Add option to remove annual cycle in percentile_doy

[huard]

Addressing a Problem?

See https://www.nature.com/articles/s41467-024-46349-x for rationale

Potential Solution

No response

Additional context

No response

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

[bzah]

Interesting read thanks @huard !
In case someone want to start with the implementation here is the paper author's code https://github.com/lukasbrunner/running_window_bias/blob/main/code/pre-processing/calc_threshold_exceedance.py#L235
The function names are a bit confusing, but it looks fairly simple to adapt to percentile_doy and xclim indices.
My understanding is that we must compute the mean of each window (1), then substract this mean to the base array (2), then compute the percentile on this base_minus_mean (3) and compute the exceedance by counting days where base_minus_mean > threshold (4).

One issue for xclim is that steps (1), (2) and (3) should happen on percentile_doy but step (4) happens before or at the beginning of the index computation. However, we can use the same technique as we did for Zhang's bootstrapping algorithm and apply this seasonal bias correction via a decorator.
It could even be integrated in the current @percentile_bootstrap decorator as the logic to get the arguments (climatology_bounds, percentile value, etc) are similar.

As a side note I'm working (again) on the percentile bootstrap algorithm to improve the performances for Cerfacs.
I'm not yet sure if this will result in a PR and I don't think this would clash too much with this seasonal cycle correction, but I prefer to warn before a git conflict pops up.

[huard]

Salut @bzah !

Thanks for looking into this and for the heads up.