TOM ADAMSON: Hello everyone, and welcome to another episode of Eyes on Earth, a podcast produced at the USGS EROS Center. Our podcast focuses on our ever changing planet and on the people here at EROS and across the globe, who use remote sensing to monitor and study the health of Earth. My name is Tom Adamson. Declassified spy satellite imagery, distributed by the USGS EROS Center, is finding fascinating uses since being made available to the public starting in the 1990s. Philipp Barthelme from the University of Edinburgh used declassified HEXAGON imagery to detect Vietnam War era bomb craters. He also used declassified CORONA imagery to identify the locations of herbicide spray lines. The purpose is to help identify the locations of unexploded ordnance, which covers much of the region, even though the conflict ended 50 years ago. We talked with Philipp about those studies. PHILIPP BARTHELME: There we go. Hi, Tom. How are you? ADAMSON: Hello. Good. How are you? BARTHELME: Yeah, I'm doing well. ADAMSON: All right. Thank you, first of all, for doing this. BARTHELME: Oh, thank you for having me. ADAMSON: Will you just go ahead and introduce yourself? Tell us about your background. BARTHELME: I'm a PhD student at the University of Edinburgh, and I'm studying the long-term impacts of the Vietnam War in Southeast Asia. So that's including Vietnam, but also Laos and Cambodia. And about my background. So I studied economics for my undergraduate degree back in Germany, where I'm from. And then I moved to Edinburgh for my master's in statistics with data science. And then after that, I was working as a data analyst and a data engineer for a couple of years. And then I think it was the first year of Covid, when I came across the PhD, and I've been doing that PhD for the last three and a half years now. So it's coming to an end soon. ADAMSON: Okay. Are you getting pretty close to getting it finished? BARTHELME: Yeah, I'll be finishing later this summer, so it's getting pretty intense. ADAMSON: Okay. I appreciate you taking your time off to do this then. BARTHELME: Oh, yeah. ADAMSON: You actually have two related studies that we're talking about. You have used the declassified HEXAGON data and CORONA data, which is data that we archive and distribute here at the USGS EROS Center. So that's why we're kind of interested in what you're doing with it. Can you give us the quick elevator speech for what those studies are about? BARTHELME: Yeah, sure. So, with the HEXAGON imagery, what I've been doing is I've been trying to automatically detect bomb craters of bombs that were dropped during the Vietnam War. So you can quite clearly see them in the HEXAGON imagery, which is very high-resolution because it was taken actually at the time of the Vietnam War and the kind of 1960s and 1970s. And then with the CORONA imagery, again, I've been looking at the Vietnam War, but this time at Agent Orange spraying. So Agent Orange was a chemical that was used during the Vietnam War by the U.S. military to defoliate forests. And you can see these spray lines in the imagery quite clearly. So for this, I've been using the CORONA imagery, which was taken a little before the HEXAGON imagery. ADAMSON: What got you interested in studying that area? BARTHELME: So actually, I didn't have any plans to come back to academia after my masters initially. But then Covid happened and I kind of was looking for opportunities. And I came across the SENSE CDT, which is the Center for Doctoral Training, which focuses on using satellite data for environmental science. And it's funded by the National Environmental Research Council in the, in the UK. And one of the projects that was advertised there was, well, what I'm doing right now, so studying the environmental impacts of the Vietnam War. A couple of weeks prior to seeing that position, I'd come across a dataset of U.S. bombing targets during the Vietnam War that had been declassified by the U.S. Air Force. And I found it immediately very interesting. And I was kind of downloading the data, playing around, looking at it for a little bit. And I'd always been interested in geospatial data, even though I don't have a geosciences background. Seeing that data being mentioned again in this PhD project, I was kind of-- quite quickly I decided that I wanted to apply for that. ADAMSON: Can you tell me a little bit more about that dataset? It was declassified bombing target data? BARTHELME: The U.S. Air Force during the war collected a lot of records of, you know, GPS locations of their targets. What kind of weapons were dropped on those targets, which planes, which, you know, the exact date and time. So it's a very detailed dataset. I think in 2016 that's been declassified and made available to the public. And actually, that dataset also includes bombing records going further to the, um, wars in Korea and also World War II. So it's a very useful resource actually. And it's very rare that you get to see, you know, these kind of military datasets that are usually, you know, highly classified. So it was very interesting to look at that. ADAMSON: And then you can line that up with this declassified spy satellite imagery from HEXAGON and CORONA. How did you find out about that imagery? BARTHELME: That was, actually it wasn't me finding that imagery. It was actually one of my supervisors, Eoghan Darbyshire, who was working at the Conflict and Environment Observatory, which is an NGO in the UK, looking, as the name says, at the interaction of conflict and the environment. And they are supporting my PhD. And he had looked at an image of Iraq at the time. And when I saw that, I was like curious and was kind of looking at, are there any images available for the area in Vietnam I was looking at, and I went to the USGS EarthExplorer website, and I found an image that had already been requested by someone else, so I could easily just download that. And I think I immediately was kind of surprised by the incredible high resolution. So I think maybe for the people who are not aware of the declassified imagery, we're talking about a resolution of about 0.6 to 1.2 meters, which is not that far off what we see on Google Maps these days. And compared to what I was initially planning to use, which is the early Landsat imagery, which had a resolution of like 60 or even 80 meters. So it's magnitudes more detailed. You can see houses, you can see streets, you can see, you know, helicopters, airplanes, military bases. So that kind of was the first thing I noticed. And then the second thing I noticed immediately was just a massive amount of bomb craters that were visible in that image. So that image I've been looking at was of Quang Tri province, which was kind of the most bombed province during the Vietnam War. And you could see tens of thousands of bomb craters. And there was, yeah, pretty shocking initially because I wasn't, you know, obviously we all learn about the Vietnam War. But I think just seeing that so visually, the amount of destruction was quite striking. And that's why I decided to use the imagery. The initial aim of my PhD was actually looking at land use changes during and right after the war, especially in areas that were heavily bombed because of this massive issue of unexploded bombs that we have in not just Vietnam, but also Laos and Cambodia. Approximately 20% of the land in Vietnam, Laos, and Cambodia is still believed to be contaminated by unexploded weapons from the Vietnam War today, and obviously, since the end of the war, there's been tens of thousands of people that have been injured or have been killed by these unexploded weapons. So it's very much still a problem in all of these countries today. ADAMSON: At this point, we need to take a step back and cover just a little bit of history. Starting in 1960, the U.S. flew spy satellites. This was the Cold War and the satellites, in their polar orbits, collected imagery with film cameras all over the world. Two of those projects were codenamed CORONA and HEXAGON. The CORONA missions provided imagery from 1960 to 1972. HEXAGON collected imagery from 1971 to 1984. Those time frames line up quite nicely with when the Vietnam War took place for the studies Philipp is working on. The reason we are even talking about them now is that the images from these systems began to be declassified in the 1990s. So far, nearly a million and a half images from these missions are available from EROS on the USGS EarthExplorer site. You might wonder, how did they get the film from space back to the ground so it could be developed and studied? This is honestly the most fascinating part. The film went into recovery buckets, which were de-orbited. Once they got down into the atmosphere, a parachute opened. An Air Force plane then snagged the parachute as it floated down. There are documentaries about this. We'll put a couple of links on the website. Philipp mentioned the impressive resolution of especially the HEXAGON imagery down to 0.6-meter resolution, about as good as Google Maps imagery today. Super impressive for satellites using 1960s tech. Okay, let's get back to Philipp and his studies. In the bomb crater study using HEXAGON, how many images did you use for that? BARTHELME: I ended up using, I think, eight images for the Quang Tri province, which I mentioned before, and maybe again as context--so each of these images actually covers a very large area that's about 250, 300 km long and about maybe 15, 20 km wide. So it's not just like small images. I mean, there's a lot of aerial historical aerial imagery out there which covers usually quite small areas, but when we're talking about these satellite images, we're talking about very, very large areas. So with these eight images, I was covering most of the province and large parts of Laos as well, actually. And then I was additionally looking at, something I call the tri border area, which is where the borders of Vietnam, Laos, and Cambodia meet, just to prove that, you know, these methods are applicable to all three countries and aren't just, you know, working in a small area. So I processed another 12 images for that area. So it's about 20 images in total. ADAMSON: What does the crater look like then in this black and white imagery? BARTHELME: Just looking at the image, you can immediately see these craters. And I think the kind of most striking characteristics are I guess the crater, what you would call the crater bowl, like just the, the hole that was created by the explosion. And then especially in more recent craters, you can also see what is often referred to as the ejecta pattern. So kind of this soil that's displaced during the explosion just comes down sprinkled around the bomb crater. So that's usually quite good indicators that it's actually a bomb crater you're looking at. The appearance of bomb craters can vary a lot, depending on a lot of factors. For example, the type of soil, the type of land cover obviously looks different if you drop a bomb on a forest or if you're dropping it on a rice paddy, or maybe even more obvious, if you're dropping it on water, you won't be able to see anything at all from a satellite image. Again, it depends quite a lot. So usually the center, kind of hole, appears, sometimes appears brighter, but often it's also filled with water and kind of appears darker as well. So you can just mostly identify that from the circular shape. And then the kind of this ejecta pattern of displaced soil often also appeared brighter in the imagery. So it really stands out quite clearly. And then obviously also the type of weapon that was used. So you have these kind of 500 pound or thousand pound bombs that leave these massive, you know, ten-meter craters. But you also have, you know, artillery shells that leave smaller craters, or cluster bombs, which were a particularly problematic weapon. So cluster bombs are bombs that kind of open up mid-air and then kind of distribute these small bomblets. And unfortunately, a lot of these bomblets were partly designed not to explode immediately. So a lot of these are still unexploded today and they're very small, so, you know, children often play with them. A lot of the accidents actually involved these kind of bomblets, cluster submunitions. But these might not leave these massive craters because they're actually quite small on impact. So all of these factors kind of play a role there. So immediately, initially, I was kind of doing a lot of online research. I have zero background in weapons or military, so I was just doing a lot of Googling. I was emailing a lot of experts trying to get feedback on what I was seeing. And as I mentioned before, there was this dataset out on, you know, target locations of bombing missions so I could start matching up these target locations, which had weapon types associated with them, and then I could see, for example, this B-52 bomb strike with 500-pound bombs were kind of referring to these, you know, 100 craters that I could see in a line in the imagery. So that's kind of how I started piecing these things together. ADAMSON: Let's also talk about the CORONA study, where you're talking about finding herbicide spray patterns, right? What did those look like in the imagery? BARTHELME: That's actually interesting because, so again, these herbicides like Agent Orange, they were used to defoliate forest. So what you're seeing is not the Agent Orange itself. What you see is kind of defoliated areas. And because vegetation absorbs light, lack of vegetation kind of reflects more light. Often you see these lines of brighter areas in these areas that were sprayed. And these spraying missions were usually also flying multiple planes, flying in formation. So sometimes you can just see like these multiple parallel lines that appear a little brighter than the kind of the surrounding forest area, which is sometimes quite easy to identify. But often the vegetation grows back quite quickly, so then it becomes a lot harder to identify. Actually this again, multiple things you can use. For example, I've been using like before and after images, when you see an area that maybe wasn't bright before, then it was bright suddenly and maybe corresponding to, again, the U.S. again declassified some of their spray lines data. So I had that and could overlay it and see, was there any recorded spraying at the time in these locations where we can see these patterns suddenly appear and kind of match these and then I could also compare this with other records, like using topographic maps, identifying where streets were located. And that way you can start to understand what you actually see in the imagery. Is that forested area, is that actually like a village maybe? It took a lot of kind of manual interpretation to identify and try to understand these images. ADAMSON: I think I understood from the from the bomb crater study at least that there's some artificial intelligence being used to identify the craters. This is a large area still that you're looking at. So how did that come into play? BARTHELME: Quite early on, I realized--obviously I started with all this manual analysis, and I, my background, obviously, is a little bit at machine learning and I-- ADAMSON: Okay. BARTHELME: realized that I would be really useful because, as you say, as one person doing a manual analysis, you can only cover a very small area. So what I ended up doing is kind of manually labeling a couple of these bomb craters in some of the areas. And then what you do is you just train an algorithm. In this case, it's a convolutional neural network, and you kind of feed it all the labeled images that you created yourself. And it starts to learn kind of to recognize and identify bomb craters in this particular imagery. And the big advantage then is that you can just take that model and apply it over your whole study area and get locations of bomb craters that it identified across very huge areas that I wouldn't have been able to do manually myself. ADAMSON: Definitely you needed those convolutional neural networks in order to get this done in a timely fashion. Otherwise, this would probably take a lot longer. BARTHELME: Yeah, multiple years I guess. ADAMSON: I was going to say this would take years otherwise, but I didn't want to assume. But it would, wouldn't it, even just that area? BARTHELME: Yeah, I think it would take quite a long time, and it would be quite boring manual work, actually. So yeah. Wouldn't recommend. ADAMSON: Yeah, wouldn't recommend. We'll let the AI do the boring work for us and get it done more efficiently that way. BARTHELME: Exactly. ADAMSON: All right, all right. Well, what were some big challenges to using this imagery, to identifying these features? BARTHELME: Well, one is the cloud cover. Vietnam is a very cloudy place, unfortunately, for most of the year anyway. And, so about like, maybe like 80% of the images were just not usable for my bomb crater analysis, just because they were only showing clouds. So that's a big challenge. And then also, I guess I went over this a little bit, just a difference in appearances of bomb craters was quite challenging. To kind of come up with a consistent methodology to label these because when you're training a machine learning model, you kind of want to have like a clear system and a clear methodology. Otherwise, if you're confused when you're labeling the models--when you're labeling the craters, the model is also going to be confused and not be very good at detecting these craters. I think the biggest challenge of using the imagery is definitely just the georeferencing. Maybe again, some context might be useful. So these images were actually taken on a film camera. And now what I guess USGS at the EROS Center, what you have is all these kind of film rolls sitting in some archive. And when I go online and request one of these images, what you do is, I guess, send someone down there and kind of scan that specific image that I'm looking at, and I just get like a picture. ADAMSON: Yeah, that's pretty much it. They're down in the basement, and somebody has to go get the correct roll, find the correct frame, scan it at high resolution and then send it to you--or maybe it's available to download for you I guess at that point, yeah, this is film. It's not digital imagery. BARTHELME: Yes. ADAMSON: Until we scan it here. BARTHELME: Even then, obviously, when I get that kind of digital scan, it's just a picture. It doesn't correspond to a location on Earth yet. So I have a rough idea of which area it was taken in. But actually what I need is kind of match every pixel to a specific location on Earth. And that process we call georeferencing that imagery. And it's a lot of manual work. Basically, it's kind of me, again, kind of looking at the image and trying to identify a couple of features that are very obvious and that I can match to locations on Earth, for example, like a kind of a road junction or a bridge or a river bend, or like a historical building, and we call these ground control points that kind of match points between the image and actual locations on Earth. And that's a very time consuming task. And you have to do that for every image you want to use. ADAMSON: Did you have to interact with anyone at EROS to get the imagery? Was it just go to the website, find it, download it? How easy was that to do? BARTHELME: That was actually quite easy. So the images that someone else already requested are just available for download. So I just went, as I said, the first image I looked at, I just went to the website, I just downloaded it, and I was looking at its after, yeah, a couple of minutes. Well, it's quite large files, so we're talking a couple of gigabytes, so it takes a while to download. But once you have it you can just look at it. And then for the images no one else has previously requested, you need to request these because, as you said, someone needs to go to the basement and scan these for you. ADAMSON: Yep. BARTHELME: So that was a bit more involved because you have to pay $30 per image to do that, so I didn't need to interact, it's all in the, like, website. You can just do that quite quickly and easily. ADAMSON: Okay. BARTHELME: It wasn't any problem. ADAMSON: Yeah, I kind of want to make it sound like we make this really easy to do for people. BARTHELME: No, no, you do. You actually do. And people can try that. I mean, just-- ADAMSON: Oh, yeah. BARTHELME: You know, I would recommend anyone listening to just, you know, go to, I think it's EarthExplorer-dot-USGS-dot-org, and just look at the area you're interested in and just see if any images are available and download them. ADAMSON: Absolutely. It is public domain worldwide: EarthExplorer-dot-usgs-dot-gov. We'll put that on the website though to make sure that's clear. BARTHELME: A lot of people just don't know about the declassified imagery at all. Most of the time I've just been like, oh yeah, this actually exists. ADAMSON: Yeah, yeah. BARTHELME: So I think it's quite good for people to become aware of this because I think it's quite useful. ADAMSON: Could the methods that you're doing, in detecting the spray lines and detecting the craters, could your methods be used in other areas for other conflicts? BARTHELME: Yeah, definitely. And actually they have been used before. So detecting and using kind of bomb craters has been done particularly in historical aerial imagery, for example, in Europe with World War II imagery. One of the main advantages of the satellite imagery is that it covers such a large area, whereas a lot of these aerial imagery only cover much smaller areas. And because you have to do this georeferencing process for every image, that actually becomes the main limiting factor when you're using aerial imagery. So having the, you know, HEXAGON imagery is actually really useful for that. And then in terms of, you know, bomb threat detection using these machine learning methods, that can definitely be applied to other conflicts as well. And that's been done, for example, in Ukraine for artillery craters, using some, you know, current satellite imagery. So that's definitely possible as well. And I'd also say for the historical imagery, it's useful for a lot of other conflicts that happened around the time the imagery was taken, in the kind of 60s, 70s, and early 80s, because there's not that much other data out there. And it's providing a lot of context, I would say, it's very valuable in providing context, because you can see locations of villages. You can see kind of where military bases were, you can see where kind of the main roads were. And I know, for example, some people in Laos and Cambodia are looking at trying to identify where these military bases were and where these roads are because they were more likely to be-- have landmines placed around their perimeter, for example. So trying to identify where these were at the time can help to kind of narrow down the search area for where these landmines might still be located today. So this kind of context information can be quite valuable beyond bomb craters. The biggest benefit is in trying to identify areas that are still contaminated with unexploded weapons today. When we're talking about the bomb craters, for example, obviously a bomb crater means that the bomb exploded, right? But it's very likely that there's unexploded bombs nearby because these bombs were often dropped in groups. So there was often, like, carpet bombing approaches used during the war. So what usually happens during clearance activities is, they start with something they call a non-technical survey. So it's basically a desk-based survey where people try to understand which areas might still be contaminated today and then narrow down the areas that they actually need to send teams in, and kind of all the machines in, to actually actively search for the unexploded weapons, because that's actually quite time consuming. It's obviously dangerous as well, and it's very expensive. And in the end it comes down to money. If we can narrow down the areas that people have to search on the ground to the areas that we know are most contaminated, we can save a lot of money and clear the most important areas. And unfortunately, there isn't that much funding available for these kind of activities. So making efficient use of the existing money is quite important. And then I think a different aspect is also-- especially in Southeast Asia, there's been so many bombs dropped and there's so much unexploded ordnance there, that it's basically impossible to clear it all. So there needs to be a kind of a long-term risk management strategy in place, and kind of having some sort of knowledge of where the risks might be higher and lower is obviously important, even in areas that might be not, you know, high risk enough to warrant clearing it. It's still good to know that there might be a danger. And then and last but not least, it's also about mine risk education a lot, especially in Vietnam, Laos, and Cambodia, there's been a lot of work just educating, for example, children at schools to not, you know, when they find these things, to not interact with them, to not touch them, to report these things so people can come and clear them. And having some sort of visual material to share with people. And that's not just children. It's also just generally for, you know, educating people about what happened during the Vietnam War. Having some of these visual images can be quite impactful. And I think it's a very valuable resource to use. Yeah, and I've been actually in Vietnam recently trying to show and present my results. And people are very interested to use it there. ADAMSON: Are you able to continue with this research or what's the next thing for you? BARTHELME: That's a good question. So, right now I'm focusing on just finishing up the PhD and kind of coming back to the original aim of the PhD, which is actually looking at land use changes after the war and how these are impacted by unexploded bombs and by the herbicide spraying as well. And then what I will be doing after the PhD, I'm not sure yet, but I'm hopefully hoping I can continue to study the long-term impacts of environmental impacts, but also socioeconomic impacts of conflict. The work I've been doing in Southeast Asia also shows that these kind of impacts can continue for decades after the war has finished. And actually satellite imagery is really a very unique and useful tool for these settings because, well, with conflict settings, obviously, accessing these areas on the ground becomes very difficult, especially during the conflict. And also it's very difficult to get objective information out of these areas. So having that source of satellite imagery that can provide some of this objective information is very valuable. And then hopefully I can continue looking into that in the future. Thank you very much to the USGS and the EROS Center for providing the imagery. As you noticed, it's been very useful for my research. ADAMSON: I'd like to thank Philipp Barthelme for joining us for this episode of Eyes on Earth, where we talked about his use of declassified imagery from the HEXAGON and CORONA satellites, distributed by EROS, to locate bomb craters and herbicide spray lines in Vietnam, Cambodia, and Laos. And thank you, listeners. Check out our social media accounts to watch for all future episodes. You can also subscribe to us on Apple and YouTube podcasts. VARIOUS VOICES: This podcast, this podcast, this podcast, this podcast, this podcast is a product of the U.S. Geological Survey, Department of Interior.