Advertisement

Machine learning may help get a leg up on blackleg

AI-backed research results may help growers better manage the risk of blackleg in canola crops

| 6 min read

By Jim Timlick

A stack of samples of canola stem discolouration due to blackleg infection. Photo: Michael Harding

Results of a recent study by researchers in Manitoba and Alberta could help growers fight blackleg disease in canola more effectively.

The study was done by a team of researchers at the University of Manitoba led by Dilantha Fernando, a professor in the school’s plant science department. They were joined by Michael Harding, a plant pathologist with Alberta Agriculture and Irrigation.

Conducted during the 2021 and 2022 growing seasons, it involved about 60 farms in Alberta each season. Results were released last December and published in the international peer-reviewed Canadian Journal of Plant Pathology.

The study is one of the first in Canada to use artificial intelligence to analyze results and make recommendations.

Fernando says the objective was to use machine learning to more precisely establish contributing risk factors that determine disease outcomes, so growers can better understand blackleg and take a more proactive approach to fighting it.

Data was collected by Harding and his staff. It was fed into machine learning models created by postdoctoral researcher Liang Zhou, and reviewed contributing factors including weather, flea beetles, root maggots, crop rotations and varieties. The risk posed by each of those was assessed.

High-risk

The study found about 66 per cent of the risk of blackleg in canola is determined by weather and rotation, as well as the variety grown and the resistance gene present in that variety.

Put another way, “these models can predict blackleg disease risk accurately 66 per cent of the time when the only information they are fed is crop rotation and weather,” says Harding.

“If you feed the models crop rotation data and weather data, they can predict blackleg risk accurately around two-thirds of the time. There’s no other blackleg risk variable that can get you close to that.

“There’s nothing we can do about the weather, so it really comes down to crop rotation and variety selection and putting the right variety in the right field to manage the risk. That’s going to get you the biggest bang for your buck as far as mitigating blackleg risk.”

Harding says stretching rotations and planting an assortment of resistant varieties is advice he and many colleagues have shared with farmers for years.

He hopes the data offered by the study will solidify that advice and convince more growers to follow it.

“For me, the key take-home message from this is we’ve got a peer-reviewed publication that’s backstopping my extension message that you need to rotate in time and space, both with crops and with different R-genes,” he says.

“If you do that and the weather isn’t pushing the disease hard, you’re going to reduce two-thirds of your blackleg risk by just practicing good crop rotation and resistance gene rotation.

“That’s a huge amount of risk that’s covered off by just that simple practice that only requires planning on the part of the grower and a willingness to sometimes go away from the host crop.

“If you build your blackleg management program with the foundation of crop and R-gene resistance rotation, you’re going to be on a really sturdy foundation to start adding in some of these other management tools like controlling flea beetles and using seed treatments or even early-season fungicides.”

Refining recommendations

“The more we can refine the value of that recommendation, the easier it is, I think, for that message to get across,” he adds.

“If I say to a grower you should rotate away from canola and that recommendation is only going to reduce the risk by five per cent, they’ll think ‘I’ll take my chances.’ But if it’s going to reduce that risk by 60 per cent, well, now they’re listening.”

Fernando praised Zhou’s contributions to the machine learning models that were an instrumental part of the study.

He says many researchers struggle with use of AI due to lack of available real-world data. That wasn’t a concern in this case because Harding provided reams of detailed information from dozens of working farms.

“I think the bottom line is the approach becomes easier with real data and real people who have collected that,” Fernando says.

“What we had with Mike was real data, real farm data. We didn’t go and (do) an experiment. We had the real data. All we did was plug that into a machine learning model and see what will be the output or the outcome.”

Harding and Fernando both say they expect AI will be used more and more as part of agricultural research. Harding says it helps scientists separate the proverbial wheat from the chaff.

“As scientists, we can generate data but we have to separate out the noise from the real picture. There’s a lot of other things swirling around out there that are incidental or confounding our ability to understand cause and effect,” he explains.

“Machine learning is a more powerful way to take really complex sets of data that have multiple inputs and sort out the cause and effect from that noise. That’s why we went the machine learning route, because it was going to do a better job than some of the classical statistical analysis we could have done.”

As for other factors, in their report the researchers note that the study didn’t take in factors such as variety resistance and pathogen population, which they say “could potentially impact the model’s accuracy.”

The different canola cultivars’ resistance levels relative to each other in different fields couldn’t be measured, as the pathogen populations may vary by field, they said.

In the future, though, pathogen genes could be monitored over crop districts in each province, which “may be more useful for defining cultivar resistance levels based on R-gene stacking.”

Implications for Prairies

Data used in the study was collected in Alberta, but Fernando and Harding believe it could have implications for the entire Prairie region.

“I feel pretty safe in saying you would get similar results in other provinces,” Harding says. “We’re growing pretty similar varieties, we have pretty similar weather patterns, our production methods are pretty comparable across the three Prairie provinces. But until someone does the work, that’s speculation.”

Blackleg was first detected on the Prairies in the 1980s and was initially kept in check thanks to genetics and varieties developed at the time. A noticeable increase in incidence has been reported in recent years as the pathogen becomes more virulent and some R-genes are less effective.

Fernando says that’s why projects like this one are so important in helping growers better manage disease risk.

Harding says he hopes study results will reinforce messaging on the tools available to fight blackleg and the importance of staying ahead of it.

“Hopefully the really important information that comes out of this is if we don’t stay ahead of blackleg, it’s going to take a bite out of somebody that’s not utilizing these tools. If it gets people’s attention and helps them to understand how big of a return on investment this could have in the long term, that I think that’s a win for us.”

Fernando agrees.

“Market price is determined in part by how much there is disease, how much is there toxins and all that. If we are to be ahead in the world as leaders in the export market, any disease that we can mitigate is to our benefit.

“It’s millions or billions of dollars, and that’s where we gain the opportunity, is by having a concrete message and that message is coming with real world data.”

Jim Timlick is a reporter for Grainews magazine. His article appeared in the April 16, 2024 issue.

Click a topic to discover more articles and insights