convert_calendar could convert day-of-year data

[aulemahal]

Addressing a Problem?

Some indicators return day-of-year data. The values are thus constrained by the calendar of the input. A 360_day timeseries will have a maximal value of 360. When mixing data from different calendar, like when doing ensemble statistics, this introduces a bias in the statistics. For example, if all members agree that this event happens on the last day of the year, some will have 360, some will have 365 and some 366. The mean will be lower than 365, and the standard deviation will be non-zero, even though every model agrees.

Potential Solution

convert_calendar could detect day-of-year data through the is_dayofyear attribute that xclim adds on such cases.

The basic conversion could be : doy_out = out.time.dt.days_in_year * doy_in / doy_in.time.dt.day_in_year.

Additional context

Caveats:

1. This introduces floating point day of year, which might not be fully expected in some cases.
2. For 360_day calendar, this is equivalent to the align_on='year' timestamp conversion method. What about the align_on='date' ? That would need a mapping of doy values, not so complex, just a bit heavier. And should the user be able to decide on which method to use separately on the time coordinate and on the data?
3. is_dayofyear is xclim-specific. There is no CF standards about this, AFAIK. This would not be portable to xarray's convert_calendar.

Contribution

• I would be willing/able to open a Pull Request to contribute this feature.

Code of Conduct

• I agree to follow this project's Code of Conduct