are using for these affected diagnostics.

[hzovaro]

In a number of diagnostics there are finite metallicity measurements when the associated line ratios are outside the range of acceptable values.

I think I know what's happening - when errors are computed via the MC routine, if the line ratio is just outside the acceptable bounds, then during some of the random iterations it's bound to jump back into within the acceptable range, in which case a finite metallicity is returned. In this case the final value will be the average of these "valid" values & will therefore be finite even though the line ratio in the DataFrame is in the invalid data range.

The question is, is this desired behaviour? Or do we not want to calculate metallicities for these spaxels at all?

Possible ways to deal with this:

• Set some threshold N such that if fewer than N% of individual MC iterations return non-NaN metallicities, then treat the metallicity measurement as unreliable.
• Create a new flag such that if the line ratios recorded in the DataFrame lie outside the range of valid values specified for each diagnostic, then mark the metallicity measurement as unreliable.

[hzovaro]

I have addressed the above issue by setting a threshold 50% fraction of NaN values below which metallicity/ionisation parameter measurements will be discarded.

[hzovaro]

In R23_KK04, there are points with R23 ratios lying far above the polynomial curves near the upper/lower branch division.

This issue was resolved by setting the 50% NaN threshold as above.

There appears to be a bug in the implementation of the Rcal/Scal diagnostics such that spaxels are being incorrectly assigned to upper and lower branches.

This was a simple typo and has now been fixed.

[hzovaro]

Checking functional forms, limits, etc:

• [x] N2Ha_K19
• [x] S2Ha_K19
• [x] N2S2_K19
• [x] S23_K19
• [x] O3N2_K19
• [x] O2S2_K19
• [x] O2Hb_K19
• [x] N2O2_K19
• [x] R23_K19
• [x] O3O2_K19
• [x] S32_K19
• [x] N2Ha_PP04
• [x] N2Ha_M13
• [x] O3N2_PP04
• [x] O3N2_M13
• [x] R23_KK04 NOTE: their eqn. 10 indicates that only the 3727 line should be included in the denominator - should we change this in the code as well?
• [x] O3O2_KK04
• [x] ~R23_C17~ decided to remove this as there are already two more R23 diagnostics
• [x] N2S2Ha_D16
• [x] N2O2_KD02 NOTE: it's unclear whether their parametrisation includes both lines in the 3727,9 doublet or just 3727. R23_KK04 seems to imply that it's the latter, so we should probably change this in the code.
• [x] Rcal_PG16
• [x] Scal_PG16
• [x] ONS_P10
• [x] ON_P10

[hzovaro]

Repeatability has been tested (see attached). In general, repeatability is very good - well within 0.01 dex when 1000 MC iterations are used. metallicity_repeatability.pdf