Irrigating by analyzing

Originally, Brian Lynch was shooting for the stars. Literally.

Lynch studied aerospace engineering with a focus on robotics throughout college, going as far as to obtain his Ph. D. in the discipline from Carleton University in Ottawa. But his and his wife’s desire to start a family led him to look to Earth for a career vs. a professorship in the field. 

“I was doing part-time teaching and my wife and I decided to try and have a baby, and school contracts are not very reliable,” Lynch says. “I then came across Vineland.” 

Lynch joined the Vineland Research and Innovation Centre, based in Ontario, in September 2018. Although he does not have a background in horticulture, he says it was a “perfect” place for him to land because it allowed him to use his expertise in robotics to solve problems. 

“Even though Vineland was outside of my primary interest in space exploration, I thought it looked like a great place to work because I get the kind of environment I want to be in at a university, but with more behind it,” Lynch says. “I’m also one of these guys that sees something shiny and I get interested in it, so any project I work on I tend to get really passionate about it.” Right now, the something shiny is smart irrigation with other projects forthcoming. 

“I still dream of sneaking in a project looking into greenhouses on Mars or something like that, but we’ll see if I can convince my superiors,” he adds with a laugh.

What a robot can do for you

Based on his prior work with robotics, Lynch says that it’s paramount to make sure a robot (or any technology) works not just in optimal conditions, but also in non-structured or unfriendly environments. In that way, a project in the greenhouse has similarities to one in space or down in a mine.

“It’s a dynamic environment,” he says. “Plants are always growing.” For smart irrigation, Lynch says most of the group’s research began “a couple of years” before he joined Vineland and that it had been carried to a point where he could take the baton and keep it going. The original idea of it was simple: make irrigation more efficient.

“Growers have so much information that sometimes they don’t even realize they are using it to decide if, or if not, to water plants on a given day,” he says, noting that specifically applies to herbs and flowers grown in pots vs. plants grown hydroponically or in a vertical farm. “I’ve seen growers take a chunk of soil out, squeeze it in their hand and, if enough water drips through, then they know it doesn’t need any water. Or they just pick up the plant to see if it’s heavy or not (to see if the plant needs water). It’s a judgement call.”

Lynch calls that process “subjective” — leaving it up to factors like the weather, the grower’s energy level on a given day, or any of the several factors that go into making the decision to water or not inexact. Oftentimes, this can lead to overwatering for fear of not giving a plant enough — which can lead to disease and pest issues.

Smart irrigation counters this by using sensors stuck right into the soil to determine when to water or not. With sensor data — specifically soil moisture, plus environmental data that growers may already be collecting in the greenhouse — the decision can be made.

“The method we ended up going with is more about teaching, so learning a grower’s pattern with artificial intelligence,” he says. “When I talk to growers, I tell them that the system isn’t meant to tell you when to water the plants. It’s not us saying ‘we’ve made this set of equations and it’s going to check the sensors and decide.’ It’s actually just matching the grower behavior — essentially seeing the patterns in the sensor data and matching the growers’ decisions.”

The idea, he adds, is to combine the two sources to make the recommendation more precise.

“Whatever a grower decides, they are making ‘yes’ or ‘no’ decisions and they get recorded,” Lynch says. “And the sensor data is applied at the same time. You’re basically identifying patterns in the sensor data that link up to the ‘yes’ or ‘no’ decisions a grower makes.”

Lynch says this is done once or twice through different crop cycles to build a map between what the grower says and what the sensor says. In time, he says, this will improve data quality and make life easier for the grower.

Applying the data and the next steps

According to Lynch, there’s a treasure trove of data already being stored by many greenhouses today. For the purposes of his projects, temperature and humidity are the two data layers most useful and found universally in greenhouses that are utilizing sensors. Light is another data point that is often easily collected. There has also been some research done at Vineland into correlations between carbon dioxide (CO2) and electrical connectivity (EC) in the soil.

In terms of applying sensors, Lynch says there are some challenges. For one, in some bigger greenhouses, plants are placed on long benches that can have different needs on one end or the other. Airflow, he says, is one thing that can affect each plant differently depending on where it’s located — specifically on the edges.

“Sometimes benches can be tilted and that can be invisible to the eye, but if you run water down it, then you’ll see run in a specific direction, thus making some plants get more water than others,” Lynch says. “So what they showed us is that you won’t necessarily be able to get away with just a few sensors — you’ll need more to make the best decisions possible. And that starts to drive up the cost.”

Lynch notes that one of the biggest benefits of smart irrigation is to save growers money, specifically on labor. In one Canadian commercial greenhouse for instance, the system reduced water usage by 15% and saved $2,800 Canadian (about $2,184 U.S.) in labor costs per acre of greenhouse space. But if more sensors are needed — and one estimate called for up to 2,000 sensors to cover a medium-sized greenhouse space — then it becomes harder to justify the cost.

The next step, Lynch says, is moving towards using more of the data already collected by the greenhouse and developing a decision support tool for growers. This model would rely on hardware already in place in the greenhouse and use data such as humidity, moisture and temperature to infer the soil’s real-time water moisture level. That data then will be used to map patterns in the greenhouse’s climate and water conditions. And from there, this can allow a grower to make a detailed prediction on whether and when to irrigate or not. Done in partnership with Dutch company Lets Grow, this model also has the added benefit of not having to worry about sensors being knocked over and placed properly — leading to inaccurate data collection.

That model, though, is still being fully fleshed out and fine-tuned for wider use by growers. Lynch, in the meantime, will keep working to push it forward with the end goal of making growers more efficient.

“Every day can be a download of the previous day’s data,” he says. “Growers are busy and this is unfamiliar to them — so I get why this can be hard for some. But growers are capable. And they will buy and invest in technology if it will help them.”