Remote Sensing and Geoinformation Lena Halounová, Editor not only for Scientific Cooperation EARSeL, 2011 Potential Open Space Detection and Decision Support for Urban Planning by Means of Optical VHR Satellite Imagery Carsten Jürgens 1, Alexander Siegmund 2, Derya Maktav 3, Filiz Sunar 3, Hayriye Esbah 3, Tuzin Baycan Levent 3, Cihan Uysal 3, Kaan Kalkan 3, Onat Yigit Mercan 3, İrfan Akar 3, Holger Thunig 2 and Nils Wolf 1 1 Ruhr University Bochum, Department of Geography, Bochum, Germany; carsten.juergens@rub.de 2 University of Education Heidelberg, Department of Geography, Heidelberg, Germany; thunig@ph-heidelberg.de 3 Istanbul Technical University, Department of Geomatics Engineering, İstanbul, Turkey; maktavd@itu.edu.tr Abstract. This work presents a remote sensing based method for the extraction of potential building ground in order to promote inner-urban development and prevent sprawl developments in the urban periphery. The potential space available in a city is detected by adapting a master ruleset for object-based image analysis to QuickBird satellite data of three urban test sites: Berlin, Istanbul and Ruhr area. Moreover, a software tool for the multi-criteria evaluation of the detected areas regarding different land use options is proposed. The software allows integrating the remote sensing based products with local ancillary data in order to support urban planners in their decision making. Keywords. Urban planning, QuickBird, Spatial decision support system, Adaptability, Objectbased image analysis 1. Introduction 1.1. Background Urban areas are complex and highly dynamic landscapes which support today more than half of the world s human population. Hence, in spite of covering only three percent of the global land surface, urban areas exert marked effects on environmental conditions at both global and local scales. Negative connotations of urban landscape changes are ground sealing, loss of valuable soil, and the fragmentation of urban and natural land. From a theoretical point of view, the aspect of fragmentation leads to an economically and ecologically disadvantageous condition because the demand of various infrastructure costs per person increases and the prosperity of ecosystems decrease by their isolation or intrusion. Facing such problems, a quantitative reduction of urban land consumption and careful handling of the limited natural land resources are key tasks for urban planning. Especially, inner-urban development towards compact and mixed-used urban structures promotes sustainable development and should take priority over fringe developments. Coping with these tasks requires precise and adaptive planning instruments and mapping the extent and, moreover, the internal structure of urban areas is therefore of great interest in order to understand and eventually manage the transformation processes.
1.2. Objective The presented study is part of the research project GAUS (Gaining Additional Urban Space, www.gaus-project.info) and aims at inventorying the open space available in urban environments and, moreover, at providing concrete decision support for its development. The method is based on VHR optical satellite data (QuickBird) and applied on three study areas: Berlin, Istanbul, and Ruhr Area. Object-based image analysis is applied to map land cover and land use and derive metrics describing urban form and inner-urban structure on multiple scales. In intersection with available GIS and local ancillary data, the outputs of image analysis serve as input for the software tool GAUSmart, a multi-criteria spatial decision support system which is developed as a tool for urban planners. Thus complex decisions are supported by numerical calculation and spatial visualization in order to come to objective solutions. 2. Satellite Data and Study Area The method is based on image data of the very high resolution optical satellite system QuickBird which provides 2.4m resolution in the 4-band multispectral (VIS-NIR) and 60cm resolution in the panchromatic mode. Exemplary, Figure 1 shows QuickBird subsets of the three study areas Berlin, Ruhr area (both Germany) and Istanbul (Turkey). The study sites reveal significant differences in the composition of the urban fabric which has to be anticipated in the urban modeling/classification process. Figure 1: QuickBird subsets showing urban landscapes of Berlin, Ruhr area and Istanbul. 3. Extraction of Open Areas 3.1. Method The extraction of open areas for inner-urban development follows an object-based approach for image analysis [1]. Basically, images are divided into homogeneous regions (image objects, segments) and afterwards classified by features such as color/tone, variance, texture and their semantic rela- 2
tionships. At this, the process of segmentation and classification is not a static two-step procedure but rather an iterative and interrelated process which successively turns raw image data into semantic information [2]. The sequences of segmentation and classification are empirically developed and finally designed as master ruleset which is adaptable to different image data inputs by manually adjusting 4 variables (visual on-screen inspection, expert system approach). Figure 2 shows the basic processing steps, starting with the multispectral and panchromatic QuickBird data as input. Firstly, basic land cover information (built-up, trees/shrubs, grass/mixed vegetation, bare, water) is extracted. As a second step, land use information is derived by aggregating the land cover layer on an upper, coarser spatial domain (defined by image segmentation). By querying the composition and configuration [3] of land cover sub-objects, a four-class land use layer is obtained (developed, forest, semi-natural and water). As a third step, by computing density maps for the class developed (proportional abundance of that class within a radial neighborhood) it is possible to stratify the four-class land use map into an urban-rural dichotomy. Combining the land use layer and the urban-rural dichotomy allows identifying the urban non-built-up regions. As a next step, such urban non-built-up ground is filtered by a defined set of shape and size constraints in order to remove all small, narrow or uncompact regions which are not suitable for any building/construction project (e.g. green stripes alongside roads or private gardens). The remaining open areas are potentially suitable for inner-urban development. There minimum size is set to 20000sqm whereas this number increases as a function of a regions shape (narrow regions are less desirable). Figure 2: Processing chain of the ruleset for the extraction of open areas. 3.2. Results The land cover classification is serves as basis for a knowledge-based derivation of the higher analysis outputs, namely the land use layer, the centrality index and the open areas. Hence, high accuracies of the land cover classification are a prerequisite for the method as a high degree of errors would be passed onto the higher products. On the other hand, minor inaccuracies are expected to be less significant as the downscaling of the spatial domain leads to a more robust model. To illustrate this point, the land use category developed requires a minimum of 10 percent proportional abundance of built-up sub-objects. Minor committed as well as omitted errors might be negligible or even level out. Tables 1 to 3 show the error matrix which is derived by comparing the classification to randomly chosen and manually labeled sample pixels. For the three study sites, accuracies of 0.84 or higher could be obtained, which can be interpreted as being reliable enough to derive the higher analysis outputs in satisfactory quality. 3
Table 1. Error matrix of the land cover classification; study area: Ruhr area. Table 2. Error matrix of land cover classification; study area: Istanbul. Table 3. Error matrix of land cover classification, study area: Berlin. Statistical evaluation and visual inspection of the extracted open areas is conducted by comparing the results to a reference map where the open areas have been manually delineated (see Figure 3). For the Ruhr area study site, a producer accuracy of 0.81 and a user accuracy of 0.86 have been obtained. As the open areas are a rather vague urban feature and independent reference data is not available, the derivation of statistical figures remains critical. Nonetheless, a visual inspection of the results supports the statistical test as it reveals a high accordance to such regions which one would intuitively consider as open area. Figure 4 overlays the extracted and the manually delineated open areas and shows that most of the areas could be spotted, while in terms of the geometric representation the results are for some cases less congruent to the reference map. 4
Figure 3: Comparison of extracted open areas to a manually derived reference map (example of the study site Ruhr area). The left image shows the manually delineated open areas (yellow polygons) which have been digitized within a subset of the image (black frame). The subset is further restricted to the automatically extracted urban footprint, namely the land use class settlement (blue polygon). Table 4. Obtained accuracies for open area detection by comparing to a manually derived reference map. 5
Figure 4: Comparison of the extracted open areas for inner-urban development (yellow) with manually generated reference polygons (black). Most polygons can be unambiguously linked with their associate polygon (apart from the middle-right panel where the extracted area has no correspondence). 4. Multi-criteria Decision Support The extracted open areas (as described in Section 3.1) mark contiguous non-built-up land within the urban footprint and it is questioned in how far they are suitable for inner-urban development projects. Beyond the remote sensing perspective, a realistic planning scenario encounters more aspects in order to valuate open areas as potential building ground. For example ownership situations, environmental protection or prohibition and risk zones are dimensions which have to be considered as well as the actual state of a city s land use structure. 6
With regard to the urban planner perspective we propose the developed software GAUSmart [4] as a tool for urban planners, which allows an integration of the local thus always specific planning scenario and the remote sensing derived products (open areas, land use, land cover). GAUSmart is a MATLAB based application which provides a graphical user interface to load and view geospatial data and to configure the decision support process in an interactive manner. GAUSmart evaluates the open areas regarding a user-defined purpose/scenario by calculating suitability maps. In our test application for the study areas Istanbul and Ruhr area we set up a scenario for the land use categories Residential, Industrial/Commercial, Open Space and Agricultural. Seven criteria are chosen to evaluate the suitability: Distance to Residential Distance to Industrial/Commercial Distance to Water Distance to Agriculture Composition of land cover (e.g. proportional abundance of vegetation) Slope (digital elevation model) Unsuitable areas/prohibition zones As preliminary steps, the attributes ranges are classified in 5 categories according to their suitability for each of the four land use scenarios. For example, when evaluating open areas regarding a residential use, a higher distance to industrial complexes is considered to be desirable while high slopes are regarded as being undesirable. Dividing the ranges (e.g. distances [m] or slope[ ]) in five ordered sub-ranges generates a criteria-specific suitability. With respect to a defined weighting factor, individually suitability maps are finally combined to form the multi-criteria suitability for a given land use scenario. Figure 4 and 5 show the results of this application for the study sites of Istanbul and Ruhr area. 7
Figure 5: Application of GAUSmart software for the study site Ruhr area. The color gradient from green to red indicates the suitability (green indicates high suitability). Top left: suitability for open space use; top right: suitability for residential use; bottom left: suitability for industrial/commercial use; bottom right: suitability for agricultural use. 8
Figure 6: Application of GAUSmart software for the study site Istanbul. The color gradient from green to red indicates the suitability (green indicates high suitability). Top left: suitability for industrial/commercial use; top right: suitability for residential use; bottom left: suitability for agricultural use; bottom right: suitability for open space use. 5. Conclusions The proposed work aims at inventorying the open areas available in urban environments and, moreover, at providing concrete decision support for its development. The method is based on VHR optical satellite data and applied on three study areas: Berlin, Istanbul, and Ruhr area. Object-based image analysis is applied to extract land cover/use information and metrics describing the urban form. In intersection with available GIS and local ancillary data, the outputs of image analysis serve as input for a multi-criteria spatial decision support system (GAUSmart) which is developed as a tool for urban planners. The workflow has been standardized in order to obtain comparable results across different test sites and datasets. 9
Acknowledgements We would like to thank the German Federal Ministry of Education and Research (BMBF) and the Scientific and Technological Research Council of Turkey (TÜBITAK) for the support of the project within the framework of IntenC. References [1] Benz, U., C. Hofmann, G. Willhauck, I. Lingenfelder & M. Heynen, 2004. Multi resolution, object oriented fuzzy analysis of remote sensing data for GIS ready information. ISPRS Journal of Photogrammetry and Remote Sensing 58, 239-258. [2] Baatz, M., Hoffmann, C., Willhauck, G., 2008. Progressing from object-based to object-oriented image analysis. In: Blaschke, T., Lang, S., Hay, G.J. (Eds.), Object-Based Image Analysis. Spatial Concepts for Knowledge-Driven Remote Sensing Applications. Springer, Berlin, 29-42. [3] Weeks, J.R., 2010. Defining urban areas. In: Rashed, T., Jürgens, C. (Eds.), Remote Sensing of Urban and Suburban areas. Springer, 33-45. [4] Maktav, D., C. Jürgens, A. Siegmund, F. Sunar, H. Eşbah, K. Kalkan, C. Uysal, O.Y. Mercan, I. Akar, 2011. Multicriteria Spatial Decision Support System for Valuation of Open Spaces for Urban Planning. 5th International Conference on Recent Advances in Space Technologies RAST 2011, Istanbul, Turkey, June 2011. 10