Local urban form measures related to land-use and development period: a case-study for Halifax, Nova Scotia.
Article Type: Case study
Subject: Public spaces (Design and construction)
Geographic information systems (Usage)
Land use (Research)
Land use (Nova Scotia)
City planning (Research)
Authors: Millward, Hugh
Xue, Guang
Pub Date: 12/22/2007
Publication: Name: Canadian Journal of Urban Research Publisher: Institute of Urban Studies Audience: Academic Format: Magazine/Journal Subject: Social sciences Copyright: COPYRIGHT 2007 Institute of Urban Studies ISSN: 1188-3774
Issue: Date: Winter, 2007 Source Volume: 16 Source Issue: 2
Topic: Event Code: 310 Science & research; 290 Public affairs Computer Subject: Geographic information system
Product: Product Code: 9107130 Urban Planning Assistance NAICS Code: 92512 Administration of Urban Planning and Community and Rural Development SIC Code: 7372 Prepackaged software
Geographic: Geographic Scope: Nova Scotia Geographic Name: Halifax, Nova Scotia Geographic Code: 1CNOV Nova Scotia
Accession Number: 179315115
Full Text: Abstract

This case-study employs micro-level measures of urban morphology to examine changing patterns and densities of development in the urban and suburban portions of metropolitan Halifax. It focuses on relationships between land use and urban form, and on how those relationships have changed through time, particularly in response to transportation technology and planning policies.

Twelve morphological indices are employed to measure aspects of the density, coverage, spacing, and pattern of buildings, lots, and streets within small (200m radius) circles, in seven districts of varying character and age. The results demonstrate strong contrasts between inner-city and suburban areas, and empirically verify historical trends towards lower densities and curvilinear street layouts. They also show significant differences in morphology related to land-use types. The most atypical or distinctive forms occur in industrial parks and residential "estate" subdivisions.

Keywords: urban form, morphology, density, pattern, streets

Resume

Ce cas-etudiez utilise des mesures de micro-niveau de morphologie urbaine d'examiner les caracteres changeants et les densites du developpement dans les parties urbaines et suburbaines de Halifax metropolitain. Il se concentre sur des rapports entre l'utilisation de la terre et la forme urbaine, et sur la facon dont ces rapports ont change par le temps, en particulier en reponse a la technologie de transport, et aux politiques de planification.

Douze index morphologiques sont utilises pour mesurer des aspects de la densite, de l'assurance, de l'espacement, et du module des batiments, des sorts, et des rues dans de petits cercles (de rayon de 200m), dans sept zones de caractere et d'age variables. Les resultats demontrent des contrastes forts entre l'interieurville et les secteurs suburbains, et verifient empiriquement des tendances historiques vers de plus faibles densites et des dispositions curvilignes de rue. Ils montrent egalement des differences significatives dans la morphologie liee aux types d'utilisation du territoire. Les formes les plus atypiques ou les plus distinctives se produisent en parcs industriels et subdivisions residentielles de grand sort.

Mots cles : forme urbaine, morphologie, density, dessin, rues

Local Urban Form: Significance and Measures

This study employs micro-level measures of local form to investigate relationships to both land use and time-period of development, in the Canadian context. We are particularly concerned to identify inter-relationships between form measures, and to employ principal components analysis (PCA) to eliminate redundancy. Localized urban form is important in its own right, since it conditions social activity and human health (Frank and Engelke 2001, Handy et al 2002, Ewing et al. 2003a), and provides the framework for transportation and land use (Boarnet and Crane 2001, Krizek 2003a). It is also important as a physical and long-lived expression of morphogenetic forces acting on urban development.

For this study, "local urban form" refers to the property cadastre and the related built fabric, as constituted by the three elements of streets, lots, and buildings (Conzen 1960, 4-5). Measures which focus on morphological properties of local urban form can be divided into the four groups of building density, building pattern, road density, and road layout design. Analysis based on such measures can reveal past trends in urban development, and contribute to predictions and planning for future development. Empirical measures of urban form can also effectively capture "on the ground" development effects of planning policies, such as industrial parks, New Urbanism design, or transit-oriented development.

In the pedestrian and streetcar eras (Adams 1970), land in North American cities was typically intensively used, with high building densities and lot coverage, and high intermixture of land uses. With the advent of widespread automobile ownership, however, cities have increased their spatial extent greatly (Johnson 2001), and have become less dense and locally more homogenous in land use and urban form (Helm 2001). Although these changes have been much discussed, there have been few empirical studies of their objective impacts on urban form. A pioneering study by Borchert (1961) examined gross road densities and road-junction densities, while Johnston (1968) measured street curvature and non-90-degree junctions. In an early statistical assessment of convergence and divergence in urban forms, Millward (1975) used gross and net road density, road junction frequency, road connectivity, frequency of non-90-degree junctions, and road curvature, all measured for 500 m square quadrats sampled in 10 Canadian and 10 British cities. He reported that cities had become increasingly similar in their urban physical form (displaying "morphological homogenization") owing to shared innovations in transportation and site design.

More recently, geographic information systems (GIS) have enabled and encouraged a renewal of interest in morphological analysis of development patterns, particularly by planners. For example, Galster et al. (2001) developed a complex and multi-faceted index to characterize sprawl in eight dimensions, while Ewing et al. (2003b) employed twenty-two variables combined into four sprawl factors using principal component analysis. However, these and similar examinations used measures based on very large units of analysis, and are thus probably too coarse to guide planning or policy decision-making at the municipal level (Knaap and Hopkins 2001).

Several researchers have applied GIS-derived morphological measures as performance indicators related to the planning principles of Smart Growth (Duncan and Nelson 1995, Daniels 1999), and architectural principles of New Urbanism (Katz 1993, Dutton 2000). Morphological methods can directly measure aspects of urban physical form, and are thus highly useful for research on the localized impacts of recent planning and design strategies (Talen 2002). For example, Cervero and Kockelman (1997) considered a large number of neighborhood variables, including proportion of blocks with sidewalk, block length, number of intersections, and retail store availability, to characterize walkable versus auto-dependent urban forms. Bagley et al. (2002) presented a method to assess neighborhood types using several subjective and objective variables derived from New Urbanism principles, while Burton (2002) developed a large set of indicators based on population density, built form density, and mix of uses, and used them to measure urban compactness in an investigation of sustainability. Handy and Clifton (2000) identified factors that contribute to accessibility at the neighborhood level, while Krizek (2003b) used housing density, neighborhood retail employment (representing land use mix), and block size to compose an index of neighborhood accessibility.

Weston (2002) compared "ideal" New Urbanist forms to conventional post-1945 suburbs using several measures of form and land use for 1,000 m by 1,000 m quadrats. The measures included land use dissimilarity and dispersion indices (measuring the variety and spatial clustering of land uses), measures of street density and connectivity, measures of the ratio of single to multiple housing units, and a measure of open space. Weston claimed that his results could help planners retrofit existing neighborhoods to more closely adhere to New Urbanism ideals.

Song and Knaap (2004) employed INDEX software (Allen 2001) to calculate five groups of measures at the neighborhood level (defined by 'block-groups', about one third the size of census tracts). The variable groups are street design (e.g. street connectivity, block size), density (e.g. lot size, home floor space), land-use mix (e.g. ratio of non-residential to residential land), accessibility (e.g. distances to the nearest store or bus-stop) and pedestrian access (e.g. percent of homes within a quarter-mile of a bus-stop or store). Applying these measures to neighborhoods in Washington county (suburban Portland, Oregon), Song and Knaap found similar trends in urban form since the 1940s: neighborhoods are becoming better connected internally, dwelling unit density has increased, single family homes are becoming larger though lots are smaller, and external connectivity is decreasing or not improving. Though highly useful, their study is somewhat problematical because block-groups vary in shape and size, and because their study area is purely suburban, with no pre-1940 neighborhoods.

In a related paper, Song and Knaap (2005) computed similar measures for the neighborhood (defined by a half-mile buffer) surrounding each home in Orange County, Florida. They confirmed many of the trends reported above, but interestingly showed that lot sizes rose to about 1970, after which land scarcity and planning policies caused decreases in lot size. Land use mix, however, declined over the whole post- 1940 period.

This brief literature review shows that, while morphological analysis is in vogue among planners to evaluate development characteristics and performance, only a modest number of studies have employed micro-level measures and disaggregated data, and very few in Canadian cities. Few studies have related form to land use types, or evaluated changes in urban form over time. Also, there is little discussion evaluating the relationships between measures of urban form. These shortcomings will be addressed in the present study. While the focus will be on a single medium-sized metropolis, the results are likely to be broadly indicative of spatial patterning and temporal trends in Canada.

Data and Methods

Because of its medium size and relatively long history of urban development, Halifax (Nova Scotia) is considered highly suitable for analysis of morphological trends. In addition, data for this study were freely available because of an existing collaboration between the authors and the regional planning unit of Halifax Regional Municipality (HRM). The study area is 25 km east-west and 20 km north-south, and contains the urbanized area and inner urban fringe.

Initially, high-resolution satellite imagery was considered for this research. It was rejected owing to the difficulty of employing raster data for automated recognition and calculation of form elements. Primary data sources were thus:

1) 1:1000 and 1:2000 digital property maps by LRIS (Land Registration Information System), providing detailed information regarding buildings and property boundaries. The lot information was current to 2005, but building footprints are based on air photographs to c. 1995 only.

2) Road and land use datasets derived from CanMap 5.0 by DMTI Spatial, with data surveyed approximately in 1997.

3) Halifax Regional Municipality zoning maps, current for 2005.

4) A time-series of topographic and street maps, at various scales. These were used to reconstruct the development sequence, and were augmented by reference to Millward (1981).

Seven sampling districts were selected throughout the urbanized and semi-urbanized areas, to exemplify a range of development periods and styles. Halifax peninsula and Dartmouth districts are older mixed-use inner-city areas developed from 1749 to approximately 1950, while Mainland North mostly comprises early suburban development (1940 to 1970), though with some recent additions. Sackville and Cole Harbour districts are large planned communities built in the period 1965 to 1980, and Kingswood district represents an exurban area of large unserviced lots, developed after 1980. Finally, the Burnside district is a major industrial park developed in phases since the early 1960s. The district areas average 10 sq. km.

Sampling points were located evenly within the districts, on a grid lattice with a 500 m interval. Circular sampling areas were created around these points, circles being preferred as the most compact shape possible. Circle radii of 100 and 200 meters (areas of 3.1 and 12.6 ha) were used, though only results for the 200 m set are reported here. Only those 219 circles which were at least 30% developed were used in the analysis. Each circle was classified by predominant time-period of development and predominant land use. The time-periods were based on generic models of urban development by Adams (1970), Borchert (1991) and Hartshorn (1992), as follows: 1) pre-1900 (walking/horse-car era), 2) 1900 to 1925 (electric streetcar/bus), 3) 1925 to 1950 (motorbus & early automobile), 4) 1950 to 1970 (early freeway), 5) 1970 to the present (late freeway). The six land use categories used are commercial, government/institutional, industrial, single-family residential (SFR), multiple-family residential (MFR), and mixed (none of the previous predominating).

All three morphologic elements--streets, lots, and buildings--were assessed, using a total of 19 measures for each of the 219 sampling circles. Those pairs of variables with Pearson correlations above [+ or -] 0.85 were identified as having redundancy. In these cases the cruder or less meaningful of the pair was dropped from further analysis: those variables discarded were gross building density, gross coverage ratio, mean building size, mean lot size, gross road density, gross junction density, and road connectivity. The remaining 12 variables form the focus of this paper, and can be divided into the four categories of building density, building pattern, road density, and road pattern. They are defined as follows:

1. Net building density (NetDen): number of buildings divided by sum area (ha) of lots containing buildings.

2. Net building coverage ratio (NetCR): sum of areas of buildings divided by sum area of lots containing buildings, expressed as %

3. Mean Proximity (Prox): mean nearest-neighbour distance between buildings (m)

4. Median Building Size (MedBS) ([m.sup.2])

5. Building Size Coefficient of Variation (BScov) (%)

6. Mean Perimeter-Area Ratio (MPAR): a building shape measure (see Moudon 1986)

7. Mean Shape Index (MSI): a non-scale-specific measure of building shape (see Barr et al. 2004)

8. Median lot size (MedLS) ([m.sup.2])

9. Lot size Coefficient of Variation (LScov) (%)

10. Net road density (NetRD): total street length (m) divided by sum area (ha) of lots containing buildings.

11. Net Junction Density (NetJD): number of road junctions divided by sum area (ha) of lots containing buildings.

12. Road Junction Frequency (JuncF): number of junctions divided by total street length (km)

Several of the measures were calculated directly from shapefiles in ArcInfo 8.3, but others required specialized software. An extension for ArcView 3.2 called Patch Analyst 3.1 (Rempel 2007) was used for the FRAGSTATS shape measures (MPAR and MSI), while the ArcGIS extension V_LATE 1.0 (Vector-based Landscape Analysis Tools) was employed for mean proximity. To measure street features (polyline shapefiles), a VBA script was created by Xue to "automatically" recognize road segments, junctions, and their X,Y coordinates within any given sampling circle. Full details of all geoprocessing are available in Xue (2005).

Form Measures Related to Time and Land-use

Urban form measures were aggregated by district, period of development, and predominant land use. The main focus of interest, however, was on cross-tabulation with development period (table 1) and land use (table 2). Medians are felt to be more useful than means as an indicator of typical values, since several variables (PROX, MedBS, MedLS, and LScov) have strong positive skew, with means well above their medians (see two columns on right of table 1). Such skew reflects the fact that both industrial areas and exurban "estate"-lot areas (Burnside and Kingswood, specifically) show development patterns quite different from the norm. Because of the skewed distributions, a non-parametric paired difference-of-ranks test (Mann-Whitney) was used on all variables to identify significant differences between pairs of districts, pairs of development periods, and land use pairs.

From table 1 and figure 1, we see that key measures of building and street density remained steadily high (and even significantly increasing) until 1950, showing that Halifax remained strongly oriented to pedestrian, street-car, and bus modes of transport to that date. It was not until 1950 that net building density (NetDen) declined significantly (at p<0.05), and 1970 before significant decline occurred in net road density (NetRD). In 1970-95 NetRD declined to only 69% of its 1950-70 value, while NetDen declined to only 20%. The different rates of decline are explained by an increase of 288% in median lot size (MedLS), from 621 to 2,415 [m.sup.2]. Median building footprint (MedBS) increased much less dramatically between the last two time-periods, but was still up from 122 to 152 [m.sup.2], or +25% (see figure 1). Building proximity (PROX) remained fairly constant in the 5-7m range until 1970, but increased significantly to 16m in the final period. Net lot coverage (NetCR) shows a steady decline through all time-periods, from 26% of lot area pre-1900 to only 13% recently. These declines, however, were only significant after 1950.

Both shape indices show that on average buildings became more compact through time, but since the MPAR ratio is sensitive to building size (larger buildings return lower values) it exaggerates the trend. MSI is not affected by building size: its range is from 1.0 (circular, or most compact) to infinity. Median MSI values fell monotonically to 1970, though not greatly, and increased slightly in the final period.

The three road measures tell a clear story. Both net road density (NetRD) and net junction density (NetJD) show overall declines through time, with the exception of 1900-25. The declines are most marked in the post-1950 period, with NetRD falling 40% from its 1925-50 level, and NetJD falling 66% (see figure 1). The greater decline for junction density indicates the shift from gridiron to curvilinear layouts, and this is confirmed by the large decline in junction frequency (JuncF), which fell 47% in the same period (curvilinear patterns have fewer junctions per km of road).

[FIGURE 1 OMITTED]

Turning to table 2, we see that both density and pattern measures vary greatly by land-use. However, caution must be exercised when evaluating these differences, since the three land-use categories of Multiple-family residential (MFR), Commercial, and Government have small sample sizes. Fortunately, these three are remarkably similar on key measures of building density (NetDen, NetCR, and PROX), and on 36 significance tests (12 variables by three land-use pairs) show only one significant difference at p<0.05. They can safely be grouped for purposes of analysis, and labeled as CGM.

Single-family residential (SFR) areas have the highest building densities (NetDen), followed by Mixed and CGM, with Industrial being very low. But SFR areas (and Mixed areas too, which tend to be in the older inner-city) also have very small building footprints (MedBS), whereas CGM buildings are larger, and Industrial buildings are very large. Lot coverage ratios (NetCR) are therefore fairly similar across all land-use categories, ranging from 16% for SFR to 26% for Commercial. Building proximities (PROX) are also similar, in the 7-10 m range, except for 26 m in Industrial areas. In keeping with their large buildings, Industrial lands show very large lot sizes (MedLS).

Building shape measures are again somewhat problematical. The low MPAR value for Industry is largely reflective of large buildings, and to a lesser extent this is true for the CGM group. The mean shape index (MSI) values show that SFR buildings are most compact, and MFR buildings are least compact.

The three street indices also highlight Industry as the most exceptional land-use. It has significantly lower levels of road and junction density, and junction frequency, than does SFR, clearly indicating the specialized layouts employed in large industrial parks such as Burnside. Both Mixed land-use and CGM have values similar to SFR, since these uses employ 'standard' street layouts. However, Commercial and Government have the high JuncF values characteristic of older grid patterns, while MFR has a low JuncF typical of modern suburban designs (extensive areas of large apartment buildings did not appear in Halifax until c. 1970).

On a district basis, the most exceptional area is the large-lot residential exurb of Kingswood, which is significantly different to all other districts on all 12 variables. Only slightly less abnormal is Burnside industrial park, which lacks significant difference only on BScov. Of conventional development areas, Halifax peninsula (developed almost entirely pre-1950) is significantly different to the newer and more exclusively residential suburbs of Mainland North, Cole Harbour, and Sackville on 10 measures, whereas Dartmouth varies from the latter three areas on only two variables. Because it contains development from most time-periods, Dartmouth's median scores are closest to the overall medians.

Relationships between Form Measures

In this section the disaggregated data for all 219 circles are employed to investigate systemic relationships between the 12 form measures. Table 3 presents Pearson linear correlations between the 12 selected variables, many of which appear highly significant (though the strict conditions for z- or t-tests are seldom present). By focusing on correlation values exceeding [+ or -] 0.50, we see that two variables are particularly intercorrelated with others. These are net building density (NetDen) and net junction density (NetJD), which gauge different aspects of development density, but are themselves correlated at r = 0.55. Of the two building shape measures, the "purer" MSI index, which is not affected by building size, has low correlation with all other variables, which suggests that it provides useful information. Its utility, however, is compromised by the complicated equation for the index, which hinders interpretation. Another variable which seems to stand alone is net coverage ratio (NetCR), and this is much easier to interpret.

Although linear correlations are useful, we should not assume that all 12 variables are linearly or even monotonically related. Because they describe elements of a system, and because they are measured at a variety of dimensionalities, we expect allometric relationships, which follow power laws (Gayon 2000). Such relationships will be linear only when both variables have the same dimensionality, or when a log-log transformation is employed.

Six measures of density or size were selected for detailed investigation of allometric relationships. They are NetDen (Net Building Density), NetCR (Net Building Coverage Ratio), PROX (Mean Proximity), MedBS (Median Building Size), MedLS (Median Lot Size), NetRD (Net Road Density). Table 4 shows linear (untransformed) and power (log-log) correlations between the six variables, for all cases with non-negative original values (n = 192). The table confirms what is also visually evident in scattergrams: all the relationships are better modeled by the power transformations. There are particularly strong correlations for NetDen with PROX and MedLS with PROX, which both seem intuitively understandable.

The power relationships are even stronger when land-use type is held constant, as illustrated in figure 2. Here, both Single-family Residential (SFR) and the CGM group have enhanced r-values, at -0.94 and -0.89 respectively, though Industry shows a poorer fit of only -0.55. More interestingly, the form of the relationship, and particularly the exponent, varies by land-use type. For the use with smallest buildings (SFR), the exponent is near -1.0: that is, a doubling in (untransformed) proximity is associated with a halving in density, since the building areas are negligible at the scale of the 200 m sampling circles (i.e., both variables can be conceptualized as one-dimensional). In contrast, Industry has large buildings, whose areas are a substantial portion of the sampling circles, so that the density variable is conceptually 2-dimensional, and the exponent is close to -0.5: a doubling in (untransformed) proximity is thus related to halving the square root of building density. The relationship for the CGM group is intermediate, with an exponent of -0.86, reflecting the medium size of buildings in this group.

Also noticeable in figure 2a is the clear separation between the main body of the SFR group and a smaller set of SFR circles with very low building densities and high proximities. Almost all this latter group are located in the Kingswood area, where subdivision regulations and zoning codes mandate minimum lot sizes of at least an acre (0.4 ha), to ensure safe operation of on-site water and sewer systems.

Underlying Components of Form

The purposes of using factor analysis in the research are three-fold: 1) to further remove redundant (highly correlated) measures and replace the entire data set with a smaller number of uncorrelated and generalized factors; 2) to discern underlying structural relationships among the original form measures, which represent styles or modes of urban development; and 3) to compare districts, development periods, and land uses using the new orthogonal (uncorrelated) factor scores. Here, the principal component (PC) method was applied with varimax rotation, since this combination explained the most variation in the data, is widely used, and supplies readily-interpretable factors.

[FIGURE 2 OMITTED]

The 12 form and density measures detailed earlier were employed as input variables, with separate runs for original and logarithmically transformed data. Comparing the cumulative percentage of variation accounted for ("explained") by the two data sets, the logarithmically transformed data showed a much higher value of 79%, compared with 66% for the original data. Given the earlier discussion of allometric (system) relationships, this result was expected.

As shown in the rotated component matrix for the log data (table 5), the first two components are almost equally explanatory, with 30.2% and 30.0% of variation respectively.

Component 1 is most highly correlated with net road density (0.81) and net road junction density (0.91), with contributions from net coverage ratio and junction frequency. This component mainly represents characteristics of "street density," but since street and building densities are somewhat inter-correlated, variables related to lot size and building density also load moderately highly on this component, so that it can be viewed more broadly.

Lot size and building density load slightly more highly on component 2 than on component 1, but the strongest loadings by far are for median building size and mean perimeter/area ratio. Recall that MPAR values are highly influenced by building size: for a given building shape, perimeter increases linearly while area increases as a squared value (i.e., there is a power relationship), so that larger buildings have smaller MPAR values. The best label for component 2 is therefore "size/shape of buildings and lots", and high scores on this component indicate areas with large lots and large industrial or institutional buildings.

The third component is most highly correlated with the two coefficients of variation, for building size and lot size, and also with the better (non-scale-specific) measure of building shape (MSI). High scores on this "variation in building/lot size" component identify areas with a mixture of lot and building sizes, and with elongated rather than compact buildings. Such areas are typically in the inner-city, and were developed prior to comprehensive zoning codes (in Halifax, prior to 1950).

Component scores on the first two factors were used as input variables for scatterplots, with cases labeled by district, land use, and time-period. There are three 'clouds' on the graphs (see figures 3 and 4): the main cloud (1) is an elongated ellipse located at the bottom of the graph, while two minor two clouds are situated at the left-middle (2) and right top corner (3) respectively. The suburban districts of Cole Harbour and Sackville predominate on the left of cloud 1, and inner-city areas of Halifax peninsula and Dartmouth predominate on the right. Most points in cloud 2 pertain to Kingswood, while cloud 3 comprises cases in Burnside.

Single-family residential (SFR) occurs mostly in the main cloud, which represents conventional (fully-serviced) developments, but also in cloud 2 (Kingswood), which comprises "estate" lots with on-site sewage disposal. Cases in the latter area were all developed in the 1970-95 period, but the main cloud has cases from the full range of time-periods. There is, however, the suggestion of a progression or shift through time: cases developed 1950-70 tend to lie to the left of earlier development, and those developed 1970-95 definitely lie even further left. This confirms that the traditional gridiron pattern had been fully abandoned by 1970, replaced by loop and modified-Radburn styles.

Mixed land uses occur centrally within the main cloud, scattered among the SFR, and do not appear to be associated with a particular time-period. Industry, by contrast, stands apart in its own cloud (3): nearly all cases here are located in Burnside industrial park, and were built in the 1970-95 period. A few cases of industry (also in Burnside) occur in low-density cloud 3, and are also post-1970. The CGM land-uses (commercial, government, and multiple-family residential) are understandably more scatterered in figures 3 and 4, but in general lie intermediate between the main (SFR/mixed) cloud (2) and the industry cloud (3).

[FIGURE 3 OMITTED]

[FIGURE 4 OMITTED]

Overall, principal component analysis provides a useful method for interpretation of the linkages between morphological styles of development, time of development, land use, and urban location. In particular, the analysis confirmed the importance of junction density and building size in differentiating both between layout styles and the land-use types accommodated within layouts.

Summary and Conclusion

q-his research provides improved methods to quantitatively characterize urban development forms at the micro level. It employs micro-level measures to investigate relationships between local form, predominant land use, and time-period of development, in the Canadian context. We are particularly concerned to identify inter-relationships between form measures, and to employ principal components analysis (PCA) to eliminate redundancy. No previous morphological study has been attempted at such a fine spatial scale, and none has used principal components to interpret micro-level variation in urban form.

The research empirically verifies well-known historical trends, in that land use intensity declines through time, buildings become larger in their footprint area and further apart, and they occupy bigger lots than ever before. Less well-known, but also revealed in this paper, are the ways in which trends in density relate to trends in road/lot layout and land-use separation, both of which have been much influenced by planning practice (both through subdivision design and zoning). Particularly indicative of these layout trends are the street-related measures of road density, junction density, and junction frequency, all of which decreased over time. Some of these results were also noted by Song and Knaap (2004), but their study is restricted to post-1940 suburban residential areas, and uses larger spatial units.

Allometric (power-law) relationships were explored among the various measures, on the basis that site-level design is an integrated system. Power-law relationships were confirmed as being stronger than linear ones, with exponents varying noticeably by land use categories. Principal component analysis using logarithmic data (i.e., based on allometry) was better able to account for data variation than PCA using untransformed data, and suggests that most data variation can be accounted for by three broad factors. These were labeled as street density, size/shape of buildings and lots, and variation in building/lot size. Graphing of component scores showed them to relate clearly and unequivocally to both time-period of development and land use.

Although not reported in this paper, the authors have repeated the entire analysis for sampling circles of 100 m, which yielded remarkably similar results, q-he smaller units have benefits for pinpointing variations in building size and pattern, but are less appropriate for analysis of street characteristics. Further work on the scale-sensitivity of morphological measures would be useful, as would applications of the techniques to other cities. Also useful would be an assessment of how 'ideal' layouts (e.g., Radburn, grid, loop) score on our measures, relative to real-world examples (cf. Weston 2000), and investigation of the implications for the social and environmental goals of New Urbanism and Smart Growth.

Acknowledgements

This research was funded in part by the Social Science and Humanities Research Council of Canada (Standard Research Grant, principal investigator Trudi Bunting).

References

Adams, J. 1970. Residential structure of Midwestern cities. Annals, Association of American Geographers 60: 37-62.

Allen, E. 2001. INDEX: software for community indicators. In Planning supporting systems: integrating geographic information systems, models, and visualization tools. ed. R. Brail and R. Klosterman. Redlands: ESRI Press: 229-61.

Bagley, M., P. Mokhtarian, and R. Kitamura 2002. A methodology for disaggregate, multidimensional measurement of residential neighborhood type. Urban Studies 39(4): 689-704.

Barr, S., M. Barnsley, and A. Steel 2004. On the separability of urban land-use categories in fine spatial scale land-cover data using structural pattern recognition. Environment and Planning B, 31: 397-418.

Boarnet, M. and R. Crane 2001. Travel by design: the influence of urban form on travel. New York, NY: Oxford.

Borchert, J. 1961. The twin cities urbanized areas: past, present, and future. Geographical Review 51: 47-70.

--1991. Futures of American cities. In Our changing cities ed. J. F. Hart. Baltimore: Johns Hopkins University Press: 218-50.

Burton, E. 2002 Measuring urban compactness in UK towns and cities. Environment and Planning B-Planning & Design 29 (2): 219-250.

Cervero, R. and K. Kockelman, 1997. Travel demand and the 3ds: density, diversity, and design. Transport and Environment 2 (3): 199-219.

Conzen, M. 1960. Alnwick, Northumberland." a study in town-plan analysis. London: Philip and Son.

Daniels, T. 1999. When city and country collide: managing growth in the metropolitan fringe. Washington, DC: Island Press.

Duncan, J. and Nelson, A. 1995. Growth management principles and practise. Chicago: APA Planners Press.

Dutton, J., ed. 2000. New American urbanism: re-forming the American metropolis. Milan: Skira.

Ewing, R., T Schmid, R. Killingsworth, A. Zlot, and S. Raudenbush. 2003a. Relationship between urban sprawl and physical activity, obesity, and morbidity. American Journal of Health Promotion 18 (1): 47-57.

Ewing, R., R. Pendall, and D. Chen. 2003b. Measuring sprawl and its transportation impacts. Transportation Research Record 1831: 175-183.

Frank, L. and P. Engelke. 2001. The built environment and human activity patterns: exploring the impacts of urban form on public health. Journal of Planning Literature 16 (2): 202-18.

Galster, G., R. Hanson, M. Ratcliffe, H. Wolman, S. Coleman, and J. Freihage. 2001. Wrestling sprawl to the ground: defining and measuring an elusive concept. Housing Policy Debate 12 (4): 681-717.

Gayon, J. 2000. History of the concept of allometry. American Zoologist 40: 748-758.

Handy, S and K. Clifton. 2000. Evaluating neighborhood accessibility: issues and methods using geographic information systems. College Station TX: Southwest Region University Transportation Center.

Handy, S., M. Boarnet, R. Ewing and R. Killingsworth. 2002. How the built environment affects physical activity: views from urban planning. American Journal of Preventive Medicine 23 (2S): 64-73. Hartshorn, T. 1992. Interpreting the city. New York: John Wiley.

Johnson, M. 2001. Environmental impacts of urban sprawl: a survey of the literature and proposed research agenda. Environment and Planning A 33 (4): 717-35.

Johnston, R. 1968. An outline of the development of Melbourne's street pattern. Australian Geographer 10: 453-65.

Katz, P. 1993. The new urbanism: toward an architecture of community. New York: McGraw-Hill.

Knaap, G. and L. Hopkins. 2001. The inventory approach to urban growth boundaries. Journal of the American Planning Association 67:314-326.

Krizek, K. 2003a. Residential relocation and changes in urban travel: does neighborhood-scale urban form matter? Journal of the American Planning Association 69: 265-281.

--2003b. Operationalizing neighborhood accessibility for land use-travel behavior research and regional modeling. Journal of Planning Education and Research, 22: 270-287.

Millward, H. 1975. Trends in the design and density of urban street layouts. Proceedings, New England and St. Lawrence Valley Geographical Society 5: 10-14.

--1981. The geography of housing in metropolitan Halifax, NS. Atlantic Region Geographical Studies, No.3. Halifax: Saint Mary's University, Dept. of Geography.

Rempel, R. 2007. Patch Analyst 3. http://flash.lakeheadu.ca/-rrempel/ patch/ (accessed October 3, 2007).

Song, Y. and G. Knaap. 2004. Measuring urban form: is Portland winning the war of sprawl? Journal of the American Planning Association 70(2): 210-225.

--2005. The fruits of growth management in the sunshine state: an examination of urban form in Orange County, Florida. Available online in PDF format at Smart Growth Network website http://www.smartgrowth.umd.edu/research/pdf/ KnaapSong_FloridaGrowthManage_022305.pdf (accessed October 3, 2007).

Talen, E. 2002. Help for urban planning: the transect strategy. Journal of Urban Design 7 (3): 293-312.

Weston, L. 2002. A methodology to evaluate neighborhood urban form. Planning Forum 8: 64-77.

Xue, G. 2005. Exploring measures of urban form: a GIS-enabled case study in Halifax, Nova Scotia. Unpublished M.Sc. thesis, Applied Science. Halifax, Saint Mary's University: 123 pp.

Hugh Millward and Guang Xue

Department of Geography

Saint Mary's University
Table 1. Medians for 12 Variables, by Development Period,
200 m Circles (n=219)

Variable     unit      Pre-1900   1900-25   1925-50

n                          24        11        19
NerDen       n/ha        13.3      10.6      16.0
NetCR          %         25.8      20.7      19.2
PROX           m            5         5         6
MedBS      [m.sup.2]      113        96        98
BScov          %          141       157        93
MPAR         ratio       .406      .435      .419
MSI                      1.26      1.24      1.22
MedLS      [m.sup.2]      369       335       460
LScov          %          174       182       103
NetRD        m/ha         212       168       218
NetJD        n/ha        1.08      0.83      0.98
JuncF        n/km         5.0       5.1       4.5

                               Overall   Overall
Variable   1950-70   1970-95   Medians    Means

n             50        115      219       219
NerDen      13.3        2.6      9.0       8.9
NetCR       18.7       13.5     17.9      17.5
PROX           7         16        8        15
MedBS        122        152      128       342
BScov         66         61       81        94
MPAR        .379       .351     .376      .359
MSI         1.19       1.21     1.22      1.23
MedLS        621       2415      629      2901
LScov         82         63       92       124
NetRD        189        131      168       178
NetJD       0.71       0.33     0.58      0.68
JuncF        3.6        2.4      3.5       3.4

Table 2. Medians for 12 Variables, by Land Use Categories,
200 m Circles (n=219)

Variable   SFR *   MFR **   Industrial   Mixed   Commercial

n           132       7         31         24         15
NetDen     12.4     6.9        1.6        8.9        4.1
NetCR        16      21         24         16         26
PROX          7       9         26          8          8
MedBS       122     218        596        115        136
BScov        35      97        100        149        166
MPAR        .39     .32        .21        .39        .33
MSI        1.19    1.36       1.27       1.22       1.27
MedLS       617     585       4388        560        460
LScov        62     175         69        182        163
NetRD       181     148         93        162        184
NetJD       .63     .30        .26        .66        .93
JuncF       3.4     2.3        2.2        4.1        5.0

                                  Overall
Variable   Government   CGM ***   Medians

n               10         32       219
NetDen         4.8        4.6       9.0
NetCR           19         22        18
PROX            10          8         8
MedBS          168        166       128
BScov          182        154        81
MPAR           .33        .33       .38
MSI           1.33       1.31      1.22
MedLS          594        487       629
LScov          160        154        92
NetRD          117        147       168
NetJD          .54        .61       .58
JuncF          4.0        4.2       3.5

* SFR--Single Family Residential

** MFR--Multiple Family Residential

*** CGM--combined Commercial, Government, and MFR

Table 3. Pearson Correlations between 12 Untransformed
Variables, for 200 m Circles (n=219)

         PROX   NetDen   NetCR   MedBS

PROX       1
NetDen   -.69      1
NetCR    -.35     .25      1
MedBS     .13    -.26     .30      1
BScov    -.20    -.08     .36    -.10
MSI       .08    -.35     .30     .17
MPAR     -.51     .65    -.34    -.58
MedLS     .30    -.30    -.10     .42
LScov    -.37     .06     .14    -.15
NetRD    -.38     .58     .23    -.16
NetJD    -.43     .55     .29    -.16
JuncF    -.48     .42     .28    -.16

         Bscov   MSI    MPAR   MedLS

PROX
NetDen
NetCR
MedBS
BScov      1
MSI       .30      1
MPAR      .09     .29     1
MedLS    -.09     .11   -.34     1
LScov     .54     .18    .27   -.22
NetRD    -.02    -.15    .31   -.21
NetJD     .07    -.01    .30   -.22
JuncF     .22     .13    .30   -.23

         Lscov   NetRD   NetJD   JuncF

PROX
NetDen
NetCR
MedBS
BScov
MSI
MPAR
MedLS
LScov      1
NetRD    -.04      1
NetJD     .06     .82      1
JuncF     .25     .43     .78      1

N = 219. Values over 0.13 are significant at p= .05, and
those over 0.18 at p=.01 (2-tailed, based on z-scores).

Table 4. Pearson Correlation Coefficients for Linear and
Power Relationships between Selected Variables, 200 m
Circles (n=192)

                                       2-D variables
1-D
Variables        Model        NetDen   NetCR   MedBS   MedLS

PROX            linear         -.67    -.36     .40     .74

            power (log-log)    -.88    -.60     .52     .88

NetRD           linear          .50     .24    -.17    -.32

            power (log-log)     .59     .41    -.26    -.52

values over 0.14 are significant at the 0.05 level
(2-tailed, based on z-scores)

Table 5. Loadings for Rotated Principal Components,
200 m circles (n = 192)

                                         Component

                  Variable               1 (30.2%)
                                          Street
Name                Label                 density

logNetJ     Net junction density            .91
logNetR       Net road density              .81
logNetC      Net coverage ratio             .76
logjunF      Junction frequency             .67
logMedB     Median building size           -.04
logMPAR   Mean perimeter/area ratio         .06
logNetD     Net building density            .60
logMedL        Median lot size             -.60
logProx    Mean building proximity         -.61
logBScv  Coef. Var. of building size        .15
logLScv    Coeff. Var. of lot size          .07
logMSI        Mean shape index             -.02

                                         Component

                  Variable               2 (30.0%)
                                       Size/shape of
Name                Label              bldgs & lots

logNetJ     Net junction density           -.19
logNetR       Net road density             -.18
logNetC      Net coverage ratio             .16
logjunF      Junction frequency            -.12
logMedB     Median building size            .95
logMPAR   Mean perimeter/area ratio        -.95
logNetD     Net building density           -.73
logMedL        Median lot size              .68
logProx    Mean building proximity          .62
logBScv  Coef. Var. of building size        .05
logLScv    Coeff. Var. of lot size         -.31
logMSI        Mean shape index              .46

                                         Component

                  Variable               3 (18.8%)
                                       Variation in
Name                Label              bldg/lot size

logNetJ     Net junction density            .00
logNetR       Net road density             -.23
logNetC      Net coverage ratio             .39
logjunF      Junction frequency             .26
logMedB     Median building size            .08
logMPAR   Mean perimeter/area ratio        -.02
logNetD     Net building density           -.11
logMedL        Median lot size             -.26
logProx    Mean building proximity         -.22
logBScv  Coef. Var. of building size        .84
logLScv    Coeff. Var. of lot size          .84
logMSI        Mean shape index              .66

Varimax rotation with Kaiser normalizarion. Rotation
converged in 5 iterations.
Gale Copyright: Copyright 2007 Gale, Cengage Learning. All rights reserved.