Stands of corn and soybeans planted weeks apart make 2019 a tough management year for growers throughout the Corn Belt.
While farmers may have access to in-season data ranging from Normalized Difference Vegetation Index (NDVI) maps to aerial imagery from companies such as TerrAvion and AirScout, there aren’t many reliable services that can pull together the various piles of data and analyze them in time for farmers to act. That’s why relatively new companies like Intelinair work to pinpoint and automatically prioritize some of the worst problems growers face, such as weeds, poor emergence and low spots.
The 3-year-old Champaign, Ill.,-based startup has developed Ag-MRI — an interface that alerts farmers to problems in the field, pushing the most pressing and addressable problem fields to the top of a list in the cloud-based web app. The standard enrollment price is $5 an acre.
Enrollment covers both Ag-MRI and frequent flights that capture high-resolution imagery for even the most expansive fields in the company’s current registry of almost a million acres, most of which are in Illinois, Indiana and Iowa.
Chief agronomist Ivan Dozier says farmers don’t even need to look at the maps supporting an alert, though if they do, they’ll see row-by-row insights thanks to the 10-cm resolution captured by the plane pilots Intelinair contracts with throughout the Midwest. The company also works with drone pilots on spread-out research fields in Hawaii and elsewhere.
“If I’m a farmer and I wake up, I know I’ve got a couple sprayers and I want to know where to send those before the next rainy weather pattern comes in,” Dozier says, adding the app’s alerts are ranked by importance in several categories, or badges: Weed Watch, Heat Seeker, Row Tracer and Yield Risk.
“You could go scouting, or you could go to the Weed Watch badge in the app, find two fields have over 20 acres of weeds and are in the same general area, and push the coordinates to someone with a quick email, all from your iPad,” he says.
Fields with heavy weed pressure or low emergence can get their rank of importance lowered in Intelinair’s alerts algorithm if the pressure is centered in an area that’s too wet to get to. Stunted areas nearer roadsides are often ranked higher.
Alerts are guided by “manual” rules generated by Dozier and his team; an example of one of the simpler rules would be that any plants present before planting are weeds. With the Row Tracer badge, Intelinair differentiates between weeds and crops while providing row-by-row stand estimation. The Yield Risk badge is available for tracking crop health after canopy closure.
Instead of waiting for the end of the season to see how different hybrids are performing using MyJohnDeere and Climate FieldView, Dozier says farmers have the option to connect those services to Intelinair for a “differing hybrid” alert, as well as other alerts tied to planting date, seeding rate and hybrid data.
“The more data we have, the smarter our analytics can be, so we highly encourage farmers to share as much as they’re comfortable sharing,” Dozier says, noting the data is anonymized.
Intelinair schedules its pilot’s flying season regionally so the first bare-soil flight starts later if conditions haven’t been ideal for planting. During the early season, flights are weekly; they occur less frequently once most field management has concluded. Frequent flights are critical to drive timely management decisions, while also building a data cache to support future insights.
“A lot of other places that are collecting imagery don’t start as early as we do,” says Wyatt Dozier, an image analyst at Intelinair and Ivan Dozier’s brother. “We start earlier because we feel there are insights you can get right away, even if nothing has emerged yet. We can look at the patterns in the soil and detect things like broken tile lines.”
In addition to strategic flight timing, Ivan Dozier says a team of machine learning programmers are working to help them increase the accuracy of various alerts, because as it stands, there are times where weed alerts are triggered by cover crops or treelines in the manual rules. By comparing false positives from two methods, they will increase the accuracy of alerts.
“If we have an 80% confidence dataset for detecting weeds, and the machine learning team produces an 80% confidence dataset, in all likelihood, when we look at where they overlap, we can combine two confident datasets to create an even better 90% confidence,” Dozier says. He adds that boundary maps are another layer that can automatically be imported into Intelinair to avoid counting wildlife habitat as weeds.
“There are manual rules we haven’t attempted to do, like automatically detecting a waterway, for example, because it’s not as high a priority to drive decisions,” he says. “A waterway follows a pretty consistent pattern through NDVI and through topography, so if they come up with a rule that can identify waterways 90% of the time based on the data we’ve accumulated, we can plan to incorporate that into the product, as well as a number of other patterns.”