Figure 3. Binary maps of (A) potential fire classes, (B) definite fire areas, and (C) the final map of fire scars. The false colour composite is shown for reference.
Rather than attempting to devise a sophisticated classification methodology that may work well in some cases, the problem of spectral overlap was solved by utilizing the operator's knowledge of the study area and image interpretation. Manually digitising of fire scars has the major disadvantage of being time consuming, but on the other hand, it yields great accuracy in recognition, especially in conjunction with ancillary data and multiple images that are available for fire history mapping. Having a multi-temporal view allows recognition of permanent landscape features. There is a possibility that very small fire events of less than 10 pixels may be missed, but such events would be very rare in this environment. The problem in relying on this methodology is the accurate boundary delineation. To trace the outline accurately increases the time requirement very rapidly. For this methodology, it is important to be accurate in eliminating image elements that may be confused with fires. However, the boundaries do not have to be very accurate where distinct land cover features adjoin, as the spectral classification will resolve this boundary with the best accuracy possible for the data resolution.
The accuracy of the final map was 91% and 94% according to field data and IKONOS data respectively. Even though the overall accuracy rate assessed by IKONOS data was reduced because of some errors occurred in the visual interpretation, this method could be an easy and extensive approach for accuracy assessment. Some of the errors can be attributed to the time difference between the acquisition of Landsat TM image and the IKONOS data as there is a distinct possibility of several fire events occurring in a single week during the middle dry season. The methodology developed is subjective as it depends on operator knowledge, and moderately time intensive depending on the number of fire events present in the data, but it is robust as no radiometric corrections are required and produces highly accurate and reliable results. This is of great importance for fire history mapping, as there is no available ground truth for past image dates.
5. Acknowledgment
This study is a part of Yue Zhang's PhD project supported by Australian Cooperative Research Centre for Sustainable Development of Tropical Savannas and Australian Research Council Key Centre for Tropical Wildlife Management. Parks and Wildlife Commission of Northern Territory, Australia, funded acquisition of images. We are grateful for these supports. Thanks should also go to Mr Joshua Forner, Mr Craig Hempel, and Mr Andrew Edwards, for their field help and suggestions.
6. References
Chuvieco, E., and Congalton, R. G., 1988. Mapping and inventory of forest fires from digital processing of TM data. Geocarto International, 4, 41-53.
Conese C., and Maselli F., 1993. Selection of optimum bands from TM scenes through mutual information analysis. ISPRS Journal of Photogrammetry & Remote Sensing. 48, 2-11.
Jacklyn, P. and Russell-Smith, J., 1998. Proceedings from the North Australia Fire Management Workshop, Tropical Savannas CRC, Darwin, pp. 1-40.
Pereira, J. M. C., Chuvieco, E., Beaudoin, A., and Desbols, N., 1997. Remote sensing of burned areas. In: A review of remote sensing methods for the study of large wildland fires, edited by E. Chuvieco, Alcala de Henares, Spain, pp. 127-183.
Press, A. J., 1988. Comparisons of the extent of fire in different land management systems in the Top End of the Northern Territory. Proceedings of the Ecological Society of Australia. 15, 167-175
Russell-Smith, J., Needham, S., and Brock, J., 1995. The physical environment. In: Kakadu: Natural and Cultural Heritage and Management, edited by A. Press, A. J. Lea, A. Webb and A. J. Graham, Australian Nature Conservation Agency and North Australia Research Unit, Australian National University, Darwin, pp. 94-126
Russell-Smith, J., Ryan, P., and Durieu, R., 1997. A LANDSAT MSS-derived fire history of Kakadu National Park, monsoonal northern Australia, 1980-94: seasonal extent, frequency and patchiness. Journal of Applied Ecology, 34: 748-766.
Williams, M. A. J., 1991. Evolution of the landscape. In; Monsoonal Australia: Landscape, Ecology and Man in the Northern Lowlands, edited by C. D. Haynes, M. G. Ridpath and M. A. J. Williams, Balkema, The Netherlands, pp. 5-17
Wilson, B. A., Brocklehurst, P. S., Clark, M. J., and Dickinson, K. J. M., 1990. Vegetation survey of the Northern Territory, Technical Report No. 49, Conservation Commission of the Northern Territory, Darwin, Australia.
