Closed jyutzler closed 7 years ago
I was instructed to take this up with the Coverage WG, which I have. They have acknowledged receipt of the question and I am now waiting for their feedback.
We concluded that the best way to handle this is to assume pixel-is-area. However, this is a client-interpretation detail and therefore not something that can be mandated as part of a data encoding. Therefore it is permissible to use pixel-is-point with the understanding that clients might not use it that way without a separate agreement to do so. We don't have a standardized way to encode the decision.
We deliberately didn't specify this because we didn't want to mandate one flavor over another. We felt that this could be handled out of band or through the metadata. The fact is that we don't have an interoperable way to express that information and mandating one way or another was deemed too heavy-handed. This is an issue that transcends GeoPackage and applies to anyone trying to do gridded coverages. I will take this up with OGC's architecture board to find out if there is any guidance for handling the situation properly.
Meanwhile, we at least need to acknowledge the issue in the text.