Geographisches Institut der

Ruprecht-Karls-Universität Heidelberg

Im Neuenheimer Feld 348 69120 Heidelberg

email: hbalzere@ix.urz.uni-heidelberg.de

homepage: http://www.rzuser.uni-heidelberg.de/~bu1/sfb/d1/index.htm

The processes of urbanisation in Africa can be considered as an extraordinary development. Despite Africa’s very low average urbanisation rate of 25% (Nigeria 38%), the growth rate of urbanisation of 6% per year, which is equivalent to a doubling of the urban population within 12 years, has to be considered as ‘very high’.

Due to this dynamic urban growth, the respective administrations are faced with serious difficulties, since they are in charge of active development planning in terms of infrastructure progress such as the improvement of access to schools and hospitals, access as well as provision of public services like drinking water supply, sewage and waste disposal and, last but not least, planning and supervision of private construction activities. Considerable difficulties are caused by spontaneously constructed and therefore unplanned and illegal private buildings. In some towns, this land of construction activity even leads to foundation of new urban wards beyond the existing administrative boundaries, causing profound changes in the landuse patterns. Such highly dynamic urban growth makes urgent town planning inevitable. Unfortunately, master plans for town planning, lay-out plans or simple street maps hardly exist. In Nigeria, accurate data on urban population is not available, since the latest and - politically influenced (FRICKE, W. & MALCHAU, G., 1994, 163) - population figures are based on the 1991 census, the regional results of which were published in the year 1998. The existing maps and aerial photographs were mostly produced in the 60s and 70s and are thus inadequate.

Therefore, the application of urban satellite remote sensing for future town planning is not only sensible but utterly necessary. Arguments in favour of the use of satellite systems are certainly the fast data access, the quick visual interpretation, the good representation on a planar surface and their great integrity of a map after the process of geometrical image correction. A further advantage is the possibility of a qualitative as well as quantitative classification, such as the generation and analysis of, for example, urban boundaries, lay-out structures, building densities and sealing degrees.

Since the launch of the IKONOS satellite in 2000, satellite images with higher ground solutions of 4x4m (multi-spectral mode) and 1x1m (panchromatic mode) are available, causing gradual improvements of intra-urban interpretation and classification.

This article presents the case study of the Nigerian town of Gombe, where IKONOS scenes were used to analyse intra-urban building structures, and other spatial developments in order to differentiate and quantificate the urbanisation process.

The town of Gombe, located in the Savannah of NE-Nigeria, has been chosen as a typical example. A socio-economic analysis of the cultural change from tradition to modernity in the urban centre of Gombe and its environs is part of the geographical project of the "SFB 268" since 1998.

In 1996, Gombe was appointed as the capital of the newly established Gombe State, which mirrors the exceptional improvement of its functional and economic position. Since 1952, due to continuos immigration, the number of inhabitants of Gombe has multiplied by sixteen, reaching 300 000 in late 2000 (i.e. average annual growth rate of 5,9%) (see Figure 1). This demographic development is mainly based on dynamic intra- and interregional migration and significantly indicates the development of the economy. The high population density and thus increasing contacts between people lead to a dynamic process and the acceleration of the present cultural transformation on various levels. New forms of economic, social and religious interaction disintegrate former common behaviour. The division of the society into two main groups, one with a more traditional view and one which favours the processes of change and modernisation, contributes to an enhancing tension between socio-economic development and cultural persistence. Such difference in individual cultural behaviour manifest themselves in spatial transformation and are recognisable in the differentiation of urban building structure development, for example in the choice between the construction of a clay or brick building, the renovation the an old compound, or the roofing of the house.

"Temporal and spatial effects might be the keys to gleaning the information sought in an analysis. The process of change detection is premised on the ability to measure temporal effects." (LILLESAND & KIEFER 1987, 173). Urban change detection mapping can be facilitated through the interpretation of different remote sensing imageries.

The Figure 1 shows the detection of suburban development around the core of Gombe by using multidate aerial photographs (1950, 1964 and 1978) and satellite imageries (1986 SPOT MS, 2000 IKONOS). The method of change detection helps to illustrate the urbanization process, e.g. the direction of urban expansion. It can also be used, for example, to detect changes in building structures and densities as well as sizes and shapes of features such as roads.

Figure 2 illustrates the high ground solution of 4x4m of the IKONOS scene and compares it with the ground solutions achieved by LANDSAT TM and SPOT XS. All three systems picture the same details of Gombe. Column a) provides an overview of the city centre, column b) shows a detail of a). Here the large pixels of the LANDSAT and SPOT images become obvious. Thus, a detailed interpretation of a smaller area is not to be recommended when using LANDSAT or SPOT images. Due to the higher resolution of the IKONOS scene, even more details can be identified. Not only single trees, but also their shadows can be distinguished.

It can be concluded that the pixel size of 4x4m immensely improves the applicability of satellite imagery for urban satellite remote sensing. The new satellite system allows a more detailed intra-urban classification and therefore is an important tool for analysing structural and spatial differentiation, as well as monitoring changes.

As demonstrated before, the development of higher ground solutions on the one hand offers better possibilities of a more detailed intra-urban classification, due to the decreasing proportion of ‘mixed-pixels’ and ‘pure-pixels’. On the other hand, the variety within a defined class increases. Unfortunately, one still has to deal with a considerable proportion of ‘mixed-pixels’ due to a high ‘local frequency’. The often significant error rates of analyses of urban areas are mostly based on such ‘mixed-pixels’, which mislead the classifier because of their unequivocal recognition.

Figure 2: Three satellite systems comparison

Here, the first results of an intra-urban classification utilising the reflecting blue spectrum, will be presented. Due to the reflection of corrugated sheet iron roofs, most African towns, seen from space, appear in blue luminosity. The spectral differences of these metal roofs in dependence on their age are an important peculiarity which allows to differentiate the structure of the town and to monitor the development.

The results of a household survey indicate a strong statistic correlation between the age of the roofs and the age of the respective buildings (sample of 1 000 households; probability of 80%).

A combination of these results and knowledge of the place allows a differentiation of ‘old’, ‘medium’, and ‘new’ roofs – and therefore ‘old’, ‘medium’, and ‘new’ buildings by classifying the satellite image and grouping the different frequencies of the blue spectral band into three classes of ‘light-’, ‘medium-’ and ‘dark-blue’ appearance.

For a sake of a better classification, the scene was first transformed into its principal components (see Figure 3). This image processing is recommended in order to emphasise or de-emphasise certain aspects of the information contained in the image. "It is fundamental to the development of the principal components transformation to ask whether there is a new co-ordinate system in the multispectral vector space in which the data can be represented without correlation; in other words, such that the covariance matrix in the new co-ordinate system is diagonal." (RICHARDS, J.A. 1994, 136).

For a particular two dimensional vector space such a new co-ordinate system is depicted in Figure 3 b); i.e. high variation of the value in x-direction, no variation in the y-direction.

By the transformation of the principle components, a new covariance matrix was generated, which then was used for classification signatures processing. Afterwards some areas of interest (AOIs) were designated, that are identical with the enumeration areas (EA), in which the household survey was conducted (see Figure 4).

The samples of pixels representing the designated AOIs were subject to an unsupervised classification in order to create signatures. The procedure of an unsupervised classification can be used to determine the spectral class of each pixel. "While the information classes for a particular exercise are known, the analyst is usually totally unaware of the spectral classes." (RICHARD, J.A. 1994, 85).

The resulting signatures and the respective classes have to undergo profound evaluation in order to identify the affiliation classes. This method reveals subordinate classes which can be merged to form a single class. Unsupervised classification is therefore useful for determining the spectral class composition of the data prior to a detailed analysis using the methods of supervised classification.

The signature editor is a tool that helps to modify the signatures, before performing a supervised classification with the maximum likelihood decision rule in order to achieve the desired classes.

Figure 3: Two-dimensional multispectral space histograms for satellite data

before transformation (a) and after principal component analysis (b)

After the satellite scene is classified by using the merged signatures, it has to be evaluated. The raster attribute editor helps to highlight the different classes for the purpose of identification. The accuracy of the classification can be visually compared with the satellite imagery (see Figure 5). For this reason, the left column shows the blown up but unmodified satellite imagery (channel 1-2-4) with the a yellow frame marking the margins of the EA. The right column illustrates results of the classification. Trees and unpaved paths are satisfactorily recognised. The darker the pixels appear on the original scene, the more difficult the classification becomes. Due to the fact, that the old town (figure 5 a) is a densely built-up area with many small buildings, huts and open shelter to host the open firewood kitchen, the proportion of mixed pixels is relatively high. The high local frequency of this area makes a classification difficult and can mislead the class decision rule.

The second row shows an EA in a newly created ward with a sparse population distribution. Most of the buildings are new flats with new roofs, which facilitate the classification. By visual interpretation of Figure 5c, a certain peculiarity of the building structure alignment towards the scan lines of the satellite system has to be noticed. Buildings arranged in a diagonal way towards the scan lines cause a higher number of mixed pixels, which, in turn, give rise to inaccuracy in classification and interpretation.

The third row depicts a larger area. Again, unsealed surfaces, trees and light blue pixels can be classified easily, whereas the road is not distinguished amongst the dark blue pixels. All further attempts to identify the road failed, although typical signatures were used.

Figure 5: Some classification examples

As mentioned above, reliable reference data on population or building structures hardly exist. Therefore, the results of a household survey conducted by the author has been used for evaluation. Approximately 1 000 households within 24 enumeration areas were surveyed with the help questionnaire. The following indicators of the reference data can contribute to an evaluation:

• Number of compounds / buildings
• Age of compounds / buildings
• Types of compounds / buildings
• Sketches of compounds
• Socio-economic indicators of the inhabitants

Figures 6 a) and 7a) both show the unchanged satellite images of the two wards Madaki and Bajoga, (channel: 1-2-4), 6 and 7 b) illustrate the results of the classification while 6 and 7 c) - d) show the differentiation of compound type of the reference data set from the Madaki and Bajoga wards. In Madaki, the respective houses or compounds are built of clay (45%) or bricks (55%). The average age of all buildings is 15 years, with varying from 1 to 40 years. In Bajoga, 70% of the building are built of clay. There, the average age of the houses is 35 years, ranging from 3 to 70 years. According to own investigations, brick buildings are rarely older than 5 years, since the brick producing industry started their production in 1996. This information can be used for further GIS application. This is demonstrated in Figures 6 c–d) and 7 c-d). It is obvious that GIS application of the reference data matches the classification to a high degree. This proves that the age of the buildings correlates with the pixel values of the respective scene recorded by the IKONOS remote sensing satellite.

Figure 6: GIS application for referenced data-based evaluation – an example of Madaki ward

Figure 7: GIS application for referenced data-based evaluation – an example of Bajoga ward

After a satisfactory classification is successfully generated, a broad variety of further steps is possible, depending on the interest of the researcher. Some of them are outlined here:

• Further spatial analysis by area class quantification as a measure of spatial distribution: The generated classified image can be smoothed by a convolution kernel 3x3 - though a spatial generalisation always implicates a loss of information - and subsequently transformed from a raster map into a vector map. This procedure results in merging all pixels of one class clumping together as neighbours. All these merged pixels form polygons, which now can be counted and statistically analysed. The bigger in size these single polygons are - always representing only one class - the more homogenous in age the buildings of the respective area are. On the other hand, the more different polygons exist – representing different classes of the age of the buildings – the more heterogeneous the area is. With this means the spatial class distribution can be described as a measure of homogeneity.
• Segmentation of the built-up areas: "The spatial patterning of housing yields very important information for the architect-urbanist, as it provides an indicator of other attributes such as population density and composition, environmental impact, historical development and a range of cultural and symbolic dimensions. Spatial pattern is a fundamental tool for the urban project, since it invites hypothesis formulation." A "... problem for urban remote sensing is how to segment urban areas which are homogeneous with respect to built form (e.g. type of housing) but which contain a range of site layouts. (PESARESI, M. & BIANCHIN, A. 2001, 63-64).
• Analysis of the urban body in terms of building structure densities: Focusing on an ‘area of interest’ (AOI) or a single ward, the proportion of the polygons of the three distinctive classes can be calculated from all cumulated polygon areas. Furthermore, the full area size of the ‘AOI’ or ward expansion divided by the area size of all polygons results in a quotient which is a measure for the building structure density ranging from zero to one. (KAWKA, R., 1998, 65).
• Satellite-based generation of an urban boundary image: An urban mask can be processed in order to disintegrate undeveloped areas on the fringe of Gombe from the clustered urban area. (OLBRICH, G., 1990, 67) For this a reclassification is necessary. All classes representing buildings and paved roads are merged into a joint class with the value one, the other classes are ascribed the value zero. After utilising a majority convolution filter with a kernel 7x7 followed by a buffering procedure, a mask of the urban area is produced. This mask serves as a provisional result and can be used for many thematic mapping purposes e.g. to clip with other thematic maps in order to obtain spatial information only within the shape of the urban boundary.
• Providing thematic information for environmental impact assessment: Instead of reclassing trees and unsealed surfaces with the value zero, they rather form two other groups to inform about the quantity of vegetation indices and the correlation between sealed and unsealed areas. (ACHEN, M., 1991, 113). This influences the infiltration of rainwater and the surface run-off, causing a falling watertable on one hand, and extreme gully erosion especially in Gombe on the other, leading to disastrous events every year.

Figure 8: Classified map of Gombe / Nigeria; based on Satellite Remote Sensing

In order to gain even more accurate results, the classes describing the age of the roofs should undergo detailed calibration before progressing to an analysis of the whole urban scene. Nevertheless, it has been attempted to classify the IKONOS scene, utilising the modified signatures as described in chapter 3. The results can be viewed in Figure 8.

The significant advantage of the process as described in this paper is that it allows a quick monitoring of the urbanization process by identifying the build-up density and the spatial and temporal patterns of the building structures. With indices such as the spatial homogeneity or with the help of the polygon sizes within one class, an approach to a quantitative analysis can be attempted.

It could be proved that the age of the buildings correlates with the intensity of the reflection in the blue spectrum, which in turn effects the value of the respective pixels. The discovery of this correlation allows a specific classification and interpretation, and enables the researcher to gain a broader view of the situation. The results, then, can be linked with demographic and socio-economic background information and therefore contribute to a better understanding of the intersection between the processes of development and transformation. Thus, the information gained might be crucial for further urban planning in Africa.

Achen, M., 1991. Nutzungsmöglichkeiten von Satellitendaten für eine ökologisch orientierte Stadtplanung: Untersuchungen über den Flächenanteil der Vegetation und den Versiegelungsgrad am Beispiel Heidelberg, unpublished Ph. D. Thesis, University of Heidelberg, Heidelberg.

Balzerek, H., 2000. The new State Capital Gombe; In: Berichte des Sonderforschungsbereichs 268, Kulturentwicklung und Sprachgeschichte im Naturraum Westafrikanische Savanne", Frankfurt am Main, pp. 132-138.

Breuer, T., Gläßer, C., Jürgens, C. (Eds.), 1997. Fernerkundung in urbanen Räumen. In: Regensburger Geographische Schriften 28, Regensburg.

Champaud, J., Chaume, R. & Cherel J.P., 1993. Croissance urbaine, environnement et imagerie Satellite: Conakry, Ibadan, Nairobi, Ouagadougou, Xaoundé. Institut Francais de Recherche Scientifique pour le Développement en Coopération. ORSTOM, Montpellier.

Cracknell, A.P. & Hayes L.W.B., 1991. Introduction to remote sensing. London, New York, Philadelphia.

Donnay, J.P., Barnsley, M.J. & Longley, P.A., 1999. Remote sensing and urban analysis. London.

Fricke, W. & Malchau, G., 1994. Die Volkszählung in Nigeria 1991. In: Zeitschrift für Wirtschaftsgeographie, 1994. H.3, 163-178, Frankfurt a.M.

Hildebrand, K., 2000. Das Bildungssystem in Nigeria: Eine Fallstudie aus Gombe/Nigeria, unpublished Diploma Thesis, University of Heidelberg, Heidelberg.

Jensen, J.R.,1986. Introductory digital image processing.

Kawka, R., 1997. Regionaltypische Probleme bei der digitalen Interpretation von Städten in der westafrikanischen Savanne. In: Breuer, T., Gläßer, C., Jürgens, C. (Eds.), 1997. Regensburger Geographische Schriften 28, Regensburg.

Kawka, R., 1998. Afrikanische Städte im Satellitenbild - die Beispiele Bobo-Dioulasso/Burkina Faso und Maiduguri/Nigeria, unpublished Ph. D. Thesis, University of Heidelberg, Heidelberg.

Lillesand, T. M. & Kiefer, R. W., 1997. Remote sensing and image interpretation. New York.

National Population Commission, 1998. 1991 Population census of the Federal Republic of Nigeria, Analytical report at the national level, Abuja.

Olbrich, G., 1990. Abgrenzung des Siedlungsbereichs von Heidelberg mit Hilfe von Landsat 5 TM Daten, unpublished Diploma Thesis, University of Heidelberg, Heidelberg.

PESARESI, M. & BIANCHIN, A., 2001. Recognizing settlement sructure using mathematical morphology and image texture. In: Donnay, J.P., Barnsley, M.J. & Longley, P.A., 1999. Remote sensing and urban Analysis, 56-67, London.

Richards, J.A., 1994. Remote sensing and digital image analysis. Berlin.