You are on page 1of 9
Urban Analytics and City Science 2017 Volume 44 Number 2 March ISSN 2399-8083 27 MAR 2017 STAFFS UNI, iSITY LiGRAA Environment and Planning B 8 journals.sagepub.com/home/epb Article Troronman and Pig Ban From paths to blocks: New ‘ot aC snc measures for street patterns on aor 25 sere courtier Bor 10 monesrss seme Marc Barthelem Jourlseepb confor Institut de Physique Theorate (CEA), Gif-sur-Yvette, France @SAGE Abstract The street network is an important aspect of cities and contains crucial information about their organization and evolution. Characterizing and comparing various street networks could then be helpful for a better understanding of the mechanisms governing the formation and evolution of these systems. Their characterization is however not easy: there are no simple tools to classify planar networks and most of the measures developed for complex networks are not useful when space is relevant. Here, we describe recent efforts in this direction and new methods adapted to spatial networks. We will first discuss measures based on the structure of shortest paths, among which the betweenness centrality. In particular for time-evolving road networks, we will show that the spatial distribution of the betweenness centrality is able to reveal the impact of important structural transformations. Shortest paths are however not the only relevant ones. In particular, they can be very different from those with the smallest number of turns—the simplest paths. The statistical comparison of the lengths of the shortest and simplest paths provides a nontrivial and nonlocal information about the spatial organization of planar graphs. We define the simplicity index as the average ratio of these lengths and the simplicity profile characterizes the simplicity at different scales. Measuring these quantities on artificial (roads, highways, railways) and natural networks (leaves, insect wings) show that there are fundamental differences—probably related to their different function—in the organization of urban and biological systems: there is a clear hierarchy of the lengths of straight lines in biological cases, but they are randomly distributed in urban systems. The paths are however not enough to fully characterize the spatial pattern of planar networks such as streets and roads. Another promising direction is to analyze the statistics ‘of blocks of the planar network. More precisely, we can use the conditional probability distribution of the shape factor of blocks with a given area, and define what could constitute the fingerprint of a city. These fingerprints can then serve as a basis for a classification of cities based on their street patterns. This method applied on more than 130 cities in the world leads to four broad families of cities characterized by different abundances of blocks of a certain area and shape. This classification will be helpful for identifying dominant mechanisms governing the formation and evolution of street patterns. Keywords Network graphs, network structure, space syntax, street networks, urban morphology Corresponding author: Mare Barthelemy, Institut de Physique Theorique (CEA) 91191, Gif-surYveete, Cedex, France, Email mare.barthelerny@ceafr Barthelemy 237 The importance of street networks Urbanization is a fundamental process in human history, and is increasingly affecting our environment and our society. Understanding the elementary spatial mechanisms that govern urbanization, leaving out specific historical, geographical, social, and cultural factors, is nowadays more important than ever, especially because policy makers (Batty, 2005) need clear directions and a scientific input in urban planning and land management (Ramalho and Hobbs, 2012; Turner et al, 2007). The street network is a long-lasting constituent of urban form (Marshall, 2006; Southworth and Ben-Joseph, 2003), a fundamental driver in the urbanization process, and understanding it might provide some important clues about the dynamics of cities. In addition to urban studies, and transportation economics, we saw in the last decade significant contributions from complex networks theory about the quantitative characterization of urban street patterns (Barthelemy, 2011). Several studies have shown that road networks not only play a central role in the spatial organization of urban areas (Cardillo et al., 2006; Lammer et al., 2006), but are important for the dynamical processes occurring on them (Balcan et al., 2009; Barthelemy, 2011; Bettencourt et al., 2007; Porta et al., 2011) and for the evolution of urban systems in general. However, a complete, quantitative analysis of the historical development of urbanization in metropolitan areas is still missing, and empirical evidence of the basic mechanisms governing urbanization dynamics is still lacking. ‘We will focus in this article on new measures helping in the quantitative characterization of street and road networks. We will focus here on the primal representation (in contrast with the dual representation, see Porta et al., 2006b) where nodes usually represent intersections of roads and links (also called edges in graph theory) represent segments of roads connecting different intersections. A more detailed description keeps the nodes with degree 2, which represent changes in direction. The important feature of these “ ‘spatial” networks is that they are embedded in space and to a good approximation, are planar graphs. A planar network is a graph that can be drawn on the two-dimensional plane such that no links cross each other (see ¢.g., Clark and Holton, 1991). Planar graphs pervade many aspects of science: they are the subject of numerous studies in graph theory, in combinatorics (Bouttier et al., 2004; Tutte, 1963), in quantum gravity (Ambjorn and Jonsson, 1997), and in biology and botanics (Katifori and Magnasco, 2012; Mileyko et al., 2012). In city science, planar networks are extensively used to represent various infrastructure networks (Barthelemy, 2011). In _ particular, transportation networks (Haggett and Chorley, 1969; Xie and Levinson, 2009) and streets patterns (Hillier and Hanson, 1984; Marshall, 2006) are the subject of many studies (Barthelemy and Flammini, 2008; Barthelemy et al., 2013; Cardillo et al., 2006; Chan et al., 2011; Courtat et al., 2011; Crucitti et al., 2006; Gastner and Newman, 2006; Jiang, 2007; Jiang and Claramunt, 2004; Lammer et al., 2006; Masucci et al., 2009; Perna et al., 2011; Porta et al., 2006a; Venerandi et al., 2014; Rosvall et al., 2005; Strano et al., 2012, 2013; Xie and Levinson, 2007) that are trying to characterize both topological (degree distribution, clustering, etc.) and geometrical (angles, segment length, face area distribution, etc.) aspects of these networks. Despite the large number of studies on planar networks, there is still a lack of global, high-level metrics allowing their structure and geometrical patterns to be characterized. Such a characterization is difficult to achieve and in this article, we will discuss various aspects of these graphs that are intimately connected to both their geometry and topology. ee OO 258 Environment and Planning B: Urban Analytics and City Science 44(2) Spatial structure of paths ‘One can think of different paths connecting a pair of nodes and we will discuss two important types here: the shortest and the simplest paths. The topological shortest is the path with the minimum number of segments. If we use weighted links (e.g., by their length, or velocity), we then obtain (euclidean) shortest paths or quickest paths. We note however that if the length of edges is not broadly distributed (ie., edges have a typical average length), the topological and euclidean shortest paths are essential proportional and display same behaviors (and unless indicated, the results presented here are obtained by considering topological shortest paths). There are however other types of paths that we can use in networks. For example, in spatial networks, we can determine straight lines (e.g., using the continuity negotiation-like algorithm, Porta et al., 2006b) and a turn is then defined as a change from one straight line to another. = Shortest path! mme Simplest path} Figure 1. Illustration of the difference between the euclidean shortest path and the simplest path in the case of Oxford (UK). In order to identity the simplest path, we first convert the graph from the primal to the dual representation, where each node corresponds to @ straight line in the primal graph, and where edges represent the intersection of straight lines in the primal graph (straight lines were determined by a continuity nnegotiation-like algorithm, Porta et al, 2006b). (This figure is taken from Viana et al, 2013). Barthelemy 259 The simplest path is defined as the one with the minimum number of turns and as shown in Figure 1, it can be very different from the shortest one. Betweenness centrality (BC): impact and spatial distribution ‘An important quantity based on shortest paths is the BC. The BC g(e) of a link e (and similarly g(i) of a node i) measures the fraction of times a given link is used in the shortest paths connecting any pair of nodes in the network: 4 aul0) 80 = RYH ow, where V is the number of nodes in the network, and the sum runs over alll nodes s and ¢. The quantity oy is the number of shortest paths going from node s to node ¢ (and o(e) is the number of paths going from s to f, through link e). The BC is thus a measure of the contribution of a link to the organization of flows in the network (Freeman, 1977). A large BC indicates that the corresponding link is essential in the network and that it can have a very specific role or function. In particular, its removal will have a large impact on the overall structure of shortest paths in this network. In nonspatial graphs, the BC is governed by the degree and the nodes with the largest BC are essentially the hubs of the network (Barthelemy, 2004), and in contrast in regular spatial lattices, the BC is a simple function of the distance to the barycenter of all nodes. For a random planar network, the spatial structure of the BC results then from the interplay between topology and space and we can expect interesting patterns. We will illustrate this on the evolution of the street network in two cases: Groane (Italy) and Paris (France). Evolution of Groane: The BC impact The Groane area is located north of Millan (Italy), covers a surface of 125 km? and includes 29 urban centres with 14 municipalities. This region has essentially evolved along radial paths connecting Milan to Como and Milan to Varese and more details about the evolution of this region can be found in Strano et al. (2012). It is however important to note that this area has never been subjected to large-scale planning efforts, one reason being that 14 different administrative bodies preside over 14 different municipalities. By importing historical topographical and photogrammetrical data into a Geographical Information System (GIS) environment, the detailed road system (including minor streets) could be reconstructed at seven different points in time corresponding to the years 1833, 1914, 1933, 1955, 1980, 1994, 2007. The road networks at different times show how the initial small separate villages have grown by the addition of new nodes and links, eventually merging together in an homogeneous pattern of streets (Strano et al., 2012). ‘We denote by the graph at time t and the difficulty here is to find a simple measure that can help in understanding the time evolution of this spatial network, and the properties of new links. Maybe surprisingly most of the standard indices that characterize the graph structure display a very smooth behavior: the network becomes denser and the fraction of nodes of degree 4 (which correspond to four roads intersections) increases with time. However, new links do not display specific statistics: we did not observe any regularity in terms of length for example, In contrast, the spatial structure of shortest paths evolved in time and the BC should reflect important features of this system. We found, for example, a clear relation between the age of a street and its BC: highly central links usually are also the eee 7 eeeeeeet eee seeee eee 260 Environment and Planning B: Urban Analytics and City Science 44(2) Barthelemy 261 s mot o maim © mone @ Figure 2. Betweenness centrality impact. (a-c): Different links according to their BC impact are shown. In ‘green densification links and in red exploration links. (4) and (e) Distribution of the BC impact. In (e), the different levels of gray correspond to different values of the smallest degree of both end points of links. We see on this graph that most exploration links have at least one end point with degree I. (Figure from Strano etal, 2012). oldest ones. More precisely, we observed that more than 90% of the 100 most central links in 2007 were already present in 1833, which reveals a “backbone” of highly central routes that have framed the Groane area in the preindustrial period, and has been driving its development across two centuries. In order to go further in the analysis and to categorize more precisely new links, we define the BC impact 5 which measures how a new link (absent at time ¢— 1 and present at time (1) affects the average BC HG) - XG\e) Gi) 50) where the bar denotes the average over the whole network. The first term of the numerator is the average of the BC over the whole network and for the second term we consider the network G, with the link e removed. The BC impact is thus the relative variation of the graph average betweenness due to the new link e. We have measured this quantity for the different times and the result is shown in Figure 2. Remarkably, the distribution of 6 displays two well-separated peaks (Figure 2(d) and (¢)). The importance of the first peak tends to increase in time while the second peak decreases, until they mostly merge into one single peak in the last time section (1994-2007). In order to understand the nature and the evolution of the two peaks, we report in Figure 2(a) to (c) the new links in different colors: in green the links whose centrality impact belongs to the left peak, and in red the links whose 6 corresponds to the right peak. The new links are divided into two classes: green links (small 6, left peak) tend to bridge already existing streets, while red links (large 6, right peak) are usually dead-ends edges branching out of existing links. The distribution of the BC impact thus suggests that the evolution of the road network is essentially characterized by two distinot, concurrent processes: one of densification (green links), which is responsible for the local density increase of the urban texture, and one of exploration (red links), which corresponds to the expansion of the network toward previously nonurbanized areas. Evolution of Paris: The spatial structure of the BC In this section, we consider the evolution of the street network of Paris over more than 200 years with a particular focus on the 19th century, a period when Paris experienced large transformations under the guidance of Baron Haussmann. It would be difficult to describe the social, political, and urbanistic importance and impact of Haussmann works in a few lines here and we refer the interested reader to the existing abundant literature on the subject (ee, e.g., Jordan, 1995, and references therein). Roughly speaking, until the middle of the 19th century, central Paris has a medieval structure composed of many small and crowded streets, creating congestion and probably health problems. In 1852, Napoleon IIT commissioned Haussmann to modernize Paris by building safer streets, large avenues connected to the new train stations, central or symbolic squares, improving the traffic fiow and the circulation of army troops. The case of Paris under Haussmann provides an interesting example where changes due to central planning are very important and where a naive modeling is probably bound to fail. By digitizing historical maps (for details about the sources used to construct the maps, see Barthelemy et al., 2013) into a GIS environment, we reconstruct the detailed road system (including minor streets) at six different moments in time: 1789, 1826, 1836, 1888, 1999, 2010. It is important to note that we thus have snapshots of the street network before Haussmann works (1789-1836) and after (1888-2010). This allows us to study quantitatively the effect of such central planning. Simple measures. The first basic measures of the network evolution include the evolution of the number of nodes N, of edges E, and the total length Ly, of the networks, These indicators display a clear acceleration during the Haussmann period (1836-1888). The number of nodes increased from about 3000 in 1836 to about 6000 in 1888 and the total length increased from about 400km to almost 700km, all this in about 50 years. It is interesting to note that this node number increase corresponds essentially to an important population increase. In particular, we observe that the number of nodes NV is proportional to the population P and that the corresponding increase rate is of order dNjdP ¥ 0.0020, The rapid increase of nodes during the Haussmann period is thus largely due to demographic pressure, and if we want to exclude exogeneous effects and focus on the structure of networks, we have to plot interesting quantities versus the number of nodes taken as a time clock. The results then display a smooth behavior: £ is a linear function of N, demonstrating that the average degree is essentially constant, and the total length versus NV also displays a smooth behavior consistent with a perturbed lattice (Barthelemy et al., 2013). All these results are at first glance rather perturbing since it seems that Haussmann had no effect whatsoever on the global structure of the network. This is however not correct as we will see in the next section about the spatial distribution of the BC. 262 Environment and Planning B: Urban Analytics and City Science 44(2) 1789 1826 1888 1999 2010 Figure 3. Time evolution of the spatial distribution of the betweenness centrality for Paris. Figure taken from Barthelemy et al, 2013). Spatial distribution of the BC. The BC is able to detect important routes and nodes in the internal structure of the network and can then be used as a structural probe of the network, and to track important modifications. We then first consider the time evolution of the node BC (with similar results for the edge BC). Apart from the fact that the average BC varies in time, the tail of its distribution remains constant, showing that the statistics of very central nodes is not modified. From this point of view, the evolution of the road network follows a smooth behavior, even in the Haussmann period. So far, most measures indicate that the evolution of the street network follows simple densification and exploration rules and is very similar to the Groane area for example, and that Haussmann’s works did not change radically the structure of the city. However, we can suspect that Haussmann’s impact is very important on congestion and traffic and should therefore be seen on the spatial distribution of centrality. In Figure 3, we show the maps of Paris at different times and where we indicate the most central nodes (such that their centrality g(i) is larger than max(g)/a with c= 10). We clearly see here that the spatial distribution of the BC is not stable, displays large variations, and is not uniformly distributed over the Paris area. In particular, we see that between 1836 and 1888, Haussmann works had a dramatic impact on the spatial structure of the centrality, especially near the heart of Paris. Central roads usually persist in time, but in this case, the Haussmann reorganization was acting precisely at this level by redistributing the shortest paths, which had certainly an impact on congestion inside the city. After Haussmann, we observe a large stability of the network until nowadays. It is interesting to note that these maps also provide details about the evolution of the road network of Paris, during other periods, which seems to reflect what happened in reality. For example, in the period 1789-1826, the maps shown in Figure 3 display large variations with a redistribution of central nodes, which probably reflects the fact that many religious and aristocratic domains and properties were sold and divided in this period in order to create new houses . Barthelemy 263 and new roads. During the period 1826-1836, which corresponds roughly to the beginning of the the July Monarchy, the maps in Figure 3 suggest an important reorganization on the east, side of Paris. This seems to correspond very well to the creation during that period of a new channel in this area (the channel “Saint Martin”), which triggered many transformations in the eastern part of the network. After Haussmann’s period, the central portion of Paris did not change substantially: the modifications observed here are due to cartographic problems (the last 2010 map does not have the same level of detail) Street networks and biological systems: The simplicity profile Definition. In planar networks, shortest paths can be very different from those with the smallest number of turns—the simplest paths. The statistical comparison of the lengths of the shortest and simplest paths provides a nontrivial and nonlocal information about the spatial organization of these graphs. In order to get some quantitative information, we denote by 1G.) the length of the shortest (Euclidean) path, which minimizes the distance between nodes / and j, and by (i,j) the length of the simplest path (if there are more than one such path, we choose the shortest one). In order to understand the structure of the network, it is useful to compare these lengths and to introduce the simplicity index as the average td > WN-Dy Kid) ‘The simplicity index is larger than | and exactly equal to | for a regular square lattice and any tree-like network for example. Large values of S indicate that the simplest paths are on average much longer than the shortest ones, and that the network is not easily and efficiently navigable (in the sense that simplest paths can be long). This metric is a first indication about the spatial structure of simplest paths but mixes various scales, and in order to obtain a more detailed information, we define the simplicity profile PO) Sd) = 5 iassagtiyna GD) 1 (d), where de(i,) is the euclidean distance between i and j and where N(d) is the number of pairs of nodes at euclidean distance d. This quantity S(d) is larger than one and its variation with d informs us about the large-scale structure of these graphs. We can draw a generic shape of this profile: for small d, at the scale of nearest neighbors, there is a large probability that the simplest and shortest paths have the same length, yielding S(d~ 0) © 1 (and increasing for small d). For very large d, it is almost always beneficial to take long straight lines when they exist, thus reducing the difference between the simplest and the shortest paths. As a result, we expect S(d) to decrease when d+ dmax (note that a similar behavior is observed for another quantity, the route-length efficiency, introduced in Aldous and Shun, 2010). The simplicity profile will then display in general at least one maximum at an intermediate scale d* for which the length differences between the shortest and the simplest path is maximum The length a thus represents the typical size of domains not crossed by long straight lines. At this intermediate scale, the detour needed to find long straight lines for the simplest paths is very large. We measure this quantity on artificial (roads, highways, railways) and natural networks (leaves, slime mould, insect wings) and show that there are fundamental differences in the organization of urban and biological systems, related to their function, navigation, or distribution. 264 Environment and Planning B: Urban Analytics and City Science 44(2) . 2 Barthelemy 265 (a ay Figure 4. Simplicity profile for natural and artificial systems. (Figure taken from Viana et a., 2013). Simplicity profile: Natural versus artificial networks. The simplicity index reveals how the straight lines in the network are distributed in space and participate in flows on the network (see Viana et al., 2013 for more details). We will focus here on the simplicity profile applied on natural (biological) and artificial systems. We discuss here (see Figure 4) city streets (Bologna, Italy; Oxford, UK; Nantes, France), the national highway network of Australia, the national UK railway system, and the water supply network of central Nantes (France). In the case of biological networks, we study the veination patterns of leaves (Ilex aquifolium) and (Hymenanthera chatamica), and of a dragonfly wing. The simplicity also allows to explore the time evolution of these networks and we refer the interested reader to Viana et al. (2013). ‘We observe in Figure 4 that for most of these systems, the simplicity profile displays the generic shape with a maximum at an intermediate scale. In typical urban monocentric systems such as Bologna and central Nantes, we have a dense center and a few important radial straight lines, leading to a simple, profile S(d) with one peak. In the case of Oxford and the Australian highway network, the polycentric organization leads to multiple peaks in the simplicity profile. Interestingly, we observe that profiles for Australian highways and railways in the UK are very different, despite their similar scale and density. In particular, the UK railway system displays small values of the simplicity (less than 1.2), while for the Australian highway network there are many pairs of nodes for which the simplest path is much longer than the shortest one. We also observe that the profile for both street and water systems of Nantes has a very similar shape, pointing to the fact that these networks are strongly correlated. In addition, the position and the height of the peak (around 1.4) observed for the Nantes water system suggest that this distribution system has similar features compared to biological systems such as vein networks in leaves (whose function is also distribution). In contrast with urban systems, the simplicity profile of biological networks has a single well-defined, and much more pronounced peak. We observe values of order Sax * 1.5 and 2.5 for d* of order 0.2 dmx, meaning that for this range of distance, the detour made by the simplest path is very large. This peak is related to the existence of domains of typical size d* not crossed by large veins. We see here a clear effect of the existence of the spatial organization of long straight lines in these systems, probably optimized for the distribution (of water for leaves). The decay for large d is also much faster in the biological case compared to urban systems: this shows that in biological systems there are long straight lines allowing to connect far away nodes. This is particularly evident on the leaves shown in Figure 4 where we can sce the first levels (primary and secondary) veins, the rest forming a network. For streets, the organization is much less rigid and the hierarchy less strict: we have a more uniform spatial distribution of straight lines, leading to a smoother decrease of S(d). ‘Our results highlight the structural differences between biological and artificial networks. In the former, we have a clear spatial organization of straight lines, with a clear hierarchy of lines (midrib, veins, etc), leading to simplest paths that require a very small number of turns but at the cost of large detours. In contrast, there is no such strong spatial organization in urban systems, where the simplicity is usually smaller and comparable to a null model with straight lines of random length and location. These differences between biological and urban systems might be related to the different functions of these networks: biological networks are mainly distribution networks serving the purpose of providing important fluids and materials. In contrast, the role of road networks is not only to distribute goods but to enable individuals to move from one point of the city to another. In addition, while biological networks are usually the result of a single process, urban systems are the product of a more complex evolution corresponding to different needs and technologies. Statistical distribution of block shapes We considered so far properties related to paths on networks, but this information does not include all geometrical aspects. In particular, visual patterns of street networks depend a lot on the arrangement of blocks and their area distribution. This is the main idea developed in this section where we consider block statistics to describe street network patterns. Classification of street patterns Information about spatial neworks is not contained in their adjacency matrix only. Geometry, encoded in the spatial distribution of nodes, plays a crucial role. A classification of cities according to their street network should then rely on both topology and geometry. We note that while classifications do not provide any ER 266 Environment and Planning B: Urban Analytics and City Science 44(2) understanding of the objects being classified, they provide a useful first insight in the different characteristics exhibited by objects of the same nature. Classifying, from a fundamental point of view, is however difficult: finding a typology of street patterns essentially amounts to classifying planar graphs, a nontrivial problem. The classification of street networks has been previously addressed by the space syntax community (Hillier and Hanson, 1984; Penn, 2003; Shpuza, 2009)—a good account can be found in the book by Marshall (2006)—and more recently a classification based on some network indicators was proposed in Strano et al. (2013). We propose to elaborate on these studies and to reach a quantitative classification of street patterns by extracting their “fingerprints.” These fingerprints allow us to define a measure of the distance between two graphs and to construct a rigorous classification of cities. From the network to blocks Existing classifications are based on the street network (Marshall, 2006), and while the information contained in the blocks and the streets is equivalent (up to dead-ends), the information related to the visual aspect of the street network seems to be easier to extract from blocks (also called cells or faces in different contexts). A block can usually be defined without ambiguity as being the smallest area delimited by roads. A natural idea when trying to classify cities is then to focus on blocks rather than streets. Blocks are indeed simple geometrical objects—polygons—whose properties are easily measured. The properties of blocks and their arrangement thus seem to be a good starting point for attempting a classification of urban street patterns. In Louf and Barthelemy (2014), we have gathered road networks for 130 major cities across the world, spanning all continents (but Antarctica). The street networks have been obtained from the OpenStreetMap database and restricted to the city center using the Global Administrative Areas database (or databases provided by the countries administration). We extracted the blocks from the street network and removed undesired features, and end up with a set of blocks, each with a geographical position corresponding to their centroids. Blocks are polygons that can be characterized by simple measures. First, the surface area A of a block gives a useful indication, and its distribution is an important information about the block pattern. We find that for different cities, area distributions have different shapes for small areas, but display fat tails decreasing as a power law P(A) ~ 1/4" with t » 2 (Barthelemy, 2011; Barthelemy et al., 2013; Fialkowski and Bitner, 2008; Lammer et al., 2006; Strano et al., 2012). Although this seemingly universal behavior gives a useful constraint on any model that attempts at modeling the evolution of cities’ road networks, it does not allow distinguishing cities from each other. A second characterization of a block is through its shape, with the form (or shape) factor g defined in the geography literature in Haggett and Chorley (1969) as the ratio between the area of the block and the area Ac of the circumscribed circle C Ac ‘This quantity g is always smaller than one, and the smaller its value, the more anisotropic the block is. There is not a unique correspondence between a particular shape and a value of @, but this measure gives a good indication about the block’s shape in real-world data, where most blocks are relatively simple polygons. The distributions of g displays important differences from one city to another, and a first naive idea would be to classify cities Barthelemy 267 Figure 5. Fingerprint of Tokyo. (Left) We rearrange the blocks of a city according to thelr area (y-axis), and their g value (x-axis). The color of each block corresponds to the area category it falls into. (Right) We ‘quantify this pattern by plotting the distribution P(gIA) of shapes per area category, and represented for different areas by colored curves. The gray curve is the sum of all the colored curve and represents the distribution of ¢ for all cells. As shown in the inset, we see that intermediate area categories dominate the total number of cells. (Figure taken from Louf and Barthelemy, 2014). according to the distribution of block shapes given by P(y). The shape itself is however not enough to account for visual similarities and dissimilarities between street patterns, Indeed, we find for example that for cities such as New York and Tokyo, even if we observe similar distributions P(y), the visual similarity between both cities’ layout is not obvious at all (Lou and Barthelemy, 2014). One reason for this is that blocks can have similar shapes but very different areas: if two cities have blocks of the same shape in the same proportion but with totally different areas, they will look different. We thus need to combine the information about both the shape and the area. In order to construct a simple representation of cities that integrates both area and shape, we rearrange the blocks according to their area (on the y= axis) and display their g value on the x-axis (Figure 5, left) We divide the range of areas in (logarithmic) bins and the color of a block represents the area category to which it belongs. We describe quantitatively this pattern by plotting the conditional probability distribution P(p|A) of shapes, given an area bin (Figure 5, right). These figures give a “fingerprint” of the city, which encodes information about both the shape and the area of the blocks. We can then characterize the street pattern of a city by P(y)A) and construct a distance between two street patterns (Louf and Barthelemy, 2014). In order to construct this distance, we first bin the area of blocks and we define the quantity fal) = N(gla)/'Ns as the ratio of the number of blocks having a shape factor and an area in the bin, divided by the total number of blocks in the city under consideration, We represent on the Figure 5 (right panel) these quantities in the case of Tokyo. The distance between two cities and for the area bin can be then defined as dita,0)= [ fe) -2e@ ld I 268 Environment and Planning B: Urban Analytics and City Science 44(2) Barthelemy 269 (O i jin classes obtained from Figure 6. Typical street patterns (shown at the same scale) for the four main classes ol Comparing the quanty F(piA): Buenos Ales (cop lef), Athens (tp right), New Orleans (bottom lf), Mogadishu (botcom right). (Figure taken from Louf and Barthelemy, 2013). (any other reasonable choice for this distance between functions will lead to similar results) and summing over all bins gives a distance between the city block structures D(a,b) = 5, de(a, 6) that can then serve as a basis for a classification of street patterns. Using this quantity, we obtain at an intermediate level basically four different types shown in Figure 6. ; ; Tn group 1 (Figure 6, top left), we essentially have blocks of medium size with shapes that are dominated by square shapes and regular rectangles. Small areas are almost exclusively squares, In group 2 (Figure 6, top right), there is a dominant fraction of small blocks with broadly distributed shapes. Patterns in group 3 (Figure 6, bottom left) are similar to the group 2 in terms of the diversity of shapes but is more balanced in terms of areas, with a slight predominance of medium size blocks. Patterns in group 4 (Figure 6, bottom right) contain essentially small, square-shaped blocks, together with a small fraction of small rectangles. We note that, interestingly enough, most European cities and American cities in our sample fall in their own subcategory, highlighting quantitatively the differences between the typical layouts of cities in both regions. Discussion and perspectives We discussed how various indicators based on paths and their structure can reveal important information about the organization of road networks. The important point for spatial networks is that relevant metrics have to encode in a useful way both topological and geometrical information about their global structure. In particular, we highlighted the importance of the spatial distribution of the BC, which appears as a central tool in these studies and we can expect further development in this direction. We also presented new measures about street networks, which are based on blocks. We have discussed a new way for representing cities's road network that can be seen as the equivalent of fingerprints for cities. This method provides a quantitative comparison of urban street patterns, and together with the specific knowledge of architects and urbanists, should lead to a better understanding of the shape of our cities. Further studies are needed in order to relate the different types of street patterns to various urban processes, a work already initiated by the space syntax community (Al Sayed et al., 2009). The discussion presented here suggests that we need to identify the mechanisms controlling the shape and the area of blocks, For example, small blocks can result from the natural fragmentation process of parcels, a large regularity of block shapes could be related to planning, and a large variety of shapes could result from urban modifications that do not respect the existing geometry. A complete characterization and exhaustive list of dominant mechanisms are however still lacking and we can expect exciting developments in this field in the near future. Declaration of conflicting interests ‘The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. Funding The author(s) received no financial support for the research, authorship, and/or publication of this article. References Aldous DJ and Shun J (2010) Connected spatial networks over random points and a route-length statistic. Statistical Science 25: 275-288. Al Sayed K, Turner A and Hanna $ (2009) Cities as emergent models: The morphological logic of Manhattan and Barcelona. In: Koch D, Marcus L and Steen J (eds) Proceedings of the 7th International Space Syntax Symposium. Stockholm: KTH. Ambjorn J and Jonsson P (1997) Quantum Geometry: A Statistical Field Theory Approach. Cambridge, UK: Cambridge University Press. Balean D, Colizza V, Goncalves B, et al. (2009) Multiscale mobility networks and the spatial spreading of infectious diseases. Proceedings of the National Academy of Science (USA) 106 21484-21489. Barthelemy M (2004) Betweenness centrality in large complex networks. The European Physical Journal B 38: 163. Barthelemy M (2011) Spatial networks. Physics Report 499: 1-101. SS 270. Environment and Planning B: Urban Analytics and City Science 44(2) Barthelemy M and Flammini A (2008) Modelling urban street patterns. Physical Review Letters 100: 138702. Barthelemy M, Bordin P, Berestycki H and Gribaudi M (2013) Self-organization versus top-down planning in the evolution of a city. Scientific reports 3 Batty M (2005) Cities and Complexity. Cambridge, MA: The MIT Press. Bettencourt LMA, Lobo J, Helbing D, et al. (2007) Growth, innovation, scaling, and the pace of life in cities. Proceedings of the National Academy of Sciences (USA) 104: 7301-7306. Bouttier J, Di Francesco P and Guitter E (2004) Planar maps as labeled mobiles. The Electronic Journal of Combinatorics U1: R68. Cardillo A, Scellato $, Latora V, et al. (2006) Structural properties of planar graphs of urban street patterns. Physical Review E 73. Chan SHY, Donner RV and Lammer § (2011) Urban road networks—Spatial networks with universal geometric features?. The European Physical Journal B 84: 563-577. Clark J and Holton DA (1991) A First Look at Graph Theory, Vol. 1. Teaneck, NJ: World Scientific. Courtat T, Gloaguen C and Douady $ (2011) Mathematics and morphogenesis of cities: A geometrical approach. Physical Review E 83: 036106. Crucitti P, Latora V and Porta $ (2006) Centrality measures in spatial networks of urban streets, Physical Review E 73: 0361251-S. Fialkowski M and Bitner A (2008) Universal rules for fragmentation of land by humans. Landscape Ecology 23: 10131022. Freeman LC (1977) A set of measures of centrality based on betweenness. Sociomerry 40: 35-41. Gastner MT and Newman MEJ (2006) Shape and efficiency in spatial distribution networks. STAT Ol: POLOIS. Haggett P and Chorley RJ (1969) Network Analysis in Geography. London: Edward Arnold. Hillier B and Hanson J (1984) The Social Logie of Space. Cambridge, UK: Cambridge University Press. Jiang B (2007) A topological pattern of urban street networks: universality and peculiarity. Physica 384: 647-655. Jiang B and Claramunt C (2004) Topological analysis of urban street networks. Environment and Planning B 31: 151-162. Jordan D (1995) Transforming Paris: The Life and Labors of Baron Haussmann. Chicago: University of Chicago Press. Katifori E and Magnasco MO (2012) Quantifying loopy network architectures. PLoS ONE 7: €37994, Lammer $, Gehisen B and Helbing D (2006) Scaling laws in the spatial structure of urban road networks. Physica 363: 89-95. Louf R and Barthelemy M (2014) A typology of street patterns. Journal of The Royal Society Interface 1101): 20140924. Marshall S (2006) Streets and Patterns. Abingdon, Oxon UK: Spon Press. Masucci AP, Smith D, Crooks A, et al. (2009) Random planar graphs and the London street network, The European Physical Journal B 71: 259-271 Mileyko Y, Edelsbrunner H, Price CA, et al. (2012) Hierarchical ordering of reticular networks. PLoS ONE 7: 36715. Penn A (2003) Space syntax and spatial cognition or why the axial line?. Environment and Behavior 35: 30-65. Perna A, Kuntz P and Douady § (2011) Characterization of spatial networklike patterns from junction geometry. Physical Review E 83(6): 066106. Porta S, Crucitti P and Latora V (2006a) The network analysis of urban streets: a primal approach. Environment and Planning B 33: 705-725. Porta S, Crucitti P and Latora V (2006b) The network analysis of urban streets: A dual approach. Physica A 369: 853-866. Porta S, Latora V, Wang F, et al. (2011) Street centrality and location of economic activities in Barcelona. Urban Studies 10 November 2011, 004209801 1422570. Barthelemy 27 Ramalho CE and Hobbs RJ (2012) Time for a change: Dynamic urban ecology. Trends in Ecology and Evolution 27: 179-188. Rosvall M, Trusina A, Minnhagen P, et al. (2005) Networks and cities: An information perspective Physical Review Letters 94: 028701 Southworth M and Ben-Joseph E (2003) Streets and the Shaping of Towns and Cities. Washington DC: Island Press. Shpuza E (2009) Evolution of street networks in Adriatic and Ionian coastal cities 1769-2007. In Proceedings of the 7th International Space Syntax Symposium Strano E, Nicosia V, Latora V, et al. (2012) Elementary processes governing the evolution of road networks. Scientific Reports 2: 296. Strano E, Viana M, Cardillo A, et al. (2013) Urban street networks: A comparative analysis of ten European cities. Environment Planning B 40: 1071-1086. Turner BL, Lambin EF and Reenberg A (2007) The emergence of land change science for global environmental change and sustainability. Proceedings of the National Academy of Sciences (USA) 104: 20666-20671. Tutte WT (1963) A census of planar maps. Canadian Journal of Mathematics 1S: 249-271. Venerandi A, Zanella M, Romice © and Porta $ (2014). The form of gentrification. arXiv preprint arXiv:1411.2984. Viana MP, Strano E, Bordin P, et al. (2013) The simplicity of planar networks. Scientific Reports 3: 3495, Xie F and Levinson D (2007) Measuring the structure of road networks. Geographical Analysis 39: 336-356. Xie F and Levinson D (2009) Topological evolution of surface transportation networks. Computers, Environment and Urban Systems 33: 211-223 Marc Barthelemy is a Senior Researcher at the Institute of Theoretical Physics (IPhT, CEA) in Saclay. He worked on applications of statistical physics to complex systems such as the structure of large networks, theoretical epidemiology, and more recently on spatial networks, From data analysis to spatial economics and modelling, he is currently working on various aspects of the emerging science of cities.

You might also like