Review Comment:
The ontology of observations remains in bad shape, theoretically as well as practically. Papers proposing to advance it are therefore more than welcome. This manuscript presents valuable and sensible practical advances over earlier work. In particular, it nicely combines insights about observation processes with pragmatic goals of providing adequate metadata vocabularies. As such it should be published, but it also needs work, mainly regarding its style and readability.
The main contribution of the paper is to propose two new OWL ontologies, om-lite and sam-lite, which appear useful and an improvement on the state-of-the-art, particularly through the ideas of sampling features and samples as items with provenance. The main argument advanced in favor of these ontologies, namely that they minimize dependencies on other ontologies, however, is a bit one-sided. The paper would have more stature if the pros and cons of such dependencies were briefly discussed and the chosen solution was rationalized based on reasoning support (rather than talking mainly about complexity and the discomfort of users). Along the same lines, claiming (in the abstract) that om-lite and sam-lite can be used to *harmonize* observational data is rather courageous. Apart from the ill-defined notion of harmonization itself, it seems obvious that terms with less semantics can be more broadly applied, but this does not mean that they harmonize or integrate anything. Also, it would be good to see a more solid discussion of the strengths and weaknesses of stub classes.
The characterization of SSN as mainly introducing the idea of a stimulus (over O&M before) is narrow. At the time, for example, O&M still had the ambiguous notion of a single feature of interest. While Cox now sensibly introduces the distinction of sampled and sampling features, it is not true that this distinction was not possible before. SSN, for example, offers the Sensor concept for this purpose, defined as "physical devices, computational methods, a laboratory setup with a person following a method, or any other thing that can follow a Sensing Method to observe a Property".
The mapping to PROV-O, as sensible as the choice is, got me rather confused. In the paper, saml: and oml:Processes are subclassed from prov:Entity, which seemed odd, but the correction in the comment (below) to subclass them from prov:Agent seems equally questionable. At least the paper should argue the avoidance of prov:Activity more substantially. A one paragraph explanation of PROV-O's basic concepts would help.
The weakest part of the paper is section 2, on the conceptual model, which oozes standard numbers and acronyms and leaves its diagrams often unexplained. In fact, figures in the paper as a whole lack explanations and sometimes appear redundant. What does Figure 6 add as an idea that Figure 2 did not contain already? Figure 8 is overloaded and not very informative, lacking explanations. Figure 9 may just not be worth the 100 words or so that it would take to explain the ideas in it, but in any case, it cannot replace these.
The main weakness of the paper overall is in fact its jargon and acronym-loaded style and its shortage on explanations. Using standardization jargon so pervasively detracts from ideas and insights. The problem starts in the abstract, which should of course be generally intelligible, but uses many acronyms and an ISO standard number without even explaining what that standard is about or what context it is taken from.
The same style continues throughout the paper, with a highlight in this phrase at the beginning of the conceptual model (!) section: "Note the use of types and classes from other ISO 19100-series models, indicated by the prefixes GF, TM, GM, MD, DQ, LI (from ISO 19109, 19108, 19107 and 19115)." Or consider the sentence at the end of 4.1, ending with: "some information that is provided in additional xlink attributes alongside the href in the GML implementation is not available locally in the RDF.".
Many phrases and sentences also need to have their English revised to become clear, starting with the tile (where the word "Basic" is used ambiguously), and continuing with spelling and grammar errors already in the first and second sentence. For another example, I could not understand this sentence (in the discussion section), even after adjusting the verb: "From this view it is clear that om-lite does provides a light-weight framework, in comparison with both SSN and OMU which any application to make a significant commitment to an existing framework".
Two more minor points:
- What are "data individuals" or "individual data instances"?
- The wfs.example.org links in the example listings are broken.
Overall, the paper reads like a submission that was somewhat hastily put together from standards documents and discussions, rather than providing the necessary reflection on and communication of the new ideas for a broader SWJ audience. It is based on good work that warrants publication, but the current exposure falls short of communicating it well, let alone making the read enjoyable.
|
Comments
PROV-O alignment
After submission I realised that the mapping between O&M and PROV-O given in 5.2.1 and 5.2.2 was incorrect, concerning saml:Process and oml:Process. A better fit is
saml:Process rdfs:subClassOf prov:Agent .
oml:Process rdfs:subClassOf prov:Agent .
and correspondingly
oml:procedure rdfs:subPropertyOf prov:wasAssociatedWith .
This will be corrected in revision.
oml:Measurement?
Listing 1 on page 9 uses the class
oml:Measurement
. I can't find this elsewhere in the paper. Should it have beenoml:Observation
?