Data literacy for farmers: lessons from IASSIST 2017

Farmers rarely encounter data as a neat dashboard. It arrives through weather applications, machinery telematics, soil tests, satellite imagery, market reports and advice from agronomists. The IASSIST 2017 sessions on big data, deep learning and digital agriculture offered a useful warning: more information does not automatically produce better decisions. Farmers need the ability to question, compare and apply evidence in the conditions of their own operation.

The archived program for “Data in the Middle: The Common Language of Research,” held at the University of Kansas from May 23 to 26, 2017, brought together researchers and practitioners working with large datasets, emerging machine-learning methods and global food systems. Its central theme was especially relevant to agriculture: information has little value unless the people using it share a common understanding of what the data means, where it came from and how uncertain it is.

Data literacy in this setting combines information literacy, numeracy and critical thinking. It means knowing how a dataset was created, whether it represents the intended situation, how missing values were handled and what a model does with imperfect inputs. It also means recognising when a prediction is a useful signal rather than a dependable instruction. The aim is not to turn every farmer into a statistician, but to give farm decisions a defensible evidence base.

Australian readers will encounter familiar complications in the details. A grain grower near Wagga Wagga may manage several enterprise and soil types, while a beef producer west of Toowoomba can face unreliable mobile coverage and sharply different pasture conditions. Across the Murray–Darling Basin, variable rainfall and water allocations make historical averages useful but incomplete. The conference material therefore works best as a framework for practical farm judgement, not as a substitute for local knowledge.

Conference theme	Central lesson	Practical check	Australian relevance
Big data	Volume creates options, not certainty	Are the records representative and complete?	Comparing rainfall stations, soil sensors and satellite maps
Deep learning	Predictions depend on training data	How were labels assigned and errors tested?	Identifying weeds, livestock or crop conditions
Digital agriculture	Technology must fit a working routine	Will the tool save time and clarify a decision?	Precision farming on mixed enterprises
Global food security	Production is only one part of food security	Who gains access, affordability and stability?	Balancing exports with domestic needs
Research communication	Shared language supports shared action	Are roles, limits and costs clear?	Working with agronomists, contractors and suppliers

Reading big data with a farmer’s scepticism

Big data is often described through its volume, speed or variety, but scale can magnify poor assumptions as easily as it can reveal useful patterns. A national soil dataset may contain thousands of observations and still describe only the fields, crops and sampling methods that were included. The first task is therefore to examine provenance: who collected the record, what instrument produced it, how often it was checked and which observations were excluded.

Weather illustrates the problem well in Australia. Bureau of Meteorology records, local rainfall gauges, soil-moisture probes and satellite estimates may all appear on the same screen while describing different things. A satellite product can smooth over cloud or vegetation differences; an individual probe may reflect a hollow or unusually wet location. A precise number can still be misleading for a particular paddock. Data literacy encourages farmers to run the numbers against visible ground truth rather than treating any digital layer as settled fact.

That scrutiny becomes more important when regional connectivity is uneven and some sensor or survey records contain gaps. Missing data should not automatically be treated as a zero, a normal value or an absence of risk. Instead, users need to know whether the gap is random or concentrated in an area that already has limited services. The practical lesson from the conference was that larger datasets should expand the range of evidence available to a decision, while transparent methods and local knowledge determine whether that evidence is credible.

Making deep learning explainable on the ground

Deep learning received considerable attention because it can identify patterns in images, sensor readings and text that are difficult to specify through traditional rules. For agriculture, possible applications include detecting crop stress, recognising weeds, monitoring animal condition and combining several measurements into an early-warning signal. The sessions also demonstrated why impressive demonstrations should be treated as a starting point rather than a finished product.

A model trained on one season, location or management system may perform poorly elsewhere. Cattle, pasture species and market conditions differ across regions; the same weed can appear under different lighting; and commercial paddocks do not resemble a clean experimental trial in every respect. Labels may also contain human assumptions. If photographs were tagged by people who routinely confuse a disease with a nutrient shortage, the model will reproduce that uncertainty at scale. Farmers therefore need information about the test conditions, error types and confidence attached to each prediction.

Explainability must also fit the decision being made. A grower deciding whether to inspect a paddock may find a marked image useful even if it is not precise enough to trigger a spray. A contractor choosing a treatment rate needs stronger evidence, while a producer checking pasture condition may need to see the trend across several dates. The lasting lesson was not that complex models should be avoided, but that their outputs should remain connected to people who can challenge them, add context and test the consequence of acting on a prediction.

Practical habits for stronger farm decisions

The conference’s practical value lies in turning sophisticated ideas into repeatable habits. Digital agriculture succeeds when information fits the sequence of work already performed: observe, compare, decide, act and review. It also succeeds when people can explain the result without needing specialist language or excessive time away from the farm.

Useful questions to ask before acting include:

• Where did the data come from, and who collected it?
• What population, field or period does it actually describe?
• Which values are measured, estimated or missing?
• What assumptions sit behind the model or forecast?
• Would the result hold in a dry year, a wet year or a different enterprise?

Habits worth building over time include:

• Start with one decision, such as nitrogen rate or grazing days, rather than buying a broad platform.
• Check outputs against field records, stock books, local trials and advice from an agronomist.
• Record corrections, including wrong predictions, so performance can be reviewed honestly.
• Use plain language when sharing results with family, staff, contractors and suppliers.
• Review data access, ongoing costs and ownership before committing to a digital service.

These habits create an audit trail for decisions that affect the whole business. They also make it easier to identify when a new tool adds value. A service that produces many maps but does not improve a fertiliser, grazing or irrigation decision may be interesting without being useful. Conversely, a simple spreadsheet linked to soil tests and rainfall records can be highly effective when its purpose is clear.

Connecting advisers, suppliers and farm businesses

Farm data is rarely generated or used by one person alone. Agronomists interpret field results, contractors operate machinery, accountants manage records, suppliers provide inputs and technology companies supply monitoring services. The IASSIST material reinforced the need for common definitions: two businesses may use the same term while recording different things. Shared language reduces duplicated work and makes it easier to compare evidence from research trials with results from commercial farms.

Australian farm markets add another layer. Many businesses remain family owned and sell through grain merchants, processors, cooperatives, exporters, stock agents or produce buyers rather than directly to the public. A producer near Wagga Wagga may share data with a grain handler to improve storage decisions, while a horticulture business may work with a processor around quality, timing and traceability. These relationships can create value, but they also raise questions about confidentiality, contractual access and who benefits from pooled information.

Plain communication is especially important in that setting. Farmers often say they want to “get a handle on” the information or “run the numbers” before making a commitment. Technology providers should respect that language instead of presenting complexity as sophistication. Contracts should identify what data is collected, how long it is retained, whether it can be combined with other records and what happens when the service ends. Interoperability matters too: information should move between machinery, accounting packages and farm records without repeated manual entry or forced changes to established routines.

Global food security through an Australian lens

The global food security sessions broadened the discussion beyond yield and production. Food security also concerns access, affordability, stability and nutrition, so agricultural data must be interpreted alongside economic and social conditions. A production system can perform well while some households still lack affordable, reliable food. Conversely, a farm may have modest productivity but remain central to a regional community because it provides stable employment, supports local suppliers and reduces reliance on distant supply chains.

For Australia, this wider view is relevant to both export performance and domestic resilience. Drought, water availability and climate variability influence production, while input costs and exchange rates affect the value reaching growers. Digital tools can improve forecasting and resource use, but they can also widen differences if smaller operations face high subscription costs, poor connectivity or limited technical support. Evaluating those effects requires the same data literacy applied to any other agricultural innovation: clear measures, understood assumptions and attention to who receives the benefit.

The most transferable lesson from IASSIST 2017 was that trustworthy farm decisions depend on a shared language between research and practice. Farmers do not need more data for its own sake; they need evidence they can interpret, challenge and connect to the realities of soil, weather, markets and daily work. Data literacy supplies that bridge, turning fragmented information into knowledge that can guide action.