[gdal-dev] HDF5: best practice

[Even Rouault]

Jose,

> 
> We're using the Python GDAL bindings to ingest many different datasets into
> our processing system, which currently outputs data as GeoTIFF files.
> However, in order to optimise processing, parts of the code use HDF5. As
> such, we would like to drop the GeoTIFF output and use the HDF5 format
> directly.

I suppose you write HDF5 with another library since the support in GDAL is 
limited to read-only ?
Well, you could create HDF5 files by using the netCDF driver in netCDF-4 mode 
(that uses a HDF5 container) (FORMAT=NC4 creation option)

> However, I can find relatively little information on having HDF5
> datasets fully characterised by a particular geographic
> projection/georregistration. I understand that adding a couple of metadata
> tags, GDAL is able to understand projections and geolocation, but is there
> anything more general?

The GDAL HDF5 driver has little support for reading projections due to the 
lack of standardization on this offered by HDF5. So from what I can see in the 
driver, there are just a few particular cases for specific HDF5 production. The 
most interesting would be the case of ODIM productions that handle projections 
expressed as proj.4 strings if the where_projdef, where_LL_lon,  where_LL_lat, 
where_UR_lon and where_UR_lat metadata item are found ( the code references 
http://www.knmi.nl/opera/opera3/OPERA_2008_03_WP2.1b_ODIM_H5_v2.1.pdf/
/* 4.3.2 where for geographically referenced image Groups )

Perhaps more interesting would be to use the netCDF driver in netCDF-4 mode 
and the netCDF CF conventions.

Even

-- 
Spatialys - Geospatial professional services
http://www.spatialys.com