Various suggestions

[m-mohr]

Various suggestions copied over from gitter:

Some thoughts:

1. see #3
2. see #1
3. see #3
4. Band math needs to be better defined. Like what operations and functions are allowed?
5. see #2
6. Let's remove raster specific things from file and clarify what should be in file and what in raster ;-)
7. I think it's a bad idea to define the values as they are and would prefer to make them like they are in file. I think the key as description can lead to conflicts. We have the case where we would have something like:

   {
   "land": 0,
   "land": 1,
   "ocean": 2
   }

   Duplicates in object keys don't really work...

8. I'm also a bit confused about the "from" in the Value object and would need a clarification

[emmanuelmathot]

some answers:

4. yes, it definitively needs a proper reference. For the moment, the baseline is the rio-tiler expression. From https://observablehq.com/@vincentsarago/the-power-of-rio-tiler-expression, it seems to be based on python numexpr: https://github.com/pydata/numexpr
5. IMO, this is the discussion about asset granularity in #3. Behind the raster extension proposal, there is the need to put metadata at band level
6. I agree on that one. I wanted a way to avoid duplicates. Not sure, you can constraint no field duplication in an array of objects
7. from field basically allows chaining several values (e.g. DN -> TOA radiance -> TOA reflectance)

[m-mohr]

7. Would the file:values specification for for you?

Example:

[
    {"value": [0], "summary": "clear"},
    {"value": [1,2], "summary": "clouds"}
]

8. A clarification in the extension itself would be good to have.

[emmanuelmathot]

For 7., I am not sure to understand what you propose. From file extension, I understand you describe each layer. Here you can describe several times the same band with different values. For instance, for typical remote sensing optical sensor the pixel can represent either the raw energy (DN), a TOA radiance or even a reflectance. Using what you propose, it would give something like:

"raster:bands": [
        {
          "values": [
            {
              "name": "TOA radiance",
              "unit": "W⋅sr−1⋅m−2⋅nm−1",
              "scale": 0.01,
              "offset": 0
            },{
              "name": "TOA reflectance",
              "scale": 0.0000185074513929,
              "offset": 0
            }]
          }]

[m-mohr]

For 7: I'm just proposing to take over the object schema from file for file:values. But it seems values actually is something different in your case. In your case it describes the values, while in my case it lists the possible values (e.g. for masks).

[emmanuelmathot]

@m-mohr I changed the values as array, does that look like what you had in mind?

[emmanuelmathot]

Discussed in meetup and transferred in #10.