Precision breeding continues to push the boundaries of cultivar development

Drones collect detailed data about plant physiology and local conditions, which helps breeders make faster, more accurate breeding decisions.

The effects of artificial intelligence (AI) are felt far and wide in the world around us. From texting on our wristwatch to letting a robotic vacuum clean our home, AI is near integral in much of our daily lives and agriculture is no exception.

Over the last number of years, virtually all ag companies have found ways to incorporate AI and machine learning into almost everything, from weed detection to colour sorting, and lately, plant breeding.

Bayer is no different. AI is a cornerstone of its Precision Breeding program, which began in 2020 to help digitize the bulk of the phenotyping process — that is, the observation and detection of flowering, stability, lodging, resistance, pod shatter and uniformity. The goal is to develop hybrids faster, with greater precision and designer genes to boost yield, resist diseases and make harvest more predictable.

And it’s working. Over the last seven years, Bayer has shaved three years off the F1-to-commercialization process. By recently doubling the number of field-testing sites in Canada and the U.S., Bayer can test more material in more environments in a single season. Advanced drone technology, with increasingly powerful photo sensors that see well beyond the human eye, has turned phenotyping into a much more accurate, informed process.

SHRINKING TIME

When Britt Ousterhout, Bayer’s digital phenotyping lead, describes this work, it’s simple and complex all at once. “We try to turn pictures into useful information,” says Ousterhout, based in St. Louis, MI. That’s the simple part.

The complex part is that while technological advances have made the phenotyping process a degree easier each year, it hasn’t been as easy as flicking on a light switch. With the volume of digitization and software advances, this is an in-depth and complicated process in a space where everyone is learning on the fly. “This is not something that’s easy for people to do, because if it was, we would already be doing it,” she says.

GENERATING ACCURATE DATA, AT SCALE

One of the ways Bayer is supporting a rapid breeding system is through its expansion of test sites, which have doubled over the last few years across Canada and the U.S.

Accurate data at scale is the goal. In the days of boots-on-the-ground field assessments, person-to-person crop ratings inevitably differed, even within the same field — what one person sees as early flower, another may see as more of a mid-flower, for example. Multiply that by thousands of plots per year in a standard breeding program and data variation is guaranteed.

But with drones programmed to scrutinize everything at pre-determined levels, the guesswork is gone because phenotype data is standardized across all test sites. Data scientists then use AI and machine learning to take that data, plus the knowledge about how specific material performs in the field, to generate millions of data points that simulate performance in virtually any climatic zone. This allows breeders to further tailor hybrids to specific growing regions for farmers, which is especially critical for those who contend with early frost and days to maturity.

As Ousterhout explains: “We’re leveraging AI to combine what we know about our material from physical testing and simulate how it would perform on any grower’s field,” she says. “Our digital simulations allow us to ask, ‘What if?’ and then measure the answer.”

A typical test plot can generate gigabytes of simulated data. Not surprisingly, data rendering is often one of the most time-consuming aspects of precision breeding. Similarly, the upkeep of software and hardware is a constant effort.

GOOD DATA DRIVES BETTER OUTCOMES

In canola breeding, phenotyping occurs before yield testing and many a potential hybrid is tossed out at this stage, while the more promising ones move on to field trials. Using AI for high throughput phenotyping is transforming this selection process.

“There’s a ton of data that enables our scientists to confidently select the best plants prior to moving them out into field trials,” says Liz Simpson, hybrid product development lead for Bayer in Calgary AB. She says the days of sifting through thousands of individual crosses, hoping to find a winner, are long gone.

Today, UAVs augments scientists’ field visits to collect data, which is automatically recorded and sent to the cloud, where it is immediately processed and catalogued. “Now we can be a lot more purposeful with a better ability to optimize the germplasm that enters the pipeline,” says Simpson.

It means that by the time cultivars advance into field trials, they often come standard with “table stakes” agronomic traits, such as pod shatter, clubroot resistance, Verticillium wilt tolerance, uniformity and improved yield, that are best suited to each grower’s environment.

Because of the improvements in imaging and other technologies that can more accurately measure traits in the field, scientists now have the power to predict hybrid performance and advance the most promising ones into commercial products.

“The combination of these accelerated breeding methods, advances in genotyping and new phenotyping technologies are allowing us to validate and deploy hybrid products faster than ever before,” says Simpson.

Farmers are already starting to see the benefits, too. Simpson says that this year, Canadian farmers will see noticeable differences in DEKALB hybrids come harvest since all of them, from the 900 series forward, will automatically carry clubroot and straight cut traits.

“We are seeing the results of the latest enhancements we’ve made,” she says. “The DEKALB 900 series launched last year, is really just the beginning of a new era of high-performing canola hybrids from Bayer. The 800 and 400 series launching in 2024, are a testament to the wave of new hybrids that will really stand out from the competition in terms of their overall agronomic and yield performance.”

Farmers can expect to see a continual increase in these and other traits, and they can expect to see them faster, says Simpson.