Statistical Mapping | Stats + Stories Episode 99 / by Stats Stories


Dr. Caitlin Kontgis is a technical lead on the Applied Science team at Descartes Labs and was honored at the New Mexico Tech Council’s Women in Tech celebration in March 2018. Since moving to Santa Fe, Caitlin began volunteering at the local Girls Inc chapter and joined the board in fall of 2017. She is passionate about supporting women in STEM fields and scientific literacy. When not at work, you can find Caitlin running, skiing, and hiking the Sangre de Cristo Mountains

+ Full Transcript

Rosemary Pennington: Scientists announced a shocking discovery in 2019. The loss of ice in Antarctica had devastated the second largest Emperor Penguin colony in the world, and it was unlikely the colony would bounce back from the loss of territory or the loss of population. The story was just one more example of the way the earth’s landscape is changing. Scientists discovered the loss of both the ice and the colony through the use of satellite survey technology. Using satellite technology to map changes to Earth’s terrain is the focus of this episode of Stats & Stories, where we explore the statistics behind the stories and the stories behind the statistics. I’m Rosemary Pennington. Stats & Stories is a production of Miami University’s Departments of Statistics and Media, Journalism and Film, as well as the American Statistical Association. Joining me in the studio are regular panelists John Bailer, Chair of Miami Statistics’ Department, and Richard Campbell former and founding Chair of Media, Journalism and Film. Our guest today is Geographer Caitlin Kontgis. Kontgis is Applied Scientist Lead at Descartes Labs, where she studies changes to Earth’s landscape using various types of satellite technologies. Caitlin thank you so much for being here today.

Caitlin Kontgis: Thank you for having me!

Pennington: As I mentioned you’re a Geographer and you’re using satellite tech to sort of study, I know changes to the environment. How did this become what you do?

Kontgis: Yeah, that’s a great question. I went into college thinking I would study English Literature, and I was at UC Santa Barbara, I took a Geography class and just really fell in love with it. They have a dynamite remote sensing program at Santa Barbara and I really fell in love with the technology. This was before Google Maps was really pervasive, and I thought the ability to see the world in this way was just so cool, and I went on to get a Ph.D. in Geography focusing on machine learning and remote sensing, or satellite image analytics. And then I wound my way to Descartes Labs.

John Bailer: That’s really an interesting combination of skills that you’re describing. Can you help us unpack a little bit, what exactly do you mean by these remote sensing data and what kind of machine learning tools are using with this data and towards what end?

Kontgis: Sure, as far as remote sensing data, so remote sensing can really mean a variety of things it just means that you’re measuring something by taking a temperature with a thermometer not remote, but any sort of weather sensor that’s out, any sort of inground sensor is a remote sensor, but the remote sensing that we do uses satellites so they are obviously remote, up in space and imaging the earth every day. And so that’s really my bread and butter, looking at the satellite imagery and using it to better understand Earth’s dynamics. As far as machine learning, we use a variety of different algorithms there, but essentially what you’re doing is automating a process to really learn from the data, and fed in different features that can help- where you’re sort of telling it what’s important and what to be looking at, and then it will take those features, learn what trees look like, for example; learn what water looks like, and then be able to scale that , to map that across the entire globe. Can you expand on that example just a little bit? For instance, the idea of what a particular problem that you’ve worked on with remote sensing data, and you classify it using these tools.

Bailer: Yeah, I was going to ask you about the rice patty stuff, which I thought it was a good example of one of your earlier studies. So how do you map a rice patty?

Kontgis: Yeah that was going to be my example, s that’s perfect. So, rice is a really unique crop, it spends part of its growing season completely submerged. So, in terms of training an algorithm to understand that I would feed in data related to wetness, or water indices, you can derive from satellite imagery. Rice also then gets very very green. We’ve all seen these beautiful photos from China or Vietnam of rice patty fields. So, feed in a similar index on vegetation. Rice tends to grow in low lying areas, so we can feed in information on elevation, temperature, it tends to be a warm weather crop. and if you feed all of this data that’s related to the crop into an algorithm and then you also feed in example points of, in this case rice, or other types of landcover that are not rice, it will go to each of those rice points and say what elevation they’re growing at, what the wetness is, what the greenness is, what the temperature is in that region, and say “okay, these are the parameters that indicate rice”. And then it can go to every other point and build a decision tool and give a probability of whether that their location is rice or not rice, based on what it learned from these specific rice points.

Richard Campbell: I noticed in some of your work you mention some of the problems that maybe cloud cover might cause, or snow, some of the problems that you face in satellite imaging like this.

Kontgis: Yeah, clouds are definitely a big problem with satellite imagery. So, you can imagine these sensors where it’s circling the Earth and it’s basically just up there with a camera taking a picture, and so if clouds are in the way you aren’t getting any look at that landscape. And so, with the work I’ve done with rice, that’s a huge problem, because rice is grown in tropical areas, so during monsoon season, you might have six months where it’s just cloud cover every single day. There’s- it’s such a boom of data right now, and there’s all different sorts of sensors going up. Commercial, government, one sensor I’m really excited by is the European Space Agency Sentinel 1 Satellite. And so, this is a synthetic aperture radar. So instead of just passively circling the earth and taking a picture, it’s emitting energy. So, it’s actively sending energy to the surface of the earth and measuring how much comes back, and the long wave microwave radiation it’s sending out can penetrate clouds. Bailer: Wow.

Kontgis: So, you can get data day or night, you can get it in all sorts of different weather. It’s a little noisy, you know it doesn’t come back where grass is green, and water is blue, that sort of thing. It’s- you have to sort of tease signals out of it, but you aren’t getting six months of data loss during a monsoon season.

Bailer: Very good.

Campbell: So how do you evaluate whether your models are any good?

Kontgis: Yeah that’s a great question. In an ideal circumstance we either work with a data vendor who has ground truth data, so somebody who has maybe labeled- they actually the United States is a great example. We have amazing ground truth in the United States. We have the USDA collecting data on what different fields are each year. Just going out and doing massive surveys. And so, we would take that data and we might set aside say 70% to train a model in the way that I described, where you’re saying this is what corn looks like, this is what soy beans look like and then we hold out 30% of the data to validate what we’ve done. So, we’ll take a model, take this other 30% of data and compare our model to what that truth is, to evaluate how well we did.

Pennington: I was just going to say, so you’re mapping these rice fields, why should anyone who is not at Descartes Labs care that you’re doing this?

Kontgis: rice is such a hugely important crop globally, 20% of the calorie supply globally is from rice. Rice is directly consumed by people, unlike corn or soy, which might be turned into derivative products. And rice is also really susceptible to climate change. So, I mentioned it’s grown in really low elevations. Those also tend to be river deltas where there’s salinity intrusion from rising seas, there’s total inundation from rising sea levels. They also tend to grow rice patties at right at the maximum temperature threshold, so any increase in temperature is going to decrease the area in yields of these crops. So, I think it’s massively important if you’re thinking about food security globally, how to distribute resources, where can it grow in the future. So, rice, specifically, I think is a pretty critical crop for people around the globe.

Bailer: I have a question did you- like I learned from reading this research these rice crops in Vietnam anyway, there’s three times a year you can get a rice crop. Which is very different than what I’m used to thinking about in terms of agriculture here. Did you know that going in to this study, or is this something you track in your satellite data?

Kontgis: Yeah, it’s something we track in satellite data, so you can look at how many times a year it’s planted, and that was actually the foundation of my Ph.D. research. Vietnam, as we all know, used to be a pretty closed economy. In the late 1980’s they instituted reforms to open the economy, and at that point these new technologies can come in, and Vietnam went from being a net importer of rice, to a net exporter, because they got these shorter duration varietals of rice. They got mechanized technology to grow rice in the Mekong and Red River Deltas primarily. And so, what I as trying to track was how these economic reforms impacted the landscape. So, rice went from primarily being grown once or twice a year, to two or three times a year and using satellite imagery you can track that through time and access how things are changing on the ground.

Bailer: You mention that one of the concerns associated with the levels of sea changing like the increased salinity of some of these-, the water sources. Do the remote sensing tools have sufficient sensitivity to detect a subtle change?

Kontgis: Yeah, I think it depends on the change. So, certainly rising sea levels, you can look- it’s a very slow process, right? So, it’s not going to be over night that you see a change but looking at decades of data you can see how coastlines have changed, and how seas have risen. Looking- thinking about… I wish it was as simple of an answer as “if sea level rises one meter, x-amount will be underwater because it’s below this one meter level”, but it tends to be pretty complicated because it’s- you know these river deltas where it’s not just the rising seas, but there are construction in pace to block sea level rise, there’s dykes and levees, there’s also sea level or salinity intrusion, so you can see it in satellite imagery, it’s just you might need a pretty large time series of data in order to really assess the impact.

Pennington: You’re listening so Stats & Stories and today we’re talking to Descartes Labs’ Caitlin Kontgis, about her work, studying Earth’s changing terrain, particularly using remote sensing and satellite technology. We’ve been talking about your work Caitlin, looking at these rice fields, but I was wondering if you had something else sort of on the pipeline, or something you’re really excited that you’re working on where you’re sort of applying this technology to a new space for you?

Kontgis: Yeah, definitely, we have a lot of different projects happening at Descartes Labs, but one of the ones that I’m most excited about is a person on my team has really been pushing forward looking at wildfire detections. So, there is a satellite up in space called Ghost-16, and it is a geostationary satellite, so instead of circling the Earth it just stays in one place and it is taking imagery of the United States every five minutes, and so it’s really the closest thing to real time streaming that we have. It’s fairly coarse resolution but we are developing an algorithm to- it has lots of new infrared bands. So, we can look at heat signatures. So, we are trying to see if we can detect wildfires earlier than they’re reported and then be able to get that information to first responders, and people who are in a position to help out during those situations.

Campbell: That’s really cool, so when you say coarse resolution, what does that mean in terms of the area?

Kontgis: So Ghose, so the radar imagery that I mentioned before for mapping rice patties that’s at about 20 kilometers- uh sorry 20-meter resolution, Ghose data is at a 2-kilometer resolution, so it’s quite a different scale that we’re looking at. That being said, we can detect fires that are smaller than two kilometers because a sub-pixel change will often result in a change for the whole pixel. So the Camp fire that affected California last year, we’ve been developing this algorithm and we were actually able to detect that when the fire was at about ten acres, that’s still- it was reported, by the time it reached ten acres it had been reported to first responders, but we’re based here in New Mexico, where there’s a much more sparsely populated state. So, we have fewer people looking at the horizon, fewer people looking out for fires to report them, and we have been able to identify fires before they’ve actually been reported.

Pennington: I was going to ask you, are you working with people where you provide this data, that help some sort of intervention? So, you’re doing this work on the patty fields, the rice patties, you’re doing this work now with these wildfires, gathering this data, but then are you turning it over to people who then use this data to create interventions of some kind?

Kontgis: Yeah, that’s absolutely the goal. It really depends on the project. So, wildfires, that’s such a sensitive issue. We don’t want to detect fires if they’re not actually there and cause a panic, so we’re trying to refine that algorithm as much as possible before we sort of release it into the wild. But with other projects we are really connecting with people who can do something about this. So, we work with various MGOs for the rice patty work specifically, I have contacts in Vietnam at different Universities and government agencies that I worked with throughout my Ph.D. and remain in contact with I’ll actually be there in June for a coupe of weeks. So that really is a goal of ours, to not just develop this in a bubble and use it internally, but actually be able to connect with folks who can help implement changes, whether in policy or in response, but really put our work to good use.

Bailer: So, I’m curious what’s the most surprising insight that you’ve gleaned as a result of analyses that you’ve worked on? Something that was a real “a-ha!”, or “I didn’t expect that”?

Kontgis: It’s a hard question to answer, because usually going in to a project we do quite a bit of literature review and reading, so, we don’t want to be totally surprised by data analysis. I think one cool project that we’ve been looking at is deforestation. So again using this synthetic aperture radar data that can penetrate clouds, so in tropical regions where deforestation can be pretty rampant, using that data to better understand where deforestation is occurring, how often and we’ve detected it in places where there are bans on deforestation and we’re still seeing it occur, so that’s been pretty fascinating and I think that’s really the beauty of satellite imagery, is that you can be this environmental watchdog, because you have eyes on the ground, whether folks want you to or not. So, you can really see what’s happening and what things are changing.

Bailer: I’d like to switch gears here and talk more about your career and you’re in a field that I suspect is dominated by men, what have the challenges been for you as a woman geographer, and I don’t think you’re teaching now, right? Or maybe you’re mentoring but, do you feel any sort of obligation to help younger women get in to science?

Kontgis: That’s a great question and I really appreciate you asking it. The first thing I want to say is that geography is such a broad field, it can really range from almost sociological research to really technical, and so I think there are gray areas in there but for what I am doing in this more technical field, it is certainly dominated by men. I am not teaching I am mentoring. I’m a team lead so in that way there’s some level of mentoring and I’m really enjoying that. I work with a local organization called girls Inc. and so that’s one way I’ve approached this., so Girls Inc. is a national organization but there’s a local chapter here in Santa Fe that’s really impressive and really serving the community. Descartes, I worked with our internal team and we got laptops donated to Girls Inc. and then worked with them on coding projects at a junior high school level, and we will be going back this summer with a few of the women here at Descartes to do more programming work during their summer programs. And internally I think it’s just trying to advocate for people and really promote the good work that women are doing. And I think being in a male dominated field it can often seem intimidating to speak up, and I really try to be that voice for women on my team if they’re feeling intimidated in this space.

Bailer: Very good.

Campbell: So how do students prepare for a career such as yours?

Kontgis: yeah, good question, I think finding what you’re passionate about I think was my main take-away. I really didn’t think this is what I was going to wind up doing, and just going into college with a really open mind and letting yourself find what you’re really god at and what you love to do. And I think for this specific field, I think getting competency in some sort of coding language is really useful, really just trying to take a variety of different classes. So, I’m mainly technical but I mentioned my Ph.D. was also looking at economic reforms in Vietnam, you know really being well-rounded in the coursework that you’re taking is, for geography I think pretty critical.

Pennington: Caitlin, so when I worked in Birmingham Alabama, I got the chance to talk to Sarah Parcack who is an archaeologist who uses remote sensing in her work, and actually she partnered with a public school of health- the school of public health- to do some remote sensing for them so I know there’s a lot of different applications for the technology that you are working with are there other areas that you think are underexplored when it comes to using this technology?

Kontgis: Yes, definitely. You know we do a lot of land cover and land use change work and that’s what I did during my Ph.D., but that field- even that field which has been pretty much explored which becomes more and more interesting with new datasets coming online. One source of data I’m really excited about is called Lidar. And so, it’s similar to radar in that it’s actively getting collected, but it’s looking at canopy density, so you can then get 3-D structure of forests, crops, things like that. To date there has never been a space born lidar, it’s all been aerial and flown on planes, it can be expensive to collect, it’s – you’re not getting the whole earth. And then recently a sensor just went aboard the International Space Station. So, it’s currently collecting data and it will be made available, but I think looking at this more 3-D aspect of the Earth will be really game changing, as opposed to just the 2-D images that we’ve been getting.

Bailer: Can you describe a situation where geospatial information and data and associated analysis have really impacted policy in some dramatic way?

Kontgis: Sure, I- there are a few good examples. I think a great one is deforestation. So, the rainforests in the Amazon was getting clear-cut for agriculture for many years, I think then being able to look at it and monitor that from space was a really great way to affect policy, to have that slow down. I mean, if we think about very specific events, so hurricanes during Katrina, after any big event your flying imagery and you can really get a sense of the scale of an event in a way that I really don’t think you can if you’re just boots on the ground. So, I can’t think of like a specific bill or anything like that, but it’s certainly affected different policy, I think particularly related to environmental protection and being this sort of environmental watch dog that I mentioned earlier.

Bailer: You know one thing I was curious about is al these geospatial tools that we encounter in our daily life. You know I was thinking about looking at my Maps, or we see so easily, or looking art the weather pattern maps that we encounter, are there any surprising apps that are using geospatial information that we may not be aware of?

Kontgis: I think every app is using geospatial information to a degree, unless you opt out of location services, it is tracking your movement which is inherently geospatial. Other than like yeah, you wouldn’t necessarily think you’re bank ap is tracking where you are, but it certainly is, and so I think it’s all of them.

Bailer: So, I got to ask you to do you turn off your geospatial location on your phone?

Kontgis: I don’t. I’m not concerned, I don’t think anybody really cares where I’ walking around Santa Fe or grabbing a beer or going grocery shopping and if they do I don’t think there’s anything menacing in what I’m doing, but I totally understand the privacy concerns of course, and why people would want to opt out of it, but I don’t personally.

Pennington: Well Caitlin, that’s all the time we have for this episode of Stats & Stories. Thank you so much for being here.

Kontgis: Thank you so much for having me!

Pennington: Stats & Stories is a partnership between Miami University’s Departments of Statistics, and Media, Journalism and Film, and the American Statistical Association. You can follow us on twitter, Apple podcasts or other places where you can find podcasts. If you’d like to share your thoughts on the program send your email to, or check us out at, and be sure listen for future editions of Stats & Stories, where we talk about the statistics behind the stories, and the stories behind the statistics.