Problem with coordinates of SImon CMIP6 datasets

[dhohn]

Hi all,

I am looking at CMIP6 datasets and encountered a problem in the SImon subset. One example is: CMIP6.CMIP.NCC.NorESM2-LM.1pctCO2.r1i1p1f1.SImon.siconc.gn

Problems I see:

1. There are unmasked values near 1e30. I tried to mask those and copy the mask over to the bounds.
2. But the bounds to not align with the points. It looks like something got shifted somehow. (See attached Fig. Here the points are green and the bounds are the remaining colours, excluding the 1e30 outliers.)
   

If this is not the right forum for this question please direct me to it.
I would be very grateful for your insight.

Sincerely,
David Hohn

[TomasTorsvik]

@dhohn
Dear David,
thanks for getting in touch. I'm not sure I fully understand the problem you have, but perhaps I can help you a bit along the way.

Using siconc_SImon_NorESM2-LM_1pctCO2_r1i1p1f1_gn_000101-001012.nc as an example, the file use the value 1.e+20 to indicate missing values. I don't see 1e30 values in this file. Are there other files where unmasked 1e30 values appear? If so, could you tell me a file where you find this problem?

The sea ice variables are defined on a tri-polar grid corresponding to the ocean grid, but th grids are not identical. See documentation here:
https://noresm-docs.readthedocs.io/en/latest/faq/postp_plotting_faq.html#different-sea-ice-and-ocean-grid

If it is not too much effort, it might be easier to remap the output to a regular Lat/Lon grid if you want to apply your own masking file to the CMIP6 data. The CICE output files should contain latitude and longitude data that can be used to remap the files, for instance using CDO.

Best regards,
Tomas Torsvik

[dhohn]

Dear Tomas,

Thanks for your reply. I also looked at the exact file you mention and my issues I reported are based on that. I downloaded from ESGF. Did you get it from somewhere else?

The issue is basically that Im unable to compare between the 1pctCO2 and esm-1pct-brch-1000PgC experiments because the coords are different.

Best,
David

[TomasTorsvik]

@dhohn
Dear David,
I downloaded directly from the data node noresg.nird.sigma2.no. For some reason the ESGF web site reports this data as unavailable. The version of the file that I look at has the attribute :creation_date = "2019-08-18T01:38:12Z".

What is grid coordinate system do you need for your analysis?

Best,
Tomas

[oyvindseland]

Hi

The esg node seems to be down, but I was able to download the files from DKRZ . I also saw the files at the other mirror sites but I did not try to download them.

Best regards

Øyvind

[dhohn]

Thanks to you both.

@ Øyvind: Did you also see values of 1e30 in lat/lon?

@ Tomas: My files have the same creation_date. Im trying to compare 1pctCO2 and esm-1pct-brch-1000PgC for which the coordinates need to be the same/compatible and don't seem to be.

[oyvindseland]

Hi

Yes I also see lat lon values of 1.e+30 but for undefined gridpoints. I think this is a standard?

I can not see any strange behavior of the field though. Random field from the file included

Øyvind

ps png format

[JensBDebernard]

Hi David, I can confirm your findings concerning the 1e+30 values for the latitude and longitude variables for the 1pct-co2 files, and these are undefined values. Actually, it seems to be so for several files under the CMIP experiments. Unfortunately, for some reason, the _FillValue attribute is set to 1e+20, so these undef-points are not shown as masked in most software. The easiest fix to get the files to be consistent is to edit the _FillValue attributes to be correct, e.g. : ncatted -a _FillValue,latitude,m,f,1.00000001504747e+30 -a _FillValue,longitude,m,f,1.00000001504747e+30 filename.nc Thank you for reporting this error, I am not sure why this has not been discovered before. Best regards, Jens Boldingh Debernard

…

On Thu, 4 Aug 2022 at 14:17, oyvindseland ***@***.***> wrote: Hi Yes I also see lat lon values of 1.e+30 but for undefined gridpoints. I think this is a standard? I can not see any strange behavior of the field though. Random field from the file included Øyvind [image: ncview siconc-1] <https://user-images.githubusercontent.com/15869190/182844519-6c22e398-7e7a-416c-8269-c164aba9c860.png> ps png format — Reply to this email directly, view it on GitHub <#357 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADTH3UQJAWELE7KRD3TXIETVXOYG3ANCNFSM54SNWZNA> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[TomasTorsvik]

I didn't realize that the problem was in the latitude and longitude variables, not the siconc itself. Thanks for clearing this up.

[adagj]

Dear all,
I also encounter problems with the SImon coordinates. It is a different problem than what is described here, but related. I raised an issue here: NorESMhub/noresm2cmor#318 , but maybe this is a better place.

The problem :
There seems to be different values for vertices_longitude in e.g siconc (SImon variable) for ssp585 and historical
This is problematic for e.g. regridding and right now I'm not sure which one to use

To Reproduce
ncdump -v vertices_longitude /projects/NS9034K/CMIP6/CMIP/NCC/NorESM2-LM/historical/r1i1p1f1/SImon/siconc/gn/latest/siconc_SImon_NorESM2-LM_historical_r1i1p1f1_gn_200001-200912.nc

and compare to

ncdump -v vertices_longitude /projects/NS9034K/CMIP6/ScenarioMIP/NCC/NorESM2-LM/ssp585/r1i1p1f1/SImon/siconc/gn/latest/siconc_SImon_NorESM2-LM_ssp585_r1i1p1f1_gn_201501-202012.nc

Feedback is very much appreciated :)

[dhohn]

Thanks everyone for the feedback.

I think that addresses my point 1 from above.

Im still wondering about point 2 from my first message: The coordinate bounds seem clearly wrong to me as some do not align with their corresponding coordinate points. Furthermore I also observe what Ada just reported that the bounds are different between 2 experiments (1pctCO2 and esm-1pct-brch-1000PgC) which indeed is problematic for concatenating and regridding. What are the correct bounds?

Thanks so much!

[JensBDebernard]

Hi Dave and Ada, Sorry, I didn't think about the boundary points. In that case all the fields, or none of them are correct, depending on how you see this... Again, there are problems with the masking, and in this case the value used in the undefined parts of the grid seems a bit random across the files. Unfortunately, the values are within the valid range of the field so this makes a fix a little more tedious. However, in short, the masking of these grid cell vertices follows the masking of the center value of the grid point. A pseudo-code in python for setting the correct value can be something like: for jj in range(j): for ii in range (i):: if latitude[jj,ii]==latitude._FillValue: vertices_latitude[jj,ii,:] = vertices_latitude._FillValue if longitude[jj,ii]==longitude._FillValue: vertices_longitude[jj,ii,:] = vertices_longitude._FillValue where the values j=384 and i=320 are defined in the files. Also the _FillValue attribute is defined in the files. Good luck, Jens

…

On Mon, 8 Aug 2022 at 10:32, David Hohn ***@***.***> wrote: Thanks everyone for the feedback. I think that addresses my point 1 from above. Im still wondering about point 2 from my first message: The coordinate bounds seem clearly wrong to me as some do not align with their corresponding coordinate points. Furthermore I also observe what Ada just reported that the bounds are different between 2 experiments (1pctCO2 and esm-1pct-brch-1000PgC) which indeed is problematic for concatenating and regridding. What are the correct bounds? Thanks so much! — Reply to this email directly, view it on GitHub <#357 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ADTH3UWK2R4V7EVAA5X5YX3VYDA3PANCNFSM54SNWZNA> . You are receiving this because you commented.Message ID: ***@***.***>

[dhohn]

Dear Jens, all,

Thanks for the suggested fix for the bounds. I have followed your advice. Unfortunately I still encounter unexpected bounds. These seem to be at the interface between land and sea. See the attached plot:

It shows the lat/lon indices of the latitude coordinate where any abs(point-bound) > 1. Changing the treshold from 1 changes a picture a bit, but I think the values is sufficiently large to call the point and bounds incompatible with each other.

You might wonder why I care so much about the bounds? I don't really. But I observed artifacts when plotting the data in anything other than the native grid, went to investigate and this issue looks like a plausible culprit.

So in the end I want to make a plot like the following without artifacts:

Do you know how I can avoid those?

I'm also uploaded a notebook to make those two plots, so you know exactly what I mean: https://gist.github.com/dhohn/198ec494bba237abc11988538bea0605

Thanks again for all the help so far. I appreciate it a lot!
David

[adagj]

Dear David,
I found an old, closed issue where this problem is discussed: NorESMhub/noresm2cmor#39

It seems like the solution is to ignore the sea ice grid and use the grid information from the ocean output e.g. tos or areacello.
The ocean and sea ice model run on the same grid, except the ocean grid an one extra j-row compared to the sea ice grid, i.e. https://noresm-docs.readthedocs.io/en/latest/faq/postp_plotting_faq.html#different-sea-ice-and-ocean-grid

The solution works, but I'm not sure how to communicate this information to a broader cmip community.