Unify extents of colorbars in comparison plots

[mahf708]

In comparison (i.e., model-vs-model or model-vs-obs) plots, the default non-specified behavior is often unsatisfactory, and the current approach in this repository is to manually set/specify "sane defaults" for each derived (or raw) variable in the ini/cfg files. See the first example below for one derived parameter I've recently aded as part of the aerosol working group efforts.

A three-panel figure without specifying the colorbars


In order to fix the plots, I had to manually set contour_levels and diff_levels so that the colorbars are more informative. See below.

A three-panel figure with specifying the colorbars


In this particular case, I made two quick edits.

1. Because the default scientific notation doesn't appear readily for the min, max, and mean fields, I decided to convert this to a unit where it will appear more readily.
2. Because I found the unspecified defaults for colorbars to be misleading, I set contour_levels and diff_levels such that: a. The top two panels share the same minimum and maximum; and b. the third (diff) panel goes from -max(abs(val)) to +max(abs(val)) with the diverging (default) color scheme.

**

I think setting the default behavior to be like 2 above makes more sense than letting the plotting libraries choose (current behavior). Thankfully, we already have all the values of interest calculated (min and max) and the plotting functionality can easily accept these new defaults without disrupting the current behavior of e3sm_diags since most (if not all) variables have their defaults overridden by the manually set configs. Therefore, I propose we adopt the items in point 2 above as new defaults.

Pros:

• Arguably clearer plots.

Cons (I chose the above example on purpose to illustrate the drawbacks of this proposal ahead of time):

• Sometimes the difference is actually all positive like above (so one can argue that having the negative portion of the diff plot is a waste).
• Sometimes there is a vast difference between the two plots being compared, also like above. The mass of SO4 in pre-industrial conditions was pretty low, and there is so much more of it in present-day conditions (so one could argue that having the same limits on the first two plots makes little sense as a general solution).
• This approach will likely force us to set the plotting defaults with a semi-log/semi-linear approach near zero to highlight the features of importance better (in differencing, we usually care about small things as well as big things).

**

I am going to adopt the above solution in my fork for our custom diagnostics, but it may be good to adopt it for all of e3sm_diags.

Let me know if you have any feedback or thoughts on this. If there is openness to this proposal, I can draft a PR with the necessary changes to make it more actionable.

[chengzhuzhang]

Hi @mahf708 Thanks for the suggested change. It would be nice to have the default behavior polished so that we can rely less on manually set the fields for colorbars. I'm also curious about your solution for convert scientific notion to a unit that can be readily printed and shown as metrics. I wondered if there is a generic solution to convert large/small values, while have reasonable metrics values to be printed.