fmaussion
commented
5 years ago Thanks for the report! Sorry you had to dig through these problems, but yes you are correct. To your points:
Are 2017 and 2018 being ignored because the model still uses CRU TS 4.0 and not 4.02 or 4.03 due to their recent release
Yes, we are still using CRU TS v. 4.01. I've also noticed that v. 4.03 is out now with data up to 2018. I will have a look at it (updating is not super trivial, because sometimes updates in CRU imply changes in the data and require a re-calibration.
...Or because the current FoG database (2017) only has SMB records through 2016?
No, this doesn't play a role. The current "5 years rule" is implemented here:
Where gdir.get_ref_mb_data() indeed checks for the baseline climate data (here CRU), ensuring that we have 5 years of overlap. I'm open to turn this hardcoded 5 years limit into a model parameter, so that users can choose their own threshold.
Perhaps these records are ill suited to be reference sites within the model calibration due to their brevity anyhow.
This is a harder question. The reason for the 5-years rule was to avoid noise, and filter out "one-time glaciers" out of the WGMS database because we don't trust these data. In your case, it might be worth using the data, in particular if you plan to look at the behavior of these glaciers in particular: for these glaciers, OGGM is going to have a "zero-bias" mass balance (although other problems like over- or under-estimated variability are likely, like we show on the tutorial).
However, because of how the OGGM mass-balance calibration works, don't expect miracles for the rest of the unobserved glaciers. Adding these reference glaciers will change their MB as well (by changing the "t* years"), but they will still be subject to considerable uncertainties at the individual glacier level.
bpelto
Using the new functions detailed in the excellent mass balance calibration tutorial on the oggm-edu page, I have created SMB dataframes for six glaciers, and assigned/appended those glaciers as reference glaciers.
At first I had trouble as only two of the six sites were successfully being appended, though this was due to the length of record. I had read that the sites need a minimum of 5 years to be considered by the model. My sites have 14, 10, 6, 6, 5, and 4 years of data respectively...up through 2018, but 2017 and 2018 are not being used (I believe) so disregarding these years meant that only two of my records were of usable length. For the time being, as a test I added some values to extend my records which were too short, back 1-3 years depending on site.
Are 2017 and 2018 being ignored because the model still uses CRU TS 4.0 and not 4.02 or 4.03 due to their recent release (bugs etc)...Or because the current FoG database (2017) only has SMB records through 2016? (both I presume)
Perhaps these records are ill suited to be reference sites within the model calibration due to their brevity anyhow.