result_meta dimension in aggregate_spatial

[soxofaan]

The devtelco of this afternoon led me to this " total_count and valid_count" feature of aggregate_spatial I was not aware of.

openeo-processes/aggregate_spatial.json

Line 81 in b81e96b

"description": "A vector data cube with the computed results and restricted to the bounds of the geometries.\n\nThe computed value is used for the dimension with the name that was specified in the parameter `target_dimension`.\n\nThe computation also stores information about the total count of pixels (valid + invalid pixels) and the number of valid pixels (see ``is_valid()``) for each geometry. These values are added as a new dimension with a dimension name derived from `target_dimension` by adding the suffix `_meta`. The new dimension has the dimension labels `total_count` and `valid_count`.",

The computation also stores information about the total count of pixels (valid + invalid pixels) and the number of valid pixels for each geometry. These values are added as a new dimension with a dimension name derived from target_dimension by adding the suffix _meta. The new dimension has the dimension labels total_count and valid_count.

I'm not sure this makes sense: you can not just "add a new dimension" to append metadata

For example, assume the input raster cube is a 2D (x,y) raster cube

Purely looking at the aggregation results, the output is also a 2D vector cube with a vector dimension (with a label for each geometry) and a "result" dimension (with just a single item, e.g. label "aggregation"). Attempt to visualize:

	"aggregation"
geom1	1.23
geom2	4.56
...	...

First observation: the "result" dimension is actually not necessary, the result could just be a 1D vector cube.

Now about the result_meta dimension and its labels total_count and valid_count. If you simply "add" this dimension, then you get a 3D vector cube, with dimensions:

• vector dimension (labels geom1, geom2, ...)
• "result" dimension (single label, e.g. "aggregation")
• "result_meta" dimension (labels "total_count", "valid_count")

but your aggregations are lost here: at what coordinates in this cube is the original aggregation stored? This result cube has only metadata.

I think what is intended is that total_count and valid_count should be new labels in the "result" dimension.
So the "result" dimension would have labels "aggregation", "total_count", "valid_count"

[soxofaan]

related to https://github.com/openEOPlatform/architecture-docs/issues/84

[soxofaan]

also related to the discussion around #341 (comment)

[m-mohr]

Yes, I also stumbled across this recently and it needs to be rephrased based also on the vector cube definition later.
One potential solution is indeed what you proposed and is also given as an example in the mentioned discussion: #341 (comment)

[soxofaan]

is also given as an example in the mentioned discussion: #341 (comment)

with the difference, I think, that the "result" and "band" dimension should be separate, instead of flattened as in that example

[m-mohr]

How should that work? You did argue above that this doesn't work if I understood it correctly?

[soxofaan]

How should that work?

It's like my original example, but one additional dimension:

start from 3D raster cube: (x, y, bands)
output should be (I think): 3D vector cube:

• vector dimension
• band dimension (like original input)
• "result" dimension with labels "aggregation" (or something alike), "total_count", "valid_count"