Reference probabilistic analog ensemble

[williamhobbs]

It would be nice to have a probabilistic reference forecast using an analog ensemble (AnEn).

Some advantages:

• The input data requirements wouldn't necessarily have to be higher than the existing (deterministic) NWP-based forecasts
• Related to the last point, it can be difficult to fetch GEFS data for a true ensemble. See [1], [2], [3]
• (I'm assuming) it could perform a lot better than the persistence ensemble options

Disadvantages:

• Could get complicated
• I'm not sure if there's an existing python-based solution to build off of
• ??

I know there's a number of higher priority items right now, but I thought I'd at least start a discussion on this topic in case it makes sense later on.

[1] #712
[2] #680
[3] #674