Steve Spittle | From Space to Sustainability: Harnessing Satellite Data for a Greener Future

Welcome back everyone to Mastering Tech Growth. Today we have a special treat. We have Steve coming from a company called Global Trust where they use satellite data to help business make better decisions. It's an award-winning company and award-winning CTO with us today. Welcome Steve.

Hi, hi Mike. Thank you very much for inviting me onto the podcast today. Very excited to have a conversation with you.

I'm very excited as well. You probably can hear it in my voice. I just want to go straight to the question. So let us know, Steve, who you are, how you ended up being in a technology business, especially satellite business, and just your journey so far.

Yeah, absolutely. Um, it's like all journeys, quite a sort of random, convoluted story, to be honest. Um, It started with my undergraduate degree, which I did environmental science. And within that degree, to be honest with you, I really focused on two key areas. I focused on meteorology and volcanology. So the study of weather and volcanoes, not necessarily together, but I always wanted to go into weather forecasting and actually to do the weather forecasting that then uses a lot of satellite technologies within itself as well. A lot of remote sensing, a lot of data analytics. But as part of that degree, I ended up doing a year in industry. And in that year in industry, I actually did that in the space sector. And I started to get a lot closer to satellites that take images of the Earth, or what we call Earth observation satellites. And that really kind of led me to sort of start to open my eyes to the power of what these different sensors and instruments could do for us. And the fact that they were taking these global images and we were able to see targets on the ground down to trees or even down to road markings, for example. And I could really start to sort of have an idea about what that might mean for the future and not just some of the kind of meteorological weather forecasting pieces, but actually what that might mean for the environment, for monitoring of communities. and it really kind of drove me then to look further into the industry. And I ended up at a company called the Satellite Applications Catapult. I ended up being there for almost 8 years. And while I was there, that was all around trying to evangelize the technology, really trying to grow the industry around us, let people know what the capability of space and satellites are. And I started to really specialize into two domains. One was into emerging technologies. So particularly trying to understand what things like artificial intelligence might mean for satellites in space more generally, what technologies like quantum or new types of visualization or new types of security protocols might mean then for the space sector. And then the other part of the job that I was doing there was really focused around working with the mining community and trying to understand what their user requirements were and how we could address those from space. And through doing that for quite a long period of time, I could kind of see this shift within their thought process, which was actually moving away from how can we just help our bottom line to actually we should be focusing on more of the environmental impact and the social impacts and giving more back then to the community and to the world around us. And so those kind of two bits of the job kind of coalesced together and I ended up creating Global Trust with my co-founder Richard Hilton about 4 years ago. I think this journey started. The company itself, I think, is close to about 3.5 years old now.

That's an amazing journey, Steve. By the way, do you mind I'm calling you Steve? Because the name is Steve, but I have a friend, so it's just my A habit.

No, that's— Steve is absolutely fine as long as it's not Stephen, otherwise I feel like I'm being told off.

So, all right, we'll keep it friendly, Stephen. So that's, um, definite innovation part of technology strategies in your journey. Can you give us maybe a little bit of what are the key products and services of Global Trust right now?

Yeah, so As I said, my background was quite a lot of time spent working with the mining community, and we could really see that they were being driven not just by new regulations that were coming out, but also two other aspects which was driving that shift towards sustainability. One was the fact that they had large investors that were really wanting them to report on what they were doing to make sure that they weren't themselves undertaking greenwashing, for example. But also, they— these mining companies were recognizing that a lot of their products were obviously being used in consumer-based markets. So, for example, metals being used in the automotive markets, and there's a lot of public pressure then to make sure that, again, those metals were sourced responsibly. And so at Global Trust, we are sort of recognizing this sort of multi-faceted and dynamic ecosystem that's going on within sustainability. And we wanted to really create a sort of scalable tool that could answer multiple questions for multiple stakeholders. And we came up with this service of GlobalGuard, which is really our flagship service today. And this takes in terabytes of data from satellites and other geospatial sources, and it turns it into real actionable information. And particularly around things like vegetation, so how much deforestation do I have, how healthy or productive is my vegetation over my site of interest, things like monitoring atmospheric emissions, so how much methane am I producing, how much sulfur dioxide, nitrogen dioxide, what is the biodiversity like at my site, and also things like what is the social impact of my site, you know, how many temporary settlements are there, is the water polluted, Are we encroaching into green space that might actually impact the quality of life for local communities? And the idea with GlobalGuard is that it can be applied to either a site such as a mining site or an oil and gas site or even a textiles factory, or it could be applied to, for example, a portfolio of assets that a financial player might be looking at. And this then takes it from looking at one or two sites to tens of thousands of sites, maybe even more. So a big part of what we tried to do within that service was not just look at each of these individual environmental parameters because there's tens, 20, 30 of these different parameters, but how do we start to combine these parameters in a way that means you don't end up with millions, if not billions of data points for each client? So for us, it's very much about bringing together all of these different pieces and then putting them into what we've called a health gauge. So you can really see which of those sites or assets have really caused either you problems, and maybe it's the sort of naughty child that needs to sit on the, on the step and be, uh, need additional attention, or which of those sites have actually increased and are they increasing in line with your sustainability policies? That's something positive that you're seeing as a result of those things that you're putting in place. And that's really what GlobalGuard tries to do. It tries to create transparency either in your operations in your portfolio or in the wider supply chain as well.

Okay, that's an amazing product, by the way. Um, it sounds amazing. So, so the output, the quality of output as I see it, really depends on the quality of input. So where do you guys source your satellite data from? Is it like Global Trace has its own satellites up in the space, or is it, um Effectively.

And I completely agree with the need for quality data in, then getting quality data, or rather quality information out the other end. For us, we don't own our own satellites and we actually see this as being a positive or part of our selling point. We're not actually linked or biased towards one dataset, so we're not trying to answer every single question with a single dataset that doesn't quite match the challenge that's put in front of us. Rather, we access all of the available data from satellites that you can get on a, on a public space. So this is both openly available data where all you need to do is put in your email address and a password and you can start to download that data today, and anyone can do that. So from places like European Space Agency's Copernicus program or places like the USGS Earth Explorer to get access to things like Landsat mission. And then we also get access then to commercial datasets as well. And the reason that we get access to commercial datasets as well as the open ones is because the commercial ones give us slightly higher resolution. That means slightly more detail on the ground than we could get from those open ones. But again, when we talk to our clients, they're always asking, what's the most detail that we can see on the ground? But typically that doesn't actually necessarily relate to their user requirements or what they're actually trying to address. So we make sure that we use the right satellites for the right challenge space. And obviously the commercial ones come with the data cost as well. So everything that we do, we try and make sure is kind of economically efficient or sustainable then for the end user as well. So there's no point in saying I want to monitor hundreds of thousands of kilometers of forest and I only want to see sort of 10% change over a period of time. And then counting individual trees, because you're just wasting both the, the fact that you're spending a lot of money to get a lot of high-resolution data and also a lot of analytical time and processing power to do that, when actually you could just be using some of the open datasets with some easy and straightforward analytics to get that sort of information out. So it really is a balance between those requirements and the datasets that one uses.

So when we talk about the quality, you mentioned resolution, and I come back to the point of the open, the commercial datasets you can get. Just thinking from the perspective of people listening and they go like, oh, I just had an idea what I could do with that data. So the difference is only the resolution, or from experience, would you say the datasets, you know, fall short on some other criteria compared to commercial ones?

So good, yeah, good question. So when thinking about satellite data, there's a whole bunch of different parameters that one needs to start to consider when understanding what the right data is for the right challenge space. So absolutely, spatial resolution, which is the amount of detail or the size of an individual pixel relates to an area on the ground, is often used as one of the only parameters that one should look at. But actually, You know, that's really not the case, and all sensors are not built equal. So there's other aspects that we need to think about here. We need to think about, for example, has the data undergone a rigorous calibration and validation program? You know, you don't want to be carrying out analytics on some of the data and claiming that some of the outputs say X, Y, and Z when actually the calibration and the validation of the data that you're playing with is so broad or not even happened that, you know, you're not playing on a kind of equal playing field. So not just the spatial resolution, not just the quality of the data, but then there's also this idea of spectral resolution. And this refers to the number of spectral bands that the satellite can observe. Unlike the picture— the cameras that we have, like, on our smartphones and things like this, satellites actually have the ability to capture data in multiple different layers. And each one of those layers refers to a different wavelength. So for example, everything around us has what's called its own spectral signature. That's the way in which the light is either absorbed or bounces off the target. So for example, we can see that wood, for example, is brown, and we can see that healthy grass is green. In the case of grass, that's because it's absorbing all the other colour wavelengths that we can see. So red, blue, and it's then reflecting back that green light. And that's why we can then see the plants as being green. These satellites aren't just limited to what our eyes can see. They can actually see multiple different wavelengths outside of that. They can start to see things like the near-infrared wavelength. So that's particularly important for things like vegetation when we're starting to quantify how healthy they are. We can start to see the, the quantity, for example, of chlorophyll within the plants. And if you have higher amounts of chlorophyll within a plant, that typically indicates that your plant is more healthy. And if you apply that to agriculture, for example, that could be, where should I be putting my fertilizers? Or if you're thinking about sustainability, that could be, if I'm looking at plants that are maybe slightly less healthy than they were last week, is that a result of pollution or some sort of anthropogenic activity that's gone on here? More than that as well, we can then start to look at even longer wavelengths, so going into the shortwave infrared, and we could use that again for looking at the type of material of targets that we're looking at. So roofs on houses, for example, be able to differentiate between different types of roofs. You could apply it to looking at mineral exploration for mining and other purposes, all the way then to looking at thermal wavelengths, which could tell us then how much heat is being radiated out of an area or a building, for example. So this idea of spectral resolution is really important. And then finally, there's, there's, there's another key aspect that I'd like to pick up on, which is also something that's referred to as the temporal resolution. How often can I get an image? So, for example, we know with the openly available satellite imagery of Sentinel-2 from the European Space Agency we know that that goes overhead every 5 days. And that's really important if you're starting to plan an application. You know, if you're trying to look at maybe something like building construction, maybe monitoring every 5 days is too frequent, and maybe you don't need to get every single one of those data points. But if you're trying to use it in the terms of maybe trying to have humanitarian aid as a result of a conflict area, or a conflict zone, then maybe you want to have an image every 1 day, for example. And there are satellites out there and constellations out there which give us that ability to do that.

You know why I love the show, Steve, is exactly this moment. I learned so much in this. I mean, all of that is new information and just sparked so much of my interest. Um, I want to ask, so if I'm using open source, and I'm gonna go and play with this just straight after the show, I'm not getting all what you just talked about, isn't it? Uh, it just feels very expensive. So from, from all the different, different types of images I can get, the open source, I would imagine, just offers the, the, the pure, just a baseline. And I might be repeating with the same question here, but just, just trying to really understand like the difference between free and paid when it comes to that.

That's an understandable conclusion, but actually weirdly not correct. And we, we base a lot of our services off the open datasets because they are so high in their quality, but also being so readily available. So if we take, for example, the Sentinel-2 satellite that I spoke about, this particular satellite is an optical satellite, so it uses those wavelengths that we were just discussing earlier, and it looks— if you wanted to create a normal color image or a natural color image, the way in which our eyes look at it, or the way you see these things on Google Earth, for example, or other tools. Sentinel-2 offers the ability not just to create those natural color images, but it also has a whole bunch of other spectral layers as well. It's actually got, I think, something like 13 spectral bands that allows us to do that analysis all across that spectrum that I just discussed. So you could use that satellite for geological exploration. You could use it to help you with your precision agriculture. You could even now use it to monitor things like methane blooms. There's a lot of academic papers and algorithms that now exist to take some of those spectral responses and monitor atmospheric conditions. So these satellites are incredibly versatile. The ones that are open, openly available. So for example, backed by the United States Geological Survey and NASA or by the European Space Agency, typically they're really grounded in a lot of science. They've had lots of calibration validation programs associated to them. So they're very high quality data more often than not. The only sort of major limitations with them end up being either the limitations of the physics and the technology itself. So for example, that particular type of satellite can't see through cloud. So here in the UK, you know, we're quite clearly not sadly going to be able to get a cloud-free image every 5 days. But also the spatial resolution that's associated to it. So the open datasets typically have slightly lower spatial resolution. So for the Sentinel-2 satellite, we're talking 10 meters on the ground. So each pixel represents 10 by 10 meters, whereas with some of those commercial satellites, you know, we're coming down to about 30 centimeters in spatial resolution. You can even use AI to increase that to even higher spatial resolutions.

Okay, a very right point about you wouldn't see anything here in UK. It doesn't matter which postcode you use, you'll get one image every 10 days. Jokes aside though, um, is there limitations on what, what area I want to explore? Is it no limitations, or would I put, you know, I don't know, something in Antarctica and it would go like, you can't get those images? Or maybe I put something, uh, like a conflict zone and it would go like, yeah, you can't see in here?

Yeah, and that's again, yeah, good, good question. It's one that we get asked quite a lot actually by our clients, even if, even if they have no interest in looking at conflict zones, they're still— they always ask that same question. Um, and from the open datasets, typically you can see anywhere in the world in which that satellite is covering. Um, the reason that I use those particular words is because you don't end up getting the really extreme poles because of the type of orbit it is. It just doesn't cover those really extreme parts of the poles. But actually you can cover and see all of those conflict zones. But again, you're seeing them at 10-meter resolution. So what does that mean in terms of what we can see on the ground? You're not going to be looking at individual cars. You're not going to be looking at people. This is going to be rows of houses. This is going to be large-scale targets that you're trying to identify. The commercial satellites are slightly different. While you have the ability to task an image anywhere in the world within reason, so for example, you could task images over any of the countries that have conflict zones, you would be likely limited by the actual small active zones or active war zones themselves. Typically, these are quite small areas that cover sort of maybe 10, 20, 30 square kilometers. And that's regulated by different international and national laws where people don't want you capturing that imagery.

Yeah, that's what I was effectively thinking when you were talking about you can put in the postcode and get that. I was like, hmm, this has to be some type of regulation around that. That clarifies things. When it comes to commercials, and I don't know if you can disclose this to you or not, what is the pricing modules? If it's a public information, I can go to their websites and look at it. It would be good to know, but if you have some unique agreements, it's okay if you don't say it.

Yeah, no, that's absolutely fine. And actually, this is one of the biggest challenges with the space sector. None of this information is necessarily transparent and easily accessible. And so there aren't that many price lists that you can go online and find. And typically that's because there's a different price whether you're wanting to buy one image or whether you're wanting to buy 100 images. And there's a different price whether you wanted to buy that image as a one-off or whether it's part of an ongoing service, whether you get it through a cloud platform. So it's never quite a straightforward question or answer, rather. So really, there's kind of two different aspects to this. And it's kind of built around the fact that over the last, really over the last decade, the industry has really sort of changed and evolved. And we've now got the miniaturization of electronics, which has driven now much smaller satellites. The cost of launch is going down as well. So it means that we're able to launch much more. We can launch smaller satellites for cheaper with very high quality outputs, and it's becoming cheaper to do that. So it means that these satellite providers now have constellations of satellites out there. You know, a single provider could have 100, 200, even 1,000 satellites at once going up there. And that is really driving the price of the data down. You still have some of those satellites which cost hundreds of millions up there, and they're really sort of beautiful Rolls-Royce of data, if you will. And they do come with a hefty price tag. You could be looking at between £1,000 and £3,000, for example, for a brand new image. But if you wanted a brand new image from one of these smaller satellites from one of the startups, then you could be looking at an order of magnitude less. You could be looking at £200 to £500. And then we have a second component as well, which is that's only relevant if you're trying to task a brand new image. But there's archives of images now going back to the 1970s, if not actually further back into the '60s from the Cold War. And if you're wanting archive imagery, then actually again, that comes with typically an order of magnitude less money. So that satellite image that cost you £1,000 to £3,000 to take a brand new image might only cost you £500 for an archive image, for example.

Okay, so effectively the pricing module is on the frequency of data, on a quality of data, and then there's a historical data sets you can buy into. That, that, yeah, that's good to know. Anyone listening, you know, because there's so many ideas product-wise coming out of having these huge and quality datasets. The one you mentioned, which, you know, was very easy to resonate, was the agricultural one where you just effectively look at the pictures and you see how the forest is doing or how your fields are doing, how your potential strategies around growing healthy plants are doing. So come back to these examples. Is there a real use case where you can give us the way to go? Like, look, This is where we came into the working with the client. That's what we had. That's how we used our satellite data to make decisions. That's what we decided, and that will happen later.

Yeah, absolutely. Um, I mean, we— for us, obviously agriculture is a really easy one from a public, uh, point of view to kind of understand because we've all driven past the field. We can all see if vegetation is healthy or not healthy. But actually, maybe some of the harder things to think about are maybe some of the more subtle environmental impacts that we can still see from space. So for example, we work a lot with organizations who create a lot of dust. That dust can be created from things like industrial activities, it could be created from things like mining activities or construction, for example. And actually, this dust creates a huge problem for those industries. It can go into local communities, sometimes it can even be toxic. Quite clearly, there are local, national, and international guidelines about what you can and can't do with this. The problem with dust is it is so diffuse and it goes everywhere. These companies that we work with, they do have sensors on the ground where they can identify where the dust goes, for example, but they end up being very point source, which means that you only know maybe 5 or 6 different locations over 10 square kilometers of where your dust is. You don't really understand then as a result of that what the real relationships are, how far are some, some types of dust going versus other, others. And so this is one of the areas that we are using satellites and some spectral analysis. So taking those different layers of data that I spoke about earlier and actually working out how far this settled dust goes. And this is a really interesting bit of work and technology because Historically, things like dust really has been a problem, especially for people like the mining community. They've had to use sprinklers to reduce the amount of dust on the ground, and that comes then with a utilization of water, which is obviously quite a limited resource, especially in some of these rural, hard-to-reach locations where these mines are. And then again, if you're using water and it's touching the dust, that in itself can then create runoff and pollution from that. Now from space, we are actually able to map out exactly where the dust goes from the activities, and we can also look at some of the high-level composition of that dust by looking at its individual spectral footprint. So what we do is we go onto the ground and we take samples of that dust. We then take it back to a lab, we put it into what's called a spectrometer, so we can work out exactly how that dust responds then to different spectral wavelengths, and we can then compare that then to the satellite imagery, and we can then map out exactly where those different chemical or different types of dust then go within this area. Now, of course, there are limitations with this technology, like you need a certain level of, a certain threshold of dust. You know, if you're only using very small amounts of dust, then maybe this isn't the right technique. But what this allows our clients to do is have this much broader picture, not just at one point in time, but multiple points in time that they can then come up with things like really clear mitigation strategies. So they can come up with ways of doing new types of sweeping on the road, for example, or they might be able to recognize, actually, we don't need to turn all of our sprinklers on because it's a waste of resource. We just need to turn them on over here and over here where we're seeing the greatest amounts of dust. And again, it's all about that efficiency and it's all about ensuring also that if there are local communities and there are local ecosystems, that the impact to those is minimized as much as possible, if not removed.

So when it comes to social environmental initiatives big companies do, is this driven by regulations or just a pure good of heart?

It's, yeah, good point. It's, I'd say it's driven mostly by the regulations. You know, everyone, everyone is going to be driven by the bottom line, and sadly that's the way it is. But we are— there are certain sort of flagship companies that we work with that really do take that more sort of moral high ground and recognize that they do have a responsibility. And what we've seen again over the last 5 years is that companies have gone from a place of having a separate sustainability report and a separate business model to now having what they call a sustainable business model, and they combine the two. So these things are really no longer separated, and companies are recognizing that if they embed this idea of sustainability right at the bottom, it gives them so much more resilience as a company. And actually, more often than not, those companies are also seeing monetary savings as well through things like using less energy, through things like being proactive in terms of their management of the environmental and social impacts rather than being reactive to them. And so I think we're going to see more of that as we go forward into the future with this combination of not splitting out the sustainability people from the commercial people and actually pushing them all together.

I love hearing that, Steve, by the way, because you always have this— maybe it's just me and maybe it's just my opinion, but when I think about big companies I don't know why, but you always have this monopoly person behind it where you, you know, you— like you mentioned, it's always bottom line, it's always the revenue. And you feel like if there would be no rules, everything would be a lot worse than it actually is. So it's good that there is a compliance to meet, there is regulations to follow. And then hearing that companies actually take that step further and go like, you know what, we all live on the same planet, so let's just make you know, look, how can we make this better? So that's good to hear.

I mean, we're not, we're not in a perfect world still, but, you know, we're seeing the advent of a lot more regulations come out. So for example, last summer there was the International Sustainability Board, the ISSB, which is trying to create sustainability frameworks on a, on a global scale. There's been things like the EU Green Deal, which is looking at creating sustainability frameworks and standards on an EU deal. And then there's obviously other similar ones that are both international and national like that. And the world is— there's been regulations out there for a long time, but that environment is now rapidly evolving. And to be honest with you, there's two reasons for that. Obviously, we're becoming a lot more aware as a population of those environmental and sustainable issues. But also we're at a sort of inflection point where technology now allows us to do this as well. You know, it's— there's no point in, in kind of creating regulations or requirements if you can't monitor them. And we're now at that point where we can do that in a trusted way and on a sort of global scale as well.

Exactly, that's a very valid point. If you cannot measure it, how can you improve it? And absolutely, the technology allowing this, yeah, that's a much brighter future, hopefully. So, um, before we switch topics, and, um, I really want to talk about how you guys look at the data and potentially utilize AI to make your data better. So we have this question, which is in the spirit of the show, which goes like this: what is the one most important factor for any technology business growth? And you can only choose one.

Good question. I'm kind of conflicted here between my passion and morality around sustainability, but then also being an owner, a part owner of a company. And especially as a startup, as a small company, you're always driven by the market and sales is everything. Now, you know, you don't have a company unless you are making those sales. But you as a business leader do have an opportunity to actually reject sales as well. You need to recognize that. And so I think if you are going to be doing this in the world of sustainability, you know, we have been in situations where companies have asked us to monitor, in Verticommerce, good activities that they're doing. And we'll ask them what they are doing and they will say things like, oh, we're we're doing massive amounts of plantations across countries X, Y, and Z. And you dig a bit deep and you realize that maybe they're using invasive species or maybe they're doing monocrops and cutting corners. And while you've got to follow the sales and the money, you've still got to make sure that you're fitting in with the vision and the culture of the company that you're trying to put in and creating, rather. So again, do follow that money, but make sure that you're doing due diligence and you are saying no to the right things as well.

I love that, actually. And you're not, you're not the first CTO to actually choose the sales. I mean, at the end of the business, at the end of the day, it's a business, right? And we live in a capitalistic economy, and you have to bring in the revenue. It's just, it's just what it is, and it's a, you know, beating heart of any business. But that you definitely can make a choice where you actually partner up what you buy, what you sell, and then goes back to your business vision, mission, and values. So that's a very— I love that answer, Steve. Very good answer. All right, so AI for sustainable future opportunities and challenges. How does Global Trust leverage AI right now?

Good question. And you know, AI has been a buzzword now for the last, what, 2, 3 years, and it's hard to go anywhere without anyone asking that question. And for me, look, AI is massively powerful and we do utilize it. But for me, it's about making sure you utilize it in the right place and in the right way. A lot of people, not just in our industry, but I'm seeing across multiple industries, use AI because they either want to get that level of exposure and say they've used AI, or maybe they don't realize that maybe it's not the most applicable tool to use in that case. And I think again, it goes back down to what is the application that you're trying to address? So each time that we have a new challenge in from a client, we sit there and we think, what is the right approach of doing this? We're not sat there going, what's the right type of AI here? No, we're not funneling ourselves down to that thought process. A lot of the time, the AI technology that's out there, for example, isn't scientifically cited or backed. So, you know, if we are trying to create outputs that says there's so many parts per million of methane or there's so much turbidity within your local water supply or, you know, your vegetation location A is more productive than location B, actually more often than not we'll go back to the scientific cited approaches than we would for something like artificial intelligence because you can't it's very hard to unravel what the AI has done. And obviously there's a whole bunch of biases and limitations that are built into it. But having said that, quite clearly there's a whole bunch of opportunities as well that surrounds it. So for us, it is, it's all around applying it to the right application at the right time, but also disclosing it to the clients as well and letting them know when, one, we've used AI and more importantly, two, when we've used AI and it's been human-checked and verified. It's very important that if you have used AI, I believe, and especially in a world where we're trying to prove credibility and validation and transparency of these companies and these supply chains, that we again are being transparent in the work in which we're doing. And quite clearly, if you're monitoring, let's say, 100,000 different sites or assets around the world, um, and you are using AI, then maybe sometimes you're not able to look at every single one of those by eye and put a human in the loop. But you need to be clear about that. You need to be clear about the limitations and the accuracies of it. So we use artificial intelligence a lot, and in particular, we use a range of both what's called shallow machine learning and also deep machine learning. And we use them, try and use them, to be honest with you, quite sparingly, so that whenever we are creating outputs, there's a real kind of chain or history that we can lean back on and say, look, this is why we're seeing this output. And if we do use the AI, then typically what we try and do is, one, make sure that we never use this kind of term of a black box so that no one really knows what's happening and out comes this magical answer. Never do that. We always kind of open up and show what we've done. But also if we create an output, we also want to be able to say how we got to that output. So that means making sure we've got a clear understanding of what our training data is and a clear understanding then of what the hyperparameters are within our machine learning models.

Okay. Can you give a quick explanation of what is shallow ML and what is deep ML just for people listening and how effectively you apply it?

Yeah.

And why you choose shallow in something circumstances are way deeper than others.

Yeah, absolutely. So let's give an example. So with machine learning, there's kind of 3 different options. You can do what's called supervised learning, whereby you're telling the computer what an answer is and then you're asking it to apply that rule set then to a new area. Or you're doing something called unsupervised learning, where you're essentially giving the computer a bunch of different datasets and you're asking it to work out what the rule sets are, or there is something in the middle where you can do this kind of semi-supervised learning. And you can apply that to both the shallow machine learning and also the deep, deep learning as well. For the shallow ones, the reason that we'd sometimes use the shallow ones is possibly because it's the right approach for that particular challenge space, or maybe it's because we don't have as much training data as we might want. So when we're talking about deep learning networks, if you're starting from scratch and let's say you're trying to create an algorithm that counts, let's say counts cars or counts trees, whatever it happens to be. If you're creating an algorithm, a model from absolute scratch, you're wanting thousands, tens of thousands of data training points. And typically we don't always have that. Now we create our own here and we, and we take trusted open ones as well, and we curate those. But typically time doesn't always allow to create a training dataset of tens of thousands. So typically that then lends itself to either shallow machine learning. So something like a random forest approach, which is like a decision tree. So if I have dataset A and I've got dataset B, then typically what that means is I'm looking at answer C, for example. That's what a decision tree might give you based on then some training data that that goes into it. And you can do that on a very small quantity of data. So if we're working with a client and they've got some ground-based data over an area of interest, but it's quite limited, then maybe that's the right approach to go down depending on their challenge space. If, for example, we've got access to hundreds of thousands of training data points and associated data that goes with it, then we could start to consider things like the deep learning And so people would have come across these things like convolutional neural networks and other types of deep learning networks as well. And that's where we apply that. There is this kind of interesting middle ground where you can apply something called transfer learning, which is where you've already built up your, your deep model on tens of thousands of images, let's say on cars, satellite images of cars. And you now want to apply that to be able to identify trees, as we said earlier. Typically because the model already understands really what a satellite image is now, although it's been trained on cars, you can feed it a lot fewer datasets of trees and actually start to transfer that model across onto another area. So instead of tens of thousands, maybe you want hundreds to thousands of images instead, and that's something that could be done as well.

Okay. Are there any good open source pre-trained modules right now which would be applicable to Global Trust? Potentially you guys even using it? And a follow-up: are you contributing to open source communities by sharing your modules?

Good question. So there are a few open source models. I mean, if you compare it to other sectors like the medical sector, for example, there's a lot more than there are in the satellite, but they are growing. The biggest limitation still remains the training data rather than the models themselves, because you can take any computer vision model, and a computer vision model is a deep learning model that you apply to images or to videos. So you can take any computer vision model as long as it's hitting the requirement for your application, and as long as you've got the training data, you can then apply it to your challenge space. There are some pre-trained models online. I think that people should use those and should learn and educate themselves, but if they're going to start using them in a kind of operational sense, then I'd suggest kind of really taking a step back and really understanding what's going into each one of those models, what the limitations are, what the training data is that they've used that goes into it. We do use a mixture of taking some open source models, or at least parts of them, and then maybe adapting them for our use. There's the ability to take models and then combine them with other models on top, for example. So we, we do that work. My— the developers that we have here would absolutely love to contribute to the open source community. Because we are a small startup, it's very difficult for us to do that and have the resources to really put some of that code out there, especially when a lot of it is focused on very sensitive information to our clients. As well.

Understandable, understandable, Steve. And shout out to the developers, by the way. Thanks, guys.

They do a fantastic job. And actually, it's important to say that, you know, within the team we've got a mixture of developers and data science, uh, and it's, it's those two specialties that come together that's really interesting. And then finally, I'd say a third component to that as well, which is the actual application specialist. So someone who understands DUST if you're looking at DUST, if you're looking at agriculture, someone who understands agriculture. And it's It's the kind of combination of all three of those different skills that's needed when you do this work.

So you mentioned black box, Steve, and, um, AI has that notion of the black box, of course, um, which makes it quite difficult to calculate return of investment of these projects. So the question is, if you would invest manpower or woman power to come up with deep learning or shallow learning or anything you do around AI, How do you know that these initiatives actually are yielding positive returns eventually? It's like, is there a thought process around this, or is just purely like, okay, we're going to automate something around our business operations, so we're going to just have better operational margins?

Yeah, good, good question. And again, I, I was going to pick up on something you said there and answer it first and then move on to the bulk of the question. You, you talked about automation, and I think there's often a tendency to confuse AI with automation and the two things end up being actually quite, quite separate from each other. You can have automated pipelines that have nothing to do with AI and in the same way that the creation of an AI module or model doesn't necessarily mean it's automated either. So, you know, I think we see this a lot within the industry and within our clients sort of confusing those two things and it's important to say you can create scalable global models without needing AI and they are automated. And at the same time, we craft AI modules here which aren't automated and to be honest, probably never will be, at least in the short term, because they need so much human intervention. In terms of resource and the amount of time it takes to do this stuff, I mean, the developers that we have here would love to have an endless amount of resource and time to to it and an endless number of people to be able to contribute to it as well from different backgrounds. Sadly, that's just not the reality that we live in. Clearly, we've got clients and projects that we need to deliver to, and we've also got staff and salaries that we need to pay for as well. So whenever we do come up against a project or a new bit of tech that we're developing, it needs to be weighed against all of the wider business considerations. And so therefore, you know, we will always choose the right solution in terms of the quality for the client, but also in terms of the resource that it's going to take and the effort it's going to take. And if there's going to be a 5% increase in quality for the client, but maybe a hundredfold increase in the cost because you're using now a really deep neural network that needs a lot of training, a lot of tweaking. Then again, is that really sort of value for money? And again, we'll tell is the creation of training data and is the validation then of the outputs as well. It's always where you've got those very human intensive moments that are built into it. But again, as we said before, so key as part of that process.

I love that. Very good answer, Steve. Very good answer. So I want to venture into ethics a little bit. It's always very bias and people's opinions, but I want to hear your opinions. So we take the data. Is there any issues, morally speaking, around data ownership? Like, you know, we're taking these pictures from up in the sky, up in the space, of people doing things. And if we take in commercial pictures, in theory we can zoom out, you know, very closely. Should I as a person be worried that that's happening?

Yeah, the quick answer, and quite clearly I'm slightly biased, is no. But I'll elaborate on that. All right. And I'd start from a place of, especially as we're both here in the UK, I think the UK is one of the most monitored countries in the world. I think I read an article that says we're captured on CCTV 70+ times a day or something like this, depending on where you are in the country, of course. And, you know, those, those are capturing our faces and our activities and our cars and things like this. It's important to say that even with the highest resolution spatial resolutions from space, you're not going to be able to see your face. You're not going to be able to make out what gender you are. You're not going to be able to work out what your number plate is, for example. So that, that's the starting point. But This idea around data privacy, this idea around access to data is an important one and it does need to be respected and does need to be talked about. And I think when you talk to these commercial data providers, they all do due diligence on your company before they sell you the data. For the open data, as I said, slightly lower resolutions that go down to 10 meters or so in spatial resolution. There is a lot less due diligence on that. But as I said, you're looking at almost like street-level information rather than kind of house-level or even car or smaller than that. But there's something more than that as well, which is, okay, so maybe that data we don't— and even the high-resolution data, maybe we don't need to put that many stringent rules around it. Clearly there are, by the way, especially around conflict zones and things like this. But more than that, what does it then mean for the derived outputs that we're creating? And that, again, there's, there's an element of responsibility there that should not be overlooked. And it is a big one, you know. So for us, for example, there's a huge amount of work that goes into the security of our systems, into making sure that there are very large firewalls between our clients so that no one gets to see that data, or those data rather. And that's incredibly important. And it's also important to drive that culture internally and have those discussions around privacy and data efficacy with everyone within the company and also your clients as well, so they're aware of what's going on.

So you mentioned due diligence. These, um, very briefly, how does that look like? What do they care about? Do they care about um, that no one as an employee have been criminally attached to some type of, you know, weird activities. Like, I, I worked for a business in gambling, healthcare, and finance, so I know how that looks like. But when it comes to providing a product as Global Trust does, what, what, what do the companies care about when it comes to due diligence to be able to sell you the satellite data?

Yeah, absolutely. Good question. Well, they do some basic things like what country are you based in? Who are your directors? Have your directors been linked to any of the illicit activities that you mentioned a minute ago? Are you in any kind of financial trouble or anything like this? So there are checks that go behind the scenes that we don't see. To be completely honest with you, not being a data provider myself, I haven't actually been involved with those checks. Quite clearly our company has undergone them, but it's not something that we're necessarily party to. It's something that we don't see ourselves.

So you're just getting— yep, sorry, that's right.

I was going to say, but when we're giving our outputs out, as I said kind of earlier, we're doing due diligence with our clients the whole way along. Are they— should we be working with these people? Should they have access to this information? If they do have access to this information, do we have the rights to give it to them? So we, for example, when we buy the satellite data from a commercial vendor, we don't own the data. We own a license to the data. Does that mean it's given us the right rights then to pass it on to our client? More often than not, if our client wants to use the data, the original satellite vendor again needs to do the due diligence on the client, which makes sense. So there is a kind of chain that goes on there. Each step of it, it is important to continue to do that due diligence.

Okay, that makes sense. I mean, at the end of the day, yeah, you're just going through a regulator who will look at you and go like, yes, you can do this business, or no, you can't. But you don't really know what exactly they check. That makes perfect sense. So I want to look at the crystal ball and ask this question, uh, or more, I'll ask the question, you'll look at the crystal ball. See, how does the future look like for this business, the satellite data people taking that information, utilizing, number crunching, sprinkling some AI on top. How does the 10 years in the future look like?

I think 10 years in the future we won't be so obsessed, I hope, with using terms such as AI, and it's going to be part of the kind of fabric of what we do on a day-to-day, and it just gets wrapped into this term of analysis, um, and we end up going from a place where we are today, which is biasing everything towards that to actually more of a balanced piece and using the right tools at the right time. But for me, in 10 years' time, the key thing is that these regulations that we spoke about before around sustainability are going to be so much more solidified. People are going to be having used them for the last decade, so they now know how to do it. You know, we're still in that learning phase. Even the heads of sustainability of these large corporates, they're still struggling to understand what it really means for them. I really hope that we're in a position where everyone is in a much more educated place around what satellites can do for them, what space can do for them. And then more than that as well, I think we're also going to be continuing to be in a place which we're seeing already, which is driven by things like fake news and stuff like this, where people are asking for more traceability in the results. You know, if I am saying to the farmer that you've got you need to put fertilizers in location X, Y, and Z, you know, him being able to trust that result and there being transparency in one, the data that's coming from the satellite to the ground, the ground then to me, the work that I do and apply to it, and then to the farmer, for example. And quite clearly, you know, that has really big implications if you think about stuff like greenwashing and companies claiming to do things that aren't right. And, you know, a lot of this stuff might have to be taken to court and you will have to prove that and you'll have to prove that every single step of the process has undergone, as you said it would. So that traceability and credibility of that process will also be really important over the next 10 years. Okay. But no, I'm excited. I think we're going to see a lot more ubiquity of satellite technologies and a lot more people using these Earth observation data in their products without even knowing. I think that's where I'd like to get to. People don't need to understand that they're necessarily always using these datasets. They need to have trust in what they're using, absolutely, but they don't need to understand every step of it unless of course they want to dive into it. But it's much more, actually it becomes part of the daily routine.

I got one speculative question. I'm not sure if you have ideas about this. So we are able to get a certain level of free data. We can pay for a much better level of pay data when it comes to satellite data. What do you think governments have?

So, good question. So the first thing, you know, you're right, with being able to pay, you can get higher resolution data. But just to be clear, that doesn't always mean it's necessarily better, but it's different. Just to be clear on that point, from a governmental point of view, a lot of those commercial data vendors actually have governmental anchor clients. So they'll have military and governmental clients who they are selling maybe 50% plus of most of their data sales to. And then the other smaller chunk of data sales go to companies like ourselves, for example. So they are using similar techniques or outputs that we are. There are things that you can do so you can So look at things like the Hubble Telescope and recognize that, or at least used to be the kind of biggest telescope that we had up in space. You could do maths and turn it around and work out what you'd see on the ground. So you could kind of have an indication of what might be possible from that perspective. But there's a whole bunch of different technologies that are going on here. It's not just about spatial resolutions that's interesting. It's about combining multiple different types of satellite data. To give you interesting inputs. And I think, you know, from a governmental perspective, more than ever they're now working very closely with industry. And I think that, that differentiation between industry and government is probably quite close these days.

Okay, that's quite good to know. I mean, I always thought— not always thought, but my— in my early days with computer, and that, that goes like 25 years back, we have this It was this presentation being done by a technician from government, and they came out and they gave the story like, you right now have a hard drive which is 1.2 gigabytes, and that was what you would get in the personal computers at that point in time. That was like Pentium 1, you know, early ages. And they were like, we already have like 250 SSD drives. And, you know, they gave the speech about like the future of drives and the You know, the distance felt like so big. And always when I think about the tech we as people get and what the governments are playing with, always think about that gap because it just stuck with me because I was like 14 years old. But effectively what you're saying, it's, it's not that big.

I think, you know, it's quite interesting because the, the commercial industry, as I said, has gone through this kind of revolution of new space. And, you know, while of course we have to be really mindful about the sustainability of space, and that's actually one of my other areas of sort of passion in itself. This new era allows us to test and innovate new sensors much more rapidly than we used to. So actually the speed of innovation that the commercial companies are now doing is much quicker than if you were, for example, only launching one satellite every couple of years, for example. Because we're able to get that immediate feedback, see what works, see what doesn't work. And you can also have the commoditization of it in the same way that we've now got amazing cameras and capabilities within our phones. Part of the reason that that works is because they're selling millions of them. You know, they've started to commoditize them. And that type of work is now happening within the space sector and within the commercial space sector, which just generally isn't possible at the scale of the government because there's only, only a set number of users then when you're talking about the government.

Yeah, that makes perfect sense. I, I need to change my mind about this. It's such a huge imprint from, from my early days. So, um, let's imagine that I'm a CTO who cares very deeply about sustainability. Let's not imagine, I am a CTO who cares very deeply about sustainability. Where do I go to participate more or participate better? So let's say maybe I got a new role or started a new business and I want to bring in those initiatives which you guys are helping with into my business. Are there any resources or just general tips you, you know, you would give to CTOs like that?

Good, good question. I mean, for me, there are the, the— and this kind of goes back to one of our initial questions about how— what advice I'd give a startup which is, I think, especially as a CTO, it's so easy to get so distracted with the technology and excited by the technology that you kind of struggle to look up from your desk. And there are so many different networking events and so many different conferences that are going on almost on a weekly basis. And I'd suggest getting involved with those and being part of the discussion. At the moment, what we're seeing within the sustainability industry, it is, is it's really within this moment of, of change, you know. I mean, if you look back only 6 months ago, we had such a different landscape. And I think the important thing is, if you are going to get into sustainability, is you need to be part of that discussion. You need to get out there, you need to start to talk to some of the people who are leading those conversations, leading those thought pieces, and become part of that ecosystem and not just sit in isolation and create a product for a year and then say, I'm going to release it in 2025 or whatever. You need to actually continually be talking to them and continually recognizing that these companies are thinking differently. The technology providers are thinking differently. The legislation is thinking differently on a week by week and month by month basis. And it is going in the right direction. But yeah, if you start to come out of that loop, it can be quite difficult. But I think what's nice is that there are so many openly available events that you can attend, whether they're workshops or webinars online or podcasts, whatever, or ideally on top of that as well, going in and talking to these people in person. And what I'd like to see is more of these conferences become more publicly available as well. They often have quite large— some of them can have large price tags associated them, which isn't really in keeping with the idea of sustainability and inclusiveness. So I hope again over the next 5 to 10 years we see them becoming even more accessible then to the wider audience as well.

Maybe we even see Global Trust running some of these events and having no price tag attached, right? Or maybe a local meetup. So that's a good answer. I mean, it does start by just engaging with the local communities around you. It doesn't need to be anything bigger than that. You don't need to pay, you know, big tickets. Uh, just, just find people talking about it and that's where the door opens, isn't it?

Absolutely. Yeah. And you'll find that everyone's in the same boat, including the clients. And you'll find the clients at these places as well, because everyone is struggling to keep up with what the latest thinking is, what the latest legislations are. So everyone is, is in a very similar place. And what I've noticed particularly around sustainability is that people are really interested and excited to have that conversation with people and have conversations with new people as well. I've seen a lot previously in the industry around other topics that it can be quite closed off, quite limited. If you're not in the inner circle, you don't get to talk and this type of stuff. Whereas actually, you know, the premise of sustainability is that part of that idea of inclusion and actually bringing all these thoughts together and recognizing that if we're gonna have any chance of making any real impact, we need to have this kind of multidisciplinary approach, having all these different skill sets coming together to address these challenges.

Thank you, Steve. Thank you for all the information, sharing your journey, giving us a glimpse of, you know, the great Business Global Trust Test and your product, giving us good tips as well. And I mean, I learned a lot and I took many, many notes, which I'll go dig deeper after the show. So thank you for being with us.

No, thank you very much for the invitation, Mike. I really enjoyed it.

I as well. Thank you, Steve.