Indicator keyword for performing calendar conversion

[Zeitsperre]

• xclim version: master
• Python version: any

Something that has come up in the process of using xclim to calculate indices that are then fed to R is the incompatibilities of some calendars, specifically the notorious 360_day calendar. If non-standard calendars are not supported in the environment that uses this indicator data, problems ensue. While this is fine (users should be aware of calendar idiosyncrasies), I wonder if we could introduce a feature to make this a bit easier.

The specific case I want to address is when we perform a temporal aggregation (e.g. xclim.atmos.tg_mean(freq="MS")) with non-standard calendar data. Due to the way that the calendar is written (days since xxxx-yy-zz), we treat the first day of each month as the month itself. With non-standard calendars, this can lead to years in which we can see multiple values in a month or the slipping of dates backwards in time (especially with 360_day). None of this is new, and I don't suggest change this behaviour.

What I would like to see is a keyword in the calculations of indicators that allows for outputs to assume a standard calendar (ie: convert calendar being called). I understand that this already is triggered when performing ensemble analysis, but being able to change the calendar (called like xclim.atmos.some_indicator(freq="QS-DEC", to_calendar="proleptic_gregorian")) would make manual comparison of data much simpler.

[aulemahal]

I think this could be interesting for some users indeed.
Currently, it is doable with:

import xclim as xc

out = xc.atmos.some_indicator(...)
out = xc.core.calendar.convert_calendar(out, 'proleptic_gregorian')

But adding this to the indicator call directly could be interesting, but I think we need to clean up the kwarg adding machinery before. Not much to do, just to decide on a clean way to inject those kwargs.

[aulemahal]

Documentation of the new xarray methods to be added while implementing #1010.