 Good morning, everybody. Thanks for your time. Thanks for the invitation. I'm sorry that I couldn't be there. I'm isolating at home with my no longer six children, but we're still home for a while. I'm going to talk a little bit about an application of my mobility and migration model to the coastal Bangladesh. I had thought I'd have a quiet four-hour plane ride and a quiet hotel room to get these together, and I didn't. So I'll do the best that I can. And I also will stress that I can in no way read the room. So in any case, if I'm going too long on any one topic or just need to keep going, feel free to shout out. So what I wanted to do for this talk in this group, given the sort of focus about extremes, was to tell a little bit about the modeling work we did in the paper and try and convey this one message on the people side, which is that for things that haven't happened yet, it's a real challenge to model what people might do because we don't have historical analogs. And we can address some of that problem by being good modelers and doing the best we can with data. And we can address other parts that just by being careful and smart about how we apply the model and how we talk about the model. So that's a key message that I'm hoping some of what I show will get to. And we just had a wonderful presentation that involved a coupled agent-based model. And so I'm going to guess, again, not reading the room, that many of you are familiar with bits of agent-based modeling. And in particular, the thing that you probably know is that they are models in which we have individual decision-making agents, whether they're people or firms or whichever, in which the properties that we're interested in at the system level emerge out of interactions among those agents and from those agents to their environments. And so they're really great tools for problems where we believe that that heterogeneity really, really matters and is driving our system's results. So the things that if you haven't done a lot of agent-based modeling you may not know are that we're still kind of figuring things out a little bit. We haven't converged in some of the ways that other modeling fields have on sort of universal ways of going about things, universal theories, universal algorithms for representing particular decisions. And to my knowledge, although I've been a part of a few attempts, there's no ab myth so far, no really good intercomparison that helps us understand what are the right ABMs for the right purposes. Another thing that's important to note is that because we're modeling individuals making decisions, doing a lot of different things, it's really rare that we have all the data that we might want to calibrate and validate the things that are happening in our models. And we make it work. We apply a range of techniques. We match patterns at different scales in order to try and reduce the issue of the equifinality problem where lots and lots of different models, different primer sets can give us the same outcome that we're looking at. And one of the last things that can be a bit of a challenge is just that as much as there's some intuition to what agent-based models are and we have individuals following rules that we can usually describe in a sentence, there's a lot of those assumptions to describe. The models get really big and it can be challenging to communicate everything that's going on at the model. And that is part of what really limits their utility. Because I think we all kind of know models aren't answers, they're outputs. And if you aren't able to think through what the set of assumptions underlying that model is, it's not easy to then make use of those outputs responsibly. So that's the sort of other side. And now that I've set things up like this, let me tell you a little bit about my model that, again, too complicated to explain appropriately and doesn't have all the data that I'd like. So the model Midas is a reasonably recent model in a long, long string of migration models that go all the way back to the 19th century. And the laws of migration and the gravity models are appropriated from physical principles to look at big cities as centers of attraction. Now, the thing about those is that they are using the city or the place as the unit of analysis and not the person. But we have now about a half a century or so of demographic theories that help us think about migration as an individual decision. And in the last 20 or 30 years, we've got the computer power to really explore those. So among the big theories that we apply to think about mobility now, particularly in the case of understanding migration decisions as a livelihood outcome, are the sort of push, pull and mooring theory. I've been around for 20, 30 years. And so Midas is an implementation of that. What it looks like is this, we have agents who are located in places. They are embedded in social networks with agents in the same places they are in other places. They're deriving utility from some set of options that are in their place, and we can interpret those as jobs or access to family and nature and use value. And yes, and they're sharing resources that they derive across their social networks. They're sharing information about the places that they are across their social networks. And periodically, they're making decisions about, am I making use of the best portfolio of opportunities for me? And if it's the case that the best portfolio for them that they know of, maybe it's in another place, if it's somewhere else, then they'll move. And so what makes this or what sets Midas apart from other models, at least at the time that I wrote it, are that it's a simultaneous consideration of pushes, pulls, and moorings. And by moorings, I sort of mean these sort of kinetic barriers to doing things. I can't leave because I can't sell my house. I need to stay here because I'm taking care of family. You can't beat these kinds of things. And what I like about it as well is that migration is a, it's an emergent strategy, among others. So it's not baked in that agents are deciding to migrate or not. They're making decisions about what's best for them. Migration is an emergent outcome from that. So the question that we wanted to address in our Bangladesh case was looking at enhanced sea level rise due to climate change and wondering how is that going to reshape migration flows towards the Bangladesh coastline. And here's what we had to work with. We had a few waves of the Bangladesh household income and expenditure survey, which is representative at the district level. There are 64 districts in Bangladesh. We had about 10 years partially overlapping of internal migration data in Bangladesh among those 64 districts. And we had a model dataset going out to 2100 of decadal worst case worst flooding predictions produced by climate central. Different pieces of a puzzle we wanted to tie together. And this is what we did. We built a minus application based on that household income expenditure data as kind of like a jobs and income access model. And we calibrated it as best as we could to internal migration data that we had. We then took those flood depth projections to get an idea of how would the worst case flood be changing over the coming century. And then this is the next piece we had to do. We had to translate those flood projections into some kind of a damage function. We had to try and find some idea of how is it that floods affect people, particularly in a space where floods are the norm. Floods happen every year. So we drew upon, there happened to be in 1998, whether the worst the worst cyclone driven flood in like the modern record in Bangladesh happened to sit between two waves of a panel study where the researchers were able to estimate the sort of wage damage effect of that shock. We're able to use that data as our best case estimate of what do we think future floods might do to wages both inside and outside of agriculture. The other thing is we have no idea what amazing things are going to happen in the economy over the next century how things are going to change. I will go downstairs with you soon but we're talking right now. Okay give us a few minutes. We don't know what's going to happen so we make the assumption that the economy doesn't change and there's really no better way to get around that for now other than to embrace that assumption and be aware of it. So what do we do? We scale our sort of simulated economy by the expected flood damage functions and then we look at what do we think might be happening to migration flows motivated by the decision model that I introduced earlier under different representative concentration pathways out to 2100. We look and see what the differences are and the the keys that we the key findings that we have are this. So we as you might expect the flood impacts on wages reduce migration towards the growth the coast but not enough to that it doesn't persist. So we observe under all rcps that we have persistent migration towards the economic opportunities that even with the damages that we can account for would persist in the coast of course thing other things may change but we don't have data on that. So what I like about this result it's a very agent-based model result which says look we're not telling you that migration is going to continue but here's this plausible story where it could continue which kind of cuts against this narrative that where there's a new and unexpected flooding people are going to leave. So let's be a little bit more careful as we think about things. Now within this set of results there are a couple of key things that matter there are lots and lots of different sensitivity analyses. One thing that matters is how much credit we make available to agents to make decisions and one of the interesting things is that more credit doesn't always lead people to move more in some cases it leads them to invest in place build themselves up and sort of become if not trapped certainly more moored than they might have been before. And then the other thing that really matters is how much this thing that we experience now is a shock then becomes a norm and this is a real challenge and it requires us to parse out you know what do people experience this uncertainty what are they experiences expected variation and how do those matter differently to the decision and we don't have a lot of good data on that we've all now just gone through over the last two years like the biggest shock that we'll probably experience in our lifetimes a lot of things have become normal like your kids joining you in presentations and it it's not obvious how universal this experience is and generalizable across different climate hazards different different things that are going to matter in climate adaptation research so this is sort of like a big unknown to try and address like I got to this one already so access to credit doesn't necessarily move people more. So again the things that are missing in all of this have no idea how we would expect coastal economies to change in the next century. We also don't have any real idea well on the physical side I think we might we don't have good data on how the experience of enhanced sea level rise and damages ensuing from that are going to match the things that we have historical analogs for so the sort of spatial and temporal correlation across Bangladesh for example of sea level rise is going to lead people to you know try and succeed through different strategies than you might see through a cyclone people you know you can you can move or you can rely on on insurance and other things to smooth damages in one way when the threat's in one place but if it's everywhere it's a very different experience so this is a thing again we don't really have great analogs we don't know as I said something that's a shock at first becomes really normal and so like across this as well as other sort of climate adaptation challenges like they're sort of out of bag problems that are going to be expensive for us to try and reduce which isn't me saying let's stop and not do anything we're going to get better at doing this we're going to be better at converging upon you know more universal modeling techniques and we're already getting better at you know not just generating data but making better use of the data that we have and I like this spectrum from like remote sensing on the one end through reliance on experts in the middle out to like the sort of like new exciting world of like high frequency mobile phone based data collection of like different ways to engage with with people and learn things we can learn a lot from being able to connect with a hard to reach populations the mobile phone it's expensive it's not necessarily easy to scale yet but it allows us to get to that point where we have like a true socioeconomic baseline that we don't have now we can talk about things in terms of anomalies in the same way that we can talk about sea surface temperature rainfall as anomalies at the other end we're getting so much better at picking up not just you know lights on lights off in cities from space but learning about whether not just that people have planted but maybe the people have switched crops or switched varieties we're starting to be able to pick up these things in smallholder plots from space which tells us a lot of behavior and thinking learning a lot about how to look at or pick out patterns of groundwater sharing and use from space and learning a bit about institutions so finding ways to be smarter about getting the data in the background that can support your models is going to be a thing that will lead us forward but we're not going to resolve everything and the implication of that is that when we think about the kinds of predictions we ought to be making from coupled models stepping out you know way into the future for things that haven't happened yet we've got to do lots and lots of big sensitivity suites sweeps we've got to end up relying on really large ensembles of potentially different valid calibrations and thinking about you know what's held in common what's you know seems to be universally true across all of these big ensembles are universally untrue and really you know I think this is always true for models of complex adaptive systems but it's particularly true with the climate adaptation challenge of like looking for qualitative insights that we think are likely to hold up and so this is a really important point I think to take forward when we make use of models in decision contexts for future projections so trying to sum that up the the application that I share was the best foot forward that we had with secondary data and the group of experts that we were able to work with and my group moving forward now is we have a similar application looking at climate mobility in Senegal we have another application looking at forest sorry at fire as a as related to deforestation in Brazil and we're hoping to do other applications stepping forward what I wanted to do in our time here was just characterize what is hard on the people side when it comes to climate adaptation problems and hopefully have a useful message about like what are the kinds of inferences that are appropriate when these kinds of models so I'll stop there I really appreciate you making time for us today and in case you have any interest in talking further you can reach me here and if I can hear them I'll take any questions Andrew I hope you're here to allow the plows there are there any questions for Andrew Jonathan thanks this very very exciting work you're doing there I'm curious as you're doing your modeling of the decision-making and the agents are kind of deciding what to do based on finding their best choice economically are you looking just at the kind of cash economy side or are you having something in there for like the hedonic utility that people like to stay in their village with the people they know yeah these are really great questions and I guess to some extent you know we can do anything we want if we're good coders right but the limitation is is at the data side and so to the extent that you have any kind of support for how one thing is valued against another we can put it in there and Midas is set up nicely to sort of plug and play like I want to consider this you know this form of utility and this form of utility and this form of utility whether you know whether it's like use value access to nature hedonic values cash incomes the trick is getting that data and where we run into so many problems is like imagining you're really wanting to engage with trade offs people make and you want to say well I think I know people care a little bit about money but they also care about time with their family and they care about not commuting and they care about their health but labor people collect labor data health people collect health data environment people collect environment data and finding ways to get you know perspectives on all of those different things from one person is the sort of critical task to better do the thing I think you're you're describing here thank you are there any other questions for Andrew if not I want to thank Andrew for giving a presentation as well as juggling with child care there and COVID so thank you thank you