Resources

• NASA Harvest
• NASA Acres
• Harvest SARA

Host and Guest

Climate Rising Host: Professor Mike Toffel, Faculty Chair, Business & Environment Initiative (LinkedIn)

Guest: Dr. Alyssa Whitcraft, Executive Director of NASA Acres and Founder of Harvest SARA (LinkedIn)

Transcript

Editor's Note: The following was prepared by a machine algorithm, and may not perfectly reflect the audio file of the interview.

Mike Toffel:

Alyssa, thank you so much for joining us here on Climate Rising.

Alyssa Whitcraft:

It's my pleasure. Thanks for having me.

Mike Toffel:

So let's begin with a brief introduction and a bit about your current roles.

Alyssa Whitcraft:

Sure. So most principally, I'm a research professor at the University of Maryland in the Department of Geography. That's actually where I got my PhD as well. Remote sensing of agriculture has really been my entire career. I did some background work in forestry, but agriculture has really been my passion.

I have a handful of other hats that I wear at different times. Most recognizable amongst them are that I'm the executive director of NASA Acres, which is NASA's US focused ag program. And I was the deputy director previously of NASA Harvest, which is NASA's global ag program. And there's a bunch of other ones that I think will come up today, so we can touch on those as we get there.

Mike Toffel:

Yeah. You're involved with so many organizations, so we probably only have time to talk about a few, but let's dive in. So much of your work, as you mentioned, involves agriculture, satellite data, machine learning, AI techniques. So maybe just give us a brief 101 on the emergence of satellite data and programs for agriculture.

Alyssa Whitcraft:

Yeah, sure thing. I'll have to take off my professor over explaining hat here, but I'll be as quick as possible. I think the history, the past is pretext, so I think it's kind of important to understanding where we are now.

NASA really started the Earth observing era in 1972, and what a lot of people don't realize is actually the very first applications were in agriculture. And they were really looking at very large scale monitoring of agriculture conditions as they were sort of likely to produce or impact trade. So it wasn't monitoring the whole globe, but really looking at these big bread baskets.

And there's a bunch of reasons why things kind of didn't progress much past that up for about 40 years, remember. So 1972 to kind of the early aughts. The first big shift was in 2008 when USGS led the way in making all their data freely and openly available. It's called Imagery for Everyone.

Mike Toffel:

It's a U.S. Geological Survey. USGS.

Alyssa Whitcraft:

Yeah. U.S. Geological Survey, and they operate a lot of the satellites together with NASA. So it's sort of like a partnership. And this was a huge departure, because before the data were not openly available and they were expensive often even to people in the federal government just to pre-process them. So not only were taxpayers not getting it, but even the people who worked for the government were having a hard time getting the data volumes that they ought to have.

And right when that imagery for everything came into be around 2008, that was when Google Earth and Google Maps and all these things could proliferate and innovate upon having free and open data.

A couple years later in 2011, 2012, there were major price hikes that were a result of multiple bread basket failures, meaning a lot of the places that we rely on internationally for food didn't have good production years. And so prices shot through the roof and the G20 was like, "We got to do something about this." And they thought, well, speculation in the market, information asymmetry is what's leading to these big price hikes.

So the G20 in 2011 started a program called GEOGLAM, which ignores the acronym, but just pretend it's called GEOGLAM, glamorous. And it was really designed to create transparent outlooks on production so that everybody kind of had level footing and hopefully that would have a calming effect on the markets.

So that's really when things started speeding up. But really cumulatively as many advances as we saw between say 1972 and 2011 when that G20 program was started, 2011 to 2016 probably had just as many if not many times more. So it was straight up.

NASA data, freely and openly available, other space agencies in the world following suit. Many new types of instruments, and of course an explosion in the commercial space sector. And suddenly we were really in a different position in 2016 all of a sudden.

Mike Toffel:

That's when NASA Harvest took off or in 2016, 2017?

Alyssa Whitcraft:

Yeah, that's right. NASA had, like I said, been working in agriculture since 1972, but it didn't have a coordinated program on agriculture like maybe it had an air quality, or climate modeling, or whatever it might be. And they were like, "Hey, we suddenly have enough data around the world to track how the growing season is progressing." And maybe not just looking at crop production or crop yield, but we could look at long-term land cover, land use change. We could look at start to get toward things like quantifying emissions that are related to agricultural management.

And so NASA was like, "It's time. It's time for us to build a coordinated effort that's really outward facing." It's an applied sciences program, not foundational research in the background anymore.

And so they put out a solicitation, publicly competed. We put in a proposal. I was one of the three people who wrote the proposal together, who ended up being the director Inbal Becker-Reshef, the chief scientist, Chris Justice, and myself as the deputy director and program manager when it was selected. And so that was NASA Harvest.

Mike Toffel:

And you were a team from University of Maryland even at the time?

Alyssa Whitcraft:

That's right. Yeah. So we're actually really unique. This was really the first time NASA had ever placed a program outside of one of its NASA centers. So they have centers all over the country that focus on different things, like Johnson is really famous for the astronauts and things like that. But Kennedy does a lot of the launches in Florida, Jet Propulsion Laboratory in Pasadena works with Caltech and does a lot of interesting stuff too. Goddard Space Flight Center, which is really near University of Maryland, is sort of the main Earth science part for NASA. And because of that, my department at University of Maryland has had a very tight relationship. Geography, all things are related, but closer things are more related. So we've had a very strong remote sensing history because of that proximity.

So yeah, we were really an experiment. And the reason was federal government historically not great at partnering with the private sector, historically not great at transitioning and pivoting quickly, being agile and flexible to respond to emerging agricultural threats. Agriculture is so dynamic in space and time. They really wanted to build a program that was fit for that purpose.

Mike Toffel:

Now, there's so many things that this technology could be used for, and I'm sure we're going to talk about several of them, whether it be predicting the best time to harvest, or predicting crop failures, or thinking about mitigation measures and regenerative age. What was NASA Harvest focused on and how did you decide on that focus?

Alyssa Whitcraft:

Yeah. So we had the history of this sort of long-term condition monitoring, international stuff. That legacy really was continued under NASA Harvest. We were looking at mostly, but not entirely, large scale within season production forecasting. And the main users of our information were market and trade analyses. We supported that G20 program GEOGLAM, and we were coordinating the ministries of agriculture all over the world, space agencies to come together and develop consensus outlooks on what we would expect to have harvested of these big commodity crops that impact prices in food security.

We worked with national governments in implementing satellite data for their national statistics, or in other cases for developing their own internal early warning systems for threats to food security. So obviously we deal with food insecurity and the US in close to 20% of households with children in the US have experienced food insecurity, meaning they don't have consistent access to food. That has nothing to do with production, that's an access and equity issue.

But there are other places in the world where they're so reliant upon what's produced in their immediate vicinity that production really is what can drive food insecurity in those regions. And so satellite data can be used, and we did this in NASA Harvest to sort of identify which areas really needed the most help, and then work with those governments so that we weren't identifying it. So it's theirs. They own the process, they own the ability to analyze and utilize the data for their needs.

Mike Toffel:

Got it. So you're predicting crop volume based on the amount of land that different crops were planted on and how those crops were progressing over the season. Is that how it was working?

Alyssa Whitcraft:

Yeah, I think that's fair. I mean, we have what we call essential agriculture variables, and these are sort of core building blocks that tell us about what the current state is in agriculture, how it's changing over time. And they can feed into our forecast of what the future will be.

And so some examples of these essential agriculture variables could be cropping type mapping. It could be cropping area estimation, it could be crop yield forecasting, it could be water use, and pest and disease pressure and stress.

And so these are things that remote sensing can actually measure from space, which is quite incredible. We do it in tandem with ground-based data to train and validate our methods, so train a model and then see how good it was. But yeah, you're pretty much correct. That's sort of how it works. Yield is the holy grail. Production, I should say, is the holy grail of understanding the outlooks. If you can figure that out, that's a very nice crystal ball to have. But it's also important to do it in such a way that doesn't privilege one set of actors and make vulnerable people more vulnerable.

Mike Toffel:

Right. So thinking about this in terms of just an equation, the X's of an equation, are the inputs coming from your satellites converting the images into variables in a sense, and then your why here is the predicted production of the crops?

Alyssa Whitcraft:

Yeah. And the thing that's nice about satellite data is it's passing over the Earth. The satellites are, a lot of them are polar orbiting it's called. So they're going over the poles over and over again as the Earth turns. And so it's kind of predictable seeing the same part of the surface within a certain cadence.

And so it's not privileging to be one specific part of the world in terms of what it can image, and you get continuous geospatial coverage. You get a map as opposed to point by point by point, which is the traditional way of developing these statistics.

And so you can kind of stitch together multiple variables. There's not just an X in the equation. There's a variety of variables you can look at together and hopefully get toward production. And we get better and better as the season progresses, because the window of possible outcomes closes. But we still many months in advance of harvest can put out some very good predictions, generate some very good predictions of what final yield will be, barring some extreme event like a tornado or something like that. We're pretty strong with it.

Mike Toffel:

So then what decisions were made either by companies, or growers, or by governments based on these predictions?

Alyssa Whitcraft:

So one of my favorite applications, this was led and due really to Dr. Catherine Nakalembe. She's also a professor at University of Maryland and the director of NASA Harvest's Africa program. She's from Uganda, and she was actually, I think she may have still been in her PhD at this point or would have recently finished, and was looking at drought in a region of Uganda called Karamoja. She was seeing on the satellites like, "We're past the point of no return. We're early in the season, but there's no way that this crop is going to turn out, and this is an incredibly vulnerable region to food insecurity."

Being from the region automatically builds more trust with people in the region. And she had a connection. Her sister's the national badminton champion of Uganda, so she had some conduit to getting to the prime minister and said, "You guys, this is coming. We need to pivot in this region." And she showed them the imagery and they were like, "Whoa." And because they were able to have that information earlier, they didn't become reliant upon for an aid to intervene.

Instead, what they did was pay the farmers to abandon those crops. They gave them new seeds to start a new crop. And in the interim, the payments that they give employed those people to do public works projects like building roads or infrastructure.

It saved millions of dollars for them and really boosted their autonomy. They still use this now operationally. And it helped something like 150,000 people. So pretty remarkable. That's one of my favorite kind of stories. So that's really the food security side.

Another cool example within NASA Harvest was supporting loss adjustment for crop insurance companies internationally. So fraud in loss adjustment in insurance claims is a very large, several billion issue in agriculture. And so a farmer may make a claim and then an adjuster will go out and they may be frauded themselves as well, but they'll go out and they'll be like, "Yep, that crop's gone, or it's this or that," and then they process the payment.

It's really costly. It takes a lot of time. And in some cases, you can't even get to the place you're supposed to view because of infrastructure issues. And this is all over the world. I'm not talking specifically about the US.

And so there was a partnership between a startup company called GreenTriangle, Swiss Re, and NASA Harvest, and we worked together to optimize both within a field where they had to visit how many crop cuts, they had to take to estimate how much was lost.

Then the regional driving plan, effectively that took into account. It was basically optimization between locations to both be able to get a field level verification, but then also a regional level estimate of what actually is going on in that country.

For us, it was great because it helped us improve our yield models and also understand from a training data perspective, we don't have to go out and take 10 points. Now we have the benefit of knowing we only need two or three, and we can use satellite data to determine where in the field that we would take those points, those samples. It helped build up the business case for GreenTriangle services as an Earth observations analytics company, and then it helped Swiss Re reduce risk from fraud and speed up and make more efficient their entire loss adjustment process.

Mike Toffel:

Wow. Now, are these companies contracting with NASA Harvest to access the data, or is this data feed available to anyone who wants it?

Alyssa Whitcraft:

So the imagery that makes it back to the satellite or the reflectance that makes it back to the satellite, that's free for everybody to download. There's a lot of mileage, so to speak, between these raw numbers, these raw values. They're not even terrain corrected at that point. You haven't even tied them to Earth. NASA and these other agencies will put out higher level products like surface reflectance. They'll convert it. So it's actually, okay, this is really what the surface is and not the atmosphere, and here's exactly where it is on Earth.

But you can move past that next to NDVI normalized difference vegetation index. It's kind of like the workhorse in a lot of ways. That's a really good indicator of vegetation condition, but it's an index. It's not a quantity.

So then if you want to get next to yield or things like that, that's where things start to transition away from freely and openly available. So NDVI and before is freely and openly available. Land cover maps can be freely openly available as well, some crop type stuff as well. But then when you move toward the things that can really move markets like yield, or information on tillage intensity, or cover crop utilization, things like that, there's a lot of power and a lot of money behind that. And that's typically not the sort of data that we immediately push out, even though we can do those analyses.

Mike Toffel:

And so are those types of analyses, that last mile analyses, are they being done by NASA and by private sector partners or one or the other?

Alyssa Whitcraft:

Certainly. Certainly, yeah. You've got NASA Harvest and other parts of NASA who have for years built up the scientific foundation, have built up the quality of the imagery, so that we call it sometimes analysis ready data. Meaning that let's demystify, let's do as much terrain correction, atmospheric adjustments so that people don't have to do 10 years of advanced radiometric equation building to utilize the satellite imagery. We can do that on the back end. The agencies can.

But the actual turning it into information, we do some of that in NASA Acres because we are an applied sciences program. We're aiming to apply the data to answer a problem. But then there's also commercial service providing companies who want to monetize that moving forward. We sometimes work with them so that their business, their science is as strong as possible. The quality of their business tool is as good as possible. And then we also, there's in other cases, some of the big actors in ag, like the really multi-billion dollar plus conglomerates in agriculture, they either require those service companies, they use those service companies.

Mike Toffel:

Got it. Now, you just mentioned NASA Acres. And NASA Acres and Harvest, they're the two other organizations I wanted to ask you about as built on the back of NASA Harvest, which continues right to this day, NASA Harvest.

Alyssa Whitcraft:

Yes. Yeah. I mean, it was an initial five-year period from 2017 to 2022, and it was just so wildly successful. I mean, we made so much impact. I just gave you two cases, but there were many, many more that NASA made not unprecedented, but very uncommon decision to continue NASA Harvest for another five years without competition.

And then they were so pleased with it that they said, "Well, we don't want to stick to the US and the international work in one program anymore. We want a whole US focused program, and we want NASA Harvest to continue as the global international program." And so in 2022, just like they had in 2017 for NASA Harvest openly competed, anybody could apply to develop a US focused ag consortium for NASA. And I put in a proposal this time as the principal investigator, what became the executive director once it was selected, thankfully. And we started just last year in 2023.

Mike Toffel:

And that's called?

Alyssa Whitcraft:

NASA Acres, because what's more American than Acres? This challenging system of measurement.

Mike Toffel:

Acres is the US version and hectares is the global-

Mike Toffel:

So then what does NASA Acres do?

Alyssa Whitcraft:

.

So when NASA Acres came around, we really had the opportunity to use finer resolution satellite data. There was even more available. There were more instruments available that were interoperable. Meaning, for example, NASA had two Landsat missions. Landsat was that very first satellite that was launched in 1972, and we suddenly had Landsat 8 and Landsat 9 launched together or operating at the same time by 2022. And then Europe had launched two similar ones called Sentinel-2. Together, those four satellites imagined the whole Earth every two and a half to three and a half days.

So that's really enough to see how crops are progressing. And we're talking about being able to see a football field of space or 10 times finer than that depending on which of those two satellites you're talking about. So whole huge amounts of information that are suddenly farm relevant.

So NASA Acres is not just looking at the national statistics piece of the US. It's really principally trying to deliver value of satellite data for on-farm decisions. So really working hand in hand with farmers by necessity to help them come up with irrigation decisions or early warning of pest and disease infestations, side dressing recommendations for nitrogen within season. Those are just a handful of projects that we have.

Mike Toffel:

So this is next, really you can see the next generation of technology. Whereas the first generation, as we talked about earlier, it was alerting governments to potential crop failures where they had to take action to perhaps begin the import process earlier or think about alternative strategies. Here, given the resolution and the more frequent readings, you're actually able to intervene, and give advice and input to farmers themselves about how to improve their yields in that moment in real time to improve that current season's output.

Alyssa Whitcraft:

Yeah, it's kind of like the granularity of the mechanism or the location of action. So yeah, policymakers can pivot within season and make different decisions. But now we're talking about farmers even changing the trajectory of their own operation.

And we're not the first people to work with farmers with satellite data. There has been this commercial space sector that's been launching these satellites, but with sort of limited success for a variety of reasons.

And so now that we have the very high quality data that NASA collects, and pre-processes, and makes available, and European Space Agency, and Japanese Aerospace, and Canadian Space Agency, all these agencies working together, this data is more reliable for a lot of different kind of decisions that people want to make because the readings are, what you see is what it is. You're not getting some relic of the measurement that has to do with where the sun was or how high quality the camera was that took the instrument.

Mike Toffel:

Sure. Okay. So we've talked about NASA Harvest and then we've talked about NASA acres, and then there's yet one more organization you're affiliated with, which is Harvest SARA, and I think that's one's more private sector oriented, dealing with big names like General Mills and Unilever. How is that different from the others?

Alyssa Whitcraft:

Yeah, so one huge way it's different is that it's operated as a 501(c)(3), and that's important for a couple of reasons. I talked about how the whole satellite space changed. There's more data, there's also more cloud computing, this sort of data democratization and also the utilization, democratization as well. But what about, that's the supply side. What about the demand side? Who's asking for the decisions?

And that was a real pivot and shift we saw 2019, 2020, 2021, which was this push toward nature-based solutions. And we had companies, both the companies who built the models that were going to try and verify that practices had been adopted or quantify how much carbon had been offset. Or for companies like General Mills who were looking to do in setting. Which means within their supply chain, they're trying to reach net-zero, but within their own supply chain as opposed to buying credits from oil and gas buys agricultural credits. That's not within their supply chain, but it is within food companies.

And so they were coming to us. The companies that were doing the verification were coming to us and saying, "Hey, can you validate our model, or can you be the person who verifies that we've undertaken this practice or that our supply chain has changed in this way?"

And I think we felt, NASA Harvest leadership, that was not NASA's role. I mean for NASA to be a verifier or put out an official decision and say, "This is what's really happened," is really not what NASA's role is and not why NASA created NASA Acres and NASA Harvest.

Mike Toffel:

All of this is talking about farmer practices now. We're talking about are they engaging in regenerative practices, for example, no-till, or crop cover, or so on. So it is very different from crop failure and crop productivity.

Alyssa Whitcraft:

Absolutely. I mean, in the public space, I think the awareness of, and also the belief in climate change grew. And I think a lot of us also realized there was this failure of the public policy space to do action about it. And so a lot of these companies and individual consumers stepped up to be like, "Well, how can I make different day-to-day decisions?" Obviously food is one of the biggest ones of them.

And so thinking about how you can make those different decisions, the whole supply chain was implicated. The whole agricultural value chain was implicated. But the agricultural bit of land is one of the biggest kind of places where they want to see this change take place. They want to see adoption of no-till, because that keeps soil, or carbon in the soil in the ground. They want to see the life cycle analysis of, they call that the cradle to the grave of a food commodity. So it's not just that you've reduced tillage or whatever. It's like you also get to account for the fact that you didn't use as much fuel in driving your tractor to do the tillage.

And so there was so much both in terms of verifying the practices that they took place, but then also being able to relate those practices to an actual mitigation outcome. So there were two camps of thinking about this. One was like, well, why belabor the point of having the right amount of carbon sequestered in the soil, when instead we should really just be incentivizing these practices because they're good for a bunch of other things? Biodiversity, resilience potentially to extreme weather events.

And then there was another camp that was like, "No, we're trying to mitigate climate change. Our commitments are to net-zero, not to just generally supporting agriculture. So we need to quantify carbon removed."

And so there was a real vacuum. There was a real lack of a public neutral actor who could intervene and say, "This is actually how well all of our collective ability to model an agricultural system, this is kind of how well it's doing."

We knew we had a role to play there in improving, again, the science. Lifting up the private sector, lifting up the businesses to make the best possible use of the satellite data, but to not do it as NASA. And so that's why we started Harvest SARA as a non-profit. People give gifts. It's totally transparent. There's no federal government oversight. Similarly, the donors don't exercise control over our results.

Mike Toffel:

So now what are you doing with the science? Are you publishing this in peer-reviewed journals? Are you trying to influence standards that are out there that come up with equations that say if you use this much cover crop and do this much no-tilling, it yields this much carbon abatement?

Alyssa Whitcraft:

So we're not really trying to create our own model or create our own measurement system because even through Harvest SARA, NASA Harvest, or NASA Acres, we're not service providers. We're here building the infrastructure and lifting the floor so everybody can do their jobs better.

So one of the things that we're doing is we've convened public sector modelers as well as private sector companies that have monetized these models or developed new ones, and soil scientists, and machine learning and AI experts and computer scientists who build different systems for processing data. We've convened them together to do a multi-model evaluation.

So the idea is we have 20 models participating that are named. We know who they are. But when we spit out the results and we do this analysis against real ground truth data of what really did change with soil carbon or what practice was really adopted, we will say model one performed like this, but nobody knows who model one is.

The modeler does. They get their results, they know where they're doing well, they know where they're not doing well. We can say the best model was this accurate, the worst model was this accurate. They can go out further to Unilever, to General Mills, to PepsiCo, to Nestle, whomever, the people who are wanting to purchase these services or these credits and say, "My model was the best one and here I can prove it."

So again, we're facilitating it, but it's such a delicate dance because none of those people will participate if they perceive us as attempting to do harm. We just want the actual measurement system to be representing the real thing so that we're not playing climate change whack-a-mole basically.

Mike Toffel:

Yeah, it reminds me of consulting organizations who convince a bunch of players in a particular industry to participate in a benchmarking study so that you can see how good the best is and where you fit relative to those. But you're not a report that names names. It's just a report that says companies A through F, and here, I guess models one through 20.

Alyssa Whitcraft:

Right. And I think one of, my interest in doing that, because when we started SARA, a big piece was, I called it building the missing evidence base. So some farmers are like, "Well, I'm interested in adopting this practice, but what is it going to do to my yield?

And so because satellite data can do these very large analyses, our knowledge is built on the field of studies. The models are often built on the field studies, but we can implement them over very large scales over very many years, and build a more robust evidence base just by virtue of how many data points you have in front of you.

But then the other question is, all right, carbon sequestration is a big thing. Reduced nitrogen emissions, big thing. So how do we look at that as one of the outcomes that we're building an evidence-based for? Well better figure out which model is the one to use for that. And that's kind of organically how we came to the benchmarking activity, the model evaluation.

Mike Toffel:

Very interesting. Okay, so I know you've done a lot of work on data governance in this space. Can you talk a little bit about what the issues are there and what progress we're seeing?

Alyssa Whitcraft:

Yeah, for sure. So we've got this rich history of satellite data for agriculture. And we've got a lot of conversations with different farmers we've had where they're like, "I tried satellite data once in this private company and it was useless, therefore it's useless," So there's a reputation issue to begin with.

And I come from a viticulture family, so my dad and my brother, my dad's passed now, but he was a winemaker and my brother's a winemaker, and I spent a lot of time in vineyards and things like that growing up. So I had sort of an innate understanding of how we can feel about outsiders coming in and saying, "You can't do this, you shouldn't do this." And so I had this kind of awareness that if we were going to make any headway with farmers, we needed to effectively build trust with them. I call it our trust infrastructure.

And another piece of it is going out and meeting people where they are. NASA does a ton of space for ag tours, and the NASA leadership comes. They're trying to make sure that the farmers know that, "We're your space agency," which really is true. We want to do the things that bring the most value to farmers in the country. So that's two pieces. Getting to know them, building face time, and building the legal agreements and frameworks that allow us to exercise whatever ethics we talk about.

And then we also have this computational infrastructure that we're building that allows farmers to share their data in secure ways and have it kind of interact with other farmers' data, and interact with satellite data, and different models from different sources without them losing control over the data or having their personal information, or even the location of their farm revealed.

Mike Toffel:

Now what data are they providing? Because the satellites of course are taking pictures you mentioned at the sub football field level of crops. They're assessing crop health. They know the weather so they know how much water, at least how much rainfall has come on those crops. What is it? So that's what NASA has from the satellites. And then there's all these layers of algorithms. What is it that the farmers are providing? Are they providing more micro detail that you can't see in agriculture, like the irrigation amounts and the fertilizer amounts, things like that?

Alyssa Whitcraft:

So agriculture is one of the most data-rich fields. And all satellite data applications, even the kind of getting us down to surface reflectance, require calibration or sometimes called ground truth. I don't like that term, because what even is truth? But an all-measurement has error. But you're right, we're tying the satellite observation to what is as close to real as we can on the ground.

And we know typically if you don't train it on the real world, you're not going to get a very good result and you certainly can't validate it or show that it compares well with the real world. And so if we want to measure yield, a satellite data observation itself isn't enough. We have to train it with yield statistics or with a yield monitor from a field. I can't train a subfield level yield model row by row on national yield statistics. It doesn't work that way. You've got to train for the scale you have.

If we're talking about this many bushels per acre, we feed that to the satellite and say, "Hey, satellite, this is what that looks like. Now go to some place where you don't have data. Make a prediction based on what you've seen before." And so farmers collect a ton of data. But being able to build predictive capacity or generalizable requires the fusion with satellite data, and that's our value proposition for them.

Mike Toffel:

So what I think I hear you saying is you're getting their actual yields in order for you to tune or calibrate your models, because you're trying to predict yield. But in order to predict yield, you have to have at least some portion of the data where you know the actual yield in order to generate the say coefficients on the X's in simple language. And then also, they could give you more X's like things that you can't observe from satellites. So is that the right way to think about it?

Alyssa Whitcraft:

I mean, it's a great way to think about it. I had a farmer ask me, "So great, I collect a ton of data. What value do you bring to me?" And it's a really important question to ask. She was being facetious because she's a big believer in innovation, but it was nevertheless an important question.

Because one example is say they give us 20 years of yield data. They already know that, but they're not asking the question about the past 20 years. They're going to harvest their corn in the end of September, middle of October. It depends where they are. They have not necessarily a way in June or July of knowing what they can expect. But because the satellite data has been collected, collected, and we've trained it for so many years, now as long as you're not having some extreme drought here that the model has never seen, we can predict things well before they have that indication. And so that's one example.

Another example is said we're trying to look at the interplay between no-till or reduced tillage and yield. Well, they might know. Well, maybe they've never even adopted it on their farm, or maybe they've adopted one type of it.

Well, are they performing as well or not as well as other neighbors for their soil type, for their operation? Does it make sense for them to even try this practice? That's another value of satellite data, is just almost the scale and data science piece of it.

Mike Toffel:

So how does climate change fit into all this? Because normally one thinks of the predictive accuracy of these algorithms as requiring somewhat of a stable system. But as climate change is injecting more variation in long-term trends, it's not really a stable system anymore. So how is that affecting the work?

Alyssa Whitcraft:

Yeah, it's a great point. We have examples that have already happened that demonstrate what we might anticipate with climate change. So for example, the derecho in 2020 in Iowa, which was this extreme high wind event that suddenly happened, and destroyed so much crops like this. That was not super typical. Now we know what that looks like because it happened, and our model is a little bit more familiar with that now. One of the questions that we've been, or I've been really wanting to answer, but it's a question of having the data to train the model, is where the people who had already adopted reduced tillage or cover crop, were they less likely to have lost their crops in the derecho? Things like that. So that's sort of a climate change adaptation way of looking at things.

But you're not wrong that we're going to see things introduced that we haven't seen before. And this kind of spins us back nicely in a way to the AI topic, because one of the fears is that you'll build a model that is so trained that it can predict anything in the future. But that's really not going to happen for a variety of statistical reasons in the way models are developed. But also just because the features themselves, the system itself is so variable.

I'll tell you that we've developed this yield forecasting model that works very well at county scale, but it worked poorly in 2012. And that is because in 2012, it was a drought year that we've never seen before or since. And so the model doesn't know how to deal with it because it's not seen it before. So that certainly will be complicating things for us in the future, but nevertheless, we persist.

Mike Toffel:

Yeah, yeah. Interesting. So I know privacy is another area. We started touching on this a bit with some farmers being reluctant to share data or wanting to make sure there's quid pro quo. That if they're going to share the data, the value they get back is at least as valuable as the value that they're donating to the project. And I know that there's not the privacy standards yet in this space that there are, for example, healthcare with personally identified data. Where are we in that privacy journey in this space?

Alyssa Whitcraft:

Yeah, so you're right. I mean, data ethics with HIPAA and healthcare, these things are pretty well described. So we have sort of the traditional data ethics model that's like minimize collection of data, really strongly limit the use. Protection is more important than anything. So it's like a leak-proof container where it sits, and then really clarify who owns the data and who has the right of revocation.

In ag data, in talking about farm data. I'm not talking about satellite observations of ag yet. There was an initiative started in 2014 called Ag Data Transparent, who we've worked with, and they've kind of tried to adapt farmer concerns about sharing their data with John Deere even, not talking about satellite data or with whomever it is. What's happening with my data and what should the principles be?

Mike Toffel:

And that's where that trucks were collecting data and then sending it to the cloud.

Alyssa Whitcraft:

Yes, exactly. And it's like, whose data is that now? And can the farmer recall it and say it's theirs? These questions were not there. That's kind of where our principles that I talked about earlier, they were an adaptation of ag data transparency principles for our cases.

And so then they kind of tried to create an ethical framework without naming it as such to deal with balancing. And last year or two years ago, the National Space Council was like, "We should maybe think about this." And they brought in an ethicist named Patrick Lin from Cal Poly who led a kind of fact finding mission about, how should we be thinking about ethics for space data? And I was one of the people interviewed for that, and talking about what I had learned in working with ag data and farmers.

And so they've put out this, or Patrick has put out this white paper to the National Space Council recommending that they develop a new ethical framework for understanding space data. We should be maximizing collection. There's really good reasons. But as NASA Harvests have shown, NASA Acres have shown, we need to be thinking about unintended consequences. What if we verify a practice has happened and it hasn't? What if somebody said, "I did no-till," and we're like, "Nope. Our model says no." Uh-oh. Things like that. Or we develop a new model working with a private sector that involves farm data, and then the private sector sells it back to the people who share their farm data to begin with.

So there's all these kinds of complications that come when you combine data from multiple sources, especially when those sources don't share the same framework for understanding the ethics that underpin the data.

Mike Toffel:

Yeah. I mean, one of the things that strikes me as particularly interesting with satellite data, it's unlike, I don't know, phone data or car tracking data. You can't opt out of satellites observing your property, for example.

Alyssa Whitcraft:

True. Yeah. It's interesting because we meet farmers. We did a survey through Farm Journal of just over 1,000 farmers and less than one in three was aware NASA did anything in agriculture. And among those who were aware, and even among those who weren't aware, they're like, "You can see our license plates since the 1960s, and we know it." And I'm like, I mean we can't. At least I can't. I don't have access to that kind of information. So typically what we're seeing is much, much, much coarser.

But like I said, because our ability to transform the satellite pictures into information is contingent upon the training data that comes from this other ethical framework, this ag data that's collected on farm, that's why whether or not we can collect the data without their opting in or opting out, our ability to transform it into actual information is contingent upon some farmers participating. Now for the farmers who aren't participating, and we can make a prediction on their field, my investment in my program has been like, okay, we need to make a commitment to not sharing that as well. So even if we can do field scale analysis or subfield even, and we can in a lot of cases, we need to not be publishing that.

Mike Toffel:

Interesting.

Alyssa Whitcraft:

For the people who have not participated in the study, they'll get their stuff back. But I'm not going to publish a super high resolution yield forecast of people. It would, I mean marketing is one of the most stressful things for farmers to deal with. I'm not going to make that harder and worse for them. I want to do as much benefit and as little harm as possible. And that's what kind of motivates this whole data governance and trust infrastructure effort.

Mike Toffel:

Right. And clearly, I mean, that makes perfect sense from the roles you've had in government and the roles you've had in this 501(c)(3) or a non-profit, or NGO. Commercial providers in this space may have a different set of motives that might in fact encourage them to publish that or at least act on that information. For example, it could influence the valuation of properties.

Alyssa Whitcraft:

There are companies out there that's their business. They use satellite data to look at the past, say productivity and use that for kind of differential rents or setting prices for land. I mean, that's a thing. So we have this Farm Innovation Ambassador Team, FIAT. And it's a place-based national test bed research site. So the farmers contribute their farm to answer the questions they want answered. So it's like our most direct conduit to farmers.

But we also use this team to ask questions about, "Okay, so we can do these county scale yield forecasts. What should we do with them? What should happen with that information next?" It's kind of almost like an advisory board.

The farmers that we talk with have no issue with commercialization. They have no issue with paying for services. I mean, of course there's anomalies, but most people understand you get a service, you pay for a service, and that's fine.

What they don't want to do is have this kind of bait and switch where they're considered as these collaborators, as this partner in innovation. And then it turns out they actually have no rights. And so in that same survey with the farm journal that was just over 1,000 farmers, we asked about incentives for participating in research studies, and a very large amount of respondents were interested in either having some stake of equity stake or royalties that came from their participation if something were commercialized or were also very happy with the service that was being developed, provided to them for free in perpetuity.

Mike Toffel:

Yeah, it makes sense. So let me pivot and look forward, and ask you really two questions. I'll just ask them to you at the same time, and you can address them however you like.

So one is, what are some new trends that you expect in this space, in the overlap between satellites, machine learning, AI, and agriculture? And then also, how can people get engaged in this space? For those listeners who think, "Wow, this is really interesting, I want to be the next Alyssa," or at least work on a team that somehow interfaces with the same issues that you have, how do you imagine they should get started?

Alyssa Whitcraft:

Well, I guess I'll tackle the second question first. I get to push something toward a vision that I have. The challenges being that ultimately I don't sign the data use agreements, even for myself. My university signs them. I have subawards at different universities around the country. I can't enforce anything that they do.

And so I'm going out and I'm trying to set forth this vision and these standards, and I'm sharing them. I'm educating our research partners and all this kind of stuff. I'm scared also that people are not going to get on board. You only get one chance to break trust, right?

Mike Toffel:

Yeah.

Alyssa Whitcraft:

But that said, if people are really interested in innovating in the public sector, in partnership with the private sector or with the vision of being like, "Yeah, let's think a little bit more carefully about this open science and transition it outward," I do think I'm going to put a plug in here for transdisciplinary research. And not just transdisciplinary.

I mean show up in the places where you're trying to work. Go on the farm and ask the question. Every single time I go to a farm, some assumption I had was revealed to me to be totally false. Even things that I had never surfaced, like I didn't realize that's how that worked.

So don't hold yourself up in an Earth observation lab. There's nothing wrong with that, by the way, but that's just going to lead you to the type of work I do. And you've got to have those people in the labs.

But go out and be there. Be with the people you want to work with. It doesn't matter if it's farming, it doesn't matter if it's healthcare, it doesn't matter if it's forestry, water resource management, whatever it is. You got to go out and meet the entire chain of people who are implicated in the set of decisions, really from day one.

Mike Toffel:

Got it. So that sounds like advice for those in the sciences. And for those in the business management area, it sounds like maybe the Harvest SARA piece where you're working with these private sector companies. Either buyers of carbon credits or developers of carbon credits, it seems like there's a lot of opportunity in that space as well.

Alyssa Whitcraft:

Yeah. I mean, it's hard, right? But my belief is that they're going to have a more successful business if they spend a larger amount of time on quality. And that is our value proposition. That is one of our corners of the market. The other is that we're doing things for everybody's benefit and not for just one company.

And I would encourage the business side, partnering with people who are invested in the quality and knowing that that might slow things a tiny bit. But I genuinely believe, based on everything I've seen working with VCs and tech companies the past 10 years, that it's going to have a longitudinal benefit. You're not going to dry up in a year or two, or implode, or break trust. You're actually going to be delivering the benefit that you envision, that you want to see happen.

Mike Toffel:

Yeah. And then as far as future opportunities, where do you think? What are you seeing?

Alyssa Whitcraft:

Yeah, so I feel like this Farm Innovation Ambassador Team is, I'm so jazzed on it as a concept because it firmly destroys that wall that we have put up between science and practice in the agricultural space. Farmers are business people. They're also incredible data collectors. They have done so much empirical research in their own operations. We have just missed the boat on collaborating with them.

And so building up this as an actual team, it's Farm Innovation Ambassador Team, these are people working together, I think has a very powerful opportunity to advance this entire field, and finally deliver on this 50+ year promise of satellite Earth observations for agriculture in the farm context.

Mike Toffel:

Great. Great. Well, I've learned a ton from our conversation, and I'm sure our listeners have too. Thank you so much for spending time with us here on Climate Rising.

Alyssa Whitcraft:

Oh, thank you so much for having me. It was a pleasure.