Brought to you by Source.ag
Smart tomato growing

Greenhouse tomato production is a significant part of the Canadian agricultural landscape, contributing substantially to both domestic sales and the export market. In 2022, Canadian growers produced 293,974 metric tons of greenhouse tomatoes, accounting for 39.1 per cent of total greenhouse vegetable production by volume, according to Canada’s Statistical Overview of the Canadian Greenhouse and Mushroom Industry report for 2022. This makes tomatoes the leading greenhouse vegetable in terms of both production volume and value. A majority are exported, primarily to the United States, highlighting the importance of this sector to international trade. 

Photo courtesy of Source.ag.

Despite the crop’s importance, accurately forecasting tomato yield remains a significant challenge for growers. This unpredictability, largely due to fluctuating climate conditions and the difficulty in assessing fruit development at early stages, can lead to inefficiencies, market instability and price volatility, impacting the entire supply chain. 

“A hard part about growing tomatoes is that growers have many decisions that impact a very long horizon,” explains Ernst van Bruggen, CPO and co-founder of Source.ag, a company with headquarters in the Netherlands and a regional office in the U.S., building AI-driven software solutions for greenhouse growers. “There are technical and strategic decisions that are more about plant management, like stem density and how that is matched to available light over the course of the season. Those decisions have a lasting impact, with additional smaller adjustments being made through climate, like artificial lighting or CO2, to balance out the crop.”

Growers rely on yield forecasting to optimize resource allocation, improve crop management and maximize sales and profitability. These predictions help sales representatives on the growers’ side set pricing strategies and confidently negotiate contracts, capitalizing on peak market demands and avoiding losses during surplus times when prices drop. This also translates to proper planning for labour, transportation and packaging, and long-term strategy. 

These trends can be analyzed over multiple growing seasons, helping growers refine their practices and making informed decisions about their operations. But, van Bruggen says, accurately predicting yield is a major challenge. For example, using historical data from previous seasons to predict a future season is often unreliable, because growing conditions change year over year. Measuring new fruit set as an indicator for yield is another method that involves calculating new fruit set weekly by measuring a sample of plants. This method also relies heavily on the assumption that fruit development time and weight will be similar to previous seasons. 

Traditionally, van Bruggen says, growers have converged to a one-week yield forecast using plant measurements and fruit count. “That’s when the fruit starts colouring on the lower trusses and you can, for the first time, visually see how much you may be able to harvest. At one week, a harvest forecast becomes more reliable than weeks in advance, and over time, this has become the point where negotiation and contracting takes place,” he says.   

The problem, he says, is the continued unpredictability of many variables. “Climate is changing all the time, and the speed of development changes accordingly,” he says, noting a sample taken one week before harvest is usually small and not an accurate representation of the entire greenhouse. 

A one-week window also presents challenges in terms of supply. Growers often rely on averages to predict what they’ll be able to produce and what contracts they’ll be able to fulfill. A short window of time between yield predictions and harvest means less time to adjust to supply and demand challenges. “If you are able to forecast three- to four-weeks out, there’s more room in the market to absorb volume. If you have oversupply and you know this four weeks in advance, there’s more room for growers to manoeuvre and find different channels of distribution, or for retailers to launch new campaigns. This stabilizes the market for all players,” van Bruggen says. 

Predicting yield three to four weeks in advance would provide some market stability, he adds, but it becomes trickier. “Growers are relying on fruit count for short-term forecasts, but this methodology doesn’t work for forecasting more than one week out. The time between fruit set and harvest typically takes six to eight weeks. In between the fruit set and colouring stages, there is no visible information that can help determine where the fruit is in its development,” he says, noting that a tomato might be green for the majority of its growth, only developing colour toward the end. 

An innovative solution, driven by AI
To combat this challenge, van Bruggen and his colleagues at Source.ag recognized the potential of using artificial intelligence (AI) in combination with grower data to track fruits from early stages of development throughout the growing season to provide an accurate four-week yield forecast. Source Cultivate, a software application powered by AI, combines important data and simulates plant growth and development based on real-time measurements, so growers can access reliable harvest forecasts no matter the condition. 

Photo courtesy of Source.ag.

The software integrates with a grower’s existing infrastructure to gather real-time data on climate conditions and plant development, including information on fruit set, development stage, greenhouse temperature, radiation levels and other important parameters. 

“Source Cultivate simulates plant development and growth through an algorithm that needs three data sources,” van Bruggen explains. A grower would register the plant development in the Source Plant App on an iPad from a selection of plants that represent the entire greenhouse. The algorithm develops a digital twin of the plant, which van Bruggen says the platform can use to track the plant’s development. 

The second data source is realized climate data, whereby the greenhouse’s climate computer is connected to the platform called Source Track to aggregate relevant climate data that will impact the growth and development of tomatoes. And finally, the grower must input their crop plan, outlining goals and strategies for the season. 

“The algorithm applies the grower’s strategy every day, simulating when pruning will be done, or when stem density changes,” van Bruggen explains. “It gives growers the ability to optimize what will happen by changing a strategy, and simulating different outcomes,” he adds, noting each simulation takes minimal time, allowing for growers to run several different scenarios in just a few minutes. 

Source Cultivate offers a higher level of accuracy in yield prediction, ranging from 85 to 90 per cent up to eight weeks in advance, by combining plant physiology with artificial intelligence. The software provides growers with a comprehensive understanding of their crop’s development, allowing them to identify potential issues, optimize their strategies, and make informed decisions throughout the season. It also assists growers in managing the introduction of new seed varieties by simulating growth patterns and further streamlining approaches based on the specific characteristics of each new variety. 

“When you start a season, it’s difficult to know how to set up a strategy to steer the plant toward maximal results. Changing something in your operation can be a high-risk move, because it’s difficult to know how it will play out. The impact is not seen until the end of the season, when you can see if you’ve done better than the season before,” van Bruggen says. 

“This is a cost-effective way of trying different things, with the benefit of knowing what might happen without actually having to take the chance.”

For more information, visit Source.ag/cultivate. 

Created by Amplify by Annex, in partnership with Source.ag and Greenhouse Canada.