Q&A: The place for AI in agriculture
share
FRUITA, Colo. — Researchers at the Colorado State University experimental farm in Fruita are trying to get soil to retain more water, find alternative livestock feeds to alfalfa, and use artificial intelligence to irrigate more efficiently.
Using less water leaves more in the Colorado River during a period of historic drought and aridification in the western United States. For farmers, using existing resources more efficiently, not spending more money then they have to on fertilizer for example, means profit margins would be better.
Perry Cabot, an extension professor of irrigation and water resources for CSU’s Western Colorado Research Center, leads the project applying machine learning, a type of supervised AI, to watering a corn field at the research center.
“I want the next generation of farmers to not have to contend with the same problems that their parents and grandparents had to deal with,” Cabot said of the project. “I want people to be able to punch a button on an irrigation system and then go to the ball game with their family.”
He says with more information and efficiency from AI systems, farmers would be able to increase crop health in individual sections, making for a more uniform field, which would mean a better yield.
The technology already exists to make a specific, granular plan for how much water should go in each section of a field. But to do that as efficiently as possible, farmers would need more information, like data from infrared and thermal sensors on a drone, which would be too much for a person to interpret. That’s where machine learning comes in.
The $230,000 project is funded by the Colorado River District, United States Department of Agriculture, and Colorado State University.
Rocky Mountain PBS recently spoke with Cabot about the project to learn how it works.
The following interview has been edited for length and clarity.
RMPBS: First could you tell me a little about what you do at the research center?
Perry Cabot: I oversee a research program that focuses on irrigation technology, consumptive use evaluation and alternative cropping systems, all of which are designed to be more efficient and more conservative with the amount of water that we use here on the Western Slope.
In general, there's applications across the world, but my particular interest is to tailor these projects with an understanding of the Upper Colorado River Basin where I work.
RMPBS: Could you give me an overview of the irrigation project? What’s the size and timeframe?
Cabot: Fundamentally, we are using machine learning in order to help write irrigation prescriptions for the most optimal use of water.
Right now, we have a 12 acre field [with a] linear move irrigation system. That 12 acre field is broken out by zones. Those zones are defined by the number of nozzles on the linear irrigation system. The zones for us are 40 feet each and we have 15 of them. So, we have a way to vary the irrigation every 40 feet.
We built [the linear] two years ago. Last year was kind of a training year, we learned a little bit and we've certainly learned that we can operate it.
We're probably going to be at least doing this project for a couple more years. We know how much water we're applying out there, what we want to do is get to a point where the yields are better than what we would expect to have under furrow [irrigation].
Many farmers would just rightfully say, I don't have time to change up my whole operation, just on this speculative basis. What they want to see is, are your yields better? Are your profit margins better?
RMPBS: What problems are you specifically looking at with this project?
Cabot: The problem we're trying to solve with the irrigation system is how we can apply less water to the field, how do we apply less water but get the same or even a better result?
Part of the answer to that question is in the form of what we call an irrigation prescription.
An irrigation prescription with a sprinkler system is tied to how much water the crop is actually using and how much water the soil can actually take.
So, you sort of design your prescription around this water balance. You always want the crop to have enough water so it can satisfy its consumptive use demand, but not so much that we're causing enormous rutting and certainly not so much that we're flooding the field.
How long, how much, how often, are the elements of a prescription. What we're now doing with variable rate irrigation is trying to then further tailor the prescription. That's where those static and dynamic variables come in.
RMPBS: So, to be more efficient, you need more information, more data, to make the prescription more precise. Where does that data come from?
Cabot: Then the next problem we're trying to solve is how can we utilize the dynamic variables to give us an even better prescription? I'll fly the drone so I can get data on a dynamic basis.
[The sensor carried by the drone] can capture information in the visible light spectrum, but also outside of the visible light spectrum, particularly the infrared and the ultraviolet spectrums.
We can actually scan the field and start trying to understand– is there something going on now or in June that we know is going to affect yields later– but let's act on it now.
It has a thermal sensor that I can literally measure the temperature from 300 feet up. It gathers discrete packets of temperature data every ten centimeters. That introduces the ability to understand differences in energy at the land surface.
We gather together signatures about the health of the crop through light, and we gather signatures through the water vapor of the crop, through the thermal temperature. We can model evapotranspiration.
RMPBS: Now with all this information, it sounds like too much for a person to sort through– is this where the AI comes in?
Cabot: I would agree that you basically start to leave the realm of human capability somewhere in the static variable realm. This goes without saying that you're talking about billions of pieces of data. That's where the machine learning comes in handy is us trying to actually understand how the yield is responding to all of these dynamic, continuous variables, which then leads us to a way to even write a better prescription.
Physically what you see is that [sprinkler] literally changing its irrigation rate continuously in response to whatever we've learned about the field, that is also changing continuously.
RMPBS: Could you spell it out for me even more? What’s going on in this application of machine learning?
Cabot: What most people are really interested in is like, what is going on inside this black box? I’m putting all this data in, but what’s really happening in there?
The easiest way I can describe it is, you can imagine like a code, a massive code that's written that is very similar to the kind of code that your phone uses to figure out what word you're trying to say or what Chat GPT uses to try to figure out what's the best place to go grab information to answer your question. Or what Google uses to figure out how often you need to see that ad to make you buy that sneaker.
These are all forms of machine learning and AI. So, it's building relationships inside the code. Whatever code you have written, it builds that relationship, and you throw another piece of data in there and it just says, okay, let's see if we can find a relationship between this and this for a billion pieces of data. So, it's a lot of complex relationship building between variables.
.
RMPBS: Who have you been working with on this project?
Cabot: Professor Raj Khosla, he's a world leader in precision agriculture, which is what you might call the precursor to AI. Precision agriculture was essentially as far as we got with our human thinking on how to do these prescriptions. We built models, and we did regression analysis, and we started to develop relationships, but that was back when [we had less] in the way of understanding these continuous variables.
The other person is Ray Hedgecoke, he’s the engineer [who built the linear]. Ray has done a remarkable job expanding the market and becoming an advocate for these tools. He's just a remarkable entrepreneur.