dimensions of opened NetCDF files

Hi,

I frequently use NetCDF files to load data in Matlab:

   ncload mydata.nc; for a quick open of new files

   nc_varget(); to get specific variables

Now, I changed to a new workplace and found here the handling with

   ncid = netcdf.open('myfile.nc','nowrite');

   varid = netcdf.reqvarid(ncid,'variable');

   data  = netcdf.getVar(ncid,varid);

This works, but the dimension is changed: While ncload loads the variable correctly as e.g. 12 x 40 x 180 x 90 (as described in the NetCDF file, ncdump),

with netcdf.getVar I get a variable of size 90 x 180 x 40 x12, which is quite inconvenient. Is this a usual behavior of netcdf.xxx or is there something I didn't recognize?

I appreciate any information, Christoph

Answer 1

[John]

When you shell out to run ncdump, you are running a C executable that is linked into the netcdf library. Since it is written in C, it used row-major order, so "time_counter" becomes the first dimension listed as it varies the slowest.

MATLAB is column-major order, so "time_counter" gets listed last, since that's where the slowest-varying dimension would be.

So, no, no mistake, it's just fallout from row-major-order vs. column-major-ordering. Also, since you are using NC_VARGET, check the snctools README, there's a discussion of the issue there.

[Christoph]

John, Thank you very much

Answer 2

[John]

Hi Christoph,

The "ncload" function referred to is part of the "netcdf toolbox", which is a very old and unsupported piece of software. It transposes the data when it reads and writes, whereas the native netcdf.getVar and putVar functions do not transpose, thus preserving the ordering of the data.

[Christoph]

Hi John, thanks for your reply. Since ncload is that old I would assume there's trouble with that, but the opposite is the case:

   !ncdump -h run10/BTST25_1m_19900101_19901231_grid_T.nc
   netcdf BTST25_1m_19900101_19901231_grid_T {
   dimensions:
  x = 454 ;
  y = 512 ;
  deptht = 64 ;
  time_counter = UNLIMITED ; // (12 currently)
  tbnds = 2 ;
   variables:
        [...]
     float votemper(time_counter, deptht, y, x) ;
          votemper:units = "degC" ;
    votemper:standard_name = "temperature" ;
    votemper:_FillValue = 9.96921e+36f ;
    votemper:coordinates = "time_counter deptht nav_lat nav_lon" ;

Then, if I use

   ncload run10/BTST25_1m_19900101_19901231_grid_T.nc;
   size(votemper)
   ans =

    12    64   512   454

Same for nc_varget:

   clear
   ncfile = 'run10/BTST25_1m_19900101_19901231_grid_T.nc';
   data = nc_varget(ncfile,'votemper');
   size(data)
   ans =
     12    64   512   454

However, with netcdf.xxx I get the following:

   clear
   ncid=netcdf.open('run10/BTST25_1m_19900101_19901231_grid_T.nc','nowrite');
   varid=netcdf.inqVarID(ncid,'votemper');
   data = netcdf.GetVar(ncid,varid);
   size(data)
  ans =
    454   512    64    12

I don't know, whether this is due to a mistake of how I handle the ncfile, or some bug in the NetCDF package used here? Kindly appreciate further ideas or comments... thanks.

Christoph

Answer 3

[Odontungalag Dorjsuren]

HI. How to open data .nc