Urbanisation

ABSTRACT

This paper demonstrates the value of satellite imagery in detecting changes in land use on the fringes of urban areas. Archived satellite data of the same geographical area, recorded over the last decade, was used to identify changes in settlement patterns. The urban growth was identified as informal low cost housing during field work. House counts were also performed in the affected areas. These areas were mapped to determine their extent in hectares. Housing densities were determined to estimate population increase. This project indicates that land use change detection from satellite imagery is a useful planning tool in the provision of services, utilities and infrastructure.

1. INTRODUCTION

The Satellite Applications Centre (SAC) is the custodian of a large archive of digital earth observation data systematically obtained from various space borne sensors, since the late 1980s. This data has been used in a variety of applications such as geology, agriculture, forestry and land use mapping. Satellite imagery is useful to determine and map geological type and structure to assist geologists with field surveys. The identification of Pinus and Eucalyptus species in forestry plantations and mapping of these areas from imagery are often used in forestry management. Similarly, satellite data is used in agriculture to map different crops and predict yields. In metropolitan areas, changes in land use on the urban fringes can be detected through satellite imagery.

As the South African population grows, changes in land use are accelerated, especially in urban areas. Urbanization has an important influence on the spatial distribution of land use. The result is that land is becoming a scarce and valuable commodity. Effective land use is therefore necessary for the optimal functioning of administrative, economic and social activities of communities. Many land use practices show typical patterns that are static for many years. This is especially true for agricultural patterns in rural areas. On the contrary, urban structures are dynamic, and spatial morphology, population structure and activity patterns are in a constant process of change and growth. As cities grow their land use patterns change. Nowhere is this change more dynamic than on the urban-rural fringe.

Urban planners and policy makers not only need current information on changes in land cover but require that it also must be accurately related to changes in land use. Furthermore the impact of these changes must be analysed within appropriate frameworks. As the changes in land use are a function of the economic, social, political and ecological context in which they occur, the impact of the changes must be assessed against these backgrounds (Green et al. 1994, p332). Effective urban development planning requires accurate data that contains the current land-use situation, the rate of the change and the boundaries wherein the change is occurring. Data for existing and historical land uses (because of their strong influence on future land use) plays a crucial role in the development of a land use and land development plan. This land use data provides a base against which to measure change, for developing land use policies and for monitoring their effectiveness (Gardner et al. 1977, p101).

Satellite imagery is becoming one of the most important sources of data for Geographic Information Systems (GIS). One of the major contributions of satellite imagery is the historical documents that are instantaneously being created at the time of sensing (Reeves et al 1975, p1824). It is also considered a flexible and fast way to obtain digital data as soon and as often as desired. It is a potentially powerful means of monitoring changes in land use at high temporal resolution and providing timely and accurate data.

1.1 Problem Statement

South African cities are experiencing an influx of people from the rural areas on a previously unknown scale, looking for work and a better quality life. Because of this rapid growth on the urban-rural fringe, planners and policy makers lack accurate, timely and cost-effective land cover and land use data which is most essential to make decisions concerning land resource management.

The urban landscape is one of diverse land-use, concentrated activity and complex economic and social function. This landscape is also the most modified by man's activities and the most crucially affected by both human and natural activities. Any natural or human induced disasters that occur in such densely populated areas would affect a far greater number of people than any hazard that occurs in less densely populated areas. This was graphically demonstrated in South Africa in recent years when squatter camps that were built within flood lines, were destroyed with loss of lives.

Land cover and land use data is needed by planners to make decisions regarding suitable land for development for urban use, i.e. to provide facilities such as accommodation, employment, public services, transportation and open space for recreation. Land use change analysis is considered to be the foundation to planning and regional policy programmes (Michalak 1993, p28).

The tern land cover is often used in association with land-use, but the two terms are not synonymous. Land cover describes the vegetation and man-made constructions covering the earth's surface. These are all directly visible from the remotely sensed imagery. Land use refers to man's activities on and in relation to the land, which are not necessarily directly visible from the imagery. (Lo 1991, p227)

1.2 Objective

The objective of this paper is to demonstrate that satellite imagery can be a fast and cost-effective means to detect land use change. Policy makers and planners needs accurate, timely and cost-effective data to plan and control urban and regional growth, the equitable distribution of services, the provision of infrastructure and to predict future demands. The rapid growth of our urban population and financial constraints are the reason that the main sources of information (i.e. official statistics, maps, etc) are usually outdated. What is essential, is data which can be obtained readily and cost effectively and which will be regularly available in the future so that change in the land cover and land-use can be detected for any needed updating.

The results of change detection analysis using satellite imagery may be interesting and revealing, but by integrating this data with other spatial layers (topography, political boundaries, property ownership or any other applicable data) this information gains powerful analysis capabilities.

2. DETERMINING LAND USE CHANGE

Satellite images used in this example for change detection, were from the SPOT SPOTPANCHROMATIC sensor. To enable reliable comparison the images were accurately georeferenced and their spectral signatures adjusted through the use of histogram matching techniques. A difference map was then produced by performing temporal image differencing. This technique is best applied on images of the same sensor and spatial resolution (Lillesand and Kiefer 1994, p261). Processing was performed on a Silicon Graphics Workstation, using the Erdas Imagine 8.2 image processing software, including a vector module.

2.1 Geographical Area

The Winterveld region north of Pretoria is experiencing rapid urban growth in the low cost informal sector. Settlements like Mapobane, Klippan, Soshanguve and Winterveld are the most important urban developments in this area. The cadastral farm boundaries were overlayed on the imagery to assist in demarcating specific areas of growth. The urban area on Rietgat 105 JR, just north of Soshanguve showed prominent growth and was selected for this study (Figure 1).

The growth of this region can be traced back to the historical political and centrifugal forces that were in place during the last fifty years. Traditionally this area can be identified by a lack of regulatory bodies and policies concerning land values, land use zoning, ownership and house structures. Hardly any formal infrastructure exists and housing usually complies with the minimum standards. The recent influx of people from rural areas has created a substantial informal low cost housing settlement. This area is characterised by a well developed commuting system, where people travel daily/weekly for work between Pretoria and their home.

2.2 Imagery and Acquisition Dates

SPOT Panchromatic imagery was chosen for this project due to its inherent high geometric quality and detail yielded by a spatial resolution of I 0 metre pixel size. In addition the SAC has a long time series of SPOT images available in their digital archives. Since the images are available in digital format, the image processing can be performed rapidly. The acquisition dates of the two images (132-401) covering the Winterveld area dated 16 December 1989 (Figure 1) and 24 July 1995 (Figure 2) respectively, approximately five years apart.

2.3 Geographical Registration

Accurate spatial registration is essential for any image that will supply information for use in a GIS as this allows for the measurement of distance and area as well as to determine position (Jensen 1986, p269; Ridd 1991, p167). The first image was registered using the image-to-map technique with ground control points (GCP) as described by Richards (1993,p58). Ground control coordinates were obtained from 1:50 000 topocadastral maps of the Surveyor General, and these points were correlated to corresponding features on the image. A Root Mean Square (RMS) error of 0,46 pixel was achieved, which is less than half a pixel. The image was then registered to a Transverse Mercator projection using the Clark 1880 ellipsoid and 29 degrees E as the longitude of origin.

The second image was then registered to the first image by performing an image-to-image procedure as described by Richards (1993, p66). Lillesand Kiefer(1994, p621) emphasises that this procedure is critical to accurately detecting land use change. The registration accuracy should be between 0.25 to 0.5 of a pixel. In this case the RMS value was 0.37 which is less than half a pixel and acceptable.

2.4 Histogram Matching

Histogram matching is necessary to modify the histogram of one image to fit that of the reference image as close as possible. This creates a new histogram for the first image and moves the distribution of brightness values of one image so that it has the same visual appearance as the reference image (Richards 1993, p102). Histogram matching prepared the images to perform temporal image differencing since it adjusted the grayscale values of the two images so that one resembles that of the other very closely. The 1989 image was used as the reference image since it had a typical Gaussian type histogram and it was acquired in summer when vegetation was growing well. The 1995 image was matched to the reference image (1989) transforming its histogram to fit as closely as possible to that of the reference image. After performing this process the 1989 and 1995 images appeared more similar, which is essential for detecting land use change.

Change detection performed on two images recorded in different years yields the best results when images are of the corresponding dates in the year, commonly known as anniversary dates (Lillesand & Kieffer 1994, p621). However, it is not always possible to use images of the same anniversary dates, since available images in the archive are very often of different dates. This was indeed the case with the Winterveld area north of Pretoria, where the only available images with a 4 - 6 year time difference were of December 1989 and July 1995. Histogram matching of the two images is used to se the variation in sun angle and seasonal differences. This technique is especially useful for change detection and mosaicking (Smith et al. 1995, p164).

2.5 Image Differencing

Once the images had been accurately registered and histogram matched as described, image differencing could be performed. This process is based on the difference of the grayscale value of corresponding pixels in the different images. Grayscale represents the range of radiance present in the imagery and ranges between zero and 255, where zero (low radiance) represents black and 255 (high radiance) represents white. This is also referred to as the digital number (DN) of a pixel that represents the level of radiance for a given pixel (Curran 1986, p183).

During the image differencing process the actual grayscale value of a pixel was subtracted from that of the corresponding pixel in the later image. Where no change occurred, the difference in grayscale values was low and close to zero. In locations where urbanization had taken place, the grayscale values were high in the second image and hence a high positive difference between corresponding pixels occured.

The difference in the pixel DN value is clearly visible in Figure 3, where a small area of both images is shown. The left image (1989) was recorded when urban development had just started and the image on the right (1995) thereafter. The two forms below the images contain the DN value for the same pixel in both images. The position of the pixel is indicated by the intersection of the white lines on both images in Figure 3.

2.6 Change Detection

The process of image differencing of the two images produced a difference map with the areas of change detected between 1989 and 1995 shown in red in figure 4. The areas of chance on the cadastral parcel of Rietgat 105 JR were mapped from the difference map, to produce a vector map indicating land use change. The land use change detected could include different types of land use and the vector map was therefore overlayed onto the 1995 image to ensure that it was indeed urban growth.

2.7 Estimating- Urban and Population Growth

Certain areas were selected from the imagery for fieldwork. The actual sizes of these sample areas were measured digitally on the image. These areas were visited to do house counts. The house counts (HC) in the sample areas and the size (S) of the area were used to determine a housing density (HD) where
HD = HC/S.
This resulted in a housing density of 9.5 7 houses per hectare, as shown in Table 1. From this density and the increase in urbanised area (UA) the total houses (TH) were estimated. The increase in the urbanised area from 1989 to 1995 was 2 032 hectares (Figure 4 & Table 1).

Once the total houses were known, the population increase (PI) could be established, using a factor of residents per house (R). The average number of residents per house used in this estimation was five. The following equation was used to establish population
PI = TH x R.
The estimated growth in population on the cadastral unit of Rietgat 105 JR for the area from 1989 to 1995 was 97 230 (Table 1).

3. DISCUSSION

The result of this project shows that digital imagery has the ability to provide information on urban land use change timely and with ease. The information extracted as a vector layer could be optimised if imported into a GIS. The area studied is small and this technique should be applied to a larger area to monitor the fringes of a metropolitan area. In this example temporal difference of approximately five years was used, but this could be adapted to a yearly frequency or determined by user requirements. Although it is not ideal that the images were from different seasons, the application of change detection on the urban fringes was demonstrated clearly in this project. However, if this technique is applied on a larger area, imagery should be selected on anniversary dates or the same season to ensure a high level of accuracy.

4. CONCLUSION

Satellite imagery provided an essential input to this project to establish urban growth in a rapid and efficient method. The land use change detected from satellite imagery in this project can provide urban planners and strategic decision makers with a timely and accurate basis for supplying required infrastructure and services. Information required through this method is particularly valuable in areas of rapid growth in the informal urban fringes, as there is often a lack of data on these areas. This information is especially useful in a GIS as an additional layer where it could be used to forecast future demand on utilities and services for a specific geographical area based on the population growth.

REFERENCES

Curran, P.J. (1986): Principles of Remote Sensing. John Wiley & Sons Inc, New York.

Gardner, W.E., Carter, P., and Smith, T.F. (1977): Digital Analysis of Multispectral Aerial and Landsat Data for Land Use Planning in Britain. In Peel, R.F., Curtis, L.F., and Barrett, E.C.

(eds.), Remote Sensing of the Terrestrial Environment, Colston Papers no.28, Butterworth & Co.,London, pp 96-107.

Green, K., Kempka, D., and Lacley, L. (1994): Using Remote Sensing to Detect And Monitor Land-Cover And Land-Use Change. Photogrammetric Engineering & Remote Sensing, 60(3) March, pp 331-337.

Jensen, J.R. (1986): Introductory Digital Image Processing - A Remote Sensing Perspective. Prentice-Hall, New Jersey.

Lillesand, T.M., and Kiefer, R. W. (1994): Remote Sensing and Image Interpretation. John Wiley & Sons Inc, New York.

Lo, C.P. (1991): Applied Remote Sensing. Longman Inc, New York.

Michalak, W.Z. (1993): GIS in Land Use Change Analysis: Integration of Remotely Sensed Data into GIS. Applied Geography, 13(1) January, pp 28-44.

Reeves, R.G., Anson, A., and D. Landen (eds.), (1975): Manuel of Remote Sensing. American Society Of Photogrammetry., Falls Church.

Richards, J.A. (1993): Remote Sensing Digital Image Analysis - An Introduction. Springer - Verlag, Berlin.

Ridd, M.K. (1991): Spatial and Temporal Scaling Issues Related to the Integration of GIS and Remote Sensing. In Star, J.L. (ed.), 7he Integration of Remote Sensing and Geographic Information Systems, ACSM-ASPRS Annual Convention. American Society For Photogrammetry and Remote Sensing, Maryland, pp 167-173.

Smith, C., Pyden, N., and Cole, P. (eds.), (1995): ERDAS Field Guide. ERDAS, Inc.

Figure 5 The images below show the rapid urbanisation characteristic of the Winterveld region, (60 km northwest of Pretoria) between 1989 and 1995. Dates of photos are 16th December 1989 and 24 July 1995.
Source: Satellite Application Centre.