gdal_rasterize produces error when burning vector dataset into netCDF dataset directly with -a_nodata nan

[hsumanto]

I am trying to perform the below with the latest gdal in trunk:

gdal_rasterize -of netCDF -tr 500.0 500.0 -te 519223.62281715404 -4538725.344003787 1452723.622817154 -3983225.3440037873 -a_srs EPSG:3112 -a_nodata nan -ot float32 -a value -l SHAPE_TEST SHAPE_TEST.shp Output_Invalid.nc

and I am getting these errors:

ERROR 1: netCDF scanline fetch failed: NetCDF: Operation not allowed in define mode
ERROR 1: Output.nc, band 1: IReadBlock failed at X offset 0, Y offset 0
ERROR 1: GetBlockRef? failed at X block offset 0, Y block offset 0

It seems that it doesn't like -a_nodata nan because I could burn vector dataset into netCDF without error if I remove -a_nodata nan. Still wondering the reason why -a_nodata nan is causing the error.

Burn vector dataset into netCDF (OK without -a_nodata nan)

gdal_rasterize -of netCDF -tr 500.0 500.0 -te 519223.62281715404 -4538725.344003787 1452723.622817154 -3983225.3440037873 -a_srs EPSG:3112 -ot float32 -a value -l SHAPE_TEST SHAPE_TEST.shp Output_OK.nc

Burn vector dataset into Geotiff Raster (OK even with -a_nodata nan)

gdal_rasterize -tr 500.0 500.0 -te 519223.62281715404 -4538725.344003787 1452723.622817154 -3983225.3440037873 -a_srs EPSG:3112 -a_nodata nan -ot float32 -a value -l SHAPE_TEST SHAPE_TEST.shp Output_Invalid.tif

[hsumanto]

vector dataset file

[hsumanto]

vector dataset file

[hsumanto]

vector dataset file

[hsumanto]

vector dataset file

[etourigny]

It seems that gdal_rasterize sets nodata values after setting projection info, which I had not tested before in the new Create() method (which gdal_rasterize uses).

IReadBlock() is missing a call to get out of define mode, so the fix is to add a call to SetDefinemode?(False), just like in IWriteBlock(). I don't think this should be detrimental to performance, as the nodata value was set to default previously and does not need to be allocated, only modified.

Try this simple fix in IReadBlock():

            Taken += start[panBandZPos[i]] * Sum;
        }
    }

+    /* make sure we are in data mode */
+    ( ( netCDFDataset * ) poDS )->SetDefineMode( FALSE );

+    /* read data according to type */
    if( eDataType == GDT_Byte ) 
    {

Can you try with other nodata values also?

[etourigny]

Oh and in the furure - please send shapefiles in a single zip files, it makes it easier to download and also dramatically smaller - thanks.

[hsumanto]

Thanks Etienne for the quick response. I will send them as a single zip next time.

I try the fix above, it fixes the issue:

WITH nan as nodata value
gdal_rasterize -of netCDF -tr 500.0 500.0 -te 519223.62281715404 -4538725.344003787 1452723.622817154 -3983225.3440037873 -a_srs EPSG:3112 -a_nodata nan -ot float32 -a value -l SHAPE_TEST SHAPE_TEST.shp Output.nc
0...10...20...30...40...50...60...70...80...90...100 - done.

WITH -9999 as nodata value
gdal_rasterize -of netCDF -tr 500.0 500.0 -te 519223.62281715404 -4538725.344003787 1452723.622817154 -3983225.3440037873 -a_srs EPSG:3112 -a_nodata -9999 -ot float32 -a value -l SHAPE_TEST SHAPE_TEST.shp Output2.nc
0...10...20...30...40...50...60...70...80...90...100 - done.

Check gdalinfo on both files, they seem OK.

Then I just try to check the content by converting the netcdf into ASCII

gdal_translate -of AAIGrid Output2.nc Output2.asc
Input file size is 1867, 1111
0ERROR 1: netcdf error #-38 : NetCDF: Operation not allowed in data mode .
at (netcdfdataset.cpp,SetDefineMode,1339)

...10...20...30...40...50...60...70...80...90...100 - done.

gdal_translate -of AAIGrid Output2.nc Output2.asc
Input file size is 1867, 1111
0ERROR 1: netcdf error #-38 : NetCDF: Operation not allowed in data mode .
at (netcdfdataset.cpp,SetDefineMode,1339)

...10...20...30...40...50...60...70...80...90...100 - done.

The ASCII datasets are still being produced but I noticed the similar warning about SetDefineMode? do appear again here.

[etourigny]

You should also get the same error with gdalinfo -stats , as it also calls IReadBlock(). The call to nc_enddef/nc_redef (in SetDefineMode?) should not happen when only reading a file, not writing.

Here is the proper fix in SetDefineMode?() (in addition to the previous one):

int netCDFDataset::SetDefineMode( int bNewDefineMode )
{
    /* if dataset is in read-only mode, do nothing */
    if( GetAccess() == GA_ReadOnly ) return CE_None;

    /* if already in new define mode, do nothing */
    if ( bDefineMode == bNewDefineMode ) return CE_None;
...
}

I will commit this in the next few days, please confirm it works well, thanks.

[hsumanto]

Yeah, that fix resolves the issues which I have encountered so far. Thanks, Etienne.

[etourigny]

fixed in 1.9 (r23760) and trunk (r23759)