Additional requirements for input climate data needed in process_gcm_data?

[anoukvlug]

I would like to start this issue to discuss if there might be additional requirements needed on the climate data that can be processed by the process_gcm_data function. This issue came to light in an experiment by @Keeptg. Using process_gcm_data on a climate dataset without year to year variability results into returning the CRU climatology. As this is likely not what the user would like to get out of it, it might be good to introduce a warning or even an error.

At first I was a little hesitant about the introducing an error, as it is in a way limiting what can be done with the function. However there are other ways to get the CRU climatology. In addition it is not what the function was not what it intended for. Because of the harm that not raising an error could do, I would at this point favour implementing an error.

What do you think of that @Keeptg, @fmaussion? Are there maybe other additional requirement needed for the input climate data?

(Like always, feel free to add others that might be interested to discuss this.)

[Keeptg]

It's a good idea for introducing an error for process_gcm_data, so people can understand more about how this function works and avoid some issue. In the meantime, I'm still thinking if it's necessary that OGGM provides some methods to make sure people can use their own climate data without bias calibration. Because, for my case, I just concern the ice volume difference simulated by my different climate experiment data.

[fmaussion]

I think I need a bit more context to decide. How exactly did the problem occur, with which datasets?

[Keeptg]

One of my colleagues provided me two experiment climate data (each data including precipitation and temperature for only one year(12 months)). What he wants to know is the glacier area/volume difference simulated by those two different climate data, after a long run time and the glacier reach a stable status. Referencing the CMIP5 dataset, I write those two climate data as .nc file and repeat 2000 times to get a 2000 years dataset. The .nc dataset was processed by process_gcm_data and then drive the tasks.run_from_climate_data model.

For the default setting, my '.nc' data triggered an assert Error in here. Then, I set scale_stddev=False in process_gcm_data function, and the function run successfully. However, the output glacier volume between different input climate data is always same. This is where the issue came from. I'm not sure if there would be more people who want to do some experiment similar with my colleagues. If there will be, then maybe OGGM provide some support will make things easier.