[PATCH] Fix flipped latitude with NetCDF CF-1 files

[ekato]

Maybe related to JoseM's comment in #2584, I've experienced getting upside-down image with NetCDF driver. So I've created a simple patch to flip latitude for CF-1.0 files.

In netCDFDataset::SetProjection?(), I assume the files with following attributes as CF-1.0 files.

• have valid nVarDimXID and nVarDimYID
• poDS->papszDimName[poDS->nDimYid] starts with "lat"
• pdfYCoord[0] < pdfYCoord[ydim-1]

[ekato]

Revised one which also works with files created by GMT's grdraster

[dylan]

Any word when/if this can be merged into GDAL proper? It would save considerable time when working with NetCDF, by removing 2 steps involved in manually flipping the grid with a combination of gdal_translate and gdalwarp.

Thanks for the patch!

[warmerdam]

This seems pretty similar to #2584 - are they related? Reassigning to Denis, our netcdf guru.

[ekato]

Real revised patch which also works with GMT's grdraster

[tbouffon]

Hi, The patch did not work for me. Files created with xyz2grd (GMT 4.2.1) had y as poDS->papszDimName[poDS->nDimYid]. So I used GMT Metada to detect Latitude, and added a warning.

I haven't tested it a lot yet.

[tbouffon]

Added support of more GMT-generated files

[mdsumner]

FWIW we've been using netcdf-y-flip-rev.patch successfully. The third one does not work at all, but I haven't explored why.

[warmerdam]

I'll take a crack at reviewing the patches and applying one. I would like a test file - mdsumner its going to try and find one.

[warmerdam]

Based on a review of the code, and mdsumner's feedback I'm reasonable confident with netcdf-y-flip-rev.patch from ekato. There seem to be doubts still about the rev2 patches efforts with additional GMT files so I'm skipping it - it might be appropriate to file that in a new ticket which includes a reference to a file or files demonstrating the problem the patch solves.

I would add I'm interested in very small netcdf files demonstrating the issues in question that can be added to the autosuite. mdsumner provide me a test file, but it is large, and it isn't known if it is suitable for redistribution.

Patch applied in trunk (r18151,r18153), and 1.6 branch (r18152).

[tbouffon]

Hi, indeed there is a problem with my rev2 patch. I thought I had posted the new version. I and a few of my colleagues have been working with the version I'm posting without any problem for one month now. I'm joining a set of files as an example : file.xyz is an ascii file containing gridded data file.gmt is generated with :

xyz2grd file.xyz -R0/10/0/10 -I1 -Gfile.gmt -fg

the version with rev.patch still flipped the raster, as the metada where not matching the criteria. But rev2.patch segfaults when the lat long name is empty. So I worked it around by setting a new string. I'm leaving the french comments in the patch to avoid any typo.

Is my answer fast enough or should I open a new ticket ?

[rouault]

Re-opening as the netcdf-y-flip-rev3.patch was added after the commits

[kyle]

Has anyone tested the third revision of this patch?

[kyle]

Is there a reference for the bottom-first convention? I didn't see it in cf-1.4 docs. Is it always true?

[etiennesky]

This problem is still found in the stable version 1.7.2. Can I test the first and 3rd patch in 1.7.2 stable or trunk?

I'd like to help fix this problem.

[etiennesky]

Sorry looks like it's been fixed, it works correctly in stable 1.7.2.

[kyle]

I can't say that this is fixed completely in 1.7.2, as I still run into datasets that open upside down occasionally. I am still working on this one though...

[JeffBullard]

Replying to kyle:

I can't say that this is fixed completely in 1.7.2, as I still run into datasets that open upside down occasionally. I am still working on this one though...

This problem is still not fixed (for me) in 1.7.2, although the problem could be with NOAA's degrib utility. This problem is reproducible: Download a grib file from NOAA's NOMAD server. Convert to NetCDF format using NOAA's 'degrib' utility Convert the NetCDF to GeoTiff? with gdalwarp. The resulting GeoTiff? is upside down, every time.

I have been dealing with this problem for the past four years and my 'workaround solution' is to gdal_translate -a_ullr the NetCDF corner coordinates (flip them upside down) before converting to Geotiff.

[tbouffon]

Replying to JeffBullard:

Replying to kyle: This problem is still not fixed (for me) in 1.7.2, although the problem could be with NOAA's degrib utility. This problem is reproducible: Download a grib file from NOAA's NOMAD server. Convert to NetCDF format using NOAA's 'degrib' utility Convert the NetCDF to GeoTiff? with gdalwarp. The resulting GeoTiff? is upside down, every time. I have been dealing with this problem for the past four years and my 'workaround solution' is to gdal_translate -a_ullr the NetCDF corner coordinates (flip them upside down) before converting to Geotiff.

Hi, did you try one of the patches above ? As long as the degrib utility writes latitude as y-coord name in the metadata, they should work.

[JeffBullard]

Replying to tbouffon:

Replying to JeffBullard:

Replying to kyle: This problem is still not fixed (for me) in 1.7.2, although the problem could be with NOAA's degrib utility. This problem is reproducible: Download a grib file from NOAA's NOMAD server. Convert to NetCDF format using NOAA's 'degrib' utility Convert the NetCDF to GeoTiff? with gdalwarp. The resulting GeoTiff? is upside down, every time. I have been dealing with this problem for the past four years and my 'workaround solution' is to gdal_translate -a_ullr the NetCDF corner coordinates (flip them upside down) before converting to Geotiff.

Hi, did you try one of the patches above ? As long as the degrib utility writes latitude as y-coord name in the metadata, they should work.

After more testing, I think the problem is with the degrib utility from NOAA. I downloaded software from NCAR to convert my GRIB data to NetCDF then converted to GeoTiff? with GDAL 1.7.2. The image was NOT upside-down. NCAR utility with GDAL gives me correct output. NOAA utility with GDAL gives me upside-down.

[kyle]

Replying to JeffBullard:

Replying to tbouffon:

Replying to JeffBullard:

Replying to kyle: This problem is still not fixed (for me) in 1.7.2, although the problem could be with NOAA's degrib utility. This problem is reproducible: Download a grib file from NOAA's NOMAD server. Convert to NetCDF format using NOAA's 'degrib' utility Convert the NetCDF to GeoTiff? with gdalwarp. The resulting GeoTiff? is upside down, every time. I have been dealing with this problem for the past four years and my 'workaround solution' is to gdal_translate -a_ullr the NetCDF corner coordinates (flip them upside down) before converting to Geotiff.

Hi, did you try one of the patches above ? As long as the degrib utility writes latitude as y-coord name in the metadata, they should work.

After more testing, I think the problem is with the degrib utility from NOAA. I downloaded software from NCAR to convert my GRIB data to NetCDF then converted to GeoTiff? with GDAL 1.7.2. The image was NOT upside-down. NCAR utility with GDAL gives me correct output. NOAA utility with GDAL gives me upside-down.

Jeff, Can you test your file with Ivan's changes in r20006? Let me know if that does anything to help with your problem.

[kyle]

Ivan's correction fixed the files I was having issues with.

[tbouffon]

Ivan's correction also fixes the problems with the file.gmt attached above.

[kyle]

I am closing this for now, if anyone finds another instance, please reopen. Thanks Ivan