Dataset selection criteria

[jveitchmichaelis]

@bw4sz following your comment in the issue, we can think about user options for filtering data. In my mind there are some obvious ones we've talked about:

• Label provenance (similar to GBIF/Inat), was the annotation done by hand or auto/semi-automated?
• Validation: field, LIDAR, none?
• Real or Synthetic
• Geographic coverage? We could expose a very coarse tag like terrestrial ecoregion ([0,13]) which is still privacy preserving
• Spatial scale? Again could expose GSD as metadata without breaking privacy
• License? Are all datasets CC-BY, etc.
• Tree Coverage (overall %) and label coverage (can estimate this with Restor model)

We'd also need to think about categories for some of these, though some seem straightforward. And how we tag data - image level I guess?

[ethanwhite]

This sounds great to me. @bw4sz and I were also just chatting yesterday about providing ways for users to filter data for projects with different size-quality tradeoffs. I'd also add providing a list of individual datasets to include for folks who want to make decisions guided by their broader understanding of the datasets that can be hard to fit into metadata