Fwd: [gdal-dev] Re: discussion on improvements to the NetCDF driver and CF-1 convention

Etienne Tourigny etiennesky at yahoo.com
Wed Aug 31 16:05:11 EDT 2011



I added an item on the wiki page under I/O issues, can you make a detailed ticket and link to it on the wiki?

In the mean time, please use Even's suggestion of using an intermediary format such as VRT.

* Support Create() method so that gdalwarp can be used on a netcdf file

It's not as simple as it may seem, because all modifications to the netcdf file (not internal gdal information) are done only in the CreateCopy() function.  in Create() you don't have information on geotransform and projection, nor raster bands.  A flag could indicate if projection/geotransform need to be written after it has been changed.  To avoid code de-duplication, the code in NCDFCreateCopy() related to projection/geotransform could be moved to a new function like SetNetcdfDimensions().  The RasterIO issues that Even mentioned can also be challenging.


I assume your file is in a projected CRS?In any case, the current implementation is not 100% CF compliant.  GDAL can read them (through custom tags), but other software cannot.   I have submitted a patch to fix geographic CRS only.
Please add this information to existing bug #2893 (which I have renamed tonetcdf driver does not read and write CF-1.0 coordinate system properly for projected CRS), along with your file (which was stripped by the mailing list)

BTW, there is a nifty CF compliance checker (python script and online checker) which I have added to the wiki. Please test your file with it. It's possible that lambert_conformal_conic:standard_parallel_1 is valid CF, it needs to be tested.


----- Original Message -----
From: Hendy Sumanto <hsumanto at vpac.org>
To: gdal-dev at lists.osgeo.org
Sent: Wednesday, August 31, 2011 1:23:54 AM
Subject: Fwd: [gdal-dev] Re: discussion on improvements to the NetCDF driver and CF-1 convention

Hi Etienne,

I have a few items which I would like to put as new topics in the wiki entry but I think I better describe them here first before putting any entry or creating tickets.

1. As part of the improvement to the NetCDF driver, is it possible to make request to implement the Create method for the driver so that tool such as gdalwarp can output as NetCDF output.

2. Recently I was trying to serve the NetCDF dataset (which were created by gdal_translate geotiff LANDSAT dataset into NetCDF)
using THREDDS data server. When I try to access it via WMS using viewer such as Godiva2, I was not able to view the dataset correctly.

When I used the netCDF toolsUI to check the NetCDF dataset, it is not recognized as a grid by NetCDF Tools (4.2).  

Looking at this documentation (http://www.unidata.ucar.edu/software/netcdf-java/reference/StandardCoordinateTransforms.html),
in order for grid data to be viewable, the NetCDF dataset must:
a) define x and y projection coordinate variables, using the correct projection units, typically km on the projection plane.
b) define your projection dummy variable which has an attribute "grid_mapping_name"
c) refer to the projection in your data variables with the "grid_mapping" attribute.

Checking on the GDAL netCDF, it only define x and y dimensions but not x and y projection coordinate variables (point a is not followed).

Thus, I wondering whether other people have experienced the same issue as what I am having.

Then I just tried to: 
-generate its CDL by ncdump the GDAL netCDF file into CDL
-and the define x and y projection coordinate variables (as shown below) into the CDL
    double y(y) ;
        y:units = "m" ;
        y:standard_name = "projection_y_coordinate" ;
        y:long_name = "y coordinate of projection" ;
    double x(x) ;
        x:units = "m" ;
        x:standard_name = "projection_x_coordinate" ;
        x:long_name = "x coordinate of projection" ;
-regenerate the netCDF file from the modified CDL using ncgen
-populate the y coordinate variable in the NetCDF file with
        ymin, ymin+resolution, ymin+2*resolution, ...
-populate the x coordinate variable with
        xmin, xmin+resolution, xmin+2*resolution, ...
-if I then serve the modified NetCDF dataset again using THREDDS data server, now it can be viewed correctly.

Note: xmin, ymin are the lower left corner and resolution is the raster pixel size.

Thus, my questions:
i)Is there a specific reason why the current driver do not define the x and y projection coordinate variables but only defining x and y dimensions. 
For the dataset to be viewable using tools such as netCDF toolsUI or when the dataset are served via THREDDS dataset, we might have to add projection coordinate variables and the correct values
into the netCDF file.
ii)For lambert conformal conic projection with more than one standard parallel values, GDAL netCDF file is showing 
lambert_conformal_conic:standard_parallel_1 = -18.f ;
lambert_conformal_conic:standard_parallel_2 = -36.f ;
but the correct one (for the dataset to be viewable using toolsUI or THREDDS) should be
lambert_conformal_conic:standard_parallel = -18.f,-36.f ; 

Could you please please comment on 1, 2(i) and 2(ii)? 

The attached files:
tile.nc - GDAL netCDF file
tile-fixed-new.nc - the one I just tried to fix and can be viewed in netCDF toolsUI or served in THREDDS



gdal-dev mailing list
gdal-dev at lists.osgeo.org

More information about the gdal-dev mailing list