how to associate a `process` with a `spatial region`?

[dr-shorthair]

I'm modelling the capture of data associated with a spatial region, e.g. in the context of remote sensing and imagery.
The sensor is pointed at a region with the intention of ultimately characterizing its contents.
But in the initial phase, while the content is not known, the spatial boundary is.
So in my application I've defined region of interest as a subclass of spatial region with the definition:

"Spatial region that is the subject of investigation or observation".

I'd like to axiomatize it properly, so this requires associating it with the process of image capture.

(The process-flow for remote sensing involves one or more Planned Acts:.
Act of Observation, Act of Measuring, Act of Information Processing are already included in the EventOntology. )

I'm looking at the object-properties that have process as the domain.
has participant (and its sub-properties has input has object affects) looks like the right connection point.
But it explicitly excludes spatial region in the range.

1. Why the specific exclusion?
2. How can I allow a process to refer to a spatial region?

@avsculley @paulhed @JJBittner @DanielleZLimbaugh

[JJBittner]

@dr-shorthair The short answer is that has participant is not the right relation for this case, and in Basic Formal Ontology (BFO) it cannot be used for a spatial region.

In BFO, has participant has domain process, but its range explicitly excludes spatial region. Its range is:

specificially dependent continuant or generically dependent continuant or (independent continuant and not spatial region)

So an assertion like:

:imageCaptureAct bfo:has_participant :regionOfInterest

would not merely be a questionable modeling choice. If :regionOfInterest is asserted as a subclass or instance of spatial region, then that use conflicts with the stated range of has participant.

The deeper issue is that a region of interest should not be modeled as the site or spatial region itself.

A site is an independent continuant. It persists as the site that it is whether or not anyone is currently observing it, targeting it, tasking against it, or treating it as operationally relevant. By contrast, being a region of interest is dependent on an information or planning context. A site may become a region of interest because a battle is occurring there, because a sensor is tasked against it, or because a Tactical Operations Center (TOC) has identified it as relevant. After the battle ends, or after the TOC moves on, that same site may no longer be a region of interest.

If RegionOfInterest were modeled as a subclass of site, then this would incorrectly make region of interest status intrinsic to the site as an independent continuant. That is not what is intended. The site persists, but its status as a region of interest does not.

So I would model RegionOfInterest as a Generically Dependent Continuant (GDC), more specifically as an Information Content Entity, that is about a site, spatial region, or independent continuant. This is similar to the way a name such as "Paris" is not the site itself, but an information content entity that designates or is about the site occupied by Paris. The site persists independently. The designation or interest status depends on an information bearing context.

For example:

:RegionOfInterest
subclass of Information Content Entity
and is about some Site or Spatial Region

or, if you want to keep the operational targeting sense clearer:

:RegionOfInterestSpecification
subclass of Information Content Entity
and is about some Site or Spatial Region

Then the planned act can use that information content entity as an input:

:ImageCaptureAct
subclass of Act of Observation
and has input some RegionOfInterestSpecification

The image or observation record produced by the process can also be about the same site or spatial region:

:RemoteSensingImage
subclass of Information Content Entity
and is about some Site or Spatial Region

This separates three things that should not be collapsed:

1. The process, such as an image capture act.
2. The independent continuant or spatial region where the process is located or toward which it is directed.
3. The information content entity that identifies something as a region of interest.

If the intent is to say where the image capture process occurs, then use a location relation:

:imageCaptureAct bfo:located_in :someSiteOrSpatialRegion .

That places the process.

If the intent is to say what the tasking order, observation plan, capture request, image, or observation record is about, then use an information content relation:

:roiSpecification123 cco:is_about :site123 .

:someCaptureAct cco:has_input :roiSpecification123 .

:capturedImage123 cco:is_about :site123 .

A generically dependent continuant, such as an information content entity, is not located in a spatial region in the same way a material entity or process is. Its bearer may be located somewhere, but the information content entity itself is about something. What it is about may be a site, a spatial region, or independent continuants located there.

This also handles the remote sensing case cleanly. In the initial phase, the boundary or site may be known, while the contents are unknown. The region of interest specification can be about the site or spatial region without asserting what material entities are contained there. Later, after observation or information processing, additional assertions may be made about entities detected at or located in that site.

So I would not define RegionOfInterest as:

"Spatial region that is the subject of investigation or observation."

That definition makes region of interest status sound intrinsic to the spatial region.

A better definition would be:

"An information content entity that designates a site, spatial region, or independent continuant as the spatial scope of an act of observation, measurement, information processing, or operational attention."

In short:

Use located in to place the process. Use is about or designates to connect the region of interest information content entity to the site, spatial region, or independent continuant. Do not use has participant for the spatial region, because BFO explicitly excludes spatial regions from the range of has participant. And do not make RegionOfInterest a subclass of site, because a site persists as an independent continuant, while being a region of interest is dependent on an information or operational context.

[pahed]

@dr-shorthair,
To answer your first question, "why the specific exclusion?", I found in the BFO book (https://www.penguinrandomhouse.com/books/657278/building-ontologies-with-basic-formal-ontology-by-robert-arp-barry-smith-and-andrew-d-spear/9780262527811/?ref=PRHB7BD76FBDB&aid=14813&linkid=PRHB7BD76FBDB), page 108, that immaterial entities divide into two major subgroups: Boundaries/Sites and Spatial Regions. Where the distinction is that boundaries and sites are directly related to material entities, and spatial regions exist independently of material entities. The definition of "process" requires the use of a material entity as a participant, therefore, since spatial regions exist independently of material entities, a process cannot be associated with them and hence the "not spatial region" in the range axiom.

I am not sure what the correct approach is for your second question either what Jay has stated in his response or perhaps your area of interest could be a fiat surface boundary in which case the use of process would work.

[JJBittner]

Thanks, Simon. I think that Barry does a good job explaining why the exclusion exists in his book, that is a must read for anyone building with BFO.

The important distinction is that BFO separates immaterial entities into sites and boundaries on one side, and spatial regions on the other. Sites and boundaries are tied to material entities, while spatial regions are not. Since processes require participants, and participation is grounded in continuants that can participate in a process, spatial regions are excluded from the range of has participant.

So I agree that has participant should not be used to connect an image capture process directly to a spatial region.

I am less sure that the right solution is to model the area of interest as a fiat surface boundary, unless the thing being identified really is a boundary of some material entity. A fiat surface boundary may work if the intended target is literally the surface boundary of a material entity, such as the surface of a vehicle, building, terrain feature, or other bounded material entity. But in the remote sensing case, the “area of interest” often seems to be an operational or informational designation, not the boundary itself.

That is why I think the safer pattern is to distinguish:

1. the material entities participating in the process, such as the sensor, platform, operator, or imaging device;
2. the site or spatial region associated with where the process occurs or toward which the sensor is directed;
3. the information content entity that designates something as the area or region of interest.

On that pattern, the area or region of interest is not itself the site or spatial region. It is an information content entity, or perhaps a directive information content entity, that is about or designates a site, spatial region, or material entity.

For example:

:AreaOfInterestSpecification
subclass of Information Content Entity
and is about some Site or Spatial Region

or, in an operational tasking context:

:AreaOfInterestSpecification
subclass of Directive Information Content Entity
and designates some Site or Spatial Region

Then the image capture act can take that specification as input:

:ImageCaptureAct
subclass of Act of Observation
and has input some AreaOfInterestSpecification

The captured image or resulting observation record can also be about the same site, spatial region, or entities located there:

:CapturedImage
subclass of Information Content Entity
and is about some Site or Spatial Region

This also addresses the temporal issue. A site can persist as an independent continuant before, during, and after the battle or observation event. But being an area of interest is context dependent. A Tactical Operations Center may designate a site as an area of interest during a battle, and later that same site may no longer be an area of interest once the operational context changes.

So I would avoid making AreaOfInterest or RegionOfInterest a subclass of site, because that would make the interest status look intrinsic to the site. I would also avoid making it a fiat surface boundary unless the intended referent is specifically a boundary of a material entity.

In short, I think the model should be:

The process has material participants.

The process may be spatially located in relation to a site or spatial region.

The area or region of interest is an information content entity that is about or designates the site, spatial region, or material entities of concern.

That preserves the BFO exclusion on has participant, avoids making the site itself temporarily become a region of interest, and still supports the remote sensing case where the boundary is known before the contents are known.

[dr-shorthair]

Thanks @JJBittner @pahed for your responses.

1. I'm comfortable that I understand the immaterial entity hierarchy. I understand that site and continuant fiat boundary are anchored to other things so are not appropriate for this purpose.
2. I proposed that region of interest is a subclass of spatial region. This followed from the discussion in https://gitlab.ogc.org/ogc/T21-GIMI/-/work_items/51 . The textual definition follows the genus-differentia form "Spatial region that is the subject of investigation or observation" - i.e. that is subject of a process or more specifically of a planned act. The dimensionality is not limited - from my geophysics background I'm very comfortable with imaging lines (flightlines, boreholes, transects ...) and volumes (in the atmosphere, ocean, and underground) as well as surfaces.
3. An individual region of interest may be the "real-world-referent" of an image, where the region but not the content is known at the time of image capture. So the relation from an Image (of a region) (a subclass of Representational Information Content Entity) to the region of interest (a subclass of spatial region) is represents (cco:ont00001938).
4. The question begged by the textual definition is how to axiomatize the connection between the region of interest and the Planned Act of "investigation or observation" (e.g., image capture). You guys have pointed out that has participant (or its sub-properties) cannot be used in this case since the range is a material entity. occurs at and occurs in are along the right lines, except (i) the range is site or material entity (ii) they seem to presume a local or in situ spatial relationship, and don't seem to allow for a remote spatial association*. Maybe this hasn't come up in the BFO or CCO world before. I've looked at the CCO ExtendedRelationOntology but can't see an existing solution. Are we are treading new ground here?

*One of the key innovations of O&M (the predecessor of the SSN Ontology) was to allow for separate locations for the feature-of-interest and the observer. This enabled in situ, ex situ and remote observations to be described using the same formalization. Previously there had been conflicting ideas about the location of the 'Observation' which had caused problems in interpretation, and difficulties in crossing over between application domains.

@JJBittner you suggest pushing region of interest over into the Generically Dependent Continuant (GDC) branch of the BFO hierarchy. I disagree. The region exists in space, it is not a dependent continuant. The camera is pointed at the region so the image is of the region. We don't know what material things are in the region, so we accept that it is merely a kind of ... a continuant part of the spatial projection of a portion of spacetime at a given time (quote from the definition of spatial region). Specifically, it is one that is subject of investigation or observation. We discussed this at length back in January. Perhaps the subclass is not needed and spatial region is enough. But we can differentiate this kind of spatial region, which usually implies a sub-class.

At some point in the discussion @alanruttenberg agreed and used the phrase 'real world referent for an image' to describe it. That settled it for me. We don't know what is there, but we know where it is. The region has a geometric description (an ICE) that provides the geospatial coordinates of interest. But we don't need another indirection. BFO's spatial region recognises that such an immaterial entity is required, so I'm planning to use it. (Unfortunately I can't locate my record of the discussion with Alan.)

@pahed I have ordered a copy of the BFO book for myself so that we can cross-reference this when necessary.