Title : Urban spatial structure, daily mobility energy consumption : a study of 34 European cities
and
Auteur : Le Néchet, Florent Le Néchet, F., Université Paris-Est, Laboratoire Ville Mobilité Transport, UPEMLV : 5 boulevard Copernic Cité Descartes F 77454 Marne-la-Vallée cedex 2 FranceMaître de conférences, Université Paris-Est Marne-la-Vallée
[email protected]
Daily mobility; polycentricity; sustainability; urban morphology.
urban
structure;
urban
In this article, we compare two databases, one providing mobility data (UITP database, 2001), and the other providing population density grid cells at 200m scale for 34 European cities. Looking towards sustainable urban forms, we assess explanatory power of socio-economic and morphological indicators, with respect to seven mobility indicators, including the energy consumption per inhabitant due to transport (base year 2001). We conclude that classical urban indicators (GDP per inhabitant, population density) and more precise morphological indicators (entropy, rank-size slope) are complementary towards such objectives.Key-words:
Titre : Consommation d’énergie et mobilité quotidienne selon la configuration des densités dans 34 villes européennes.
Dans cet article, nous confrontons deux bases de données, l’une portant sur la mobilité dans 34 villes européennes et l’autre donnant une grille fine (au pas de 200 mètres) de densité de population pour ces mêmes villes. Dans la perspective d’un développement urbain soutenable, nous mettons en évidence la complémentarité d’indicateurs urbains classiques (PIB par habitant, densité de population) et d’indicateurs morphologiques plus fins (entropie, pente de la loi rang-taille) pour expliquer les différences inter urbaines de sept indicateurs de mobilité quotidienne, mettant en particulier l’accent sur la consommation d’énergie par personne due aux transports (en 2001).
Mots-clés : Mobilité quotidienne ; polycentrisme ; soutenabilité urbaine ; morphologie urbaine.
structure
spatiale ;
Introduction The complex relationships between urban structure and daily mobility are widely discussed in the literature (Camagni, Gibelli, & Rigamonti, 2002; McMillen, 2003; Aguilera, Mignot, and Madre, 2005; Acharya & Morichi, 2007; Milakis et al. 2008; Bertaud et al., 2009). Increased car use raises questions about the negative consequences of urban sprawl. In reaction with that observation, scholars and planners have tried to identify which city structures might be compatible with sustainability. Daily mobility, which is at the heart of sustainability concerns at the city level, is often linked in the discourses with urban form; Cervero (2002), among others, recommends a higher use of public transport system as well as shorter distances to jobs: this is the « compact city » model. In contrast, Breheny and al. (1997) questions the acceptability of this model and emphasize the necessity to conciliate sustainability objectives with individual inclination towards nature and space. More generally, empirical works identify statistical relationships among various attributes of urban shapes and mobility, in different geographical contexts. Some approaches emphasize the microscopic interactions among individuals, seeking to identify the individual determinants of mobility (transportation costs, proximity to different transport infrastructures, Massot et al., 2006). Other approaches characterize a city by a small number of attributes (compactness, average household size, number of trips per day), as a way to identify correlations among them. The latter approach is more planning-oriented, because it can more easily be distilled into political actions, even though it is very hard in the complex framework of cities to identify causalities, as Fusco (2004) does. Although these two approaches can be complementary (Wegener & Fürst, 1999), the approach taken in this article is clearly macroscopic, based on the initial results of an analysis of the relationships between urban form and daily mobility in 34 European cities. We demonstrate the contribution of a multivariate quantification of urban form in addition to classical socio-economic indicators, in a context of sprawling cities, and increasing polycentricity. It raises questions about the sustainability of new forms of mobility (increasing distances, heavy use of the automobile, and high energy consumption due to transport).
1 – Urban spatial structure and urban daily mobility In this section, we present the issues related to the transformation of the urban structure in European cities in the twentieth century (section 1.1): the possible relationships between configuration of the population distribution within a city and patterns of daily mobility observed are analysed by an extensive interdisciplinary literature, (section 1.2). In the last section we present the detailed objectives of this article (section 1.3).
1.1. Sustainability issues and the transformation of urban structure Automobiles did facilitate urban sprawl by providing faster travel (EEA, 2006), this encouraging residential decentralization and the intensive urbanization of suburban areas. People can live further from the center, and yet benefit from access to the core business district. According to Gordon and Richardson (1997), these people find more living space and better proximity to nature on the outskirts of cities. Wiel (1999) described the historical dynamic of cities as follows: after the industrial revolution, a « walking » city, small and radioconcentric, evolves into a « rail city », organized around rail stations in the suburbs, and then into an « automobile city » which sprawled and filled in the gaps between the rail corridors by lower density urbanisation – a few dozen people per hectare instead of a few hundred a century before, according to Newman and Kenworthy (1999), in the case of American cities. In addition to shorter travel times, the use of cars is almost uniform on the territory, while journeys made by public transport are only possible to and from a small part of today's cities: transit systems requires a minimum population density to work efficiently (Department for Transport, 2008). It follows that urbanisation now consumes a lot of space, resulting in diffuse urban forms, and causes various problems (soil waterproofing Djellouli et al., 2010, transformation of the social links, automobile dependence, Dupuy, 1999, high energy consumption related to the heating of detached houses). The automobile could be a sign of technical progress when we consider its speed and low cost of transport. At the same time, it is now criticized from many perspectives. A combination of arguments questions the extensive use of cars in today's cities. Mobility patterns and lifestyles (organized nearly exclusively around the car use in some parts of cities) also cause problems. Car emissions are still much higher than those of trains. Furthermore, it may become impossible to use oil energy over the long term, because its price might rise dramatically over the next decades. Newman and Kenworthy (1989) have popularized a simplistic curve linking human density and the energy consumption per inhabitant due to transport in 50 cities. Figure 1 shows how energy use varies between low-density U.S. and Asian cities; energy use appears to be correlated with these high densities. These studies are based on a database that excludes cities in Africa and South America, which may hinder the scope of this result.
Figure 1: Urban density and energy consumption per inhabitant due to transport of 46 world cities (based on Newman & Kenworthy, 1999)
The study by Newman and Kenworthy (1989) was criticized methodologically (Johnson et al., 1997) and from the point of view of the acceptability of urban densification policies (Breheny, 1997). This approach created a space for many other studies of the theoretical or empirical interlinkages between urban form and mobility patterns. The compact city model, which is still highly debated in today's planning literature (Dempsey, 2010), became for some authors a policy that might respect both people and the environment (Cahn, 2003).
The transformation of urban spatial structures in the second half of the twentieth century was marked by an increasing mobility and the decentralisation of activities in European cities. Beyond the diminution of residential density, the density of jobs in central cities also decreased, and the monocentric vision of a city may not be relevant anymore (Heikkila et al., 1989). Many studies have emphasized the emergence of employment centers on the outskirts of large cities in the United States (Edge Cities, Garreau, 1991, Edgeless Cities, Lang, 2002) and Europe (Bontje, 2005; Berroir et al., 2005). Knapp et al. (2006) talk about the emergence of polycentric urban regions, after the functional integration of urban centers that were historically functionally distinct. To design and evaluate new forms of urbanization constitutes a scientific challenge: how is it possible to overcome the crude description of the city given by the classical “duet” population and density?
1.2. Links between urban structure and daily mobility Empirical approaches to the interactions between land use and mobility patterns are studied on two scales: inter-urban (Newman & Kenworthy, 1999) and intra-urban (Fouchier, 1997; Pouyanne, 2005). We present (Figure 2) the classical transport - land use interaction (Wegener and Fürst, 1999). According to the authors, five parameters of land use are likely to have a significant impact on mobility patterns: – residential density, which, combined with shorter trips, may increase the modal share of public transport. – density of jobs, which has complex effects on commuting distances : - longer if the jobs are grouped into a small number of centers - shorter if the cost of transport rises and if residential areas and places of employment are balanced (see Aguilera, 2006). – local urban form, which may be able to influence modal choices (especially soft modes, walking and cycling) (see Lin, 2009). – the size of the city; a more populated city is supposed to be associated with shorter travels and higher transit use. – overall urban form (location factor).
Figure 2: Interactions between transport and land use, adapted from Wegener and Fürst (1999)
In the context of emerging polycentric metropolis (Hall, 2006), we explore possible links between polycentricity and car use, and between polycentricity and trip length. According to Schwanen (2001), who studied several Dutch cities, polycentric patterns induce greater use of the car, far from the sustainability aims outlined above. To achieve these objectives, the authors distinguish four types of Dutch cities, depending on the intra-urban flows. The authors distinguished the forms of relationships between places, "centralized" (classical monocentricity),
"decentralized" (strong links between all urban centers, even secondary ones), "crosscommuting" (strong links between suburbs) or "exchange-commuting" (monocentric flow with symmetric relationship between center and periphery). Car use appears to be statistically linked with variations of these patterns, contrary to trip length. Aguilera (2006) focus on the proportion of local trips, as opposed to metropolitan trips, which is crucial to assess the distance from the urban model of the "self-consistent city" suggested by Korsu and Massot (2004). In addition to an "injunction to densify" (Wiel, 2008), these authors would advocate some "urban consistency" (proximity of houses to workplaces) in accordance with environmental requirements. In addition to these forms of "functional polycentricity," the authors quantify and analyze the forms of "morphological polycentricity ", the spatial configuration of inhabitants through space. Kutter et al’s (1998) study of several German cities suggests a greater distance for trips in a monocentric city (Hamburg, Munich) than in a polycentric city (Stuttgart, Francfort). It suggests that this model of development may only be sustainable at the cost of significant investments in the transit system, which makes possible the pooling of the energy consumption per kilometer. For the largest cities, it may be impossible to maintain a radio-concentric model, as illustrated by the new-towns policies around London and Paris. In this study, we will only tackle the latter aspect, because we lack disaggregated data on flows of goods and people for the 34 European cities studied.
1.3. Objectives of the article The city is a complex system and the two following visions of the city, static (where people are) and dynamic (where people go) are certainly not dual. Detailed knowledge of the urban structure cannot completely determine the characteristics of mobility and vice versa. We aim at identifying such links, however weak by construction, partly because of the possibility for the policy maker to influence urban form. The literature identifies several key socio-economic indicators. Cameron et al. (2004) show the crucial role of household car ownership and population density to explain the use of private cars in seven cities. The degree of wealth, the history of the city (former presence of public transport infrastructure, for example) are other important factors to explain the distance traveled (Newman & Kenworthy, 1999). In this article, we analyze the statistical regularities between urban form specificities and mobility specificities in 34 European cities, based on two databases: UITP1 (2001) for mobility and EEA – European Environment Agency (2002). The focus will be on energy consumption per inhabitant due to transport for intra-metropolitan trips (all purposes). Is energy consumption related, all others things being equal, with a more or less intense degree of sprawl, with more or less hierarchical configuration of densities within a city? The intensity of car use plays an 1
Union Internationale des Transporteurs Publics- International Union of Public Transporters.
important role in this question. According to figures from RATP (2006), automobile energy consumption is about twice as big per passenger-km as the energy consumption of a bus and almost ten times more than the energy consumption of an underground train, a tramway or a suburban train. Distance of travel is another key dimension of sustainability of mobility patterns. The aim of this article is to investigate the contribution of a richer quantification of the urban spatial structure to improve the understanding of the determinants of daily mobility. For instance, the concept of morphological polycentricity has not, to our knowledge, been confronted with the practices of mobility in the city, too often leaving to the density the task of representing alone the concept of "urban spatial structure" (Anas, 1998) or "urban form" (Tsai, 2005). We will in the next two sections describe the mobility indicators available in the UITP database (2001) and the morphological indicators computed for the same urban boundaries from the density grid provided by the European Environment Agency (EEA, 2002; Gallego, 2008). We then try to compare indicators from these two databases in order to highlight correlations and coproductions between forms of urbanization and forms of daily mobility.
2 – Indicators from UITP database (2001) In this section, we present available data for European cities involved in this study (section 2.1). We then raise, as a methodological question, the blurred boundaries of cities and their impact on the analysis of links between urban structure and daily mobility (section 2.2). Finally, we highlight the diversity of daily mobility patterns in European cities (section 2.3) as well as the need for urban indicators to complement the traditional socio-economic and geographic attributes to explain this diversity.
2.1. Cities studied Thirty-four European cities from the UITP database were used (Table 1 and Figure 3). The socioeconomic and daily mobility indicators corresponding to these cities were obtained from the UITP database (2001) and correspond to the metropolitan boundaries defined by this database.
Popula City Amsterdam
Studied area Amsterdam agglomeration
Barcelona
Barcelona metropolitan region
Berlin
Berlin district
Bilbao
Biscaye province
Bologna
Bologna and two municipalities
Brussels
Brussels-Capital Region
Budapest
Budapest municipality
Clermont-Ferrand
Transport Area
Copenhagen
Greater Copenhagen Region
Glasgow
Strathclyde TAA
Hamburg
Transport Area (HVV)
Helsinki
Yhteistyovaltuuskunta
Krakow
Krakow municipality
Lille
Lille metropolitan area
Lisbon
Lisbon metropolitan area
London
Greater London
Lyons
Lyon metropolitan area
Popula
tion 850 000 4 390 000 3 390 000 1 120 000 434 000 964 000 1 760 000 264 000 1 810 000 2 100 000 2 370 000 969 000 759 000 1 100 000 2 680 000 7 170 000 1 180 000
City Madrid
Studied area Madrid Community
Manche
Greater Manchester
Marseil
Marseille municipality
Munich
Landeshauptstadt München Nantes agglomeration
tion
ster les
Nantes Newcas
Tyne and Wear County
Paris
Ile-de-France region
Praha
Mesto Praha
Rome
Rome municipality
Rotterd
Stadsregio Rotterdam
Sevilla
Transport Area
Stockh
Stockholms Lan
Stuttgar
Stuttgart and four Kreis
tle
am
olm t Torino
Vienna
Turin and 25 municipalities Metropolitan Transport Area Vienna Municipality
Warsa
Warsaw Municipality
Valenci a
w
5 420 000 2 510 000 800 000 1 250 000 555 000 1 080 000 11 100 000 1 160 000 2 810 000 1 180 000 1 120 000 1 840 000 2 380 000 1 470 000 1 570 000 1 550 000 1 690 000
Table 1: Thirty-four European cities studied, from UITP database (2001).
The sample of cities was dictated by the availability of data in both the UITP database (2001) (for instance, three European cities were deleted due to missing mobility data), and the EEA (2002 database) (Swiss cities were not covered by the satellite data). We worked with a heterogeneous distribution of population between 10 million inhabitants (Paris) and less than 300,000 (Clermont-Ferrand). In the following section, we study the specific effect of heterogeneous boundaries on urban form and daily mobility comparison.
Figure 3: Localisation of 34 European cities of the study.
2.2. Scope and scale of urban analysis We present a study of the influence of the scale of analysis on the observed relationships between urban form and daily mobility. To draw city boundaries is a difficult task (Guérois & Paulus, 2002). In addition to administrative definitions (the borders of Paris have not changed since 1929), two types of definitions are commonly used in urban studies: morphological, based on the spatial distribution of dwellings and functional, taking into account daily mobility of inhabitants. Geographical definitions of cities – morphological and functional areas defined by INSEE in France, "Functional Urban Regions" defined by Hall (2006) - hardly correspond to the boundaries for which mobility data are available. The delimitation of cities from UITP database (2001) often corresponds to the jurisdiction of public transportation agencies, which permitted the collection of a set of indicators from the metropolitan bodies. In some cases, including the Ile-de-France (Paris region), the transport area is simultaneously an administrative division. The territories in this study do not have the same geographical meaning; London is represented in the database by the Greater London Authority
territory, which only corresponds to the 7 million people in the most central part of the bigger metropolis described by Hall and al. (2006). At the same time, Paris is represented by the Ile-deFrance region, with size around 10 million inhabitants (Figure 7). The work of Appert (2005) attests, however, that the two regions have much closer population sizes. However, UITP has attempted to harmonize the data: administrative boundaries of suburban towns were included in Turin, Bologna and Stuttgart, for instance. It might also be argued that comparing areas of public transport authorities makes sense, because it does correspond to one among many metropolitan functions (at least for users of these networks).
Figure 4: Boundaries of Paris and London, at the same spatial scale according to the density grid of the European Environment Agency (2002).
To illustrate this problem, we show on Figure 5 a curve similar to the curve of Kenworthy and Laube (1996), comparing the use of cars and urban density, for 46 cities. For Paris and London, three nested scales were used (see Figure 4): the core area (respectively central Paris and within the "ring-road" of London); the dense area (Paris and the three contiguous departments, the most central boroughs in the case of London); and the largest areas are respectively the Ile-de-France region and Greater London.
Figure 5: Use of private vehicles and urban density of 46 world cities (based on Newman & Kenworthy, 1999, Courel et al. 2005, Mayor of London 2006).
We acknowledge the high sensitivity of data with respect to the scale of study. In the case of Paris, the assumption of Kenworthy and Laube (1996) is met: a higher density tends to indicate a greater use of public transport. For London the results show less regularity: the London « Central Area » (within the ring-road) is a business center with low residential density. If the dots corresponding to "Greater London" and "Inner London" also follow the curve of Kenworthy and Laube (1996), the Central London dot do not (the scale might be too small to be relevant to the study of daily mobility). With this sensitivity analysis, we want to relativize the importance of the error due to the scale of analysis. In the case of Paris, the shape of the scatter plot would be little affected by the scale adopted for the study, and this is also true for the most relevant geographical entities that may be called « London ». Other databases might have been used for this study: Urban Audit (Eurostat) for instance provides harmonized data for about 350 functional areas in 30 European countries (the « Larger Urban Zones » are conceptually close to functional urban regions). This interesting database could be subject to work similar to that presented here. However, the available indicators are not the same as in the UITP database (2001), and may be of lesser quality when it comes to mobility data; for instance, it does not include energy consumption caused by transport. Given the objectives of this article, we chose to work with the UITP database (2001), which we found to be a good compromise between availability and quality of data.
2.3. Corpus of mobility indicators from UITP database (2001), and daily mobility in Europe. The UITP database (2001) includes hundreds of aggregate indicators on the cost of travel, frequentation of transport infrastructures, as well as the social and environmental consequences of everyday practices of mobility. We focus on five indicators that we find particularly important to study the sustainability of mobility patterns: the number of trips per day, and the modal shares of travel modes: proportion of trips or proportion of miles achieved by soft modes (walking and cycling) and mechanized modes (by car or public transport). We will add to this description two indicators related to the consequences of these practices: energy consumption per inhabitant due to transport (from the political perspective, it could be seen as a source from the perspective of urban ecology) and the number of fatalities due to transport, per million inhabitants. The UITP database (2001) also provides geographical indicators such as population or population density. We use classical socio-economic indicators such as GDP per capita and car ownership, as well as an indication of the supply of transport (infrastructure available, by mode2). Table 2 details the indicators that have been used in this study, divided into four categories. Out of the 34 cities studied, five are located in the former countries of the Eastern Europe political block (Warsaw, Prague, Krakow, Budapest and, eastern Berlin). The historical presence of extensive public transport networks in these cities could be a limitation to the correlations that we try to underline. We postulate that these socio-economic parameters (GDP per capita ratio of transport infrastructure or individual) can partly compensate for the methodological issues that may arise from these cultural differences.
2
We use the logarithm of the ratio (km of motorways) / (km of lanes dedicated to transit).
Indicators Socio-economic Geographical
Daily mobility
Mobility externalities
UITP database GDP per capita car ownership log(ratio km motorways / km rail and bus lanes3)4 population density of population number of trips per day per person average length of a mechanized trip5 proportion of trips by soft modes6 proportion of mechanized trips by car6 proportion of mechanized kilometers by car6 energy consumption per inhabitant due to transport number of fatalities due to transport7
Abbreviat ion GDP OWN INFRA POPU DENS #TRIPS LENGTH %SOFT %CAR kCAR ENERGY RISK
Table 2: List of indicators used (UITP database, 2001).
The correlations among these indicators are rather loose, even if, for instance, energy consumption per inhabitant due to transport is positively correlated (α = 0.05) with the average trip distance, which echoes the concerns about ongoing urban sprawl and disconnection of growth between houses and workplaces in European cities (Aguilera, 2006). We developed a typology of European cities based on the three following indicators of daily mobility: the relative share of car trips among all the mechanized trips [% CAR], the relative share of soft modes among all trips [% SOFT], and the energy consumption per inhabitant due to transport [ENERGY] These indicators are independently distributed (to a 95% or more confidence degree), and we performed an ascending hierarchical classification which highlighted three types of strongly differentiated mobility patterns8. In the first class, containing the most elements, energy consumption per inhabitant due to transport is high on average and inhabitants more frequently use car to travel. The five cities of the class 2 (Amsterdam, Bilbao, Rotterdam, Sevilla, Valencia) see important use of soft modes mobility (foot or bike represent 46% of trips on average) and therefore have a low energy consumption even though most of the mechanized trips are made by car (76% on average). In the last class (10 cities), 45% of mechanized trips are undertaken via the transit system, which allows moderate energy consumption even if the soft modes are rarely used.
3
Road or rail lane dedicated to transit. For Hamburg, Rotterdam and Amsterdam, missing data in UITP database was replaced by the value zero. 5 For Warsaw, Amsterdam and Valencia, missing data was replaced by the average value of the other 31 cities. 6 « Soft modes » (walking and cycling) are opposed to « mechanized modes », which can be individual (car, motorbike) or collective (bus, train). 7 Value per million inhabitant ; for Krakow and Warsaw, missing data was replaced by the average value of the other 32 cities. 4
8
Ascending hierarchical clustering was produced using SAS software « Ward » method: inter-class inertia is the weighted distance between centers of gravity of classes and intra-class inertia is the average distance to the center of gravity of each class.
Figure 6 shows the diversity of development patterns in European cities and justifies the approach proposed by the article. We call attention to the differences between the level of development and the type of urban development. - The horizontal axis indicates the ratio between the number of kilometers of motorway network and the total size of dedicated lanes for transit (road or rail). - The y-axis is the GDP per capita of the city. Level of development in itself cannot explain the shape of today's cities: this can be an illustration of the influence of policy makers in the cases of urban development based mainly on public transport (Budapest) or auto mode (Sevilla). These clusters do not form homogeneous subsets (Figure 6). If cities with the least sustainable mobility (Class 1) seem to correspond to higher levels of wealth than the others (Classes 2 and 3), this relationship is weak: Torino, Vienna, Helsinki, Amsterdam and Rotterdam have been able to balance long-term sustainable mobility and a high level of wealth. Glasgow, in contrast, is one of the cities with the most energy use per inhabitant due to transport and the highest car use, even though it has built a strong public transport infrastructure and is not a particularly affluent city.
Figure 6: Urban development and practices of mobility in 34 European cities.
This approach highlights the need for a complementary set of indicators which would take into account factors other than level of development and availability of transport network; we
focus on morphological indicators in the following section: to what extent can the compact, sprawled and polycentric aspects of a city be linked with the observed mobility patterns?
3 - Indicators of urban structure « Urban form » of « urban structure » are ambiguous terms, often taken for granted by researcher in ecology and planning engineering; it can be observed from different perspectives, either focusing on the shape of cities as seen from the sky (Guérois & Pumain, 2008), on the urban fabric of the streets (Allain, 2005), or on the inner organization of urban areas (geographical configuration of urban centers) (Anas et al., 1998). We will use the terms "urban form" or "urban structure" for the configuration of inhabitants throughout urban space. This simplification is dictated by the description of the city that we will use, based solely on the population density grid from EEA (2002). Thus, the localisation of activities will be absent. In addition, the UITP database (2001) is macroscopic, therefore it will not be possible to work with detailed patterns of daily mobility. In this section, we detail data available to evaluate the inner configuration of densities within European cities (section 3.1); we then provide a family of urban structure indicators, intended to quantify the degree of urban sprawl (section 3.2) or morphological polycentricity (section 3.3). Finally, the correlation among morphological indicators highlights the complementarity of the innovative morphological indicators with respect to more classical ones, total population and population density.
3.1. Type of data To quantify the urban structure of a city, we use the population density grid provided by the European Environment Agency (EEA, 2002). This grid is based on satellite observations, categorized as land use typology (Corine Land Cover database), using national census from European countries. Each pixel is assigned a population, depending on the type of land use -urban continuous, urban discontinuous, among others - and the total population of the NUTS9-5 area in which it is located (Gallego, 2008). Despite the limitations of the methodology (this approach cannot completely solve the problem of heterogeneous geographic entities, but it does smooth it), we found relevant, at the scale of large urban areas, to work with this 100-meters harmonized population density grid-cell. Figure 7 illustrates the spatial information given by EEA database. We show in this figure the cities of Stuttgart (with diffused urbanization) and Barcelona (more compact, also with a large spatial extension of the core city center). Note that the maps of Figure 7 correspond to a grid of 100 meters wide, the finest available. For reasons of computing time, we merged these cells into 200-meters ones, from which we computed the indicators of urban structure described below.
9 NUTS means « Nomenclature des Unités Territoriales Statistiques » and refers to harmonized type of administrative boundaries throughout Europe.
Figure 7: Map of population of Stuttgart and Barcelona (EEA, 2002).
According to Tsai (2005): "Metropolitan form can be analysed as four distinguishable dimensions"10- size (total population), intensity (population density), the degree of inequality of distribution (concentration of population in a small proportion of the urban space) and the degree of clustering, which is the tendency for dense areas to be located next to each other. Having analyzed the 219 U.S. Metropolitan Statistical Areas (MSA) of less than 3 million inhabitants, he established the complementarity of these indicators, the correlation coefficients between them not being significant. We retain this approach to illustrate our will to quantify urban morphology beyond the density of urbanisation: we further investigate the degree of urban sprawl on one hand and the forms taken by the polycentric urban development, on the other hand, in addition to conventional indicators.
3.2. Indicators of urban sprawl The indicators below come from the literature: they merely attempt to quantify and differentiate compact and sprawled urban forms as well as monocentric and polycentric patterns. Entropy quantifies the level of organisation of the distribution. If the entire population were located in a single cell, the entropy is 0 (perfect order). In an evenly distributed population (arguably the highest form possible of urban sprawl), the entropy would be equal to 1.
(2) 11 [ENTROP]
10
Which means that urban form can be characterized by four independent indicators. M ≤ N is the number of occupied cells in the city; pi is the population of zone i et PM = PN the total population of the city 11
We will also use the average distance between two individuals within the city (3). We make the hypothesis that the relative proximity of individuals is statistically linked with the patterns of mobility observed in cities.
(3) [DIST] Grasland (2008) discuss the relevance of this indicator to analyze the spatial structure of a country. In an intra-urban context, the average distance between two people will be lower if the population is concentrated around a pole than if it is dispersed evenly. For example, the average distance between two individuals is 26 km within the Ile-de-France region (100 km times 100 km square), half of distance that would be theoretically obtained if the population was randomly distributed throughout that territory. In order to compare the values obtained for different city sizes, we will use a relative indicator (equation 4), obtained by dividing the distance between two individuals by the radius of a circle with the same area as the city. (4) [RELDIST] where (5) A morphological index of compactness (Cole, 1964, quoted by Guérois, 2003) is computed using the radius of the city we just introduced, and the maximum distance between two individuals in the city (denoted Dmax). This indicator is used to quantify the distance of the urban form to the model of a circular city. We will use the following formula: (6) [ELONG] This indicator equals 1 in the case of a circular city and is bigger than 1 in all other cases. A high value of this indicator would indicate a particularly elongated urban form, which might be incompatible with the imperatives of sustainability mentioned. We will call this indicator « index of elongation ».
3.3. Morphological polycentricity In addition to urban sprawl, we discussed the emergence of polycentric urban form, due either to the interaction between several formerly independent historical centers or polynuclear forms of diffusion from the center. Two indicators are used in the literature to discriminate monocentric or polycentric conurbations. It is important to underline that these morphological indicators are not necessarily related to functional aspects (a large literature on polycentricity focus of inner flows between subcenters). We present below various tentatives from the literature to quantify the morphological aspects of polycentricity. The slope of the rank-size rule (Zipf, 1949; Pumain, 2003) is commonly used in inter-urban context studies to quantify the hierarchy of a given territory (most frequently, a country): the two extremes models for urban networks are a « macrocephalic » country with one city much bigger
than all the others (France, UK), and a more balanced model (Germany, Netherlands). At the intra-urban level, Batty (2001) and Tsai (2005) used this indicator to quantify the degree of polycentricity of a city. The slope α of rank-size rule - formula (7) - will be referred to as hierarchical index [RSS] for “Rank-Size Slope”. In this formula, P1 is the population of the most populous cell and pk is kth ranking cell, in term of population. Out of the 34 cities studied, this equation well approximates the statistical distribution of population. (7) However, this indicator has no spatial dimension. Other indicators can be used to understand the monocentric or polycentric nature of urban region. For instance, the coefficient of hierarchy of Thiessen polygons areas corresponding to major urban centers in Europe (ESPON, 2005) has a clear spatial dimension. This approach is intended to discriminate, at the national level, countries with major centers which are concentrated in space (thus having Thiessen polygons of unequal area) from countries with a more balanced territorial organization. We will not use this idea here, because the problem of defining urban subcenters will not be tackled here; it is a complex issue according to Giuliano and Redfearn (2005). The Moran index (8) is an index of spatial autocorrelation, ranging from (-1) to 1. A value close to 0 tends to indicate a random distribution (absence of spatial autocorrelation) and a value close to 1 a spatial grouping of the densest areas (high spatial autocorrelation). A value of (-1) could be found in the hypothetical case of a « chess-board » like urbanisation (Tsai, 2005), that is to say that the boxes adjacent to one urbanissed boxes are empty and vice versa. This never happens in real cities.
(8) 12 [MORAN] The Moran index is intrinsically spatial: the relative proximity between populated areas of the grid is important to its calculation, in contrast with the entropy and hierarchy indices that are based solely on the distribution of population size. Note that Tsai (2005) uses the Moran index to quantify the level of polycentricity of a city, which he justified by a few tests on hypothetical configurations. The Moran index is a statistical indicator that assumes a normal distribution of individuals on the grid, which is probably not appropriate in most cases, since the rank-size rule (a power law) works well at the intra-urban scale: the spatial autocorrelation indicator could have been adapted accordingly. For example, it would have been possible to compute Moran index for a normalized distribution of the population, which could be approached by studying the log of the population rather than the population itself. However, because of the limited scope of this study, we retain the indicator mentioned above, as used by Tsai (2005), although more recent work (Le Néchet, 2010) highlights the sensitivity of this indicator to the size of the grid. 12
p is the population of zone i and P the total population of the N zones.d is the geographical distance i N ij between zones i et j .
Indicators from the fractal literature (De Keersmaecker et al., 2004) could have enriched the empirical approach produced, but because of time constraints it has not been implemented.
3.4. Correlations among morphological indicators Table 3 summarizes the morphological indicators described in this section. We add to the seven morphological indicators computed from the population density grid of the EEA (2002) the urban population and population density from the UITP database (2001). Indicators Polycentricity
Urban Sprawl
Geographical
Name hierarchic coefficient (slope of the ranksize-rule) Moran index entropy (degree of disorder) distance between two individuals relative distance between two individuals (compacity index) elongation index (shape of city) population (UITP) density (UITP)
Abbreviat ion RSS MORAN ENTROP DIST RELDIST ELONG POPU DENS
Table 3: Morphological indicators (sources: EEA, 2002 and UITP, 2001).
Table 4 indicates the correlations among the indicators that have been presented (the values of these indicators are given in Table 8). Note that we do not analyse statistical distributions of these indicators; the approach that we develop relies on the hypothesis that the indicators follow a Gaussian distribution, which might not appropriate for all indicators. In addition, the small number of cities involved in this study and the fact that the sample does not stem from a random sampling among all European cities, but only from availability of data, constitutes important limitations to any generalisation of the results that will be obtained. Instead, we insist on the benefits of an exploratory approach that allows to the formulation of an innovative hypothesis that might be supported by more robust statistical methods. We observe a rather low correlation (R = 0.19) between population of cities [POPU], density [DENS] and the morphological indicators. We also note that the urban sprawl indicators (entropy [ENTROP] and relative distance [RELDIST)] are strongly interrelated, and are largely independent of polycentricity indicators ([RSS] and [MORAN]).
Variable s POPU DENS RSS ENTRO P DIST RELDIS T MORA N ELONG
POP U 1 0,19 0,09
DENS
RSS
1 0,24
1
-0,02 0,61* **
-0,01 0,49***
-0,26
-0,06
-0,15 0,31*
0,36** 0,05
-0,23 -0,12 0,29* 0,01 -0,01
ENTR OP
DIST
RELD IST
MOR AN
1 -0,23 0,78*** 0,24 -0,08
1 -0,12 0,48*** 0,17
1 -0,05 -0,06
1 -0,13
Table 4: Correlations between morphological indicators ***: α = 0.01; **: α = 0.05; *: α = 0.10
Table 4 suggests that the indicators are complementary with more classical ones and enrich the description of a city (population, population density). For instance, the slope of the rank-size rule [RSS], and entropy index [ENTROP] are statistically independent with population and population density.
4 - Urban spatial structure, socio-economic indicators and mobility patterns In this last section, we compare indicators from the two databases used as a way to identify statistical links between urban forms and mobility patterns. The aim of this approach is not to identify causal links (a pitfall mentioned by Mignot et al., 2004), but instead to provide empirical insights on the complex dynamic processes involved in the production of city. We will first detail the methodology used in this study (section 4.1), then we will focus on the correlations between all indicators of daily mobility and urban structure described in the previous two sections (section 4.2). Lastly, multivariate regression will be used to highlight the complementarity between morphological indicators and socio-economic indicators to explain some of the variability of daily mobility between European cities studied (section 4.3).
4.1 Methodology Here we compare the traditional socio-economical and geographical indicators used to characterize a city and morphological indicators that we have just presented. We explain first the inter-urban differences in mobility practices from the following five indicators: population [POPU], density [DENS], GDP per capita [GDP], car ownership [MOTOR] and logarithm of the ratio between linear of transit (rail or road) and linear of highways [INFRA].
Table 5 suggests positive correlation between energy consumption per inhabitant due to transport [ENERGY] and GDP per capita [GDP] (see figure 6). However, the results from Newman & Kenworthy (1989) and Kenworthy & Laube (1996) could not be reproduced in the European context: neither energy per inhabitant due to transport [ENERGY] nor proportion of trips made by car [%CAR] were significantly correlated with the urban density [DENS]. Variables RISK
ENERGY %SOFT
#DEPL
LENGTH kCAR
%CAR
GDP
OWN
POPU DENS GDP OWN INFRA
0,21 -0,24 0,57*** 0,23 -0,22
-0,13 -0,21 0,42** 0,14 -0,08
0,25 -0,03 0,28 -0,02 -0,22
-0,03 -0,19 0,28 0,28 0,47***
0,22 -0,19 1 0,27 0,01
-0,06 0,06 – 1 0,32*
-0,09 0,12 -0,36** 0,32* 0,53***
-0,03 0,11 0,22 -0,27 0,06
-0,20 -0,12 0,19 0,26 0,53***
INFR A -0,30* -0,19 – – 1
Table 5 : Correlation matrix between mobility and socio-economic indicators. *** : α = 0.01 ; ** : α = 0.05 ; * : α = 0.10
We then show the contribution of a multidimensional quantification of morphological structure of European cities to partly explain the variance of daily mobility indicators across these cities.
4.2 Morphology and mobility Table 6 shows the observed correlations between mobility and morphological indicators. Among the significant correlations that were obtained, we confirm the result already found by various mobility studies (Schwanen et al., 2001). Average distance of travel [LENGTH] appears positively correlated with the average distance between two individuals [DIST], a proxy for the degree of sprawl of the city. The energy used for transport, per capita, is correlated with the slope of the rank-size rule [RSS] as well as the distance between two individuals [DIST]. Sprawled (high distance between two individuals) and polycentric (small rank-size coefficient) urban forms appear hardly compatible with energy-efficient forms of mobility according to these simple linear regression. As already mentioned, this type of result does not induce any causal relationship between variables. Note that from Tables 5 and 6, the share of journeys made by car or soft mode are not statistically correlated with any socioeconomic or morphological attributes of these cities, which constitutes an interesting result : local policies might be more efficient than metropolitan ones in this context (Cervero, 2002). Variables RSS ENTROP DIST RELDIST MORAN ELONG
RISK 0,34* -0,45*** -0,22 -0,42** 0,01 -0,06
ENERG Y -0,44*** 0,23 0,38** 0,15 -0,23 0,29
#TRIPS -0,30* 0,36** -0,06 0,28 -0,06 0,25
LENGTH -0,19 -0,23 0,41** -0,09 -0,17 -0,08
%SOFT -0,16 -0,19 0,03 -0,13 0,37** -0,14
%CAR -0,26 0,01 0,16 0,05 0,01 0,14
kCAR -0,21 -0,02 -0,01 0,03 0,11 -0,00
Table 6: Correlation matrix between indicators of mobility and morphological indicators ***: Α = 0.01, **: α = 0.05; *: α = 0.10
Figure 8 illustrates this result, where cities rather "monocentric", Torino, Warsaw, Valencia, seem more energy efficient than other, more "polycentric" ones : Glasgow and Brussels, for instance13.
Figure 8: Correlation between the hierarchical index [RANGT] (slope of the rank-size rule) and the average energy consumption per inhabitant due to transport.
From Table 6, entropy index [ENTROP] (degree of disorder of the spatial distribution) seems appropriate to partly explain the variance of the risk associated with transport [RISK] and the average number of trips per day per person [#TRIPS]. A high degree of disorder (high entropy) seems to correspond to lesser danger from urban transportation and an increased number of trips per person. Urban concentration (accumulation of inhabitants in a small number of cells, corresponding to lower entropy) appears in this perspective incompatible with some aspects of sustainable mobility.
13
This scatter plot may appear "pulled" by the presence at both ends of Barcelona, very hierarchical, and London, much more dispersed, having respectively a rather low and a rather strong energy consumption. In fact, to remove these two cities do not affect the coefficient of determination, R ² = 0.19. However, we acknowledge that the sample of cities is dictated by the availability of data in the UITP database (2001) and can not be considered as representative of all European cities; these regressions have to be interpreted with care.
4.3 Complementarity between urban indicators As suggested in Figure 6, the complementarity between socio-economic indicators and indicators of urban structure on the other hand, shall be further explored. We establish below the contribution of a multidimensional description of the city to identify statistical regularities with some indicators of daily mobility. Given the potential explanatory power of the different morphological indicators and the correlations between them, we only keep in this part the following indicators: entropy [ENTROP] (indicator of urban sprawl), slope of the rank-size rule [RSS] (indicator of polycentricity) and average distance between two individuals [DIST], which are statistically independent with each other. The degree of elongation [ELONG], although statistically independent from these three indicators, was excluded from the analysis because it fails to help explaining inter-city variations of daily mobility. Moreover, the three socio-economic indicators: GDP per capita [GDP], car ownership [OWN] and type of infrastructure [INFRA] (road or public) are preserved, despite the existence of a weak statistical correlation between car ownership and preferred type of infrastructure (the correlation is significant for the error rate α = 0.1). We present below the results of several multiple regressions. Initially, we confronted various mobility indicators with the three socio-economic indicators, and the three morphological indicators, separately. We then selected the most significant variables through a stepwise procedure, using SAS software14. In tables 7a and 7b, we confront the explanatory powers of various models obtained, which are modest for most indicators of mobility. In Table 7c, we bring together the socio-economic indicators and indicators of urban structure, proving in most cases an important complementarity between the two families of indicators, significantly increases the explanatory power of the regressions. Variabl e
Pr > Socio-economic indicators
R² 0,1
#TRIP S
GDP [+]
7
%SOF T
0,1 OWN [–]
ENER GY
1
GDP [+]
6
4 0,3
GDP [+]
INFRA [–]
8
4 0,4
RISK
INFRA [+]
GDP [–]
OWN [+]
9
7 0,2
%CAR LENG 14
INFRA [+] GDP [+]
GDP [+]
9
4 0,0
F F 6,7 0,014 3 3,0 0,062 5 9,3 0,000 7 9,5 0,000 1 6,4 0,004 6 2,7 0,106
Step by step selection of explanatory variables, with input and output thresholds of 0.15, which means that the panel of explanatory variables is reassessed at each iteration ; some of the variables can be removed after the entry into the model of another potential explanatory variable with bigger own explanatory power. At every step, all variables in the model are significant with the error rate α = 0.15
TH
8
6 0,2
kCAR
INFRA [+]
6 12,
8
46
0,001 3
Table 7a: Socio-economic indicators relevant to the explanation of seven indicators of mobility for 34 European cities studied. The explanatory variables are presented in the order of their entry into the stepwise model regression. Pr > Variable
Urban form indicators
#TRIPS %SOFT ENERG
ENTROP [+] no explanatory variable was significant
Y
DIST [+]
RSS [–]
R² 0,1 3 X 0,3
ENTROP [+] 6 0,3
RISK
DIST [–]
ENTROP [–]
1 0,0
%CAR LENGT H kCAR
RSS [–]
7 0,1
DIST [+] no explanatory variable was significant
7 X
F F 4 0,03 ,66 85 X X 5 0,00 ,5 39 7 0,00 ,05 30 2 0,13 ,38 27 6 0,01 ,59 52 X X
Table 7b: Morphological indicators relevant to the explanation of seven indicators of mobility for 34 European cities studied. The explanatory variables are presented in the order of their entry into the stepwise model regression.
Variable
Indicators
#TRIPS
GDP [+]
%SOFT ENERG Y
OWN [+]
GDP [–]
ENTROP [–]
DIST [–]
OWN [+]
DIST [+]
RSS [–]
ENTROP [+]
RISK
GDP [–]
OWN [+]
ENTROP [–]
INFRA [+]
%CAR LENGT H
OWN [+]
RSS [–]
INFRA [+]
kCAR
RSS [–]
ENTROP [+]
DIST [+] INFRA [+]
R² 0,2 6 0,3 6 0,5 6 0,5 3 0,4 0 0,1 7 0,3 8
F 5, 53 4, 08 9, 28 8, 07 6, 79 6, 59 9, 68
Pr > F 0,00 88 0,00 96