The changing urban landscape of the continental United States

Landscape and Urban Planning 110 (2013) 74–86

Contents lists available at SciVerse ScienceDirect

Landscape and Urban Planning journal homepage: www.elsevier.com/locate/landurbplan

Research paper

The changing urban landscape of the continental United States Nikhil Kaza ∗ Department of City and Regional Planning, University of North Carolina at Chapel Hill, 110 New East, Campus Box 3140, Chapel Hill, NC 27599-3140, USA

h i g h l i g h t s    

Land cover data from 2001 and 2006 were used to study the patterns of urbanisation in the United States. Landscape metrics are calculated for each county in the US for the urban areas. Overall, urban counties are becoming more contiguous. Some rural counties in South and Mountain West are becoming more fragmented.

a r t i c l e

i n f o

Article history: Received 2 December 2011 Received in revised form 28 August 2012 Accepted 18 October 2012 Available online 27 November 2012 Keywords: Landscape metrics Urban patterns Urbanisation drivers

a b s t r a c t The hallmark of development in the United States has been sprawling urbanisation. However, most of the analyses about urbanisation have focused on specific regions or metropolitan areas. This study provides a comprehensive overview of the changing urban landscape patterns between 2001 and 2006. Using the land cover data from the US Geological Survey for these 2 years, I characterise the landscape metrics for each county in the continental United States. The changes in these metrics are correlated with the drivers of urbanisation including socioeconomic variables. Uneven and heterogeneous patterns of growth in a country this large are not surprising. However, the metrics reveal that while urban counties are becoming less fragmented, some rural counties in the Southern and Western United States are experiencing significant leapfrog development. © 2012 Elsevier B.V. All rights reserved.

1. Introduction Urban form in the United States (US) is characterised by sprawl, fragmentation and leapfrog development(Downs, 1999). Despite many laments about such form, it shows no sign of abatement and is only exacerbated by public policy (Knaap, Song, & Nedovic-Budic, 2007). Most of the indices used to describe these phenomena are based on demographic and economic indicators such as the density of housing and the mix of land uses (Cutsinger & Galster, 2006). However, aforementioned characterisations have not made effective use of widely available land cover data that lends itself to continuous monitoring of urban patterns. This study provides an empirical assessment of urban patterns through landscape metrics and reveals a more nuanced picture of urbanisation pattern in the US. By characterising changes in the urban landscape in each county of the continental United States using landscape metrics, this study finds that patterns and rates of urbanisation differ throughout the country and depend greatly on the degree of rurality. The characterisation of urban patterns through indices is not new, although different types of indices are created and used for

∗ Corresponding author. E-mail address: [email protected] 0169-2046/$ – see front matter © 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.landurbplan.2012.10.015

different purposes (Clifton, Ewing, Knaap, & Song, 2008). Landscape ecologists interested in environmental protection use land cover data to study habitat fragmentation, whereas economists and transportation planners use employment and population data to uncover economic inefficiencies and accessibility patterns. In a comprehensive study of the US, Burchfield, Overman, Puga, and Turner (2006) constructed sprawl indices for metropolitan areas based on land cover data from 1976 and 1992 and argued that these indices have not substantially changed. For metropolitan regions, Ewing, Pendall, and Chen (2003) created an index that included geographic, demographic and economic data which accounted for land use mix, density, street connectivity and centrality variables. However, indices such as the above are only comprehensive in so far as they are usually cross-sectional indices that compare metropolitan areas with respect to one another. In a study for a single metropolitan area of Phoenix, Shrestha, York, Boone, and Zhang (2012) explain the patterned fragmentation of the landscape through institutional factors such as the availability of infrastructure (e.g. water, roads) and constraints on development (e.g. parks, military bases, topography). Two main issues are overlooked in the cross metropolitan comparisons: (1) dynamics of urbanisation and (2) transition of rural areas. Harvey and Clark (1965) long ago argued “the sprawl of the 1950s is frequently the greatly admired compact urban area of

N. Kaza / Landscape and Urban Planning 110 (2013) 74–86

the early 1960s. An important question on sprawl may be, ‘How long is required for compaction?’ as opposed to whether or not compaction occurs at all.” Echoing these sentiments, many urban economists argue that discontinuous development is important in the dynamic land market by allowing land closer to the edge be preserved for higher value uses at a future date (Ohls & Pines, 1975; Peiser, 1989). Restricting attention to metropolitan areas leaves rural areas understudied. Johnson (2001), concluding his review of the literature, identified that older and less dynamic metropolitan areas are rarely given attention. This paper uses indices that, as long as the satellite data are available, lend themselves to a continuous and comprehensive characterisation of urban form. To my knowledge, this is a first comprehensive characterisation of urban landscape in the continental United States. Characterising the urban form itself, however, is not sufficient to advance our understanding of development patterns. Sociopolitical and institutional factors also drive the evolution of urban form. In a pioneering empirical work, using a cross-sectional analysis of US cities, Brueckner and Fansler (1983) found that population has the largest effect on city size with per capita income and agricultural rent having significant but smaller effects. These results were validated in other studies (McGrath, 2005; Song & Zenou, 2006). In their analysis of urbanisation trends in the US, Alig, Kline, and Lichtenstein (2004) identify population density, the metropolitan character of the area and per capita income as key drivers of urbanisation. They also allude to regional differences in the effects of these drivers. These studies are primarily concerned with the extent of urbanisation and the relationship between the drivers of urbanisation rather than the pattern of urbanisation. In this study, I focus not only on the extent but also on the pattern of urbanisation in the US by accounting for socioeconomic and infrastructure variables. This study differs from other studies in a number of ways. It characterises the urban development pattern for the entire United States (except Alaska and Hawaii), ascertains the change in patterns during a short period (5 years) and correlates these changes to socioeconomic and infrastructure variables. There are three main reasons for choosing this time period. First, longterm urban dynamics are different from short-term dynamics and it is useful to study both, even while this study focuses on the latter. Second, consistent, comprehensive and comparable landcover datasets are available for these 2 years for the first time. Similar datasets are expected to be available every 5 years henceforth, and make continuous monitoring possible. Third, the early part of 2000s represented an unprecedented housing boom in the US since 1975 (see Fig. 1); 8.9 million new residential units were constructed during this time and at its peak 2 million of them are constructed in a single year. Most of these houses are built as second homes and speculative investment properties fuelled by subprime mortgages (Wheaton & Nechayev, 2008). Significant numbers of new residential units were added in the sunbelt states of California, Arizona, Florida and to a lesser extent in already existing urban areas in the Northeast and Chicago regions (Fig. 1). The results show that while the US may be experiencing long-term fragmented development, most urban counties are experiencing short-term contiguous development. The study also reveals interesting differences between urban and rural counties by focusing not just on metropolitan regions, but on the whole conterminous US. Furthermore, by correlating demographic and economic variables with landscape patterns, the study finds that differences exist by types of counties. For example, density of highways, while being negatively correlated to urbanisation in urban counties, is associated with fragmentary urban development in rural counties. The paper follows the following structure. First I motivate the landscape metrics and their role in examining urbanisation patterns. I then discuss the levels and patterns of urban development

75

in the US at the turn of the century. I then examine the short-term effects and argue that landscape metrics provide us with a geographically comprehensive, if incomplete, picture of urbanisation. I conclude with caveats of using land cover data for these kinds of analyses as well as future research questions that arise from this study.

2. Landscape metrics Landscape metrics have been extensively used in landscape ecology to examine habitat fragmentation, biodiversity and epidemiology (Haines-Young & Chopping, 1996; Stephens, Koons, Rotella, & Willey, 2004). Stemming from the pioneering work of O’Neill et al. (1988), many such indices and their metrics have evolved over time to suit different disciplines (Gustafson, 1998; Uuemaa, Antrop, Roosaare, Marja, & Mander, 2009). More often than not, these metrics are used to characterise categorical maps rather than continuous fields. Contiguous areas of the same category are considered patches; the metrics are used to describe the shape, composition, and configuration of these patches. Due to wide implementation in various software environments (McGarigal & Marks, 1995; Rempel, 2008), they are now commonly used for ecological analyses. Landscape metrics have been used to study urban morphology by planners, geographers and urban economists (Buyantuyev, Wu, & Gries, 2010; Irwin & Bockstael, 2007; Yeh & Xia, 2001). Seto and Fragkias (2005) used landscape metrics to describe the patterns of change in the 1990s in China’s rapidly growing Pearl River Delta and found many patterns of urban growth, especially during early stages of economic growth. Similarly, Xu et al. (2007) characterised the diffusion-coalescence phases of urban growth whereby cities experience leapfrogged expansion in edge areas that then become contiguous. Herold, Scepan, and Clarke (2002) also used landscape metric signatures to characterise patterns in addition to the amount and location of urban growth. While Torrens and Alberti (2000) may have articulated how landscape metrics can be used to study urban patterns, they have not demonstrated it in an empirical case. As urban form consists of many land uses, these studies use different kinds of land cover categories to study the patterns of urban growth. These include low- and high-density residential, commercial and industrial categories. Since the purpose of the paper is to characterise the urban landscapes of the continental United States, I treat all urban land as a single class. Therefore, many of the finegrained metrics such as contagion and dispersion indices, while possible and useful at a regional or city scale, are not used in this study. Some key landscape metrics are used to characterise the urban area. The total urban area is essentially the sum of the cells that are classified ‘urban’. The number of patches reflects the contiguity of the urbanisation (Fig. 2). A large total urban area reflects a high level of urbanisation, while a low number of patches reflect contiguity of urbanisation and a compactness of the urban pattern within the landscape itself. Similarly, a high average patch area is indicative of the presence of larger subdivisions or neighbourhoods, whereas low values indicate scattered small developments. The standard deviation of the patch area reflects the distribution of the patch sizes. High values indicate significant differences in the sizes of developments, whereas low values indicate similar sizes. The shape index is a ratio of the actual perimeter of a patch to the patch of the same size that is maximally compact, and ranges from 1 to ∞, with 1 being the index for a patch that is square-shaped. Mean shape index is the average of the shape indices of the patches in the county. In this index low values reflect the compactness of individual developments. Fractal dimension is the indicator shape complexity and takes on values between 1 and 2, with 2 being an indicator that

76

N. Kaza / Landscape and Urban Planning 110 (2013) 74–86

Fig. 1. Estimated permits for housing units 2001–2006.

the development patterns are area-filling lines. These aforementioned metrics thus effectively characterise the spatial structure of a region with some caveats (Hargis, Bissonette, & David, 1998). Landscape metrics can characterise the composition and configuration of the landscape (Turner, 1989). In this analysis, I am not so much concerned with the composition as there are only two classes (urban or non-urban); the only composition metric is the total urban area. The rest of the metrics used in the study are fundamentally about the configuration of urban patches in the landscape. The landscape in this case is a county, an arbitrary political entity that makes sense from a political geography perspective, if not necessarily from an ecological perspective. Because county boundaries are stable and a great deal of economic data is available at this level, I use county as the unit of analysis.

3. Data description For this analysis, I utilised land cover data from the United States Geological Survey (USGS). The 2001 (version 2) and 2006 land cover data for the lower 48 states in the US are retrieved from MultiResolution Land Characteristics Consortium websites (Fry et al., 2011; Homer et al., 2007). At the time of the analysis, comparable datasets for Hawaii and Alaska were not released that prevented their inclusion in the analysis. These land cover products are 30 m × 30 m resolution raster datasets categorised in National Land Cover Data (NLCD) Level II classes. Each of these nine billion cells is categorised into one of the 16 different classes with four categories of urban land: developed open space, low intensity developed, medium intensity developed and high intensity developed. For this analysis, I treat all these categories as urban. A strong reason for using the 2001 and 2006 datasets is that the land cover in each year is consistently classified and compatible with the classification scheme used in the other year. While a 1992 land cover dataset is also available, I did not use it to study the longer-term trends due to differences in the classification method and the scheme. For this reason, USGS requires

that the 2006 NLCD product be only compared to the 2001 NLCD (version 2) product. Accuracy of NLCD datasets is of concern in the literature (Cunningham, 2006; Greenfield, Nowak, & Walton, 2009; Irwin & Bockstael, 2007). Many of these studies are regional in scope, however, a comprehensive national assessment for the 2001 NLCD (version 1) suggests that the Anderson level I accuracy is 85.3%, while the level II accuracy is 78.7% (Wickham, Stehman, Fry, Smith, & Homer, 2010). This suggests a significant cross hierarchy misclassification in the dataset (e.g. developed open land classified as pasture). However, forest and urban areas are primarily misclassified within the hierarchy (e.g. high intensity developed classified as medium intensity). In this work, we are interested in the undifferentiated urban pixels and hence Anderson level I accuracy is more pertinent. While the accuracy assessment of 2006 land cover dataset is not yet readily available, Wickham et al. (2010) reported a 5 percentage point improvement in the 2001 dataset compared to 1992 land cover data. Furthermore, Wickham et al.’s assessment was based on a prior version of the 2001 dataset, not the version 2 used in the current analysis. The 2001 version 2 data includes corrections that were identified during the production of the 2006 dataset and therefore, the accuracy of the datasets used in this analysis is likely to be higher than the reported in the literature. Qualitative differences exist between patterns of urbanisation within rural and within urban areas. Therefore, I use two metrics that expand upon the simple metro and non-metro characterisation by the Office of Management and Budget (OMB) in order to capture the degree of rurality: the Index of Relative Rurality and a threshold-based classification system formulated by Isserman (2005). The Index of Relative Rurality (IRR) is a metric that characterises the rural nature of the county on a continuous scale from 0 to 1, with 1 being the most rural (Waldorf, 2006). The index is calculated from four other variables, population size, density, % population in urban area and the distance to the closest metropolitan area. On the other hand, Isserman uses thresholds of population density, % urban population and metropolitan characterisation

N. Kaza / Landscape and Urban Planning 110 (2013) 74–86

77

Fig. 2. Schematic illustration of urban landscape metrics.

to determine four categories of counties: urban, mixed-urban, mixed-rural and rural. Furthermore, different categories are not evenly distributed geographically. For example, while over 21% of the counties in the Northeast Mid-Atlantic region are urban, rural counties dominate the Mountain West (66%). However, most of the urban counties in the US are in the South Atlantic region (35%), whereas the Northeast (Mid-Atlantic as well as New England) has relatively few rural counties (<2%). While ‘urban’ counties have IRR ranging from ‘0 to 0.3’ and rural from ‘0.3 to 1’, there is a considerable overlap of values between mixed-urban and mixed-rural with the expected skewed distribution. This analysis uses both

the IRR and the Isserman’s metric as complementary measures of urban–rural distinctions. The drivers of urbanisation, as described in the literature on levels of urbanisation, are mainly socioeconomic variables such as population, employment and income growth. Data and projections from the US Census, Bureau of Economic Analysis (BEA) and the Regional Economic Information System (REIS) are also used to correlate changes in demographics and economics with changes in the landscape (Table 1). Three categories of key explanatory variables are considered: demographic changes, economic changes and levels of infrastructure (Fig. 3).

78

N. Kaza / Landscape and Urban Planning 110 (2013) 74–86

Table 1 Summary statistics of key variables.

Gross area (sq.mi.) Highway miles IRR in 2000 % Poverty  Population (annual)  Employment (annual)  Income  Urban area  Number of patches  Mean patch size  Standard deviation of patch size  Mean fractal dimension  Mean shape index

Min

1st Qu

Median

Mean

3rd Qu

Max

2.0 0.0 0 6.44 −26.28 −7.99 −19.74 −0.49 −53.46 −71.22 −46.31 −16.07 −52.50

436.7 162.0 0.39 28.42 −0.50 −0.36 15.11 0.05 −1.15 0.00 0.00 −0.16 −0.06

623.2 244.0 0.51 35.41 0.17 0.57 19.62 0.46 0.00 0.47 0.45 −0.01 0.00

967.9 305.7 0.50 35.57 0.30 0.67 20.14 1.75 0.26 2.52 2.20 −0.24 −0.25

932.5 364.0 0.61 42.37 0.95 1.54 24.51 1.92 0.00 3.03 2.67 0.00 0.23

20,110.0 5310.0 1.00 79.94 8.90 11.73 175.60 37.98 255.93 134.31 61.64 2.91 23.29

NA

1 5 7 53

4

n = 3109; all changes are percentage changes between 2001 and 2006.

In the first category, a county’s development stage is described through demographic changes. Maturity of the county is determined by the decade in which the county achieved 75% of the current population. A county that achieved three quarters of its population before 1950 is labelled ‘very mature’; and a county that reach that point sometime in the 1950s and 1960s is labelled ‘mature’. ‘Developed’ and ‘developing’ counties are those that respectively achieved their three quarters population between 1970 and 1980 or 1990 and later, respectively. I use this variable to test whether the historical period in which a county urbanised has influence on the current changes in patterns. While population can decline, urbanisation is largely irreversible. Counties with declining population show up as mature and thus the metric by itself is problematic. To capture the concurrent population growth and urban growth, I calculate the annualised change in population from 2001 to 2006 from the population estimates of the US Census. Values above zero indicate population growth while values below zero indicate population decline; values differing by one standard deviation away from the national average are ‘rapidly’ changing. In a second category, I use poverty rates to complement the per capita income and employment changes to assess the relative wealth of a county. The BEA publishes annual statistics on county level employment and per capita income. The employment statistics include both proprietors and wage and salary (Fig. 3). Per capita income includes wages as well as transfers such as interests and dividends. I measure changes in per capita income between 2001 and 2006 as a change in the county relative to the change in the US as a whole. For this study, poverty variable is defined as the fraction of the population, whose income in 1999 falls below 200% of the national poverty level. Counties with higher than average poverty fractions are labelled ‘high’ while counties with lower than average poverty fractions are labelled ‘low’. Counties with poverty fractions more than one standard deviation from the mean value are suitably noted as ‘very high’ or ‘very low’. Note that this measure does not account for the regional differences in the cost of living. In addition to the socioeconomic variables, I use a third category, the density of lane miles, to capture the level of infrastructure of a county. I calculate this variable using the 2002 roads data from National Highway Planning Network version 2.1 published by the Bureau of Transportation Statistics as part of the National Transportation Atlas Databases. Lane miles normalised by gross area of the county are binned according to quartiles. One can attribute the distinct difference in the density of highway lane miles east and west of the Mississippi River in rural and mixedrural counties to larger geographic county size in the West. Most urban and mixed-urban counties have a high density of lane miles, because they have large number major roads and smaller areas. Infrastructure is usually considered leading factor in urbanisation.

Furthermore, consistent temporal datasets were not available and therefore changes in lane miles are not used in the analysis. Given the past trends in urbanisation, it is unclear if population and employment growth contribute to sprawling and fragmented development or to infill development that reduces urban fragmentation. One expects that, given the preference for detached single-family houses in the US, income growth would lead to more fragmentation. Also, one would expect that higher levels of highway miles would indicate increasing fragmentation considering that the automobile—along with the infrastructure it requires–is cited as a prime explanation of spatial structure of population settlement in the US. In this analysis, I use various software environments and packages including SDMtools (VanDerWal, Falconi, Januchowski, Shoo, & Storlie, 2011), raster (Hijmans & van Etten, 2011), RColorBrewer (Neuwirth, 2011), spdep (Bivand, 2011) inside R environment. (R Development Core Team, 2009). Data and programs used in the analysis are available at http://maps.dcrp.unc.edu/nkaza/ projects/urbanisation metrics/index.html. 4. A snapshot of US at the turn of the century This section provides a brief description of the broad patterns of the urbanisation of the US. Less than 10% of the total area in a county on average is urbanised. In over 3000 counties or county equivalent areas in the country, St. Louis City, Missouri has the largest proportion of the total area classified as urban land (93%) followed by Alexandria, Virginia (90%). The counties of Petroleum, Missouri and Keweenaw, Michigan, have less than 0.2% urban area, while San Bernardino, California has the largest total geographic area. Furthermore, there is a distinct difference in the area of urban and rural counties. Urban counties have the smallest average area (480 square miles (sq.mi.)), while mixed-rural counties have the largest area (1150 sq.mi.). The size of the county is much larger in the West (of the Mississippi River) than the East. The median county size is smallest in the South Atlantic census region (428 sq.mi.) and largest in the Mountain West census region (2242 sq.mi.). However, among the various census regions (Fig. 1), the Mountain West has the lowest levels of urbanisation and correspondingly lowest levels of highway lane mile density. Most of the population in the US resides in urban counties with an average gross density of 2885 (per sq.mi.), while rural counties have less than 30. The pattern of density in US is also quite uneven. Urban counties exhibited a wide variation of densities across various census regions (1063–6900 persons per sq.mi.), but mixed-urban, mixed-rural and rural counties have fairly even distribution. While it is unsurprising that urban counties are economic engines of the country, there are distinct regional differences. The Mountain West and West North Central Midwest regions have

N. Kaza / Landscape and Urban Planning 110 (2013) 74–86

79

Fig. 3. Spatial patterns of explanatory variables.

the lowest job density (∼30 jobs per sq.mi.) while the Mid-Atlantic region has a concentration 25 times that (∼733 jobs per sq.mi.). This is reflected in the preponderance of urban counties in the MidAtlantic region and rural counties in other regions. Mixed-urban counties have job densities five times greater than the mixed-rural

counties (247 compared to 52 jobs per sq.mi.), a pattern consistent across all regions. These patterns of population and job density are mirrored in the per capita income and poverty indicators. On average, 35% of households are below twice the national threshold for poverty. 80% of

80

N. Kaza / Landscape and Urban Planning 110 (2013) 74–86

household incomes in Buffalo, South Dakota fall into this category, while Los Alamos, New Mexico has the least number of these households (∼6%). Large portions of rural counties in Appalachia, the upper Midwest and the South, historically known for poverty belts defined by racial lines, also have above average rates of poverty and low per capita income. While inner cities are notorious for high poverty levels, urban counties themselves have relatively low poverty rates (Fig. 3). Urban counties have the highest per capita incomes, and it is the mixed-urban counties that have the lowest poverty rates. This is indicative of uneven distribution of incomes in urban counties relative to homogenous income distribution in mixed-urban counties. The West South Central census region consistently has the highest poverty rates across every type of county except rural, for which the East South Central region has the highest rate (45%). The Pacific West has the highest per capita income for urban counties; New England has the highest income in all other types of counties. Outliers on both ends of the income spectrum heavily influence per capita income, but not poverty rates. Therefore, changes in income growth reflect the relative economic growth and decline of the county, whereas the poverty rate provides a static view of the wealth of the county. Very little correlation exists between these two variables (<0.1). Landscape metrics reveal similar but a slightly different story. Circa 2001, urban counties have the lowest number of urban patches on average (435) followed by rural counties (502). A low number of patches in urban counties reflects larger connected or contiguous urban areas; in rural areas, few patches reflect low levels of urbanisation. Mixed-rural counties have the highest number of patches indicating a more severe fragmentation of the landscape. The mean patch area in urban counties is almost five times larger than in rural counties. However, the standard deviation of patch sizes indicates that there is significant variation in the patch sizes in urban counties, whereas rural counties have the least variation in the patch sizes, reflecting the relative homogeneity of size of developments in these counties. Regional differences are also striking. The West Pacific region has the largest number of patches in all types of counties except rural counties. Among rural and mixed-rural counties, the West South Central region is both the least fragmented region with the lowest number of patches as well as the region with the largest average patch size. The South Atlantic region is least fragmented among urban and mixed-urban counties. The averages among indicators vary by the type of county and reveal regional differences in socioeconomic variables, infrastructure levels and patterns of urbanisation. That these differences exist in and of themselves is not striking. But these various indicators paint a different picture of the landscape of the United States than earlier studies have portrayed and, thus, we can now get a more complete picture of the changes in the urban form.

5. Patterns of change in early 2000 There are significant regional differences in the type of economic and demographic changes between urban and non-urban counties (Fig. 3). While most urban and mixed-urban counties have been growing at rates that are much higher than the US average, the largest population growth has been in the mixed-rural counties in the West. The rural counties in the Midwest have continued to experience population decline that is moderate (either negative or less than one standard deviation) to rapid (more than one standard deviation). However, some rural counties in the South Atlantic region have experienced significant population increases. Employment during this period increased in most urban counties, however there were some significant exceptions. While

many Midwestern urban counties continued to experience decline in employment due to deindustrialisation, the most rapidly declining employment rates in urban counties were in California’s Bay Area due primarily to the readjustment of the technology sector following the “dot-com bust”. For the most part, per capita income growth in urban and mixed-urban counties has been lower than the national average, while patterns have been much more mixed in the rural and mixed-rural counties. In general, between 2001 and 2006, rural counties made up the bulk of counties that registered rapid per capita income gains relative to the nation. While the Louisiana counties of St. Bernard and Orleans registered the largest shift in the per capita income change, that shift was largely a result of the outmigration of low-income households following Hurricane Katrina. Apart from these two counties, Billings, North Dakota, a rural county, had the highest positive change in income; Hamilton, Kansas registered a 20% decline. Surprisingly, it is the urban and mixed-urban counties that had less than stellar income growth, while mixed-rural and rural counties in the West registered larger than average income growth. This is likely due to the existence of already high average incomes in the urban counties. Infrastructure level and population growth are positively correlated in rural counties (0.30), spearman rank correlations unless otherwise noted, and negatively correlated in mixed-urban and urban counties (−0.38). However, no substantively significant correlations are observed for employment changes for this short time period. Income growth and poverty rates exhibit statistically significant positive correlations in all types of counties except rural counties, which show no correlation. In other words, counties with a significantly high proportion of poor households experienced high per capita income growth, but the correlation is not substantively significant (0.27, 0.32 and 0.19 for urban, mixed-urban and mixed-rural counties respectively). Counties that experienced a burst of population growth in the 1990s have experienced the largest change in the total urban area (Table 2). However, they are also the counties that experienced the largest decrease in the number of patches and increase in mean patch sizes. On average, these counties have 10.5% increase in the average patch sizes and 3% decline in the total number of patches. In contrast, patch sizes decreased by 0.15% in very mature counties and the number of patches increased by 1.5%. Furthermore, counties that are urbanising now (developing and developed) have very large changes in the standard deviation of patch sizes, which represent the large variation in types of new urban development in these counties as well as the location of these developments. Between 2001 and 2006, the total urban areas within a county grew, with rapid land conversion occurring in most urban, mixedurban and some mixed-rural counties, while the total urban area in rural counties remained the same (Fig. 4). Urban counties in the Northeast and Pacific West registered moderate gains (in terms of percentage), while rapid changes were observed in Southern and Western (except Pacific West) urban counties. On average, the proportion of urban areas in mixed-urban counties grew by 5.7%, whereas in mixed-rural counties, it grew by 2.7%. The fastest growing urban counties are predominantly in the Mountain West (11% on average), whereas the fastest growing mixed-urban counties are in the South Atlantic and West South Central regions (∼7.5%). The variance in these differences, both by type of county and by region, is statistically significant. Regional differences in levels of urbanisation notwithstanding, the heterogeneous patterns of urbanisation are remarkable. The number of patches in a county declined by and large over this time period, suggesting that urban growth became more contiguous. The largest declines in the number of patches were observed in the urban and mixed-urban counties across the US. The number of patches in the mixed-rural counties was marginally lower in

N. Kaza / Landscape and Urban Planning 110 (2013) 74–86

81

Table 2 Average percentage change in the landscape metrics between 2001 and 2006 by maturity of the county.

Developing Developed Mature Very mature

Total area

No. of patches

Average size

Standard deviation

Average fractal dimension

Average shape index

6.33 2.59 2.07 0.58

−2.92 −1.45 1.14 1.53

10.49 5.00 3.16 −0.15

8.84 4.06 2.63 0.13

−0.34 −0.24 −0.38 −0.20

0.25 0.12 −0.64 −0.44

Categories are based on when 75% of the 2000 population is achieved. Very mature—before 1950; mature—1950-1970; developed—1970–1990; developing—after 1990.

2006 than in 2001. Mean patch size increased in almost all types of counties except rural counties. Given the relentless urban growth in the preponderance of counties, it is apparent that the urban form is becoming more contiguous in the US, at least in the short term. With a few exceptions, the patch sizes are becoming more divergent within each county, irrespective of the type of county or the region as evidenced by the increase in standard deviation of urban patch areas. Two different mechanisms are likely. In counties with decreasing number of patches, infill and coalescent development is creating larger patches from existing similar size patches. In counties with an increasing number of urban patches, the new fragmented development is of varying sizes and different from existing urban areas. Only some rural and mixed-rural counties in the Mountain West and in Texas are experiencing a convergence of patch sizes. These are also counties that have experienced a large increase in number and a decrease in average patch size. All three indicators, point to a pattern; new development in these counties that are unconnected to existing development, but are of smaller sizes than before (e.g. smaller ranches and subdivisions). Thus, fragmented urban development in these counties continues as before. In analysing that shape index and fractal dimension index, which are indicators of the shape of the patterns of urbanisation, I find only relatively minor changes. The mean fractal dimension index stayed the same in most rural and mixed-rural counties. However, this index exhibited a decline in many urban counties, especially in the Northeast and Midwest. The decline reflects the simplification of shapes in these urban counties. Very few counties exhibited an increase in this index. The correlations between socioeconomic variables and landscape metrics reveal strong differences in the types of counties as well. The correlations (whether negative or positive) are frequently not large, but they are statistically significant. While larger urban counties experienced a larger amount of urbanisation, the converse is true in mixed-rural and rural counties. It is the change in the population more than employment that has a significant impact on the total urbanisation in all types of counties. Population change is inversely correlated to fragmentation, as evidenced by the negative correlation with the number of patches and mean patch size. IRR is only strongly correlated to most of the landscape metrics for rural and mixed-rural counties. Since IRR expresses the relative rurality within a particular type of county, the correlations of the landscape metrics provide a richer picture of the urbanisation pattern. Within urban counties for example, the less urban is the county, the more likely it is to experience higher levels of urbanisation (0.38). Likewise, within rural and mixed-rural counties, the more rural they are, the more likely they are to have experienced low levels of urbanisation (−0.45 and −0.48) within the period of study. In particular, with greater rurality in a mixed-rural county, the county experienced higher fragmentation. However, that relationship is not as pronounced in rural counties and is completely absent in urban and mixed-urban counties. The relative change in income is not significantly correlated with any of the landscape metrics, but the poverty indicator is relevant in mixed-urban and mixed-rural counties. Poverty is both moderately

and negatively correlated with change in urban area, mean patch size and the standard deviation of patch size. In other words, high poverty counties are likely to be more fragmented. Surprisingly, the density of lane miles is negatively correlated with total urban area in urban and mixed-urban counties, but is positively associated with urbanisation in mixed-rural and rural counties. An increase in the density of lane miles is also associated with increasing fragmentation in urban counties, but it also indicates more continuous development in rural and mixed-rural counties. The metrics of patch shapes are not necessarily correlated to many of the socioeconomic variables, suggesting that deliberate designs rather than natural processes have determined the shapes of individual urban developments. Ordinary least squares regression models are used to summarise the relationship between socioeconomic and infrastructure variables and the changes in the landscape metrics (Table 3). The dependent variables are percentage changes in the metrics from 2001 to 2006. The table should be interpreted with caution. Because the entire population, not a sample of the counties is used, the interpretation of the coefficients differs from the typical inference from sample relationships to population relationships. Instead, these regression models should be used as pattern recognition tools that reduce the dimension of data. In general, the R2 of these models are low, in particular for metrics such as the change in number of patches. This suggests that the socioeconomic variables and landscape metrics are orthogonal. Thus, an analysis using indices that rely on both kinds of information will give a more complete picture of urbanisation patterns in the US. The type of county is statistically significant for all six landscape metrics. Even controlling for employment and population growth, urban counties show a decrease in the number of patches, a higher increase in average patch size and a larger variation in patch sizes than other counties (Table 3). Increases in population result in more contiguous patterns of development, perhaps due to infill development, while increases in employment result in more fragmentation. The higher the per capita income change, the more urban patches increase in the county and the more the average size decreases. Controlling for other variables, the higher the poverty rate in a county, the lower the amount and higher fragmented urbanisation. Surprisingly, with higher levels of infrastructure density, we see lower urbanisation resulting in the county. Western urban counties are also geographically large counties and therefore have low lane mile density, but have experienced high urbanisation. Lane mile density is correlated to an increase in the number of patches and a decrease in the mean patch size (i.e. higher fragmentation). This effect is primarily due to the fragmentary urban development in rural and mixed-rural counties. The analysis above is about changes in patterns within counties. Multi-county spatial clustering is also apparent in the pattern of urbanisation (Fig. 5). Local Indicators of Spatial Autocorrelation (LISA) are calculated for various metrics and local clusters are tested for statistical significance (Anselin, 1995). Regions around Los Angles, Chicago, Denver, Boston, Charlotte, San Jose and Houston have experienced high urbanisation rates between 2001 and 2006 (Fig. 5(a)). However, only rural Utah and Texas

82

N. Kaza / Landscape and Urban Planning 110 (2013) 74–86

Fig. 4. Changes in urban landscape metrics 2001–2006.

have experienced more fragmentation of the urban area (Fig. 5(b)). Regions around Chicago, Sacramento, Boston, Atlanta and suburban Virginia counties around Washington, DC, have experienced coalescent urban growth (Fig. 5(c)). These results point to regional institutional differences at work that make and remake urban form.

6. Discussion While no single landscape metric can provide a full picture, a combination of metrics can be used to supplement our understanding of the pattern of urbanisation. While increasing number of patches reflect on fragmentary urbanisation, this coupled with

N. Kaza / Landscape and Urban Planning 110 (2013) 74–86

83

Table 3 Relationships between urban landscape metrics and explanatory variables.

(Intercept)  Population  Employment  Income Poverty level Lane mile density Area (sq.mi.) Mixed-rural Mixed-urban Urban N R2 Adj. R2

Total area

No. of patches

Average size

Standard deviation

Average fractal dimension

Averge Shape index

0.005 (0.007) 0.010 *** (0.000) 0.001 ** (0.000) 0.013 * (0.007) −0.025 *** (0.005) −0.004 *** (0.001) −0.000 * (0.000) 0.011 *** (0.001) 0.037 *** (0.002) 0.029 *** (0.003)

−0.268 *** (0.035) −0.009 *** (0.003) 0.007 ** (0.002) 0.241 *** (0.035) 0.095 *** (0.025) 0.008 (0.005) 0.000 ** (0.000) −0.011† (0.006) −0.052 *** (0.012) −0.109 *** (0.014)

0.168 *** (0.022) 0.019 *** (0.002) −0.003 † (0.001) −0.125 *** (0.022) −0.094 *** (0.016) −0.018 *** (0.003) −0.000 * (0.000) 0.022 *** (0.004) 0.101 *** (0.008) 0.178 *** (0.009)

0.109 *** (0.015) 0.016 *** (0.001) −0.001† (0.001) −0.078 *** (0.015) −0.063 *** (0.011) −0.010 *** (0.002) −0.000 *** (0.000) 0.019 *** (0.002) 0.069 *** (0.005) 0.099 *** (0.006)

0.011 *** (0.002) −0.000 (0.000) −0.000 * (0.000) −0.011 *** (0.002) −0.004 * (0.002) 0.001 ** (0.000) −0.000 (0.000) −0.001 * (0.000) −0.004 *** (0.001) −0.006 *** (0.001)

0.052 *** (0.008) 0.002 ** (0.001) −0.001 ** (0.000) −0.048 *** (0.008) −0.020 *** (0.006) −0.001 (0.001) −0.000 * (0.000) 0.002† (0.001) 0.006 * (0.003) 0.008 * (0.003)

3041 0.426 0.424

3041 0.082 0.079

3041 0.307 0.305

3038 0.335 0.333

3041 0.040 0.038

3041 0.043 0.040

Standard errors in parentheses. Independent variables are percentage changes in metric between 2001 and 2006. Mixed-rural, mixed-urban and urban counties are contrasted with rural counties. † Significant at p < .10. * Significant at p < .05. ** Significant at p < .01. *** Significant at p < .001.

decrease in patch sizes can be used to deduce that new suburban developments are smaller in size than existing neighbourhoods. The increases in average patch size as well as the number of patches is indicative of large subdivisions such as retirement communities and planned unit developments. The decrease in the number of patches are indicative of coalescent urbanisation, coupled with an increase in average patch size, is indicative of infill development. Without the decrease in the number of patches, the increase in average patch size and the increase in total urban area, is indicative of contiguous development and not necessarily infill development. The changes in the variance of the patch area are also indicative, however, only when coupled with other metrics. Divergence in patch sizes alone can be due to fragmentary development, where most new development are of different sizes than existing neighbourhoods or due to coalescent urbanisation that dramatically changes both the number and size of the urban patches. Therefore, a decrease in the range of patch sizes in coalescent urbanisation is indicative of urban form becoming more mature as neighbourhoods become of similar sizes. Whereas, an increase in the range is indicative of infill development with smaller stand-alone urban patches at the edges. The key finding of the above analysis is that there are differences in the patterns of urbanisation based on the type of the county and the region. The urban counties in the US, by and large have experienced coalescent urbanisation either through infill development or through contiguous growth. The counties that tended to have fragmentary growth in the short term tended to be mostly rural and, in some cases, mixed-rural counties. While large portions of developing and developed urban counties are in the South and in the West, more mature counties are in the Midwest and the industrial Northeast. These mature counties experienced fragmentary urbanisation, whereas developing counties experienced coalescent urbanisation. Also, the more rural a mixed-rural county is, the higher the fragmentation of urban development, and such a relationship is absent or marginal for other type of counties. As subprime mortgage markets drove much of development during this part of

the decade, poorer counties experienced more fragmentary development. A comparison of Figs. 1 and 5 points to the different patterns. By and large, the clusters of large urbanisation happened in California, Colorado and Florida as evidenced both by large numbers of housing permits and newly urbanised area. However, more than the similarities in these two maps, it is the differences that are worth noting. Regions around Seattle, WI and Detroit, MI have high number of permits but insignificant clusters of urbanisation. Seattle is a booming service economy and Detroit is a declining old industrial region (Fig. 3). One can surmise from these differences that new housing permits in Detroit area represent redevelopment of already urban areas. Whereas the Seattle region, with its strong growth management regimes have restricted rapid land conversion from other uses to urban uses. Urban landscape metrics are, therefore, an useful way to characterise a dimension of urban development pattern. However, coalescent urbanisation is not equivalent to efficient urbanisation, nor fragmentary to wasteful. Both coalescent and contiguous urbanisation can occur with increasing per capita consumption of land. Discontinuous development at the edges could be beneficial in the long run as they preserve the option of coalescence at a future date. Conclusions drawn from this analysis should be tempered with such recognition. 7. Caveats and further research Li and Wu (2004) identified three kinds of issues with landscape pattern analysis, singling out correlation analyses as problematic due to conceptual flaws and the ecological irrelevance of the landscape indices. Their critique does not apply here, because Li & Wu are concerned about correlations within the landscape metrics and not with other variables that may explain the patterning. Nevertheless, these analyses are not without caveats. Counties are political boundaries in the US that are relatively stable, but from the perspective of the metropolitan regions, they are arbitrary.

84

N. Kaza / Landscape and Urban Planning 110 (2013) 74–86

Fig. 5. Local Indicators of Spatial Autocorrelation (LISA) for change in metrics between 2001 and 2006. In (b) counties that experienced increases number of urban patches by 2% or more are considered and in (c) counties experienced decrease in the number of urban patches by 2% or more are considered.

Many cities and their suburbs span multiple counties. Even within a county, urbanised and non-urbanised areas exhibit different characteristics of patterns. In other words, this analysis is vulnerable to the Modifiable Areal Unit Problem (MAUP) that plagues many geographical analyses (Openshaw, 1984). This analysis, therefore, remains a initial stab at characterising the changes in landscape patterns throughout the United States; future effort could be directed at refining the metrics at regional and local scales. The classification errors also pose some limitations on the analysis. Most landscape metrics are sensitive to the classification errors as patches are determined by contiguous pixels of the same category and errors in classification affect the contiguity. However, because the classification accuracy is likely to be more than 85% for the 2 years used in the analysis and because this is only dataset that

is available for the whole conterminous US, the analysis should be considered at a continental scale. Further regional analyses, more highly refined satellite data that is contextualised with the local land use data should be used (Coppin, Jonckheere, Nackaerts, Muys, & Lambin, 2004; Morgan & O’Sullivan, 2009). Furthermore, the resolution of the data is 900 m2 pixel (0.25 acre). This is roughly the size of a single-family residence in a medium-sized urban area (5 units per acre). Rural areas are characterised more by development that is much less dense (2 acres per unit or more). Because of this, the rural areas have more houses surrounded by grasslands or forests or other land cover, but still be considered developed from a land use perspective, if not from satellite imagery (Irwin & Bockstael, 2007). Thus urban land in rural areas would appear more fragmented in any given year. However,

N. Kaza / Landscape and Urban Planning 110 (2013) 74–86

this problem is persistent in both comparison years and, thus, the change in the metric still provide some indication of changes in urbanisation patterns in rural areas. This is one possible explanation for the paradoxical pattern of high population growth accompanied by low urbanisation rates exhibited by some counties. For areas that are usually classified as developed open space—even when urban cells are adjacent to non-urban cells, I treat them as urban areas and thus the metrics are more accurate in capturing the shape of the urban areas. However, as roads are linear urban features, they skew the metrics in rural areas (e.g. see Fig. 2). Future research and more spatially fine-grained analyses should address these issues. The main advantage of these landscape metrics is the ability to use them to monitor urbanisation patterns. The metrics I use in this analysis can be used to study further relationships between urbanisation and public problems such as air and water quality and downturns in the housing or financial markets. For example, air quality is dependent on the pattern and level of urbanisation since transport emissions are highly correlated with driving distances. The patterns of urbanisation can also reveal clues about how to control runoff and manage water quality. Furthermore, data from the latter part of this decade reflecting the collapse of the housing market and dramatic slowing of the economy may provide a different picture of the urbanisation trends and could be pursued as soon as the 2011 landcover dataset is released.

8. Conclusions This study is an analysis of patterns of urbanisation during one of the boom periods in US history. The routinely used urbanisation metrics in the US have missed an important element. The landscape of the US is continuously monitored through satellites. This monitoring can be used to study the changes in urban form along with the changes in natural habitats. This study shows that some elements of urbanisation patterns are not adequately captured by demographic and economic variables alone and need to be supplemented with information from land cover data. Population and employment may increase or decrease in a given period, but urbanisation is largely irreversible. Furthermore, while increases in urbanisation correspond to increases in population and employment, the patterns of urbanisation are not, which makes this study important. Leapfrog development has often contributed to the destruction of sensitive ecosystems and it has been argued that one of the aims of urban policy should be to ‘grow compactly’, which usually meant contiguously. In this paper, I show that landscape metrics are particularly useful to capture the ‘fragmentation at the edges’ in the periurban areas. Urban counties are, perhaps, growing contiguously or becoming more contiguous and it is the rural and mixed-rural ones are areas of concern. Most studies usually focus on Metropolitan Statistical Areas (MSA) and have missed the increasing fragmentary development in mixed-rural and rural areas. This study shows that continuous and comprehensive monitoring of urban areas is necessary and is possible. Urban landscape metrics are useful to capture the fragmentary and coalescence dynamics of urban form. However, they are unlikely to capture the full range of characteristics of urban patterns that are important. For example, low-density development exemplified by low-rise office parks and large lot sizes, even when they are contiguous pose serious environmental problems. Therefore, this study should be seen as complementary, not a substitute, to other indices of urbanisation. While different regions in the US show different urbanisation patterns, the metrics suggest that overall the coalescence phase of the urbanisation is in effect between 2001 and 2006 within urban counties. Rural counties in specific regions (such as Texas and Utah)

85

are experiencing fragmentary development patterns. Understanding the reasons of these differing patterns would require more detailed analyses of the institutional structures that facilitate them. Nevertheless, this study points to the need for nuanced understanding of heterogeneous urban development patterns and their implications. Acknowledgements I am grateful to Brigitte Waldorf who provided data from the Index of Relative Rurality for each county in the US and Mallory Rahe who provided me with the county classification scheme developed by Andrew Isserman. Marlon Boarnet, Todd BenDor and Noreen McDonald provided substantive comments. I also wish to thank Travis Green for excellent research assistance and Ashley Tindell for editing. Remaining errors are my own. References Alig, R. J., Kline, J. D., & Lichtenstein, M. (2004). Urbanization on the US landscape: Looking ahead in the 21st century. Landscape and Urban Planning, 69, 219–234. Anselin, L. (1995). Local indicators of spatial association—LISA. Geographical Analysis, 27, 93–115. Bivand, R., et al. (2011). spdep: Spatial dependence: Weighting schemes, statistics and models. R package version 0.5-43. Brueckner, J. K., & Fansler, D. A. (1983). The economics of urban sprawl: Theory and evidence on the spatial sizes of cities. Review of Economics and Statistics, 65, 479–482. Burchfield, M., Overman, H. G., Puga, D., & Turner, M. A. (2006). Causes of sprawl: A portrait from space. Quarterly Journal of Economics, 121, 587–633. Buyantuyev, A., Wu, J., & Gries, C. (2010). Multiscale analysis of the urbanization pattern of the phoenix metropolitan landscape of USA: Time, space and thematic resolution. Landscape and Urban Planning, 94, 206–217. Clifton, K., Ewing, R., Knaap, G., & Song, Y. (2008). Quantitative analysis of urban form: A multidisciplinary review. Journal of Urbanism: International Research on Placemaking and Urban Sustainability, 1, 17–45. Coppin, P., Jonckheere, I., Nackaerts, K., Muys, B., & Lambin, E. (2004). Digital change detection methods in ecosystem monitoring: A review. International Journal of Remote Sensing, 25, 1565–1596. Cunningham, M. A. (2006). Accuracy assessment of digitized and classified land cover data for wildlife habitat. Landscape and Urban Planning, 78, 217–228. Cutsinger, J., & Galster, G. (2006). There is no sprawl syndrome: A new typology of metropolitan land use patterns. Urban Geography, 27, 228–252. Downs, A. (1999). Some realities about sprawl and urban decline. Housing Policy Debate, 10, 955–974. Ewing, R., Pendall, R., & Chen, D. (2003). Measuring sprawl and its transportation impacts. Transportation Research Record: Journal of the Transportation Research Board, 1831, 175–183. Fry, J., Xian, G., Jin, S., Dewitz, J., Homer, C., Limin, Y., Barnes, C., Herold, N., & Wickham, J. (2011). Completion of the 2006 national land cover database for the conterminous United States. Photogrammetric Engineering and Remote Sensing, 77, 858–864. Greenfield, E., Nowak, D., Walton, J., et al. (2009). Assessment of 2001 NLCD percent tree and impervious cover estimates. Photogrammetric Engineering and Remote Sensing, 75, 1279–1287. Gustafson, E. J. (1998). Quantifying landscape spatial pattern: What is the state of the art? Ecosystems, 1, 143–156. http://dx.doi.org/10.1007/s100219900011 Haines-Young, R., & Chopping, M. (1996). Quantifying landscape structure: A review of landscape indices and their application to forested landscapes. Progress in Physical Geography, 20, 418–445. Hargis, C. D., Bissonette, J. A., & David, J. L. (1998). The behavior of landscape metrics commonly used in the study of habitat fragmentation. Landscape Ecology, 13, 167–186. Harvey, R. O., & Clark, W. A. V. (1965). The nature and economics of urban sprawl. Land Economics, 41, 1–9. Herold, M., Scepan, J., & Clarke, K. C. (2002). The use of remote sensing and landscape metrics to describe structures and changes in urban land uses. Environment and Planning A, 34, 1443–1458. Hijmans, R. J., & van Etten, J. (2011). raster: Geographic analysis and modeling with raster data. R package version 1.7-46. Homer, C., Dewitz, J., Fry, J., Coan, M., Hossain, N., Larson, C., Herold, N., McKerrow, A., VanDriel, J., & Wickham, J. (2007). Completion of the 2001 national land cover database for the conterminous United States. Photogrammetric Engineering and Remote Sensing, 73, 337–341. Irwin, E., & Bockstael, N. (2007). The evolution of urban sprawl: Evidence of spatial heterogeneity and increasing land fragmentation. Proceedings of the National Academy of Sciences of United States of America, 104, 20672. Isserman, A. (2005). In the national interest: Defining rural and urban correctly in research and public policy. International Regional Science Review, 28, 465–499.

86

N. Kaza / Landscape and Urban Planning 110 (2013) 74–86

Johnson, M. (2001). Environmental impacts of urban sprawl: A survey of the literature and proposed research agenda. Environment and Planning A, 33, 717–735. Knaap, G. J., Song, Y., & Nedovic-Budic, Z. (2007). Measuring patterns of urban development: New intelligence for the war on sprawl. Local Environment, 12, 239–257. Li, H., & Wu, J. (2004). Use and misuse of landscape indices. Landscape Ecology, 19, 389–399. http://dx.doi.org/10.1023/B:LAND.0000030441.15628.d6 McGarigal, K., & Marks, B. J. (1995). FRAGSTATS: Spatial pattern analysis program for quantifying landscape structure. FRAGSTATS Manual, 97331. McGrath, D. (2005). More evidence on the spatial scale of cities. Journal of Urban Economics, 58, 1–10. Morgan, F., & O’Sullivan, D. (2009). Using binary space partitioning to generate urban spatial patterns. In 4th International Conference on Computers in Urban Planning and Urban Management. Neuwirth, E. (2011). RColorBrewer: ColorBrewer palettes. R package version 1.0-5. Ohls, J. C., & Pines, D. (1975). Discontinuous urban development and economic efficiency. Land Economics, 51, 224–234. O’Neill, R., Krummel, J., Gardner, R., Sugihara, G., Jackson, B., DeAngelis, D., Milne, B., Turner, M., Zygmunt, B., Christensen, S., Dale, V., & Graham, R. (1988). Indices of landscape pattern. Landscape Ecology, 1, 153–162. Openshaw, S. (1984). The modifiable areal unit problem. Number 38 in concepts and techniques in modern geography. Norwich, UK: Geo Books. Peiser, R. B. (1989). Density and urban sprawl. Land Economics, 65, 193–204. R Development Core Team. (2009). R: A language and environment for statistical computing. ISBN: 3-900051-07-0. Rempel, R. (2008). Patch analyst. http://flash.lakeheadu.ca/rrempel/patch/ Seto, K. C., & Fragkias, M. (2005). Quantifying spatiotemporal patterns of urban landuse change in four cities of China with time series landscape metrics. Landscape Ecology, 20, 871–888. Shrestha, M. K., York, A. M., Boone, C. G., & Zhang, S. (2012). Land fragmentation due to rapid urbanization in the phoenix metropolitan area: Analyzing the spatiotemporal patterns and drivers. Applied Geography, 32, 522–531. Song, Y., & Zenou, Y. (2006). Property tax and urban sprawl: Theory and implications for US cities. Journal of Urban Economics, 60, 519–534.

Stephens, S. E., Koons, D. N., Rotella, J. J., & Willey, D. W. (2004). Effects of habitat fragmentation on avian nesting success: A review of the evidence at multiple spatial scales. Biological Conservation, 115, 101–110. Torrens, P., Alberti, M. (2000). Measuring sprawl (Working Paper 27). University College London: Center for Advanced Spatial Analysis. Turner, M. (1989). Landscape ecology: The effect of pattern on process. Annual Review of Ecology and Systematics, 20, 171–197. Uuemaa, E., Antrop, M., Roosaare, J., Marja, R., & Mander, U. (2009). Landscape metrics and indices: An overview of their use in landscape research. Living Reviews in Landscape Research, 3. VanDerWal, J., Falconi, L., Januchowski, S., Shoo, L., & Storlie, C. (2011). SDMTools: Species distribution modelling tools: Tools for processing data associated with species distribution modelling exercises. R package version 1.1-5. Waldorf, B. (2006). A continuous multi-dimensional measure of rurality: Moving beyond threshold measures. In Annual Meetings of the Association of Agricultural Economics Long Beach, CA. Wheaton, W., & Nechayev, G. (2008). The 1998–2005 housing “bubble” and the current “correction”: What’s different this time? Journal of Real Estate Research, 30, 1–26. Wickham, J., Stehman, S., Fry, J., Smith, J., & Homer, C. (2010). Thematic accuracy of the NLCD 2001 land cover for the conterminous united states. Remote Sensing of Environment, 114, 1286–1296. Xu, C., Liu, M., Zhang, C., An, S., Yu, W., & Chen, J. M. (2007). The spatiotemporal dynamics of rapid urban growth in the Nanjing metropolitan region of China. Landscape Ecology, 22, 925–937. Yeh, A., & Xia, L. (2001). Measurement and monitoring of urban sprawl in a rapidly growing region using entropy. Photogrammetric Engineering and Remote Sensing, 67, 83–90. Nikhil Kaza is an assistant professor of City and Regional Planning at University of North Carolina at Chapel Hill. His research is at the intersection of land use, transportation, energy and environment.