Document consistent design patterns for use of overlaps relation for areas

[cmungall]

ENVO uses overlaps assertions to connect areas (immaterial) to environments (biomes, environments).

Note that while an intuitive reading of the string "overlaps" might suggest a significant spatial overlap, in fact RO follows standard mereology, where overlaps is trivially satisfied only if a single atom is shared. Thus overlaps is often trivially true, and asserting it could be considered an antipattern. It is better to either

• make two assertions chaining has-part and part-of, naming the intermediate(s). E.g. "nervous system overlaps visual system" is true but not useful; "visual system has-part eye, eye has-part retina ganglion layer, retina ganglion layer part of nervous system" entails overlap and explains why the overlap exists
• use a stronger relation (especially for connecting to shadow classes, where definitions mirror)

Another common mistake with overlaps is to assume that ontology assertions are symmetric, because it's a symmetric object property. In fact, for ontologies (TBoxes) using existential restrictions, the inference only holds in one direction

With that said, let's look at all of the ENVO overlaps assertions where either subject or object is an area:

(white=area, cyan=biome/ecosystem/environment/ABD)

• sometimes the assertion is in the direction of area->material, sometimes the other (but never both)
• frequently the assertion connects a shadow (area) to a base concept, sometimes the two concepts are distinct

while the assertions are all true, the underlying use cases and SOP for when to assert is not clear. When adding new terms, which direction should these be asserted in? For a concept X, should the biome or ecosystem be asserted, or both? How do we know when these assertions are complete?

There are also a small number of cases of overlaps being used to define areas in terms of overlap with ecosystems or other areas:

[Term]
id: ENVO:00000043 ! wetland area
def: An[sic] vegetated area which overlaps a wetland ecosystem.
intersection_of: ENVO:01001305 ! vegetated area
intersection_of: RO:0002131 ENVO:01001209 ! wetland ecosystem

[Term]
id: ENVO:00000111 ! forested area
def: An area with a high density of trees. A small forest may be called a wood.
intersection_of: ENVO:00000109 ! woodland area
intersection_of: RO:0002131 ENVO:01001243 ! forest ecosystem

[Term]
id: ENVO:01000242 ! desert scrubland
def: A scrubland which is located in a desert
intersection_of: ENVO:00000300 ! scrubland area
intersection_of: RO:0000086 ENVO:01000230 ! arid
intersection_of: RO:0002131 ENVO:00000097 ! desert area

The first two are likely incorrect, due to the weakness of overlaps. Consider a very large vegetated area that shares 1m^2 of land with a neighboring wetland ecosystem. This will be classified as a wetland area using these OWL definitions.

The third is less problematic, although the overlaps clause is doing no real work here, arid is acting as the main differentia. Also of concern:

• it's dubious whether immaterial entities can have material qualities such as being arid
• the textual definition is considerably different from the logical definition

Recommendations

1. Use a stronger http://purl.obolibrary.org/obo/RO_0002379 spatially coextensive with in logical definitions to connect an area-of-X shadow to X{ecosystem,biome}
2. assert in the direction of immaterial shadow to concrete, for consistency
3. remove non-defining overlaps assertions, optionally replace these with something more useful

[cmungall]

Note: since this ticket was created, #1390 was merged, which moves many of the environment areas out from immaterial entity