Data-rich, innovation-poor

The Value of Data in Agriculture

Data, data, data... the world urges us to collect and store it properly, which is a rather tedious task, but why is it so important?

Data is proving to be just as transformative in agriculture as in other sectors that embraced the digital revolution earlier. Even a domain deeply rooted in natural cycles and ancestral wisdom is now forced to exploit data to evolve. Today, data has become essential for improving agricultural practices' efficiency, sustainability, and productivity.

Although humans have long interpreted past experiences to make decisions, it wasn't until the second half of the 19th century that a data-driven approach to agriculture was formally born. Gregor Mendel, a monk passionate about biology and mathematics, precisely recorded observations of his garden plants. Those seemingly simple notes laid the foundation for genetic science and exemplify how past data can be used to predict future outcomes (Gliboff, 1999).

Indeed, the primary utility of data lies in its ability to anticipate future phenomena. In ancient times, the Romans might have consulted an oracle, who, after watching the flight patterns of birds, would hazard a guess about when rain might arrive (Beard et al., 1998). Today, we consult data scientists, whose predictions, while still imperfect (the accuracy depends on what we want to predict), are based on far more rigorous foundations.

When the Past makes us look into the Future

Every day, we are bombarded with enormous amounts of data, most of which is irrelevant, but a small, critical portion is vital for informed decision-making. Consider a seemingly banal but familiar scenario: choosing a restaurant to impress your partner.

Your brain, functioning like a real-time processor, begins compiling relevant saved data: your partner's culinary preferences, aesthetic tastes, preferred locations, etc. After internal calculations, it was recommended that "Thai Kai," a contemporary New Zealand fusion restaurant, be recommended. Problem solved?

Perhaps. But the reliability of this "decision" must be supported by three fundamental questions:

Is the quantity of data sufficient for a reliable analysis? (Do we have enough information to choose a restaurant that satisfies our partner's preferences?)
Does the data accurately reflect reality? (Are the insights we have about our partner's tastes correct and up to date?)
Is the data free from external biases? (Did we choose "Thai Kai" because it's genuinely the best option for our partner or simply because we want to try it?)

The same principles apply when deciding how much nitrogen to apply to a maize field. A datum is a recorded fact from the Past, but using it correctly requires a valid process for collection and interpretation.

Practical Applications of Data in Agriculture

Farm management is a continuous exercise in decision-making aimed at maximizing economic returns, minimizing environmental impact, and optimizing time. Decision Support Systems (DSS) are increasingly prevalent and effective tools in this context. Whether it's fertilization, pest control, or long-term strategy, predictive models are becoming indispensable.

These models must not replace the farmer; they assist. "Assistance" is a key term in Agriculture 4.0.

Just as a computer processes data, so does the farmer, whose local knowledge and intuition remain crucial. A well-developed DSS serves as a sort of "oracle", offering scientifically grounded predictions based on large datasets. However, one needs high-quality, representative data to build and validate such systems.

The Challenges of Agricultural Data

Despite the promise, significant challenges slow down the full exploitation of data in agriculture, unlike in other sectors:

1. Agricultural Data Is Expensive

Collecting agronomic data requires considerable time and financial resources. For instance, to model the lifecycle of a harmful insect, one must install traps across multiple fields, repeatedly visit those sites, and manually count captured insects. Capturing other insect life stages, like eggs, demands further expertise and effort. And once data is collected and used to create a model, it probably results in an unsatisfactory DSS: a single season and one location rarely capture the necessary variability to train a robust model. Natural phenomena are known for their complexity and variability (Jones et al., 2017)

2. Agricultural Data Is Difficult to Sample

Even with years of effort and investment, data quality can drop if collection protocols aren't followed precisely. For example, assistance from a well-meaning but untrained friend may help us get useful data needed for model optimization, but his samples will probably be compromised due to sampling errors.

3. Local Data Limits Model Transferability

Suppose, after years of fieldwork and methodological refinement, our model finally performs well. A foreign farm becomes interested in adopting it, but we discovered that the model does not work properly in that country. This often happens when models are trained exclusively on local data, limiting their generalizability to different agroclimatic zones.

Unlocking Innovation Through Data Sharing

Thousands of researchers across Europe spend their days collecting data on soils, crops, and plant diseases. However, due to bureaucratic constraints or competitive concerns, most of this data remains inaccessible, locked away on university or private servers.

This is understandable. Gathering such data is labor-intensive and expensive, so researchers are understandably protective. But this jealousy over (in many cases, publicly funded) data slows progress.

Speaking about the Agri-tech sector, my brief time in academia made me realize how rarely these data (and the tools that accompany them) reach farmers. This gap motivated my transition to the private sector, where I now develop agronomic models aimed at real-world applications.

However, many startups and smaller companies lack the capacity for comprehensive data collection campaigns, which only the largest firms can afford. As a result, much of the agricultural sector suffers from a chronic data shortage that prevents the adoption of innovative solutions.

Universities often develop highly effective models, pure research at its best, but they're rarely implementable on the farm. It is up to private enterprises to turn these models into accessible, farmer-friendly tools.

Every time I build a model, I think about the terabytes of valuable unused data. I imagine a future in which academia and industry can collaborate more, sharing data to accelerate agricultural innovation, optimize resource use, and create real benefits for both the environment and the farming community.

How can we make this vision a reality? Honestly, I don't have a definitive answer. My intent with this article is simply to draw attention to a crucial opportunity, one that could define the next phase of growth for the Agri-tech sector, as suggested by the 2020 European Commission report about community data sharing, which says:

"A dataset may become particularly valuable when it is (re)used together with other datasets, providing insights for decision-making processes or the development of services. Data is unlike any other resource. The non-rivalrous and non-excludable nature of data is the fundamental economic driver of socioeconomic welfare gains in data-sharing operations: many parties can use and reuse the same dataset for a variety of purposes and an unlimited number of times, without any loss in its quality or quantity" (EU Commission, 2020)

If you want to explore more about this topic and the possible path through effective solutions, the report by the EU Commission proposes a comprehensive strategy to exploit the full societal value of data through responsible, scalable, and transparent data sharing, enabled by coordinated policy actions, legal certainty, technical infrastructure, and dedicated governance frameworks across the EU.

References

Gliboff, S. (1999). Gregor Mendel and the Laws of Evolution. History of Science, 37(2), 217–235. https://doi.org/10.1177/007327539903700204
Beard, M., North, J., & Price, S. (1998). Religions of Rome: Volume 1, A History. Cambridge University Press.
Jones, J. W., Antle, J. M., Basso, B., Boote, K. J., Conant, R. T., Foster, I., ... & Rosenzweig, C. (2017). Toward a new generation of agricultural system data, models, and knowledge products: State of agricultural systems science. Agricultural Systems, 155, 269–288.
https://doi.org/10.1016/j.agsy.2016.09.021
European Commission: Directorate-General for Communications Networks, Content and Technology, Towards a European strategy on business-to-government data sharing for the public interest – Final report prepared by the High-Level Expert Group on Business-to-Government Data Sharing, Publications Office, 2020, https://data.europa.eu/doi/10.2759/731415