Open Science
Research data management in agriculture
The unglamorous layer that decides whether dispersed agricultural research is searchable at all - metadata standards, controlled vocabularies and the discipline that produces them.
Every search that works depends on a layer nobody sees. If one repository records a field as creator and another as author, if one uses a controlled vocabulary for crop species and another uses free text, and if neither records what units a measurement used, then aggregating them produces a heap rather than an index.
This was the problem the research project that formerly occupied this domain worked on, and it is worth understanding because it has not gone away.
Why agriculture is a hard case
Most scientific data is context-dependent. Agricultural data is unusually so.
A yield number is meaningless in isolation. It needs the cultivar, the soil, the weather during the season, the management history, the plot size, the harvest method and the moisture basis the weight was recorded at. A soil moisture reading needs the depth, the sensor type, the calibration and the soil texture. An aquaculture growth rate needs species, temperature, stocking density, feed and water quality.
Two datasets that appear to record the same variable often measured different things. This is why agricultural metadata standards carry so much apparatus, and why simply publishing a spreadsheet does not make data reusable.
The FAIR principles, plainly
Findable means a persistent identifier and enough descriptive metadata that a search can locate it.
Accessible means a documented route to obtain it using an open protocol. This does not mean unrestricted - it means the conditions are clear and the metadata remains available even if the data itself is protected.
Interoperable means using formats, vocabularies and identifiers that let the data be combined with other data. This is where controlled vocabularies earn their place.
Reusable means enough provenance, method detail and explicit licensing that someone else can legitimately use it and know what they are using.
The common misreading is that FAIR means open. It does not. Commercially sensitive or personal data can be fully FAIR while access remains controlled, and conflating the two has caused a lot of unnecessary resistance.
Controlled vocabularies
AGROVOC, maintained by FAO, is the multilingual thesaurus that underpins much of agricultural information management. It holds tens of thousands of agricultural concepts with equivalents across dozens of languages, structured as linked data.
Its practical value is that it decouples searching from wording. A record indexed with an AGROVOC concept can be found by someone searching a different term, or a different language, for the same thing. Given how much agricultural research is published in national languages, this is not a marginal benefit - it is often the difference between finding regionally relevant work and not knowing it exists.
Persistent identifiers do the same job for objects rather than concepts: a DOI for a dataset, an ORCID for a person, a ROR identifier for an institution. Each removes an ambiguity that would otherwise break automated linking.
A workable plan for a small trial
Formal data management plans are a funder requirement, and for a small on-farm or institutional trial the useful version fits on one page.
Record what is being measured and why. Record how - the instrument, its calibration status, the units, the sampling interval and the exact locations. Record who is responsible and where the files live, including the backup. Record what happens to the data at the end: which repository, what licence, what identifier.
Then, critically, write a short description of every column in every file, in words, at the time the file is created. Almost all practical data loss comes down to a spreadsheet whose headings made complete sense to the person who typed them and are opaque to everyone afterwards, including that person two years later.
Frequently asked questions
What does FAIR actually require?
That data be Findable, Accessible, Interoperable and Reusable. In practice: a persistent identifier and rich metadata so it can be located; a clear and documented route to obtain it, which need not mean fully open; formats and vocabularies that let it be combined with other data; and enough provenance and licensing information that someone else can legitimately reuse it. Notably, FAIR does not mean open - sensitive or commercial data can be FAIR while remaining restricted.
Why is agricultural data especially hard to combine?
Because it is intensely context-dependent. A yield figure means nothing without soil type, weather, cultivar, management history, plot size and measurement method. Two datasets recording the same nominal variable frequently measured it differently, at different depths, on different scales, in different units. Combining them requires metadata that captures all of that context, which is why agricultural metadata standards are unusually elaborate.
Do I need a data management plan for a small trial?
If anyone will use the results later, including you, then yes - though it can be one page. Recording what was measured, how, in what units, with what instrument, and where the file lives, takes minutes at the time and is close to impossible to reconstruct two years afterwards. Most data loss is not dramatic; it is a spreadsheet whose column headings nobody can now interpret.