Implement convenience mode

[konstntokas]

"Convenience mode" describes a user-friendly mode, which shows the "measurement" group as a xr.dataset. This means specifically for the individual Sentinel products:

Satellite	Product	Data structure
Sentinel-1	GRD	2 data sets (VH=cross-polarization) and (VV=co-polarization) is presented as 2d dataset with dims (azimuth_time, ground_range), present the two data sets as
Sentinel-1	SLC	Includes VV and VH, three sub swaths IW1, IW2, IW3, and multiple bursts along the azimuth time; multiple burst could be stitched together along the azimuth time axis. is there stitching along the 3 sub-swaths possible?
Sentinel-1	OCN	contains Ocean Wind field (OWI), Ocean Swell spectra (OSW), Surface Radial Velocity (RVL), all of them have difference dimensions
Sentinel-2	L1C & L2A	Bands are available in different spatial resolutions; user selects bands and spatial resolution; if spatial resolution is not available, resampling is performed; nearest neighbor and bilinear interpolation shall be supported;
Sentinel-3	OLIC EFR	rectification ?
Sentinel-3	SLSTR FPR	rectification ?

We can get inspiration from SNAP (https://github.com/senbox-org)

[konstntokas]

Sentinel-2 Product Access Modes

Mode	xr.open_datatree	xr.open_dataset
mode="native"	No modifications needed; Zarr is represented as an xr.DataTree as-is.	Applies only to Zarr groups (r10m, r20m, r60m). No processing needed—just filters and error handling before opening the data.
mode="analysis"	❓ (Unclear behavior)	Allows band and spatial resolution selection, returning a flat xr.Dataset with resampling if needed.

@forman please edit the table for further adjustments

[forman]

Sentinel-2 Product Access Modes

Mode	xr.open_datatree	xr.open_dataset
mode="native"	No modifications needed; Zarr is represented as an xr.DataTree as-is.	Applies only to Zarr groups (r10m, r20m, r60m). No processing needed—just filters and error handling before opening the data.
mode="analysis"	Allows band and spatial resolution selection, keeping the tree structure with resampling if needed. Add CF-compliancy, e.g., grid mapping.	Allows band and spatial resolution selection, returning a flat xr.Dataset with resampling if needed. Add CF-compliancy, e.g., grid mapping.

EDIT

"analysis" mode: Add CF-compliancy, e.g., grid mapping.

[konstntokas]

For Sentinel-2 add information on the resampling to the attribute of each band/data variable.

[forman]

For Sentinel-2 add information on the resampling to the attribute of each band/data variable.

This should apply to all product types.

[konstntokas]

For Sentinel-2 L2a, add colors for SCL:

https://custom-scripts.sentinel-hub.com/custom-scripts/sentinel-2/scene-classification/

    cube["scl"].attrs = dict(
        flag_values=[0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11],
        flag_meanings=(
            "no_data saturated_or_defective_pixel topographic_casted_shadows "
            "cloud_shadows vegetation not_vegetation water "
            "unclassified cloud_medium_probability "
            "cloud_high_probability thin_cirrus snow_or_ice"
        ),
        flag_colors=(
            "#000000 #ff0000 #2f2f2f #643200 #00a000 #ffe65a #0000ff "
            "#808080 #c0c0c0 #ffffff #64c8ff #ff96ff"
        ),
    )

[konstntokas]

@forman FYI

ESA would like to set the interpolation / aggregation mode to align with S2 current way defined in https://step.esa.int/thirdparties/sen2cor/2.10.0/docs/S2-PDGS-MPC-L2A-ATBD-V2.10.0.pdf:

• down-sampling (i.e. from 10 m to 20 m resolution) a method based on block means is used (note SCL is not down-sampled in the current product, thus not defined)
• up-sampling (i.e. from 60 m to 20 m resolution) a cubic spline method is used. SCL map is resized from 20m to 60m with an order of 0 for spline interpolation, in order to keep the classification values.

[forman]

note SCL is now down-sampled in the current product, thus not defined

What do you mean by now? In our code, or in the current Zarr sample products? Why is it not defined?

[forman]

SCL map is resized from 20m to 60m with an order of 0 for spline interpolation, in order to keep the classification values.

SCL and other categorical variables are not handled yet. Indeed they need a special treatment when resampling them.