Climate change is impacting many industries across the globe, and the farming and agriculture industry is no exception. Changes in rainfall patterns, temperature, and increasingly extreme weather place farmers under immense strain and threaten food security. However, there is a solution, and Manal Elarab, the COO at Regrow Ag, is here to explain how Regrow is changing the farming industry.

Regrow Ag aims to empower food and agriculture industries to adopt, scale, and monetize resilient and regenerative agricultural practices. We start by learning about Manal, her professional career journey, the how and why behind Regrow Ag, and the company’s overall mission. We then discuss why agricultural practices need to change, and unpack the complex relationship climate change has with agriculture. Hear about how Regrow Ag is leveraging machine learning to enhance regenerative farming, what makes regenerative farming practices different, the different technological toolkits Regrow Ag has developed, and more. Tune in to discover how technology is being used to revolutionize the farming industry with Manal Elarab from Regrow Ag!

“Compared to other industries, agriculture almost finds itself on both sides of the climate change equation. It is what agriculture is doing to the climate and what climate change is doing to agriculture.” — Manal Elarab

“Machine learning plays a big role at Regrow. Our core offering is built on two key elements. A process-driven carbon model and the machine learning-based toolkit model.” — Manal Elarab

“I would say machine learning is a tool that powers a product.” — Manal Elarab

“I imagine myself and my kids walking down the grocery aisle, picking up snacks and cereal boxes, and pasta, and selecting products that the growers have used regenerative practices in producing those crops.” — Manal Elarab

Transcript:

[INTRODUCTION]

[00:00:02] HC: Welcome to Impact AI, brought to you by Pixel Scientia Labs. I’m your host, Heather Couture. On this podcast, I interview innovators and entrepreneurs about building a mission-driven, machine-learning-powered company. If you like what you hear, please subscribe to my newsletter to be notified about new episodes. Plus, follow the latest research in computer vision for people in planetary health. You can sign up at pixelscientia.com/newsletter.

[INTERVIEW]

[00:00:32] HC: Today I’m joined by guest, Manal Elarab, COO of Regrow Ag, to talk about regenerative agriculture. Manal, welcome to the show.

[00:00:40] ME: Hi, Heather. Thanks for having me.

[00:00:42] HC: Manal, could you share a bit about your background and how that led you to Regrow Ag?

[00:00:46] ME: I sure can. My name is Manal. I’m born and raised in Lebanon. My interest in how crops and plants have been growing started way back in high school. We had a full semester dedicated to plant biology and plant physiology and biochemistry. I was fascinated by plants and crops. I started my education in agriculture engineering and obtained my master’s in navigation engineering back in Lebanon. Then my journey continued here in the US. I met my mentor and advisor Dr. Mac McKee at Utah State University. I was trained with his team to obtain a Ph.D. in remote sensing and data science.

My Ph.D. revolving around machine learning algorithms like RVMs and SVMs. How they could be used with multispectral high-resolution imagery to serve precision agriculture. I was very involved and heavily trained in machine learning and data science during my Ph.D. How did I get into Regrow Ag? I started my professional career in startups, addressing precision agriculture, either on the software or the hardware side of trying to solve the problem. I actually met my founder and boss, Anastasia at one of the SPIE, the International Society for Optics and Photonics Conference. I was very inspired by how she’s building the company and how she’s thinking, her vision to what Regrow Ag can be. At that point, we were still called FluroSat. I met Anastasia and I was very fascinated by her vision of what the technology can serve in precision ag and sustainable ag. I joined the mission of Regrow Ag. I was actually the first US hired. I’ve been on board for the past four years trying to make a legal ag and impactful company.

[00:02:40] HC: What does Regrow Ag do and why is this important for agriculture and for climate change?

[00:02:45] ME: Agriculture and climate change have a, what you would call a complex relationship. Compared to other industries, agriculture, almost finds itself on both sides of the climate change equation. It is what agriculture is doing to the climate and what climate change is doing to agriculture. Agricultural production on scale that can feed the 8 billion people that we want to feed as a major contributor to climate change, as well as land use change, and biodiversity loss, and depletion of water resources. Just 31% of the world’s GHG is produced in the food and agriculture value chain, 70% of the world’s water consumption goes to agriculture. Obviously, agriculture is the play that I call the woodly art on the climate agenda.

On the other hand, the climate also impacts agriculture. All the new locations where we’re witnessing lots of floods or droughts, or the climate is changing. Agriculture is also a player and trying to understand how those changes are going to impact what can be grown and what should be grown. That’s why I say Ag finds itself on both sides of the equation of what’s happening to Ag and what Ag is also inflicting on climate change. What Regrow does and our standards, as we look at agriculture. We think of agriculture along with many partners in this community, we think of agriculture as a way to solve for climate change, and at the same time, build the resilience needed to fight or meet where climate change is taking the agricultural regions.

Strong advocates at Regrow of soil health, and how it can be restored through regenerative farming practices, which brings us to what does Regrow Ag do? So Regrow Ag, with an enterprise software provider that provides software solutions to enhance the measurement and monitoring capabilities that food companies, and food and fiber companies need in order to scale the adoption of climate-smart regenerative practices. Then as a result for them also to reduce their scope three mission. Our course has products are to sustainability insights, which is dedicated to help organizations plan investment to reduce their emissions, as well as an MRV, which is a Measure Report and Verify platform, which enables leaders to a path of credible measure and report and verfies the impact of the regenerative farming programs at scale.

[00:05:16] HC: For those who might not be familiar with regenerative agriculture, what is this? How is it different than traditional agriculture?

[00:05:22] ME: Great question. Regenerative agriculture is the original way of farming. It is taking into consideration, your soil health, as well as your water consumption. Your biodiversity on your farm and on your land and being a good steward of the land, so the definition of regenerative, there’s a lot of different definitions out there that describe it more from either a soil health perspective or from an application. You have to be doing A, B and C to call yourself – you’re doing regenerative agriculture. I think the point is looking at that regenerative agriculture approach, it’s a box, it’s a toolkit, and being able to match the right practice that ensures that your GHG footprint is accounted for and your decisions, your farming decisions are not creating a huge GHG footprint. At the same time, you’re still profitable. You’re taking care of your soil health. You’re rejuvenating, recharging, taking care of your soil through erosion and water, and all of that.

[00:06:24] HC: Going back to Regrow Ag itself. What role does machine learning play in your technology?

[00:06:30] ME: Machine learning plays a big role at Regrow. Our core offering is built on two key elements. A process-driven carbon model and the machine learning-based toolkit model. They’re both dedicated to detecting the current and historical agriculture practices at the field level. Both of these are the core of our products being MRV and sustainability insights. Our models, currently use a range of satellite imagery, weather data, and other publicly available data to identify the crop on a field, to identify a cover crop present on a field, as well as tillage intensity and other practices. Then all of that data gets passed to the DNDC, Denitrification Decomposition model, which enables us to estimate a baseline level for carbon in the soil and predict the effect changes in management practices, and how will that be reflected on an SOC.

[00:07:29] HC: How do you gather all this data of different modalities that you need in order to do this? How do you annotate it in order to train machine learning models?

[00:07:37] ME: There are several ways we collect ground source data for modeling, training, and validation. I’ll start by talking about a program that we’ve created called Sustainability Watch. We operate a network of trained ground truth data collectors, which are agronomists. They are already in the field doing what they need to do from soil sampling or providing consultancy to growers. We reach out, we created a network of those agronomists.

Through this program, we recruit them, we train them, and when they’re doing the field visits at different times, during the year, they collect a range of data about crop practices, residue cover, and other use of cover crops and other inputs. In addition to the data that we collect directly, we also leverage a range of third-party data that are available in the literature and other locations. We also have a strong academic partnership and ties to academic and government parties. We also have access to data from those partners as well.

[00:08:40] HC: What kinds of challenges do you encounter when working with these datasets? You mentioned satellite imagery and weather data. Are they easy to use straight out of the box for machine learning or are there certain challenges that come up that you have to deal with?

[00:08:53] ME: Yeah. Imagery, specifically, imagery is it becomes tricky in two dimensions, its timing, and its availability. There’s the timing and availability of imagery. Then how do you actually work with imagery data? The most significant issue with imagery is trying to match the availability of the image with the agriculture practices that you want to be able to see from the image. Sometimes the imagery is not available when you want to be able to look at it. When the events that we’re monitoring happen, they can happen over a few hours, or they can happen over a few weeks. You might have imagery that is cloudy. You’re unable to use those imageries intelligently. That’s when we identify proxies or second effects that persist on the landscape as a key strategy for us here to be able to monitor and identify what we wanted to look at and track indicators of specific practices rather than the practice itself.

It’s pretty much the availability of the imagery. That’s one thing. The other piece of working with imagery and its challenges is, how do you work with that data? It’s challenging to work with imagery. We at Regrow have developed a number of tools and best practices that are designed to overcome some of these problems. For example, using the Google Earth Engine, as our geospatial data processing platform has been very pivotal, because that takes the burden of data ingestion, of posting of moving the pixels around for our team to work with and allows us to focus on the model development and the deployment itself. Also, we’ve heavily invested into model frameworks such as transformers that are robust, robust to clouds, to haze, and to other image quality issues that would commonly lead to poor performance.

[00:10:43] HC: Sounds like tools are very critical and choosing the right type of model to handle some of the challenges there. You mentioned a couple of variations that you see in your data like cloud cover, but how do you ensure that your models work for other variations that you might see, maybe different crops, different geographic locations, or climates? How do you make your models robust to those types of variations?

[00:11:05] ME: We’ve adopted a transfer learning approach to the expansion of our models to new geographies and new crop types. Often the core model can be trained in an information-rich environment such as the US or Europe. We train and we deploy in those information-rich environments. Then we retained and deployed in new areas of address using a much smaller data set. Our focus here is on identifying the differences in practices between the location we’re transferring to. An example of this is monitoring cover crops in Europe. There are several activities related to the cover crop in Europe that are fundamentally different from the cover crop model that we’ve developed for the US. It’s really important to understand the differences, the agronomical understanding of the practice, and bring that knowledge back into the model for it to be representative of that crop type and for that layer that you’re trying to create.

There is data available for testing the accuracy, which our scientists look at it as – it’s a way to always reference back and understand how the training and adoption of this new model are going on in this new geography. There is data available for testing the accuracy, but generally, it’s not enough to take a brute-force approach to model development. We go through a process of training the model to report on the state of the field over time, which enables us to identify specific practices, rather than just predicting them directly. That’s been our approach to taking the model to new geographies.

[00:12:47] HC: You mentioned the example of different cover crops. That’s very important information to know in training models and where you can apply them. I suspect the average machine learning developer, at least in the beginning didn’t understand the importance of that. How do your machine learning engineers collaborate with other experts in order to get this information?

[00:13:08] ME: Many of our machine learning experts, they come from environmental science or agronomy background, which we’re very lucky, which provides a shared language for the agronomist and the environmental scientists and policy experts that we need to collaborate with. Also much of our development process is driven by defining the scope of the problem. We’re trying to solve early in any project. Identifying the success criteria we have for any new product model or data. These experts are also available internally at the company. As a team, you would see the agricultural expert, also joining forces and being coupled with environmental scientists that understand the nature of that data and the agronomical context needed.

[00:13:53] HC: How do you measure the impact of your technology?

[00:13:55] ME: The impact of our technology, we think about it in two ways. We think about, what do we enable the product to do for our users. For example, being able to monitor more crops, being able to bring to the market, more programs that advocate for regenerative practices, and being able to account for scope three emissions in a very reliable, credible way. That’s one way we think about our impact. The other way is all of the efficiencies that we introduce into the product and into this solution that we’re building the efficiencies around being able to scale a regenerative agricultural program. So having a platform where a grower in Louisiana can join a program offered by the food brands that they source from that geography, a simple grower enrollment for them to choose a program to start their journey of regenerative agriculture. Do something different and decrease their scope. The efficiencies that we introduce and the ability to scale for this program are how we measure the impact of our technology.

[00:15:03] HC: Is there any advice she could offer to other leaders of AI-powered startups?

[00:15:07] ME: I would say machine learning is a tool that powers a product. It reduces the amount of data needed from a customer or increases the scale at which a question can be answered. It’s when AI or machine learning becomes the focus. It’s easy to get stuck in a loop, a loop where prioritizing for a better model overshadows and becomes bigger than actually solving a problem, or it can impact on the revenue or on a customer need. I would say machine learning is a tool.

[00:15:42] HC: Finally, where do you see the impact of Regrow in three to five years?

[00:15:46] ME: I imagine myself and my kids walking down the grocery aisle, picking up snacks and cereal boxes, and pasta, and selecting products that the growers have used regenerative practices in producing those crops. I’d see Regrow, enabling those food companies, enabling a system where growers and food companies can kick start their journey of regenerative agriculture and decrease the footprint. I see Regrow being the catalyst to scale these products and these practices across the food aisle in your grocery store.

[00:16:33] HC: This has been great. Manal, your team at Regrow Ag is doing some really interesting work for agriculture and sustainability. I expect that the insights you’ve shared will be valuable to other AI companies. Where can people find out more about you online?

[00:16:46] ME: Please visit our website, Regrow Ag where you learn more about regenerative agriculture and some of the amazing projects that we’ve been supporting food brands like Kellogg’s, Cargill, and General Mills and some of the great work that the team has been so proud of doing. Also, check out the blogs, there is a lot of information if you’re looking to learn more about the space of GHG accounting and scope one, scope three, and more about this market.

[00:17:16] HC: Perfect. Thanks for joining me today.

[00:17:18] ME: Thank you. Thanks, Heather. Thanks for having me.

[00:17:21] HC: All right, everyone. Thanks for listening. I’m Heather Couture, and I hope you join me again next time for Impact AI.

[OUTRO]

[00:17:31] HC: Thank you for listening to Impact AI. If you enjoyed this episode, please subscribe and share with a friend. If you’d like to learn more about computer vision applications for people in planetary health, you can sign up for my newsletter at pixelscientia.com/newsletter.

[END]

Resources for Computer Vision Teams:

LinkedIn – Connect with Heather.
Computer Vision Insights Newsletter – A biweekly newsletter to help bring the latest machine learning and computer vision research to applications in people and planetary health.
Computer Vision Strategy Session – Not sure how to advance your computer vision project? Get unstuck with a clear set of next steps. Schedule a 1 hour strategy session now to advance your project.
Foundation Model Assessment – Foundation models are popping up everywhere – do you need one for your proprietary image dataset? Get a clear perspective on whether you can benefit from a domain-specific foundation model.