Decoding Complexity: AI in Agricultural Research

Dr. Debjani Sihi of NC State explores how artificial intelligence is helping scientists make sense of complex agricultural and environmental systems, uncover patterns in vast datasets and improve predictions. She also explains why AI alone isn’t the answer and why scientific knowledge, human intelligence and real-world expertise remain essential to turning these powerful tools into meaningful solutions for agriculture. 

Transcript

Copy Transcript Copied

[00:00:00] AgTech 360 discusses breakthrough technologies that are impacting growers, businesses, and consumers. Hear from industry and academic experts about what's on the horizon.

As researchers collect more data than ever before, artificial intelligence is helping them make sense of complex agricultural and environmental systems in new ways.

In this episode of AgTech 360, we're joined by Dr. Debjani Sihi of the NC Plant Sciences Initiative, but also, uh, of the Department of Plant and Microbial Biology, and also the Department of Crop and Soil Sciences here at NC State. Debjani is gonna discuss how AI is transforming agricultural research and accelerating scientific discovery.

So welcome, Debjani.

Thanks, Adrian. I'm glad to be here.

We're glad to have you. So let's start with you. Um, tell me a little bit about your [00:01:00] background and what got you into the world of agricultural research.

So by training, I'm a biogeochemist, which is a very fancy way to say I'm an environmental scientist.

Another way to say is, like, I work in interdisciplinary space, biology, geology, chemistry, in environment. So I started my undergrad in agriculture, then got really attracted to a field where I can apply all different basic sciences together, and I think that sparked my interest very early in my academic career.

And then throughout the process, I transform into more natural system science, going beyond agriculture, studying forest, wetlands, more ecosystems that those are more naturally relevant then I feel like this current position is kind of homecoming for me to ag field, and I feel like this is a perfect time to be in ag tech because of, like, we [00:02:00] are really in the digital world in terms of access to datasets, and we have a lot to unpack from that.

And to that point, you are actually an NC State AI cluster hire.

Mm-hmm.

So for people that don't really know what that means, what is a cluster hire, and why is NC State, looking to invest in bringing new faculty, uh, to work with researchers who are focused on AI and in, particularly in agriculture?

So I'm part of NC State's College of Agriculture and Life Sciences Applied AI Cluster, so that's a mouthful.

In general, cluster hire is, you can think about, like, a group of faculties that work together to a

Space, a field which has a shared interdisciplinary program or theme. So in our case, this is an applied AI cluster, so we are in charge to apply different agricultural- different AI applications to agricultural domain.

So my role is connected [00:03:00] with plant sciences initiatives. I'm working in plant and soil system, overall different challenges that agriculture faces within that domain, like ill forecasting, thinking about, uh, any type of stress that, crops, uh, experience, starting from extreme events. We have colleagues in the cluster who are working on the livestock system and, we are looking forward to have another colleague who would work more on the extension human facing. So we are technically covering the- Research on different sectors, but also teaching and extension. So , the Applied AI Cluster, I taught for the first time a very new course, which is targeted for agricultural workforce, particularly CALS undergrad.

So really one of the, task is also to empower the future generations in so that they are more competitive in the job market, especially related to ag tech.

I'm also, the faculty contact [00:04:00] person for CALS, for a new AI initiative across the university, like, a AI liaison. So kinda wearing those hats to do sort of, uh, connections, research, connecting with people and resources as and when needed.

So wow, that's a lot. That's a lot in terms of your research and education and workforce development.

And I know you've not been at the university so long, so it's really impressive to see you have s- so many fingers in so many pies.

Now, one of the things that your work does, I know, is really focusing on understanding really complex biological environmental systems, as you've already kind of alluded to. Can you give us a sense of why this is so hard to study and why it's so complex, using traditional research approaches, and how now with AI and other approaches, we can understand these things a little bit more easily?, 

So I study ecosystem, which is like outdoor. So things in outdoor are [00:05:00] uncertain. What we do practice in very controlled lab environment, that does not always apply in the field because, there are so many different complexities, feedbacks that are constantly going on.

All the approaches, all the functions are nonlinear, so there are like this biology. It's, uh, like plants and microbes in soil, they're interacting constantly with minerals, which are like dead, living in the soil. Or maybe the biophysical environment, the climate. And in agriculture especially, like we have this human dimension that is the center of the whole system.

So it is a lot to unpack in the sense of things are nonlinear, they are interactive and, and often, uh, we have to deal with a scaling issue.

So what do we mean by scaling issue? Let's say we are thinking about, uh, the backbone of agriculture [00:06:00] or, or food system is nutrition. So nutrition comes from like the nutrients and minerals in soil and, and often how sustainably we manage nutrient in soil and ecosystem that depend on, , the tiny microbes that are living in the system.

So how much they're active under which,, type of environment, how much, that would impact or interfere with the chemicals that we apply in traditional agriculture or, or maybe organics that we apply in organic agriculture. And, and then what are the Downscale effect of how much of that is being lost to environment, to air, to our groundwater.

So it's rather complex, and the effect that we see is at the large scale, but often the drivers are operating at very small scale. And also, you think about what we practice here in North Carolina, we want to apply in Midwest, that might not be [00:07:00] very easily translated because we have to think about different climate, different soil.

So for that reason, that underlying complexity comes as a challenge. But it's also an opportunity . You don't feel your job is boring- Mm-hmm ... because you are always constantly challenged to identify solution.

The North Carolina Plant Sciences Initiative impacts lives through innovative applications and discoveries. By leveraging cutting-edge research and technology, we address global challenges related to agriculture, sustainability, and human health.

And so how are AI and machine learning tools kind of fitting into this, understanding this complexity and helping you or other scientists, you know, make decisions , and, and gain understanding?

So AI could help us, uh, in simplifying this complexity by identifying trends, identifying patterns, the threshold, maybe the risks that are not here but maybe [00:08:00] upcoming. So AI could help us building those, features that are not in our system now. For example, I can think about, we are in a big drought now, so there are consequences.

And, and this is not just true here. We are seeing episodes of different types of extreme events in agriculture system in general.

So if we know that this is coming, uh, in real like near term or real term, then we can help, aI could help, by understanding what is the known, predict the unknown. And that prediction, if it comes at a scale, at a time that is relevant for farmers, stakeholders or extension agents to give recommendations, I think AI could do that rather quickly than traditional approaches. Because traditionally what we have been doing in ag field or, or generally in environmental ecosystem field, we, we do long-term practices, we do trials.

I'm not saying those are not needed. Those are [00:09:00] absolutely needed. What AI could do is really extract value, extract information from those already, uh, vast dataset so that we can come to a conclusion, we can come to a solution rather quickly and in real time.

So AI is helping us with the output, but when we talk about the input, you talked about data sets.

So what kind of data are you using? Are you either collecting or accessing? What, what are those data sets? What do they look like?

Uh, different types of data sets, multiple sources, multiple scale, uh, across space and time. I use data sets from like very small scale. We sometimes call it a, a pore, like a soil pore, which is at micron scale or microbial data set, which is like really small scale, nano to micron scale, to data that are collected in the lab, to plot, to field, to watershed, and regional and continental.

So [00:10:00] really depend on the question that we are asking. And there are different ways that we could do. For example, like even in the same field you get a, you can have different types of soil. Based on that, you need to have different management practices. Like you would not likely in more economically feasible way not apply the same amount of fertilizer or same amount of irrigation water.

So if we have spatial data that are then, uh, which is like spatial imaging using drones or satellite or flights, again, depending on the resolution, and then connect that with, let's say, management practices and, and other, data that are in temporal dimension, that things change very dynamically, like nutrients and soil moisture, or our climate, the temperature, the microclimate. You have to combine that space and time together and that's where all the complexity comes. But also, as I say, like, that, that's where your job [00:11:00] becomes much more, uh, exciting because, it comes with data across different modalities.

So one of the things I, um, you kind of alluded to this, Debjani, but I'm, I'm interested kind of from a bigger picture perspective, how do you see these tools

and predictive modeling, all of th- the things that you work on helping us help farmers become more resilient, more sustainable, more efficient in their farming practices?

So

I think we definitely have to start with the, uh, the existing, uh- classical approach of collecting data that we, we have done that over several years.

Now what we could do is we can, pack those data sets in a way that makes sense and, and much of my work, for the last few years, in, in fact for the last decade, was like really working on process-based models. We have yield data, we have soil nutrient data, we have emissions [00:12:00] data. What I do is I connect all those different data sets a- along with like microbial data or, or other chemistry data to see like how we can predict the yields, the emissions, the, environmental losses of nitrogen or carbon or phosphorus. And then, that requires a lot of understanding of what is happening, like at the process scale, which is like rich.

What we need to do to really make it as a... And that sounds contradictory maybe to like making it quick, making it efficient, time, uh, real time, is really learn from that and make a system, and that's where AI plays into, uh, uh, like have a big role in my work, is, create a simpler system of that complex model.

And that simpler system, like in, , terms we use are surrogate models. That's one example. We often use terms like hybrid model. So AI learn from process, and then it can [00:13:00] also help understand new process because the process understanding is like we already have some assumption. This is how our system works.

But AI could help us understand is you are missing a big piece because that is something I find this feature is important, and that can help us build this process. So it a two-way process, and through that process we can make our AI models, surrogate models, which often is a pathway to create decision tools.

And that is what we have to have, to hand it over to our extension agent, to hand it over to our stakeholders, and really test it against the real life, like real world scenario.

It might work, it might not work. We need to learn, and then eventually we, we will, uh, actually improve the performance.

Yeah. I mean, you, you're kinda going the direction I was gonna ask you, which is like for the future, which I know is somehow difficult because AI and all these tools [00:14:00] are evolving at such a rapid rate.

Mm-hmm. But also you're on the cutting edge of this, so you must have some idea how in the next ten years do you feel, these types of tools are gonna evolve and, and help or accelerate your research?

Um, I think, maybe some feel like AI is evolving

at a pace that we are not keeping track of, uh, which is a genuine feeling because it is really, coming up. What I would always say, and, and this is coming from like an applied researcher, is like 

that as a tool n- and Like, look at, at the background, what's happening, try to understand.

So for, for example, in environmental application, AI has been treated as black box for a long time. But now we have new technology, new algorithms where we can actually explain, make it more explainable, interpretable AI. And that's what I mean is like, well, we don't want to just trust it blindly, and this is always coming [00:15:00] from like an applied end.

We need to unpack it rather further. And there are even new tools that we are applying in our research is like we are trying to build equations, and we are trying to find, like, not just why it is happening, but also, like, where it is happening, how it is happening. So that we need to ask those questions always so that we can improve the performance.

And this is something I'm talking about predictive AI, but I think another sector of AI that is emerging is the generative AI. So yes, it hallucinate, but again, like, as we can improve it by our knowledge, by the domain science, the science needs to be there, the human intelligence needs to be there. So we need to work it as a tool so that we can improve over the time through our integrated effort.

And for that, we need disciplinary scientist who actually have the [00:16:00] knowledge of the system. We need the software scientists who can give it to life, and we need the practitioners who are actually facing that in real time, and we need to work together. I think that, that needs to be there in the near future.

Well you've just summarized very nicely what we've heard from other guests on this kind of podcast around AI being a tool- Mm-hmm ... help with efficiency and, uh, how human intelligence, which is I think reassuring for all of us is- Yeah.

Yeah ... is still gonna be very important. But also how understanding how the AI is working and the transparency will build trust, not only with- Yeah

Yep

consumers and the public, but also with the scientists themselves who are using it. Absolutely. So now, again, very general, uh, Debjani. So when you think about the future of agricultural research, where do you, where do you see AI having the biggest impact?

Um, I think if you can think about [00:17:00] what agriculture is like, who is being served through this sector, the human being, uh, the people, the, the producers, and also the consumers.

So we really need to make AI, uh, as a tool so that we can identify patterns, we can understand what is happening, but we can also make it do good forecasting for us.

, Like economic analysis. At the end, we need to have like better ROI. That's, that's coming from the producer side because that is important.

And also from the consumer side we also need to understand like, where is this coming from? And AI could help us understand the whole supply chain together. And I think like there are a lot of technology to tap our fingers into it. We really need to be cognitive about what works and not blindly [00:18:00] using just an AI. Like it really matters what kind of dataset work for what algorithm, and we need to understand that so that we can apply it correctly.

Debjani, thank you so much for joining me today. Uh, AI is certainly helping unlock new avenues to understand the complex systems within agriculture, and it's really exciting to think about how these advances could help shape the next generation of scientific discovery and innovation. I've really enjoyed this conversation.

Thank you so much. Thank

you so much. Thank you.

AgTech 360 shares relevant news and breakthroughs with audiences across the globe. Stay connected and join the conversation by following NCPSI on social media.