The AI Telling Farmers When to Harvest

When apple harvest season began in Washington State last year, the fruit was ripe and farm workers were ready to pick it. But extreme heat quickly brought operations to a halt.

“It was like 38C… it’s not safe for people to work in that heat,” recalls Joel Carter of Okanagan Specialty Fruits. “We had to stop at 10 o’clock in the morning.”

Carter believes artificial intelligence (AI) could help farmers make better decisions about when to harvest. AI models that combine crop data with weather forecasts can estimate the ideal harvesting window and help growers plan around difficult conditions.

“You need to know more than just when your fruit is going to be ripe. How long do you have to pick it?” he says. “That’s where these models are really helpful.”

Okanagan Specialty Fruits operates more than 1,250 acres of apple orchards in Washington State. Its apples are primarily grown for sliced portions sold to customers such as hotels and schools. The company is investing in technology aimed at improving productivity—including apples that have been genetically engineered to resist browning after they are cut.

But knowing exactly when to harvest is far from simple.

New technologies are emerging that can count and analyse fruit growing on trees and vines while also predicting when crops will reach maturity. Such information can be particularly valuable for high-priced fruit such as strawberries and blueberries, where market prices can change rapidly.

A mistake can be costly. Farmers may hire seasonal workers too early, leaving them paying for labour they do not need, or harvest too late and miss the period when prices are highest.

Okanagan Specialty Fruits is already testing camera technology developed by Canadian company Vivid Machines. Mounted on tractors, the cameras capture detailed images of apple trees as they move through the orchards. AI then analyses the footage, identifying buds, flowers and fruit.

“Right now, Vivid is telling us crop estimates and harvest dates,” Carter says. He adds that the technology can identify extremely small flower buds that would be difficult for workers to spot with the naked eye.

However, the quality of an AI forecast depends heavily on the quality of the historical data used to train it.

“This isn’t something where an AI can scrape the internet and figure out what’s the average [yield] for Granny Smith,” Carter explains. “It’s going to be bespoke to your farm.”

Apples are relatively forgiving because some varieties have a harvest window of several weeks. Carter says Granny Smith apples, for example, can typically be harvested over about three weeks.

For berries, however, farmers may have only a few days.

“If a strawberry crop is on, you have to harvest it – otherwise your entire crop gets diseased very, very quickly,” says Raymond Martin, co-founder and chief operating officer of UK-based FruitCast.

FruitCast provides growers with forecasts for strawberries, raspberries, blackberries, blueberries and tomatoes. The company also plans to expand into grapes.

But can AI really outperform experienced farmers?

Martin says growers generally know when their crops are likely to ripen. The challenge is monitoring an entire operation, particularly when farms cover large areas or include both outdoor fields and indoor growing facilities.

“We do exactly what the farmers could do but we just do it on a scale that they can’t,” he says.

FruitCast’s system analyses images of ripening crops captured in several ways. Farmers can use drones, smartphones while walking through fields, or cameras mounted on agricultural vehicles.

Weather is another major factor.

This year has been particularly difficult for UK growers, Martin says, with high temperatures and severe drought affecting fruit farms. Stress from extreme conditions can push plants into thermal dormancy, slowing fruit development and making harvest timing even harder to predict.

FruitCast says its forecasting model incorporates weather and irrigation conditions. According to the company, its forecasts are within 10% of the actual volume harvested one week in advance and within 17% three weeks ahead. It also guarantees an error rate of less than 20%.

For farmers working with crops that can deteriorate within days, even a modest improvement in forecasting could make the difference between a successful harvest and a costly loss.