Ethical frontiers: data stewardship in global food security research

Global food security depends on information as much as it depends on rainfall and fertile soil. Researchers analysing crop yields, climate patterns, and supply chains generate terabytes of data every harvest season, and the choices they make about collecting, storing, and sharing that information carry real consequences for farming communities. At conferences like IASSIST 2017, held in Lawrence, Kansas, scholars have gathered to discuss the common language of research, and the ethical questions surrounding data use remain among the most pressing concerns of our time.

In Australia, where vast pastoral stations stretch across the Outback and boutique vineyards line the coast of South Australia, food security research intersects with unique legal and cultural considerations. The country's Privacy Act 1988, its commitment to Indigenous data sovereignty, and its reliance on institutions such as CSIRO shape a distinct environment for ethical inquiry. Understanding how data ethics operates in this context offers lessons that resonate far beyond the Southern Hemisphere.

Foundations of ethical data stewardship in agricultural studies

Ethical data stewardship begins with the recognition that numbers represent lives. When agronomists in Wagga Wagga collect soil moisture data, when economists in Perth track grain exports, or when epidemiologists in Brisbane map the spread of crop diseases, they are handling information that can influence policy, prices, and the wellbeing of farmers. The principle of beneficence, borrowed from medical ethics, requires researchers to maximise benefit and minimise harm. In food security research, this translates into questions about who benefits from predictive models and who bears the costs of algorithmic errors.

The Australian context adds layers of complexity. Researchers working with Aboriginal and Torres Strait Islander communities must navigate principles of Indigenous data sovereignty, which assert that communities hold rights over data pertaining to their lands, waters, and traditional knowledge. The Maiam nayri Wingara Indigenous Data Sovereignty Collective and the Australian Indigenous Governance Institute have articulated frameworks that challenge Western assumptions about open access. These frameworks ask researchers to move beyond extraction and toward genuine partnership.

Researchers also face practical questions about data quality and representativeness. A yield prediction model trained primarily on data from broadacre farms in the Murray-Darling Basin may perform poorly when applied to smallholder operations in Southeast Asia or to regenerative agriculture plots in Tasmania. Ethical stewardship demands transparency about the limitations of datasets and humility about the generalisability of findings.

Consent, privacy, and smallholder farmer data

Smallholder farmers produce roughly a third of the world's food, yet their data is often captured by external researchers and corporations with limited feedback to the communities themselves. Ethical concerns arise when consent is obtained through processes that farmers cannot fully understand due to language barriers, literacy gaps, or time pressures during planting seasons. In Australia, the Privacy Act 1988 provides protections for personal information, but agricultural data collection often occurs through aggregators, technology providers, and research partnerships that fall outside traditional consent frameworks.

The concept of dynamic consent offers one solution. Rather than treating consent as a one-time signature, dynamic consent allows participants to modify their preferences over time, often through digital interfaces. Such approaches recognise that farmers may wish to withdraw data when they discover how it is being used, or to grant permission only for specific research questions. Implementing these systems requires investment in user-friendly design and ongoing engagement, costs that many research institutions struggle to absorb.

Privacy in agricultural contexts also involves questions about de-identification. A farmer's yield data, when combined with satellite imagery and weather records, can often be re-identified with surprising ease. Researchers must therefore consider whether data has been stripped of obvious identifiers, and whether the combination of variables could lead to identification of individuals or properties. The Australian Bureau of Statistics has developed methods for handling such risks, drawing on techniques used for census data and applying them to agricultural surveys.

Algorithmic bias in yield prediction and climate modelling

Algorithms have become indispensable tools for forecasting harvests, predicting pest outbreaks, and modelling the impacts of climate change on food systems. Yet these algorithms inherit the biases present in their training data. If historical yield records over-represent industrial monocultures, the resulting models may systematically undervalue the contributions of polyculture systems and small-scale producers. In Australia, where the wheat belt stretches from Western Australia through South Australia and into western Victoria, such biases could distort recommendations that affect thousands of operations. Tools that rely on programming frameworks for data analysis, including those showcased at the csharpcon conference, must be developed with awareness of these structural issues.

Climate modelling presents similar challenges. Models trained on temperature and precipitation patterns from the past century may fail to capture the increasing frequency of compound extreme events, such as the simultaneous droughts and heatwaves that affected Australian agriculture in the early 2000s and again in 2019. Researchers have a responsibility to communicate these uncertainties clearly, particularly when their findings inform government policy or commodity market decisions.

The black-box nature of some machine learning systems compounds the problem. When a neural network predicts that a particular region will experience yield declines, stakeholders deserve an explanation that they can interrogate. Explainable AI methods, including attention mechanisms and surrogate models, offer partial solutions, but they require expertise that not all research teams possess. Building capacity for ethical algorithmic auditing represents an investment that funders and institutions must prioritise.

Framework	Primary focus	Key strengths	Limitations
FAIR Principles	Technical accessibility and reuse	Wide adoption, clear metrics	Limited attention to justice and equity
CARE Principles	Indigenous data sovereignty and rights	Centres community authority, addresses historical harms	Less familiar to mainstream institutions
Privacy Act 1988 (Australia)	Personal information protection	Legal enforceability, established case law	Designed for commercial contexts, gaps for aggregated data
Dynamic consent models	Participant control over time	Respects autonomy, adapts to changing preferences	Resource-intensive, requires technical infrastructure
Data cooperatives	Collective bargaining by data contributors	Empowers producers, balances commercial interests	Coordination challenges, legal complexity

Open data versus commercial interests in Australian agriculture

The open data movement has transformed many fields of research, and food security stands to benefit from greater sharing of datasets, methodologies, and findings. Commercial interests complicate the picture, however. Agribusiness companies collect vast amounts of data through precision agriculture tools, and much of this data remains proprietary. Researchers who wish to study the environmental impacts of certain farming practices may find themselves locked out, forced to rely on incomplete public datasets or to negotiate costly access agreements.

In Australia, the National Farmers' Federation and various commodity organisations have argued that farm data should be treated as a commercial asset, with farmers retaining control over how it is used. The Australian Farm Institute has explored models for data cooperatives, in which farmers pool their information and collectively bargain with technology providers. Such models attempt to balance openness with the legitimate interests of data contributors.

Government agencies have taken different approaches. The Australian Bureau of Agricultural and Resource Economics and Sciences publishes aggregated data on production, trade, and climate, providing a foundation for public research. CSIRO has open-sourced datasets related to soil moisture, crop monitoring, and climate projections. These resources demonstrate that public investment in data infrastructure can support both academic inquiry and policy development, even amid commercial pressures to restrict access.

Indigenous knowledge and sovereign data governance

Indigenous communities across Australia hold sophisticated knowledge systems about land management, weather patterns, and food production, knowledge developed over tens of thousands of years. This knowledge appears in oral traditions, seasonal calendars, and agricultural practices that differ markedly from Western approaches. Ethical data use requires recognition that such knowledge is not simply data to be extracted, but a living body of wisdom with its own protocols for sharing and use.

The CARE Principles for Indigenous Data Governance, formulated by the Global Indigenous Data Alliance, complement the more familiar FAIR Principles. CARE stands for Collective benefit, Authority to control, Responsibility, and Ethics. These principles assert that Indigenous peoples have rights to govern data about their communities and territories, and that researchers must engage with Indigenous governance structures throughout the research lifecycle.

In practice, this means that food security research involving Indigenous lands in places like the Northern Territory or parts of Western Australia requires partnership agreements, benefit-sharing arrangements, and ongoing consultation. It also means recognising that some knowledge may not be appropriate for publication in open-access journals or databases. Researchers who treat Indigenous knowledge as freely available risk perpetuating colonial patterns of extraction, regardless of their good intentions.

Building frameworks for responsible food security analytics

Responsible frameworks for food security analytics must address technical, legal, and cultural dimensions simultaneously. Technical standards for data documentation, such as the FAIR Principles mentioned earlier, provide a foundation, but they are insufficient on their own. Legal frameworks, including Australia's Privacy Act and emerging regulations around algorithmic transparency, offer important safeguards, yet they often lag behind the pace of technological change. Cultural frameworks, particularly those developed by Indigenous communities, remind researchers that data is always situated within relationships and histories.

International coordination adds another layer of complexity. Food systems span borders, and data collected in one country often informs decisions in another. Researchers who travel to conferences like IASSIST 2017, often by air travel arrangements, carry with them assumptions shaped by their home contexts. Cross-cultural dialogue can illuminate blind spots and foster more nuanced approaches to ethical data use.

The ethics of data use in food security research is not a destination but a journey. It requires ongoing reflection, willingness to revise practices, and genuine engagement with the communities whose lives are represented in datasets. As climate change intensifies pressures on food systems worldwide, the stakes of getting this right have never been higher. Researchers, funders, and policymakers must work together to build institutions and norms that honour both the power and the limitations of data.

Guiding principles for researchers working with food security data

• Recognise that data represents real people, lands, and livelihoods, not abstract variables.
• Engage with communities before, during, and after data collection, not only when convenient.
• Document the provenance, limitations, and potential biases of every dataset used.
• Distinguish between data that should be openly shared and knowledge that requires restricted access.
• Build skills in algorithmic auditing and explainable AI to interpret model outputs responsibly.

Practical steps for institutions supporting ethical research

• Invest in training programmes that cover both technical methods and cultural competencies.
• Establish review boards that include community representatives alongside academic peers.
• Fund long-term partnerships rather than short-term data extraction projects.
• Develop clear policies on data ownership, benefit-sharing, and dispute resolution.
• Support open data infrastructure while respecting commercial and cultural sensitivities.