Navigation – Plan du site

AccueilRubriquesAménagement, Urbanisme2017Using the Urban Atlas dataset for...

2017
815

Using the Urban Atlas dataset for estimating spatial metrics. Methodology and application in urban areas of Greece

Utilisation de données Atlas Urbain pour l'estimation des métriques spatiales. Méthodologie et application pour les villes grecques
Poulicos Prastacos, Apostolos Lagarias et Nektarios Chrysoulakis

Résumés

Cet article propose d’utiliser la base de données Atlas Urbain pour estimer des métriques spatiales permettant d’analyser la morphologie urbaine. Un processus de généralisation cartographique permet la transformation des données originales en fusionnant des unités voisines appartenant à la même classe et donc de créer de plus grandes unités d’occupation du sol. La base qui en résulte est utilisée pour estimer les métriques spatiales et la dimension fractale au niveau de la classe et du paysage pour neuf villes grecques. Les métriques estimées avec cette procédure reflètent la distribution des utilisations du sol en dépassant le problème posé par la haute résolution de données de l'Atlas Urbain (dans lequel chaque unité spatiale est un îlot urbain). Les indicateurs sont utilisés pour analyser et comparer la forme des différentes villes. Les résultats montrent que ces mesures spatiales et en particulier les métriques de classe, permettent de quantifier l'hétérogénéité spatiale et la fragmentation de l'urbanisation, ainsi que la structure et la composition des utilisations des sols, contribuant ainsi à caractériser les morphologies de chaque ville.

Haut de page

Texte intégral

This work was performed in the framework of the PEFYKA project within the KRIPIS Action of the GSRT. The project was funded by the Greek government and the European Regional Development Fund of the European Union under the NSRF and the O.P. Competitiveness and Entrepreneurship.
The authors would like to acknowledge the useful comments from two anonymous referees.

Introduction

1Comparative analysis of the form of European metropolitan areas is a research issue that has preoccupied researchers and planners for a long time. Lack of consistent and detailed land use/cover datasets, however, has impeded research in the field. The common approach has been to develop land use maps for urban areas of about 30 m resolution using Landsat images. However, these are usually developed for individual cities, often following ad hoc land use classification schemes thus making difficult comparisons across cities. Until recently, the CORINE Land Cover database (Bossard et al., 2000; Büttner et al., 2004) that covers all Europe was the only consistent source of land use maps. Released in 1990 and updated for 2000, 2006 and 2012 it has a resolution of 100 meters and land uses are classified into 44 different classes. However, only 2 of these classes refer to the density of the built up areas, which is the focus of analysis when studying urban areas.

2Understanding the need for an in depth analysis of European cities, the European Environment Agency (EEA) through the Global Monitoring for Environment and Security (GMES) program released Urban Atlas (UA), a freely available dataset with land use information for year 2006 for almost all cities in European Union with at least 100,000 population (EEA, 2010). There are 20 land use classes, 6 of which represent different built-up density levels. EEA is in the process of releasing a new version of UA with data for year 2012 (http://land.copernicus.eu/​local/​urban-atlas/​urban-atlas-2012).

3With the UA dataset, European cities can be analyzed in many different ways. One way is to estimate percentage coverages for different land use classes (Gisat, 2012). However, this approach does not consider the aggregation/dispersion/proximity patterns of land uses. These could be revealed through the analysis of spatial metrics. Spatial metrics originally introduced in landscape ecology (Turner, 1989) are indicators estimated from a patch based representation of the landscape. Patches are areas that are homogenous with respect some characteristic such as land use class, habitat type etc. Spatial metrics are quantitative and aggregate in nature. According to Herold et al. (2005), spatial metrics can be defined as indicators obtained from the analysis of thematic-categorical maps exhibiting spatial heterogeneity at a specific scale and resolution, and bring out the spatial component of the urban form and the dynamics of change and growth processes.

4In the last 15 years spatial metrics are used extensively as indicators for describing urban form. The typical application estimates metrics for different time periods to assess land use changes (Herold et al. 2002; Seto and Fragkias, 2005; Ji et al., 2006; Weng, 2007; Skupinski et. al, 2009; Zhao and Murayama, 2011; Pham et al., 2012; Ramachandra et al., 2012). Metrics are also used to illustrate changes in urban form (Dietzel et al., 2005a, Aguilera-Benavente et al. 2014), to assess the goodness-of-fit of urban growth models (Herold et al., 2003; Dietzel et al., 2005b; Aguilera et al., 2011) and to compare the form of different cities/regions (Huang, 2007; Schwarz, 2010; Triantakonstantis and Stathakis, 2015).

5Spatial metrics is not the only way to analyse urban form. Guérois and Pumain (2008) compare 40 European cities using linear functions that describe the gradient of built-up surfaces. Based on methodological concepts developed by Frankhauser (2004), fractal analysis has been also used to determine urban form; Thomas et al. (2012) analyze 18 European cities, while Tannier and Thomas (2013) compare Belgian cities.

6Estimating metrics with the UA dataset presents some methodological issues that must be addressed in order to have a reliable and unambiguous city comparison. The UA maps are of high resolution and every city block is a separate land use patch. Metrics estimated with these maps (Prastacos et al., 2012) are affected by both the patterns of land use distribution and the organization of city blocks. The continuity of land use distribution is not evident since neighboring patches of the same type are city blocks separated by a road, nonetheless they are of the same class. Additionally, UA contains information for an extensive number of classes, some of which represent a very small portion of the urban area. To overcome these problems, the original land use classes are aggregated to a smaller number of classes and neighboring patches of the same type are merged to form larger contiguous areas. With this approach the emphasis is on land use continuity rather than on the distribution of a class in many smaller patches.

7In this paper, a methodology for computing metrics using the UA maps is presented and then used for the estimation of spatial metrics in nine Greek cities. Greek cities are monocentric, with high population density and parts of them have developed without strict adherence to planning regulations. There are relatively few quantitative studies of their form (Lagarias 2007; Prastacos et al, 2012; Tsilimingas, 2014; Salvati 2014). In this study the different metrics are analyzed and compared across classes and cities. The analysis demonstrates that the metrics describe urban form in a parsimonious way and highlight the differences that exist.

8The paper consists of six different sections. In the next section there is a discussion of the UA dataset and the issues that must be considered when using this dataset for estimating metrics. In the following section the methodology for computing metrics from the UA is elaborated and then applied for the estimation of metrics for nine Greek cities. The metrics are analysed and compared in the next section, while in the following section there is a brief overview of the form of each city. The findings of the paper are summarized in the conclusions section.

Spatial metrics and Urban Atlas

Land use classes in UA

9The land use classification system in UA identifies 20 different land use classes, 17 are ‘artificial surfaces’ that is, developed/built-up areas, and 3 non developed/natural areas. Six artificial surfaces classes, the ‘urban fabric’, describe built up/density levels, they could be therefore considered land cover rather than land use classes. There are 5 different classes for transport infrastructure (fast transit roads, other roads, railroads, ports and airports) and 6 classes for other uses (industrial/commercial/public facilities, mineral extraction/dump sites, construction areas, land without use, green urban areas and sports/leisure facilities). The database is in vector format (every land use entity is a polygon), has been developed from satellite images of 2006±1 year and map scale is 1:10.000. Minimum mapping unit is 0.25 ha (50x50 m). Land use data are available for 305 urban areas in EU participating in the Urban Audit program. The Urban Audit, recently renamed as ‘City Statistics’ (http://ec.europa.eu/​eurostat/​web/​cities/​overview), is a Eurostat program through which a range of statistical information and indicators are published on a regular basis for cities in the EU, Norway, Switzerland and Turkey. In Urban Atlas data are available for almost all cities in EU with population exceeding 100,000, but also some with less population.

10The patches of the 6 urban fabric classes are delineated on the basis of the imperviousness/soil sealing degree. Soil sealing measures the loss of soil resources because of the coverage of land by housing, roads or other construction and takes values from 0% to 100% (100% representing fully built up). The patches of the 6 urban fabric classes in UA were identified using the FTS high resolution soil sealing layer (Maucha et al., 2010). Urban fabric areas were first classified as continuous (sealing degree >80%) and discontinuous (sealing degree <80%). The latter were further broken down to four separate classes (dense, medium density, low density, very low density) with 50%, 30% and 10% being the soil sealing degrees cut-off points. ‘Isolated buildings’ are classified as a separate urban fabric class. The high resolution of the map, the relatively small size of the mapping unit and the procedure of identifying urban fabric classes on the basis of the sealing degree result in a database that provides detailed information on how densities change through space, both in the city core and in the periphery where densities are lower.

11Urban areas boundaries in UA are specified as in Urban Audit, with cities defined by considering the Larger Urban Zone (LUZ). The LUZ represents the functional urban area (http://ec.europa.eu/​eurostat/​web/​cities/​spatial-units), not just the core city and often covers an area significantly larger than what is usually considered a metropolitan area. Therefore, based on the UA data in most cities the artificial/developed land accounts for less than 30% of total land (often just 10%-20%) while the remaining is classified as natural areas (agricultural, forests etc.). For example, in Athens 74% of total area is classified as Agricultural land/forests and only 13% as urban fabric and another 4% as industrial areas (Table 1). This is because the boundaries of ‘Athens’ in UA are the administrative boundaries of the Attica region, an area by far exceeding what is referred to as Athens Metropolitan Area.

Table 1: Land use classes distribution in Athens (UA)

Code

Land use category

Number of patches

Total Area (ha)

% Patches

% Area

% Perimeter

Mean Patch Area (ha)

Mean Patch Perimeter (m)

11100

Continuous Urban Fabric (S.D. > 80%)

28,562

10,833

28.8

3.6

8.1

0.4

270

11210

Discontinuous Dense Urban Fabric (S.D. 50% - 80%)

19,638

12,536

19.8

4.1

7.6

0.6

370

11220

Discontinuous Medium Density Urban Fabric (S.D. 30% - 50%)

11,260

9,554

11.4

3.1

5.3

0.8

450

11230

Discontinuous Low Density Urban Fabric (S.D. 10% - 30%)

7,428

7,492

7.5

2.5

3.9

1.0

496

11240

Discontinuous Very Low Density Urban Fabric (S.D. < 10%)

1,329

1,091

1.3

0.4

0.6

0.8

417

11300

Isolated Structures

4,378

2,034

4.4

0.7

1.4

0.5

296

12100

Industrial, commercial, public, military and private units

7,466

13,155

7.5

4.3

4.2

1.8

535

12210

Fast transit roads and associated land

80

665

0.1

0.2

0.3

8.3

3,792

12220

Other roads and associated land

697

14,103

0.7

4.6

40.3

20.2

55,022

12230

Railways, associated land

33

298

0.0

0.1

0.3

9.0

9,703

12300

Port areas

96

1,269

0.1

0.4

0.2

13.2

2,191

12400

Airports

5

1,293

0.0

0.4

0.0

258.6

8,673

13100

Mineral extraction and dump sites

191

1,460

0.2

0.5

0.2

7.6

1,125

13300

Construction sites

317

628

0.3

0.2

0.2

2.0

539

13400

Land without current use

1,324

665

1.3

0.2

0.5

0.5

331

14100

Green urban areas

1,682

2,796

1.7

0.9

1.0

1.7

561

14200

Sports and leisure facilities

689

1,876

0.7

0.6

0.5

2.7

694

20000

Agricultural + Semi-natural areas + Wetlands

11,547

191,873

11.7

63.1

20.7

16.6

1,709

30000

Forests

1,793

29,255

1.8

9.6

4.0

16.3

2,114

50000

Water bodies

549

1,347

0.6

0.4

0.7

2.5

1,177

Total

99,064

304,224

100.0

100.0

100.0

3.1

834

Spatial metrics

12Spatial metrics are indicators describing land use patterns in an urban area. They are defined as mathematical expressions of patch characteristics, such as the area, the perimeter, the geometry (shape), the relative position in the urban area and others. Class metrics are computed by analyzing patches of the same land use class and therefore describe characteristics of that class. Landscape metrics are estimated by analyzing patches of all land use classes, and therefore describe with one indicator an aspect of the land use patterns in the whole urban area.

13An extensive list of metrics has been proposed; the list includes simple metrics such as the percentage of a land use class in the urban area (PLAND) but also several mathematically complex metrics used as indicators of the geometry (shape) of the patches, of the average distance between patches, of the relative dispersion of a land use class and others. In most studies only a relatively small set of metrics is estimated since there are significant correlations among them. Commonly applied metrics are edge density (ED), nearest neighbour distance (ENN), size of the largest patch (LPI), fractal dimension (FRAC) and contagion (CONTAG) (Herold et al., 2003). Geoghegan et al. (1997) use Shannon’s Index (SHDI) as diversity index to measure land uses heterogeneity, perimeter-area (PARA) as fragmentation index, fractal dimension index (FRAC) to identify convoluted patch geometry and the contagion metric (CONTAG) to represent diversity of land use distribution. The literature concentrating on the analysis of urban areas using fractal theory has proposed the fractal dimension (FD) as the key indicator for describing urban areas.

14Overall, the trend has been on describing urban areas with landscape metrics rather than class metrics with the focus of the research being on the analysis of the developed/non /developed land dichotomy. Lack of datasets with multiple land uses classes has hampered the use of class metrics. Landscape metrics, however, do not bring forward the diversity of built-up densities that exist in an urban area since they represent averages of all land use classes and therefore provide a relatively aggregate description of urban form. This is the reason that in the following sections there is an extensive discussion of class metrics.

15The metrics are usually computed with Fragstats (McGarigal et al., 2012), a public domain software available since 1995 (current version 4.2). Fragstats contains a large library of metrics that can be estimated at the class and landscape level and has been used for almost all the spatial metrics studies referenced earlier in this paper. The Patch Analyst (Rempel et al., 2012), an add-on to the ArcGis software, can be also used for computing metrics, it does not, however, offer a large library of metrics. Researchers concentrating on fractal analysis of urban areas have used Fractalyse (http://www.fractalyse.org/​), another public domain software that estimates the fractal dimension. Research efforts relying on Fractalyse include Tannier and Pumain, (2005), Thomas et al. (2008a), Thomas et al. (2008b), Thomas et al. (2012) and others. Zaragozi et. al (2012) have proposed an open source library for estimating landscape metrics.

16Most applications of metrics in urban areas concentrate on the analysis of the urban growth trajectory comparing metrics for at least two time periods. The computation process consists of two steps; through satellite image classification (Landsat, ASTER, SPOT or other sensor) a raster land use map is produced and then it is used for estimating the metrics. Urban extent is defined on the basis of administrative boundaries or topographic landmarks or some ad hoc procedure (the boundaries of the study area). Map resolution is usually in the range of 30-50 meters with no specific standards placed on the minimum mapping unit.

17With the UA dataset the situation is different. Land use information is available as vector polygons with every city block identified by a land use class while roads between blocks are classified as a separate land use category (other roads). There is therefore a spatially detailed classification of land use distribution that follows the city blocks organization. However, land use continuity is not evident since the road network separates blocks of the same type and therefore some spatial metrics are not particularly useful. For example the Mean Patch Area (MPA) and the Largest Patch Index (LPI) metrics estimated with the UA map do not convey meaningful information since they reflect city block characteristics rather than those of a land use.

18To estimate metrics that consider land use distribution irrespective of the layout/ organization of city blocks, the UA map must be transformed. To reduce fragmentation, neighbouring blocks of the same land use must be aggregated into a single block. This can be accomplished by implementing a map generalization process; two polygons of the same type separated by a road are combined to form a single one, while the road that separates them is reclassified to the land use type of the merged polygons. An improved procedure is to consider the width of the road and allow only polygons separated by roads of width less than some threshold to merge. With an aggregation distance (road width threshold for merging polygons) of 10 or 20 meters many polygons will merge but the form of the city with respect the large roads/highways remains intact. If the aggregation distance is higher, 40 m for example, blocks of the same type that are on different sides of a freeway will merge into a single block and therefore the function of large roads as barriers/containers of neighbourhoods will not be recognized.

Methodology for estimating spatial metrics for the nine Greek cities

Land use reclassification

19Several of the land use classes identified in UA account for a very small percentage of the urban area and a detailed analysis would not provide a meaningful insight on the form of the city. The 20 UA classes were therefore reclassified into 8 classes (Table 2). Three of the 8 classes represent urban fabric areas. Transport infrastructure (fast roads, other roads, railroads, ports, airports) were aggregated into one class and the same was done for green areas and sports facilities.

Table 2: Land use classification for the estimation of spatial metrics

Class

Name

UA code

Sealing degree

C1

Continuous urban fabric areas

11100

80-100%

C2

Discontinuous dense urban fabric

11121

50-80%

C3

Discontinuous urban fabric

11220, 11230, 11240, 11300

< 50%

C4

Industrial/commercial areas

12100

C5

Transport infrastructure

12210, 12220, 12230, 12300,12400

C6

Mine/Dump sites, Construction/

Land without use

13100, 13300, 13400

C7

Green areas and sport facilities

14100, 14200

C8

Agriculture, Forest, Water

20000, 30000, 50000

20Spatial metrics were computed for the nine Greek cities for which data are available in UA (Table 3, Figure 1). Population (2011 Census) ranges from 69,849 (Kalamata) to 3,827,624 (Athens), and total area from 30,425 ha (Volos) to 304,223 ha (Athens). Artificial surfaces represent 27% of total area in Athens, while in other cities they account for less than 15% of the area. Population density (population divided by artificial surfaces) is 14 people/ha in Ioannina and 47 people/ha in Athens and Thessaloniki. The distribution of urban fabric among the three land use density classes (C1, C2 and C3) differs significantly between the various cities.

Table 3: Population, area, population density and urban fabric distribution for the nine cities in Greece

Population

(2011)

Area

(ha)

% Artif.

Population

density

Urban fabric distribution (%)

C1

C2

C3

Athens

3,827,624

304,223

27

47

25

29

46

Thessaloniki

960,302

142,581

14

47

34

44

22

Patras

217,504

51,263

11

39

16

29

54

Heraklion

211,370

60,444

10

34

12

23

64

Larissa

195,120

154,941

8

17

26

42

31

Volos

136,936

30,425

14

32

32

27

41

Ioannina

132,979

132,526

7

14

13

45

42

Kavala

70,501

35,161

7

28

29

35

36

Kalamata

69,849

44,171

6

25

15

20

65

Figure 1: Land use maps of the nine Greek cities (Legend of classes: see table 2)

Figure 1: Land use maps of the nine Greek cities (Legend of classes: see table 2)

Polygon aggregation/map generalization

21The UA database consists of polygons that correspond to city blocks. Applying the ‘Aggregate polygons’ procedure provided in ESRI’s ArcGis software (a procedure that combines polygons within a specified distance of each other into new polygons https://pro.arcgis.com/​en/​pro-app/​tool-reference/​cartography/​aggregate-polygons.htm) the original UA map was transformed into a map in which neighbouring blocks of the same land use class are aggregated into a single polygon. A 20 m aggregation distance was assumed and same class polygons on the sides of roads less than 20 m wide merged and the road segment between them was reclassified to that class, thus forming a single larger patch. Other parameters of the procedure were defined in such a way as to maintain all block holes, blocks of a different land use type than the surrounding, irrespective of their size. With the aggregation procedure the number of polygons is reduced drastically since adjacent polygons of the same class merge together. In Athens, the number of continuous fabric (C1) polygons decreases from 28,562 to 1,921 while the associated area increases from 10,833 ha to 12,697 ha since the road network between merged polygons is reclassified as C1 class. Mean polygon area (MPA) increases from 0.4 ha (a city block of about 60 x 60 m) to 6.6 ha (250 x 250 m). Maps of an area in Athens before and after polygon aggregation are displayed in Figure 2.

Figure 2: Maps for a subarea of Athens, before and after polygon aggregation (Legend of classes: see table 2)

Figure 2: Maps for a subarea of Athens, before and after polygon aggregation (Legend of classes: see table 2)

22Since Fragstats 4.2 accepts as input only raster files, the vector map obtained after polygon aggregation was rasterized at a resolution of 20 meters per pixel. The rasterization process affected also to a small degree the number of patches. In the following discussions the vector data of the original UA database are referred to as ‘polygons’, while the data used by Fragstats software, the transformed and rasterized data, are referred to as ‘patches’.

23The map generalization procedure affects both the Number of Patches (NP) and the area of the three urban fabric classes. After aggregation and rasterization the NP of C1, C2 and C3 classes represent respectively only 6%, 26% and 34% of the original polygons while class area increases by 17%, 8% and 6% respectively. That is, as built-up density (measured by the sealing degree) decreases, the merging rate of polygons decreases. The high compaction rate of continuous urban fabric (C1) polygons implies that these polygons are clustered together, usually in the older parts of the urban area where roads are narrower. The merging rate of discontinuous medium and low density class (C3) polygons is the lowest, however it is not very different from that of C2 although this class includes the ‘isolated buildings’ which usually are not affected by the aggregation procedure (18% of C3 polygons in Athens are classified as ‘isolated buildings’).

24To assess how road width affects polygon aggregation, additional runs were performed for Athens and Patras using 10 m and 40 m as aggregation distance. As shown in Figure 3 the number of polygons declines drastically with an aggregation distance of 10 m, while for larger road widths the marginal decrease is smaller. This is true in both cities and for all three urban fabric classes, with the decrease in the number of C1 polygons’ being the steepest. In Patras, a much smaller city, the curve is smoother. The reduction in the number of polygons leads to significantly larger mean patch area (Figure 4).

Figure 3: Number of polygons after aggregation for different road widths

Figure 3: Number of polygons after aggregation for different road widths

Figure 4: Mean polygon area after aggregation for different road widths

Figure 4: Mean polygon area after aggregation for different road widths

Metrics

25The metrics for analyzing the Greek cities are tabulated in Table 4. Class metrics computed include NP, MPA, MPE, PD, ED, LPI, ENN_MN and SHAPE; landscape metrics used were PD, ED, SHAPE, CONTAG, SHDI and FD. Metrics were estimated with the Fragstats software version 4.2 (McGarigal et al., 2012). The PD, ED and LPI metrics were computed on the basis of artificial surfaces rather than total area as implemented in Fragstats. This was necessary because artificial surfaces in the nine cities represents between 6.2% and 26.9% of the total area, therefore, values of these metrics are affected by the urban area boundaries assumed and comparison of urban form of different cities might be biased. The fractal dimension (FD) is estimated with the Fractalyse software, since in this software the indicator is defined in a way that is consistent with fractal theory. Fragstats also estimates the Fractal dimension index (FRAC), however, the approach used is not fully compatible with fractal theory.

Table 4: Spatial metrics description

Name

Value ranges

Description

Number of Patches (NP)

NP > 0

The number of patches of a land use class

Mean Patch Area (MPA)

MPA > 0

The area of all patches of a land use class divided by the respective number of patches

Mean Patch Edge (MPE)

MPE > 0

The edge (perimeter) of all patches of a class divided by the respective number of patches

Patch Density (PD)

PD > 0

The number of patches of a land use class divided by the artificial surfaces area

Edge Density (ED)

ED > 0

The total edge of all patches of a land use class divided by the artificial surfaces area

Largest Patch Index  (LPI)

0 < LPI ≤ 100

The area of the largest patch of a land use class divided by the artificial surfaces area

Shape Index (SHAPE)

SHAPE > 1

The patch perimeter divided by the square root of patch area; a measure of the shape/geometry complexity of the patches of a land use class

Euclidean Nearest-Neighbor Distance (ENN)


ENN > 0

The mean distance to the nearest neighboring patch of the same class, based on shortest edge-to-edge distance from cell center to cell center

Contagion (CONTAG)


0 < CONTAG ≤ 100

Measures the overall probability that a cell (pixel) of a land use type is adjacent to cells of the same type

Shannon's Diversity Index (SHDI)

SHDI ≥ 0

Measures the diversity of the distribution of patches in the urban area

Fractal dimension

1 ≤ FD ≤ 2

A measure of the complexity of urban areas with respect to how built-up patterns change across different scales

26‘Agricultural land, water areas and forest land’ (C8) patches were not considered when estimating landscape metrics since the characteristics of this class do not describe the form of built-up areas; also, because of their large size, they would affect the results. This was implemented using a feature of Fragstats that permits to define a class as ‘background’ and be excluded when estimating landscape metrics.

Results

27The following discussion concentrates on the urban fabric (C1, C2, C3), commercial/industrial (C4) and green areas/sports facilities (C7) metrics. The transport infrastructure class represents a significant portion of the urban area -- in Athens roads account for 17% of artificial area--, however, since some roads were reclassified to another land use the metrics computed with the remaining road network do not provide a valid picture of the transport network. Additionally, the large patches of airports and ports cannot be used for any meaningful analysis. Class C6 (other land uses) metrics are also not reported, since they represent a small percentage of the artificial area (less than 4% in Athens) and do not convey information useful for describing urban form.

28The number of patches (NP), the mean patch area (MPA), the coefficient of variation of the size of the patches (CVarea, ratio of standard deviation to MPA) and the mean perimeter/edge (MPE) metrics (Table 5) provide some first information on the urban form which is further refined with the other metrics. In every city, as built-up density decreases (sealing degree 80-100%, 50-80%, > 50%), the NP increase while MPA and MPE decrease. C1 patches are the largest (highest MPA) and exhibit the largest variation in size (measured by the coefficient of variation CVarea). In all urban areas there are a few very large C1 patches while most patches are much smaller (median size is 0.65 ha in Athens). The largest MPA is in Volos (10.5 ha), while in Athens and Thessaloniki MPA is 7.4 ha and 10.2 ha respectively. The CVarea of C1 patches takes the highest values in the large cities of Athens and Thessaloniki. These urban form characteristics arise from the polygon aggregation procedure that merges polygons of the original UA data to form larger contiguous areas. Had this procedure not been implemented the differences in the MPAs of the three urban fabric classes would be marginal. In Athens MPAs for the three classes would be respectively 0.4 ha, 0.6 ha and 0.8 ha (Table 1) while with the aggregation procedure they are 7.4 ha, 2.8 ha and 2.6 ha. The former figures account for the land use distribution and the organization of the city blocks, the city layout, while the latter describe the spatial organization of land uses. MPA of C2 patches and the variation of their size are for most cities larger than those of C3. The MPA for industrial/ commercial activity (C4) ranges between 0.9 ha and 2.8 ha (Kalamata –Athens).

Table 5: NP, MPA and MPE metrics for selected land use classes

Class

NP

MPA

Coeff.

Var.(patch area)

MPE

NP

MPA

Coeff.

Var.(patch area)

MPE

Athens

 

 

 

Volos

 

 

 

C1

1,712

7.4

11.6

1,577

74

10.5

4.7

1,935

C2

5,040

2.7

5.7

975

262

2.2

2.6

909

C3

8,238

2.6

9.8

898

614

1.4

4.0

600

C4

4,825

2.8

8.6

779

510

2.0

7.1

633

C7

1,524

3.1

4.5

907

80

1.5

1.3

689

Thessaloniki

 

 

Ioannina

 

 

 

C1

358

10.2

9.3

1,942

158

3.1

5.8

943

C2

1,497

2.8

3.1

1,022

609

2.8

2.4

1,018

C3

2,228

0.9

1.7

495

1,337

1.2

1.8

570

C4

2,341

2.3

5.2

702

2,385

0.9

3.5

477

C7

480

2.1

1.7

800

106

2.3

1.9

782

Patras

 

 

 

Kavala

 

 

 

C1

182

2.9

8.6

953

43

8.6

4.6

1,704

C2

513

1.7

2.3

825

110

3.7

2.6

1,255

C3

969

1.6

3.2

752

317

1.3

3.3

541

C4

526

1.6

2.3

627

192

2.4

2.8

753

C7

310

1.1

1.7

571

55

1.6

1.6

708

Heraklion

 

 

 

Kalamata

 

 

 

C1

75

6.5

5.7

1,480

62

3.6

3.6

976

C2

189

4.4

3.8

1,363

202

1.3

1.8

635

C3

1,000

2.2

3.5

840

685

1.2

2.3

658

C4

473

2.5

3.7

738

328

1.5

2.4

611

C7

59

2.6

2.1

783

51

1.5

1.0

750

Larissa

 

 

 

C1

304

3.9

4.4

1,140

C2

544

3.3

3.0

1,110

C3

1,333

1.0

2.7

483

C4

943

2.5

3.0

773

C7

145

1.6

3.3

677

Class metrics

29The estimated class metrics are presented in Table 6 and discussed in this section.

30PD (Patch Density): PD is a measure of the fragmentation/ spatial distribution of the patches of a land use class. Low PD values imply fewer patches and indicate land use continuity, whereas higher values denote more patches, spatial dispersion and discontinuity. For the three urban fabric classes the PD metric increases as built-up densities (sealing degree) decrease. Highest values are attained for C3 indicating fragmentation. PD metrics are inversely related to the MPA metrics discussed in the previous section, MPA is the smallest for C3 and highest for C1. Differences of PD metrics across urban areas are relatively marginal. The C1 PD metric takes values between 1.2 and 3.3 patches per 100 hectares, while for C2 the range of values is wider (3.1 to 9.3). The lowest PDs for both classes are in Heraklion. The range of PD values of class C3 is more wide (values between 10.0 and 25.0), however, in most cities the PD metric is between 10 and 15. Lowest PD values, indication that patch distribution is less fragmented, are in Athens, Thessaloniki and Larissa. For class C4 (industrial/ commercial) the lowest PDs are in Athens, Heraklion, Patras and Larissa. For Green/sport facilities areas (C7) the PD for most cities is between 0.9 and 1.9, whereas for Patras it is 5.6 indication of the wide distribution of recreation areas in this city. PD metrics are affected by the polygon aggregation procedure. In the original UA map the PD of the three urban fabric classes in Athens are respectively 35, 24 and 30 patches per 100 ha, while after the polygon aggregation/rasterization procedures the PD metrics change significantly and are respectively 2.1, 6.2 and 10.1 patches per 100 ha.

Table 6: Class metrics for selected land use classes

Class

PD*

ED*

LPI*

SHAPE_AM

ENN_MN

PD*

ED*

LPI*

SHAPE_AM

ENN_ MN

Athens

Volos

C1

2.1

33

2.4

9.5

189

1.8

34

9.1

4.7

158

C2

6.2

60

0.7

4.3

128

6.2

56

1.6

2.8

124

C3

10.1

91

1.9

5.8

124

14.6

87

1.8

2.7

146

C4

5.9

46

1.6

3.1

191

12.1

77

7.4

2.2

209

C7

1.9

17

0.3

2.7

297

1.9

13

0.3

1.7

460

Thessaloniki

Ioannina

C1

1.8

34

7.2

9.1

357

1.7

16

2.4

4.0

495

C2

7,4

75

1.0

2.8

112

6.5

66

0.8

2.7

202

C3

11.0

54

0.1

1.6

164

14.3

81

0.3

1.8

192

C4

11.5

81

1.3

2.5

184

25.4

121

1.0

1.6

174

C7

1.8

14

0.1

1.8

409

1.1

9

0.3

1.7

999

Patras

Kavala

C1

3.3

31

6.1

7.3

223

1.7

29

10.3

4.5

374

C2

9.3

77

0.6

2.6

160

4.4

55

3.0

3.0

246

C3

17.6

132

1.8

2.7

148

12.7

69

1.8

2.2

236

C4

9.5

60

0.7

1.8

220

7.7

58

2.5

1.8

224

C7

5.6

32

0.4

1.7

198

2.2

16

0.7

1.8

398

Heraklion

Kalamata

C1

1.2

18

5.1

6.0

407

2.2

22

3.2

3.0

266

C2

3.1

42

3.5

3.6

409

7.3

46

0.6

2.0

184

C3

16,2

136

2.9

2.9

156

24.8

163

1.4

2.4

112

C4

7,7

57

2.9

1.8

239

11.9

73

1.4

1.9

246

C7

0.9

7

0.5

1.5

1128

1.9

14

0.2

1.9

605

Larissa

C1

2,6

30

2.2

3.6

282

C2

4,6

52

1.1

3.0

129

C3

11.4

55

0.6

1.8

275

C4

8.1

62

1.0

1.9

263

C7

1,2

8

0.5

2.8

753

* The PD, ED and LPI metrics are estimated on the basis of the artificial land.

31ED (Edge Density): The ED metric addresses the spatial distribution of a land use by considering the size and the complexity of the shape/geometry of the patches. Low values imply fewer patches and compact shape, whereas large values denote many patches with simpler shapes. In all cities ED metrics, similarly to PD metrics, are the lowest for C1. With neighboring polygons merging there are fewer C1 patches more compact and therefore total edge/perimeter is reduced. ED values for C1 range from 18 to 34 m/ha (lowest in Ioannina/Heraklion, highest in Volos/ Athens/ Thessaloniki). C2 and C3 ED metrics are higher (C2 ED values range from 42 to 76 m/ha, Heraklion lowest, Patras/Thessaloniki higher while for C3 values range from 54 to 163 m/ha with the lowest value in Larissa and the highest in Kalamata, Heraklion and Patras). Correlations between the PD and ED metrics are significant and depend on the land use class (0.81, 0.88 and 0.98 respectively for classes C2, C3 and C4). Τhe correlation of PD and ED metrics of C1 is lower (0.38) and not significant since with polygons merging the reduction in total edge is not necessarily related to the reduction of the NPs.

32LPI (Largest Patch index): LPI, the percent of the area of artificial surfaces the largest patch represents, is a measure of the importance of the largest patch. In every urban area the C1 LPI metric is significantly higher than that of C2 and C3 classes denoting the relatively larger size of the largest C1 patch. The highest C1 LPI metrics are in Kavala, Volos and Thessaloniki (10.3, 9.1 and 7.2 respectively), while in Athens. a large metropolitan area, LPI is significantly lower (2.4). For industrial/commercial areas (C4) the highest LPIs are in Volos and Heraklion (7.4 and 2.9 respectively) and are attributed to the large industrial area in the periphery of these cities. In Athens the C4 LPI is 1.6, whereas in other cities it takes lower values.

33A different way to assess the significance of the largest patch is to estimate the percentage of class area it represents (Table 7). A similar approach was used by Herold et. al (2003). In every urban area the largest C1 patch accounts for a significant portion of the C1 area. In Athens, it is 16% of the C1 area, but in several cities it accounts for more than 50%. Further analysis in Athens shows that the four largest C1 patches, all four in close proximity to the downtown area account for 54% of the C1 area. For C2 and C3 land use classes the domination of the largest patch is much smaller; Heraklion is an exception to this rule, the largest C2 patch accounts for 26% of the C2 area. The results demonstrate that Greek cities are relativaly monocentric with large patches of continuous urban fabric in the downtown area dominating the urban area.

Table 7: Area of the largest patch as a percentage (%) of the class area

C1

C2

C3

C4

C7

Athens

16

4

7

10

5

Thessaloniki

40

5

1

5

4

Patras

64

4

6

5

6

Heraklion

64

26

8

15

19

Larissa

22

7

5

5

26

Volos

50

12

9

31

9

Ioannina

45

5

2

4

13

Kavala

70

19

11

14

19

Kalamata

40

6

5

8

9

34ENN_MN (Euclidean Nearest-Neighbor Distance): ENN is a measure of the segregation/spatial separation of the patches of a land use class. Low values imply that patches are relatively close to each other, whereas high values imply spatial separation. In the nine Greek cities, C1 ENN is higher than the ENN of the other two urban fabric classes since as discussed above there are large C1 patches in the downtown area while the remaining patches of this class are interspersed in the urban area. ENN metrics for C2 are significantly smaller and increase for C3, probably because this class includes the low density built-up areas and the ‘isolated structures’. These results conform to urban economic theory and the way cities evolve through time with high density areas acting as centers of activity and built-up densities decreasing around them. C1 patches in Athens are closer together (lower ENN) than in other cities. In less populous cities there are relatively fewer C1 patches and usually at higher distance between them; there is a large patch in the downtown area and the C1 patches in periurban areas are concentrated in few areas and at a distance. The highest ENN values for both C1 and C2 are in Heraklion and are about the same, that is patches of these two classes exhibit similar patterns of spatial separation/coherence. ENN metrics for C4 are relatively similar for all cities ranging between 200 to 250 meters. A comparative view of the PD, ED and ENN metrics for the four largest cities appears in Figure 5.

Figure 5: The PD, ED and ENN class metrics*

Figure 5: The PD, ED and ENN class metrics*

* To be displayed in the same graph, ED metrics are divided by 10 and ENN by 100

35SHAPE (Shape Index); the SHAPE metric is estimated through a mathematical expression that considers patch perimeter divided by patch area and is an indicator of the complexity of the shape/geometry of the patches. It can be evaluated as area weighted (SHAPE_AM) that is each patch weighted by its area or without considering the patch size (SHAPE_MN). Values close to 1 denote simple shapes while higher values indicate convoluted shapes. Differences between the C1 and C2 SHAPE_MN metrics are marginal while the C3 metric is significantly smaller, implying simpler geometry (SHAPE_MN values are not reported). For the SHAPE_AM indicator highest values are always recorded for C1 since because of the polygon aggrgation the large C1 patches are more complex/irregular in shape than those of C2 and C3. The highest values are in Athens and Thessaloniki, 9.5 and 9.1 respectively. In most cities, the SHAPE_AM metric of C2 is larger than the equivalent metric for C3, although this is not true for Athens (4.3 and 5.8 respectively for C2 and C3). The explanation for the lower C2 SHAPE_AM metric in Athens might be that since it is a large city, the ‘merged’ C2 patches are more orthogonal than those of class C3 which are at the outskirts of the urban area. This is also evident from the coefficient of variation of the patch size (5.7 for C2 versus 9.5 for C3).

36Landscape metrics
The landscape metrics of the nine Greek cities are presented in Table 8 and discussed below.

37CONTAG (Contagion); this metric considers adjacencies of similar land use types. It is computed at the pixel level (raster cell), and not at the patch level as most metrics, and takes values between 0 and 100. Low levels of patch dispersion (high proportion of like adjacencies) and interspersion (inequitable distribution of pairwise adjacencies) result in high contagion, and vice versa. Therefore in urban areas consisting of large, contiguous patches the index takes high values, whereas in areas that patches are smaller and dispersed CONTAG is lower.

38SHDI (Shannon's Diversity Index); is an indicator of the diversity of land use distribution. It is estimated using Shannon’s information content function and takes a value of 0 when there is only one patch (no diversity) and its value increases when the proportional distribution of the urban area among different patch types is more equitable.

Table 8: Landscape metrics for the nine Greek cities

City

CONTAG

SHDI

PD

ED

Athens

39.2

1.80

109

122

Thessaloniki

42.1

1.78

98

94

Patras

35.9

1.81

136

143

Heraklion

46.1

1.66

87

85

Larissa

40.5

1.80

100

84

Volos

42.0

1.79

103

94

Ioannina

42.4

1.77

134

82

Kavala

40.4

1.87

103

82

Kalamata

41.2

1.67

149

107

39CONTAG values are between 35.9 (Patras) and 46.1 (Heraklion). An intuitive explanation of the differences between the lower and higher contagion indexes is provided by the MPA and ENN_MN metrics. MPA values (Table 5) in Patras are the lowest for almost all classes and contrast with those of Heraklion, a city of about the same population; the ENN_MN metric (Table 6) is relatively low in Patras and high in Heraklion. The second lowest contagion value (39.2) and one of the highest SHDI values (1.80) are in Athens.

40The correlation between CONTAG and SHDI metrics is -0.61. Both metrics denote dispersion/diversity/fragmentation, nonetheless a comparison of their values shows some important differences. The CONTAG metric, is computed at the pixel level and therefore its value is affected by all pixels that comprise the large patches, while the SHDI does not fully account for these since it considers only the proportional distribution of area among patches. The SHDI index for Patras and Athens is similar, but the CONTAG metric of Athens is almost 10% higher than that of Patras because of the large patches.

41A comparison of the CONTAG metrics estimated with and without polygon aggregation (original UA data) is presented in Figure 6. The original UA data result in lower CONTAG values. This is reasonable since with polygons merging some urban fabric patches and parts of the road network disappear and therefore both patch dispersion and interspersion decrease leading to a higher CONTAG. The difference between the two CONTAG metrics is affected by the form of the urban area. However, for both urban area configurations the lowest CONTAG metrics are in Patras and Athens and the highest in Heraklion. The correlation between the merging rate of the polygons of the original UA data and the change of the CONTAG index is 0.36 but not very significant.

Figure 6: CONTAG values for the Greek cities (with and without polygon aggregation)

Figure 6: CONTAG values for the Greek cities (with and without polygon aggregation)

42The highest values for the PD and ED landscape metrics are respectively in Kalamata and Patras. Patras, the city with the lowest CONTAG metric, indication of fragmentation/dispersion has also a relatively high PD metric. Although the distribution of urban fabric among the three classes is different (Table 3), the PD metric for Thessaloniki, Larissa, Volos, and Kavala is about the same, which implies that the PD metric at the landscape level might be relatively coarse since it aggregates the characteristics of the three urban fabric classes into one indicator. Same is true for the ED metric which for Heraklion, Larissa, Ioannina and Kavala is about the same.

43Fractal Dimension is an indicator that describes the complexity of urban areas with respect the degree to which the patterns of developed/built-up areas change across different scales that is whether they are reproduced at different scales of spatial analysis. It is an index of how built up areas conform to the urban landscape with respect their space-filling capacity; that is whether they fill the area in an organized, methodic way or whether they are distributed in unsystematic, uncoordinated ways. FD values range between 1 and 2 with areas characterized by fragmented disorganized built-up patterns having low FD, whereas in compact areas or in areas where the fragmentation is systematic and reproducible the FD is higher.

44The FD of an urban area can be estimated with the Fractalyse software. Fragstats also estimates the Fractal Dimension Index (FRAC), the methodology implemented however, results in an indicator that is a proxy of the complexity of the shape of the patches rather than the scalability of the built-up patterns. The Fractalyse software includes several methodologies for estimating the FD. For the nine Greek cities the FD was computed using the box-counting and dilation methods. In the box-counting method to determine the change/differences of the patterns of built-up areas across different scales an iterative approach is followed. At each iteration a grid of boxes of size r is superimposed on the region under study and the number of boxes N(r) needed to cover the built-up areas is counted (Frankhauser, 1998). The process is repeated with the size of the box reduced from one step to the other until a minimum size rmin is reached. This approach relies on the mathematical definition of a fractal which is based on the expression N(r) = k r–D (where N(r) is the number of parts necessary to cover the fractal shape; k is a constant; r represents the scale and D the fractal dimension). In the dilation method each pixel that represents built-up areas is surrounded by a square window of size ε. The size of these squares is then gradually enlarged, and the total surface A(ε) covered at each stage is measured. As the squares are enlarged, any details smaller than ε are overlooked and in this way an approximation of the original form is gradually obtained. By dividing this total surface by the surface of a test square (ε2), an approximation of the number of elements N(ε) necessary to cover the whole pattern is estimated.

45To implement these procedures in the Greek cities, binary maps of built-up areas (0 non-developed, 1 developed) were prepared and cities were examined at a scale 1:500,000 with tiff images produced at a resolution of 600 dpi. In the box-counting method different grids of size ranging from 1 to 256 pixels (1, 2, 4, 16, 64,128, 256) were used and the FD was estimated through statistical regression. In the dilation method 20 steps of dilation were used. In all cities the value of the correlation coefficient of the regressions in the box-counting method was higher than 0.97 while for the dilation method it was higher than 0.99. The FDs estimated by each method are presented in Table 9.

Table 9. Fractal dimension for the nine Greek cities

Box counting

Dilation

Athens

1,69

1,64

Thessaloniki

1,55

1,53

Patras

1,48

1,46

Heraklion

1,47

1,41

Larissa

1,40

1,31

Volos

1,52

1,49

Ioannina

1,42

1,34

Kavala

1,41

1,35

Kalamata

1,37

1,37

46The fractal dimension values obtained by the two methods are comparable (correlation coefficient 0.96), there are, however, some minor differences. With both methodologies the highest FD values are in the largest cities of Athens and Thessaloniki, with the FD of Athens being significantly higher and indication that it is a more compact city. Volos, which based on the spatial metrics can be considered a compact city, has a FD that is almost the same as Thessaloniki’s although the city is significantly smaller in area and population. The dilation method estimates lower FD values for all cities. However, the decrease in the FD values of Heraklion, Larissa, Ioannina and Kavala obtained with this method is significantly higher.

47Analysis was performed to determine whether FD values are related to landscape metrics and other urban area characteristics. Of course, landscape metrics consider all classes independently, while for computing the FD artificial land classes are aggregated into one class (roads excluded). Good examples of this difference are Patras and Heraklion which have respectively the lowest and highest CONTAG but the same FD. The only significant correlation is between FD and ED (r=0.63, significance at the .10 level). There are however strong and very significant correlations between FD and population density (r=0.80) and percent of artificial area (r=0.98). These correlations are of course obtained from a sample of only nine cities that differ significantly in terms of population and area and must be therefore validated with larger datasets.

Discussion

48In the previous section there was an extensive discussion of the various metrics and how their values differ across classes and cities. In this section each city is described on the basis of its metrics.

49Athens, or more correctly the Attica region is a large metropolitan area with almost 4 million people; there is a large port (Piraeus) and many municipalities. Population density is 47 people per ha. The three urban fabric classes C1, C2 and C3 consume 25%, 29% and 46% respectively of the urban fabric area. Fifty four (54%) of C1 area is in 4 large patches that include most of the city of Athens and several surrounding municipalities. The large patches lead to an MPA for class C1 that is almost three times higher than the MPA of the other urban fabric classes. Land use classes are mixed with mean shortest distance (ENN_MN) between patches of the same class being less than 200 m, smaller than in other cities. Green areas/sports facilities (C7) ENN_MN metric is among the lowest, higher only of the equivalent metric of Patras. The Fractal dimension metric is the highest, an indication that the urban area is compact.

50Thessaloniki is a large metropolitan area, a port city with population of one million. Its form follows the coast of Thermaikos Bay extending inland for several kilometers (Figure 1). Compared to Athens there is a 4:1 population and artificial area difference, therefore population density is the same for both cities, however the difference in the size of the LUZ area is only 2:1, that is Thessaloniki as defined in UA covers an area half of the area of Athens. Average size of C1 patches is 10.2 ha, 38% larger than Athens and the distance between them is 357 m, double the distance of patches of this class in Athens. There are large patches in the downtown area and C1 patches in periurban areas are distributed at a distance between them. This implies that there are gaps of the continuity of the built up areas. Only 21% of urban fabric is classified as C3, lowest percentage than other cities and patches of this class are smaller than in other urban areas (MPA 0.9 ha), indication of fragmentation and sprawl. Compared to its population, it is the city with one of the highest percentage of land classified as industrial/commercial activities. All of these characteristics result in a CONTAG metric that takes a value that is on the high side and a fractal dimension smaller than that of Athens.

51Patras is a port city located in the north coast of Peloponnesus. It is the city with the lowest CONTAG indicator and for all three urban fabric classes MPAs are smaller than in other cities. Only 17% of urban fabric is classified as C1 and C1 patches are at an average distance of 223 meters one from the other (one of the smaller ENN’s for this class). Green areas/sports facilities patches are at an average distance of only 200 m (in Athens they are at a distance of 280 m while for all other cities the distance is more than 400 m) and the PD metric for this class is almost 6, by far the highest of all cities. In terms of compactness as measured by the FD metric, Patras is less compact than Volos but more compact than Heraklion.

52Heraklion, with population density of 34 people per ha, is a port city in the island of Crete. It is one of the few urban areas in Greece in which there was significant population growth between 2001-2011 (4.3%). Only 12% of urban fabric is denoted as C1 and a record 63% is classified as C3, highest proportion than other cities (together with Kalamata). Patches of all three urban fabric classes are more concentrated than in other urban areas; PDs of C1 and C2 are the lowest and ENN metrics are the highest (400 m for both classes). The largest C1 patch –the area contained by the Venetian castles- represents 60% of total C1 area, whereas the largest C2 patch accounts for 26% of the C2 area. The PD metric for C3 is among the highest, a sign of discontinuity; however C3 patches are among the largest in size for this class, an indication that the large PD metric results from the large size of the urban area classified as C3. All these land use distribution patterns lead to a CONTAG metric that is the highest.

53Larissa is a large city in Thessaly about 60 km from Volos, another urban area. Located in the flatlands with relatively abundant land for development its population density is only 17 people per ha. Land use classes C1, C2 and C3 represent 28%, 42% and 30% respectively of the urban fabric. Land uses are split into relatively many smaller patches. C3 area is half of the equivalent area in Heraklion (1,294 vs 2,233 ha), there are however, more C3 patches in Larissa than in Heraklion (1,333 vs 1,000) and at a greater distance between them (275 vs 156 m). Although Heraklion is located in a relatively flatland its expansion is constrained in the east by the airport and in the south partially by the archaeological site of Knossos (and in the north by the sea) and this explains the low PD metrics. The fractal dimension indicators estimated with the box-counting method also show that Heraklion is a more compact city (1.44 in Heraklion, 1.36 in Larissa). Compared to its population, Larissa is the city with the largest proportion of area occupied by Industrial/commercial activities (together with Thessaloniki).

54Volos is a port city consisting mainly of two municipalities, Volos and Nea Ionia and is well known for its rectangular street layout. Population density is 33 people per ha and urban fabric is relatively evenly split between C1, C2 and C3 (35%, 26% and 39%). The MPA and LPI metrics of C1 in Volos take higher values than other cities (LPI second only to Kavala). In the downtown area there is a C1 patch of 385 ha, about 50% of the total C1 area. Compared to its population and area there is significant industrial/commercial activity. Fractal dimension is the highest for the second tier cities (that is excluding Athens and Thessaloniki, the large urban areas), another indication of its compact structure.

55Ioannina, the largest city in the region of Epirus, is located inland in an area surrounded by mountains; population density is only 14 people per ha, the lowest of all nine cities. C1 class accounts for only 13% of urban fabric with the remaining land split between the other two urban fabric classes. There is a sharp contrast with Volos a port city of the same population with much higher population density and different distribution of urban fabric. MPA for C1 is only 3.1 ha whereas in Volos it is 10.5 ha and average distance between patches of this class is 495 m in Ioannina and only 158 m in Volos. The PD metric for industrial/commercial activity is the highest of all cities (25 versus 12 the next lower one).

56Kavala is a port city in the north of Greece about 130 km east of Thessaloniki. Population density is about 28 people per ha. Urban fabric is almost evenly split between the three urban fabric classes (31%, 35% and 34%) and the topography of the area, built on a hill overlooking the sea, has affected its form. The city is bounded by the sea in the south and by steep-slope hills and mountains in the north, and so it presents a rather compact form with activities expanding mainly towards the east and along the shore.The largest C1 patch represents 70% of the C1 land and average distance between patches of this class is 374 m one of the highest. ENN_MN for C2 and C3 are also among the highest. The FD metric is among the lowest, although it could be considered a compact city.

57Kalamata is a city by the sea in the south Peloponnesus, with a population density of 25 people per ha. C1 accounts for only 17% of urban fabric, while 65% is classified as discontinuous medium and low density (C3). MPA for C1 is 3.6 ha and the distance between patches of this class is 266 m. There is an interesting contrast between Kavala and Kalamata, cities of about the same population, both by-the-sea but with different form because of the area’s topography. Although Kalamata is surrounded by mountains there is an extensive amount of land available for development and with agriculture and tourism being the main economic activity, this has resulted in lower densities, what could be considered as sprawl in larger urban areas. The city of Kavala has developed amphitheatrically on the hill overlooking the sea and therefore built-up densities are significantly higher.

Conclusions

58An important issue brought up by the research reported in this paper is that, before estimating spatial metrics, it is necessary to modify the original UA dataset merging neighboring polygons of the same type to form larger continuous areas. This approach distinguishes between land use distribution and city blocks layout and makes explicit the organization of an urban area in contiguous areas of the same type. Metrics estimated with this modified UA database, provide a better representation of the land use distribution. With the regular update of the Urban Atlas maps every 6 years, this methodology could be used also to monitor the changes that are occurring and provide a quick explanation for the changes, for example C3 polygons changing to C2 polygons in the vicinity of C2 polygons can be considered as “infill” etc.

59The comparative analysis of the nine Greek cities reveals that metrics depict the similarities and differences that exist among urban areas.The estimated metrics provide an insight on the form of the cities and support our knowledge about their form. Continuous urban fabric is concentrated in the core of the cities, with densities diminishing as the distance from the center increases. There are of course differences in the urban form of the cities; Athens a large metropolitan area has different characteristics than other urban areas, land uses are more mixed, average distance between patches of the same land use are shorter, and is therefore more compact. In smaller cities there is usually one large C1 patch dominating the city. There are cities such as Volos that are very compact with well distinguished large downtown center and industrial areas. In Heraklion the downtown center is one large C1 patch, and the rest of the urban area is split between large C2 and C3 patches.

60Some of the differences in the form of the Greek cities can be attributed to topography, cities located in flat plains with relatively abundant land for expansion, versus areas ‘constrained’ (Larissa vs Heraklion), or located at the foothills of a hill (Kalamata vs Kavala). Another explanation for the differences might be the land use zoning restrictions adopted and enforced and/or the enactment of urban development plans in the past (Larissa vs Heraklion). The comparison of several metrics between cities of about the same population size demonstrates that the metrics portray reasonably well the differences in urban form.

61The issue of finding the most meaningful and appropriate set of metrics for describing urban form, is an issue that always attracts attention. The analysis of the Greek cities demonstrates that class metrics, rather than landscape metrics, provide a more comprehensive picture of the urban form. With the UA dataset identifying several land use types class metrics communicate a richer picture of the urban morphology. Additional metrics are probably needed for measuring the interaction of different classes, that is how land use classes interact across the urban gradient. Landscape metrics combine all land use classes and differences in built-up densities are not necessarily translated to significantly different values. Cities with different percentages of urban fabric classes might have the same landscape metric. In a departure from most spatial metric studies fractal dimension was estimated using the principles of fractal theory as implemented in the Fractalyse software. The results complement those obtained from the spatial metrics, and fractal dimension provides an insight on the degree of the compactness of an urban area that is not found in the other metrics.

62In conclusion, the availability of the UA for the cities in Europe, a massive database developed with the same standards and updated on a regular basis, might lead to new insight on how cities grow and how activities as measured by land use/cover occur. For planners, it provides at no cost a database that can be used for identifying and quantifying the existing urban form and therefore can be used for addressing planning issues and monitoring urban growth.

Haut de page

Bibliographie

Aguilera F., Valenzuela L., Botequilha-Leitão, A., 2011, "Landscape metrics in the analysis of urban land use patterns: A case study in a Spanish metropolitan area", Landscape and Urban Planning, Vol.99, No.3/4, 226-238.

Aguilera-Benavente, F., Botequilha-Leitao, A., & Díaz-Varela, E., 2014, "Detecting multi-scale urban growth patterns and processes in the Algarve region (Southern Portugal)", Applied Geography, Vol.53, 234-245.

Bossard M., Feranec, J., Otahel, J., 2000, CORINE land cover technical guide – Addendum 2000, Technical report No.40. European Environment Agency.

Büttner G., Feranec J., Jaffrain G., Mari L., Maucha G., Soukup T., 2004, "The CORINE Land Cover 2000 project", EARSeL eProceedings Vol., No.3, 331-346 http://www.eproceedings.org/static/vol03_3/03_3_buttner2.pdf.

Dietzel C., Herold M., Hemphill J.J., Clarke K.C., 2005a, "Spatio-temporal dynamics in California’s Central Valley: Empirical links to urban theory", International Journal of Geographical Information Science, Vol.19, No.2, 175–195.

Dietzel C., Oguz H., Hemphill J.J., Clarke K.C., Gazulis N., 2005b, "Diffusion and coalescence of the Houston Metropolitan Area: evidence supporting a new urban theory", Environment and Planning B: Planning and Design, Vol.32, No.2, 231-246.

EEA, 2010, Mapping guide for a European Urban Atlas, European Environment Agency, Copenhagen, http://www.eea.europa.eu/data-and-maps/data/urban-atlas#tab-methodology.

Frankhauser P., 1998, "The fractal approach: A new tool for the spatial analysis of urban agglomerations", Population, Vol.10, No.1, 205-240.

Frankhauser P., 2004, "Comparing the morphology of urban patterns in Europe – a fractal approach", European Cities, insights on outskirts, Report Cost action 10, 79-105.

GISAT, 2012, "Urban Atlas Exploration tool", http://urbanatlas.gisat.cz/.

Geoghegan J., Wainger L., Bockstael N., 1997, "Spatial landscape indices in a hedonic framework: an ecological economics analysis using GIS", Ecological Economics, Vol.23, No.3, 251-264.

Guérois M., Pumain D., 2008, "Built-up encroachment and the urban field: a comparison of forty European cities", Environment and Planning A, Vol.40, No.9, 2186-2203.

Ji W., Ma J., Twibell R. W., Underhill K., 2006, "Characterizing urban sprawl using multi-stage remote sensing images and landscape metrics", Computers, Environment and Urban Systems, Vol.30, No.6, 861–879.

Herold M., Clarke K.C., Scepan J., 2002. "Remote sensing and landscape metrics to describe structures and changes in urban land use", Environment and Planning A, Vol.34, No.8, 1443-1458.

Herold M., Goldstein N.C., Clarke K.C., 2003, "The spatiotemporal form of urban growth: measurement, analysis and modelling", Remote Sensing of Environment, Vol.86, No.3, 286-302.

Herold Μ., Couclelis Η., Clarke K.C., 2005, "The role of spatial metrics in the analysis and modeling of urban land use change", Computers, Environment and Urban Systems, Vol.29, No.4, 369–399.

Huang J., Lu X., Sellers J.M., 2007, "A global comparative analysis of urban form: Applying spatial metrics and remote sensing", Landscape and Urban Planning, Vol.82, No.4, 184-197.

Lagarias A., 2007, "Fractal analysis of the urbanization at the outskirts of the city: models, measurement and explanation", Cybergeo : European Journal of Geography, article No.391. DOI : 10.4000/cybergeo.8902

Maucha G., Büttner G., Kosztra B., 2010, European validation of GMES FTS Soil Sealing Enhancement data, European Topic Center Land Spatial Information, European Environment Agency.

McGarigal K., Cushman S.A., Ene E., 2012 FRAGSTATS v4: Spatial Pattern Analysis Program for Categorical and Continuous Maps. Computer software program produced by the authors at the University of Massachusetts, Amherst.

Pham H.M., Yamaguchi Y., Bui T.G., 2012, "A case study on the relation between city planning and urban growth using remote sensing and spatial metrics", Landscape and Urban Planning, Vol.100, No.3, 223-230.

Plexida S., Sfougaris A., Ispikoudis I., Papanastasis V., 2014, "Selecting landscape metrics as indicators of spatial heterogeneity—A comparison among Greek landscapes", International Journal of Applied Earth Observation and Geoinformation, Vol.26, 26–35.

Prastacos P., Chrysoulakis N., Kochilakis G., 2012, "Spatial metrics for Greek cities using land cover information from the Urban Atlas". Proceedings of the AGILE'2012 International Conference on Geographic Information Science: Science Multidisciplinary Research on Geographical Information in Europe and Beyond, Avignon, France, 24-27 April 2012.

Ramachandra T.V., Bharath H. A., Durgappa D. S., 2012, "Insights to urban dynamics through landscape spatial pattern analysis", International Journal of Applied Earth Observation and Geoinformation, Vol.18, 329–343.

Rempel R.S., D. Kaukinen., Carr A.P., 2012, Patch Analyst and Patch Grid. Ontario Ministry of Natural Resources. Centre for Northern Forest Ecosystem Research, Thunder Bay, Ontario. http://www.cnfer.on.ca/SEP/patchanalyst/Patch5_2_Install.htm.

Salvati L. 2014 "The spatial pattern of soil sealing along the urban-rural gradient in a Mediterranean region", Journal of Environmental Planning and Management, Vol.57, No.6, 848–861.

Seto K.C., Fragkias M., 2005, "Quantifying spatiotemporal patterns of urban land-use change in four cities of China with time series landscape metrics", Landscape Ecology, Vol.20, No.7, 871-888.

Schwarz N., 2010, "Urban form revisited—Selecting indicators for characterising European cities", Landscape and Urban Planning, Vol.96, No.1, 29–47.

Skupinski G., Binh Tran D., Weber C., 2009, "Les images satellites Spot multi-dates et la métrique spatiale dans l’étude du changement urbain et suburbain – Le cas de la basse vallée de la Bruche (Bas-Rhin, France)", Cybergeo : European Journal of Geography, article No;439. DOI : 10.4000/cybergeo.21995

Tannier C., Pumain D., 2005, "Fractals in urban geography : a theoretical outline and an empirical example", Cybergeo : European Journal of Geography, article No.307. DOI : 10.4000/cybergeo.3275

Tannier C., Thomas I., 2013, "Defining and characterising urban boundaries: A fractal analysis of theoretical and real Belgian cities", Computers, Environment and Urban Systems, Vol.41, 234-248.

Thomas I., Frankhauser P., Biernacki C., 2008a, "The morphology of built-up landscapes in Wallonia (Belgium): A classification using fractal indices", Landscape and Urban Planning, Vol.84, No.2, 99-115.

Thomas I., Tannier C., Frankhauser P., 2008b, "Is there a link between fractal dimension and residential environment at a regional level?", Cybergeo : European Journal of Geography, article No.413. DOI : 10.4000/cybergeo.16283

Thomas I., Frankhauser P., Badariotti D., 2012, "Comparing the fractality of European urban neighbourhoods: do national contexts matter?", Journal of Geographical Systems, Vol.14, No.2, 189-208.

Triantakonstantis D., Stathakis D., 2015, "Examining urban sprawl in Europe using spatial metrics", Geocarto International, Vol.30, No.10, 1092-1112.

Tsilimingas G., 2014, (in Greek) "Consequences of land use structure in the development of landscape characteristics: Quantification of structure and spatial pattern in the Larger Urban Areas", Aeichoros, Vol.19, 24-37.

Turner M., 1989, "Landscape ecology: the effect of pattern on process", Annual Review Ecological Systems, Vol.20, No.1, 171–197.

Weng Y.C., 2007, "Spatiotemporal changes of landscape pattern in response to urbanization", Landscape and Urban Planning, Vol.81, No.4, 341-353.

Zhao Υ., Murayama Υ., 2011, "Urban Dynamics Analysis Using Spatial Metrics Geosimulation" : in Murayama Y., in Thapa R.Β. (eds), Spatial analysis and modeling in geographical transformation process, Springer, 153-168.

Zaragozí B., Belda A., Linares J, Martínez-Pérez J.E., Navarro, J.T., Esparza, J., 2012, "A free and open source programming library for landscape metrics calculations", Environmental Modelling & Software, Vol.31, 131-140.

Haut de page

Table des illustrations

Titre Figure 1: Land use maps of the nine Greek cities (Legend of classes: see table 2)
URL http://journals.openedition.org/cybergeo/docannexe/image/28051/img-1.png
Fichier image/png, 1,5M
Titre Figure 2: Maps for a subarea of Athens, before and after polygon aggregation (Legend of classes: see table 2)
URL http://journals.openedition.org/cybergeo/docannexe/image/28051/img-2.png
Fichier image/png, 902k
Titre Figure 3: Number of polygons after aggregation for different road widths
URL http://journals.openedition.org/cybergeo/docannexe/image/28051/img-3.png
Fichier image/png, 20k
Titre Figure 4: Mean polygon area after aggregation for different road widths
URL http://journals.openedition.org/cybergeo/docannexe/image/28051/img-4.png
Fichier image/png, 18k
Titre Figure 5: The PD, ED and ENN class metrics*
Légende * To be displayed in the same graph, ED metrics are divided by 10 and ENN by 100
URL http://journals.openedition.org/cybergeo/docannexe/image/28051/img-5.jpg
Fichier image/jpeg, 48k
Titre Figure 6: CONTAG values for the Greek cities (with and without polygon aggregation)
URL http://journals.openedition.org/cybergeo/docannexe/image/28051/img-6.png
Fichier image/png, 13k
Haut de page

Pour citer cet article

Référence électronique

Poulicos Prastacos, Apostolos Lagarias et Nektarios Chrysoulakis, « Using the Urban Atlas dataset for estimating spatial metrics. Methodology and application in urban areas of Greece », Cybergeo: European Journal of Geography [En ligne], Aménagement, Urbanisme, document 815, mis en ligne le 11 mai 2017, consulté le 29 mars 2024. URL : http://journals.openedition.org/cybergeo/28051 ; DOI : https://doi.org/10.4000/cybergeo.28051

Haut de page

Auteurs

Poulicos Prastacos

Regional Analysis Group
Institute of Applied and Computational Mathematics
Foundation for Research and Technology-Hellas
Heraklion, Greece
E-mail: poulicos@iacm.forth.gr

Articles du même auteur

Apostolos Lagarias

Regional Analysis Group
Institute of Applied and Computational Mathematics
Foundation for Research and Technology-Hellas
Heraklion, Greece
E-mail: lagarias@iacm.forth.gr

Articles du même auteur

Nektarios Chrysoulakis

Regional Analysis Group
Institute of Applied and Computational Mathematics
Foundation for Research and Technology-Hellas
Heraklion, Greece
E-mail: zedd2@iacm.forth.gr

Haut de page

Droits d’auteur

CC-BY-4.0

Le texte seul est utilisable sous licence CC BY 4.0. Les autres éléments (illustrations, fichiers annexes importés) sont « Tous droits réservés », sauf mention contraire.

Haut de page
Rechercher dans OpenEdition Search

Vous allez être redirigé vers OpenEdition Search