Ouranosinc / xsdba / 24407598516

79%

push

github

Extremes default (#262)

<!-- Please ensure the PR fulfills the following requirements! -->
<!-- If this is your first PR, make sure to add your details to the
AUTHORS.rst! -->
### Pull Request Checklist:
- [x] This PR addresses an already opened issue (for bug fixes /
features)
  - This PR fixes #xyz
- [x] (If applicable) Documentation has been added / updated (for bug
fixes / features).
- [x] (If applicable) Tests have been added.
- [x] CHANGELOG.rst has been updated (with summary of main changes).
- [x] Link to issue (:issue:`number`) and pull request (:pull:`number`)
has been added.

### What kind of change does this PR introduce?

* Changes some of the default values to values that should generally
produce better results.
* Fixes a bug when using an even value for `power`.

### Does this PR introduce a breaking change?

* Yes. Defaults are changed.

### Other information:

* Tests were done on a sample of ~10 locations in Southern Quebec (with
small variations between models to add some noise), 12 CMIP6 models, 2
references (ERA5-Land and CaSR v3.2).
* 560 parameter sets were tested:
  * `q_thresh` between 0.95 and 0.98.
  * `frac` between 0.25 and 0.85.
  * `power` between 1 and 4.
* Using the full timeseries vs. seasonally-coherent extremes (DJFMAM /
JJASON).
* Results were done by comparing the bias-adjusted data vs. the
reference for 9 indicators covering high quantiles and annual/seasonal
return values.

0.95 / 0.70 / 3 performed the best with seasonal data.
0.97 / 0.70 / 3 performed the best with the full timeseries (but this is
not a recommended setup, as it mixes high intensity events in summer to
large-scale spring events).

With 50 years of data instead of 30, a `frac=0.60` seemed better than
`0.70`.

The current defaults (0.95 / 0.25 / 1) consistently produced very high
biases, especially with high quantiles.

5 of 5 new or added lines in 2 files covered. (100.0%)

2834 of 3599 relevant lines covered (78.74%)

6.14 hits per line