Éric Dupuis

Regarding the spi errors, let me show a case study on the first error ("ML-gamma-D-window==1"). First, here are many values, the error is relatively small but not negligible:  (yes...

Also, regardless of performance, `keep_longest_run` should treat alll nan slice differently IMO That's not right -> ```python rl.keep_longest_run(xr.DataArray([np.nan,np.nan,np.nan],dims="time")) >>> [True, False,False] ```

The function gives an output and yields correct results, so we decided this is not a bug. With the recent improvements, this should be much better, but I'm not sure...

Hi! Thanks for your contribution, this will be a very nice addition to `xclim`! For the `Callable`, did you have some relation in mind that is sometimes used? Since it...

> This involves the modification of prob_of_zero variable generated from parameter calibration. Right now, only (pp_rank_{n}_pr) is stored. (pp_rank1) may also need to be stored when f ≠ 1 as...