Understanding Cities - Part 2

In part one, you heard all about the key concepts of complexity science and how they apply to cities. Well, in this episode part two, we are joined again by Michael Bade, professor of planning at University College London. In this episode, Michael's going to talk about how we put all these concepts together and model cities. This is simplifying complexity. A podcast where we explore the underlying principles of complex systems. Systems that seem to defy our rational view of the world. Like economies, ecologies, or even you or me. I'm forensic engineer, Sean Brady, and I'll be your host. So these laws, you can list of a bunch of these laws like Metcalfslaw, Westslaw, Brandslaw. Do you want to just talk just a little bit about a few of those, Michael? Yeah, I mean, in some senses, complexity theory has thrown on to the agenda. And I should say that this agenda is not specifically peculiar to complexity theory. It was there before complexity theory, but there are a number of different regularities, we should say, in the way things are distributed in cities and society, which have acquired the status, these regularities have acquired the status of laws. So for example, Metcalfslaw basically says that the importance of a network depends on the the number of nodes and the number of connections. And as we get more nodes basically, then the number of potential connections increases as the square of the number of nodes. So if you think of 10 people, you've got potentially 10 squared or 100 interactions. Assume you interact with yourself and you interact with all of the 99 basically. Each of you in the group of 100 is interacting with your self and also 99. So you've got basically 100 squared, which is 10,000 possible links. If you have 1,000 people, you have a million possible links basically. So as you see that as the number of nodes or individuals or objects increases, then the amount of interaction increases as the square of that. Now, if you're in a big city, you have potentially a lot more numbers of people who you can interact with than if you're in a small city, basically. If you're in a village of 10 people, you can only interact with the other 10. So you can have 100 possible links. But if you're in a city of a million basically, then potentially, Ilco, we never realize all of these. But potentially you could have a million times a million potentially interactions. Of course, we could never realize these. You only have a, I think Robin Dunbar who's an anthropologist at Oxford, basically says that Dunbar's number is something like the maximum number of people in our acquaintance network is about 250 or something like this. In fact, I think you would probably find that the maximum number of people in the acquaintance network in a village is much smaller than if you lived in a big city. We don't know. I mean, there's a lot of work done. Now, that's an example, if you like, of Metkoff's law. So Metkoff's law is very much related to the number of things, basically, and the size of things in that particular context. Now, there's lots of other laws which really relate to things such as the density of activity in cities that there's a set of laws that really relate to all cities, basically, or more than one system of cities. And there's lots of laws that relate to individual cities. And in individual cities, one very widely known law is that the density of population or lots of other things, too, such as industry and so on, the density of population drops inversely with the distance from the centre. So if you think about Brisbane, you've probably got higher densities in the centre of Brisbane than on the edge of Brisbane, basically. And that's pretty universal. And that's really a power law. It can be a power law. You can literally use the power law equation to actually model the density of Brisbane, basically. And essentially, that really has been observed for maybe, you know, 100 years or more, where it certainly goes back to, well, an Australian, Colin Clark, basically, Australian economist, wrote a very famous paper in the 1950s, quite the '51, called urban population densities. And those sorts of laws are essentially fractals in that sense. Now, West's law, as I said, some of these things depend on what goes on inside a city and others depend on lots of cities between cities. So West's law, basically, which is probably also referred to as Marshall's law after the economist Marshall, really relates to agglomeration economies. As cities get bigger and bigger, they increase in wealth per capita. So, if you want to optimise the amount of wealth that you can gain control of, then live in a bigger city, small cities tend to be less rich on a per capita-based system, big cities. That's a very sweeping generalisation, I know. It's called a lomatory. It changes in the quality of size of something. It changes in the size of something as something gets bigger, basically. It's the notion of the fact that if you're a mouse, basically, your heart rate works at the same rate as a human or something, and therefore, because of that, the size of a mouse is tiny, and its metabolism works much, much faster than ours does, basically. And there are scaling relationships involved in how animals function different sizes, and what size can be supported for a certain kind of metabolism. All of that really relates to what's called a lomatory, but all of it really relates to the idea of fractal structures. These systems will only work if they're fractured, something like that. And that's presumably Jeff West, who gives them such West's law, effectively in the economic context of wealth in cities, then it really relates to the economist Alfred Marshall who's lived in the late 19th century, who's an English economist, very well known. But Jeff and the Santa Fe group, basically, which we should mention because it's central to the emergence of complexity theory in general. But Jeff pioneered this idea that if you look at different sizes of cities, you see different degrees of wealth per capita, in that sense. Yeah, we had Jeff on the show. We're back near the beginning, actually, and it was a real three. Just one more law because this one sounds fascinating. Brands law. The bigger a city gets, the greener it gets. Some of these laws are contingent. We have to make a distinction between the industrial city and the post-industrial city. And most of the growth in cities now is in post-industrial cities, or it's a post-industrial nature, and the still a large residual effect of the industrial city. But generally speaking, in cities that are dominated by changes in the post-industrial world, then they're getting cleaner, cleaner and greener in that context. And you can really begin to see this in contemporary life in the last 20, 30 years with the emergence of active travel in cities, a lot of concern for how we travel, what we eat, a whole range of different things in that sense. And so it's speculation. I mean, all of these things are, they're not really laws. I mean, I probably don't have to tell your audience that these are actually laws in the mechanical sense in terms of Newton's laws and so on. These are speculations to some extent. Although some of them are really quite strong, what's called Zip's law, which is if you take the largest city in a nation and divided by two, you get the value of the second largest divided by three, the third largest in song, it's called the rank size rule, that's a power law basically. Those sorts of laws really pertain to a very wide range of different objects in a language, for example, different words within a language. So again, they're all examples of power laws, they're all examples of fractals in this sense. So when we take all of that Michael, the other fractals, the parallels, the scaling, the emergence, positive feedback, all of those things and start to put them together and use them. How do we actually go about this business of modeling cities and understanding cities? We should probably make quite a big distinction between what we've been talking about, which is largely theory, but it's not theory in a vacuum. I mean, there are lots of examples that one can think about. Most of the things we've talked about are ideas that help us to inform how cities work and how they might work in the future and so on in this particular context. If we're in the business of trying to actually produce definite plans for cities using some of these ideas about what we call city science or urban science, sort of catch all phrase for these things, then we really need to encode them into forms where we can make predictions, basically. Conditional predictions, I mean, I'm very much a strong believer that we cannot predict the future anyway. All we can do is invent it and guess it and so on. But we can still use these tools to inform us about what some of the key problems are and the models that have been built and are built. being built slowly and surely with lots of problems, of course, to make these models useful to us. The models that are built often depend on some of the things that we've been talking about. The idea of modeling a city we need to figure out where we draw the boundary in some sense. We need to figure out what level of detail we need to build in this particular context. We need to think about how activities which determine what we're modeling basically, the location of population, employment, the movement of traffic, all of these things sort of fit together. How are they modeled or simulated so that we can make predictions? There are several different types of models that have been developed over the last 50 years or more in parallel with what's gone on in urban science and complexity theories, so we can extend them. These models basically are computer models. We couldn't build them without computers in the sense that digital depends on data and so on. There are really three or four different classes of model. The classic model which does use quite a lot of stuff related to power laws and delometry and so on in terms of how people travel and population densities that I've talked about earlier on. The Land Use Transportation models which we developed really from the 1950s onwards, almost as soon as computers got invented, people began to build Land Use Transport models basically for the main highway schemes in the United States and so on to look at those. The Land Use Transport models are still built fairly routinely. They basically enabled you to make predictions about if you put a thousand jobs or 10,000 jobs here, if you put a new road there or a new rail line and so on, it'll look at the impact of that on everything else in the city basically, good or bad. And those are the sort of models that were first built. From complexity theory, the sort of fractal type models that have been built are what are called cellular automata models and these divide the city into cells which has the name suggests is sort of a grid but a grid across the city and the big difference between cellular automata and Land Use Transportation models is that cellular automata models tend to be dynamic. They look at how the cells change over time in that sense whereas the Land Use Transportation models really look at the picture of the city as though it's in equilibrium. In that sense, I did say that equilibrium is a no-no to some extent in complexity theory but there's emerging of lots of these ideas into one another of course in that way. Cellar automata models as the name suggests enables you to make predictions about what happens in the cells according to principles about why people would want to live there, why they would want to develop their own son. And what are also now called agent based models were rather than dealing with large aggregates of people or even aggregates of people in cells we deal with individual agents in that sense and so there's a variety of transport models such as MatSim, for example, which is developed in the United States and in Switzerland at ETH. These are models that simulate every car, every traveler on the road basically. So these are very intensive models basically that actually look at how people behave in terms of transportation behavior and those are agent based models or sometimes called micro simulation models. So you have this great range of models in terms of how they deal with statics or dynamics and how they deal with the resolution of the system whether it be the Land Use Transport models tend to divide the city up into rosary essentially block groups or wards or units with between 3 and 4,000 people in and so on whereas the cellular automata models deal with the resolution of the city out 100 meter square or kilometer square and so on. The agent based models deal with individual agents household basically and try and model everyone of those and there are different tools and techniques which you used to affect this modeling basically. So let's dig into the three of those types of models and more data. So it's start with the Land Use Transportation model. And then how is it set up and what does it do? Well the employment is divided into different types of activities so typically industrial employment of various sorts, commercial employment, retailing and so on. The employment side of the house is in terms of where activity is located. Where people are located and the strong link in Land Use Transportation model is where people live and where they work although many of them are now extended to where people live and where they shop, where they live and where they go to school, where they live and where they get their health care and so on. Whole range of things like this. So in other words the models essentially build the networks basically and we're back to complexity theory networks. The networks are channels on which activities are linked to one another so where people live and where they work for example and how they travel basically that we often have different networks, rail, road, bus and so on in that particular context. So essentially those models basically simulate what happens if you change where people live or where they work or if you change the way they go from work to home or whatever in terms of the mode for example. So you put in new roads, you change the travel time on the road system, things of that. So that's the way those models work. They make predictions. Well, most models are used to make conditional predictions if something then what so to speak if then type modeling. If you put in a new central business district on the edge of the city or something what would be the impact on transportation journey to work and so on. So that's in essence the way land use transport models work. The focus is on land use meaning activities located on land parcels and transport of course is the actual movement patterns. Now having said that the other models I mentioned don't necessarily stress transport or movement to the same extent. And this land use transportation model this is obviously being used all over the world and it's where you design our we need a new industrial area to the city over here or we need a new suburb over here is that what it's basically used for. Basically yeah and it varies enormously that as you say they've been used they're used worldwide different countries of development to different degrees basically they're used to inform rather than predict basically they're used to think about what the future might be like rather than to actually produce a plan in some sense that that's a major issue this whole question of how how you use them to predict. Have there been any big successes or failures with their application? At the very beginning when these models were first built and there were enormous numbers of failures you know people never anticipated how much time it would take to do these things the data was terrible the computer time was terrible basically. In other words everything that could go wrong did go wrong and of course the problem was that the theories on which these models were built were not well structured it was early days basically the theories were not adequate in some sense. As time has gone on many of these problems have changed the data is much better computer time is no longer an issue at all in these things some of the theories got better but that's to some extent the Achilles heel in all of this everything we're talking about today is problematic when it comes to judging how good these theories are basically and it the field wanes between thinking it's that we have some success in explaining the world to a very little success and partly this is due to the fact that the cities themselves are becoming ever more complex you know new modes of being added behavior patterns are changing which would be on our ability to predict and so on. So in some sense is the succession failure has to be judged against this wider back cloth of uncertainty really in that sense. And then moving on to the second group cellular automata starch a bit more general with them because I think if you listen to the show you probably get your head round the land use transportation model you know certainly here of agent based models perform but cellular automata are a fascinating field in among themselves maybe if you can start there Michael and then we'll talk about sort of models they produce. In some sense the cellular automata models are the simplest of any of these things because essentially in building a model of a city we have to produce a representation of the city a spatial representation. Then there are models of cities which are non spatial. J Forest is a dynamics model from the late 60s was essentially a model of central Boston it didn't have any spacing it it just talked about what was happening in central Boston. And it's quite valuable in some senses but it was very different from the kind of spatial models we're talking about here. The cellular automata model the spatial representation is to divide the world up as the name suggests into a series of cells. So in other words we take a landscape a city for example like Brisbane and we divide it up into let's say a thousand cities. basically a thousand grid squares. And what happens in each of those grid squares is the essence of what the city is. It's where people live, where they work, where there's land available, all the attributes of what goes on in that grid square are coded and coded into the model in that particular way. And then there's a series of rules basically. In the case of solar autonomous model, there are more to do with a set of rules that suggest if something happens in a cell, then it makes an impact on another cell. Their interaction effects, a typical example is the challenge segregation model. I'm sitting in a cell. OK, and if something happens in an adjacent cell and I don't like it, then I will vacate the cell that I'm sitting in. So typically if this is a model based on ethnic or income segregation. And if I'm a rich person, the rule is that I will only live in a cell which is surrounded by the cell in question is surrounded by let's say eight, eight other cells basically. If I'm living in that cell, the certain income and the number of people come into the adjacent cells with either a much bigger income or a much lower income, I will no longer feel comfortable and I will vacate it. So essentially the model works at the level of cells. OK, how the cells behave with respect to what's happening around them basically. So in other words, the whole set of rules, which are encoded in terms of how the city develops at the level of individual developers in each of the cells. And we have transport in those cells, but generally speaking, transport is not really built into this. There may be some element of what we call accessibility, how near is that cell to all the other cells based on some measure of accessibility through the road network and so on. Generally speaking, those models of ignored for a whole variety of technical reasons, the idea of transportation and very often you may say, well, that's crazy, you need transportation. And the answer of course is yes, but the assumption is that you would use more than one model in a real context, you use both the cellular automata model and couple it to a land use transportation model basically. So coupling models together is a way of adding more and more detail, a more and more theory and technique and method, which can't be integrated in and all singing all dancing model. Yeah, that's interesting because it did strike me as you were talking like the cellular automata is almost the opposite of the land use model. If you ignore transportation, but the land use model must be ignoring the relationships between the cells. Not so much that it's ignoring what happens over time. The land use transportation model doesn't ignore the relationship between the cells at any point in time, but it does ignore what's happening with those cells at a future point in time or the past. In other words, it's a static equilibrium model, which assumes that the world is in equilibrium. And if you make a change, it assumes that the world adjusts negative feedback like it assumes that the world adjusts to this change. I didn't think of it in those terms, and it did right. I mean, that's what the cellular automata must be so powerful because you can look at it over time because you cellular automata model that would explain wouldn't it the Georgian houses and the gentrification that is very much so. But as one cell gets upgraded and done up and people go, I can have that as well and moving next door and suddenly you gentrify and you see that as you say over time as a model sort of moves towards that new point. And then to our last model, which you will have heard on the show before the agent based model. So what is it different? Is it as simple as we define the agents and define the interactions between them and let them run loose and see what happens? Well, and agent based modeling is almost a style of modeling that land use transportation modes are peculiar to the world of cities. And modeling is quite generic. It can be used from everything in physics through to sociology really in some sense. So there is a difference at that level of difference of specificity one really needs to become fairly specific about what sort of agent based model in our particular world. The easiest way to think about it if you think of a cellular automata model where each cell is an agent then there's an immediate translation between cellular automata and agent based model. So in other words, if you put a 100 meter grid across the city and then if you increase the resolution of that grid to a one meter grid across the city, then you'd be able to capture the location of every particular agent and some of the cells would be agents and some of them would not have agents. And the same sorts of rules would apply. So to some extent, an agent based model is a generic modeling technique. All of these models can sort of fuse into one another to some extent. But the agent based model is much more general, I think than either the other two. It's more general for other systems other than cities, I should say. Yeah, but the really common application of agent based model is in traffic modeling when you model each vehicle on the road, isn't that right? And see the image and phenomenon that flows from that. So when we build agent based models of cities, is that really looking at a pair person level or is it a bit more granular than that? Well, it's the bigger ones tend to work with households basically the household unit is a better unit than the individual variety of reasons a lot of decision making in cities really related to what's happening to the household. The scale is very important. A lot of the agent based models, which are at the person level are at a very fine scale. Pedestrian modeling, for example, which is not a million miles away from traffic modeling, pedestrian modeling is one of the early applications of agent based models. And essentially those have been developed quite widely for a whole variety of local design type situations, you know, crowding in airports, particularly crowding in entertainment complexes and so on. Those models are quite widely used and the individuals in question tend to be persons. The other feature of course of agent based model is the agents need to be persons. They can be movable objects in the environment. So we built a model of the Notting Hill Carnival. That's a two day carnival in London in August. And this model of the carnival had the crowds and the paraders and the police were all separate sorts of agents. But also a whole set of movable objects such as barriers in the streets and so on. These were also agents basically. So anything that was mobile and of significance to the interactions was clustered as an agent when you go to the agent based models what sort of phenomena they throw an up that we aren't seeing in the other model. What sort of insights are they giving us? Well local interactions between the physical environment and the agents themselves because at their scale the physical environment the actual configuration of the buildings and so on. None of that really relates to any of the more aggregate of land use transportation model stuff. So you can think of the land use transportation model has dealing with aggregates of population and employment in the order of maybe three or four thousand. Okay lumps three or four thousand agent based models are either individuals themselves or households in this particular context. And therefore the level of representation is such that they're able to link with anything at the same scale at which they're at and so the physical environment can be encoded rather differently in those cases. And what sort of I suppose apart from necessarily the travel aspects in terms of pedestrians and vehicles and that sort of model and what are these agent based models telling us about the modern cities were living in? Well to some extent the gaining in popularity because of this move towards active travel people shifting their mode of transport to new forms of vehicular traffic in that context. So rather than telling us a lot about cities it's sort of informing us how we need to model cities to deal with some of these potential changes in this particular context. In terms of adding things that the other models don't do then I think that the potentially there are much stronger interaction effects to non spatial markets in agent based modeling. So for example if you're buying a house etc. Then a good deal of what goes on is not in the spatial context. It's not to do with how close you live to various things and so on. It's very much to do with the mortgage market and so on. So a variety of markets can be added as a good model developed by Rob Axtel and Dorn farmer housing market model were. They've got a bunch of agents and were they buying houses and so on but also it's looking at agents well in terms of where they're buying houses but it's also looking at how they acquire housing finance and things of that sort really so it's extension into those sorts of markets that agent based models have the potential connections to make. So they're picking up as you say this there's more complexity in our lives and in our cities exactly and they're starting to pick that up and get it into the model definitely it's a fair to say that it's very difficult to predict using these models. But these models help us explain and understand phenomena that we can see and also give us options. They allow us to run simulations and look at the effect of different decisions. Is that reasonable? Yes. I mean, all of the models are able to do that, to look at the impact of changes. In the case of the agent-based models, I'm not familiar with models that have actually introduced additional agents, basically. But they certainly introduced new behaviour patterns, right? I mean, they're more likely to evolve new behaviour patterns, agent-based models. Land use transportation models are more to do with lumps of employment and population being projected for the city. Both an increase and also a decrease, really, in that sense. It can look at the impact of negative things. I think land use transportation models are much easier to look at, plus and minus, say, agent-based models that really are rather focusing on a different thing. I think the key thing is that, although there is some mixing between these different types of models, generally speaking, the class of models are quite separate. And if we were using them together, we would fashion away of integrating the kind of contents of one model with another, in some sense. That would probably be the main thing. What does the future, if you look like in this research, in terms of where do you see the big wins would come from, in terms of impacting how we think about cities and live our lives and cities? I think there's a general move towards agent-based modelling. In fact, there are developments on all fronts, really, in some sense. The land use transport models are potentially more useful because they do with large aggregates and planning policies, urban planning deals with large aggregates rather than agents, basically. At the urban design level, agent-based models are quite useful, but they tend to be relatively useful to small pockets of intense development in the city. If you broaden it out to the whole city, you can build agent-based models for everywhere, basically. But generally speaking, moving to the whole city scale, you'd be using a different type of model, such as land use transportation, basically. Agent-based models, solar, automata and land use transport all have different sets of principles that can be used in relation to one another, particularly when it comes to interaction effects over space in this particular context. So, movement, patterns, etc. and the influence of different locations on one another. So we've talked a lot about fractals and emergence and similarity in the parallels. How does it all come together inside these models? Do we see these things in the results of the models or in how we build the models? I think we see an obvious focus on different aspects of the urban system in different models. So in other words, what's in a land use transportation model is unlikely to be in a solar, or a tomato model, unlikely to be in an agent-based model, but there are exceptions that are linked. So we see it in the tools being used. And having said that, there's a whole range of tools out there which we've not talked about, which are a bit more analytic rather than simulation in a sense which can also be used at the same time, really. So, we see a lot of different different types of systems and so on. In that sense, we would probably see some differences between different groups of people working with in a planning context, for example, with different focus, different perspectives on what the ultimate plan would be. And we're not planning to do this in a planning practice, and possibly Australian too, I don't know, but a lot of what we've talked about is not really developed because we don't have very strong strategic planning in Britain. You sort of need strategic planning, urban and regional citywide planning before you these models really come into their own. Apart from, of course, very detailed agent-based models and some transport models that are embedded within the mandates of the local municipalities themselves. And it comes to the fractals and the power laws. We see that in some of these models, we see those characteristics that we know exist in real cities. Yeah, very much so. I mean, these models have to be calibrated. They have to be tuned to the existing situation. And in tuning them, it's a bit like saying, there are power laws contained in some of the land use transportation models and some of the other models too, I think. But they have to be tuned. They have to be tuned to the situation. They have to be estimated to the situation because the way we travel in Bangladesh is very different from the way we travel in central London. And this kind of thing. So to some extent, there needs to be a calibration or tuning that would be reflected in the models and their applications. Well, though we think of a system, you know, you could think of it in one sense as this fine tune top down control machine. And then the other end of the scale, we could think of it very much as almost a random chaotic process. But what we really see is it's somewhere in the middle where we do have organization and we do have structure in it. And that's reflected in all these laws that we've talked about in results from positive feedback and emergence results from the positive feedback as well. And that allows us to look at these models and tune them and get them represented of something that's in real life and presumably it's because all this scaling exists because the fractals exist in cities. That's why we can model them in the first place. That's right. To some extent, the development of these models is continuing a pace. And it seems to me that we're going to have a lot more integration between the different models in the future than we've had in the past. What will happen in terms of practice is quite problematic because we haven't really talked about this. So we implied it that cities are getting more complex. The whole idea of development planning in cities is increasingly problematic from a whole range of issues to do with local politics, participation and so on. These tools are potentially expensive in terms of the way they're developed. We don't have good professionals working with these tools because they're hard to train and it's expensive to train them in that sense. There's a whole range of bigger problems that this discussion sort of fits into really. And that sounds very pessimistic. It's not meant to be pessimistic because we've actually made quite a lot of progress. But it's a kind of progress that is such that we have to run to stand still because the world is changing so rapidly in this particular context. And that's even true about some of these models. Arguably, these models have been developed over over a number of years. And some of the earlier ones are, if you like, obsolete. Some of the land use transportation models are less relevant because the way of looking at the city is a little bit less relevant through that lens than through say solar automata. But that's really saying that our focus of interest has changed. It's not that the models are wrong or in that sense, but the focus of interest has changed. We're building a bigger portfolio of different models dealing with this enlarged and enlarging focus of interest. Michael Barney, thank you very much for being on the show. Thank you very much Sean. Look forward to hearing it. Thank you. Thanks for listening to Simplifying Complexity. When we look at the key concepts of complexity science with expert minds from across the world, concepts like emergence, self-organization, adaptation, networks, scaling, tipping points, and much more. This podcast was produced by Brady Aeword and Wadland Creative. To make sure you don't miss an episode, be sure to subscribe to or follow the show in your podcast app. I'm Sean Brady, and I'll see you in our next episode.