Field Notes

Orchard Math Has Entered Its Managing Era

A Yakima startup has filed a patent on telling apple growers when to thin their crop, a decision the science still cannot predict reliably.

Clay-diorama editorial illustration of a miniature orchard with fruit and measuring tools.

What changes in this design

A reader might otherwise think of a plant picture as just a picture and of other plant information as separate notes, rather than inputs used together to reach a next step.

Here, pictures of the crop and other information about it are used together to identify growth-related aspects, forecast target metrics, and generate an action.

A reader may find that worth knowing because the system connects observations about growth and forecasts to a generated action. The particular action is not stated.

Ten millimetres, and one spray decision

On a morning in May, in a Washington apple block, a grower walks the rows with a caliper and measures fruitlets. Around 10 to 12 millimetres is the stage where thinning sprays have traditionally worked best. Spray then and the tree sheds part of its crop, and what stays gets bigger and sells for more. Miss the window and you're paying a crew to pull fruit by hand, or selling small apples.

The Cornell researchers who formalised this practice put a number on getting it wrong: $10,000 to $20,000 per hectare in crop value between over- and under-thinning. They call it the single most important management strategy affecting annual profitability, and they say plainly that it is hard to control and predict. A trade article from 2010 ended with a line I keep coming back to: it would be very useful to be able to accurately assess thinning response.

Sixteen people in Yakima file a patent

Thirteen years later, that want ad got a filing. US20260231847A1, published in August 2026, is titled Systems and Methods for Automated Crop Load Management. The applicant is Automated Fruit Scouting Inc., which trades as FruitScout and works out of Yakima, Washington, and it claims priority to a provisional from February 2023. The described loop is simple: crop images plus crop data go in, growth-related aspects get identified, target metrics get forecast, an action comes out.

The lead named inventor is the company's chief executive, who came to fruit after nearly thirty years consulting on IT systems. Its pitch since its first outside investment has been the phone: same-day deployment, no hardware to buy. One caution: the public record here is thin, and the only copy I could retrieve came from a commercial patent database.

Where the counting falls apart

Cornell extension researchers tested two commercial vision systems across 23 orchards in four states between 2021 and 2023. The vehicle-mounted camera tracked trunk measurements but consistently underestimated them. The smartphone app counted flower clusters accurately in one block and significantly under in another. Its yield estimate held up to about 75 fruit per tree; above that, foliage hides fruit. Both it and an older app were accurate below 200 fruit per tree and overestimated above it. Commercial trees routinely carry far more than 200.

Washington trials the next season sharpened the picture. On a four-year-old high-density block, a multispectral scanner averaged 15.5% error on fruit count and 3.8% on size. On a twenty-year-old organic Honeycrisp block, the same system ran 51.9% early and was still 28.5% off at harvest. It works on young uniform plantings and degrades on the old blocks where thinning economics bite hardest.

A free model that uses no cameras

Cornell also gives away a carbohydrate balance model that produces the same kind of answer. It takes temperature, solar radiation and day length from a weather station in the block, plus two dates the grower types in, and returns a four-day carbohydrate balance mapped to explicit spray rules: above zero, raise the thinner rate 30%; below -80 grams a day, skip thinning entirely. It's public and free and it uses no imagery at all. Its caveats are stated honestly: it is only as good as the weather forecast. A camera system has to beat that, not merely work.

The spray at the end of the recommendation

So what does the action come down to? A spray. One of the newer thinning agents is a repurposed herbicide that blocks photosynthesis for a few days, starving the tree until it drops fruitlets. The EPA registered it for apples and pears in March 2025, two years after this application's priority date. Across fourteen years of testing it significantly reduced fruit set in 55% of trials and improved harvested size in 43%. It still beat the alternatives.

What the filing is really betting on

The company has raised $9.76 million across four rounds since 2021, all of them seed, with 16 employees as of mid-2026. A rival founded in 2022, building tractor-mounted cameras rather than a phone app, raised $22 million in one Series A in September 2025. Two and a half times the money, in a third of the time, on the opposite bet.

So read the filing for what it claims, which is not a better way to count apples. It claims the layer above the counting: pictures and data in, a recommended action out. That is the valuable position and the one the evidence supports least. The counting breaks down inside normal crop loads. A free public model already gives the same category of recommendation with no camera. The spray it ends in works well slightly more than half the time. I don't read that as cynical. I read it as a bet that the certainty will arrive, and that owning the route to it will pay.

Sources

Reporting for this piece drew on the published filing and the references below.

The record, briefly. US20260231847A1, “Systems and Methods for Automated Crop Load Management” The abstract describes using crop images and crop data to identify growth-related aspects, forecast target metrics, and generate an action.

More angles on this

The Leveling Bar Has a Very Firm Opinion About LevelAnother reading from the same A01B classification family, soil working in agriculture or forestry.
A Work Vehicle, Taking Its Job Quite PersonallyAnother reading from the same A01B classification family, soil working in agriculture or forestry.
CPC A01B: Soil working in agriculture or forestry; parts, details, or accessories of agricultural machines or implements, in generalThe official classification family this record was filed into, published in the USPTO Cooperative Patent Classification scheme.
CPC G06T: Image data processing or generation, in generalThe official classification family this record was filed into, published in the USPTO Cooperative Patent Classification scheme.
When a Vehicle Notices Something, Who Decides What the Moment Means?Another reading from the neighbouring G06V classification family, image or video recognition or understanding.

Read around this one:Public patent record USPTO published figure Patent issue-date index