Spatial resolution cardinality

[jakubklimek]

In GeoDCAT-AP 3.0.0 revision, a discussion about the relations and cardinality of Spatial resolution on different classes (Dataset, Data Service, Distribution) was had, resulting in resolution of SEMICeu/GeoDCAT-AP#100.

It resulted in the following conclusions:

1. For Distribution, spatial resolution [0..1] represents the spatial resolution of the described file.​
2. For Data Service – spatial resolution [0..*] describes the capabilities of the data service, i.e. in which spatial resolutions it can serve data.​
3. For Dataset – spatial resolution [0..*] describes the spatial resolution of the original data in the dataset, i.e. regardless of how it is distributed using distributions.​

This is now in conflict with DCAT-AP 3.0, where for Dataset, the spatial resolution as a cardinality [0..1] (related to #372)

[bertvannuffelen]

Dear community,

Do you agree with the relaxation of the cardinalities as described in the proposel by the GeoDCAT-AP community?

[HendrikBorgelt]

As I can only interpret the "conclusions from abocve" based on the definitions in the DCAT Vocabulary, spatial resolution would de defined as "Minimum spatial separation resolvable in a dataset, measured in meters.", "The minimum spatial separation resolvable in a dataset distribution, measured in meters." and/or "The minimum spatial separation resolvable in a data service, measured in meters.". From my knowledge of various Ontologies, this poses some challenges, because since every distribution is a dataset, it inherits the possibility of having a minimum resolvable spatial resolution of the original dataset while simultaneously having aspatial resolution based on being a distribution. As these definitions basically conflict, this is already critical. You can't distinguish between the definitions the definition of the spatial resolution of a dataset and its distribution. A dataset that is distributed might as such want to describe how precise the measurements of the distribution could have been, based on the original dataset, while it may still want to describe how precise it is based on just being a subset of the dataset. So I think it would be best to differentiate between the "spatial resolution of the original dataset" and the "spatial resolution of a distribution".

Coming back to the example and ignoring the inconsistencies in the definition. A dataset would in theory only have the best of all spatial resolutions, by the definition of spatial resolution. Why should one have multiple "Minimum resolvable spatial separation[s] "? x_1 < x_2 < x_3 --> min of (x_1, x_2, x_3) = x_1 ! Something multivalued would not be feasible and as such not interpretable.

[bertvannuffelen]

@HendrikBorgelt thanks for your feedback.

As I can only interpret the "conclusions from abocve" based on the definitions in the DCAT Vocabulary, spatial resolution would de defined as "Minimum spatial separation resolvable in a dataset, measured in meters.", "The minimum spatial separation resolvable in a dataset distribution, measured in meters." and/or "The minimum spatial separation resolvable in a data service, measured in meters.". From my knowledge of various Ontologies, this poses some challenges, because since every distribution is a dataset, it inherits the possibility of having a minimum resolvable spatial resolution of the original dataset while simultaneously having aspatial resolution based on being a distribution. As these definitions basically conflict, this is already critical. You can't distinguish between the definitions the definition of the spatial resolution of a dataset and its distribution. A dataset that is distributed might as such want to describe how precise the measurements of the distribution could have been, based on the original dataset, while it may still want to describe how precise it is based on just being a subset of the dataset. So I think it would be best to differentiate between the "spatial resolution of the original dataset" and the "spatial resolution of a distribution".

Two points of consideration:
-In DCAT-AP we clearly make a distinction between a Dataset (a conceptual thing) and a Distribution (a file based sharing of that dataset). In DCAT-AP a Distribution shall not exists on its own. There is always a Dataset for a Distribution. Thus in your reflection, the situation you have to consider is that a publisher has denoted a collection of data points as a dataset and shares this via an API (Data Service) or via a file download (Distribution). Without going into a more conceptual discussion I think this the case we are considering.

• in GeoDCAT-AP workshops, the community made clear that Datasets can be managed with multiple resolutions (eg. city areas are fine grained, while rural areas are coarse grained.) In order to facilitate the sharing of that information multiple values must be allowed. We agreed with the community that in case of a Distribution a single value is acceptable.

Coming back to the example and ignoring the inconsistencies in the definition. A dataset would in theory only have the best of all spatial resolutions, by the definition of spatial resolution. Why should one have multiple "Minimum resolvable spatial separation[s] "? x_1 < x_2 < x_3 --> min of (x_1, x_2, x_3) = x_1 ! Something multivalued would not be feasible and as such not interpretable.

Yes it is ignoring the "minimum" in the definition. But if the community of practice (e.g. INSPIRE) informs us that multiple resolutions are the practice then this definition is maybe inadequate. We will raise it also to W3C.

As for the multivalued, if it exists in practice, then we have the following options:

• force users to adapt their view on the cardinality,
• adjust the specification

In this case we have chosen the second and with this issue we raise the decision to the DCAT-AP level.