Synergism of CIR aerial photographs and SPOT images for monitoring of project housing in Bandung, Indonesia

Synergism of CIR aerial photographs and SPOT images for monitoring of project housing in Bandung, Indonesia

Drs. Victor F.L. Polle
ITC, P.O. Box 6, Enschede, 7500 AA Enschede, The Netherlands

Aim and method of the research.
Until 1986 only aerial photographs or map made by extensive ground surveys could provide an overview of the urban development. Although satellite remote sensing had already started in 1972 with the LANDSAT MSS from the USA, for a long time we could only order images with a resolution of 70 meters. These images were too vague to be of any use for the planning of high density cities in developing countries.

With the arrival of Landsat TM the resolution dropped to 30 meters, which was a considerable improvement. And in February 1986 the /French arrived on the market with the Spot images of still higher resolution. Two types of images could be ordered:

colour resolution 20 meters or Panchromatic black & white resolution 10 meters.Suddenly satellite remote sensing also became interesting for urban planners.Urban planners usually order aerial photographs of a large scale: 1:5,000- 1:20,000, enlarged sometimes to 1:2,000 or 1:1,000. Occasionally smaller scales, up to 1:50,000 were used for large metropolitan areas (e.g. Colombo 1981). The resolution of SPOT images is comparable to aerial photographs of scale 1:200,000 or smaller. Town planners will have to become familiar with these smaller. Town planners will have to become familiar with these smaller scales, which require a different interpretation technique. Since 1986 we have been doing research with SPOT images of Bandung in Indonesia. (1) The planning of Greater Bandung is the responsibility of two ministries (Public Works and Interior Affairs) and of two authorities (the Municipality and the Region). Aerial photographs of the area have been taken recently to make orthophotomaps at scale 1:2,000.

A part of Java was covered in 1981 by CIR photography at scale 1:30,000. But these photographs have probably not rully been used, although they covered the Greater Bandung area, and even the whole Region Bandung, because it is not well defined which organization is responsible for the mapping of Greater Bandung. These images were, however, of great help during our research, especially because their false colour is comparable to the SPOT false colour.The aim was to assess the usefulness of SPOT PAN-XS images in combination with already available CIR aerial photography for the monitoring of the housing projects, which are spread over a large area on many big small sites, and which until now have not been properly mapped in their spatial distribution. See fiigure1. What is the efficiency and the accuracy of this method?

Figure 1: The spread of the Bandung project housing over an area of 19 km by 14 km, as identified on a SPOT PAN-XS image scale 1:25,000. The areas north of the generalized 700 m line are for 75% from before 1981 and have been identified on CIR aerial photography 1:30:000 from 1981 (See table 1).Note that most project housing is found outside the old municipal boundary and therefore not built under the planning jurisdiction of the municipality. Only in 1989 the area of jurisdiction of the municipality. Only in 1989 the area of jurisdiction of the municipality has been enlarged. Still some major developments are taking place outside the 2989 boundary.

The photographic products of SPOT
A city like Bandung, population 1.5 m., is not recorded by SPOT regularly. Recording is only done if a "programming request" has been placed with SPOT IMAGE in France. An extra problem for Indonesian cities is the almost constant cloud cover. It may take months before a successful recording can be made. But for our research, which started in 1986, we could now use an Xs recording from 6 July 1986 and a stereo pair in PAN from 29 June 1987 and 21 August 1987. For ground truth we obtained colour small format aerial photographs 1:18,000 from August 1987.

For the urban planner the photographic products from spot (based on pre-processing level IB) are the most practical until now. On a photograph the size of a map, a whole city can be seen, together with its environment in which it is likely to expand in, let us say, the coming fifteen years.

For the colour images the maximum useful scale for the enlargement is 1:50,000 although these enlargements do not look sharp anymore. The usual colour "translation" is the false colour. The vegetation becomes red, while manmade objects and bare soil become yellow, blue or green in a variety of shades.

Black & white images can be enlarged to 1:25,000. It is also possible to combine the XS and PAN images (with image processing equipment like ERDAS or ILWIS) to coloured images (PAN-XS) with a resolution of 10 m x 10 m which can also be enlarged to 1"25,000.

The enlargements should be interpreted with the naked eye. Use of a magnifying glass is meaningless because the picture elements (pixels) become so much larger than the resolution of the eye (approximately 0.111 mm) that our brains cannot perceive the terrain objects anymore.

Resolution and the size of the urban objects
The SPOT photographs 1:50,000 or 1:25,000 are much less sharp than aerial photographs of the same scale. If we loo8k at air photos with a magnifying glass we can see the individual houses even in the most densely built-up kampung. We can see details like the pitched roof, the flat roof extension overt a terrace and the occasional trees in the gardens.

The scene elements of SPOT form a regular grid over the city. The position of the grid is randomly selected. Also the pixels are therefore representing a random part of the terrain. In the city the terrain characteristics change over short distances, often less than 10 meters. The largest part of the pixel set will be mixed pixels. That means that the reflected light from a scene element comes partly from buildings, partly from vegetation in gardens and partly from the roads or railroads. We cannot see small objects separated from their environment.

Large objects however, like an airport, a golf links, a sports stadium, are represented by more than one pixel. We can see their shape. Also the outline of objects like "a residential area" or "an industrial area" can be seen in this way.

Recognition by the shape of the objects
For an interpreter of urban areas the most normal way to recognize objects is by looking at their shape and size. But we can only do that when the objects are big enough or near enough to be much bigger than the resolution of our eyes. They should "occupy" more than one "pixel" to have a distinct shape that can be recognized

For recognition on a SPOT image "large objects" are objects measuring, let us say, more than 100 m x 100 m. Then they can be recognized by their shape. In the colour images they are represented by at least 5 pixels by 5 pixels, each with its own colour and brightness. In PAN images or PAN-XS images they are even represented in 10 pixels by 10 pixels. This will reveal even more of the shape. We say then: the image is clearer, sharper. Let us consider some examples. An airport will be recognized by its runway which is 2-3 km long. A golf links will be recognized by the elongated lawns separated from each other by rows of trees and bushes, which are darker (red) because of more light absorption. Also sports stadium have a characteristic shape (two half circles connected by straight lines) that will be recognized.

Big housing projects, once we have studied their shape, can also be recognized by their shape every time that we look at the picture. We can also detect changes in their shape if we compare pictures from different dates. In this way change detection, or urban growth detection is made possible.

Most urban objects however are smaller than these examples and cannot be recognized by their shape. We have to use colour, tone and texture as clues to their identity.

Recognition by the colour and the texture of the object (SPOT-XS)
If we move away from an object its details cannot be seen anymore. First the small details "merge" with their environment and gradually even the larger details disappear. The details make with their environment a "texture", a surface characteristic, which depends on the composition of the surface elements.

With a SPOT image we are 800 km away from our objects. We see whole areas as a single object. We see its outlines (shape), separating it from its surroundings, but that shape is not characteristic. Residential areas for example may show an endless variety in shape. In order to determine its identity we have to rely on its colour and texture.

Unfortunately colour in itself is not too helpful. Residential areas may have many different colours, depending among other things on the density of housing, the amount and type of vegetation, the colour of the roof tiles and the width of the streets. It is more the combination of color and texture that makes identification possible. (2)

Housing areas are characterized by a high frequency texture. The colour and brightness (tone) varies from pixel to pixel. Once we become sensitive to these differences, we can recognize residential areas. With training we may even be able to distinguish between different types of housing area. A rough classification of housing areas may be possible. But this depends also on good local knowledge.

In the panchromatic (black & white) images we miss, of course, the colour. only the brightness and its variation is left as help for recognition. But it is possible to distinguish between smooth and rough textures, and therefore between types of land use.

Recognition by the tone and the texture of the object (SPOT PAN)
With SPOT PAN we have a higher resolution (10m) but without colour. Interpretation is often easier because of the higher resolution, more shape can be seen, but some urban areas are more difficult to interpret without the help of a color difference.

The resolution of PAN images is so much higher that in housing areas the texture from the colour images is often replaced by a tone pattern that can be interpreted as the "street pattern". Counting the number if darker lines gave the number of streets: the correctness of this information was confirmed by counting the streets on air photos. Also smaller objects are recognized by their shapes. Industrial buildings get a more rectangular outline, compared with the curved outlines on the color image.

Interpretation through synergism of aerial photographs and SPOT images : the Example of identifying project housing

In Banding's fringe large new residential areas have been developed for different types of houses. These developments data from the last 10 to 15 years. Some of these types are project housing (houses built in series), others real estate (individually built houses) and kampung. A further subdivision, e.g. on income level is possible.

Our research aim was however the mapping of the project housing of Greater Bandung with the help of CIR photographs of 1981 and a geo-corrected SPOT PAN-XS image 1:25,000 from 1987.

The mapping could be carried out in a short time. This was possible because in areas where interpretation was difficult, that is in the project housing from hbefore 198111, we could use the 1:30,000 false colour photographs for a first identification. The difficult areas are mainly in north Bandung where project houses have been built on sloping terrain on small sites and where considerable changes to the to the houses have been made by the people themselves. With the identified project areas we learned the typical colour and texture of this type of development.

On the aerial photographs from 1981 about 50 of these project housing areas could be identified. Many of these areas are very small (less than 10 ha) and apparently on pieces of land which do notallow further extension.

At least 30 areas did not show any growth between 1981 and August 1987 (the date of the last SPOT image used). The SPOT image was clear enough to state that with a great certainty.

In about 40 new areas projects started in the period 1981-1987, while 10 areas have expanded.

All areas with project housing (152) have been plotted on an overlay map on the PAN-XS image scale 1:25,000 and have been measured separately. The aggregated data have been given in table 1. It may be interesting to note that the interpretation of the image and the measurement of the delineated areas took only 2 Mondays, which proves the efficiency of the SPOT PAN-XS in combination with small scale older photographs for this kind of study.

The housing areas do not look all the same in the false color image. They show a pattern of lighter and darker spots. And their overall color varies from whitish, through bluish to greenish. These color differences reflect among other things stages of development. Some areas are still building sites. Other project housing areas are covered by houses, which either are not yet inhabited, or have just recently been occupied by renters: again others have been inhabited some time shown by abundant vegetation in the gardens and new additions made to the houses.

Verification in the field indeed revealed that areas with a different colour and texture differed as to the stage of development. But also the roof material, the spacing of the houses and the garden size proved to be different.

The SPOT XS image seems to show main roads in the area, but only when a strip of green vegetation makes them appear wider. The normal streets which divide the residential areas in "blocks" are not visible. They are too narrow to influence the colour of pixels enough to make a line discernable.

Comparison between SPOT XS and SPOT PAN reveals that the higher resolution of the black & white (or- in our case-the PAN-XS) image leads indeed to the detection of smaller structures and objects for this type of area. Direction as well as spacing of minor streets can be seen on the image.

Table 1: Development of project housing in Greater Bandung 1981-1987 in 2 sectors (Areas in ha.)

	Existing in 1981	Built between 1981-July 86 (5 year)	Built between July 86-Aug 87 (1 year)	Total
North Banding above 700m	202	68	1	271
relative	75 %	25 %	0 %	100 %
South Banding	180	368	118	666
relative	27 %	55 %	18 %	100 %
Total	382	436	119	937
relative	41 %	46 %	13 %	100 %

Information such as in the table 1 gives the planner an insight in and a quantification of the overalls development in Banding's project housing. Although south Banding's is considered as less favorable for housing (mosquitoes abundant in the paddy fields) a considerable shift to that area has taken place because of a lack of suitable land in the northern hills. In 1981 there was more project housing in the hill, north part (above 700 m) than in the south (202) ha against 180 ha). In 1987 the situation was greatly reversed (271 ha against 666 ha). While in the south construction continued in 1987 (118 ha), in the north it came to a standstill (1 ha).

Accuracy of area measurement
Local adjustment of the image to the topographical map makes plotting of the new residential areas possible with an accuracy in the order of .5 or .4 mm i.e. the usual width of a line drawn with a pen. For most planning and management purposes useful enough. For an accuracy test see table 2.

Table 2: New housing areas built between 1981 and 1987 in a test area of Bandung (180 ha),
delineated on SPOT PAN-XS (1:25,000, 1986/1987) and on a small format aerial
photograph (1:10,000, 1987).

Delineated area built 1981-1986	Area on SFAP ha	Area on SPOT PAN-XS ha	Difference %
1.	6.5	6.8	6.1
2.	11.7	13.5	11.15
3.	2.3	2.0	-13
4.	8.2	8.3	+1
5.	3.7	4.3	+16
6.	0.4	0.5	+25
1986-87
7.	8.6	9.8	+13
Total	41.4	44.6	+8

in table2 we see that for the total area of 41 ha the area delineated on spot is only 8% larger than the area as measured on the detailed small format aerial photograph for the smaller photograph for the smaller p[arts of the area the error lies between planning is fast developing areas this error is fully acceptable. of course the area are grass areas on the small format aerial photographs it is possible to delineate separately the included undeveloped land the difference between such net area and the grass area the test area or less than the growth in one year still fully acceptable for practical use.

The test area is typical for all new housing areas in the southern part of the city where housing projects have been developed in the former paddy fields conclusions for the test areqa may be expanded to all new housing areas in south banding the developing of the project areas between July 1986 and august 1987 could be interpreted from the combined image new project housing areas will have a light tone on the PAN image in the 1986 image these area could have the following colours .

So detection of light red or light blue gray areas usually not far from existing project housing indicated the newly built houses

No confusion with industrial areas
Several large industrial complexes are visible on the SPOT images as well as smaller individual industries they are characterized by their light tone almost white due to the high reflection of the light through the big roofs an industrial building is usually so large that more than one pure roof pixel can be found confusion with new project housing did not occur due to the difference in pattern could be studied first on the false color photographs 1:30,000 from 1981

Other forms of urban growth a subject for research.
Not only project housing and industry an be studied with spot images to monitor the urban growth also other urban land uses can be mapped and measured although with various degrees of accuracy and certainty the growth of industrial areas kampung areas real estate utilities are all examples orf land uses that garbage dumps are can be monitored .for most land uses we do not yet know what the signature on the spot images is we do not yet know whether theswe things can all be seen some further research in this field is the purpose of our current project in binding .

If we compare the spot images the other remote sensing product the aerial photograph the later is b y far superior through its much higher resolution wherever aerial photographs are available they should be used as much as possible how ever research both aerial photographs and the maps derived from them become outdated very soon in a dynamic city like Bandung.

New missions for aerial photography are expensive an can probably be ordered every five with years only in the years in between monitoring could be on done on spot images so that the strategic planning can be reviewed every year. Interpretation of new housing areas on Spot Pan XS the original is a false color image made by combing as 1986 XS with a 1987 PAN image enlarged to1:25,000 this illustration shows a detail in blackn and white enlarged to 1,12,500 the illustration shows a detail in black and white enlarged to 1:12,500.

Areas marked 1 are housing areas existing in 1986
Areas marked 2 and 3 were rural in 1986 but urban in 10987.
Areas marked 2 are light red : red from the paddy in 1986 and light from new urban in 1987.
Area marked 3 are light blue : blue from flooded paddy fields in 1986 and again light from new urban in 1987.

The brightness of the light red and blue depends on the new urban objects usually land prepared for developments a street pattern on the textured of new housing caused by the mix of house garden and street in mixed pixelsthe light red or blue areas are islands in a darker sea and are therefore easily detected. The areas marked 4 and 5 still rurall in 1987n they are dark red indicating growing paddy or dark blue because of flooded paddy land.

Conclusions

Interpretation of housing areas on spot XS or PAN XS is greatly enhanced if small scale CIR aerial photography of some years earlier is available the higher resolution of the photographs makes identification much more sure even in areas mixed use the housing areas thus identified on spot as training areas for learning the typical color and texture.
Mapping of the new housing areas of the city of Bandung with a population of 1.5 million could with this method by accomplished by one experienced interpreter in two working days.
The delineations of the new housing areas in SPOT PAN -XS ro be accurate enough for metropolitan planning purposes when compared with delineations on color small format aerial differed not more than 15% aerial photography with a scale of the gross housing areas differed not more than 15% for areas larger than 4 ha fore the test area is compared difference found was 8% if the gross housing area from spot is compared with the nett housing area from the airports then the error for the test area as a whole in 14%.
Incongruence between the area of jurisdiction of a municipality and its actual urban area is found in many cities spot can be very useful for identification of housing areas outside the area outside the area of jurisdiction of a planning authority reliable maps derived from spot may influence positively decisions formore realistic jurisdiction boundaries.

References

V.F.L. polle monitoring and measuring human settlement changes by remote sensing methodsa case study with spot images of bandung Indonesia itc urban survey and human settelment analysis report fromther bcrs beleids commsssie remote sensing /policy commission on remote sensing from the nether lands government BCRS 88-16c project no 4533 t03 .11 December 1988 copies of this report be made available.
V.F.L polle detection of city growth areas using spot image interpretation survey pepper for the 16th congress of the internatoional society for photogrammetry and remote sensing kyoto Japan 1988 in international archives of photography and remote sensing volume 27 PART B9 PP VII 3564.
V.F.L. polle spot image interpretation for information on human settelment growth in proceedings of the asian conference on remote sensing 114 Oct 22-27 198 jakarta indonesia.