Spurious SST values when using ERA-Interim data

[shakka]

Hi helpdesk,

I'm having a bit of a nightmare trying to get ERA-Interim SST and sea ice data into the global model, and I wonder if you can help. I am attempting to use ERA-Interim data as input to force the global model at N512, before downscaling to my 4 km LAM.

The SST and SIC data are contained separately in files with ERA-Interim SSTs and SIC, interpolated onto the N512 grid (e.g. /projects/polar/elgil/hindcast/N512_SIC_SSTs/2008_SST_n512).

I have reconfigured the global model (glm_um > reconfiguration and ancillary control > configure ancils and dump fields) to use these files rather than the default climatologies usually used.

When I run the model, the job.out files show that these files are picked up successfully, and indeed, when I examine the output, I can see that the sea ice is downscaled successfully from ERA > GLM > LAM (see attached file '20080101_SIC_inheritance_glm_ERA_native.png').

However, the SSTs are completely bizarre - despite supposedly reconfiguring the SSTs from my input file, there are very artificial edge effects, where the model has set any values over sea ice to -1.7 deg. However, the sea ice fraction used is not the sea ice fraction I have supplied (see attached file '20080101_SST_inheritance_glm_ERA_native.png'). Further, the location of these areas of spurious values vary on a daily timescale.

In the areas that are *not* strangely masked, the ERA-Interim data has evidently been correctly interpolated - this is true both of the GLM and LAM. This suggests that the UM is re-writing the SST data based on incorrect (and time-varying) sea ice ancillaries.

Can you help me find out whether this is hard-wired into the code, and crucially, how to change it?

I have also plotted both SSTs and SIC with the UM's climatological ancils, and although they are more similar to each other, the area with spurious values is not exactly the same (see attached files '20080101_clim_SIC_inheritance_glm_ERA_native.png' and '20080101_clim_SST_inheritance_glm_ERA_native.png'). After discussing with colleagues, we wonder whether this is because the UM is interpolating the monthly climatologies into daily values.

This may suggest that the UM is using monthly climatologies interpolated to produce a daily sea ice field, which is then used to mask the SST data wherever sea ice is present.

Can anyone advise?

Many thanks in advance,

Ella

[shakka]

sea ice using climatological values

[shakka]

SSTs using climatological values

[shakka]

SSTs using ERA-Interim

[shakka]

sea ice using ERA-Interim

[grenville]

Ella
What's the suite id?
Grenville

[shakka]

Hi Grenville,

I thought I had replied to this, but evidently my comment didn't post. Both suites u-bh076 and u-bh069 should have the same set-up, and experience the same problem.

Thanks,
Ella

[shakka]

Hi Grenville,

Have you had a chance to look into this?

Ella

[grenville]

Hi Ella

Sorry, it's been a very busy time. I'll do my best to look at this.

Grenville

[shakka]

Thanks Grenville, it would be great if you could when you have a minute, as I cannot make any progress on my work without getting these runs completed.

Best,
Ella

[grenville]

Hi Ella

I think the model is doing what it is supposed to - I believe it is using the correct ice field to set the sst to -1.7; it's a little difficult to tell from the .png, can you point me to the raw data.

I am not clear how the model is running: does the global model pick up a new start file periodically (at each cycle for example), which has been created to have ERA ssts and sic? Is this same start file (reconfigured to the LAM) then used to run the LAM?

Does the model fail?

Grenville

[shakka]

Hi Grenville,

The input ECMWF grib files are located at /projects/polar/amworr/startdumps/AntarcticCORDEX/atmos/YYYmmdd_HH.grib, and the input SST and SIC files (on an N512 grid) are at /projects/polar/elgil/hindcast/N512_SIC_SSTs/

The suite id is u-bh069 and both glm/LAM data are archived at moose:/devfc/u-bh069/field.pp/ or /home/d04/elgil/cylc-run/u-bh069/share/cycle/

The driving model starts from a new grib file at 12-hourly intervals (a standard ECMWF startdump). I then use the reconfiguration to read in data from an ancillary file that contains sea ice and sst fields on the n512 grid at 12-hourly intervals to match. I then allow the model to downscale these fields from the glm —> LAM. The model runs without failing, but produces these weird step-wise edges seen in the attached images.

Ella

[grenville]

Hi Ella

Thanks, I'll look.
Have you been in contact with Andrew Orr - he has been doing something similar (as far as I understand) -it'd be worth comparing your configuration with his.

Grenville

[shakka]

Hi Grenville,

Andrew is my PhD supervisor so I speak to him quite regularly. Once I discovered this problem he went back to check some of his previous runs, and has noted the same problem, although this is less of an issue for him because his domain is so large and at lower resolution.

Let me know what you think.

Thanks,
Ella

[grenville]

Ella

I don't have MASS access - there are no data files in /home/d04/elgil/cylc-run/u-bh069/share/cycle/ (i didn't check every directory).

Grenville

[shakka]

Hi Grenville,

Apologies - I've put two examples in ~/cylc-run/u-bh069/share/cycle/20080101T0000Z/Peninsula/4km/hindcast_2008/um/

Ella

[grenville]

Ella

We're stuck with the model setting sst to -1.8 where sic > 0 (I'd not recommend touching the bit of code setting this). However, I think the problem here is that you are reconfiguring a 25km sst field (with its jagged edges from its sic) on to a 4km domain without supplying a 4km sst or sic field. Could you create 4km ssts from the ERA data - they could then be passed into the glm dump when it is reconfigured?

You'd need to trial this first to check.

Grenville

[shakka]

Hi Grenville,

Do you think this is definitely the case? I'm just not sure because the sea ice seems to be reconfigured correctly, and was read in from the same resolution. Besides, it doesn't seem to me that the sea ice edge is in the same place where the stepped edge in the SSTs is (see attached plot, GLM SST SIC.png). The first panel in the plots I attached above is the N512 SST/SIC data plotted directly from the netcdf file. It is indeed lower resolution, but the step-wise edge is only there in the GLM output, not the input data.

Producing additional ancils for the LAM is something I'd like to avoid, if possible.

[grenville]

Can you plot where sic is non-zero in the glm start file?

[shakka]

Hi Grenville, you're on to something! (see attached). The non-zero part of the sea ice field differs - along the same edge that we're seeing. I think this will require some digging to find out what the reason behind this is.

Thanks

[grenville]

Hi Ella

Ok -the model is at least doing what it claims to do, so your results are correct model behaviour (maybe not what you want). You would get a less boxy sst if you use 4km sst ancils (the sst would still be set to -1.8 where sic >0 'though.

Grenville

[shakka]

Hi Grenville,

I am changing this ticket to closed, as I have found a way around the problem by mimicking the behaviour of the UM with my ERA forcing data - simply changing my input ancillary files so that all data below SIC = 0.1 is reset to 0 using xancil.

Thanks
Ella

[grenville]

Glad you found a workaround

Grenville