A novel global urban typology framework for sustainable mobility futures

Urban mobility significantly contributes to global carbon dioxide emissions. Given the rapid expansion and growth in urban areas, cities thus require innovative policies to ensure efficient and sustainable mobility. Urban typologies can serve as a vehicle for understanding dynamics of cities, which exhibit high variability in form, economic output, mobility behavior, among others. Yet, typologies relevant for sustainable urban mobility analyses are few, outdated and not large enough in scope. In this paper, we present a new typologization spanning 331 cities in 124 countries. Our sample represents 40% of the global urban population and contains the most recent data from 2008 to date. Using a factor analytic and agglomerative clustering approach, we identify 9 urban factors and 12 typologies. We discuss the implications of this new framework for researchers and planners and investigate the relationships between mobility and environmental sustainability indicators. Notably, we show an immediate application of the urban typologies to better understanding travel behavior and also describe their usage for detailed large-scale simulation in representative prototype cities for insights into sustainable future mobility policy pathways. Our data and results are publicly available for further exploration and will serve as a foundation for future analyses toward desirable urban and environmental outcomes.

In the first paragraph of the Introduction (page 1), the emissions due to transportation as (IEA 2017) were incorrectly reported. The corrected sentence is: Transportation remains a key driver for carbon dioxide emissions, accounting for nearly a quarter of the 32.3 GtC0 2 -e emissions from fuel combustion globally in 2015 (IEA 2017).
In the same paragraph, estimates from the International Transport Forum report (ITF 2017) were incorrectly reported. The corrected sentence is: The number of motorized urban passenger-kilometers is expected to nearly double to 48.4 trillion in 2050, while the number of motor vehicles on the road is estimated to grow to 2.4 billion within the same period, from the 2015 level of 1 billion (ITF 2017).
Furthermore, in describing the MassTransit Moderate typology in the last row of table 6 (page 11), 'low congestion' should be 'moderate congestion.' Finally, in the last sentence of section 3.2.4 (page 11), there was a typographical error. The corrected sentence is: The Hybrid Moderate typology is largely represented in South America (e.g. Havana, Cordoba, Panama City) and Central Asia, while the Hybrid Giant cities are chiefly found in Eastern Europe and East Asia (e.g. Daegu, Hiroshima, Sofia).
We stress that these typographical and reporting errors were limited to the paper, and have no bearing on the modeling framework presented in the manuscript.

IEA 2017 CO2 emissions from fuel combustion 2017-highlights
Technical report Int. Energy Agency Paris, France ITF 2017 ITF Transport Outlook 2017 Technical report Int.

Introduction
The rapid pace of growth in urban populations and vehicle ownership worldwide has led to an increasing demand for mobility and its associated impacts. Transportation remains a key driver for carbon dioxide emissions, accounting for nearly a quarter of the 32.3 MtCO 2 -e emissions from fuel combustion globally in 2016 (IEA 2017). Urban passenger traffic further accounts for about 30% of this contribution. The number of passenger-kilometers driven is expected to more than double to 100 billion in 2050, while the number motor vehicles on the road is estimated to grow to 2.5 billion within the same period, from a current level of 1 billion. Even more critically, CO 2 emissions are projected to increase by 60% in 2050 in the absence of concrete mitigation measures (ITF 2017).
To tackle mounting environmental challenges, metropolitan and city agencies are making greater efforts to develop and share resources for understanding city dynamics. Instruments arising from these efforts can facilitate the impact assessment of future mobility solutions at the city level. To this end, various programs have been established. We note the following key examples. The International Council for Local Environmental Initiatives 5 was founded in 1990 and has influenced over 1500 urban areas worldwide. The city vitality and sustainability 6 program was Original content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence.
Any further distribution of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI. 5 https://iclei.org/. 6 https://civitas.eu/. established in 2002 to promote cleaner transportation in European cities. Created in 2001, the transportation planning capacity building program 7 coordinates information sharing and decision-making among planners in the US toward more efficient and sustainable mobility. Another noteworthy example is the Smart City Challenge 8 launched by the US Department of Transportation in 2015. Under this initiative, 78 cities proposed technology-driven solutions to address urban mobility challenges and reduce carbon emissions, with implementations currently underway in 5 of them (U.S. DOT 2017). The unofficial Chinese four-tier city classification system (Bland and Hernandez 2016) highlights longstanding efforts to coordinate policy for development and mobility in the world's most populous country (Li 2007). The example of China also signifies its increasingly progressive approach in developing accessible and resilient transportation systems.
Yet, the above initiatives often suffer from a lack of comprehensive and quantitative global analyses of city dynamics and their heterogeneity. This type of analysis is especially critical in today's globalized transportation market, where solutions pioneered in one city are swiftly deployed in others with mixed outcomes. Consequently, results from academic efforts in understanding functional patterns in city dynamics can be harnessed to improve outcomes for sustainable urban mobility. Over the past 80 years, researchers have approached this challenge by classifying cities chiefly in terms of geography, urban form and economy. Price (1942) proposed using factor analysis to reduce the dimensions of 15 socioeconomic and demographic variables across 93 US cities. Four key factors were identified to ultimately aid the prediction of social change. Harris (1943) was then the first to empirically classify cities, with his seminal work grouping US metropolitan areas based on primary economic function: manufacturing, diversification, transportation, tourism, wholesale, retail, education and mining. Bruce and Witt (1971) also applied factor analysis in their classification, producing 13 economic types of US cities from 6 factors, following earlier efforts by Hadden and Borgatta (1965).
The economic classification of cities continues to be an active topic with recent work still being put forward by researchers, enhancing the methods used and enlarging the scope of analysis. Martin et al (2008) proposed a 15 indicator-based socioeconomic classification of 300 cities, advancing the idea of a 'canonical city' to allow for an 'economy of effort' in policy analyses. Similarly, the 'City 600' project conducted by the McKinsey Global Institute (Dobbs et al 2011) attempted to track the economic power of established and emerging global cities from 2007 to 2025 by ranking them on age, GDP and household income, although no typologies are generated.
The economic classification was soon followed by urban form analysis. Huang et al (2007), for example, clustered 77 cities using seven spatial metrics to typologize urban form in order to provide a framework for urban development analyses. More recently, these urban form dynamics were linked to sustainability and environmental indicators. Le Néchet (2012) analyzed morphological and functional indicators of 34 European cities in order to find measures of sustainable urban form. The study further determined that energy consumption is positively correlated to wealth, automobile density, sprawl, diffusion, and polycentricity. In a recent analysis of 274 cities, notable for its scope and relevance to sustainability, Creutzig et al (2015) obtained 8 typologies of energy consumption, using socioeconomic and environmental indicators.
In contrast to the existing literature above, very few studies have integrated and focused on the transportation dimension in large-scale urban classification. Thomson (1977) collected and analyzed data from 30 megacities and in the process defined 5 archetypes that broadly captured the urban characteristics of the cities. We note Priester et al (2013) and Zegras and Gakenheimer (2005) who analyzed 41 megacities on a global scale to determine future mobility characteristics. From 59 indicators, Priester et al (2013) obtained 13 factors, including congestion, taxi traffic, public transit usage, parking charges, among others. Seven typologies emerged, namely: paratransit, auto, non-motorized, hybrid, traffic-saturated, transit and the singleton, Manila. However, this work is based on data collected in 1995 9 and did not include environmental or economic factors. Subsequently, McIntosh et al (2014) investigated how urban form influenced car ownership from 1960 through 2000. This study incorporated prior typology information but was only limited to 26 cities.
To effectively address urban efficiency and environmental concerns, a mobility-oriented global urban typologization based on recent relevant data is required. As noted, the majority of existing comprehensive urban typology work has been limited to a few regions: Europe, China and the US. Those with a global scope are either now outdated or do not account for transportation-related variables. Given the significant contribution of mobility to CO 2 emissions and consequently climate change, effective pathways to sustainability must include sufficiently detailed transportation variables. Most critically, there is a dearth of publicly available data to inform decisions by planners and policymakers. The new urban typologies we have discovered in our effort target the aforementioned gaps. Our analyses span the most recent data from 331 cities worldwide, which is the largest in scope in the academic literature to the best of our knowledge. We incorporate economic, demographic, urban form, mobility and environmental indicators. Our compiled data and results are freely available and applicable to (a) further understanding city dynamics and behavior, and (b) enabling the coherent impact analysis of future scenarios towards sustainable urban mobility.

Methods
For the purposes of this urban typologization, we defined a 'city' as an urban agglomeration with a population of at least 750 000. Over 700 cities met this criterion. However, owing to the ready availability of open data sources, we further reduced our city sample to 331, ensuring global and national representativity. This final sample spans 124 countries and represents 40% of the global urban population. We collected urban data from a variety of open sources, with the aim of representing the most recent snapshot, thus limiting the variables to no earlier than 2008. Following data collection, we conducted an exploratory factor analyses to obtain latent urban attributes and consequently reduce dimensionality for further differentiation. Finally, we applied clustering methods, using the identified attributes, to obtain the typologies. We validated our results by examining typology characteristics across the factors and key variables.

Data
The data collected for this study consists of 64 indicators across seven urban dimensions: mobility, economy, environment, social development, urban form and geography. We compiled these data from public and open sources and have made them publicly available (along with necessary metadata) at http://its. mit.edu/typologies. Key sources and variables are summarized in table 1. To obtain the most recent and accurate mode share data, metropolitan, regional and national documents were accessed. We validated and cleaned the dataset by inspection. In cases where appropriate and where city-level data were unavailable, urban average values at the country level were used. Network data was obtained from OpenStreetMap. Given the inconsistencies in the boundaries of metropolitan regions, we computed the network statistics (Boeing 2017) based on the urban core of each city, which we found to be consistent across all the available cities. To prepare the data for factor analyses, each variable was scaled to a maximum value of 10.

Factor analysis
We conducted an exploratory factor analysis (EFA) to unearth the latent structure of the dataset and also to reduce dimensionality for the clustering which followed. Indicators related to metro, BRT and bikeshare were treated as left-censored variables truncated at 0, given that a sizable fraction of the cities do not have these services available. Thus, we used the generalized Tobit factor analysis approach, defining the vector of J latent response variables y*, where the actual measurement is given by y j in each case. The factor analysis model is thus: where v is the vector of variable means, L a J×P matrix of factor loadings; P is the number of factors, h the P×1 vector of variable scores on each factor, and e the J×1 vector of independent error terms The covariance structure of y* is given by where Y and Q are the factor and error covariance matrices, respectively. Further details on the formulation of this factor analytical approach can be found in Muthen (1989) and Kamakura and Wedel (2001). A maximum likelihood approach was used to estimate the model. Thus, standard errors and a test statistic were computed as robust to non-normality and non-independence of observations. To handle missing data, full-information maximum likelihood was incorporated. In the estimation procedure, the conditional distributions of variables with missingness were therefore assumed to be multivariate normal on variables with complete observations across the sample, ensuring that all observations were utilized in model estimation (Hirose et al 2016, Cham et al 2017. The rate of missingness in the dataset is 15%, and we assumed that observations were missing at random.
We note that the goal of EFA is to produce an interpretable parsimonious representation with similar correlation structure to that obtained from the original sample. Therefore, as an initial estimate of the appropriate number of factors required, we consider the eigenvalues of the sample correlation matrix. This scree plot is shown in figure 1. From this, we see that extracting any number between 8 and 12 factors will sufficiently capture the variance in the dataset.

Model fitness
To formally judge model fitness, we used likelihoodbased approaches. We first employed the likelihoodratio test (LRT) to successively test m-factor models against (m+1)-factor models. The LRT has been shown to have limitations in that it could lead to overfactoring. However, this property is not undesirable, as the greater the number of factors, the closer the estimated structure will be to the true structure (Hayashi et al 2007). These test statistics are shown in table 2. diff 2 c is the LRT statistic (Satorra and Bentler 2001) used to test the null hypothesis (H 0 ) that the data structure has at most m factors against the alternative of at least m+1 factor. In all successive models the test statistic is significant enough to reject H 0 . Clearly, we cannot rely on the LRT alone in this case for the most parsimonious choice. However the relative sizes of diff 2 c indicate that within the current range, the larger models will not drastically improve the fit.
As has been long established (Jöreskog 1969), statistical testing alone is inadequate to make a final decision. Given that the AIC and BIC both decreased as the number of factors increased to 11, we also had to consider interpretability and simplicity to guide the model selection between 8 and 11 factors. Thus, we computed the loading simplicity index (LSI) (Lorenzo-Seva 2003), which measures the simplicity of a factor solution on a scale of 0 (most complex) to 1 (simplest). These values are shown as well in table 2, providing further support for the 9-factor model. LSI * denotes the case in which only the statistically significant loadings are used. Finally, we visually inspected each solution and chose the 9-factor model as it gave the most parsimonious and interpretable results. The factor loadings L were obliquely rotated using the Geomin method (Yates 1988) for optimal interpretation. We based our choice of the oblique rotation on the hypothesis that the latent factors are correlated. Further, Geomin has been shown to provide stable loadings and better results when the original loading pattern is unknown, as was the case here (Celimli Aksoy 2017). The objective of the rotation is to maximize interpretability by minimizing the criterion Q using an iterative approach that begins with a random starting point. λ jp are the elements of the factor loading matrix, while ε is a small number added to prevent indeterminacy when λ jp =0. We used the recommended 100 random starts with a maximum of 10 000 iterations to find the best local minimum. In the objective function, the rotated loading matrix AT L = , where A is the unrotated loading matrix. Thus, the solution finds the optimal rotation matrix T . The rotated factor loadings and standard errors are shown in table 3. Factor correlations are shown in table 4 10 .
Finally, we computed the factor scores, i.e. the reduced dataset, using the components method. Specifically, we found the weight matrix W R where R is the sample correlation matrix. The reduced dataset was then given by Y Y W 331 9 331 64 64 9 =´´.

Hierarchical agglomerative clustering
Hierarchical clustering has been shown to be effective at pattern recognition, particularly because it is unsupervised and thus requires no predefined assumptions on the nature of the data (Jain et al 1999, Jain 2010. Given a separation metric between elements in a multidimensional set, hierarchical clustering proceeds by iteratively agglomerating groups based on various criteria (e.g. Ward's method, method of averages (UPGMA), single linkage, among others).
Using the 331×9 factor score matrix obtained from the factor analytical procedure, we computed a 331×331 dissimilarity matrix based on the Manhattan metric. Upon comparing the results of several agglomerative approaches, we finally selected Ward's method (Murtagh and Legendre 2014) to cluster the cities based on the factor score dissimilarities. Ward's method has historically been shown to consistently perform best in a variety of applications (Kuiper and Fisher 1975, Blashfield 1976, Hands and Everitt 1987, Milligan and Cooper 1988, Ferreira and Hitchcock 2009. In this case, it produces the most balanced clusters with assignments adjudged to be the most valid. In addition to selecting the clustering method, the other important question was the number of relevant clusters in the dataset. Various measures and heuristics have been proposed to elicit the cluster number that produces the optimal separation of members into groups, but the objective of interpretability ultimately dictate the final choice. We compared the results of 30 clustering validity indices using the NbClust package in R (Charrad et al 2014). With the minimum number of clusters set to 4, nine of the indices (including Hartigan, Tau and Ratkowsky) indicated that 5 was the optimal number of clusters. As we increased the minimum cluster number constraint, successively optimal cuts were found at 6, 8, 9, 11 and 13 clusters, respectively, as can be seen in the dendrogram (figure 2).
We observed that a 6-cluster result produced the higher-level typologies as described in section 3.2, namely: Auto, BusTransit, Congested, Hybrid, Metro-Bike, and MassTransit. The 13-cluster result splits the MassTransit Heavyweight into two further groups which are distinguished by population size. Moving further up the tree to 12 clusters produced the pairwise members of all the aforementioned higher-level groups. Given the interpretative benefit of the 12 cluster solution, however, we chose this as the final result. Our decision was supported by a precedent for not always following the majority rule in choosing the optimal cluster number (Charrad et al 2014). After the first clustering solution was obtained, we iterated by imputing the cluster means into the dataset at the factor score computation stage. This iterative process was repeated until the cluster results were stable in order to obtain the final typology assignments.

Results and discussion
3.1. Urban factors Nine urban factors were discovered from the factor analysis, namely: Metro Propensity, bus rapid transit (BRT) Propensity, Bikeshare Propensity, Development, Sustainability, Population, Congestion, Sprawl and Network Density. These are the dimensions upon which the typologies are defined. The factor descriptions are summarized in table 5. Further, we show the key significant variables and their respective loadings on each factor in figure 3.

Typology descriptions
The 12 typologies can be grouped into pairs, namely: Auto, BusTransit, Congested, Hybrid, MetroBike and MassTransit. A summary of each typology showing its key features and representative cities is provided in table 6. The factors also enable us to better characterize the typologies, as seen through the spider plot profiles in figure 4. The geographic distribution of the typologies is shown in figure 5.

Auto Innovative and Auto Sprawl
The Auto cities are modern and highly industrialized, but marked by a history of car-driven development.
Thus, both typologies have the lowest mass transit modeshare (figure 6(c)). Auto Innovative comprises the subset of relatively dense North American agglomerations with extensive rapid transit systems (e.g. Boston, Toronto, Chicago). Further, it has the highest average Development score (0.82) of all the 12 typologies.  Bike mode share, street length, safety, efficiency, low congestion Sprawl Road deaths, high car mode share, low gas price, CO 2 emissions, street length Network density High intersection density, high street density, low street length average, low circuity Meanwhile, Auto Sprawl ranks second highest on the Development factor (0.74), Auto Sprawl tellingly scores much lower in Network Density (0.41 compared to 0.58 in Auto Innovative). Of all the typologies, Auto Sprawl has the highest car mode share and ownership (figure 6(c)), highest emissions ( figure 6(b))

BusTransit Dense and BusTransit Sprawl
These typologies consist of cities with high bus (public transit) usage, along with some of the lowest Development and Sustainability average scores. The BusTransit Dense cities have the highest BRT Propensity score (0.56), and they outscore BusTransit Sprawl in Population, Congestion and Network Density. However, BusTransit Sprawl has the highest Sprawl average (0.72 compared to 0.71 in Auto Sprawl). Key examples of BusTransit Dense cities are Rio de Janeiro, Jakarta, Tehran and Mexico City. These are highly populated, dense cities with good mass transit systems, particularly with respect to BRT. BusTransit Sprawl includes several Latin American cities (Caracas, Puebla, San Salvador), along with others across the world, including Almaty, Cape Town and Isfahan.

Congested Boomer and Congested Emerging
The Congested Boomer cities are characterized by joint highest scores in Population and Congestion, compared to other typologies. Congested Emerging also features the second highest Congestion average score (0.69). However, both typologies rank the lowest in Development (Boomer: 0.14; Emerging: 0.08). The Boomer cities have the highest Population score (and also the highest average population growth and density figure 7(g)). Notable in this typology are the Indian subcontinental urban centers (e.g. Delhi, Mumbai, Dhaka) along with Lagos, Manila, among others. Remarkably, Congested Emerging has the lowest CO 2 emissions per capita and the highest public transit mode share (figures 6(a), (b)). This could be explained by its low levels of industrialization and wealth.

Hybrid Giant and Hybrid Moderate
The Hybrid cities are fairly dense, growing urban centers, with solidly average performance across all urban factors. They also have among the highest mass transit mode shares (about 40%) and share similar CO 2 emissions (figure 6(a)) and car usage characteristics ( figure 6(b)).
The chief distinguishing factor between the Hybrid Giant and Hybrid Moderate typologies is Metro Propensity, as Giant cities have an average score of 0.37, compared to 0.09 for Moderate cities. Giant also ranks slightly higher on Congestion and Sprawl compared to Moderate. Furthermore, Giant cities have an average population of 2.7 million, compared to 2 million for Moderate cities. The Hybrid Moderate typology is largely represented in South America (e.g. Havana, Cordoba, Panama City) and Central Asia, while the Hybrid Moderate cities are chiefly found in Eastern Europe and East Asia (e.g. Daegu, Hiroshima, Sofia).

MetroBike Emerging and MetroBike Giant
These MetroBike cities are heavily populated and rapidly urbanizing metropolitan areas in China, where rapid transit and bikeshare systems are widely available. The pace of bikeshare growth in China has been so rapid in recent years that it has posed recent challenges in urban management (Shi et al 2018).  is dominated by European cities (e.g. Antwerp, Brussels, Helsinki), along with major Israeli cities and Ottawa.

Emissions and urban sustainability
The typology classification provides a unique way of viewing and understanding urban and environmental metrics. Box plots of selected urban and environmental indicators are shown in figure 6 by typology. We see that the Auto cities, for example, have the highest car mode share and CO 2 emissions per capita. The MetroBike cities stand out with the highest bike mode shares. They also feature the highest highway proportion of all typologies.
Scatter plots of two environmental variables, annual CO 2 emissions per capita and Pollution Index (Numbeo 2017), against selected urban and economic indicators are shown in figure 7. (The Pollution Index is calculated on a scale of 0 to 100, in increasing order of perceived pollution of air, water and the environment.) The data points shown are the average values for each typology. The Vehicles per capita variable was not used in the factor analysis given that it was highly correlated with car usage, and also given that city-level data were not readily available for many cities. From the plots (figures 6(a), (b)), its relationship to CO 2 emissions and Pollution provides validation of the typology classification. For MetroBike Giant and MetroBike Emerging, the relatively higher emissions/vehicle indicates the impact of the economic and industrial activities that also contribute to their emissions.
On one end of the spectrum are the Auto Innovative and Auto Sprawl cities (colored light and dark blue) with the highest CO 2 emissions per capita and the lowest Pollution Index. Congested Boomer and Congested Emerging cities are on the other end of the spectrum with the opposite ranking on those two indicators. Denser cities tend to be more polluted but contribute less to CO 2 emissions per capita. Unsurprisingly the wealthiest cities (those with the highest GDP per capita) contribute the most to per-capita CO 2 emissions. Highway proportion, an indicator of a car-centric mobility infrastructure culture, appears to be a good indicator of CO 2 emissions per capita. The recently developing MetroBike cities are an exception here: with the highest highway proportion of around 4%-5%, their contribution to CO 2 emissions is comparable to cities with half that percentage.
We note that the variables discussed here were chosen based on their relevance to environmental sustainability. Further relationships among the variables can be explored across the typologies to yield findings of interest. These could be treated in more depth in a future study.

Potential applications
4.1. Prototype city generation for simulation of future mobility A major application of the typologies presented in this paper is in the creation of prototype cities for largescale future mobility simulation. The typologies provide a framework for coherently analyzing the impacts of relevant future mobility scenarios across different city types. Thus, we have developed an approach for high-fidelity large-scale simulation-based analyses of the urban typologies. This hinges on the generation of a prototype city designed to be broadly representative of the cities within a given typology. We generate and validate the prototype city based on population, landuse, demand (activity and mode choice) and supply characteristics, building on earlier efforts to model the Boston metropolitan area in the US (Viegas de Lima et al 2018). Using this approach, we have created and simulated prototype cities for Auto Sprawl and Auto Innovative in a state-of-the-art simulator, Sim-Mobility (Adnan et al 2016). These detailed Agent-based simulations on such prototype cities can yield insights into the broad impacts of future scenarios in these typologies. Considerable time and effort can thus be conserved by using the typology-level evaluations to guide the selection of viable sustainable mobility pathways for more detailed analyses at the city level. Precedents for this sort of application can notably be found in the work of Thomson (1977), where archetype strategies were empirically determined and analyzed for groups across 30 megacities. Fielbaum et al (2016) also created a parametric representation of cities as a tool for benchmarking and evaluation of mobility scenarios. In this case, we leverage on the representational power of an agent-based simulator to facilitate typology-relevant scenario explorations.

Latent class choice model framework for behavioral implications
We have obtained access to a unique behavioral dataset sampled globally from cities across 52 countries (Dalia Research 2017). The survey, conducted by Dalia Research, provides individual characteristic and mobility preference data (car ownership, travel mode, travel time, distance of driving), tendency to purchase and use electronic vehicles (EV), and safety perceptions of autonomous vehicles (AV). Using a latent class choice modeling (LCCM) framework informed by the 9 urban factors discovered, we plan to estimate a model to explain travel behavior to further confirm the validity of the typologies presented in this paper. Moreover, capturing taste heterogeneity in mobility behavior across typology-specific models, the LCCM approach has the added advantage of predicting current, and potentially future, travel-related choices, which are relevant for sustainable urban mobility. The latent urban typology structure is identified by maximizing the total likelihood of observing the choices of individuals from different cities. Ultimately, we can obtain a probabilistic typology profile for each city, which provides richer information compared to a deterministic assignment to only one typology.

Conclusion
We have discovered a new set of urban typologies based on 64 urban indicators from 331 cities spanning 124 countries across all continents. The cities represent 40% of the global urban population in 2016, while the indicator data have been compiled from open sources dating back to 2008. Our effort is the most comprehensive mobility-oriented classification to date that also incorporates environmental variables. Using factor-analytic and agglomerative clustering approaches, we identified 9 urban factors and 12 typologies. Each typology captures distinct urban outcomes and serves as a potential testbed for sustainable mobility implementations.
Our results uncover critical typology patterns that will enable researchers to focus future efforts in mitigating environmental concerns. In particular, the MetroBike cities in China are the fastest growing, along with the greatest proportion of highways. The Congested Boomer typology is also notable for its large population density and congestion problems, yet relatively low CO 2 emissions. Predominantly in North America, the Auto cities are exceptional in their car usage, wealth and CO 2 emissions. The MassTransit Heavyweight and MassTransit Moderate typologies, represent potentially desirable outcomes in sustainable mobility. Further detailed comparative analyses on these typologies can yield valuable insights for urban planners and policymakers.
We have indicated how the typologies can directly impact policy through agent-based simulation of prototype cities. We plan to conduct simulations of alternative mobility scenarios beyond automated mobility-ondemand and on an expanded set of prototype cities, in order to generate insights for optimal policy approaches that cities can adopt to effectively harness new vehicle technologies and mobility services for overall social and environmental benefits. We have also indicated an ongoing extension of the typology discovery effort by incorporating behavior and perceptions using a recent survey. With the aforementioned LCCM approach, we expect to further enhance our typology specifications and provide insights into the likelihood of EV and AV adoption across various urban typologies.
Our data and results have been made publicly available at http://its.mit.edu/typologies. We expect that these will be valuable to researchers and planners as a foundation for further research, as well as aid policy efforts in reducing the environmental impacts of urban mobility for a sustainable future.