Environmental migration in Bangladesh and agent-based modeling developed by Dr. Kelsea Best

[MUSIC PLAYING] Welcome to Through the Human Geography Lens Podcast. We're looking forward to sharing an exciting season 3 with you. Here's what we have planned. Another season of interesting and thought-provoking interviews are on topics related to human security. More highlights on openly available human geography data. And as we prepare for our next in-person WWHCD event in April, we'll bring in some topical discussants to better understand the ripple effects of the Ukraine conflict, which is the focus of that event. We have a lot planned, and we look forward to sharing it with you. We'd also like to invite you to be a guest on the show. If you have openly available data or tools to share with this community, please go to www.whgd.org today and apply. Hi, and welcome to Through the Human Geography Lens, a podcast brought to you by the World Wide Human Geography Data Working Group, or www.hd. I'm Terry Ryan, and I'm One of Holtz. And today we're here with our guest, Dr. Kelsey Best, a post-doctoral researcher at the University of Maryland, who studies environmentally-induced migration in Bangladesh. Kelsey, thank you for joining us today. It's great to see you again. Hi, thanks so much for having me. So we met Kelsey back in November at our climate migration and human security event at the Department of State. She gave a great panel discussion, and we were excited to bring her back to kick off season three of the human geography lens. So Kelsey, can you describe to our listeners your research? Sure. Yeah, thanks again for having me. So as you mentioned, I study environmentally-induced migration in Southern coastal regions in Bangladesh. And I do that by working with really interdisciplinary teams. So on this project, specifically, I worked with social scientists, and psychologists, engineers, natural scientists, and modelers to try to understand the really complex dynamics of how environmental change interacts with human decision making around migration. So my role in that project, I like to think of myself, as kind of the integrator of those different sources of knowledge. And I do that by using a range of quantitative data driven and modeling approaches. So a lot of my work has been using social survey data with machine learning and agent-based modeling to try to really understand these multifaceted, multi-spatial temporal dimensions of how people decide in the region to migrate or to stay in a place that might be impacted by environmental stress. That's really interesting. Yeah, what kind of data do you use? You mentioned these surveys. And then can you also remind us what agent-based modeling is? Sure. Yeah. So I was lucky coming onto this project because I was kind of given a very rich social survey data set that had already been collected by some colleagues of mine. And they went to more than nine communities in the Southwestern region of Bangladesh and interviewed more than 3,000 individual people about their migration histories. So across their entire lives, how many times they had moved, a little bit of information about why they moved. And then a lot of data about their households, their employment, even things about their perceptions of their communities and environment. So it was this really large, rich data set that I had to work with from the start. So then part of why I applied some machine learning to that data is because I was relatively new to the project and wanted to use this kind of data-driven approach to understand, can we have all of this information? What is actually important if we're trying to predict migration and mobility? So from there, I did just some exploration of the data, which then informed the agent-based model. And to your question, agent-based modeling is this really broadly applicable technique of modeling that essentially focuses on behaviors of individuals or agents? So a simpler example could be, if we're interested in studying how geese flock, your individual agent would be a goose. And you can see if they're behaving a certain way and have a certain set of properties, then what kind of formations? We call it emergence emerge from that individual behavior. So in my case, the individuals are individual people and households within this kind of simulated Bangladeshi community. And I can code those agents to have their own characteristics and make decisions about migration, which then allows us to see, OK, if individuals and households within this community are behaving a certain way, what larger level patterns of migration do we see emerging from the model? So I just have to take a step back, because we're talking about the work that you're doing bringing all of these groups together to understand data. And then having to apply machine learning models because this is a vast data set, how did you get started? I mean, when you went to college for the first time, what were you studying? And how did that evolve to where you are today? That's a great question. So if we go way back in my undergraduate program, I studied chemical engineering, actually. Oh, wow. Yeah. And I did that because I loved math and science, and I wanted to solve problems. And I thought, why not go for chemical engineering? I very quickly learned that that was not my passion. And so while I was in undergrad, I found I kind of discovered environmental studies and took a range of classes and ended up getting a minor in environmental studies. But it was things like the history of environmentalism and intro to climate justice. And then some of the more climate science courses. And it showed me that there are so many pieces to this puzzle. And what motivates my work is really that climate change is a human issue and a justice issue at its core. So that was just very eye-opening for me. And I had a bit of a non-linear path and it upworking in industry for a couple years. And then coming back to graduate school, when I did that, I knew I wanted to stay. I was really interested in this idea of human dimensions of climate change and especially how what we call vulnerable groups might be impacted by climate. So I found the program at Vanderbilt that I joined that the project in Bangladesh seemed like a really good combination of all of those interests that I was just starting to develop. And then like I mentioned, I was handed this huge data set, essentially on day one, not knowing really much at all about climate migration or general migration. So the machine learning was in some ways my approach to just starting to get familiar with that data set and really saying, OK, I myself and I think generally in the field and the discipline as well, I don't have a great idea of which variables out of this set of 2,000 variables are the most important. So let's ask the machine to tell me and then go from there. Wow. So then when you did ask the machine to tell you, what kind of variables came out as being the most important and did you learn anything surprising about these individuals through that process? Yeah, I think we did learn quite a bit from that. And one of the things that was maybe unsurprising was that there was a wide range of variables that emerged as the most important just statistically in terms of predicting our migration outcome. And it was a range of variables that some were environmental. So I remember that past cyclone experience emerged as an important predictor in our data set for number of migration trips. So that indicated some kind of relationship between a natural hazard and mobility. But then we also had a lot of more economic variables emerged. So things like, does your household own a business? If so, when did that business start? And then other factors that we interpreted as kind of coding for socioeconomic status or wealth as well, which were like, do you own a refrigerator, or do you own a gas stove? And in that area, that would be a sign of a level of wealth. So that exercise could show us what variables were mathematically important. And we could start to kind of lump them into, maybe this is environmental, maybe this is economic. But what it didn't tell us is how those variables were relating to migration. And that's because we used an algorithm. And this is kind of a general challenge with these really advanced methods. That they can be a little bit black box. So we used an algorithm called a random forest model. That's powerful because it can get at these really complex relationships between variables and between predictors and outcomes. But then we don't have a great way to lift the hood and look at really what's going on because it's so complex. So that's where the modeling kind of came in as a next step. Because the first step was OK, what matters? And the second step was really, can we tease out why and how these things matter? So in all of this research, was there anything that surprised you? Yeah, so something that surprised me that's one of our main results from the agent-based modeling work and was really a non-intuitive result is that we found that the degree of inequality within an origin community-- specifically, we were looking at the distribution of land ownership, which in Bangladesh is a really important signifier of wealth. We found that the level of inequality in that land ownership-- so how many households were potentially dominating the amount of land owned was really important for us to be able to replicate the larger level patterns of migration that we were interested in. So basically, we found that when there was more inequality within the community, meaning fewer households owned more of the land, going to dominated more of the market, then we were able to see what we expected based on what we knew from the region. And that was something-- and I think is something that hasn't been explored quite so much is we talk more about perhaps household or individual resources and wealth. But what really mattered was how that was distributed across the larger community. So that's something that we still are trying to explore further. But it was a cool finding from the model that we weren't necessarily expecting. And again, like you were talking about at the beginning when you had this huge data set, making that connection would have been something that you maybe would have done without your model. Right. Yeah, exactly. And so that's why I am kind of a personal proponent of, like, let's look at all the tools we have and see how we can use them and use them in combination and learn different things from those different tools. So speaking of tools-- and I mean, I know you talked a lot about that really large data set. So is that raw data set available publicly? Or is that something that you and your team had accessed to just for this project? Yeah, unfortunately, that data set is not publicly available. And that was data that my colleagues, Amanda Kerrico and Catherine Janado collected over a couple of years. And they're still building on that data. They're still doing new phases and collecting more. So I don't own that data. I'm hoping it would be publicly available at some point. But I will say that I also benefited from a data set, a similar large social survey in Bangladesh that is publicly available that was collected and published by Helen Adams, a researcher who's very prominent in migration and non-migration research. So I benefited from that data being out there. There were a couple times when I was developing my model where my data set didn't have quite the right question or the right variable that I needed to incorporate into the model. And I could go to Dr. Adams' data set. So I think it was a great example of having that data publicly available and benefiting from it. Yeah, that's awesome. Yeah, one of the things we remember from your presentation at the Climate Migration and Human Security Symposium-- and it really is stuck with us-- is you mentioned that people actually want to stay and not migrate even if there's extremely challenging situations. So how do you think the community of researchers working on this topic can begin to gather that information into data? Is it like survey data? Or how would you get around that? Yeah, so I think this is a direction that I see the field going, which is understanding that perhaps some people do want to migrate. And I think maybe that was even a shift at some point in the field. At first, we were talking a lot about displacement. And there was this idea of people being forced to move, perhaps not wanting to do so, which of course happens. And then we started to think about this voluntary migration or maybe migration as adaptation. So people actively choosing to migrate to better their circumstances. And I think the piece that is still a bit underdeveloped is this question of what about people who don't want to move? And that might be the majority of people in certain settings. And that's because it's hard to move. And people are tied to their homes and have history there and culture and family ties, so this kind of deep place attachment. And I think that there's different ways to collect data to get at those questions. I know I have a colleague, Dr. Bisha Jeet Mali, who works in Bangladesh as well, who is very interested in what he calls non-migration and has started to ask more questions around migration aspirations or migration intent. So trying to get at that, do you want to move or not? Did you move or not as an in-between step whereas a lot of the data I've worked with has just did you move or not without getting at that intent? So I think there's still a role for social surveys there. Absolutely. I think, as I've talked about my agent-based modeling, which an important piece of that is the decision-making process. So I am really interested in maybe finding new ways to collect information about that process of decision-making. There's some kind of interesting tools out there, like more participatory methods. I've thought about even developing something like a board game where I could take this into the field and have people play out different migration decisions and explain their thought process and collect information that way. So yes, I think that there's just a lot of opportunity in that space and we might need to get creative with how we collect that data. You know, another thing that you mentioned that's related to this discussion during the event was this idea of narratives as data. You talked a lot about this quantitative data but there's so much qualitative information out there about people's decision-making processes, like you mentioned here. And it's really like stories about people. And I think that for us with the W.W.A. Shidi focusing on human geography, that's really like we like to say its data about why people do what they do, where they do it. It's a really great quote from Roland Ellic. And you know, it really makes me think about your idea of narratives as data because that's really what human geography is about trying to understand these very A-spatial, sometimes things in a spatial way. So how do you start to figure that out? Like what kinds of mechanisms are you using to crack that? Have you seen some exciting new stuff that you can talk about? Yeah, I think this is something that I started to think a lot more about when I was really in the weeds of my modeling and my quantitative data. I had kind of a personal moment of like, wait a minute, this is people. And this is people who are living this right now. And I found myself a little bit removed from that for a moment. So I really wanted to come back to that and found some internal funding at Vanderbilt to just collect stories. And my plan was to have them be publicly available, put them out there. As just, let's share that this is happening now. Let's give people who are going through this a bit of a chance to be seen and try to elevate some of these voices. And so I did work with a colleague in Bangladesh to talk to just 10 current environmental migrants who were living in a informal community in DACA, the capital city of Bangladesh. And it was unstructured. We had a couple of questions kind of like this, it was conversational though. And really just asked people to tell their stories of where they came from and why they moved and what happened and their future aspirations. And just listening to those stories was so informative. I got, there's so much information in one person's story. I remember a man who described that he was a fisherman in his hometown got caught in the middle of a cyclone, described this incredible story of drifting at sea and eventually finding his way home. And his house was destroyed. So he decided to move and explained his experience in DACA, kind of challenges the black of electricity. And there's just so many things that you can tease out from a story like that. So I would say that that exercise has been informing a lot of my research, even the questions that I'm interested in asking now. And I'm very interested in finding ways to maybe more formally incorporate those kinds of things into modeling. So I mentioned this idea of participatory modeling. And there's different techniques for that, everything from a board game that I described, which is something I hope I can do at some point. Or even from the beginning co-creating these models with people who have the lived experience and iterating over models with this kind of continuous feedback. So I think people are thinking about very creative, exciting ways to do this. And I personally am just kind of scratching the surface of that. But I think we are missing out on a lot of information if we limit our definition of data to spreadsheets when these stories just have so much useful and interesting and important information. >> I do recall now you're sharing some of those stories at the event at Department of State and they really just touched the heart. And I remember the man caught in the cyclone. I mean, didn't it take him like weeks to eventually make his way back home? I mean, it's yeah, to be able to get to the stories behind this data. So you talked a lot about models and the WWE HD is very interested in openly available data, models, techniques. Where could some of our listeners find this model? >> Yeah, thanks for asking. And my model is publicly available and it's in Zanoto, which is just a repository. But you can find it on my Google Scholar page. There's just a link to it right there. It's in Python, but anyone can download it and play around. I tried to code it in a way that it kind of has this building block structure. So people could take pieces of it or hopefully alter it to context that they're interested in studying. I'm also very open to feedback on it if people do play around. >> And very good. >> Thank you. >> Well, we'll make sure that, yeah, we put that in the show notes and make sure we have all the other links to any data sources that you might have that we might be able to share. So thank you, Kelsey. This has been so wonderful. We love this conversation with you and just to see you again after seeing you a couple months ago. And we wish you so well and all your research and your endeavors. >> Thanks so much for having me. >> And to the audience, thank you so much for joining us. Please join us next time for another conversation on human geography and human security on through the human geography lens. Our human geography library has more than 5,000 data sources and references, including data applicable to environmentally induced migration from the international organization of migration, World Pop at the University of Southampton with population distribution, demographics, and dynamics data, and data from the internal displacement monitoring center. If you're interested in learning more about human geography and the WWHD, check us out at www.hd.org, where you can find the human geography library and access presentations and recordings for more than 50 data-driven events. If you have any interesting, openly available geospatial data to share, feel free to contribute your data at www.hd.org. I'm Gwyneth Holt, and I'm Terry Ryan. Thank you for joining us and we hope to see you again next time. We really appreciate your support. If you enjoy this show, please take a moment to leave us a review and a rating on Spotify, Apple Podcast, or your favorite podcast platform. And we hope you'll share the podcast with your friends on social media. Thanks again for listening. [Music]