Connecting research data across disciplines at IASSIST 2017

IASSIST 2017 brought researchers, technologists and information specialists together at the University of Kansas in Lawrence from 23 to 26 May 2017. Its theme, “Data in the Middle: The Common Language of Research,” captured a practical problem: promising studies become difficult to combine when teams describe, store and govern information differently. The archived programme records plenaries and contributed sessions on big data, deep learning, digital agriculture and global food security, showing how the same methodological issue appeared in radically different research settings.

For an Australian audience, the archived schedule, plenary information and presentation listings offer more than historical interest. They reveal how people working with sensors, models, statistics and policy evidence negotiated a common research language. That concern remains immediate for teams distributed between metropolitan universities, regional campuses and field stations, and for collaborations that must satisfy privacy, ethics, security and Indigenous data expectations while moving information across institutional boundaries.

Why interdisciplinary data still matters

Interdisciplinary work exposes assumptions that remain invisible inside a single discipline. A crop scientist may treat rainfall as an environmental measurement, an economist may treat it as a production input, and a machine-learning specialist may treat it as a training feature. When each group uses a different unit, time interval and quality threshold, combining the records can silently change the meaning of the evidence.

The value of the IASSIST archive lies in this collision of perspectives. Its programme placed computational methods beside practical problems in food systems and agriculture, while plenaries framed data as something that must be communicated and governed, not merely transferred. The event therefore documented a wider definition of interoperability, one that includes language, documentation, rights and trust alongside file formats and technical compatibility.

A common language also determines whether reuse is possible. If a dataset lacks provenance, uncertainty information or a stable identifier, another team may spend months reconstructing what happened. If access conditions are unclear, a technically open resource may still be unusable. The conference model points toward a practical middle layer: shared descriptions, accountable stewardship and rules that let unrelated specialists evaluate the same evidence.

Lessons from the Lawrence conference

The archived site includes ordinary conference information—check-in, transportation and local recommendations—alongside the programme. Those details reveal a design principle that travels well beyond Lawrence: participation depends on the surrounding experience as much as the formal sessions. A registration desk, reliable local transport and clear guidance affect who can attend and how confidently newcomers engage with unfamiliar material.

Australian teams encounter a similar issue across distance and time zones. Researchers in Sydney, Melbourne, Brisbane, Perth, Adelaide and Canberra may work different hours, while partners at rural campuses and farms face patchy connectivity. Current work often unfolds through video meetings, instant messages and shared cloud drives, with field technicians collecting information on phones before synchronising it later. Good coordination records decisions and data locations so that a handover does not depend on memory or one busy individual.

The model becomes especially useful when a project treats coordination as documented infrastructure. Pairing an event schedule with practical project guidance can help teams turn participation into maintained outputs. In Australia, that work can draw on university e-research services, the Australian Research Data Commons and national compute facilities, while recognising that commercial storage, software and analytics services form a substantial local market with different costs and support models.

Shared data needs across domains

The programme’s sessions revealed that the structure of data may differ sharply even when the management questions converge. A cross-disciplinary comparison makes those common requirements clear:

Research context	Typical evidence	Shared interpretive need	Frequent friction
Digital agriculture	Satellite images, soil samples, sensor readings and farm records	Consistent location, season and crop identifiers	Satellite, laboratory and farm systems use incompatible codes
Deep learning	Training corpora, labelled examples, model versions and evaluation results	Traceable preprocessing, consent and data lineage	Source rights, hidden test sets and undocumented transformations
Global food security	Prices, production estimates, household surveys and supply-chain records	Comparable dates, units and geographic scales	Definitions and coverage differ across agencies and countries
Cross-institutional studies	Clinical, environmental and administrative datasets	Governance rules and secure linkage procedures	Consent, disclosure risk and incompatible approval timelines

These contrasts matter because interoperability cannot be achieved by converting everything to one format. Technical harmonisation may make columns align while leaving the underlying concepts incomparable. Teams need a shared account of what was measured, where and when it originated, which transformations occurred, and what limitations should travel with the data.

A grain-production study illustrates the issue. Satellite-derived biomass, rainfall totals and farmer observations may all contribute to a yield forecast, but each has a different error profile. The resulting model is scientifically stronger when those errors and overlaps remain visible. Common metadata should identify scale, coverage, uncertainty and responsibility for updates instead of hiding differences behind a clean merged file.

This is why the “middle” in the conference theme can mean a governance and communication layer. It can include data dictionaries, consent conditions, access categories, provenance records and named stewards. Those artefacts allow a nutrition researcher, software developer and policy analyst to understand not only where a file came from, but whether using it is appropriate for their question.

Big data, machine learning and agriculture

Big data sessions at IASSIST 2017 highlighted scale, but scale also magnifies poor decisions. A larger volume can conceal inconsistent categories, duplicated records and unrepresentative coverage. For Australian collaborators, national datasets must also be read against regional differences: drought in inland areas, intense rainfall events and isolated production zones cannot be treated as minor variations around an average national case.

Deep learning adds another layer of documentation. A trained network may perform well because its training data were selected, cleaned and labelled in a particular way. Other disciplines need records of model version, preprocessing steps, performance measures and the population represented by the examples. Without that lineage, a model can travel easily while the conditions supporting its conclusions remain opaque.

Digital agriculture connects both issues to decisions made under uncertainty. Sensors, satellite imagery, weather services and farmer diaries can support forecasting, yet their value depends on timely interpretation and usable delivery. Australian producers often work long distances from research centres, so mobile capture, low-bandwidth transfer and locally meaningful alerts matter. Global food security research then places those local signals alongside international trade, price and production data. The presence of major retailers such as Coles and Woolworths in Australian consumer markets makes that chain especially relevant, from wholesale prices to household food access.

Practical patterns for Australian teams

A useful starting point is a short set of agreements made before complex files begin moving between groups:

• Name a research question, data steward and decision owner for each partner.
• Create a shared data dictionary with agreed units, dates, locations and category labels.
• Assign persistent identifiers to datasets, code lists, models and the versions used in analysis.
• Record provenance and transformation steps alongside each analytical product.
• Use staged access, beginning with open metadata and sharing sensitive files under controlled conditions.

A second layer protects the people and environments represented in the evidence. This work should be reviewed with relevant institutional, community and legal expertise before data are collected:

• Assess obligations under the Privacy Act 1988, the Australian Privacy Principles and applicable state or territory health-records legislation.
• Apply Indigenous data sovereignty principles through consent, governance, ownership and benefit arrangements shaped with the relevant communities.
• Document de-identification, aggregation and disclosure-control methods, including residual re-identification risks.
• Distinguish open, mediated, restricted and embargoed access in machine-readable access conditions.
• Preserve essential records, licences, software environments and audit trails for long-term reuse.

Australian institutions can turn these agreements into routine practice through research data-management plans, repository deposits and cross-disciplinary data forums. They can also build them around familiar work rhythms: office-based teams may coordinate during shared commuting or meeting periods, while field partners need records that work on mobile devices and remain useful when connectivity is limited. Standard templates help, but recurring conversations with farm groups, health services, data providers and community representatives are what keep the rules aligned with actual research.

From archive to active collaboration

The archived conference schedule can support a durable model because it makes a temporary gathering visible as a sequence of knowledge exchange. Plenary presentations show how senior researchers frame broad questions; contributed talks reveal operational details; discussion and informal contact show where collaboration emerges. Institutions adapting this pattern can preserve recordings, slides, abstracts and session notes in a managed repository instead of allowing useful material to disappear with event accounts.

A common data workflow can then connect the intellectual and practical sides of a project. Before work starts, teams can publish a question, data map and governance note. During the meeting, shared notes can connect each presentation to specific datasets, methods or unresolved issues. Afterwards, contributors can deposit materials with licences, version information and links to the relevant outputs. The same workflow works for a hybrid workshop, a project consortium or a national data initiative.

Success appears in concrete behaviour: fewer duplicate collections, faster resolution of inconsistent fields, earlier identification of consent limits and more projects able to reuse existing evidence. The Kansas programme illustrates why this matters across borders. When research communities in Lawrence and Australia treat data description, stewardship and communication as shared responsibilities, information travels with enough context to support new questions. The lasting achievement is a common scholarly infrastructure built around transparent evidence rather than a collection of isolated files.