[Go-essp-tech] Problems with Sea Ice Mass Transport variables (transix, transiy)

[Jennifer Adams]

Dear Experts, 

Short story: 
1. A random sampling of Sea Ice Mass Transport variables (transix and transiy) reveal that GFDL-ESM2G and MPI-ESM-LR contain bad data. These are the only two models I have checked, I am 0-for-2. 
2. The transport vectors appear to be grid-relative and I need information on how to rotate them to be earth-relative. 

Long story (with pictures):

Almost all of the variables in the ocean and seaice realms are on non-rectilinear grids, a feature that renders them unreadable by GrADS. I have been working to remedy this problem by using the ESMF stand-alone regridding tool (ESMF_RegridWeightGen) to generate the weights so I can interpolate the data to a linear grid. GrADS uses the "PDEF" interface to do this on the fly so I do not have to reformat the data. 

My first look was at these files: 
transix_OImon_GFDL-ESM2G_historical_r1i1p1_186101-186512.nc
transiy_OImon_GFDL-ESM2G_historical_r1i1p1_186101-186512.nc

I have plotted transix and transiy for the 2nd time step. The images are posted here:
   http://iges.org/jma/cmip5/demo_transix_GFDL.png
   http://iges.org/jma/cmip5/demo_transiy_GFDL.png
Note weird discontinuities along the 80E/280W meridian (the native grid edge) and a data gap in transix along 80E poleward of 70N. 

My first reaction was that I had done something wrong during the execution of ESMF's tool or the creation of the PDEF file for GrADS. I checked by converting the native data to a GrADS station data set, which treats every grid point as a randomly located station. Here are similar plots, with similarly colored dots at each native grid point: 
  http://iges.org/jma/cmip5/demo_stntransix_GFDL.png
  http://iges.org/jma/cmip5/demo_stntransiy_GFDL.png
These plots confirm that the weirdness and missing data are present in the CMIP5 files, and that ESMF regridder is doing an awesome job, and there's nothing wrong with my PDEF code.  

Next step, check another model: 
transix_OImon_MPI-ESM-LR_historical_r1i1p1_185001-200512.nc
transiy_OImon_MPI-ESM-LR_historical_r1i1p1_185001-200512.nc

Here I have plotted transix and transiy for the 5th time step. The images are posted here:
   http://iges.org/jma/cmip5/demo_transix_GFDL.png
   http://iges.org/jma/cmip5/demo_transiy_GFDL.png
There is a stripe in the data for both variables stretching from Ellesmere Island to Wrangel Island. This doesn't align with any fixed lon or lat. 

The station versions of the plots are here:
   http://iges.org/jma/cmip5/demo_stntransix_GFDL.png
   http://iges.org/jma/cmip5/demo_stntransiy_GFDL.png
These reveal that the stripe is along a single model column. Once again, I cannot blame ESMF or my own code -- this native data is also bad, but not as bad as GFDL's ESM2G. 

Out of the many petabyes of CMIP5 data, I have checked on two models chosen at random from single experiment, looking at a fairly obscure subset of variables, just a handful of grids … what are the odds that they would BOTH be flawed?

Setting these troubling issues aside, I pressed on by looking at the vector field for the MPI-ESM-LR data. GrADS doesn't draw vectors very well at high latitudes, so I have drawn the streamlines instead (ignore the kinky streamline in the path of the stripe in the data). Streamlines are colorized according to the magnitude of the vectors:
  http://iges.org/jma/cmip5/demo_transport_MPI.png
This doesn't look right --  the flow in and around the north pole is highly suspect. I checked other time steps and they all have a circulation feature centered exactly on the pole. That leads me to the conclusion that these vectors are grid-relative rather than earth-relative, and need to be rotated. But this is a highly-subjective guess based on a cursory visual inspection of the data; actual metadata would be better. 

Does anyone have a clue where I can find out if vectors are grid-relative or earth-relative, and what is the rotation angle to convert between the two? I checked standard_output.xls and couldn't find anything in the "fx" or "OImon" tables. My colleague has also been poking at this problem and he found a file on GFDL's data portal, gridspec_seaIce_fx_GFDL-CM3_historical_r0i0p0.nc. This file appears to have a variable called "angle_dx" with the standard name "grid_vertex_x_angle_WRT_geographic_east" but the grid is only partially populated (intermittent zeros in between valid anlges). 

I have cc'd this message to the contact emails given in the global attributes of the data files in question. Is this my fate, to be forced to contact each model representative independently to unlock the mystery that is Sea Ice Mass Transport?

I apologize for the extreme length of this email. If you are SWAKS (someone who actually knows something), please advise. 
--Jennifer


--
Jennifer M. Adams
IGES/COLA
4041 Powder Mill Road, Suite 302
Calverton, MD 20705
jma at cola.iges.org



-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mailman.ucar.edu/pipermail/go-essp-tech/attachments/20120524/03f91ce3/attachment.html