Add support for ICCLIM custom indices

[bekozi]

• Custom indices assigned a standard function key like 'icclim_custom'
• The 'kwds' function dictionary value will be the keyword arguments to icclim.indice
• icclim documentation for custom indices: http://icclim.readthedocs.io/en/latest/python_api.html#example-3-indice-tg90p-with-callback

Ping @nilshempelmann, @pagecp

[huard]

Well, that's a lot more complicated than I anticipated. The custom user indices computation is deeply buried in icclim and there seems to be a lot of data preparation before it's called, which as I understand would have to be replicated within ocgis. Christian, suggestions ?

Also, is it possible that currently ocgis does not support passing some icclim function parameters. For example RR1 has a parameter called "threshold" and it doesn't seem possible at the moment to assign any value to it from ocgis.

[bekozi]

Well, that's a lot more complicated than I anticipated. The custom user indices computation is deeply buried in icclim and there seems to be a lot of data preparation before it's called, which as I understand would have to be replicated within ocgis.

When I looked at it awhile ago, my thought would be to have ocgis as a very thin layer that mostly just passes through to icclim. It may not be worth it though unless there is a big advantage to using ocgis in all cases.

Also, is it possible that currently ocgis does not support passing some icclim function parameters. For example RR1 has a parameter called "threshold" and it doesn't seem possible at the moment to assign any value to it from ocgis.

It would be nice to modify the ocgis-icclim interface to allow arbitrary keyword arguments allowing icclim to evolve without much change to ocgis. That interfaces needs some TLC.

[pagecp]

I agree with Ben's comment. Just having ocgis to act as a thin layer on top of icclim probably the way to go.

For icclim parameter passing, I agree that having an interface to pass arbitrary arguments would be the best. It could be gcc-like for example.

[bekozi]

Received a request from another user to implement these. It would be good to identify a path forward for integrating custom indices. Here is the request:

1. The frequency of dry day (Similar to the number of wet days "RR1")
2. Precipitation intensity but only for days exceeding the specific percentile (similar to SDII but only for the days exceeding the specific percentile "e.g. pr > 99th percentile")
3. Modified indices for the frequency of day exceeding percentile thresholds, instated of fixed percentile thresholds ("R99p") externally specifiable thresholds (e.g. 10, 20, ... 75, 90, 95, 99)