Insect natural history is an extremely broad domain, which means that identifying an appropriate scope for a new data vocabulary and ontology is not a simple task. Our approach to this problem was to assemble example natural history data, drawn from real data sources, for each of the five major insect orders. This served two purposes. First, examining a well-drawn set of example data is a practical method for delimiting the scope of a new vocabulary and ontology, and second, a good example dataset also provides valuable test cases for use during vocabulary and ontology development.

To generate the example dataset, we worked in five small groups. Each group was assigned one of of the five major insect orders, and we ensured that each group included at least one entomologist with expertise in the assigned order. Then, each group gathered example natural history data for their insect order, with the goal of compiling a concise dataset that represented the various kinds of natural history information recorded on specimen labels for each major insect order. We attempted to capture both the breadth of biological information and the range of observational detail found in label data. Although we focused on information from insect specimen labels, we also included some data from literature sources and online databases such as iNaturalist (https://www.inaturalist.org) and GloBI (Poelen et al. 2014). For data from specimen labels, we used specimens and labels with digital images available on iDigBio (Page et al. 2015) whenever possible. Example data we gathered at the workshop were supplemented by additional example data that a few participants gathered both prior to and after the workshop.

Our final dataset includes 189 natural history observations covering a wide range of concepts and observation types (see next section). We expect that this dataset will have value to other researchers as well, so we have included it with this manuscript as two supplemental files, with one file formatted as a PDF document (Suppl. material 1) and one file in tabular comma-separated values (CSV) format (Suppl. material 2). Both of these files are also available in a public git repository hosted on GitLab which provides the example data in other formats, too, including styled HTML, Markdown, and a SQLite database (https://gitlab.com/stuckyb/inhd_ontology/tree/master/example_data).

After assembling the example data, we used them to delimit the high-level scope of the new vocabulary and ontology. Again working in small groups, we analyzed the kinds of information contained in the example data, with each group focusing on one of the five major insect orders. For each order, we summarized the kinds of biological information that were observed (e.g., various multi-organism interactions, developmental data) and the ways in which the information was recorded (e.g., qualitative or quantitative). Then, we reconvened as a large group, each small group reported their findings, and we synthesized the results to arrive at a set of 10 high-level conceptual areas required for the final ontology (Table 1).

Together, these conceptual areas cover virtually all of the kinds of information contained in the example data we assembled, and we therefore propose that an ontology that provides suitable coverage of all 10 of these areas will be sufficient for modeling nearly all insect natural history data from specimen labels as well as a substantial proportion of insect natural history data from other sources, including literature-based data. This conclusion is dependent, of course, on the extent to which our example data capture the conceptual breadth and depth of all available insect natural history information. Although we were not able to formally evaluate this, given the collective entomological expertise of the workshop participants (many of whom have years of experience examining specimens and labels from entomology collections around the world) and the effort spent compiling example data, we are confident that we at least came close to achieving this goal for natural history data from insect specimen labels.

We also note that several of these conceptual areas overlap with the domains of extant ontologies, and in Table 2, we list the ontologies that are most relevant to each conceptual area. To ensure broad compatibility, reusability, and extensibility, we plan to use existing ontological resources wherever possible and contribute (or suggest) new entities for extant ontologies, when appropriate.

Of the 10 conceptual areas we identified (Table 1), we determined that observations and observing processes, relationships and interactions, and positional (spatial) information were the most critical for developing an immediately useful vocabulary and ontology. Our decision to prioritize these areas was based on three considerations. First, observations and observing processes underlie all insect natural history data and encompass the crucial "who", "when", and "where" information about such data. Second, relationships and interactions are of broad scientific interest because they provide the raw ecological information needed for a wide variety of research topics (e.g., understanding trophic relationships, discovering potential disease vectors, or predicting the consequences of ecosystem changes). Third, we found that positional or spatial information is often included on specimen labels and in literature-based natural history observations, and we therefore concluded that even a minimal data standard should be able to capture such information. After prioritizing these three conceptual areas, we again worked in groups to begin sketching out data models (Simsion 2007, Simsion and Witt 2005) and ontology design patterns (Gangemi 2005) for all three areas and to identify the entities (concepts) to include in each conceptual area.

This initial design work revealed several critical data modeling challenges, the thorniest of which is the problem of recording metadata about natural history observations that include interactions between organisms. Such observations are common in natural history data and include, for example, observations about feeding relationships, parasite/host relationships, courtship, and many more. As with any other natural history observation, it is important to be able to record metadata about interaction observations, such as who made the observations, when they occurred, and so on. Without plunging into too much technical detail, the central problem is that the technology most often used for implementing ontology-enabled data, the Resource Description Framework (RDF, Miller 2005), currently has poor support for expressing metadata about interactions or relationships (Hartig 2017). A number of workarounds have been proposed (e.g., Hartig 2017, Nguyen et al. 2014, Hernández et al. 2015), but most of them have undesirable consequences, such as artificially increasing database size, complicating query statements, or slowing query response times (Hartig 2017, Hernández et al. 2015). Our work on this is ongoing, and we are actively investigating several different implementation strategies.

A second important data modeling problem is the challenge of accurately capturing information about what organisms were observed, which means dealing with the myriad difficulties posed by the use of taxonomic names (Zermoglio et al. 2016, Hardisty et al. 2013, Remsen 2016, Pyle 2016, Patterson et al. 2016). These issues are especially severe when dealing with data about insects, simply because insects are so extraordinarily diverse: many species remain undescribed and specimens in collections are often not identified to species (indeed, for some diverse insect families, the majority of specimens in a collection might not be identified to species). Relatively frequent – and sometimes dramatic – taxonomic changes mean that the names used in publications and labels can quickly become inaccurate or obsolete. These issues are certainly not unique to insect natural history data, and we have not attempted to add to the substantial work already done in this area (e.g., Franz et al. 2017, Franz and Peet 2009, Hardisty et al. 2013, Pyle 2016). For now, though, taxonomic integration remains a major challenge for virtually all biodiversity-related data aggregation efforts, and insect natural history data are no exception.

The last major task of our preliminary design and development work was drafting detailed ontology competency questions and identifying potential users and user cases. Ontology competency questions (OCQs, Grüninger and Fox 1995, Ren et al. 2014) provide a means for testing an ontology by providing specific queries that an ontology (along with an associated database) ought to be able to answer. In other words, OCQs specify how an ontology will be used to ask questions of real data. Thus, writing OCQs goes hand-in-hand with determining an ontology's users and use cases. To give a couple of examples, OCQs for an ontology of insect natural history data might include, “On what substrates does species A lay its eggs?” or “Has species B been collected at artificial lights?”

To identify use cases and develop OCQs, we divided into three groups on the last day of the workshop, with each group working independently and recording their results. After the workshop, one of us (BJS) synthesied the results of each group’s efforts into a single, comprehensive set of use cases and OCQs. The use cases we identified cover seven main user groups or domains:

1. Entomology (e.g., insect collecting and rearing, forensic entomology).
2. Taxonomy and systematics (e.g., field guides, systematic revisions).
3. Ecology and evolutionary biology (e.g., disease ecology, comparative studies).
4. Conservation biology and natural resource management (e.g., ecological restoration, environmental monitoring).
5. Agriculture and forestry (e.g., identifying potential pest insects, identifying potential disease vectors).
6. Education (e.g., classroom education, public outreach).
7. The general public (e.g., researching garden pests and control agents, hobby insect collecting).

The full sets of use cases and OCQs are too large to report in the main text, so we instead provide them in Suppl. material 3. The use cases and OCQs are also available in a public git repository on GitLab, which includes a SQLite database of use cases and OCQs along with example queries (https://gitlab.com/stuckyb/inhd_ontology/tree/master/OCQs).