allow setting tiff scale + offset + custom metadata

[jdries]

Add a format option to set tiff metadata. Some of these are regular tiff tags, others will have to be encoded as gdal band metadata.

The options in geotrellis are limited, but there are very basic tifftools available in linux:
https://linux.die.net/man/1/tiffset
The idea is to be able to set metadata tags without touching the rest of the file, avoiding a full rewrite. Ideally this is done right after writing the tiff in geotrellis.

Format option will have to be passed through via save_result.

[VictorVerhaert]

example of custom metadata: https://github.com/VITO-RS-Vegetation/lcfm-production/issues/18

[jdries]

For tiffset, we would need to add libtiff-tools to the container.

[JorisCod]

The current metadata, visible through gdalinfo, is:

Metadata:
PROCESSING_SOFTWARE=0.39.0a1
AREA_OR_POINT=Area

both can stay, but what we are looking for is:

Metadata:
AREA_OR_POINT=Area
bands=['s2-B02-p10', 's2-B02-p25']
copyright=LCFM project 2020 / Contains modified Copernicus Sentinel data (2020) processed by LCFM consortium
creation_time=2024-07-19 00:35:04.693848
license=CC-BY 4.0 - https://creativecommons.org/licenses/by/4.0/
product_crs=EPSG:32629
product_grid=Sentinel-2 UTM tiling grid
product_tile=29TNE
product_type=LSF monthly median composite for band B02
reference=TODO
time_end=2020-02-29T23:59:59Z
time_start=2020-02-01T00:00:00Z
title=LCFM Monthly Land Surface Features (LSF-MONTHLY) product at 10m resolution for year 2020
version=v002-satio

So this would be something quite flexible. We are setting the metadata through rasterio (python) and passing a dictionary:

        if metadata:
            with rasterio.open(final_vrt_fn, 'r+') as dst:
                dst.update_tags(**metadata)

        if bands_names:
            with rasterio.open(final_vrt_fn, 'r+') as dst:
                for i, b in enumerate(bands_names):
                    dst.set_band_description(i + 1, b)

Additionally, there is a difference in how we set the band description and how OpenEO does it:
OpenEO:
Band 2 Block=64x64 Type=Int16, ColorInterp=Undefined
NoData Value=-32768
Overviews: 1033x1542, 517x771, 259x386, 130x193, 65x97, 33x49
Metadata:
DESCRIPTION=B02_P25

We:
Band 17 Block=1024x1024 Type=UInt16, ColorInterp=Undefined
Description = s2-B08-p25
NoData Value=65535
Offset: 0.0031999999191612, Scale:8.16687767161827e-06

The description in OpenEO is at metadata level, which makes that the band names are not displayed in the symbology in Qgis.

Relatedly, as you see in the output, there's an offset and scale on every band. In principle, the offset and scale can be different per band.

[JorisCod]

Just checking whether this issue is already planned?
If it's too extensive, can the scaling and offset be done first and the metadata in a separate issue?

[JorisCod]

We set the scale and offset through gdal_translate:
https://gdal.org/en/latest/programs/gdal_translate.html#cmdoption-gdal_translate-a_scale
but there are likely other options.