IN SEARCH OF THE EASY BUTTON
BY IAN DOIG • ILLUSTRATION BY SHUTTERSTOCK
The next great leap in digital agriculture, the adoption of intelligent technologies is a complex work in progress. Artificial intelligence (AI) and machine learning (ML) are engineered to produce agronomic insights farmers can act upon from a vast flow of information that includes precision ag field data, weather, soil and yield data as well as drone and satellite imagery.
“New ML algorithms and building AI solutions will be the next big thing in the industry,” said Alex Melnitchouck, Olds College Smart Farm chief technology officer, digital ag. He uses GPS technology as an analogy: to locate yourself on the planet, the system must connect with multiple satellites. Similarly, to tell the farmer where they are agronomically, the Smart Farm’s HyperLayer Data Concept project aims to cross-reference many layers of ag data. “Can we model soil properties and predict what to expect in the soil without going to the field? That’s a fundamental question with agronomy and something we’re trying to develop,” said Melnitchouck.
Ten college employees have contributed to the internally funded project. Partners include several large companies, agronomic consulting firms and farmers. A central component is the employment of AI to create a digital twin of the Smart Farm’s physical cropland. This growing data set has been described as exquisite by one partner research organization, much to Melnitchouck’s satisfaction. It has been used as the basis for highly accurate fertilizer rate plans on the Smart Farm’s fields. The HyperLayer project can also be used as a means to test new technologies. For instance, a business with a newly developed algorithm for soil analysis can test its process on those same Smart Farm acres. The results can then be compared to the facility’s own. “We can conclude if that technology works well or needs improvement,” said Melnitchouck.
Farmers and agronomists have long relied on multiple software programs to inform their decision-making, said Melnitchouck. This has been necessary given the 140-plus factors that affect crop development, most of which are unpredictable or difficult to model. Add to this the many types of field equipment and crop varieties in play, and completely accurate predictions are impossible. “I tell everybody agriculture is more complex and difficult than rocket science,” said Melnitchouck. “It’s not because I want to humiliate rocket scientists, it’s because agriculture is so unpredictable and relies on so many factors.”
From this profusion, a modest investment in human labour and commercially available software programs can produce great results, he explained. By this principle, the HyperLayer project strives to create next-generation digital ag systems that can be practically implemented on farms. “It’s one of the main laws of computer science that you can easily operate the system without having any idea how the computer works. That concept applies to utilizing data in agriculture. We’re trying to streamline and automate that process—to build an easy button.”
Terry Aberhart farms wheat, barley and canola near Langenburg, SK, and operates Sure Growth Solutions, an agronomy consulting company. An enthusiastic fan of precision ag and eager adopter of new software technology and processes, he offered to supply HyperLayer field data to ground truth its methodology.
“A lot of what we do in agriculture still comes down to gut feeling. The more we utilize data to make the decisions, the better off we are.” As an example, he said the apparent dry conditions on his farm this spring belied plenty of ground moisture detected by soil probes. Whereas one might have assumed it was game over, timely rain would yet save the day.
One of the HyperLayer partners is xarvio Digital Farm Solutions, a subdivision of BASF. A marketer of digital tools such as the SCOUTING and FIELD MANAGER apps, intelligent technologies are central to xarvio’s ambitions. As the global company’s head of technology and data, Jeff Spencer leads the development of what it has dubbed the “agronomic decision engine” at the core of its digital services.
The term “decision support systems” has been used in digital ag in the past, he said, and it remains a good description of intelligent technology platforms. “I hope we’ve left a phase of ag where it’s information overload without prescribing what to do with that information,” he said. “How can I take that information overload to now turn it into a clear and tangible decision that can be made on a farm and in a field?”
As the project partners puzzle this out, they will create a dense “particle or parcel” of data for each field being studied over multiple years starting with 2020. “We’re convinced we’ll find numerous insights that have escaped us in the past, because we haven’t had the depth of concentrated data,” said Spencer. The challenge is to make this process cost effective and package it in a single tool à la Melnitchouck’s easy button. “Olds College is providing this platform, and we need many partners to be successful in this intelligent technology space.”
As with the Olds College HyperLayer project, Manitoba’s Enterprise Machine Intelligence and Learning Initiative (EMILI) relies on teamwork to push the implementation of intelligent tech forward in the ag sector. Founded in 2016, the non-profit group’s funders and board of directors include corporations, multiple levels of government and academic institutions. EMILI and its eight staff members take an “ecosystem approach” with support of research, development and integration as well as skills training, talent development and the capital required to make it all happen. EMILI aims to support the launch and scaling of businesses employing this technology in Manitoba and across the Prairies. To do so, it has already raised more than $500 million.
Board chair Ray Bouchard believes intelligent technology is akin to historical hinge points in farming such as the invention of the plow and the first tractor. “It’s transformative and a massive opportunity,” he said.
Bouchard estimates 90 per cent of farmers collect field data. “But probably less than 10 per cent of growers collect good data, clean it and drill into it to gain insights that lead to productivity, efficiencies or profitability improvement. To increase production in a more sustainable way, we’re really just on the front edge of that curve.” He cited the sensor-driven selective spraying of weeds, which may potentially be 90 per cent more efficient than blanket application.
As software tools are created to extract layers of data, predictive modelling tools are also being built. The “easy button” concept will only be realized with incremental advancements throughout the entire ag value chain, said Bouchard. “We will be able to gravitate to an easier button, or one or two buttons versus 25 buttons.”
In April of 2019, EMILI and the University of Winnipeg set out to create an image database of Prairie crops and weeds. Overseen by computer science professor Chris Henry and physics professor Chris Bidinosti, EMILI and U of W, the three-year project will collect and label plant images. The project was inspired by ImageNet, a massive public database of described images from hyenas to helicopters that acts as a database reference-style library for intelligent technology tools.
“There was no publicly available data set of Prairie crops sufficiently large and accurately labelled to fuel innovation across Canada,” said Jacqueline Keena, EMILI managing director. “With this labelled data set it will be possible to train algorithms and deploy drones and other implements in the field that are able to make smart decisions in a fast manner right in the field.”
This is how the value of intelligent technology in farming will be realized, said Melnitchouck. “If you can complete that full circle of data analysis, from data collection to accurate automated decisions, that’s the Holy Grail of digital agriculture.” Like Bouchard, he predicts the evolution of these systems will remain incremental. “I don’t know when we’ll see the actual easy button for agriculture. We just try to find an optimal solution. And when it comes to our data layers, we’re in the process of constant evolution. It’s a long process.”