Rasql_Wrong transposition between X and Y axis when encode in netcdf

[Bang Pham Huu]

The problem is when encode a coverage in netcdf, the output is transposed between X and Y (in case of 2D).

for example: encode rgb 2D in netcdf (the i1 and k3 as dimensions can be random names so try to swap these variables does not help (e.g: i,j to j,i).

Also the transpose option from ​http://rasdaman.org/wiki/CommonFormatsInterface does not work (i.e: set values to [1,0] or [111,0] does not have any affect).

rasql -q 'SELECT encode(c[0:10,0:3].0, "netcdf" , "{\"dimensions\":[\"i1\", \"k3\"],\"variables\":{

	\"i1\":{\"type\":\"double\",\"data\":[0,2]},

	\"k3\":{\"type\":\"double\",\"data\":[1,2]},
	\"red\":{\"type\":\"unsigned char\",
\"metadata\":{\"units\":\"10^0\"},\"name\":\"red\"}},
\"metadata\":{}}")

 FROM test_rgb AS c' --out file

when use gdalinfo to show bounding box (wrong: size X(4), size Y (11))

Size is 4, 11
Coordinate System is `'
Metadata:
  red#units=10^0
  red#valid_max=255
  red#valid_min=0
Corner Coordinates:
Upper Left  (    0.0,    0.0)
Lower Left  (    0.0,   11.0)
Upper Right (    4.0,    0.0)
Lower Right (    4.0,   11.0)
Center      (    2.0,    5.5)

if encode in png

rasql -q 'select encode(c[0:10,0:3].0, "png") from test_rgb as c' --out file

the bounding box is expected (size X: 11, size Y: 4)

Size is 11, 4
Coordinate System is `'
Corner Coordinates:
Upper Left  (    0.0,    0.0)
Lower Left  (    0.0,    4.0)
Upper Right (   11.0,    0.0)
Lower Right (   11.0,    4.0)
Center      (    5.5,    2.0

[Dimitar Misev]

Yes transpose is not implemented at the moment.

[Dimitar Misev]

We need to implement the transpose format parameter, documented here: ​http://rasdaman.org/wiki/CommonFormatsInterface

It can be restricted to the last two dimensions for now which makes it very simple to support. The function to do this is already implemented:
​http://rasdaman.org/browser/conversion/convertor.hh#L173

And it's used in conversion/grib.cc at the moment. We just need to invoke it in conversion/netcdf.cc in case transpose is specified in the format parameters.

[Bang Pham Huu]

NOTE: not every netCDF file is in variable order (Lat, Long) which will be flipped when displaying. If input coverage is in order (Long, Lat) respectively, it will be displayed correctly.

​http://www.unidata.ucar.edu/software/netcdf/examples/tos_O1_2001-2002.cdl

so WCPS can know when should use the transposition from the axis order.

[Vlad Merticariu]

CF compliant ones will always be flipped.
But indeed, we should be able to differentiate, maybe at import. We need to discuss this in a f2f meeting next week.

[Dimitar Misev]

Danut, can you attach a patch of the modifications (including the test) that you did so far?

[Dimitar Misev]

Bang, do we have transpose option in wcst_import recipes?

[Bang Pham Huu]

Replying to dmisev:

Bang, do we have transpose option in wcst_import recipes?

don't have this option yet.

[Vlad Merticariu]

Let's discuss how we want to achieve this in the group meeting today. We have 2 options when importing netcdf:

• transpose at ingest time, deliver as is for any format except netcdf and apply traspose again when delivering netcdf
• store the data as is, deliver as is for netcdf, deliver transposed for all others.

We will need to think about how to store this information in the coverage.

[Bang Pham Huu]

I thik it should be consistent, when ingesting with 1 option tranpose_enable: true, it will transpose when ingesting coverage and getCoverage without regarding to any format types.

[Dimitar Misev]

It would be good in any case to be able to specify during ingestion which option is preferred, as both are effective in different situations:

1. Preferable if the data is primarily exported for visualization in common 2D formats
2. Preferable if the data is primarily exported in netcdf