How to convert from HDF5 to GeoTIFF with Python

Question

I'm coming from Reading metadata in HDF5 file.

I have a HDF5 file with some radar information. I'd like to read this data and geospatial metadata contained in the file to produce a GeoTIFF. Using the accepted answer, I can read the metadata, but I'm not sure how to convert it to something that rasterio can understand to produce a valid CRS and transform that allow to georeference the field.

Right now, the metadata I can get from the HDF5 are these two dicts:

{'projection_indication': b'Y',
 'projection_name': b'STEREOGRAPHIC',
 'projection_proj4_params': b'+proj=stere +lat_0=90 +lon_0=0 +lat_ts=60 +a=6378.14 +b=6356.75 +x_0=0 y_0=0'}


{'geo_column_offset': array([0.], dtype=float32),
 'geo_dim_pixel': b'KM,KM',
 'geo_number_columns': array([700], dtype=int32),
 'geo_number_rows': array([765], dtype=int32),
 'geo_par_pixel': b'X,Y',
 'geo_pixel_def': b'LU',
 'geo_pixel_size_x': array([1.0000035], dtype=float32),
 'geo_pixel_size_y': array([-1.0000048], dtype=float32),
 'geo_product_corners': array([ 0.      , 49.362064,  0.      , 55.973602, 10.856453, 55.388973,
         9.0093  , 48.8953  ], dtype=float32),
 'geo_row_offset': array([3649.982], dtype=float32)}

From the former I can easily obtain the CRS, but I don't fully understand the meaning of the parameters in the latter, although I'm pretty sure they implicitly define the transform.

How can this be used to produce a regular CRS and tranform understandable by rasterio?

Answer 1

You are correct, this information implies the transform and you will also want to fill out the profile so you can write the data to a GeoTIFF via rasterio.

You are going for a python dict that will look something like this:

{'affine': Affine(1.0000035, 0.0, x_corner, 0.0, -1.0000048, y_corner), 'count': 1, 'crs': CRS.from_string('+proj=stere +lat_0=90 +lon_0=0 +lat_ts=60 +a=6378.14 +b=6356.75 +x_0=0 y_0=0'})}), 'driver': 'GTiff', 'dtype': 'float32', 'height': 700, 'interleave': 'band', 'nodata': -9999.0, 'tiled': False, 'transform': (x_corner, 1.0, 0.0, y_corner.0, 0.0, -1.0000048), 'width': 765}

I've tried to fill those values as best I can guess from the data you posted - you'll need to figure out what to substitute for 'xcorner' and 'ycorner' though.