GISdevelopment.net ---> AARS ---> ACRS 1989 ---> Mapping from Space

Multi-Sensor image interpretation applications in Sabah

F. G. Bercha, D.H. Currie, J. A. Dechka, P. Fuenning
Bercha International Inc.

P. Jordan
Sun Malysia Petroleum Company


Abstract
In conjunction with the conventional analog and digital geologic interpretation of Synthetic Aperture Radar (SAR) imagery being applied in the exploration of Sabah, Malaysia, certain multi-band digital applications were required in order to extract information on subtle geologic features, cultural features, and land use. This information was required primarily in support for the planning of exploration program logistics, specifically, a data set consisting of digital SAR and SPOT data, which were co-registered and interpreted for the requisite information. Digital techniques applied to the SAR and multi-spectral SPOT images included filtering, radiometric enhancements, and statistical analysis. Results of a ground truth program were used to corroborate some of the findings. Generally the information produced on vegetation and terrain was of high accuracy for the purposes of logistic planning. Detailed information on the culture and transportation infrastructure: including roads, rivers, villages, and other settlements was found to be of high accuracy. The multi-sensor data set was found to provide an order of magnitude more cultural information than the SAR alone. The techniques utilized form a good basis for the generation of strategic and resource mapping information in accessible tropical terrain.

Introduction
Persistent cloud cover and inaccessibility are often associated with tropical environments. Sabah, Malaysia is no exception. A request was made by Sun Malaysia Petroleum Co. (SMPC) to extract cultural information and geological features from SAR and SPOT data for the defined area of interest. SMPC felt that the use of remotely sensed data could assist with the location of potential anomalous

Structures indicative of geologic closures. Also these data were assessed in the planning of exploration program logistics for the generation of strategic and resource maps for the area as well as accessibility routes. This paper focuses on the cultural interpretation conducted using these data. Analysis of SAR data for the purpose of geological interpretation can be Bercha et at. (1989).

Study Area
The study area is located in Sabah, on the island of Borneo, and covers approximately 11, 300 square kilometers. This area of Borneo is well drained, densely vegetated and mountainous with forestry and resource recovery providing the main industry. A geological description can be found in Bercha et. al. (1989). There are numerous settlements and towns scattered throughout this region.

Data
The cultural and geological interpretation of the Sabah area is based on the SAR data acquired in early 1989. These data were acquired in the X-band with 6m resolutions. In all, 13 flight lines were acquired, providing stereoscopic coverage of the area. Data products supplied were in the form of 1:50,000 and 1:100,000 analog strips and computer compatible tapes.

An image produced by the system Pour 1 "Observation de la Terre (SPOT) satellite was also acquired. This multispectral image was recorded on April 17, 1988 with 20 m resolution. The image quality is good and contains scattered cloud cover of less than 20 percent the entire scene.

Ancillary data included one general geological map covering the whole state of Sabah (1:500, 000) and 14 topographic maps at a scale of 1:50, 000 covering half of the SARR flight lines. These maps had been annotated by field crews with observed dips at various accessible locations. In addition, photographs and a video tape, providing partial visuals coverage of the location was provided to acquaint the interpreters with the area.]

Methodology
  • SAR Data
    The SAR digital data were loaded into the PCI EASI/ACE Image Analysis System. These data were then mosaiced using both image to image and map to image registration to provide a 1:100,000 image product. The area was divided into a north and south section simply because the images exceeded the maximum size constraints permitted by the digital scanner for hard copy output. Each mosaic was intepretedfor cultural features using manual proceduress.

  • SPOT and SAR Data
    When the SPOT image was displayed on the image analysis system it was decided that an intensity reduction for the visible bands should be conductedto reduce the effects of the high humidity haze in the tropics. A training area of water was created in the infrared band and compared to the visible bands. The difference in the means of the IR-band and the visible band data wasw determined to reduce the intensity of the visible bands.

    Once this was complet, the SPOT and SAR data wer registered in the areas of overlap. The SAR scene was reduced from 6metre to 20 metre resolution to facilitate the registration of the two data sets.

    When these data were merged, various enhancement and transformmms were initiated in order to display the topogaphic featuresssexhibited by the SAR data and to maintain the multispectral nature of the the SPOT imagery. The best result was obtained using the intensity hue and saturation conversation (IHS) (Drury, 1986). The intensuty was represented by the radar data and the hue and saturation were represented by components one and two respectively. Once this procedure was completed a look-up table was created to enhance the subtle features displayed on the imagery such as the roads, settlements and topographical features.
Results And Discussion
  • SAR Imagery
    The SAR imagery was useful for providing general information regarding roads, drainage, some settlements and forestry activities. In certain cases, ancillary data sources were useful for identifying roads, especially where the presence or absence of roads became questionable. However, the features of Interest had to run in a direction parallel to the aircraft flightlines to be detected. This is a dcirect result of the radar pattern where by shadowing effects of features highlight cleared areas, such as roads, by changing contrast. If the feature was perpemndicular to the signal, it would not generally be visible because there would be no shadowing effects and therefore minimal contrast change.

    Road networkks included both major routes and secondary access roadss. Major rroutes were verified where 1:50,000 mapcoverage was available and by using the 1:5000,000 geological map. In the area selected, for its potential geological closure, roads were mapped under neglecte category rather than differentiating between primary and secondary routes. This was done to avoid a visual overload of information since rthe structural geology was of primary importance. Figure 1 illustrates an example of an interpreted area, with Table 1 representing the associated interpretation key.


    Figure 1- SAR Stereopair of a cultural interpertation

    Table 1 Cultural Interpretation Key

    Drainage networks included major rivers and associated trirbutes. Tertiary drainage was only mapped if it was well defined and visible. As expected, drainage was often situated along lineaments and fault trends although this was not always observed. . The extent of potential boat passage could not be interpreted due to limitations in the scale and limited ground truth data.

    Areas interpreted as a settlement or structures relating to some form of industry (i.e. forestry) were denoted under the heading of settlement. In all cases, these aqreas were interpreted based on their characteristic relative brightness,s a direct result of high reflectance.

    The use of radar data alone sas limited in the cultural interpretation process even though the texturally variable signatres of the image were importrant to the interpretative process. The areas providing providing significant textural contrast allowed for more accurate interpretations to be made while areas of limited contrast made interpretation difficult without the benefit of alternative sources of information. An example of this was observed in clear out areas near topographically level tidal flats. In these areas road networks were often difficult to differentiate due to the lack of contrast, resulting from the absence of shadowing.

    Further, in mountainous area, there was a loss of information resulting from radar shadowing. To compensate, these areas were interpreted and determined using the 1:50,000 topographic map coverage or the 1:500,000 Sabah geologic map (Ketiga, 1985). Where background information was not available , a "best estimate" was made in sitations of less than 2 km. Areas greater than 2 km were left uninterpreted simply due to the lack of available data. In most cases, the questionable areas occuredalong secondary roads.

    Although the radar imagery provided excellent topographical representation, in many areas it did not provide adequate data for the interpretation of road networks, drainage and settlements. As a result, a SPOT image was acquired for the area so that it could be merged with the radar imagery in order to assesswhether the results could be improved.
  • SPOT and SAR Imagery
    The SPOT image provided minimal topographic information except in areas of maximum slope but clearly provided better definition of roads, settlements and drainage patterns when ccompared to the radar image of the same area. The satellite image contained some cloud cover, however, the clcoud shadows were not a problem because of diffusescattering of light.

    The first step in the multi-sensor image processing was the removal of someof the atmosphere haze caused by the high humidity typical of thestudy asrea. This was done for each band by determining themean intensities of training areas located on larger water bodies. Since water is expected to have a near zero reflectance in the infra-red region the the difference in intensity for the training sites between the visible and infra-red bnands was calculated. This value was then substracted from each digital number in each visible band. This provided an acceptable, albeit incomplete, haze reduction.

    The second step was to register the SPOT image to the SAR moosaic. A standard polynomiAl transformation was used resulting in an RMS error of 4 pixels. Since the radar mosaic had not been corrected for terrain variation this was considered acceptable. As expected, the greatest distortion occured in areas of high relief.

    The result of the registration was a four band image at a ressolution of 20 m. Various presentations of thisdata set may be devised to emphasize selected characteristics. In order to compress the data into a three band image suitable for hardcopy presentation the following procedure was used. First, a principal components analysis was performed on the three SPOT bands. The first two principal comonents were used as hue and saturation for input to an IHS transform with the radar data used as intensity. Custom look-up tables were then developed for each of the output bands to produce a hardcopy product suitable for interpretation which displays both the geological and cultural features.

    The process described demonstrated small settlements, roads in the tidaal areas, selective logging areas and suitable drainage patterns were displayed which had been previously undetected on the SAR imagery. Therefore, the multi-sensor image clearlly showed the previously mentioned features in addition to the basic topography.
Conclusions
Radar imagery has been of tremendous value in the exploration of tropical environments primarily because of its cloud penetrating capabilitiessss. Multispectral sensors, on the other hand, have been limiteddue to the impenetrable cover of clouds and hight humidity haze, but providedbetter cultural definitions. When both types of imagery were used, the benefits of boh images were acquired.

As a result of this project several conclusions havebeen obtained.
  • Radar data provided excellent topographical representation and geological structure.
  • Radar imagery didnot adequately allow cultural interpretations in areas hidden by radar shadow and in areas of little relief such as tidal flats.
  • It was difficult to determine accurately if a small settlement was present on a radar image without ancillary data.
  • The SPOT image provided easy identification of cultural activity because of the greater spectral range allowing for increased detail as a result of the multispectral wavel.engths.
  • Roads, drainage, settlements and forest activities were all evident using the SPOT imagery, even in areas of little relief.
  • The merging of SPOT and SAR data yielded topographic information in addition to the representation of all cultural features.
  • The SPOT and SAR registered image provided themost useful data for cultural interpretatiton because any questionable feature such as a road could be quickly assessed in relaltion to topography and either accepted or dismissed.
Acknowledgements
The support of Sun Malaysia Petroleum Company is gratefully appreciated. In addition, we would like to acknowledge the initial assistance of D. Werle of AERDE Research for the initial cultural interpretation in some of thecloooosure areas.

References
  • Bercha, F.G.,O. Sawicki, P. Fuenning, and P.Jordan, 1989. Geologic Interpretation of Sabah Sar Data. Petroleum Geology Seminar '89. Geological Society of Malaysia.
  • Drury, S.A., 1986. Image Interpretation in Geology. Allen & Unwin. Boston, Mass., 213 pp. Ketiga, C.Y., 1985. Geological Map of Sabah. Geological Survey of Malaysia.