Applying a novel urban structure classification to ... - Urban Ecology Lab

Applied Geography 53 (2014) 427e437

Contents lists available at ScienceDirect

Applied Geography journal homepage: www.elsevier.com/locate/apgeog

Applying a novel urban structure classification to compare the relationships of urban structure and surface temperature in Berlin and New York City A. Hamstead c, Peleg Kremer d, Dagmar Haase a, e, Neele Larondelle a, b, *, Zoe d Timon McPhearson a Humboldt University Berlin, Institute of Geography, Lab of Landscape Ecology, Mat-Nat Faculty II, Institute of Geography, Unter den Linden 6, 10099 Berlin, Germany b Potsdam Institute for Climate Impact Research, Climate Impacts & Vulnerabilities e Research Domain II, Telegrafenberg 31, 14473 Potsdam, Germany c Milano School of International Affairs, Management and Urban Policy, The New School, 72 Fifth Avenue 5th Floor, New York, NY 10011, USA d Tishman Environment and Design Center, The New School, 79 Fifth Avenue 16th Floor, New York, NY 10003, USA e Helmholtz Centre for Environmental Research e UFZ, Depart. Computational Landscape Ecology, Permoser Straße 15, 04318 Leipzig, Germany

a b s t r a c t Keywords: Land cover classification Cities Urban structure Temperature regulation Comparative approaches

This study introduces a novel approach to classifying urban structure using land cover and building height. The goal of the study was to improve comparability of urban structureefunction relationships across cities through development of a novel classification framework that can facilitate urban studies of ecological patterns and processes. We tested the suitability of the classification in two very different urban settings e continental Berlin and coastal New York City. Using Landsat temperature data as an ecological function variable, we compared how urban structures in both cities relate to temperature. Results show that in both cities a large range of urban structure classes show similar trends with respect to land surface temperature, despite differences in climate and structure of the two cities. We found that approximately 68% of the area of Berlin and 79% of the area of New York City can be represented with the same fifteen urban structure classes. Results indicate that these common classes share very similar temperature patterns and may indicate broader utility of the classification framework. Among the classes which have the most dissimilar temperature trends between the two cities, we find large differences in inner-class composition and neighboring classes. Findings also show that the presence of water has a strong influence on temperature regulation, as classes containing water have the lowest surface temperatures, indicating a need for prioritizing aquatic ecosystems in urban planning and management. © 2014 Elsevier Ltd. All rights reserved.

Introduction Characteristics of urban form and structure, such as the presence of vegetation and other land covers, as well as spatial configurations of structural elements such as buildings, can influence ecological functioning and human well-being in cities. Structural elements are linked to health of urban residents through air quality,

* Corresponding author. Humboldt University Berlin, Institute of Geography, Lab of Landscape Ecology, Mat-Nat Faculty II, Institute of Geography, Unter den Linden 6, 10099 Berlin, Germany. Tel.: þ49 30 2093 6843. E-mail addresses: [email protected], [email protected] (N. Larondelle), [email protected] (Z.A. Hamstead), kremerp@newschool. edu (P. Kremer), [email protected] (D. Haase), [email protected] (T. McPhearson). http://dx.doi.org/10.1016/j.apgeog.2014.07.004 0143-6228/© 2014 Elsevier Ltd. All rights reserved.

local climate impacts, hydrologic processes, and by influencing mental well-being among other impacts (Alberti, 1999; Hamin & Gurran, 2009; Jackson, 2003; Marquez & Smith, 1999; Stewart & Oke, 2012; Stone & Rodgers, 2001; Velarde, Fry, & Tveit, 2007; Yu & Hien, 2006). Understanding ways in which structural elements are related to functions is necessary for understanding ecology in cities (Jansson, 2013), and in order to guide urban development in a way that utilizes ecological functions to enhance environmental performance. However, since most ecological studies have focused on areas with low human population densities, knowledge of these relationships is not developed enough to glean generalities about the structure and function of cities. Urban comparative approaches can help to build our understanding of urban system properties that underlie structure/function relationships, examine causal factors

428

N. Larondelle et al. / Applied Geography 53 (2014) 427e437

and generate general principles to help guide planning and urban design (McDonnell & Hahs, 2009). Comparative approaches can also help to evaluate the extent to which local conditions versus €, global drivers influence structure/function relationships (Niemela Kotze, & Yli-Pelkonen, 2009). However, a lack of standardized variables and common datasets between cities makes it challenging to conduct cross-city comparison (Hahs, McDonnell, & Breuste, 2009). In order to conduct cross-city comparative research on relationships between urban structures and ecological functions, classification systems that use comparable structures and spatial scales appropriate to the functions of interest are necessary. Many studies have associated specific land use/cover types with indicators of human well-being and environmental performance (Breuste, Haase, & Elmqvist, 2013; Burkhard, Kroll, Müller, & Windhorst, 2009; Kroll, Müller, Haase, & Fohrer, 2012; Larondelle & Haase, 2013; McPhearson, Kremer, & Hamstead, 2013) However, few studies have tested structure/function relationships across cities, particularly across continental boundaries. A review of comparative approaches categorized studies according to their use of urbanization gradients, ruraleurban gradients and change over time analysis (McDonnell & Hahs, 2009). For instance, urban to rural gradient comparative studies have included a comparison of minimum temperature in Baltimore, MD and Phoenix, Arizona (Brazel, Selover, Vose, & Heisler, 2000), a comparison of avian diversity and richness in Quebec, Canada and Rennes, France (Clergeau, Savard, Mennechez, & Falardeau, 1998), and a comparison of biodiversity and ecosystem performance along measures of urban form in five cities in the United Kingdom (Tratalos, Fuller, Warren, Davies, & Gaston, 2007). These studies all obtained direct measurement of function at study sites, rather than across the entire city area. Other studies have conducted comparative analysis over entire urban extents, including a comparison of relationships between urban green space and population, residential area, number of households and urban compactness in 300 European cities (Kabisch & Haase, 2013), a comparison of relationships between urban green space coverage, city area and population size in 396 European cities (Fuller & Gaston, 2009), and a comparison of ecological connectivity and urban form in 66 United States urban areas (Bierwagen, 2005). These studies use pre-existing land use/ land cover classifications, such as the European Corine system, or U.S. National Land Cover Database (NLCD) for urban structure variables, assuming certain relationships between structure and function. Land uses indicate ways in which humans employ land, whereas land cover describes the character of the earth's surface (Meyer & Turner, 1992). Boone et al. (2012) argue that comparative approaches “must be bound by some common denominators, such as the biophysical, social and historical commonalities of regions.” Additionally, in order to conduct comparative studies that test structure/function relationships across continents without obtaining direct measurement from a limited number of study sites, a classification system capable of characterizing urban environments in a more standardized way is necessary. Because the diversity of land cover types tends to be higher in urban areas than surrounding landscapes, and land cover elements change more quickly over fine spatial scales in urban environments (Cadenasso, Pickett, & Schwarz, 2007), urban structure classifications need to have relatively fine spatial resolutions. A standard classification also needs to include land cover or land use types that have similar meanings across cities. For instance, assumptions about the proportion of vegetation in residential land may lead to substantial inaccuracies when comparing across cities or even within a single city. The objectives of this study are 1) to apply an approach to developing an urban structure classification, based on

combinations of structural elements that emerge in the urban environment, to a comparative study of a European and North American city, 2) to test its sensitivity to an indicator of ecosystem functionesurface temperature and 3) to examine the patterns in the relationship between urban structure and surface temperature in the two cities in order to assess the generic nature of the approach. We build on the generic classification procedure proposed by Stewart and Oke (2012), which is based on biophysical properties of the urban surface. In combining land cover elements with building height types, we derive urban structure classes that commonly occur in each city and use these classes to examine the relationship between urban structure and surface temperature. Although other studies have examined relationships between land cover or land use and temperature, our approach is unique in that rather than defining classes a priori, we instead generate a classification based on common combinations of structural elements emergent in the urban landscape. To evaluate the transferability and broader applicability of this classification approach, this study is carried out in two cities: the European continental city of Berlin, Germany and the coastal city of New York in the United States. Study sites New York City (NYC) is the largest city in the U.S. with over 8 million inhabitants (U.S. Census Bureau, 2012), and an average population density of 10,640 inhabitants per km2. It is located at a natural harbor area on the east coast and contains more parkland than any other U.S. city. Due to the City's efforts to further increase green infrastructure, by 2030, the city's tree canopy cover proportion is projected to increase from 21% to 30% (Grove, O'Neil-Dunne, Pelletier, Nowak, & Walton, 2006). However, population growth and climate change impacts will put increased pressure on the city's ecosystems to reduce pollution, regulate temperature, manage stormwater, and provide green recreational spaces for NYC residents (McPhearson, Maddox, Gunther, & Bragdon, 2013). Berlin is the capital and largest city in Germany with over 3.5 million inhabitants (Senatsverwaltung für Stadtentwicklung und Umwelt, 2012) and an average of population density of 4000 inhabitants per km2 (Senatsverwaltung für Stadtentwicklung und Umwelt, 2013a). Located in the northeast of Germany, Berlin is one of the most populated areas in Europe and at the same time one of Europe's greenest cities, with a ratio of unsealed surface of almost 50% (Kabisch & Haase, 2014). Population growth is creating significant pressure for planning and management in Berlin while climate change impacts, such as heat and drought, are expected to increase stress on the city's ecosystems in the next decades (Senatsverwaltung für Stadtentwicklung und Umwelt, 2013b) (Fig. 1). Material and methods Data In order to generate urban structure classes, we combined preexisting land cover and building height data for each city (Table 1). For Berlin the Urban Atlas database (EEA, 2006) was used in combination with the LULC model for Berlin (Lauf et al., 2012) and assumptions for building height for certain building types. The Urban Atlas service offers a high-resolution land use map of urban areas with an overall minimum accuracy for all classes of 80% (European Commission, 2011). Based on Earth observation satellite images with a 2.5 m spatial resolution, the Urban Atlas provides comparable land use data for all of the European larger urban zones with more than 100,000 inhabitants. For NYC, we combined a 3 3 ft raster land cover dataset (NYC Parks and Recreation, 2010) with


429

Fig. 1. Land use classes including building height (low, mid, and high-rise), tree canopy, low plants (shrubs and herbaceous plants), bare soil, pavement, and water and their distribution in Berlin (left) and New York City (right). The original input datasets are used here, New York with a resolution of 1 1 m and Berlin with a resolution of 10 10 m. Two zooms are provided for NYC, to better show the contribution of the classes, not visible in this scale.

building height data based on NYC's tax lot database (MapPLUTO, 2011). The overall per pixel accuracy of the NYC land cover classification is 96% (MacFaden, O'Neil-Dunne, Royar, & Lu, 2012). Utilizing the different sources of data in the two cities with their unique construction and spatial scales imposes a comparability challenge commonly encountered in the pursuit of quantitative cross-city comparisons. For example, in Berlin, the administrative boundary defines the limits of the study area. However, for NYC, the available land cover data excluded most of the water in the Hudson River and East River, which represents a potential limitation in properly assessing the comparative impact of water in the two cities. In order to analyze the relationship between the urban structure classes and surface temperature (ST), we derived ST from the Landsat TM 7 imagery thermal band (6_1). Two different Landsat 7 scenes were selected for each city. Scene selection was based on the following criteria: a) summer day with minimal cloud cover (9 stories) (Stewart & Oke, 2012) (Table 2). The combined land cover and building height datasets were stored as raster files and generalized to a 50 m resolution to achieve comparability between the two cities. In generalizing the urban structure dataset, 50 m pixels were assigned the structural type that covered the most area of the pixel. A fishnet with a cell size of 100 100 m (1 ha) was created and aligned to the cells of

Table 1 Land cover, building height, and surface temperature data sources. Data

Sources for Berlin data

Sources for New York data

Land cover Buildings Surface temperature

Urban Atlas (EEA), 2006 Land use model 2008 (Lauf et al., 2012), expert assumptions Landsat TM 7 imagery thermal band (6_1): July 3 2008 with data gaps filled by July 28 2008 June 1 2008 with data gaps filled by June 12 2006

Land cover dataset (Mcfaden et al., 2012) MapPluto tax lot database (2011) Landsat TM 7 imagery thermal band (6_1): July 15 2011 with data gaps filled by August 20 2010 July 22 2011 with data gaps filled by July 3 2010

430


Table 2 Land cover terminology translated from the original land cover datasets. NYC

Berlin

Land cover terminology

Tree canopy Grass/shrub

Forest Agricultural, semi-natural areas, green urban areas, land without current use, isolated structures Road/rail network and associated land Sport and leisure facilities, mineral extraction and dump sites Water bodies, wetlands

Tree canopy Low plants

Road and other paved surfaces Soil Water

Paved

Table 3 Surface temperature observations for two days. Number of original observations

Date of Landsat temperature data

Number of observations with temperature data

Mean temperature ( C)

Berlin

89,519

New York City

81,510

July 3, 2008 June 1, 2008 July 15, 2011 July 22, 2011

89,555 89,448 79,854 80,566

33.69 27.29 32.57 32.88

Soil Water

the urban structure datasets, allowing for a maximum of four land cover and building height types within each class. These combinations of land cover and building height form the urban structure classification. Using a zonal operation in ArcGIS, we calculated the area of each land cover or building height type for each cell. Fig. 2 illustrates the possible combinations of surface properties that make up a class, the maximum number of classes derived, and an example of how land cover and building type can be combined within a spatial unit. Indicator assessment and data processing Surface temperature was used as a proxy for ambient temperature (Haase, Schwarz, Strohbach, Kroll, & Seppelt, 2012; Schwarz, Bauer, & Haase, 2011), an indicator of the cooling effect of vegetation and water. Landsat 7 thermal (band 6_1) scenes were used to extract ST. In total, 8 scenes were extracted and processed using ENVI 5.0. The processing procedure included radiometric calibration converting Landsat data to radiance using the following equation:

. Ll W m2 mm sr ¼ Gain*pixel value þ Offset Gains and offset values are available from the Landsat scenes metadata (Exelis, 2012b). Atmospheric correction was then performed on all scenes using an In-Scene Atmospheric Compensation algorithm. The resulting corrected radiance data was converted to ST (K) using the emissivity normalization method (Kealy & Hook, 1993). In order to reduce cloud cover effects, pixels with temperature below the threshold of 290 K were removed and

assigned a no data value. Data gaps were filled using the ENVI 5.0 mosaicking with color balance procedure as outlined by USGS (2013). In this procedure gaps in each selected base Landsat 7 scene are filled by an additional Landsat 7 scene (Furby & Wu, 2006) using geo-referenced mosaicking and color balancing. Details of all base and gap filling Landsat 7 scenes are indicated in Table 1. Color balancing through a gain compensation algorithm is a common, per pixel, adjustment in mosaicking (Liu, Li, & Gu, 2011) that was applied to the gap filling images to statistically match the data range between the base and gap filling scenes (Exelis, 2012a). Finally, temperatures were converted to degrees Celsius using a Band Math calculation. Analysis of class temperature profiles We created four tabular datasets at the hectare level, representing two unique days occurring during summer months for each city. We computed the mean, range and standard deviation of the temperature pixels that fell within each cell for each of the two Landsat-derived ST datasets for each city. In this operation, each temperature pixel was weighted equally. These statistics were then merged with the class variables, resulting in two datasets linking urban structure classes to average temperature. Since the two ST datasets had some gaps even after the gap-filling procedure, the final datasets did not include as many observations as the class datasets. For both cities, 98e99% of the observations from the class datasets had associated temperature data. Using the two datasets, we examined the number of classes necessary to represent 90% of land area in each city, overlaps between the classes needed to represent 90% of land area in each city, as well as temperature trends between the two cities. All temperature trend data was analyzed based on the mean temperature values for each class (Table 3). Uncertainties Uncertainties of the classification approach include a) the sensitivity of the comparative temperature analysis to resolution of the original land cover and building data and the generalized urban structure data (Hamstead, Kremer, Larondelle, McPhearson, & Haase, in review), b) the potentially wrong class assignment within the original dataset, c) large within-class differences in urban structure types between the two cities, d) the estimations for building height and e) the potential effects of context on the temperature profile of classes. Additionally, ST data highly depends on satellite data availability and represents a temperature state at a specific point in time, which may be limited relative to long-term or continuous measures. Results

Fig. 2. Possible combinations of surface properties in cities (top) and example where a maximum of four land uses (lowrise building, water, soil, and tree canopy) is combined into one class. The maximal number of combinations is 162 different classes.

In comparing the classification results, we find that 90% of the area of NYC is described by 19 classes and 23% of the city is covered


431

Fig. 3. Spatial distribution of the 15 classes, which both cities have in common. These 15 classes cover 68% of the area in Berlin and 79% of the area in New York City.

by classes with only pervious surfaces. In Berlin, 90% of the area is described by 25 classes, 39% of which is covered by classes with only pervious surfaces. Of the classes which cover 90% of each city's land area, NYC and Berlin have 15 classes in common. In NYC these classes cover 79% of the area and in Berlin 68%. The 15 shared classes contain six classes with buildings; of which five contain lowrise buildings, one midrise and one a mixture of low- and midrise. Another nine classes do not contain buildings. Fig. 3 shows the spatial distribution of the 15 common classes for both cities. Table 4 displays aerial photographs for both cities for these 15 classes. Table 5 summarizes the commonly-occurring classes that cover 90% of the cities' surfaces. Classes which are not shared include four classes in NYC and nine in Berlin. Of the NYC classes not shared with Berlin, three include buildings and one includes vegetation and paved surface. Six of the eight Berlin classes not shared with NYC include buildings. For both cities one class which is not shared includes high-rise buildings, which is not a component that falls within the 15 common classes. Analyzing the relationship between the classes and ST, we find that temperature trends similarly across the classes common to both cities (Fig. 4). Classes that include combinations of buildings and paved areas tend to be warmer while classes that include combinations of tree canopy and water tend to be relatively cooler. Comparing common classes in NYC and Berlin on a standardized scale (Fig. 4, right side) it is clear that the classes generally exhibit similar surface temperature trends. Because the overall city-scale structure of these two cities is different, this result indicates that the classification approach is generic and capable of describing urban structure at the units that we selected. We suggest that the classification procedure developed and tested in Hamstead et al. (in review) was able to identify common units of urban structure for multiple cities that may be related to urban ecological processes and functions, in this case relating similarly to indicators for urban climate. Further statistical information on the classes and the respective temperature analysis can be found in the supplement, Table A.

However, several classes are more dissimilar between the two cities than others (see Fig. 4, right side). The most noticeable difference in relative temperature behavior occurs in the following classes: “tree canopyepaved” (0.33 standardized degrees difference), “soil” (0.26 standardized degrees difference), “low plants-water” (0.22 standardized degrees difference), “lowriseetree canopy” (0.19 standardized degrees difference) and “low plants” (0.19 standardized degrees difference). To understand why these classes differed, we analyzed the within-cell composition of the classes. For the cases “lowriseepaved” (case A) and “tree canopyepaved” (case B) we show that differences in temperature could relate to differences in the distribution of urban structural components within the class (Table 6). For example the class “lowriseepaved” shows relative higher standardized temperatures in New York City than in Berlin. The inner class comparison shows, that on average, this class contains more paved surface in New York City (57%) than in Berlin (34%), whereas in Berlin this class contains on average more lowrise (66%) than New York City (38%). As lowrise often is connected to lowrise detached and semi-detached houses with gardens, the corresponding temperature was lower than a paved surface. Other conglomerate classes in which relative temperatures differed between cities are the “lowriseetree canopy” class (case C) and the “low plantsewater” class (case D). Here, an analysis of the within-cell composition of the class showed that the distribution of the two land cover types is similar in the two cities. Furthermore the homogenous classes “lowrise” (case E), “soil” (case F) and “low plants” (case G) show relative temperature differences, which due to their homogeneity, cannot be explained by an inner class analysis. However, performing an analysis of the classes of neighboring cells could provide a possible explanation for the differences we observed in surface temperature pattern. Therefore, a buffer analysis was performed on neighboring classes around each cell of analyzed classes, using a 200 m doughnut buffer. In NYC, cells within a 200 m buffer of “lowrise-

432


Table 4 Aerial photographs for the 15 common classes.

tree canopy” contained mainly combinations of “lowriseepaved” or other combinations with “paved”. In Berlin, the main class neighboring “lowriseetree canopy” within a 200 m buffer is “tree canopy”, second and third are “lowrise” and “lowriseetree canopy” and forth is “tree canopyewater”. The same analysis was conducted for the class “lowrise” where NYC shows significantly higher

relative surface temperatures. The neighborhood analysis revealed similar results (case E). In NYC the five most occurring, neighboring classes of “lowrise” are all either paved or built, while in Berlin three out of five classes are neither built nor paved. For the class “low plants” (case G) Berlin displayed relatively higher temperatures than NYC. The analysis of the neighboring classes shows that

N. Larondelle et al. / Applied Geography 53 (2014) 427e437 Table 5 Classes displaying 90% of the area of Berlin and New York City, in percent [%] and area [ha]; Classes listed in bold are shared in the top 90% of area in both cities. Class Lowriseemidriseepaved Lowriseemidrise Lowriseetree canopyepaved Lowriseetree canopy Lowriseelow plantsepaved Lowriseelow plants Lowriseepaved Lowriseesoil Lowrise Midriseetree canopyepaved Midriseelow plantsepaved Midriseelow plants Midriseepaved Midrise Highrise Highriseepaved Tree canopyelow plantsepaved Tree canopyelow plants Tree canopyepaved Tree canopy Low plantsepaved Low plantsesoil Low plantsewater Low plants Pavedesoil Paved Soil Water Total

NYC [%]

NYC [ha]

2%

1518

5% 2% 1%

4213 1549 1121

16%

12,605

2% 1%

1301 1010

3%

2819

1% 4% 6% 11% 7% 6%

735 3373 4525 8843 5604 4847

1% 7%

793 5478

13% 1% 2% 90%

10,324 575 1636 72,869

BER [%]

BER [ha]

2% 3%

1485 2992

1% 1% 3% 6% 1% 12%

719 1175 3133 5476 692 11,029

1% 2% 5% 8% 1%

787 1810 4135 6770 859

1% 1% 16% 2% 2% 1% 12% 1% 1% 3% 4% 90%

1096 871 13,885 2038 1389 694 10,568 1049 1015 2905 3932 81,181

even though more than 56% of the class cells are surrounded by the same class “low plants” in Berlin, the second class is “low plantsepaved”. Whereas in NYC the main class neighboring “low plants” is also “low plants” with almost 50% but followed by “tree canopyelow plants” with about 12%, which shows significantly lower temperatures than “low plants” and “low plantsepaved”. Differences in neighboring classes may also explain the differences in temperature between the two cities for the classes “soil” (case F) and “low plants-water” (Case D) (Table 6). Both analyses show that differences in relative temperatures shown in Fig. 4 can be explained by inner-class composition and intra-class neighborhood.

Discussion Overall, temperature signatures of urban structure classes show similar patterns both within and between Berlin and NYC (see Fig. 4). However, some urban structure classes show more difference then others when comparing different dates within a city as well as the two cities. In Berlin, July 3, 2008 shows warmer temperatures for all classes than June 1, 2008, which is expected given that the average temperature on July 3, 2008 was over 6 C warmer than June 1, 2008. However, classes with tree canopy and water combinations show smaller temperature differences between the two days, which may indicate that these elements have a higher capacity to remain relatively cool during hotter days, a finding that is supported elsewhere (Arnfield, 2003; Hathway & Sharples, 2012; McPherson et al., 1997). In NYC, classes with paved surfaces, such as “midriseepaved,” “low plantsepaved,” and “paved,” showed very consistent temperature between July 15, 2011 and July 22, 2011, whereas classes such as “tree canopy,” “low plants,” “low plantsewater” and “water” showed stronger differences between the two days, although the two days vary by only 0.31 standardized degrees overall. These differences may indicate a need to examine a

433

longer-range climatic context, as some classes may have a different capacity to cool down or propensity to store heat over time, and may also point to ways in which other contextual factors matter. Applying a matrix approach (Müller, Werner, & Kelcey, 2010) to this analysis would provide additional knowledge of how the broader landscape context influences ecological outcomes. In general, exploring these differences further may enable us to identify the proximate causes and social processes that influence differences in urban structural patterns between cities, and ultimately help to distinguish contextual factors from more generalized principles that describe structure/function relationships. Comparing ecological function across cities using a common classification system provides a way to observe local versus global trends. Analyzing a larger number of diverse cities is necessary to determine which trends are global in a broader respect, though data availability may be limiting factor. The results of this paper support the rapidly developing notion that urban green and blue infrastructure provide people in urbanized environments with important ecological benefits, including local climate regulation (Chiesura, 2004; Pataki et al., 2011; Peters, Elands, & Buijs, 2010; West, Shores, & Mudd, 2012). Vegetation reduces the urban heat island effect and mitigates negative impacts of climate change (Hamin & Gurran, 2009). For example, the presence of street trees and parks increases thermal comfort (Stathopoulos, Wu, & Zacharias, 2004) by providing shade and can improve air quality by removing air pollutants, thereby positively affecting urban health and wellbeing (Seymour, Wolch, Reynolds, & Bradbury, 2010). Water bodies in and around cities also provide cooling functions (Hathway & Sharples, 2012) and hence contribute to good urban living conditions. Additionally, € lker and Kistemann (2013) showed that water bodies have some Vo overall therapeutic value and reduce stress. Many of these studies analyze certain green and blue infrastructure, but do not connect the urban structure of a place to the ecological function. Cities consist of a mix of urban structural elements where gray, green and blue infrastructure exist in close proximity, and dynamically influence the environmental performance of each other. Therefore, it is important not to study these infrastructures in isolation, but rather to examine the emergent structures formed by interactions of structural elements. Our approach builds on recent urban structure classification efforts (Cadenasso et al., 2007; Stewart & Oke, 2012) to create a broadly-applicable method for urban structure classification, and in this way better enable city comparisons which are otherwise limited due to very different land cover data and respective land cover type descriptions. Developing common urban structure units of analysis to conduct cross-city analysis helps to distinguish between general trends and important differences among cities, which may point to ways in which contextually-related social processes drive unique urban structural conditions. Our results show that the composition of the classes and neighborhood in which a given class is situated play important roles in the structure/ function relationship for urban structure types. Cross-city comparative analysis allows for the identification of patterns in how urban structure elements spatially relate to each other. We find that the majority of land area in Berlin and NYC is described by repeating patterns of similar urban structure combinations, and those combinations are related in similar ways to an indicator of ecological function. More generally, results imply that combinations of urban structural elements at relatively fine scales can describe the building blocks of functions in urban areas. We argue that although the classification system used here is more complex than land use or land cover classifications commonly used to examine functions (Pauleit & Duhme, 2000), this approach is more meaningful in comparative analysis because it allows for an

434


examination of emergent patterns by making fewer a priori assumptions about the classes that are relevant to structureefunction relationships in urban environments. By building the classification in a way that applied to both cities, and potentially to other cities, we were able to identify urban structure classes common to both urban contexts, which allowed for a comparison of similarities and differences in urban structureefunction relationships. Because urban areas are highly heterogenous coupled socialeecological systems, urban classifications need to identify and relate the type, amount, and arrangement of specific urban structural features (Cadenasso et al., 2007) such as land cover and built infrastructure. Urban structure in any particular city can be driven by many kinds of urbanization processes and may depend on the regional, historical, or cultural context (e.g., European, U.S., Asian, or Arabian) (Schwarz, 2010), development state, biogeophysical context (e.g., desert, coastal, tropical, or temperate), or other urban structure drivers (Elmqvist et al., 2013). To advance the understanding of urban structureefunction relationships and how they

compare across cities requires considering land cover, and built infrastructure elements in a flexible and more nuanced approach for the analysis of urban structure, particularly when comparing across cities or applying a classification in multiple or different urban contexts. Conclusions The classification approach developed and tested in this paper describes urban structure of two case study cities, NYC and Berlin, and allows us to identify and represent general properties of urban structure and relate them to a particular ecological function, surface temperature. We found that 79% of the area of NYC and 68% of Berlin can be described with as little as 15 urban structure classes. Further, these same urban structure classes describe surface temperature similarly, demonstrating a surprising overlap in the structure/function relationship despite major differences in the two cities' climate, size, levels of built

Fig. 4. Temperature trends observed across 15 classes shared between New York City and Berlin, divided into six built classes (left of black line) and nine non-built classes (right of black line). The left side of the figure shows the temperatures for two days in Berlin and NYC separately, and the right side shows the temperatures for both cities in a single panel (top: temperatures for both cities; bottom: temperatures for both cities, minemax standardized).


435

Table 6 Analysis of inner-class composition and of neighboring classes. Colors indicate certain land use (gray ¼ classes including paved; red ¼ classes including built not paved; green ¼ classes including tree canopy or low plants without paved or building component; blue ¼ only water; yellow ¼ only soil).

versus green space, and cultural, economic and ecological histories. The overall goal of this study was to develop a generic urban structure classification method with which we can lower the uncertainty associated with cross-city assessments of urban structure/ function relationships in order to generate general principles to help guide urban planning and design. We have identified urban structure classes that exhibit similar temperature trends in different cities, as well as potential contextual factors, such as

neighborhood context, that may account for different trends in some classes across the two cities. We therefore advance this new classification approach to provide a novel basis for analyzing structure/function relationships more comprehensively and comparatively in heterogenous urban landscapes. We suggest that the power of this classification approach lies in its ability to describe land cover mixtures rather than single, land cover types, which is better suited to address complex and heterogenous urban areas and is more appropriate for comparative

436


analysis. The classification method can be used for city comparisons, even when e as is often the case e data are not directly comparable. Further development of a generic classification should apply this approach to other city contexts, and evaluate additional processes and functions of importance to city planners, managers, and residents, such as noise level, stormwater absorption, patterns of biodiversity, and air pollution removal. Further, more refined temperature data, including wind speed and ambient temperature would link the classification more directly to an indicator connected to human well-being for specific times of day, when exposure to heat stress may be especially high. We suggest that a more complex and nuanced urban structure classification, such as the one developed here, is important for understanding urban structure/function relationships for both within- and cross-city studies. This approach has potential utility for deriving general urban patterns and differences, and ultimately for conducting more accurate and meaningful comparative analyses. Acknowledgments We wish to thank our colleagues in the URBES Project (www. urbesproject.org), the Biodiversa Programme of the EU, the Tishman Environment Design Center at The New School, and Stephan Pauleit for very helpful comments on this manuscript. Appendix A. Supplementary data Supplementary data related to this article can be found at http:// dx.doi.org/10.1016/j.apgeog.2014.07.004. References Alberti, M. (1999). Urban patterns and environmental performance: what do we know? Journal of Planning Education and Research, 19(2), 151e163. http:// dx.doi.org/10.1177/0739456X9901900205. Arnfield, A. J. (2003). Two decades of urban climate research: a review of turbulence, exchanges of energy and water, and the urban heat island. International Journal of Climatology, 23(1), 1e26. http://dx.doi.org/10.1002/joc.859. Bierwagen, B. G. (2005). Predicting ecological connectivity in urbanizing landscapes. Environment and Planning B: Planning and Design, 32(5), 763e776. http:// dx.doi.org/10.1068/b31134. Boone, C. G., Cook, E., Hall, S. J., Nation, M. L., Grimm, N. B., Raish, C. B., et al. (2012). A comparative gradient approach as a tool for understanding and managing urban ecosystems. Urban Ecosystems, 15, 795e807. http://dx.doi.org/10.1007/ s11252-012-0240-9. Brazel, A., Selover, N., Vose, R., & Heisler, G. (2000). The tale of two climatesdBaltimore and Phoenix urban LTER sites. Climate Research, 15, 123e135. Breuste, J., Haase, D., & Elmqvist, T. (2013). Urban landscapes and ecosystem services. In S. Wratten, H. Sandhu, R. Cullen, & R. Costanza (Eds.), Ecosystem services in agricultural and urban landscapes (pp. 83e104). Wiley. Burkhard, B., Kroll, F., Müller, F., & Windhorst, W. (2009). Landscapes' capacities to provide ecosystem services e a concept for land-cover based assessments. Landscape Online, 1e22. Cadenasso, M. L., Pickett, S. T. A., & Schwarz, K. (2007). Spatial heterogeneity in urban ecosystems: reconceptualizing land cover and a framework for classification. Frontiers in Ecology and the Environment, 5(2), 80e88. Chiesura, A. (2004). The role of urban parks for the sustainable city. Landscape and Urban Planning, 68(1), 129e138. http://dx.doi.org/10.1016/j.landurbplan.2003.08.003. Clergeau, P., Savard, J.-P. L., Mennechez, G., & Falardeau, G. (1998). Bird abundance and diversity along an urbanerural gradient: a comparative study between two cities on different continents. The Condor, 100(3), 413e425. Elmqvist, T., Fragkias, M., Goodness, J., Güneralp, B., Marcotullio, P. J., McDonald, R. I., et al. (2013). Global urbanisation, biodiversity and ecosystem services: Challenges and opportunities. Springer. European Environmental Agency EEA. (2006). Retrieved from http://www.eea. europa.eu/data-and-maps/data/ urban-atlas (17.08.12 ). European Commission. (2011). Mapping guide for a European urban atlas. Retrieved from http://www.eea.europa.eu/data-and-maps/data/urban-atlas/mapping-guide. Exelis Visual Information Solutions. (2012a). ENVI help mosaic images. Exelis Visual Information Solutions. (2012b). EVNI help radiometric calibration. Fuller, R. A., & Gaston, K. J. (2009). The scaling of green space coverage in European cities. Biology Letters, 5(3), 352e355. http://dx.doi.org/10.1098/rsbl.2009.0010.

Furby, S., & Wu, X. (2006). Evaluation of Landsat 7 SLC-off image data for forest cover mapping. CMIS technical report 06/154. CSIRO Mathematical and Information Sciences. Grove, J. M., O'Neil-Dunne, J., Pelletier, K., Nowak, D., & Walton, J. (2006). A report on New York City's present and possible urban tree canopy. Prepared for: Parks & Recreation. New York. Haase, D., Schwarz, N., Strohbach, M., Kroll, F., & Seppelt, R. (2012). Synergies, tradeoffs, and losses of ecosystem services in urban regions: an integrated multiscale framework applied to the Leipzig-Halle region, Germany. Ecology and Society, 17(3). http://dx.doi.org/10.5751/ES-04853-170322. art22. Hahs, A. K., McDonnell, M. J., & Breuste, J. H. (2009). A comparative ecology of cities and towns: synthesis of opportunities and limitations. In M. J. McDonnell, A. K. Hahs, & J. H. Breuste (Eds.), Ecology of cities and towns: A comparative approach (pp. 574e596). Cambridge University Press. Hamin, E. M., & Gurran, N. (2009). Urban form and climate change: balancing adaptation and mitigation in the U.S. and Australia. Habitat International, 33(3), 238e245. http://dx.doi.org/10.1016/j.habitatint.2008.10.005. Hamstead, Z., Kremer, P., Larondelle, N., McPhearson, T., & Haase D., A classification approach for relating surface temperature to urban structure at multiple scale. under review, available from the authors upon request. Hathway, E. A., & Sharples, S. (2012). The interaction of rivers and urban form in mitigating the urban heat island effect: a UK case study. Building and Environment, 58, 14e22. http://dx.doi.org/10.1016/j.buildenv.2012.06.013. Jackson, L. E. (2003). The relationship of urban design to human health and condition. Landscape and Urban Planning, 64, 191e200. Jansson, Å. (2013). Reaching for a sustainable, resilient urban future using the lens of ecosystem services. Ecological Economics, 86, 285e291. http://dx.doi.org/ 10.1016/j.ecolecon.2012.06.013. Kabisch, N., & Haase, D. (2013). Green spaces of European cities revisited for 1990e2006. Landscape and Urban Planning, 110, 113e122. http://dx.doi.org/ 10.1016/j.landurbplan.2012.10.017. Kabisch, N., & Haase, D. (2014). Green justice or just green? Provision of urban green spaces in Berlin, Germany. Landscape and Urban Planning, 122, 129e139. Elsevier B.V. http://dx.doi.org/10.1016/j.landurbplan.2013.11.016 Kealy, P. S., & Hook, S. J. (1993). Separating temperature and emissivity in thermal infrared multispectral scanner data: implications for recovering land surface temperature. IEEE Transactions on Geoscience and Remote Sensing, 31(6), 1155e1164. Kroll, F., Müller, F., Haase, D., & Fohrer, N. (2012). Ruraleurban gradient analysis of ecosystem services supply and demand dynamics. Land Use Policy, 521e535. Elsevier Ltd, http://dx.doi.org/10.1016/j.landusepol.2011.07.008. Larondelle, N., & Haase, D. (2013). Urban ecosystem services assessment along a ruraleurban gradient: a cross-analysis of European cities. Ecological Indicators, 29, 179e190. Liu, J., Li, H. T., & Gu, H. Y. (2011). Study of color balance for remote sensing imagery mosaic. In 2011 International symposium on image and data fusion (pp. 1e4). http://dx.doi.org/10.1109/ISIDF.2011.6024299. MacFaden, S. W., Jarlath, P. M., O’Neil-Dunne, J. P. M., Royar, A. R., Lu, J. W. T., & Rundle, A. G. (2012 September 12). High-Resolution tree canopy mapping for New York City using LIDAR and object-based image analysis. Journal of Applied Remote Sensing, 6(1), 063567. http://dx.doi.org/10.1117/1.JRS.6.063567. http:// remotesensing.spiedigitallibrary.org/article.aspx?articleid¼1358118. Marquez, L. O., & Smith, N. C. (1999). A framework for linking urban form and air quality. Environmental Modelling & Software, 14(6), 541e548. http://dx.doi.org/ 10.1016/S1364-8152(99)00018-3. McDonnell, M. J., & Hahs, A. K. (2009). Comparative ecology of cities and towns: past, present and future. In M. J. McDonnell, A. K. Hahs, & J. Breuste (Eds.), Ecology of cities and towns: A comparative approach (pp. 71e89). Cambridge, UK: Cambridge University Press. McPhearson, T., Kremer, P., & Hamstead, Z. A. (2013). Mapping ecosystem services in New York City: applying a socialeecological approach in urban vacant land. Ecosystem Services, 5(2), 11e26. http://dx.doi.org/10.1016/j.ecoser.2013.06.005. McPhearson, T., Maddox, D., Gunther, B., & Bragdon, D. (2013). New York City biodiversity, green space, and ecosystem services. In T. Elmqvist, et al. (Eds.), Cities and biodiversity outlook: Urbanization, biodiversity and ecosystem services: Challenges and opportunities (pp. 355e383). McPherson, E. G., Nowak, D., Heisler, G., Grimmond, S., Souch, C., Grant, R., et al. (1997). Quantifying urban forest structure, function, and value: the Chicago urban forest climate project. Urban Ecosystems, 1, 49e61. Meyer, W. B., & Turner, B. L., II (1992). Human population growth and global landuse/cover change. Annual Review of Ecology and Systematics, 23, 39e61. Müller, N., Werner, P., & Kelcey, J. G. (Eds.). (2010). Urban biodiversity and design. Wiley-Blac. €, J., Kotze, J., & Yli-Pelkonen, V. (2009). Comparative urban ecology: chalNiemela lenges and possibilities. In M. J. McDonnell, A. K. Hahs, & J. Breuste (Eds.), Ecology of cities and towns: A comparative approach 9 (pp. 9e24). Oxford, UK: Oxford University Press. NYC Parks and Recreation. (2010). NYC open data: landcover raster data. Retrieved from https://nycopendata.socrata.com/Environmental-Sustainability/ Landcover-Raster-Data-2010-/9auy-76zt. NYC Department of City Planning. (2011). MapPLUTO. New York City: NYC Department of City Planning. Pataki, D. E., Carreiro, M. M., Cherrier, J., Grulke, N. E., Jennings, V., Pincetl, S., et al. (2011). Coupling biogeochemical cycles in urban environments: ecosystem

N. Larondelle et al. / Applied Geography 53 (2014) 427e437 services, green solutions, and misconceptions. Frontiers in Ecology and the Environment, 9(1), 27e36. http://dx.doi.org/10.1890/090220. Pauleit, S., & Duhme, F. (2000). Assessing the environmental performance of land cover types for urban planning. Landscape and Urban Planning, 52, 1e20. Peters, K., Elands, B., & Buijs, A. (2010). Social interactions in urban parks: stimulating social cohesion? Urban Forestry & Urban Greening, 9(2), 93e100. http:// dx.doi.org/10.1016/j.ufug.2009.11.003. Schwarz. (2010). Urban form revisiteddselecting indicators for characterising European cities. Landscape and Urban Planning, 96(1), 29e47. http://dx.doi.org/ 10.1016/j.landurbplan.2010.01.007. Schwarz, N., Bauer, A., & Haase, D. (2011). Assessing climate impacts of planning policiesdan estimation for the urban region of Leipzig (Germany). Environmental Impact Assessment Review, 31(2), 97e111. http://dx.doi.org/10.1016/j.eiar.2010.02.002. Senatsverwaltung für Stadtentwicklung und Umwelt. (2012). Population prognosis for Berlin and districts 2011e2030 (p. 23). Berlin. Senatsverwaltung für Stadtentwicklung und Umwelt. (2013a). 06.06 Einwohnerdichte (Ausgabe 2013). Berlin. Senatsverwaltung für Stadtentwicklung und Umwelt. (2013b). Stadtentwicklungskonzept Berlin 2030 Berlin heute: Der Statusbericht kompakt Ein Stadtentwicklungskonzept für 2030. Seymour, M., Wolch, J., Reynolds, K. D., & Bradbury, H. (2010). Resident perceptions of urban alleys and alley greening. Applied Geography, 30(3), 380e393. http:// dx.doi.org/10.1016/j.apgeog.2009.11.002. Stathopoulos, T., Wu, H., & Zacharias, J. (2004). Outdoor human comfort in an urban climate. Building and Environment, 39(3), 297e305. http://dx.doi.org/10.1016/ j.buildenv.2003.09.001.

437

Stewart, I. D., & Oke, T. R. (2012). Local climate zones for urban temperature studies. Bulletin of the American Meteorological Society, 93(12), 1879e1900. http:// dx.doi.org/10.1175/BAMS-D-11-00019.1. Stone, B., & Rodgers, M. O. (2001). Urban form and thermal efficiency: how the design of cities influences the urban heat island effect. Journal of the American Planning Association, 67(2), 186e198. http://dx.doi.org/10.1080/01944360108976228. Tratalos, J., Fuller, R. A., Warren, P. H., Davies, R. G., & Gaston, K. J. (2007). Urban form, biodiversity potential and ecosystem services. Landscape and Urban Planning, 83, 308e317. U.S. Census Bureau. (2012). 2012 Population estimates. American Factfinder. Retrieved 14.01.14, from http://factfinder2.census.gov/faces/tableservices/jsf/ pages/productview.xhtml?src¼bkmk. USGS. (2013). Filling the gaps to use in scientific analysis. Retrieved from http:// landsat.usgs.gov/sci_an.php. Velarde, M. D., Fry, G., & Tveit, M. (2007). Health effects of viewing landscapes e landscape types in environmental psychology. Urban Forestry & Urban Greening, 6(4), 199e212. http://dx.doi.org/10.1016/j.ufug.2007.07.001. €lker, S., & Kistemann, T. (2013). Reprint of: “I’m always entirely happy when I'm Vo here!” Urban blue enhancing human health and well-being in Cologne and Düsseldorf, Germany. Social Science & Medicine (1982), 91, 141e152. Elsevier Ltd. http://dx.doi.org/10.1016/j.socscimed.2013.04.016 West, S. T., Shores, K., & Mudd, L. (2012). Association of available parkland, physical activity, and overweight in America's largest cities. Journal of Public Health Management and Practice, 18(5), 423e430. Yu, C., & Hien, W. N. (2006). Thermal benefits of city parks. Energy and Buildings, 38(2), 105e120. http://dx.doi.org/10.1016/j.enbuild.2005.04.003.