What if you could measure water quality from pixels on an aerial image? This episode is about the emerging geospatial analytics field that combines data-hungry algorithms and AI with remote sensing data. This technology can change how we identify, measure, and solve challenging environmental problems.
We talk with the Sean Donegan, President & Chief Executive Officer of Satelytics. Learn more about Satelytics here.
Transcript
Robert Osborne:
Welcome to the Outfall where we share the back stories about our water world. This is Robert. What if you could measure water quality from pixels on an aerial image? Maybe this image was taken from a drone or a satellite, but somehow there was a way you could figure out how much, for example, phosphorus was in the water without even measuring it on site. Crazy, not possible, well that’s exactly what I thought a few years ago. I was wrong. This episode is about the emerging geospatial analytics field that combines data, hungry algorithms and AI with remote sensing data. Watch out, this technology can change how we identify, measure, and solve challenging environmental problems. We talk with the CEO of Satelytics.
Sean Donegan:
My name is Sean Donegan or until I was 11, I used to go by the name of come here you little idiot. And I am the CEO and President or President and CEO of a company called Satelytics based in Perrysburg Ohio or as we like to call it the 751st most visited city in the nation.
Robert Osborne:
Like all good ideas, Sean’s idea for this software company started with caffeine and a napkin.
Sean Donegan:
We literally were visiting Google and we were in a Starbucks and I was drawing it on a napkin. They said, “No, I don’t know if we can do that.” And I said, “Yeah, but we’ve never tried. We got to try.”
Robert Osborne:
So, what was Sean’s idea? It was an algorithm born in a sense out of algae blooms in Lake Erie. Now the term algorithm gets tossed around a bunch, but it really is just a set of steps in a computer program to accomplish a task.
Sean Donegan:
So, it is a software company that develops algorithms, and we wrap our algorithms around a software wrapper if you will, that sits exclusively on the cloud. The secret sauce are the algorithms and the data that they use. So, we use a genre of technology called Geospatial Analytics. Basically it means the sun shines on the Earth’s surface or a body of water to a depth of 18 inches, the so-called column of water, and what we are looking at is the infrared spectra. So, near infrared, shortwave infrared, and we’re looking at the reflectance signature of many of the constituents. And we started life looking at Lake Erie for toxic alga blooms or as you have to say it more wokely today, harmful algae blooms. The three first algorithms that we developed were phosphorus, phycocyanin, and chlorophyll. Now, not only did we want to identify them, but as a company, we wanted to really push the envelope. We wanted to be able to quantify those from above the earth surface.
And the second part of that equation is we always want, and we still are today, so we’ve proven it out but we wanted to be in the forward operating area of the industries we served with the explicit goal of really three major minimizing events if you will. We wanted to minimize the event itself if we could, we wanted to minimize any of the remediation efforts, because they escalate very quickly. And then the third thing industries we serve, they’re often under the scrutiny of NGOs and regulators and the press. So, anything that we could keep small and confined would have less outcomes that would be distasteful for all involved, quite frankly, including the environment. Today, we have 40 different algorithms. They can all run at the same time. At the heart of our algorithms is a convolutional neuro network or a very fancy word for artificial intelligence. And we develop three to four new algorithms every year and it’s just gone from strength to strength.
Typically, the data is coming from satellite because of the goal that I set out just a few minutes ago, we want to give you the results within three to four hours of collecting the data. We’re 40 people, the backgrounds are typical technology, so artificial intelligence, software development, but the key and the core to our science is geology. So, we have some of the finest, young, brilliant minds. And of course I take all the glory and do none of the work. And they really are at the leading edge of developing these algorithms, looking that infrared spectra. So, for example, when we started life on Lake Erie and Toledo has the auspicious label of shutting down its water supply in 2014, 2015, that was caused by harmful algae blooms. Now the key is those three algorithms, if you are going to get to quantification, the challenge is how do you build a robust model that when you look at that area, you know that is 10 parts per billion?
And today our phosphorous algorithm, one of our most mature, we know that we’re accurate to plus or minus six parts per billion, wherever you apply that in the world. So, when we first go out, particularly water it’s very challenging because it moves. So, we take a boat out onto Lake Erie and what we’re looking for and Robert, you’ll recollect this, what we’re looking for are we’re looking for measurements that are in the range from low through the medium to the high ranges of the values we’d expect of that constituent. And we grab samples, physical samples, they are shocked with a backpack spectrometer at the same time as a satellite that we’ve tasked has over passed. So, that all has to be very uniquely lined up. The samples are sent to an independent laboratory, like a Pace or an Alloway. And when those measurements come back, we calibrate back the algorithm and what we’re seeing in the backpack spectrometer with the physical measurements.
And then it’s all about taking more and more data in those medium, low, and high range values and training the synthetic data sets from this convolutional neuro network to understand that pattern is that measurement, if that makes sense. Land is not quite as delicate because you don’t have to have the exact overpass of the satellite when you grab the soil sample, but you don’t want there to be months. You want to be in close proximity, even up to a couple of hours, but water is the critical one. And then once you’ve run that algorithm, you start inhaling data of other water bodies, tributaries from around the world. The AI develops very quickly. Today that algorithm for phosphorus is in the 98.65 percentile of accuracy.
Robert Osborne:
Wow. Yeah, so water I guess I could believe it maybe a little more just because it seems more uniform and you can get down to the 18 inch depth. But land just seems messier because you would have grass growing and you couldn’t really penetrate down. So, tell us a little more about that. How can it look at a chemical on the land?
Sean Donegan:
Well, it doesn’t penetrate the ground. We are looking at the surface and in the body of water and so-called column of water to a depth of 18 inches. But there are many surrogates that we use. So for example, a pipeline leaking under the ground, drip, drip, drip, oil. Vegetation becomes our canary down the mine. Vegetation is one of the most sensitive reactors, if you will. And we’re looking for not just direct signatures of liquid hydrocarbon in that instance, but we’re also looking for corroborators or surrogates. Most of what Satelytics is asked to do is to say, “Where are those concentrations and how bad are they?” Or “Have we done a good job remediation?” So, we’re not doing the work, we’re telling you where you need to concentrate your programs to rehabilitate that area or remediate that area.
Amy Anderson:
So, you helped me better understand the how and something that occurred to me when I first heard Robert talk about this is what are the limits as we look at you have your model, you have your constraints, what are you hoping to bring next? And at what point will you have exhausted your options?
Sean Donegan:
First of all, Satelytics is a company when we first started, there were a lot of disbelievers and you have to find visionaries, BP, duke energy. And you have to kiss a lot of frogs before you find the visionaries. And with a small company at the time, very few of us, we were the needy, not the greedy. So, we wanted people to listen to us, but we quickly got over that when we met BP, who said, we like what you did on Lake Erie, we’ve got 25,000 bodies of water on the north slope of Alaska. And we’d like to see if you could detect arsenic, barium, iron, manganese, and copper. And we met a very influential person, to this day he retired from BPR for 40 years and he’s a full-time grandpa and now he works part-time on our science team, Dr. Jim Chatham.
And he’s a geochemist and he’s extraordinary because he could see the vision of what we had. So, the challenges are like all other businesses number one, finding really high quality people. We want to be at the top end of the spectra, if you will, the top end of the tree for delivering high quality product, very high results, very high accuracies in our measurement. But we do depend on third party data. So, today we task satellites from companies like Airbus, Maxar TripleSat, Satellogic, Planet Labs. The amazing thing is in the early days, there wasn’t that availability of satellite when we first started. So, we had to use both aircraft and satellite. Today, the amount of money that’s being spent above the Earth’s surface is staggering. Today, you can revisit anywhere in the world four to five times a week with a satellite. By 2026, there’s so much money being spent that could be up to every couple of minutes.
Robert Osborne:
David, Amy, and I, we come to you. We just had a think tank session over coffee and we said, man, let’s call Sean. We think there’s some titanium in the hills here. In the Hills of Clemson, tell us the steps and how long it would take before you could get something back to us?
Sean Donegan:
So, first of all, we’d have to identify is there a spectral signature for that constituent that you’re looking for titanium? And if there isn’t, is there a surrogate that bonds to it? Is there something that would give us a propensity to say, look, if this exists, then we can assume that. So for example, arsenic is a good indicator of gold. So, most recently before the whole Russian Ukraine element that blew up, there were several Russian companies that wanted us to look for arsenic concentrations for the very same reason. So, if there is a spectral signature, then what we would do is we would again, if you just want a presence rather than a value or a measurement because to get to measurement, we’d have to do the same sampling episode or have some idea of quantities of titanium and what that spectral signature and concentration look like.
So, that process would take three to four months. And then at the end of that, three to four months, we may well say to you, “Let’s run a test, run on an area that you know where there’s titanium. Don’t tell us, just give us a big area.” We’ll task a satellite, we’ll run the algorithm, and we’ll give you the results. And as you get the results, what we need from you is feedback to say “Yes, that was accurate, that wasn’t” because that then starts to train the AI. Within three captures, you’re going to be above the 75%, 80% of accuracy and detection.
Robert Osborne:
We have touched on some of what Satelytics can analyze. If you want to see the list of constituents that Satelytics can analyze today, click the link in the show notes. It’s impressive. We want to thank Sean for joining us. I love his passion. I can’t wait to see how this technology evolves. Thanks again, for listening to the Outfall. If you enjoy our podcast, please subscribe and share the podcast with your friends. See you next time.
