Faster Math = Better Growth: How AI is Bringing Farmers New Insight on Soil Dynamics

Faster Math = Better Growth: How AI is Bringing Farmers New Insight on Soil Dynamics

Phillip Owens is a research soil scientist and the research leader of the Dale Bumpers Small Farms Research Center in Booneville, AR. His research focuses on the interactions between soil, water, and nutrient interactions over large areas of land.

Welcome, Dr. Owens, to Under the Microscope

UM: Why is it important to understand the kinds of interactions that you study between soil, water, and nutrients? How does the information you discover influence how farmers run their operations?

PO: Interactions between soil, water, and nutrients create micro-environments that can help plants thrive or cause them to fail. Soils store water and nutrients and provide structural support for plants, so conditions in the soil are key to helping plants flourish. When we discover information about how the properties of soils vary across landscapes, this information can allow farmers to target the right management practices in the right places. 

UM: What were some of the major challenges in your research before you started using AI?

PO: Before AI, programming software to predict soil properties was much slower and more tedious. With AI, we can generate more precise computer programming that would be very time-consuming to generate by referencing older code from different sources. Put simply, it has helped us with the nuts and bolts of programming complex scientific models that can account for soil variability and performance, which allows us to move forward with discovery much faster than we would have otherwise. 

Dr. Phillip Owens in the field

UM: What is AI, from your perspective, and how are you incorporating it in your work?

PO: I think of AI as the process of incorporating human intelligence into computers through programming languages. You can think back to when you learned to drive a car and made lots of mistakes in the beginning. You learn from your mistakes and eventually become a proficient driver. In a similar way, the algorithms of AI incorporate data and feedback to improve over time. 

In our work, we are focusing on goals like precision optimization of farm inputs by using data to help AI systems learn positive and negative outcomes, with the goal of tailoring inputs to the precise amounts needed to save farmers time and money. 

UM: What tools do farmers use to make decisions now? How would AI improve on them?

PO: Farmers, in general, have a great understanding of the land they work with. They have tremendous knowledge and sharp intuition; they work with an intelligence that is vastly superior to AI. Nonetheless, AI can be a useful tool in certain scenarios. For instance, AI can be used to analyze thousands of images taken from satellites to develop models of moisture patterns in their fields. Any farmer can, and does, get snapshots of such images every time they go out and observe their land. But AI can integrate data from many sources to make a cohesive picture over large areas for extended periods of time. These composite pictures can enhance and inform the intelligence and intuition that farmers use every day to do their work.

Dr. Phillip Owens in the lab

UM: How advanced is the use of AI in your work now? Is it already being used in ways that affect farmers?

PO: AI is extremely useful for integrating vast and diverse datasets, but not very advanced when it comes to seeing basic patterns that humans can see quite easily. For example, if you look at an aerial photo, you can quickly see patterns of green which are related to plant health; however, it is a complicated task to program a computer to recognize what our eyes easily pick out. We use AI to give us summaries of thousands of images of landscapes and elevation models that we can use to understand how soils perform their basic functions, but AI can’t understand soils in the way that soil scientists can, and it certainly can’t interpret the images in the way users of soils, such as farmers, can. 

Today, we are using AI to do certain tasks on a much greater scale than we could do without it. In the past, a field soil scientist who was trying to create maps for farmers to use for land management would look at a few aerial photographs of an area of interest and rely on their own knowledge and experience of soil, topography, and other factors to guide them. Now scientists can use AI to aggregate thousands of those kinds of images with multiple other datasets to test hypotheses on how the soil performs with respect to water movement and drainage. The AI output can show patterns that are evident over many images, but that a single image could not capture. These patterns are linked in time and space, which can directly affect the ways we manage soils. 

UM: What questions do you have about what AI can do in your research? Are you using it in ways that haven’t been explored much before?

PO: We would like AI to learn fundamental physical principles that govern water movement and water availability in soil. To date, this has not happened, and AI can’t show any understanding of how soils will perform in basic conditions. For example, where will the moisture go after rainfall? AI, as we know it, has no answers to this question for a particular landscape because it is not equipped with an understanding of basic physics. We think it will get there someday, but it’s not there yet. Right now, we are using fundamental elements of AI such as machine learning to develop high-resolution products related to soil moisture movement at field and farm levels.  

UM: How reliable is AI? How would you decide when, or if, it’s ready for farmers to use?

PO: AI is reliable for basic tasks involving computer programming and model development. Models generated by AI can be very useful inputs for decision-making, but humans still need to make the ultimate, final decisions. If we rely on AI for decisions like when to plant, when to apply fertilizers or irrigation, when to engage in other management interventions, we could be making a mistake, because there is not enough data yet to answer those kinds of complex questions. AI is a long way from replacing human intelligence, particularly that of an experienced farmer. 

AI in Agriculture Image

UM: What are some specific examples of how your use of AI might show up in the day-to-day operations of a farm?

PO: AI-generated summaries of remote sensing data can be used to determine patterns in yield or soil moisture that could be very useful in delineating agricultural management zones for farmers. The farmer can see in these summaries what the historical patterns of yield and soil moisture have been and can determine which management strategies are best to address field variability. In the near future, farmers may be able to pose questions about management systems on a ChatGPT-type system and get options based on the goal of the producer. 

UM: In the long run, how do you think the way we handle issues of soil, water, and nutrient interaction may be changed by AI? What are your hopes for how AI can improve farming?

PO: AI can integrate complex data collected over time in ways that allow for very detailed plot work to be scaled up to large regions; we can test hypotheses about the effects of different treatments on agricultural outcomes in ways we couldn’t before. One particular superpower that AI has is the ability to do tedious things millions of times and not get tired. From a research perspective, we can rely on and benefit from AI to organize and examine research data to generate results that can help us learn and optimize farm production.