How Data Drives Greenhouse Production

AI could have been the Word of the Year for 2023. Maybe it will be for 2024. From the explosion of generative AI like ChatGPT to fake video clips and robocalls, something that seemed science fiction only a handful of years ago is now in our daily lexicon.

There is no consensus on what AI will look like in the greenhouse or when we will have achieved it in controlled-environment agriculture (CEA). To some folks, AI is the current cutting edge of machine learning and assisted growing. For others, the definition is more of a human-free approach. Either way, the state of the art is advancing.

AI: More Than Another Set Point Controller

“The majority of environmental control systems on the market right now don’t use artificial intelligence at the moment,” says James Whalen, technical sales manager at Total Energy Group. While they are more advanced than those available a decade ago, they still turn equipment on or off based on the current input from a sensor. Think of it like a snapshot in time: If it’s too cold, turn on the heat.

Higher-end environmental controllers provide more finite control, but they’re essentially doing the same thing, with more factors, more systems, and better programming. These controllers have been fairly grower-centric. They rely on the grower to decide about the needed changes (if any) and then manipulate the settings or controls to affect a change in the physical greenhouse environment. All that is starting to change.

“Over the years, we’ve really refined controlling greenhouses,” says Henry Vangameren, Regional Marketing Manager Americas for Priva. New tech, through advances in machine learning, machine vision, intelligent algorithms, and data analytics, is learning how to control the environment based on human-derived strategies.

“What’s been built into the system is logic. It’s anticipating what’s going to happen and making changes to the environment–heating, venting, lighting,” says Vangameren. “The idea is to make these adjustments to systems so the plant doesn’t feel large or sudden changes which can really affect some crops. Typically we can keep a large facility within a 10th of a degree, which sounds really finite, when it comes to controlling temperature. But, that’s important for growers.”

The new generation of AI and smart algorithm-enhanced systems are forward-looking, not just historical. Luis Trujillo, President of Hoogendoorn, USA, explains that historical data is important, but we can do better.

“You have historical climate data, but that won’t tell you tomorrow will be a sunny day. Highly changeable climatic conditions make historical data less reliable than looking forward,” he states. “We can’t look at what happened in the past to improve a crop moving forward. We have to look at the conditions coming up and how we adjust to maximize crop growth.”

 Gathering Data

Data analysts talk about clean data, and it’s a requirement for a smart greenhouse. Data that isn’t accurate or missing identification — wrong tags, no dates, format errors — isn’t useful as a historical record and won’t be actionable.

“It’s going to require a ton of clean data, a ton of inputs,” says Whalen. Good crop registration data over time is essential. “Bad data yields bad outputs. It’s true for a control system and true for an automation AI system as well. It has to be traceable. You’ve got to know where it came from and have that fidelity.”

But while the data needs to be high-quality, the methods of gathering it remain the same, even if an AI system will crunch it.

“Data is effectively gathered the same, no matter if it’s an AI system or grower managed,” says Will Justis, software team lead for Wadsworth Controls. “The main difference is what happens with the data after it’s collected. For simple feedback mechanisms, the data can remain local at the control. For large data sets and complex algorithms needed for AI, the data often needs to be transferred off-site for additional processing. Usually, this means a cloud service or other off-site provider.”

Different and more inputs may be needed, including weather forecasts, energy prices, leaf temperature sensors, and more of the typical temperature, light, and humidity type sensors than before. Whalen explains, “On an acre block, there may have been one temperature and humidity sensor for the environmental control system to operate off of that set point. Now, we might deploy ten sensors in a wireless system and get a more comprehensive vision of what the greenhouse is actually doing from a climate perspective.”

How Is AI Different From a Programmed Climate Controller?

Current high-end control systems are already programmable for predicted events. For example, beginning to reduce the load on the heating system an hour before the lighting is scheduled to come on in anticipation of the heat load. However, these systems use data that represents a snapshot in time and are adding in some known changes due to scheduling. “When systems can predict the way parameters are changing based on weather data and trends, then you’re taking that AI step,” says Whalen.

“The role that AI plays is different than just managing the set points of an environmental control system. It will be able to look at the variables and much more data. It’s going to use machine vision data of crop performance and production over the course of that growth cycle,” says Whalen.

An AI controller will combine crop registration information and the environmental control parameters during the crop cycle to evaluate the outcome and identify parameters to change on the next run. “They’re going to start processing that registration data on what the crop is actually doing and how it’s responding to those set points–how well the greenhouse achieves them. AI is going to close that loop. Each progressive cycle will get better over time,” Whalen explains.

Grower or AI: Who Makes the Decisions?

A grower may need to steer a crop to be more vegetative or generative. Interfacing with the AI, the grower can input a desired condition, and the AI will translate that into the environmental controller settings needed. It could change the employment of curtains, supplemental lighting, temperature, CO₂, or fertigation. That’s the extra layer where artificial intelligence will live on top of the environmental control system. Instead of manipulating dozens of individual parameters, a grower could adjust one, leaving more time for tasks other than being an environmental control technician.

“Of course, the idea is that the AI will be making those decisions, working with the grower to allow the grower to do more and cover more acreage,” says Whalen. He explains this using the analogy of an airplane on autopilot: “You’re still in the cockpit, but not pushing as many buttons. The computer is taking care of most of that for you.” With the AI system taking care of day-to-day manipulation, growers can take on a more managerial role.

“You start with set points a grower sets–the typical greenhouse controls. If there is a situation that you want to fall back on your traditional controls, these need to be set. Then, on top of that, you have artificial intelligence or intelligent algorithms'” explains Pieter Kwakernaak, General Manager of Hoogendoorn America. He explains that the system can be set as more of an advisor, showing the recommendations and data used to arrive at those actions. “If you like the results, it can be set up to make the adjustments to your facility, and will auto-adjust to the parameters within the range that you set. At that point, it can be viewed as semi-autonomous, but the grower always needs to stay in control,” he says.

Experienced Growers are Essential

If you’re wondering about autonomous greenhouse growing, human-free, it seems we’re not there yet, and that doesn’t seem to be the goal. Growers are vital to the equation, providing plant physiology knowledge and overall direction. They’re also needed to supervise the AI.

For now, “AI should only be providing suggestions. The grower should always be at the helm. AI feedback should never be making direct changes to a system unless a grower specifically allows it,” says Justis. However, he explains that the suggestion and approval process itself provides an AI system invaluable guidance on whether the automated improvements are acceptable for the end-user, helping to tailor future suggestions.