Urban Ecosyst (2016) 19:823–833
DOI 10.1007/s11252-016-0535-3
Defining urban, suburban, and rural: a method to link
perceptual definitions with geospatial measures
of urbanization in central and eastern Massachusetts
Anne G. Short Gianotti 1 & Jackie M. Getson 1 &
Lucy R. Hutyra 1 & David B. Kittredge 2,3
Published online: 16 February 2016
# Springer Science+Business Media New York 2016
Abstract Developing greater understandings of socio-ecological relationships across urbanizing
areas is increasingly recognized as important for the conservation and management of natural
resources in a variety of development contexts. Efforts to do so have been hindered by a lack of
consistent measures of urbanization and the challenge of integrating socio-cultural characteristics
into definitions of urban. We present a novel method for linking perceptual definitions of urban,
suburban, and rural to geospatial characteristics and demonstrate how the method can be used to
map urban, suburban, and rural areas at multiple scales in central and eastern Massachusetts. Our
method can facilitate comparative approaches to urban ecology, be used to scale up socioecological studies, and inform conservation research and practice in urbanizing areas.
Keywords Urbanization . Urban–rural gradient . Definition of ‘urban’ . Social science . Urban
ecology . Conservation
Introduction
Urbanization—including the development of high-density cities as well as suburban sprawl
and the expansion of exurban areas at the urban–rural fringe—is a dominant driver of land
Anne G. Short Gianotti and Jackie M. Getson contributed equally to this work.
* Anne G. Short Gianotti
[email protected]
1
Department of Earth and Environment, Boston University, 685 Commonwealth Avenue, Boston, MA
02215, USA
2
Harvard Forest, Harvard University, 324 N Main St., Petersham, MA 01366, USA
3
Department of Environmental Conservation, University of Massachusetts Amherst, 60 Holdsworth
Way, Amherst, MA 01003, USA
824
Urban Ecosyst (2016) 19:823–833
change across the globe (Grimm et al. 2000). The growing population densities and associated
changes in land use and management that drive urbanization transform local ecosystems.
Urbanization changes the biophysical landscape through increased levels of impervious
surface (McKinney 2002) and changes to the forest structure (McDonnell and Pickett 1990);
alters the flux of energy, nutrients, organisms, and water (Grimm et al. 2000; Pickett et al.
2011; Hutyra et al. 2014); increases pollution levels; and leads to changes in average ambient
temperature levels (McKinney 2002; Pickett et al. 2011).
A growing body of literature investigates the ecology in and of cities as well as the
variation in the structure and function of ecosystems across urban-to-rural gradients (see
Grimm et al. 2000; McKinney 2002; Pickett et al. 2001, 2011). Improved understandings
of urban ecology and differences across urban-to-rural regions are viewed as important for
addressing environmental problems and management of natural resources in cities and
suburbs, planning, and conservation efforts (Grimm et al. 2000; McDonnell and Hahs
2008; McKinney 2002; Pickett et al. 2008, 2011). A parallel but smaller body of
research aims to understand the human dynamics of land management and conservation in
urbanizing areas to better understand what shapes decision making and policy as a means
to mitigate environmental degradation and influence conservation efforts (e.g., Kittredge
et al. 2015; Pickett et al. 2008, 2011).
Among these studies, there is little consistency in methods used to define urban,
suburban, and rural and to quantify urbanization gradients (Raciti et al. 2012; Theobald
2004; McIntyre et al. 2000). Urban ecology studies frequently rely on subjective classification schemes and/or vague definitions that define urban or rural-ness in contrast to
other categories or simply assume the definition of urban (Theobald 2004; McDonnell and
Hahs 2008; McIntyre et al. 2000). Where urbanization has been quantified, common
measures of urbanization include linear distance from a major city (e.g., McDonnell and
Pickett 1990; Medley et al. 1995), population density and other demographic measures
(e.g., McDonnell et al. 1997; Medley et al. 1995), and physical measures such as road
density or the percent of impervious surface (e.g., Medley et al. 1995; Raciti et al. 2012;
see McIntyre et al. 2000 and Theobald 2004 for a review of common definitions and
measures of urbanization). These metrics are often criticized as over-simplifying urban
structure and/or as relying on arbitrary cut-offs (McDonnell and Hahs 2008; McIntyre
et al. 2000; Ramalho and Hobbs 2012).
The inconsistent, vague, and arbitrary definitions are problematic as they limit comparability of urban ecological studies (McDonnell and Hahs 2008; McIntyre et al. 2000) and
research findings can be dependent on the definition of ‘urban’ used in each study (Raciti et al.
2012). Developing clear definitions and meaningful quantitative metrics that place research in
the landscape context and facilitate comparable work is thus a major challenge for researchers
(Theobald 2004; Hahs and McDonnell 2006). The development of such metrics would
facilitate comparison of ecological processes across cities and further our understanding of
urban ecology (McDonnell and Hahs 2008).
The integration of abstract socio-cultural characteristics into definitions of urban is another
challenge to urban ecology and conservation studies. The importance of understanding the
links of social and ecological processes in human dominated landscapes has led to the call for
interdisciplinary approaches to urban ecology and conservation studies (Dow 2002; McIntyre
et al. 2000; Grimm et al. 2000; Pickett et al. 2011). While demographic measures such as
population density and income have been frequently used to quantify urbanization, it has been
hard to integrate more abstract cultural and political characteristics. McIntyre et al. (2000)
Urban Ecosyst (2016) 19:823–833
825
suggest that these more abstract characteristics and residents’ perceptions of their environment
are important for understanding urban ecology because Bperceptions are integral to
people’s motivations and actions^ (p 11). They go on to suggest that using Ba perceptually
based definition of urban provides a key link in the cultural, political, physical, perceptual, and
economic aspects that must be integrated in urban ecology^ (p 11).
In this paper, we address the dual challenge of establishing meaningful and comparable
metrics of urbanization and incorporating perceptual definitions of urban into urbanization
metrics. We present a method for linking abstract perceptual definitions of urban, suburban,
and rural to geospatial characteristics that can be quantified and demonstrate how the method
can be used to map urban, suburban, and rural areas at multiple scales in central and eastern
Massachusetts. We utilize a mail survey to assess landowners’ perception of urban, suburban,
and rural and a decision tree approach to examine the relationship between those perceptual
definitions and geospatial characteristics and classify other parcels and towns as urban,
suburban, or rural. While we acknowledge that no single definition of urbanization will capture
all of the social or ecological processes of interest to urban ecology or conservation studies
(Ramalho and Hobbs 2012), we argue that this method (1) provides a useful approach for
generating a broad yet meaningful measure of urbanization to facilitate comparative approaches to the ecology of cities, (2) can be used to scale up socio-ecological studies, and
(3) can be used in the development of conservation policies.
Study area and methods
Study area
Our study area includes towns located along two 100 km transects that stretch westward from
Boston, Massachusetts to the central part of the state (Fig. 1). Development patterns, land uses,
and human communities vary along the study transects. At the broad scale, the northern
transect forms a somewhat idealized urban-to-rural gradient with development patterns
transitioning from urban areas in the east to dense suburbs followed by less dense suburbs
and rural areas as one travels west. The southern transect parallels a major transportation
corridor (I-90) and contains two urban centers, Boston in the east and Worcester (approximately 65 km west of Boston). Development patterns and land use shift between high and lowdensity suburbs between these two urban centers and quickly transition to more rural characteristics west of Worcester.
Surveying perceptual definitions of urban, suburban, and rural
We used a mail survey to assess landowners’ perception of urban, suburban, and rural along
the study transects. The survey was part of a larger study on land management and conservation along urban-to-rural gradients and contained a question that asked respondents to selfidentify as living in an Burban,^ Bsuburban,^ or Brural^ location.
Survey recipients were selected using a stratified random sampling drawn from assessor tax
records containing information on the location, size, and use of parcels as well as landowner
names and mailing addresses. Five of the transect towns were excluded from the study because
tax assessor data was not available. The sample was stratified by town and parcel size:
landowners with less than or equal to 10 acres and landowners with greater than 10 acres.
826
Urban Ecosyst (2016) 19:823–833
Boston
0 50 100 km
0
10
20 km
Class
Mostly Urban
Urban-Suburban
Suburban
Rural-Suburban
Mostly Rural
Rural
Not Included
Fig. 1 Location of study towns and the Burban-ness^ of towns. The urban-ness of towns was classified using
survey respondents’ self-perceived classifications of Burban^, Bsuburban^, and Brural.^ Individual responses were
aggregated at the town level and towns were classified in the following categories: rural: all respondents selected
rural; mostly rural: respondents selected a mix of rural and suburban with more than 50 % of town respondents
selecting rural; suburban-rural: respondents selected a mix of suburban and rural with more than 50 % of town
respondents selecting suburban; suburban: all respondents selected suburban, urban-suburban: respondents
selected a mix of suburban and urban with more than 50 % of town respondents selecting suburban, or mostly
urban: respondents selected a mix of suburban and urban with more than 50 % of town respondents selecting
urban
Where possible, 27 landowners were randomly selected from each size category for each town.
For towns that had fewer than 27 landowners in one of the size categories, additional
landowners were selected from the other category to ensure a total of 54 landowners for each
town.
A total of 1758 surveys were mailed following a modified Tailored Design Method
(Dillman 2007). Of the mailed surveys, 114 were returned as undeliverable or disqualified
because the respondent was deceased or no longer owned land in Massachusetts. A total of
414 surveys were returned, giving an effective response rate of 25 %. The identical survey was
also administered to an additional 14 landowners within the Boston Metropolitan Statistical
Area (MSA) in conjunction with an ongoing biogeochemical study. These 14 landowners were
identified through personal networks and agreed to allow fieldwork in their yards and complete
the survey. Though not randomly selected, we do not believe that their inclusion creates a bias
in our data since their selection was not based on characteristics likely to shape their perceptual
definition of urban, suburban, or rural. We include them in this analysis to increase our sample
size and thus improve model performance.
Survey responses were paired with the MassGIS parcel database, which includes parcel
boundaries as well as other attributes of the parcel. To ensure that our analysis of
perceptual definitions of urbanization reflected the experiences of resident landowners in
the transect towns, returned surveys were excluded from further analysis if the respondent
indicated that their survey responses were based on a property in a different location, if the
land use code for the corresponding parcel was not for a single family or multi-family
residence, or if the respondent did not answer the question on urban status. The final
analysis included 314 responses.
Urban Ecosyst (2016) 19:823–833
827
Predicting and mapping urbanization
We used a decision tree approach to see if it was possible to predict the perceptual urbanization
class (i.e., Burban^, Bsuburban^, Brural^) of each survey parcel using demographic, physical,
and landscape measures of urbanization (Table 1). We selected attributes that are spatially
explicit, have been used as measures of urbanization in other studies, and are widely available.
To allow us to develop predictive models at multiple scales, each urbanization measure was
calculated at the parcel level, in square 1 and 2.5 km2 buffers around the parcel centroids, and/
or at the town level (Table 1).
Our goal was to find parsimonious predictive model(s) using commonly available data.
Geographical spatial analysis was performed using ArcGIS (version 10.1) and statistical
analysis was carried out using R (version 2.15.3). The R-packages rpart, party, and
randomForest were used to develop the decision tree. A decision tree allows for a sample,
in this instance a survey parcel, to be assigned to a specific group based on classification
rules (Breiman et al. 1984). The 10 most important variables, as determined by
randomForest with 5000 trees, were: town population density, town ISA, 2.5 km2 population density, 1 km2 population density, parcel value, 2.5 km2 ISA, distance to urban
edge, 1 km2 ISA, town area, and town percent canopy. These ten variables were then
compared in different combinations in order to produce the simplest and most informative
model to predict urban status. The model prediction was then compared to the survey
results. Our sample size of 314 was split unevenly between urban, suburban, and rural.
Due to the limited sample size we used 90 % of the data for model development with the
remaining 10 % reserved for model validation. Samples were randomly assigned to either
group. This method was repeated using different random assignment of samples to the
development and validation groups to ensure that model structure would avoid overfitting
and be robust to random variations in the groups.
Once decision trees were generated at the parcel and town levels, we used the decision trees
to classify the urban-status of all single family and multi-family parcels (n = 848,605) and
towns (n = 232) located in the seven counties that include and/or border towns included in the
study area: Franklin, Middlesex, Worcester, Hampshire, Suffolk, Hampton, and Norfolk
counties.
Results
Just under half of the survey respondents classified their location as rural (n = 153), 43 %
classified their location as suburban (n = 136), and 8 % self-classified their location as urban
(n = 25).
Town level decision trees
Two models at the town level performed equally well. Both mean town ISA and mean
town population density predicted the landowner self-perceived urban-status well (accuracy ≅ 84 %). Parcels within a town with mean ISA <8 % were classified as rural, between
8 and 36 % were suburban, and >36 % were urban (Fig. 2a). Parcels in towns with mean
population density <94 people km−2 were classified as rural, between 94 and 1321 people
km−2 were suburban, and >1321 people km−2 were urban (Fig. 2b).
LandScan 2013 Global Population Database
(ORNL 2014)
Impervious Surface 2005 (1 m2) (MassGIS)
Urban and Rural Classification (US Census
Bureau 2010)
Level 3 parcel data (MAPC 2013)
Town Boundaries (MassGIS)
NLCD Canopy (Homer et al. 2015)
Population
ISA
Distance to urban edge
Value
Area
Canopy
Mean canopy coverage (percent)
Value of parcel (dollars m−2)
Area (m2)
Parcel centroid from the edge of a census designated
urban area or cluster (UA/UC) (m)
NA
X
NA
X
NA
NA
Mean daytime population density (people km−2)
Mean impervious area (percent)
Parcel
Description
NA
NA
NA
NA
X
X
1 km2 buffer
NA
NA
NA
NA
X
X
2.5 km2 buffer
X
NA
X
NA
X
X
Town
The Massachusetts Office of Geographic Information (MassGIS) (2009) datalayer was also included in this analysis. The percentage of each land use class was calculated within each
parcel and their corresponding buffers. Every land use class was included as well as summarized into 7 broad categories based on use and vegetation
Square buffers were created around the centroid of each parcel
Source
Variable
Table 1 Description of variables included in decision tree model
828
Urban Ecosyst (2016) 19:823–833
Urban Ecosyst (2016) 19:823–833
a
829
c
0 10 20 km
b
d
Class
Urban
Suburban
Rural
Not Included
Fig. 2 a decision tree based on mean town ISA (%) b decision tree based on mean town population density
(people km−2) c map of central MA towns predicted urban status based on town ISA model d map of central
MA towns predicted urban status based on mean town population model. Bold area indicates sample
transect towns
We used both the percent ISA and population density decision trees to classify towns in
counties overlapping or adjacent to transect towns as urban, suburban, or rural (Fig. 2c and d).
The models agree on the classification of 97 % (n = 224) of the towns. Both models classify
approximately 6 % of towns as urban (though the models differ on the classification of three
towns). The ISA tree classifies approximately 47 % of towns as suburban and 47 % of towns
as rural while the population density tree classifies 50 % of towns as suburban and 44 % of
towns as rural.
Parcel level decision tree
We also generated a decision tree that allowed variation among parcels within the same town.
A decision tree consisting of only mean population density within a square 1 km2 area of the
parcel centroid accurately predicts 78 % of the survey responses. Adding a classification based
on distance to urban edge (defined as the distance from the parcel centroid to the edge of the
closest UA/UC polygon) explains an additional 3 % of the response variation (Fig. 3a presents
the tree based on the 1 km2 population density and distance to UA/UC decision tree).
830
a
Urban Ecosyst (2016) 19:823–833
b
Class
Urban
Suburban
Rural
Water
Not Included
Non-residential
0 10 20 km
c
d
Population
(ppl km-2)
> 2730
0 2 4 km
0
20 40 km
< 2730
UA/UC
Fig. 3 a decision tree based on sub-town data including mean population (people km−2) within a 1 km2 buffer of
the parcel centroid and distance of parcel centroid to the closest UA/UC edge(m) b map of central MA parcels
predicted urban status based on 1 km2 square mean population and distance to UA/UC model. c detail for town of
Worcester indicated by thick black line in Fig. 3b above. d map of central MA with the decision tree population
threshold displayed and the UA/UC edge outlined in black. Bold area indicates sample transect towns
Both models were cross-validated with ten randomly selected subsets including 10 % of the
data. Both models could out perform the other depending on the specific subset evaluated. On
average the model that included distance to urban edge was 2 % more accurate for data subsets
and 3 % more accurate when all data is included.
Since the performance of the tree with only population density and the tree with population
density and distance to the urban edge were similar, we further assessed the performance of the
trees by comparing the number of mis-classified parcels at each terminal node. We created a
metric of Bnode contamination^ where
X
node contamination ¼
fraction of minority classðesÞ=fraction of majority class
Smaller values indicate less node contamination. The average value for node contamination
in the tree based on population density was 0.41 while the value for the tree that also included
distance to urban edge was 0.21. Given this difference, we determined that the tree including
both population density and distance to urban edge out performed the tree that only used mean
1 km2 population density.
All parcels with a mean population density greater than 2730 people per km2 within the
surrounding 1 km2 buffer were classified as urban. Parcels with lower population densities
were classified as suburban if they were located relatively deeply within an UA/UC (more than
711 m from the UA/UC edge) and rural if they were outside of the UA/UC or if they were
within about three-quarters of a kilometer (711 m) from the edge.
All single family or multi-family residential parcels located in the seven counties adjacent
or overlapping the transect towns (n = 848,605) were classified as urban, suburban, or rural
based on the population density and UA/UC decision tree (Fig. 3b). Using this decision tree,
3 % of the parcels were urban, 64 % were suburban, and 32 % were rural. The majority of
towns (59 %) contain parcels in multiple classes with nine towns (4 %) having all three classes.
Approximately one third (n = 76) of the towns contain only rural parcels, 9 % (n = 20) contain
only suburban parcels, and no town contains only urban parcels.
Urban Ecosyst (2016) 19:823–833
831
Discussion
We have presented an approach for assessing landowners’ perceptual definitions of urban,
suburban, and rural and linking those definitions to widely available geospatial data at the
parcel and town level. Using this approach, we found that landowners’ perceptual definitions
of urbanization in central and eastern Massachusetts can be predicted using very simple
decision trees and we used the resulting decision trees to classify parcels and towns across a
larger region of Massachusetts as urban, suburban, and rural.
We have shown that this approach can be used to link perceptual identification of urban/
suburban/rural status to measures of urbanization that are easily quantified and that residents’
perceptual definitions can be captured with surprising simplicity. The decision trees at the town
level each achieve 84 % explanatory power with a single demographic or physical attribute
and are consistent in the classification of 97 % of the towns in the study area. The decision tree
at the parcel level can explain 81 % of the variation in landowners’ perceptual definitions using
only two attributes.
The approach discussed responds to two challenges in urban ecology and conservation: (1)
creating simple measures of urbanization that are meaningful and can be standardized and (2)
integrating social experiences into urbanization measures. While we recognize that the boundary of urban areas is blurry (McIntyre et al. 2000), this approach provides a justification for
selecting broad measures of urbanization to quantify urban-to-rural gradients and provides a
justification for break-points that can be used to distinguish between urban, suburban, and rural
areas.
Our decision tree approach has the potential to facilitate comparison across different urban–
rural regions and generates a meaningful method for socio-ecological studies to be scaled up
through modeling efforts. While our classification and mapping exercise was limited to central
and eastern Massachusetts and may not reflect the social and cultural characteristics in other
areas, future studies could employ a similar approach to identify the geospatial variable(s) that
best capture perceptual definitions of urbanization across multiple geographic contexts and test
the degree to which population density, ISA, and distance to UA/UC are good predictors of
perceptual definitions of urban, suburban, and rural in other contexts.
Our method also has applications for policy relevant conservation research. Using the
approach to map perceptual definitions of urbanization, which may be closely tied to attitudes
on conservation, provides opportunities to study, design, and incorporate conservation strategies that are tailored to the specific social and ecological contexts of urban, suburban, or rural
areas.
A final benefit of this approach is that it can be applied at multiple scales. We developed
decision trees at the town and parcel level. Towns and cities are heterogeneous and often
include areas of different levels of development. This approach can be applied at the parcel
scale to capture some of the heterogeneity within towns. When we applied the approach to
central and eastern Massachusetts, we see that what looks like a simple urban-to rural
progression is more complicated. Indeed, the majority of towns contain parcels in multiple
classes. There are suburban areas within both ‘urban’ and ‘rural’ towns and ‘urban’ areas in
‘rural’ towns.
Although the method presented here suggests a pathway towards a more standard approach
to measure urbanization, we want to be clear that we are not suggesting that this approach
replaces the use of more specific urbanization metrics that are driven by individual research
questions or management problems being addressed. As several researchers have suggested, a
832
Urban Ecosyst (2016) 19:823–833
single universal definition of urbanization is not a useful endpoint. Individual studies should
incorporate more specific metrics derived from the ecological and social processes being
examined (McIntyre et al. 2000; Hahs and McDonnell 2006; Ramalho and Hobbs 2012;
Tavernia and Reed 2009) and should also reflect the unique context as well as the spatial and
temporal heterogeneity of the study area (Ramalho and Hobbs 2012; Tavernia and Reed 2009).
Conclusion
Improved understandings of urban ecology are important for natural resource management,
conservation, and addressing environmental problems in urbanizing areas. Previous work in
urban ecology has been limited by the lack of rigorous and consistent measures of urbanization
that integrate social and ecological experiences and facilitate comparison across urbanizing
areas. Though this study, we have shown that it is possible to develop simple classification
models using perceptual definitions of urban, suburban, and rural. Though such broad
definitions should not preclude the use of more specific metrics tailored to individual studies,
this approach is useful because it provides a method for quantifying the social and perceptual
aspects of urbanization, facilitates comparison across different urban–rural regions, allows
studies to be scaled up through modeling efforts, and provides information that can be useful
for the design of conservation programs and policies.
Acknowledgments The authors are grateful to the landowners who completed the mail survey and for the assistance
and feedback provided by Mark A. Friedl, Leah Nothnagel, and George Andrews. This work was primarily supported
by the National Science Foundation, Coupled Natural Human Systems program (NSF-BCS-1211802), and is a
contribution to the Harvard Forest Long-Term Ecological Research Program (NSF-DEB-0620443). Hutyra was also
supported through a National Science Foundation CAREER award (NSF-DEB-1149471).
References
Breiman L, Friedman JH, Olshen RA, Stone CJ (1984) Classification and regression trees. CA, Wadsworth &
Brooks/Cole Advanced Books & Software, Monterey
Dillman DA (2007) Mail and internet surveys: the tailored design method. John Wiley & Sons, Hoboken, NJ
Dow K (2002) Social dimensions of gradients in urban ecosystems. Urban Ecosyst 4:255–275. doi:10.1023/
A:1015767231137
Grimm NB, Grove JM, Pickett STA, Redman CL (2000) Integrated approaches to long-term studies of urban
ecological systems. Bioscience 50(7):571–584
Hahs AK, McDonnell MJ (2006) Selecting independent measures to quantify Melbourne’s urban – rural gradient.
Landsc Urban Plan 78:435–448. doi:10.1016/j.landurbplan.2005.12.005
Homer CG, Dewitz JA, Yang L, Jin S, Danielson P, Xian G, Coulston J, Herold ND, Wickham JD, Megown K
(2015) Completion of the 2011 National Land Cover Database (NLCD) for the conterminous United StatesRepresenting a decade of land cover change information. Photogramm Eng Remote Sens 81(5):345–354
Hutyra LR, Duren R, Gurney K, Grimm NB, Kort E, Larson E, Shrestha G (2014) Urbanization and the carbon cycle:
current capabilities and future outlook on observing and modeling urban carbon flows. Earth’s Future 2(10):473–495
Kittredge DB, Short Gianotti AG, Hutyra LR, Foster DR, Getson JM (2015) Landowner conservation awareness
across rural-to-urban gradients in Massachusetts. Biol Conserv 184:79–89. doi:10.1016/j.biocon.2015.01.001
Massachusetts Office of Geographic Information (MassGIS) (2009) 2005 Land use, impervious surface, town
boundaries datalayers. http://www.mass.gov/anf/research-and-tech/it-serv-and-support/application-serv/
office-of-geographic-information-massgis/. [Last accessed December 2014]
McDonnell MJ, Hahs AK (2008) The use of gradient analysis studies in advancing our understanding of the
ecology of urbanizing landscapes: Current status and future directions. Landsc Ecol 23(10):1143–1155. doi:
10.1007/s10980-008-9253-4
Urban Ecosyst (2016) 19:823–833
833
McDonnell MJ, Pickett STA (1990) Ecosystem structure and function along urban–rural gradients: an unexploited opportunity for ecology. Ecology 71(4):1232–1237
McDonnell M, McDonnell M, Pickett S, Pickett S, Groffman P, Groffman P, …, Medley K (1997) Ecosystem
processes along an urban to rural gradient. Urban Ecosyst 1:21–36
McIntyre NE, Anez KK, Program LS, Marcos S, Hope D (2000) Urban ecology as an interdisciplinary field:
differences in the use of Burban^ between the social and natural sciences. Urban Ecosyst 4(1997):5–24. doi:
10.1023/A:1009540018553
McKinney ML (2002) Urbanization, biodiversity, and conservation. Bioscience 52(10):883–889
Medley KE, McDonnell MJ, Pickett STA (1995) Forest-landscape structure along an urban-to-rural gradient.
47(14):159–168
Metropolitan Area Planning Council (MAPC) (2013) Massachusetts Land parcel database
Oak Ridge National Lab (ORNL) (2014) LandScan 2013. http://web.ornl.gov/sci/landscan/
Pickett STA, Cadenasso ML, Grove JM, Nilon CH, Pouyat RV, Zipperer WC, Costanza R (2001) Urban
ecological systems: linking terrestrial ecological, physical, and socioeconomic components of metropolitan
areas. Annu Rev Ecol Syst 32:127–157
Pickett STA, Cadenasso ML, Grove JM, Groffman PM, Band LE, Boone CG, …, Wilson MA (2008) Beyond
urban legends: an emergine framework of urban ecology, as illustrated by the Baltimore ecosystem study.
BioScience, 58(2):139–150. doi: 10.1641/B580208
Pickett STA, Cadenasso ML, Grove JM, Boone CG, Groffman PM, Irwin E, …, Warren P (2011) Urban
ecological systems: scientific foundations and a decade of progress. J Environ Manag 92(3):331–362. doi:
10.1016/j.jenvman.2010.08.022
Raciti SM, Hutyra LR, Rao P, Finzi AC (2012) Inconsistent definitions of Burban^ result in different conclusions
about the size of urban carbon and nitrogen stocks. Ecol Appl 22(3):1015–1035. doi:10.1890/11-1250.1
Ramalho CE, Hobbs RJ (2012) Time for a change: dynamic urban ecology. Trends Ecol Evol 27(3):179–188.
doi:10.1016/j.tree.2011.10.008
Tavernia BG, Reed JM (2009) Spatial extent and habitat context influence the nature and strength of relationships
between urbanization measures. Landsc Urban Plan 92(1):47–52. doi:10.1016/j.landurbplan.2009.02.003
Theobald DM (2004) Placing exurban land-use change in a human modification framework. Front Ecol Environ
2(3):139–144. doi:10.1890/1540-9295(2004)002[0139:PELCIA]2.0.CO;2
US Census Bureau (2010) 2010 census urban and rural classification and urban area criteria, https://www.census.
gov/geo/reference/ua/urban–rural-2010.html