In the vast, sun-baked fields of Western Australia, a quiet revolution is taking place, one that could reshape how farmers manage nitrogen fertiliser use. At the heart of this transformation is a study led by Jonathan Richetti from CSIRO in Floreat, Western Australia, published in the journal *Smart Agricultural Technology* (translated from the original title, *Inteligencia Artificial en la Agricultura*). Richetti and his team have taken a significant step towards creating a data ecosystem for AI in agriculture, focusing on optimising nitrogen decision-making in wheat production.
The research tackles a critical challenge in modern agriculture: how to deploy AI systems that are not only statistically robust but also practical and easy to implement on farms. “Designing an AI system in agriculture involves several key steps,” Richetti explains. “First, you need to collect the right data and surrogate information to determine feasibility and ensure the system aligns with existing knowledge.” This foundational work is crucial for building an operational AI model that farmers can trust and use effectively.
The study zeroes in on nitrogen (N) fertiliser management, a cornerstone of wheat production. Agronomic knowledge tells us that optimal N rates depend on a delicate balance between crop demand and soil nutrient supply. To identify the key variables for predicting the best N rates, Richetti and his team conducted on-farm strip trials over three wheat seasons (2019–2020). They collected a staggering 43 variables, including weather, soil, plant, and reflectance data.
Here’s where things get interesting. The researchers compared two variable selection methods: AIC stepwise regression with ANOVA and recursive feature elimination, and SHAP (Shapley Additive exPlanations) via random forest. The results were revealing. The random forest approach identified just five key variables: NDVI and NDRE ratios between farmer and rich strips, total soil carbon at depth, chlorophyll canopy content index, and available N at mid-season. In contrast, stepwise regression selected eight variables, including total N and carbon at various stages and photosynthetically active radiation.
The implications of this research are profound. The random forest method, with its simpler, more streamlined approach, offers a more reproducible system for N decision-making in dryland wheat production. “This study highlights the importance of efficient database design in agricultural AI systems,” Richetti notes. By reducing the number of variables needed, the random forest approach could make AI-driven nitrogen management more accessible and practical for farmers, ultimately leading to better crop yields and more sustainable farming practices.
So, what does this mean for the future of agriculture? The study suggests that AI systems in farming must be designed with both statistical performance and practical deployment in mind. “The system’s evaluation assesses not just statistical performance but also ease of deployment, including data collection and recommendation speed,” Richetti explains. This dual focus could pave the way for more widespread adoption of AI in agriculture, transforming how farmers manage resources and optimise yields.
As the world grapples with the challenges of climate change and food security, research like Richetti’s offers a glimmer of hope. By leveraging AI and data-driven approaches, farmers can make more informed decisions, leading to more efficient and sustainable agricultural practices. The study, published in *Smart Agricultural Technology*, is a testament to the power of interdisciplinary collaboration and the potential of AI to revolutionise the way we grow our food.
In the end, the research underscores the importance of designing AI systems that are not only technically sound but also practical and user-friendly. As Richetti and his team continue to refine their approach, the future of agriculture looks increasingly bright—and data-driven.