Computers, Environment and Urban Systems 33 (2009) 179–188
Contents lists available at ScienceDirect
Computers, Environment and Urban Systems journal homepage: www.elsevier.com/locate/compenvurbsys
Urbanization in India – Spatiotemporal analysis using remote sensing data H. Taubenböck a,b,*, M. Wegmann b, A. Roth a, H. Mehl a, S. Dech a,b a b
Julius–Maximilians-University Würzburg, Geographic Institute, Am Hubland, D-97074 Würzburg, Germany German Remote Sensing Data Center (DFD), German Aerospace Center (DLR), Oberpfaffenhofen, D-82234 Wessling, Germany
a r t i c l e
i n f o
Article history: Received 8 April 2008 Received in revised form 10 September 2008 Accepted 10 September 2008
Keywords: Urban growth Multitemporal remote sensing Landscape metrics Gradient analysis Classification of cities Mega cities India
a b s t r a c t Urbanization is arguably the most dramatic form of irreversible land transformation. Though urbanization is a worldwide phenomenon, it is especially prevalent in India, where urban areas have experienced an unprecedented rate of growth over the last 30 years. In this uncontrolled situation, city planners lack tools to measure, monitor, and understand urban sprawl processes. Multitemporal remote sensing has become an important data-gathering tool for analysing these changes. By using time-series of Landsat data, we classify urban footprints since the 1970s. This lets us detect temporal and spatial urban sprawl, redensification and urban development in the tremendously growing 12 largest Indian urban agglomerations. A multi-scale analysis aims to identify spatiotemporal urban types. At city level, the combination of absolute parameters (e.g. areal growth or built-up density) and landscape metrics (e.g. SHAPE index) quantitatively characterise the spatial pattern of the cities. Spider charts can display the spatial urban types at three time stages, showing temporal development and helping the reader compare all cities based on normalized scales. In addition, gradient analysis provides insight into location-based spatiotemporal patterns of urbanization. Therefore, we analyse zones defining the urban core versus the urban edges. The study aims to detect similarities and differences in spatial growth in the large Indian urban agglomerations. These cities in the same cultural area range from 2.5 million inhabitants to 20 million (in the metropolitan region of Mumbai). The results paint a characteristic picture of spatial pattern, gradients and landscape metrics, and thus illustrate spatial growth and future modelling of urban development in India. Ó 2008 Elsevier Ltd. All rights reserved.
1. Introduction For many decades, centuries in some cases, cities have been spreading (Anas, Arnott, & Small, 1998). A United Nations report (2003) projects that almost all global population growth in the next 30 years will be concentrated in urban areas. While urbanization is a worldwide phenomenon, it is exceptionally dynamic in India, where unprecedented urban growth rates have occurred over the last 30 years. During the last 50 years the population of India (today 1.2 billion) has more than doubled, but the urban population has grown nearly five times. The number of Indian mega cities will increase from the current three (Mumbai, Delhi and Kolkatta) to six by the year 2021 (including Bangalore, Chennai and Hyderabad), when India will have the largest concentration of mega cities in the world (Chakrabati, 2001). This phenomenon will necessitate advanced methodologies such as space technologies, which will help city planners, economists, environmentalists, ecologists and resource managers solve the problems which accompany such growth (Maktav & Erbek, 2005). Urban planners need information
* Corresponding author. Address: German Remote Sensing Data Center (DFD), German Aerospace Center (DLR), Oberpfaffenhofen, D-82234 Wessling, Germany. E-mail address:
[email protected] (H. Taubenböck). 0198-9715/$ - see front matter Ó 2008 Elsevier Ltd. All rights reserved. doi:10.1016/j.compenvurbsys.2008.09.003
about the rate of growth, pattern and extent of sprawl to provide basic amenities such as water, sanitation, and electricity, etc. Since planners currently lack such information, most of the sprawl areas lack basic infrastructure facilities. Within the fields of geography and planning, there is a long tradition of research on the description, mapping, characterization, measurement, understanding and explanation of form, morphology, and evolution of urban environments. The classic theories of urban morphology define urban patterns as concentric rings with different land use types (Burgess, 1925), as sectors. The transportation network modifies the form of the concentric zone pattern (Hoyt, 1939), and the multiple nuclei theory model a patchy urban form with multiple centres of specialized land use (Harris & Ullman, 1945). Since the 1960s various theories were used to characterise urban form: for example fractals (Batty, Longley, & Fotheringham, 1989), cellular automata (Tobler, 1979), dissipative structure theory (Allen & Sanglier, 1979), or landscape metrics (O’Neill et al., 1988). In general, the application, performance and outputs analysing and comparing the development of urban form of various cities depend strongly on the data available for parameterisation (Longley & Mesev, 2000). Remote sensing techniques have already proven useful for mapping urban areas at various scales and obtaining data
180
H. Taubenböck et al. / Computers, Environment and Urban Systems 33 (2009) 179–188
for the analysis of urban land cover change (Batty & Howes 2001; Donnay, Barnsley, & Longley, 2001; Herold, Scepan, & Clarke, 2002). Recent research has used remotely sensed images to quantitatively describe the spatial structure of urban environments and characterise patterns of urban morphology. Spatial metrics are critical in the description, analysis, and modelling of urban form and its changes (Herold, Goldstein, & Clarke, 2003). Researchers can use these indices to objectively quantify the structure and pattern of an urban environment. Most of the studies on urban landscape metrics focus on a single city (Herold et al., 2002; Herold et al., 2003; Ji, Ma, Twibell, & Underhill, 2006; Luck & Wu, 2002; Taubenböck et al., 2008a; Zhang, Wu, Zhen, & Shu, 2004), and there are few studies that compare cities in developing countries at about the same development stage in the same cultural area (Seto & Fragkias, 2005; Taubenböck et al., 2008b). In this study, we conduct a spatiotemporal analysis using time series of Landsat data to detect urban footprints and their changes in the 12 largest Indian cities (currently ranging from 2.5 to 20 million inhabitants). The land-cover classification is based on an object-oriented hierarchical classification approach (Berger, 2007; Pengler, 2007; Taubenböck, 2008; Taubenböck et al., 2007). The geometric capabilities of Landsat data are not cluttered with microscopic detail, but let us differentiate urbanised and nonurbanised areas with high accuracy. The main objective of this study is to identify similarities and dissimilarities in the urban characteristics of the largest Indian urban agglomerations. We quantify the spatiotemporal growth with a combination of zonal statistics, landscape metrics and gradient analysis to characterise types of urban development. Our strategy for quantitatively describing urban footprints is twofold. First, we calculate parameters of the full extent of urban areas to identify types of spatiotemporal growth in Indian cities. Second, we analyse two example parameters (built-up density and SHAPE index) with respect to location (city centre versus periphery). In particular, we use zones with increasing distance to the urban centres to map the spatial development of the parameter. The results paint a characteristic picture of spatial pattern, gradients and landscape metrics, and thus support to understand spatial growth and future modelling of urban development in India. In this study, we address several specific questions about spatiotemporal urbanization
and 14.3 million inhabitants respectively in 2005 and high average annual population growth rates of 3.1%, 4.1% and 2.0%, respectively, between 1975 and 2000 (United Nations, 2005), the mega cities represent one type of urban agglomeration in India. Three different groups can be classified based on population (Table 1). The three mega cities clearly stick out from incipient mega cities such as Chennai, Bengaluru, Hyderabad and Ahmadabad that currently have between 5 and 7 million inhabitants. The third group includes urban agglomerations between 2.5 and 5 million inhabitants such as Poona, Surat, Kanpur, Jaipur and Lucknow. Based on this pre-classification of cities by population, the study concentrates on analysing parallel or opposed spatial development over time within these groups or across the board. The crucial question in this study is whether urban agglomerations that share a culture area also share spatial characteristics over time.
How can spatiotemporal growth be quantified? Which urban growth types can be detected? Is there a spatiotemporal analogy for cities within the same cultural area? Is the future of cities currently in the range of 2.5–7 million people similar to today’s mega cities? 2. Study sites and data In 2005, 22 mega cities (urban agglomerations of 10 million inhabitants or more) around the world were identified; three of the cities, Mumbai, Delhi and Kolkatta, were on the Indian subcontinent (Münchner Rück, 2005; United Nations, 2005). Two of the cities, Mumbai at 3.1% and Delhi at 4.1%, have among the highest population growth rates of all mega cities in the world. Less attention is paid to ‘‘smaller”, explosively fast growing cities, whose high growth rates may precipitate transition into mega city status. Besides the three current mega cities, nine more urban agglomerations in India (Ahmadabad, Bengaluru, Chennai, Hyderabad, Jaipur, Poona, Kanpur, Lucknow, Surat) currently have more than 2.5 million inhabitants. Fig. 1 shows the geographic locations of the twelve cities on the Indian subcontinent analysed in this study. The three mega cities, Mumbai, Delhi and Kolkatta, clearly stand out from the other urban agglomerations in India. With 18.2, 15.1
Fig. 1. Location of India’s large urban agglomerations.
Table 1 Population growth in the 12 largest Indian cities in million inhabitants City/Year
1975
1990
2000
2005
2015
Mumbai Delhi Kolkatta Chennai Bengaluru Hyderabad Ahmadabad Poona Surat Kanpur Jaipur Lucknow
7,01 4,43 7,89 3,61 2,11 2,09 2,05 1,35 0,64 1,42 0,78 0,89
12,31 8,21 10,89 5,34 4,04 4,19 3,26 2,43 1,47 2,00 1,48 1,61
16,08 12,44 13,06 6,35 5,57 5,45 4,43 3,66 2,70 2,64 2,26 2,22
18,20 15,05 14,23 6,92 6,46 6,12 5,12 4,41 3,56 3,02 2,75 2,57
21,87 18,60 16,98 8,28 7,94 7,42 6,30 5,52 4,62 3,72 3,47 3,18
H. Taubenböck et al. / Computers, Environment and Urban Systems 33 (2009) 179–188
To answer this question, this study uses remote sensing data sets. Landsat data have been continuously available since 1972, featuring a geometric resolution of 79 m (multispectral scanner since 1972), 28.5 m (thematic mapper since 1982) and 15 m (enhanced thematic mapper since 1999). Therefore, this system lets us analyse extended time series. The chosen level of spatial resolution with Landsat features is not with microscopic detail, but incorporates specific features of the urban system. In return, the requirements for the differentiation of classes are limited to the classification of urbanized and non-urbanized areas. With its field of view of about 185 km, the satellite can survey the large metropolitan areas of the study sites. Measurement of both areal coverage and spatial distribution are needed to describe the morphology of an urban area adequately (Schweitzer & Steinbrink, 1998). All Landsat data sets at hand for the 12 study sites show no cloud cover and were subject to atmospheric corrections with the ATCOR (atmospheric and topographic correction) Software (Richter, 1996). Fig. 2 shows an example: Landsat ETM data from the year 2000 for the metropolitan area of the mega city of Mumbai. Clearly visible are the urbanized peninsula, the city’s large extent, the linear axial growth and the existing satellite cities. The main goal is to identify the urban built-up areas to measure changes in the urban extension over the available time steps. For that purpose, we base our classification methodology on an object-oriented hierarchical approach (Berger, 2007; Pengler, 2007; Taubenböck, 2008; Taubenböck et al., 2007). We used the objectoriented fuzzy-based methodology to combine spectral features with shape, neighbourhood and texture features for classification. We did a separate land cover classification, extracting the classes of built-up areas, bare soil, vegetation, and water on the available
181
Landsat images at three time stages for each city. Thus, we adjusted the classification approach based on the particular spectral characteristics and spatial capabilities of each data set. Consequently, classification accuracies vary due to sensor dependent spatial resolutions. Due to the large amount of mixed spectral information in such a coarse ground resolution – caused by multiple land covers within the lowest granularity of one pixel – the accuracy is limited. But for the requirement of mapping the large city footprint, its spatial dimension and the spatial developments over the years, the Landsat images provide enough information for an assessment of urban change. We assessed the accuracy of every classification result with 250 randomly distributed pixels. Due to missing ground truth data, we then assessed the accuracy visually by comparing classification results to the Landsat data. Thus, this assessment of accuracies already includes uncertainties. Even so, the high overall accuracies range from 86% to 93% correctly classified pixels. Postclassification comparison was found to be the most accurate procedure and had the advantage of indicating the nature of the changes (Mas, 1999). Therefore, independently, we implemented a comparative analysis of land cover classifications for the available times to monitor and analyse the land cover changes in the 12 metropolitan areas. The classification results are sampled up to the highest available geometric resolution of Landsat ETM. Subsequently, we did pixel-wise change detection to individually check the land cover classes of the available years. Fig. 3 shows the results of change detection for the 12 largest Indian urban agglomerations, displaying the urban footprints and their spatiotemporal evolution since the 1970s.
3. Spatiotemporal analysis of urbanization in India Urbanization may be linked with details of topography, transportation, land use, social structure and economic type, but is generally related to demography and economy in a city (Li, Sato, & Zhu, 2003). In the following, we analyse consequences of the plurality of these influencing factors on spatial urbanization. We analyse spatiotemporal processes of urbanization by urban form and its changes over time. The methodology to analyse spatiotemporal urban growth is two-fold: At first the analysis aims to identify types of spatiotemporal urban patterns at the city level. Therefore, we analyse parameters like absolute areal growth or built-up densities (sealed areas per spatial entity), as well as landscape metrics. In general, spatial metrics can be defined as quantitative and aggregate measurements derived from digital analysis of thematic-categorical maps showing spatial heterogeneity at a specific scale and resolution (McGarigal, Cushman, Neel, & Ene, 2002; Herold et al., 2003). The second step provides insight into detailed spatial gradients, like the change of built-up density with respect to increasing distance to the urban centre. Thus, location-based analysis of urban structure satisfies this requirement. The main idea is to learn mechanisms of the complex process of spatial urban growth by finding similarities and differences between cities’ past development at these two scales. 3.1. The urban footprints
Fig. 2. Landsat ETM+ 2000; Mumbai.
The results of the change detection display stark differences in the urban footprints of the Indian metropolitan areas and their development over time. While the urban footprints of Mumbai, Chennai and Jaipur are influenced by coastal or hilly orography, the remaining urban footprints are not subject to orographic restrictions. Fig. 3 shows the different types of spatial urbanisation in India from the 1970s to 2001. The pre-classification based on absolute population is reflected in the urban footprints displaying
182
H. Taubenböck et al. / Computers, Environment and Urban Systems 33 (2009) 179–188
Fig. 3. Spatiotemporal urban footprints of the large Indian urban agglomerations.
the mega cities’ extent first, with the incipient mega cities following and the smallest urbanized areas for the remaining cities. Different urban footprints evolve. Dichotomies include monoversus polycentric growth types, axial versus ring-shaped urbanization, and laminar versus dispersed settlement. This is reflected, for example, in the maximum difference in the diameter of the core cities from 60 km (for the oval, lengthwise urban footprint of mega city Kolkatta) in comparison to 11 km (for the compact, ringshaped urban core of Kanpur). In the following, we calculate quantifiable parameters defining growth and its pattern, to measure temporal development and compare different cities. 3.1.1. Classifying urban types Urban growth is characterised by complex diversity of spatial types. Growth may be laminar or punctual; it may increase density or it may sprawl; it may be mono- or polycentric. In the following, we try to quantify measures describing urban growth types The main idea to classify spatial urban types is to span a spider chart, using the avail able quantitative parameters as axes to characterise the spatial urban landscape (Fig. 4). We display the type of urban growth by combining various parameters and metrics. Eight
parameters quantifying the urban footprint at each time are calculated and combined in the net diagrams: absolute urban area (Area), built-up density (BD), landscape shape index (LSI), largest patch index (LPI), number of patches (NP), patch density (PD), total edge (TE), edge density (ED). Riitters et al. (1995) found high collinearity between 26 landscape metrics and postulated that a subset of six univariate metrics are good representatives. In statistical analysis, this problem would have to be targeted, but since this study includes descriptive statistics only, we disregarded the problem. Nonetheless, parameters which ought to show up with a high collinearity (e.g. PD and TE) did not show similar patterns across all cities. The absolute urban area defines the dimension of the urbanized areas as the first parameter to compare extents as development stages. As a further parameter, built-up density is calculated in a 20 km circle around the urban centre, as a quantitative measure of land consumption. We chose this approach to provide better comparability, due to the size of the smaller cities. The landscape shape index provides a standardised measure of the observed perimeter of all patches versus the perimeter of a compact patch of the same size of one land cover type (here: urbanised areas) in
183
H. Taubenböck et al. / Computers, Environment and Urban Systems 33 (2009) 179–188
LEGEND: Area = absolute area of urbanized areas; BD = Built-up density in 20 km circle around centre; LSI = Landscape shape index; LPI = Largest 1975 1990 2000 Patch Index; NP = Nearest Patch; PD = Patch density; TE = Total edge ; ED = Edge density;
Delhi
Mumbai
A rea
ED
LSI
PD
TE
LP I
PD
BD
LSI
BD
TE
LP I
LSI
0
Poona BD
Jaipur
LSI
BD
PD
TE
LP I
ED
Kanpur
A rea
BD
PD
Surat
LSI
ED
BD
LP I
TE
LSI
LP I
TE
LSI
0
PD
LP I
NP
NP
NP
BD 50
0
PD
A rea 100
50
0
LP I
NP
A rea
ED
LSI
0
100
50
PD
TE
NP
100
ED
LSI
LP I
NP
BD 50
0
PD
A rea 100
50
0
LP I
NP
A rea
ED
LSI
PD
NP
50
BD
0
100
ED
TE
TE
LP I
100
Lucknow
ED 50
PD
Ahmadabad A rea
A rea 100
ED
NP
TE
Bengaluru
A rea
50
0
PD
LP I
NP
100
50
LSI
0
PD
LP I
Hyderabad
A rea
TE
TE
NP
100
ED
LSI
0
NP
BD 50
50
0
Chennai
ED
BD
50
A rea 100
ED
BD
TE
Kolkatta
A rea 100
100
Fig. 4. Spider charts characterising spatiotemporal urban development.
the landscape (McGarigal et al., 2002; Schneider, Seto, & Webster, 2005). Hence, it is a measure of aggregation or clumpiness: if the urbanised area comprises one single compact area, the LSI will be small, approaching 1.0. If the landscape contains dispersed patches
with complex and convoluted shapes the LSI will be large. Thus, this parameter is used as a measure of complexity of urban growth. The largest patch index gives the proportion of total area occupied by the largest patch (Luck & Wu, 2002). It is a measure that
184
H. Taubenböck et al. / Computers, Environment and Urban Systems 33 (2009) 179–188
represents the separation of the urban landscape into smaller individual patches, as opposed to a dominant urban core. The LPI reaches 100 when the entire landscape consists of one single patch and approaches zero as the largest patch becomes increasingly small (McGarigal et al., 2002). A further landscape metric quantifying the urban landscape is the parameter number of patches. This parameter complements the LPI; it is a simple measure of the extent of subdivision. The patch density, which is the number of urban patches per area, is a measure of discrete urban areas in the landscape. Patch density is expected to increase during periods of rapid urban nuclei development, but may decrease if urban areas expand and merge into a continuous urban fabric (McGarigal et al., 2002; Seto & Fragkias, 2005). Yet another complementary parameter is the Total Edge, which measures the complete length of the borderlines of urbanized areas. TE approaches zero when the total landscape consists of one single patch. The edge density, which defines the ratio of the complete length of edges to the area of urbanisation, is related to TE but takes the area into account. This makes comparisons between landscapes of varying sizes feasible (McGarigal et al., 2002). The spider chart is calculated as a relative diagram, using for every axis the maximum value of the particular parameter of one city as 100. For comparison, the values of the remaining cities are calculated relative to the maximum value. The spider chart allows us to spatially visualise the urban type by quantifying landscape metrics that spatially describe the urban environment. In the following, the 12 largest urban agglomerations of India are displayed at three different time stages, beginning in the 1970s. Thus, we can analyse the development of spatial urban growth for individual cities, as well as in comparison with other urban developments. The resulting nets of the individual cities aim to classify spatial urban patterns and spatiotemporal growth. Analysing the spider charts, we see that the values of the parameters used to span the diagram almost without exception increase over time for the individual urban agglomerations. Thus, we can conclude that in general, the size of the enclosed area of the graph represents a temporal development stage of a city. This interpretation is reflected by the fact that the largest areas are for the three mega cities, with decreasing enclosed areas for the remaining smaller cities. Furthermore, the form of the graph of the individual cities retains a single consistent shape over time. Thus, spatiotemporal urbanization in India does not change its particular basic urban type. With urbanization, area, built-up density and complexity (LSI) consistently increase (Taubenböck et al., 2008b). This is confirmed through temporal evolution of the individual parameter in a particular city as well as a cross-city comparison at a given time. For example, rapid urban sprawl apparently involves dramatic increase in urban complexity. Thus, this part of the spider chart results in a precise reflection of the temporal development stage of one city. For the remaining landscape metrics, urbanisation results in a less uniform spatiotemporal evolution, but reveals similarities in a detailed analysis. The mega cities show consistent similarities in their graphs, emphasising an extensive LSI to TE axis and a graph development giving the net an Area, BD and LSI leaning shape. Hence, the mega cities consist of disaggregated patches and parallel a high built-up density. With respect to LPI, the three mega cities differ because Delhi has a high LPI. This indicates that Delhi has a trend consisting of one single large patch for the entire landscape. In detail, the preclassification of urban types based on the population is also reflected in characteristic gradients of the graphs. The mega cities, Mumbai, Kolkatta and Delhi, show a consistent increase from NP to PD, as well as a decrease from TE to ED. The other nine remaining cities display converse gradients. Nevertheless, local particularities evolve. For example, the combination of typical, unplanned highly dense Indian organic urban structure with the planned,
structured, low density urbanization engendered by English colonisation in Delhi causes by far the largest urbanised areas. Notwithstanding Delhi’s laminar ring-shaped growth, in contrast Kolkatta shows very patchy growth. Thus, Indian mega cities do not converge toward a standard spatiotemporal norm of urbanisation, but show similarities with respect to many parameters, resulting in a characteristic spider chart. It becomes apparent that the pre-classified incipient mega cities, Chennai, Hyderabad, Bengaluru and Ahmadabad show very similar shapes in each of their spider charts. The gradients from the various parameter axes, as well as the enclosed areas, almost consistently correspond to each other. A striking characteristic of the spider charts is the change of the longest direct connection line between opposing parameters from the mega cities to the incipient mega cities. Mega cities consistently feature the longest connection line from TE to LSI, while the incipient mega cities consistently show the longest line from ED to LPI. A typical trait of these cities is their high ED, indicating complex and unstructured urban sprawl. At the same time, the incipient mega cities have the highest rates of growth in LPI over time, indicating redensification as one type of spatial absorption of urban expansion. Similar to the mega cities in general, a characteristic, obviously evolution-dependent urban type is identified for the incipient mega cities. The urban agglomerations from 2.5 to 5 million inhabitants also result in homogeneous graphs in the net diagrams and parallel development of gradients. Likewise, in the incipient mega cities, the longest straight connection line is between ED and LPI. They consistently show a high LPI with a laminar growth resulting in low values of NP and PD. Again, with certain variations in the details, the spatiotemporal growth types correspond to each other. Thus, the hypothesis that spatial urban types correlate with their development stage is confirmed. Solely Poona sticks out, displaying an atypical urbanization type in India. For its assumed temporal development stage, Poona shows high values for ED, PD and LSI. This reflects a polycentric, punctual and dispersed settlement structure and thus a very complex spatial growth pattern. Its spacious, scattered and low density spatial urbanization is contrary to the observed urbanisation processes in the other Indian cities and thus, a separate urban growth type is observed. Overall, the pre-classified urban types based on population are clearly reflected in their characteristic shapes displayed in the spider charts. Thus, spatial urbanisation at city level proves broadly similar for cities at the same temporal development stage, with variations in the details. While the mega cities, do not converge toward a standard norm (due to historic urban planning processes and orographic restrictions), in general, we observe similarities in spatial urbanisation on the Indian subcontinent. 3.2. Zonal-based urban structure analysis In contrast with the urban type analysis using the complete spatial extent of the cities, the parameters can also illuminate the zonal-based spatial pattern. In the following, two parameters, builtup density and the SHAPE parameter, are subject to a detailed zonal-based gradient analysis to characterise spatiotemporal urban sprawl. Using six ring-shaped zones around the main urban centre, zonal-based analysis is carried out. Zone 1 is defined as a 5 km circle around the hub, while zone 2 entails a ring in 5–10 km distance, zone 3 in 10–20 km distance, zone 4 in 20–30 km distance, zone 5 in 30–40 km distance and eventually zone 6 in 40–50 km distance to the particular city centre. At first spatial characteristics over time are analysed for the two inner urban zones with respect to the built-up density. Fig. 5 shows all 12 large Indian urban agglomerations mapped against each other. The built-up density clearly shows similar trends for the majority at the urban centres of the 12 large Indian agglomerations. The
H. Taubenböck et al. / Computers, Environment and Urban Systems 33 (2009) 179–188
first noticeable trend is convergence of built-up densities to a saturation effect of around 80% in zone 1, due to low land availability or open spaces. Ahead of the other metropolitan areas Mumbai and Kolkatta show the highest built-up densities in zone 1, with little redensification since the 1970s. The difference for the third mega city Delhi, comes from the double structure of its urban core, with an old city showing even higher built-up densities than Mumbai and Kolkatta. The mix with New Delhi, a planned low density area, is the reason for its different appearance. All three mega cities show very similar decreasing built-up densities in zone 2 at around 58–61%. A second trend can be clearly identified at the urban cores of the incipient mega cities Chennai, Hyderabad, Bengaluru and Ahmadabad. While the built-up densities of the mega cities in the 1970s already showed values of about 70% in zone 1 and 30–40% in zone 2, the incipient mega cities only had around 30% and, respectively, 5% at that time. Immense redensification processes in the urban cores led those cities to values comparable to the values of today’s mega cities 20 years ago. For example, Hyderabad shows for the year 2001 a built-up density for zone 1 of 72% and 31% in zone 2. This corresponds with the built-up density values of Mumbai and Kolkatta at the end of the 1970s. The trend of concurrent redensification processes in Indian cities at particular temporal development stages can clearly be followed focusing on the cities of Jaipur, Lucknow, Kanpur and Surat. The year 2000 shows values in these urban agglomerations comparable to the incipient mega cities at 1990. Thus, for the third group of cities, again, a time shift of about 10 years in urbanization is detected. Thus, a clear concurrence is identified in the inner zones regarding built-up densities over time. For the smaller cities, this parameter may predict future expectations for spatial urbanization. The only city showing a different trend is, once again, the urban area of Poona. The uniqueness of Poona has already been displayed in Fig. 4, and becomes confirmed at the zonal-based structure analysis. The difference is exhibited in the low built-up densities in the main urban core, due to the absorption of spatial growth in polycentric and punctual, dispersed types of settlement gain. With the exception of Poona, in general, parallel development of built-up densities in Indian cities is identified with respect to their temporal development stages.
185
As a second zonal-based gradient analysis, the SHAPE parameter is calculated for the six zones with increasing distance to the main urban centre. The SHAPE parameter defines the complexity of each individual patch, and extracts the average value for the particular zone. Hence, it is a measure of patch compactness in its zone. High values correspond to complex shapes (average over zone) and vice versa. Fig. 6 shows development of the SHAPE parameter with increasing distance to the urban centre at the available times per city. Generally the spatial characteristics fall into two categories. The first class is that of cities which show increasing complexity with increasing distance from the urban centre. The second class is that of cities which are characterised by increasing compactness with distance. In comparison to the spatiotemporal correlation of built-up density in zone 1 and 2, as well as the detected similarities of many parameters at city level, the spatial development of the SHAPE parameter does not significantly coincide with the pre-classification of cities based on population. Nevertheless, it becomes apparent that most cities become more complex over time. In detail, the zonal-based analysis of the SHAPE index emphasises parallel growth types in the three mega cities. As typical characteristic appear, constant low SHAPE values from the urban core to the periphery indicate a widely developed stage of urban spatial exploitation. This goes along with very little change over time. A trait of the incipient mega cities shows a decreasing trend of the SHAPE index from the hub to the periphery at higher SHAPE values compared to the mega city. Concluding for incipient mega cities, the spatial options (especially in the urban core) are not yet fully developed and used up. This results in a more complex land use pattern, which is also reflected in yet lower built-up densities. In comparison, the urban agglomerations from 2.5 to 5 million show high complexity at their peripheries, indicating uncontrolled and complex sprawl. But generally, the zonal-based spatial development shows heterogeneity between cities at the same development stage. In particular, Lucknow and Kanpur show specific spatiotemporal development. Lucknow features an inversion of SHAPE complexity over time. While until 1990 the complexity values rise from the centre to the urban outskirts, the converse gradient becomes apparent around the year 2000. This suggests complex spatially disconnected development in the urban core,
Fig. 5. Spatiotemporal relationship analysis using the parameter built-up density.
186
H. Taubenböck et al. / Computers, Environment and Urban Systems 33 (2009) 179–188
Mumbai
5
2
3 2
2
1
1
0
0
0
2 3 4 5 Distance to center
1
6
Chennai
5
2 3 4 5 Distance to center
6
1
Hyderabad
5
3
3
3
2
Shape
4
Shape
4
2 1
1
0
0
0
6
1
2 3 4 5 Distance to center
6
1
4
4
4
3
3
3
1
Shape
5
Shape
5
2
2
2 3 4 5 Distance to center
6
1
2 3 4 5 Distance to center
Lucknow 5
4
4
Shape
3 2 1
Kanpur
2 3 4 5 Distance to center
1975
1990
6
6
4
3 2
3 2 1
0 1
2 3 4 5 Distance to center
Surat
5
1
0
1
6
Shape
5
2
0
0 1
6
1
1
0
2 3 4 5 Distance to center
Jaipur
Poona
Ahmadabad
5
6
2
1
2 3 4 5 Distance to center
2 3 4 5 Distance to center
Bengaluru
5
4
1
Shape
3
1
1
Shape
4
Shape
Shape
Shape
3
Delhi
5
4
4
Shape
Kolkatta
5
0
1
2 3 4 5 Distance to center
6
1
2 3 4 5 Distance to center
6
≈ 2000
Fig. 6. Location based structure analysis using the SHAPE parameter.
and at the same time, laminar urban sprawl at the periphery. In Kanpur, the complexity rises significantly over time at the suburban area due to non-laminar growth and the development of satellite cities. In comparison, the urban core stays compact and ringshaped and thus shows constant lower SHAPE complexity. 4. Main findings and conclusion The study has demonstrated that urbanisation and its spatiotemporal form, pattern and structure can be quantified and compared across cities using a combination of landscape metrics and gradient analysis. Landsat data proved to be an independent, area-wide, and (with respect to the limited geometric resolution) adequate data source for the analysis of large and fast-changing areas of Indian mega cities and incipient mega cities. The main findings and results address the three questions we defined earlier
in the introduction. (1) How can spatiotemporal growth be quantified? (2) Which urban growth types can be detected? (3) Is there a spatiotemporal analogy for cities within the same cultural area? (4) Is the future of cities currently in the range of 2.5–7 million people similar to today’s mega cities? (1) Spatiotemporal growth can be quantified by a number of parameters measuring dimension, pattern and form. This study uses absolute urban area, built-up density, landscape shape index, largest patch index, number of patches, patch density, total edge and edge density to quantify the classified urban footprints at various times. The usage of normalized spider charts to display the multiplicity of landscape metrics for a spatial visualization of urban growth types lets us monitor temporal development as well as compare different cities. Thus, a large scale assessment provides a first
H. Taubenböck et al. / Computers, Environment and Urban Systems 33 (2009) 179–188
quantification of spatiotemporal urban growth types by the form and enclosed area of the graph. In addition to the characterisation of the entire urban footprint, we used location based gradient analysis to analyse inner urban structural patterns. With distance to the main urban centre, the developing of parameters like built-up density reveals detailed insight into the urban pattern. (2) Generally speaking, there are four spatiotemporal urban growth types identified at city level in the 12 largest Indian agglomerations. The mega cities display an Area, BD and LSI leaning spider chart, consistent with the largest enclosed area as expression for the furthest temporal development stage. The incipient mega cities display a smaller enclosed area in the spider charts and feature a consistent ED–LPI directional shape and also consistently high LSI values. This group has uncontrolled, complex urban sprawl in the periphery, with redensification in the urban core catching up with development processes detected earlier in mega cities. The third group contains cities from 2.5 to 5 million inhabitants except Poona, which exemplifies the fourth spatial urban type in India. Jaipur, Kanpur, Surat and Lucknow show the smallest enclosed areas of the graph, symbolizing the least temporal development, but already showing a spatial spider chart shape similar to incipient mega cities. In contrast Poona has an atypical spider chart shape for India. The result shows at an early temporal development stage already a highly complex, punctually scattered and polycentric urban growth type. Analysing urban growth types from a different, more descriptive perspective, we identified three spatiotemporal growth types. First, there is the urban agglomeration characterised by a monocentric urban core with laminar – axial or ring-shaped – spatial growth. Members of these groups are the mega city Delhi and the smaller urban agglomerations of Ahamadabad, Jaipur, Lucknow and Kanpur. Second, there is a growth type showing a transition of mono- to polycentric growth with punctual spatial growth. Among this growth type are the cities of Chennai, Hyderabad and Bengaluru. Third, there is a polycentric growth type characterised by widespread punctual growth. This is predominantly the city of Poona, and with deduction Mumbai and Surat. (3) Overall, we detected similarity in spatiotemporal urbanisation at the 12 largest Indian urban agglomerations. It becomes apparent that the temporal development stage correlates with the enclosed area of the graph in the spider chart. In addition, the form of the net diagram generally corresponds with the pre-classification of Indian cities based on absolute population. In detail, individual parameters, like the built-up density with respect to their location highly correlate over time. Effects like redensification and saturation are found to match over time and converge toward a standard norm in Indian cities. The study reveals that aspects of spatial urban growth largely proceeded very similarly. But exemplifying another individual parameter, the study proves that an overall standard form cannot be detected. The SHAPE parameter reveals a trend not correlating with the development stage of a cities’ spatial evolution. Thus, generally speaking, the spatiotemporal urbanization and the particular urban footprints largely show similarity in the same cultural area. In detail, individual parameters reflect a large spectrum of possible growth patterns. Thus, an overall fixed standard form of spatial urban growth cannot be detected. (4) The study suggests that in general, today’s mega cities are good predictors of the future of morphology of incipient mega cities. We observe that the urban growth type of one particular city, defined by the spanned net, does not change
187
its shape significantly over time. Thus, characteristics of urbanization do not alter over time for an individual city. Along these lines, detected correlations of spatial parameters regarding urbanization across the board enable prediction for yet smaller cities. For example, the correlation of builtup densities in the inner district clearly shows the trend and what to expect in these locations in incipient mega cities in the coming years. The dimension of increasing areal growth, increasing built-up densities, rising complexity, with redensification resulting in decreasing TE can be expected. Naming just a few findings, planners can act on the trend towards urban growth in smaller cities. Thus, we conclude that in the same cultural area, the spatiotemporal development follows certain characteristics typical for a particular temporal development stage. Overall, the analysis shows that urban structure is very scaledependent. A multi-scale analysis starting at the entire urban footprint, as well as a location based gradient analysis, are found to provide insight to both sides. In future analyses, a highly detailed structural analysis of the large-scale and heterogeneous inner structures of urban morphology using satellite data with higher geometric resolution (e.g. Ikonos or Quickbird) are expected to augment information for planning purposes. Restrictions on the large extensions of these cities are the small swath widths, limiting area-wide analysis and the handling of the large amount of data. The time series of gradient analysis and landscape metrics is important for describing, understanding and monitoring the spatial configuration of urban growth. A comparative analysis is crucial for urban growth trajectories across cities. Measuring the development stages of the large Indian urban agglomerations, conclusions about incipient mega cities in the same cultural area like Hyderabad, Bangalore or Chennai may support planning, future modelling, and thus decision-making for sustainable and energyefficient urban futures. Acknowledgements This research was completed by cooperation between the German Aerospace Centre and the Geographic Institute of University of Würzburg. The authors would also like to thank the SHAKTI-Project Team for the productive teamwork. Furthermore we would like to specifically thank Isabelle Pengler, Markus Breunig, and Christian Berger for their great support. References Allen, P. M., & Sanglier, M. (1979). A dynamic model of urban growth: II. Journal of Social and Biological Structure, 2, 269–278. Anas, A., Arnott, R., & Small, K. (1998). Urban spatial structure. Journal of Economic Literature, American Economic Association, 36(3), 1426–1464. Batty, M., Longley, P., & Fotheringham, S. (1989). Urban growth and form: Scaling, fractal geometry, and diffusion-limited aggregation. Environment and Planning A, 21(11), 1447–1472. Batty, M., & Howes, D. (2001). Predicting temporal patterns in urban development from remote imagery. In J. P. Donnay, M. J. Barnsley, P. A. Longley (Hrsg.), Remote Sensing and urban analysis (pp. 185–204). London: Taylor and Francis. Berger, C. (2007). Raum-zeitliche analyse indischer megastädte mit Landsat–Daten (p. 82). Bachelor-thesis. Institute for Geography, Friedrich–Schiller-University Jena. Burgess, E. W. (1925). The growth of the city. An introduction to a research project. In R. E. Park, E. W. Burgess, & R. McKenzie (Eds.), The city (pp. 47–62). Chicago: University of Chicago Press. Chakrabati, P. G. D. (2001). Urban crisis in India: New initiatives for sustainable cities. Development in practice, 11(2–3), 260–272. Donnay, J. P., Barnsley, M. J., & Longley, P. A. (2001). Remote sensing and urban analysis. London: Taylor and Francis. Harris, C. D., & Ullman, E. L. (1945). The nature of cities. Annual American Academic of Political and Social Science, 242, 7–17. Herold, M., Scepan, J., & Clarke, K. C. (2002). The use of remote sensing and landscape metrics to describe structures and changes in urban land uses. Environment and Planning A, 34, 1443–1458.
188
H. Taubenböck et al. / Computers, Environment and Urban Systems 33 (2009) 179–188
Herold, M., Goldstein, N. C., & Clarke, K. C. (2003). The spatiotemporal form of urban growth: Measurement, analysis and modelling. Remote Sensing of Environment, 86, 286–302. Hoyt, H. (1939). The structure and growth of residential neighborhoods in American cities. Washington, DC, USA: Federal Housing Administration. Ji, W., Ma, J., Twibell, R. W., & Underhill, K. (2006). Characterizing urban sprawl using mutli-stage remote sensing images and landscape metrics. Computers, Environment and Urban Systems, 30(6), 861–879. Longley, P. A., & Mesev, V. (2000). On the measurement and generalization of urban form. Environment and Planning A, 32, 473–488. Li, L., Sato, Y., & Zhu, H. (2003). Simulating spatial urban expansion based on physical process. Landscape and Urban Planning, 64, 67–76. Luck, M., & Wu, J. (2002). A gradient analysis of urban landscape pattern: A case study from the Phoenix metropolitan region, Arizona, USA. Landscape ecology (Vol. 17, pp. 327–339). Kluwer Academic Publishers. Maktav, D., & Erbek, F. S. (2005). Analyse of urban growth using multitemporal _ satellite data in Istanbul, Turkey. International Journal of Remote Sensing, 26(4). Mas, J.-F. (1999). Monitoring land-cover changes: A comparison of change detection techniques. International Journal of Remote Sensing, 20(1), 139–152. McGarigal, K., Cushman, S. A., Neel, M. C., & Ene, E. (2002). FRAGSTATS: Spatial pattern analysis programm for categorical maps. Amherst: Computer software produced by the authors at the University of Massachusetts. Münchner Rück (2005). Megastädte – Megarisiken. Trends und Herausforderungen für Versicherung und Risikomanagement. . O’Neill, R. V., Krummel, J. R., Gardner, R. H., Sugihara, G., Jackson, B., Deangelis, D. L., et al. (1988). Indices of landscape pattern. Landscape Ecology, 1, 153–162. Pengler, I. (2007). Stadtstrukturtypenkartierung mit Methoden der Fernerkundung am Beispiel der Stadt Hyderabad (Indien) (p. 141). Diploma thesis. Fachhochschule München, Fachbereich Geoinformationswesen. Richter, R. (1996). A spatially adaptive fast atmospheric correction algorithm. International Journal of Remote Sensing, 17(6), 1201–1214.
Riitters, K. H., Neil, R. V. O., Hunsaker, C. T., Wickham, J. D., Yankee, D. H., Timmins, S. P., et al. (1995). A factor analysis of landscape pattern and structure metrics. Landscape Ecology, 10, 23–29. Schneider, A., Seto, K. C., & Webster, D. R. (2005). Urban growth in Chengdu, Western China: Application of remote sensing to assess planning and policy outcomes. Environment and Planning B: Planning and Design, 32(3), 323–345. Schweitzer, F., & Steinbrink, J. (1998). Estimation of mega-city growth. Applied Geography, 18/1, 69–82. Seto, K. C., & Fragkias, M. (2005). Quantifying spatiotemporal patterns of urban land-use change in four cities of China with a time series of landscape metrics. Landscape Ecology, 20, 871–888. Taubenböck, H., Pengler, I., Schwaiger, B., Cypra, S., Hiete, M., & Roth, A. (2007). A multi-scale urban analysis of the Hyderabad Metropolitan area using remote sensing and GIS (p. 6). Urban remote sensing joint event, Paris, France. Taubenböck, H. (2008). Vulnerabilitätsabschätzung der erdbebengefährdeten Megacity Istanbul mit Methoden der Fernerkundung. PhD Thesis. University of Würzburg, p. 178. . Taubenböck, H., Esch, T., Thiel, M., Wurm, M., Ullmann, T., Roth, A., et al. (2008a). Urban structure analysis of mega city Mexico City using multi-sensoral remote sensing data. In Proceedings of SPIE-Europe (international society for optical engineering) conference, Cardiff, Wales. Taubenböck, H., Wegmann, M., Berger, C., Breunig, M, Roth, A., & Mehl, H. (2008b). Spatiotemporal analysis of Indian megacities. In Proceedings of the international archives of the photogrammetry, remote sensing and spatial information sciences (ISPRS) (Vol. XXXVII) (pp. 75–82). Part B2. Beijing. Tobler, W. R. (1979). Cellular geography. In S. Gale & G. D. Olsson (Eds.), Philosophy in geography (pp. 3–16). Dordrecht, The Netherlands: Reidel Publishing Company. United Nations (2003). World urbanization prospects, The 2003 Revision, New York. United Nations (2005). World Urbanization Prospects, The 2005 Revision, New York. Zhang, L., Wu, J., Zhen, Y., & Shu, J. (2004). A GIS-based gradient analysis of urban landscape pattern of Shanghai metropolitan area, China. Landscape Urban Plan, 69, 1–16.
