Change `ReferenceEnsemble` to `HindcastEnsemble`

Before the v1 release, I want to be confident in the naming practice of ReferenceEnsemble. Philosophically, it seems like we desire splitting the perfect model setup (e.g., MPI) from the ensemble setup from observations/assimilations. This largely affects the way the output is analyzed and it’s easiest to just split them into two separate object types.

However, does ReferenceEnsemble as a name make sense? Is this a made up term? I feel like I might have picked it up in a textbook chapter on predictability. Regardless, this is something I want to think about for documentation, object naming, etc.

[bradyrx]

From Meehl 2014:

Finally, the terms “hindcast” and “retrospective forecast” are perhaps more prob- lematic. The ocean reanalysis community refers to hindcast for an atmosphere-forced ocean-only simulation. But this term is also used in the literature for initialized predictions of past cases. In the decadal climate prediction context, retrospective prediction, retrospective forecast, and hindcast are used inter- changeably to refer to this type of initialized forecast of past cases.

That bolded bit is what I was trained on. I use “hindcast” to reference something like our FOSI. Or like a CMIP5 historical run.

Although I see your point now on all this. I was thinking of the perfect-model setup as also being “hindcasts” (retrospective forecasts), but they’re not, since they aren’t trying to reproduce real past cases. Let’s go ahead and switch ReferenceEnsemble over to HindcastEnsemble to avoid confusion for the user.

[bradyrx]

Completed in https://github.com/bradyrx/climpred/pull/124