Discovery of Urban Typologies

Urban Morphology (2012) 16(1), 27-40 © International Seminar on Urban Form, 2011 ISSN 1027-4278

On the discovery of urban typologies: data mining the manydimensions of urban form

Jorge GilAfdeling Urbanism, Faculteit Bouwkunde, Technische Universiteit Delft, Julianalaan

134, 2628 BL Delft, The Netherlands. E-mail: [email protected]

José Nuno BeirãoAfdeling Urbanism, Faculteit Bouwkunde, Technische Universiteit Delft, Julianalaan134, 2628 BL Delft, The Netherlands, and Faculdade de Arquitectura, UniversidadeTécnica de Lisboa, Rua Sá Nogueira, Pólo Universitário, Alto da Ajuda, 1349-055

Lisboa, Portugal. E-mail: [email protected]

Nuno Montenegro and José Pinto DuarteFaculdade de Arquitectura, Universidade Técnica de Lisboa, Rua Sá Nogueira, Pólo

Universitário, Alto da Ajuda, 1349-055 Lisboa, Portugal. E-mail: [email protected]; [email protected]

Revised version received 21 June 2011

Abstract. The use of typomorphology as a means of understanding urbanareas has a long tradition amongst academics but the reach of these methodsinto urban design practice has been limited. In this paper we present a methodto support the description and prescription of urban form that is context-sensitive, multi-dimensional, systematic, exploratory, and quantitative, thusfacilitating the application of urban typomorphology to planning practice. Atthe core of the proposed method is the k-means statistical clustering techniqueto produce objective classifications from the large complex data sets typical ofurban environments. Block and street types were studied as a test case and acontext-sensitive sample of types that correspond to two differentneighbourhoods were identified. This method is suitable to support theidentification, understanding and description of emerging urban forms that donot fall into standard classifications. The method can support larger urbanform studies through consistent application of the procedures to different sites.The quantitative nature of its output lends itself to integration with othersystematic procedures related to the research, analysis, planning and designof urban areas.

Key Words: urban typologies, data mining, GIS, urban design

The use of typomorphology as a means ofunderstanding urban areas has a long researchtradition (Moudon, 1994, 1997). However, thereach of these methods into urban designpractice has been limited (Hall, 2008),encountering resistance in the establishedurban development processes and within thearchitecture and planning communities

(Samuels and Pattacini, 1997; Trache, 2001).This is despite recognition of the importanceof typology-driven approaches in achievingresponsive and responsible urban environ-ments (Habraken 1988; Kelbaugh 1996;Rapoport, 1990; Samuels 1999) and increasedinterest in their application in urban design andeducation (Beirão and Duarte, 2009; Lee and

28 On the discovery of urban typologies

Jacoby, 2011; Parish and Müller, 2001).Several possible causes are advanced by the

various authors: first, the analytical process islaborious and not entirely objective; secondly,the classic urban typomorphology studies arebased on very specific geographical regionsand focus on restricted urban form traditions;thirdly, more work is required on recent cityforms (Maller,1998); and fourthly, there is aneed to integrate different morphologicalapproaches to obtain a more complete andcomplex set of urban environmental attributes(Conzen, 2010; Osmond, 2010; Wineman etal., 2009). When type is applied to practiceother problems are identified. These include alack of conceptual rigour and incompleteunderstanding of the typological approach; adifficulty in going beyond a type’s superficialtraits; and an ignorance of the deepersignificance and history of specific types(Grant, 2001). In particular, Shane (2011)identifies the current challenge to the typo-logical urban design approach posed by veryrapid large-scale urbanization in non-Westerncountries and the increasing number ofinformal settlements. He doubts the existenceof typologies suitable for those specificcontexts and the adequacy of the methods inresearch and practice to cope with the dynamicnature of the problem.

The use of computer technology to supporturban morphological studies and to buildbridges to contemporary urban designprocesses is essential and its effectiveness hasbeen shown in data analysis and the visual-ization of urban form at all scales (Lee et al.,2006; Lo, 2007; Moudon, 1997; Osmond,2010).

In this paper a method is presented thatsupports the description and prescription ofurban form in typomorphological studies andtypological urban design processes. The goalis to facilitate the application of urban typo-morphology in planning practice by develop-ing a method that is context-sensitive, multi-dimensional, systematic, exploratory, andquantitative. The proposed method takes alarge number of attributes of the characteristicsof a given urban area, uses data miningtechniques to reveal the block and street types

present in that area and presents the results ina detailed quantitative format amenable toapplication in parametric modelling of urbandesign. At the core of the proposed method isthe k-means statistical clustering technique thatdeals with large complex data sets typical ofurban environments and produces objectiveclassifications that are site and project specificand not derived from previously defined types.The proposed method thereby helps to revealthe intrinsic nature of local types and identifypreviously unknown morphological types.

The paper is structured as follows. First, theapplication of quantitative methods to urbanform classification is reviewed, introducing theconcept of data mining as a technique formultivariate classification that can be appliedto architecture and planning. The stages in theproposed typological exploration method arethen described, explaining the variousoperations and the outcomes of each stage. Inthe following section, the results are presentedof a test case that demonstrates the capabilityof the method to identify different types thatare consistent with two distinct urban fabrics.The discussion focuses on the possibilities andbenefits of applying the proposed method totypomorphological research and typologicalurban design practice. In conclusion themethod is assessed, highlighting strengths andshortcomings and considering possible futurework.

Quantitative classification of urbantypologies

Several urban form studies provide detailedanalysis, description and quantification ofurban environments, offering differentmethods to classify urban entities in order toobtain urban typologies in search of a betterunderstanding of city form and its qualities.Some focus on the typologies ofneighbourhoods (Peponis et al., 2007;Wineman et al., 2009) or identification of‘urban structural units’ (Haggag and Ayad,2002; Osmond, 2010). Others focus on theoverall form of settlements (Marshall andGong, 2009). Here we shall concentrate on

On the discovery of urban typologies 29

studies that address the components of theselarger scale entities, namely the urban blockand the street.

Urhahn and Bobic (1994) identify principlesof good city life and catalogue urban neigh-bourhoods through a quantitative andqualitative description. The study coversseveral scales, from city to district, block andbuilding, and includes different classificationbases, namely form, density, land use, andmobility infrastructure. The final presentationof the typology is textual for more complexdimensions, such as urban context andaccessibility, but it is quantitative for the builtform dimensions. Overall it is highly visual,displaying the various attributes of each area ina disaggregate format. Interestingly, the streetas a classification entity is formally ignored,although it receives a brief mention in somedescriptions.

Streets receive full attention from StephenMarshall (2005), underlining the importance ofurban layout and configuration for urbanquality. He exposes the limitations of certainclassifications and catalogues of types as theyoffer a univariate interpretation on a theme,resulting in a fragmented view. Marshall usesquantitative attributes relating to configuration,composition, complexity, and constitution ofstreets, combined in triangular multivariatecharts, to define street typologies.

Berghauser-Pont and Haupt (2004, 2010)take a similar multivariate approach in relationto urban blocks around the theme of develop-ment density, using a set of four indices. Anovel aspect is that they create an interactiveon-line tool so that users can systematicallymeasure neighbourhoods and compare themwith the ones in the main catalogue, thusclassifying new designs or identifying newtypologies (http://www.permeta.nl/spacemate/index2.html). They restrict their themes tothree or four variables in order to achieve away of defining and visualizing the typology.However, other methods allow a higherdimensional classification of types.

Systematic classification using data miningtechniques

The data mining process is characterized by arecursive withdrawal procedure supported bya statistical platform leading to informationdiscovery, and is commonly used to performthree different tasks (Fayyad et al., 1996):first, classification – arranging the data intopredefined groups; secondly, clustering –where the groups are not predefined and thealgorithm creates natural groups of similaritems; and thirdly, regression – to find afunction that models the data with the leasterror.

Technically data mining is the process offinding data correlations or data patternsamongst dozens of fields in large relationaldatabases. Data mining seems to facilitate thediscovery of data patterns that would be verydifficult, if not impossible, to reveal by manualmeans and to quantify. The relevance of thesetechniques to urban morphological researchand design is that they allow the users toanalyse the complex urban environment fromdifferent angles simultaneously, categorize it,and summarize the relationships identified.

Recent studies use clustering as aclassification technique in the comparativestudy of buildings: for example, in definingarchetypal office building layouts (Hanna,2007), and Arabic house types (Reffat, 2008)and in identifying residential building typesaccording to energy use, correlated withbuilding age (Alexander et al., 2009). At theurban scale there have been studies on urbanblock shape and density (Laskari et al., 2008),of neighbourhoods (Thomas et al., 2010) andof whole cities (Crucitti et al., 2006;Figueiredo and Amorim, 2007). Theseexamples demonstrate that the use oftechniques of semi-automatic classification ofdata patterns according to multiple variablesreveals building and urban form types in asystematic way.

The method we propose focuses on theidentification of individual block and streettypes using a k-means clustering technique,which is described next.


A method for the discovery of urban formtypes

The method for identifying urban form typesbased on data mining techniques uses therecommendations of Witten and Frank (2005).It has three main phases: representation,analysis and description. These are brokendown into the following tasks:

1. Representationa. Preparation of the planb.Selection of classification attributes

2. Analysisa. Spatial analysis of the planb.Statistical clustering of attributes

3. Descriptiona. Statistical profiling of typesb.Semantic description of types

Representation

The representation phase involves thepreparation of the geometric data of the planand the selection of classification attributes.The information can be gathered in ageographical information system (GIS) tofacilitate management, analysis andvisualization of the large amounts of data,although it could conceivably be done in otherplatforms.

The selection of classification attributes isan important step in the process and it isessential to reach agreement on whichattributes to use to describe the urban form,how they relate to performance, and how tocalculate them. Because different attributesgive different meanings, one needs ameaningful set tailored to address the specificproblem in order to obtain useful typologies.‘The best way to select relevant attributes ismanually, based on a deep understanding ofthe learning problem and what the attributesactually mean’ (Witten and Frank, 2005, p.289).

There is not a right or wrong set ofattributes. But there are different perspectives(Habraken, 1988; Marshall and Gong, 2009);focusing on structural, geometric, relational,

physical, stylistic, historical or socio-economiccharacteristics. One of the benefits of thismethod is the ability to combine a large andvaried set of attributes originating fromdifferent aspects of urban morphology, thusfacilitating an integrated approach (Conzen,2010).

Analysis In the analysis phase the attributes of the siteplan are measured and their importance andthe relation between them are evaluatedstatistically. After completing spatial analysisit is important to visualize the individual urbanform attributes through maps, as it helps toverify the representation and identify mistakesin calculation. This is also a first step inbecoming familiar with the individualmorphological characteristics. At this stage,traditional urban morphological studies resortto town-plan analysis to understand anddescribe the urban environment (Maller, 1998;Osmond, 2010), but this is when clusteringbecomes a useful support method by analysingall the attributes simultaneously.

Before proceeding it is still necessary totransform the attributes to obtain a normaldistribution of values required by moststatistical operations and perform pair-wisecorrelations to identify and exclude dependentattributes that would bias the study towards aparticular theme.

A classic k-means clustering technique(Witten and Frank, 2005, p. 137) is thenapplied to identify urban form types within thegiven area. Clustering allows the classificationof instances in multi-dimensional space wherethere are no classes defined beforehand. Thek-means algorithm, as found in most standardstatistical analysis packages, is a partitioningprocess that subdivides a large data set into ak number of clusters seeking to minimize themean distance between all members of eachcluster. To determine the best number ofclusters (k) one can use a scree plot. This chartplots the sum of squared distances of everyinstance to its cluster centroid, for all clustersand for an increasing number of clusters. As


the number of clusters increases, this distancewill naturally decrease. As a result of the k-means clustering analysis, for every element inour plan, the cluster number it belongs to andits distance to the cluster’s centroid areobtained. This allows the most central elementof the cluster to be selected as an archetype.

Description

The description phase translates the results ofthe clustering process into urban form typesusing a quantitative profile and a semanticdefinition; two formats that are useful forurban analysis and design. To facilitate thedescription process we translate the attributes,which in most cases are continuous numericalvalues (for example, area or length), intoclasses of values. This is called data discret-ization (Witten and Frank, 2005, p. 296) andcan be achieved with quantiles, equal intervals,natural breaks or domain knowledge classes.Ideally there are domain knowledge classesthat are meaningful to the community ofexperts or practitioners.

The quantitative profile of each urban formtype indicates the range of values of thevarious attributes and their composition interms of classes of values. For the semanticdescription of the types we only focus on thosecharacteristics that are dominant or unique inorder to highlight the specificities of each type.

Demonstrating and testing the method

The method presented in the previous sectionis now applied to an urban data set fordemonstration purposes and to test whether itoffers the desired characteristics: context-sensitive, multi-dimensional, systematic,exploratory and quantitative.

The test case consists of two neighbour-hoods in Lisbon, Portugal that are adjacent butdifferent in character. The first is the Expo 98PP4 site, the northern part of the 1998 worldexhibition site, which is a contemporaryneighbourhood, planned from scratch on abrownfield site and developed over the first

decade of the twenty-first century. Theadjacent Moscavide is a neighbourhoodfounded in 1928 and developed more slowlyover the following decades: it has sufferedfrom densification, in particular inside theurban blocks, owing to a strongly boundedlocation without room for expansion. Can thismethod identify different urban form typesbetween these two sites?

Representation

We first prepare the features describing thetwo neighbourhoods, both in terms ofgeometry and plan information (Figure 1), andload these data layers in a GIS:

1. Building: any built-up object, both publicand private.

2. Open space: empty space within blocks,both public and private.

3. Plot: the legal boundary of a property,containing buildings and open space.

4. Block: group of plots and private or publicopen space, forming an island surroundedby transport network.

5. Pavement: the public space between theblocks and the roads.

6. Road centre line: linear representation ofthe street network.

We then select a combination of differenttypes of characteristics to demonstrate thepotential for multi-dimensional and inter-disciplinary urban morphological studies(Table 1). This includes built form and openspace dimensions for blocks and streets,density metrics for blocks, and networkconfiguration metrics for streets (Berghauser-Pont and Haupt, 2010; Figueiredo andAmorim, 2005; Hillier, 2007; Marshall, 2005).

Analysis

Using GIS, we perform spatial analysisoperations to obtain all the required attributevalues listed in Table 1 and produce maps ofthe individual attributes of blocks and streets


(Figure 2). This information is used in theclassification of urban form types, but afterinspection of the maps it becomes clear thatindividual characteristics are unable to capturethe different character of the two neighbour-hoods. One would normally resort to asubjective and laborious process of cross-referencing the various maps to piece togethera collection of possible urban form types or tomatch the instances in the plan to the types ofa pre-defined typology.

Using the proposed method, we run theclustering analysis on all attributes simul-taneously and obtain sets of block and streetclusters with different numbers of classes (k)each. To select the most suitable number ofclasses we produce a scree plot (Figure 3) andchoose the smallest k where the plot shows akink after which the curve becomes flatter.

As a result we obtain six clusters of blocksand four clusters of streets, where each cluster

represents an urban form type. To verifywhether the distinction between types is clearwe create maps of the blocks and the streetsdistinguishing the cluster identificationnumber of each instance (Figure 4). Visualinspection of the cluster instances on the planof the urban area (Figure 4) demonstrates theextent of typological overlap betweenneighbourhoods. Some blocks in Moscavideare more recent and correspond to the typesfound in the Expo 98 site, and some streettypes, such as the cul-de-sac, are widespreadand can be found in both areas.

Description

The resulting clusters can be describedquantitatively and semantically to convertthem into meaningful and useful urban formtypes. The quantitative descriptions of the

Figure 1. Plan of the two test areas (Moscavide and Expo98PP4).


Table 1. The block and street characteristics selected for analysis and typologicalclassification

Entity Attribute Theme Code Calculation

Block, street Length Dimension LEN mWidth Dimension W mOrientation Dimension DIR degreesSolar orientation Dimension SOLO N,S,E,WNumber of buildings Density BLDN integer

Block Area Dimension TA m2

Built-up area (footprint) Dimension BA m2

Gross floor area Dimension GFA m2

Perimeter Dimension PER mProportion Shape PROP LEN / WArea perimeter ratio Shape APR TA / PERFloor area ratio Density FAR GFA / TAGround space index Density GSI BA / TALayers (number of floors) Density L GFA / BAOpen space ratio Density OSR (TA-BA) / GFAPrivate space area Land use PRVA m2

Public space area Land use PUBA m2

Street Pavement width Land use PAVW mPedestrian area Land use PEDA m2

Connectivity Network CON DegreeContinuity (angular) Network CNT DegreeGlobal accessibility Network ACCG ClosenessLocal accessibility Network ACCL ClosenessGlobal movement flow Network MOVG BetweennessLocal movement flow Network MOVL Betweenness

types include a series of reference values –maximum, minimum and mean – of eachattribute in a given type, and the representationof the type in a profile chart (Figure 5). Toachieve this the range of values of eachattribute is separated into quartiles. In theprofile chart each attribute is represented by abar displaying the share that it has of eachclass of values (high, medium-high, medium-low and low) in a particular type.

A quantitative analysis of the profile chartprovides the dominant and unique character-istics of a type. Dominant characteristics areconsidered to be those that have a 70 per centor higher share of a single class value – forexample, 94 per cent of type 3 blocks have anarea of public space classified as very low.Unique characteristics are regarded as being

those that have a class share that is 50 per centabove or below the average of that classamong all types: for example, only 6 per centof type 2 blocks have a very low open spaceratio as compared with the average of 48 percent. The various dominant and uniquecharacteristics are translated into a succinctdescription of the six block types and fourstreet types (Table 2) and are presentedtogether with a sample of the ‘archetype’blocks and streets (Figure 6), where we define‘archetype’ as the instance in each type that isclosest to the centre of its cluster.

Results

The results of applying the proposed method


to the test case are encouraging because it hasbeen possible to classify the blocks and streetsin a meaningful way that identifies the twoneighbourhoods. By statistically correlatingthe instances of the types to their neighbour-hood, Expo 98 or Moscavide, the degree towhich the types are characteristic of aneighbourhood can be observed. We find thatsome types clearly correspond to one of the

areas, while a few types have an even share ofinstances in both areas; for example, ‘Blocktype 3’ and ‘Street type 3’ (Table 3). Theoverall coefficient of determination (R2)between clusters and neighbourhoods is 0.67for the block clusters and 0.58 for the streetclusters, where a value of 1 would correspondto complete identity between the two variables.

Figure 2. Blocks distinguished according to area (TA) and floor area ratio (FAR). Fromindividual morphological characteristics it is not possible to distinguish the different

neighbourhoods or identify prevalent types.

Figure 3. Scree plot of blocks and streets clusters. The circle indicates the selected knumber of clusters.


Table 2. Description of block and street types based on their dominant and unique characteristics______________________________________________________________________

Type Description______________________________________________________________________

Block 1 Closed block, medium density with private courtyard onlyBlock 2 High density, compactness and pressure on open spaceBlock 3 Low density with private open spaceBlock 4 Open block of medium density with privileged public spaceBlock 5 Open public space with no built-up areaBlock 6 Large, low density block with equipment and associated public space

Street 1 Very low or no continuity and movement flowStreet 2 High connectivity and continuity streetsStreet 3 Low continuity streetsStreet 4 Long streets with wide pavements and high average of tall buildings______________________________________________________________________

Discussion

Using the method described in this paper it hasbeen possible to identify a series of differentblock and street types that correspond to twodifferent neighbourhoods, creating a context-

sensitive sample of types. This method is thussuitable to support the identification, under-standing and description of types of emergingurban form types that do not fall into standardclassifications either because they are in newurban areas, or in informal settlements, or in

Figure 4. The six block clusters and the four street clusters. The different types belongto the different neighbourhoods and only a few types occur on both sites – these are

more recent blocks in the Moscavide neighbourhood that do not follow its traditionalurban form, and street types that are more universal such as the cul-de-sac.


Table 3. Percentage of block and street instances from the two neigh-bourhoods present in each type___________________________________________________________

Type Total instances Expo 98 (%) Moscavide (%)___________________________________________________________

Block 1 45 0.00 100.00Block 2 16 93.75 6.25Block 3 17 52.94 47.06Block 4 22 90.91 9.09Block 5 2 50.00 50.00Block 6 2 100.00 0.00

Street 1 14 28.57 71.43Street 2 96 3.13 96.88Street 3 44 63.64 36.36Street 4 66 95.45 4.55____________________________________________________________

Figure 5. Sample profile charts for streets of types 1 and 2. These charts show thecharacteristics that differentiate between types (box A). The traits of a type that is

dominant (box B) or unique (box C) are also identified.

Figure 6. Plan of the archetypes of block types 1 and 4 and street types 1 and 2, asdescribed in Table 2.


cultures and geographical locations that have not been studied before.

The method is systematic and can supportlarger studies through consistent application ofthe procedures to wider areas or different sites.It facilitates the process of grasping thecomplex relations of attributes. However, it isimportant to recognize that skill is required inthe selection of attributes and in the inter-pretation of the results.

The method can support site-specific andproject-specific sets of urban form attributes.The set of attributes used in the test case is byno means optimal or finite and can becustomized according to the requirements ofthe specific research or urban developmentproject. For example, the temporal dimension,frequently used in the form of building age orhistorical period, was explicitly excluded fromthe test set so that it could be used forvalidation of the results, but in other cases itcan be included as a classification attribute.

However, it is necessary to consider thepossibility of defining a basic set of attributesthat can be consistently applied by differentpeople across various projects and locationsirrespective of their local significance. This isa requirement for larger comparative morph-ological studies and for the identification ofmore universal types (Marshall and Gong,2009). Since the proposed method can operatewith a very large set of attributes, it wouldsupport an integrated approach using attributesfrom different pieces of urban morphologicalresearch. One of the outputs of quantitativetypological profiling is the degree of signifi-cance of each attribute within the local sample.This contributes to the contextual nature of themethod despite having started with a moregeneral set of attributes.

Finally, the quantitative nature of the outputlends itself to further integration with othersystematic procedures related to research andanalysis or to planning and design of urbanareas. An example of the first case is thedefinition of ‘urban structural units’ (Haggagand Ayad, 2001; Osmond 2010) or other types

of neighbourhoods that should becharacterized by typological homogeneity.Once a local set of types has been identifiedusing the proposed method, the boundaries ofthose analytical areas can be more easilydefined, either visually or using computationalmethods such as spatial clustering.

With regard to the second case, integrationwith planning and design practice, this methodcan be applied when the objective is to use acontext-sensitive typological approach basedon precedents: on the one hand, supporting thecreation or expansion of design patternlibraries; on the other hand, providingnumerical constraints in the form of parametervalues that produce varied solutions that fit theexisting context or the specific urbanprogramme.

Parametric rule-based design structures

The results of the proposed method can beused to extract the characteristics of local typesto build parametric rule-based designstructures. Such a design approach is beingdeveloped as part of the City Inductionresearch project (Duarte et al., 2011), in whichdesign patterns are codified into a smallparametric shape grammars (Stiny and Gips,1972) and stored in a design pattern library(Alexander et al., 1977). Each patterngrammar encodes information related to thequantitative descriptions of the correspondingtype and its rules and parameters can bemanipulated for applying the same type in anew design. As the quantitative descriptionspecifies ranges for the attribute values, thereuse of types in design seems a way ofguaranteeing context-sensitive solutions whilemaintaining reasonable design flexibility.

The use of a pattern-based system forgenerating urban designs has been shown byBeirão et al. (2011). They propose twodifferent design patterns based on shapegrammars; one for generating street grids andanother for defining urban blocks with varyingsizes. These patterns are applied in sequenceso that, after the grid is generated, twodifferent types of blocks are distributed


according to their location on the grid. Theseblocks adapt to the size of the grid cellsaccording to the parametric shape grammarrules used to encode them. The methodproposed in this paper could be used to refineand extend the design pattern library byextracting accurate information regarding thecontext in which the blocks types can beapplied, and by identifying other types fromexisting situations. This would help togenerate more adequate and more complexdesigns.

Conclusions and further work

In this paper a method has been described thatsupports the understanding of urban areasthrough urban typomorphological analysis andresults in the description and prescription ofurban types. The proposed method has thefollowing characteristics that are important infacilitating the application of urban typo-morphology to planning practice:

1. Context-sensitive to geography and culture;2. Multi-dimensional, considering an array of

urban form characteristics;3. Systematic, for replication in comparative

studies of different urban areas and/or to becarried out by non-expert teams underexpert supervision;

4. Exploratory, offering a ‘blind’ discovery oftypes independent of pre-existingclassifications or taxonomies of types;

5. Quantitative, amenable to translation intoparameters and rules.

Ultimately, it offers a method of creatingurban form typologies derived from the localcharacteristics of a place and tailored to theobjectives of a given project.

However, a consistent framework of urbanform characteristics is needed to make the linkbetween the generated types and urbanenvironmental quality (Marshall and Gong,2009; Osmond 2010). Further research isrequired to define this framework with a morecomplete set of attributes related to socio-economic characteristics, such as population

demographics and land use, or to thebuilding’s surface characteristics, such asentrance types and façade transparency. Thesetypes of indicators may allow us to identifyrelations between socio-economic data andmorphological characteristics of the urbanfabric.

Testing the proposed method in urbandesign seems to offer a promising researchavenue and can be first introduced in urbandesign education within the context of designstudios. Considering the difficulties ofstudents during the pre-design phases of theurban design process, the proposed methodcould complement the urban design methoddescribed by Beirão and Duarte (2009), as itprovides an enhancement of the analyticalphase and complements the synthesis phase byintroducing a systematic way of developingdesign patterns.

Acknowledgements

We should like to thank the reviewers for theirinvaluable comments and suggestions. Thisresearch originates in the ‘City Induction’ projectsupported by Fundação para a Ciência e Tecnologia(FCT), Portugal, hosted by ICIST at the TechnicalUniversity of Lisbon (PTDC/AUR/64384/2006)and co-ordinated by Professor José Pinto Duarte. J.Gil is funded by FCT with grant SFRH/BD/46709/2008. N. Montenegro is funded by FCTwith grant SFRH/BD/45520/2008. J.N. Beirão isfunded by FCT with grant SFRH/BD/39034/2007.The spatial network of the study area and itssurroundings was taken from the ‘axial map’ of theLisbon metropolitan region by João Pinelo,University College London.

References

Alexander, C., Ishikawa, S. and Silverstein, M.(1977) A pattern language: towns, buildings,construction (Oxford University Press, NewYork).

Alexander, D., Lannon, S. and Linovski, O. (2009)‘The identification and analysis of regionalbuilding stock characteristics using map baseddata’, unpublished paper presented to theEleventh International IBPSA Conference,Glasgow, UK, July.


Beirão, J. N. and Duarte, J. P. (2009) ‘Urban designwith patterns and shape rules: devising amethodology to create solutions’, in Stolk, E.and Brommelstroet, M. (eds) Model town: usingurban simulation in new town planning(Uitgeverij SUN, Amsterdam) 148-65.

Beirão, J. N., Duarte, J. P. and Stouffs, R. (2011)‘Creating specific grammars with genericgrammars: towards flexible urban design’,Nexus Network Journal 13 (1) 73-111.

Berghauser-Pont, M. and Haupt, P. (2004)Spacemate: the spatial logic of urban density(Delft University Press Science, Delft).

Berghauser-Pont, M. and Haupt, P. (2010) Spacematrix: space, density and urban form (NAiPublishers, Rotterdam).

Conzen, M. P. (2010) ‘The elusive commondenominator in understanding urban form’,Urban Morphology 14, 55-8.

Crucitti, P., Latora, V. and Porta, S. (2006)‘Centrality in networks of urban streets’, inChaos 16 (015113), 1-9.

Duarte, J. P., Beirão, J. N., Montenegro, N. andGil, J. (2011) ‘City induction: formulating,generating, and evaluating urban plans’, inMüller Arisona, S., Wonka, P., Aschwanden,G.,and Halatsch, J. (eds) Digital urbanmodelling and simulation Communications inComputer and Information Science Series(Springer, Berlin).

Fayyad, U. M., Piatetsky-Shapiro, G. and Smyth, P.(1996) ‘From data mining to knowledgediscovery: an overview’, in Fayyad, U. M.,Piatetsky-Shapiro, G., Smyth, P. and Uthuru-samy, R. (eds) Advances in knowledge discoveryand data mining (American Association forArtificial Intelligence, Menlo Park, CA) 1-34.

Figueiredo, L. and Amorim, L. (2005) ‘Continuitylines in the axial system’, in Proceedings of the5th International Space Syntax Symposium (TUDelft, The Netherlands).

Figueiredo, L. and Amorim, L. (2007) ‘Decodingthe urban grid: or why cities are neither treesnor perfect grids’, in Proceedings of the 6thInternational Space Syntax Symposium(Istanbul Technical University, Turkey).

Grant, J. (2001) ‘The dark side of the grid: powerand urban design’, Planning Perspectives 16,219-41.

Habraken, J. (1988) ‘Type as a social agreement’,unpublished paper presented to the Third AsianCongress of Architects, Seoul, November.

Haggag, M. A. and Ayad, H. M. (2002) ‘The urbanstructural units method: a basis for evaluatingenvironmental prospects for sustainable

development’, Urban Design International 7,97-108.

Hall, A. C. (2008) ‘Bridging the gap: applyingurban morphology to successful planningpractice’, Urban Morphology 12, 54-7.

Hanna, S. (2007) ‘Automated representation ofstyle by feature space archetypes: distin-guishing spatial styles from generative rule’,International Journal of ArchitecturalComputing 5, 2-23.

Hillier, B. (2007) Space is the machine: aconfigurational theory of architecture (SpaceSyntax, London).

Kelbaugh, D. (1996) ‘Typology - an architecture oflimits’, Architectural Theory Review 1 (2), 33-52.

Laskari, A., Hanna, S. and Derix, C. (2008) ‘Urbanidentity through quantifiable spatial attributes:coherence and dispersion of local identitythrough the automated comparative analysis ofbuilding block plans’, in Gero, J. S. and Goel,A. K. (eds) Design computing and cognition '08:Proceedings of the Third InternationalConference on Design Computing andCognition (Springer, Dordrecht) 615-34.

Lee, C., Moudon, A. V. and Courbois, J.-Y. P.(2006) ‘Built environment and behavior: spatialsampling using parcel data’, Annals ofEpidemiology 16 (5) 387-94.

Lee, C. C. and Jacoby, S. (2011) ‘Typologicalurbanism and the idea of the city’, ArchitecturalDesign 81, 14-23.

Lo, C. P. (2007) ‘The application of geospatialtechnology to urban morphological research’,Urban Morphology 11, 81-90.

Maller, A. (1998) ‘Emerging urban form types in acity of the American Middle West’, Journal ofUrban Design 3 (2) 137-50.

Marshall, S. (2005) Streets and patterns: thestructure of urban geometry (Routledge,London).

Marshall, S. and Gong, Y. (2009) SOLUTIONS:urban pattern specification WP4 DeliverableReport (Bartlett School of Planning, UniversityCollege, London).

Moudon, A. V. (1994) ‘Getting to know the builtlandscape: typomorphology’ in Franck, A. (ed.)Ordering space: types in architecture anddesign (Van Nostrand Reinhold, New York)289-314.

Moudon, A. V. (1997) ‘Urban morphology as anemerging interdisciplinary field’, UrbanMorphology 1, 3-10.

Osmond, P. (2010) ‘The urban structural unit:towards a descriptive framework to support


urban analysis and planning’, UrbanMorphology 14, 5-20.

Parish, Y. I. H. and Müller, P. (2001) ‘Proceduralmodeling of cities’, in Proceedings of the 28thAnnual Conference on Computer Graphics andInteractive Techniques (ACM Press, New York)301-08.

Peponis, J., Allen, D., Haynie, D., Scoppa, M. andZhang, Z. (2007) ‘Measuring the configurationof street networks’, in Proceedings of the 6thInternational Space Syntax Symposium (IstanbulTechnical University, Turkey) 002:1-002:16.

Rapoport, A. (1990) History and precedent inenvironmental design (Plenum Press, NewYork).

Reffat, R. (2008) ‘Investigating patterns ofcontemporary architecture using data miningtechniques’, in Proceedings of the 26thConference on Education of Computer AidedArchitectural Design in Europe (Higher Instituteof Architectural Sciences, Antwerpen) 601-08.

Samuels, I. (1999) ‘A typomorphological approachto design: the plan for St Gervais’, UrbanDesign International 4 (3) 129-41.

Samuels, I. (2008) ‘Typomorphology and urbandesign practice’, Urban Morphology 12, 58-62.

Samuels, I. and Pattacini, L. (1997) ‘From descrip-tion to prescription: reflections on the use of amorphological approach in design guidance’,Urban Design International 2 (2) 81-91.

Shane, D. G. (2011) ‘Transcending type: designingfor urban complexity’, Architectural Design 81,128-34.

Stiny, G. and Gips, J. (1972) ‘Shape grammars andthe generative specification of painting andsculpture’, Information Processing 71, 1460-65.

Thomas, I., Frankhauser, P., Frenay, B. andVerleysen, M. (2010) ‘Clustering patterns ofurban built-up areas with curves of fractalscaling behaviour’, Environment and Planning

B: Planning and Design 37, 942–54.Trache, H. (2001) ‘Promoting urban design in

development plans: typo-morphologicalapproaches in Montreuil, France’, Urban DesignInternational 6 (3) 157-72.

Urhahn, G. and Bobic, M. (1994) A pattern image:a typological tool for quality in urban planning(Thoth, Bussum).

Wineman, J. D., Marans, R. W., Schulz, A. J., vander Westhuizen, D. L., Grant-Pierson, S. andMax, P. (2009) ‘Contributions of accessibilityand visibility characteristics to neighborhoodtypologies and their predictions of physicalactivity and health’, in Proceedings of the 7thInternational Space Syntax Symposium(Stockholm, Sweden) 131:1-131:10.

Witten, I. H. and Frank, E. (2005) Data mining:practical machine learning tools and techniques(Morgan Kaufmann, San Francisco) 2nd edn.

Documents

Discovery of Urban Typologies