excessive adding of "band" dimensions

[soxofaan]

Originates from discussion on forum https://discuss.eodc.eu/t/exclusion-mask/441 (unfortunately user started this as private message instead of forum topic)

user does a couple of operations like

s2_cube_water = s2_cube.reduce_dimension(reducer = water_function, dimension = "bands")
s2_cube_water = s2_cube_water.add_dimension("bands", "water_prob", type = "bands") 

...

s2_cube_water_threshold = s2_cube_water.apply_dimension(dimension="bands", process=lambda x: if_(x > 0.75, x,0))
s2_cube_water_threshold = s2_cube_water_threshold.add_dimension("bands", "w_T75", type = "bands")

...

s2_cube_swf = s2_cube_water_sum.resample_cube_spatial(s2_count) / s2_count
s2_cube_swf = s2_cube_swf.add_dimension("bands", "swf", type = "bands") 

The first one is ok: first eliminate the "bands" dimension and add it back. But the other two cases are risky: apply_dimension or resample_cube_spatial do not eliminate the "bands" dimension, but the user is still adding another (?) "bands dimension.

Further down the line, when executing the whole graph, the user gets non-helping errors (e.g. "java.lang.IllegalArgumentException: Invalid band index 5, only 4 bands available.").

I think we should throw a more clear error here.
Maybe even this can be done client-side

[soxofaan]

from https://processes.openeo.org/#add_dimension:

Errors/Exceptions

    DimensionExists
    Message: A dimension with the specified name already exists.

[soxofaan]

should be covered now

[sulova]

"reduce_dimension" is applied I am getting also an error message OpenEoApiError: [400] DimensionExists: A dimension with name bands already exists. (ref: r-49f1fdda306a4c8abd63667dd0e51653)

This is the script:

start_date           = '2021-06-01'
spatial_extent       = {'west': -74.06810760, 'east': -73.90597343, 'south': 4.689864510, 'north': 4.724080996, 'crs': 'epsg:4326'} #colombia
zone                 = "tropical"

start_date_dt_object = datetime.strptime(start_date, '%Y-%m-%d')

end_date             = (start_date_dt_object + relativedelta(months = +2)).date() ## End date, 1 month later (1st Feb. 2021)
start_date_exclusion = (start_date_dt_object + relativedelta(months = -2)).date() 
bands                = ['B02', 'B03', 'B04', 'B08', 'CLP', 'SCL' , 'sunAzimuthAngles', 'sunZenithAngles'] 



s2_cube = connection.load_collection(
    'SENTINEL2_L2A_SENTINELHUB',
    spatial_extent  = spatial_extent,
    temporal_extent = [start_date_exclusion, end_date],
    bands           = bands)


s2_count = s2_cube.filter_bands(bands = ["B08"])
s2_count = s2_count.reduce_dimension(reducer=lambda data: data.count(), dimension = "t")

s2_count = s2_count.add_dimension("bands", "count", type = "bands")
# s2_count = s2_count.rename_labels("bands",["count"])


s1_s2_mask_save = s2_count.save_result(format='netCDF') #GTiff #netCDF
my_job  = s1_s2_mask_save.create_job(title="s2_count")
results = my_job.start_and_wait().get_results()
results.download_files("s2_count") 

I thought that for reduce_dimension should not be a problem.

[soxofaan]

s2_count = s2_count.reduce_dimension(reducer=lambda data: data.count(), dimension = "t")

s2_count = s2_count.add_dimension("bands", "count", type = "bands")

In first statement you do reduction of the "t" dimension, which indeed eliminates the "t" dimension, but the "bands" dimension is still there, so you can not add another bands dimension in the second statement