Key Words: interactive approach, DEM, bush fire mapping, savannas, Australia
Abstract Tropical savannas occupy one third of Australia's landmass, and are composed of dense grasslands and scattered trees. Both deliberately lit and natural wildfire sweep across the vast monsoonal savannas of northern Australia each year and the consequence of these modern fire regimes can be catastrophic for fire-sensitive plants and animals. This paper presents an interactive methodology which improves the mapping accuracy for mapping fire history from Landsat Thematic Mapper data of an area with typical savanna landscape characteristics of northern Australia. A satellite image captured in the middle dry season of the year 2000 was used. With the visible red, NIR and MIR bands, an unsupervised digital image classification was carried out to delineate the burnt patches. These patches were labelled by using field knowledge as well as by on-screen assessment of the raw data and signature files for previously confirmed fire scars. Spectral overlap between fire scars, water bodies, shadows and miscellaneous geological features was observed and was eliminated by using of a binary spatial mask and the Digital Elevation Model (DEM). The validity of the fire mapping was assessed with the help of field data as well as using a high spatial resolution IKONOS image acquired at the time of the TM data recording.
1. Introduction
Wildfire has been considered as one of the major landscape characteristics in the Australian tropical savannas. Earlier studies have estimated that over 50% of the savannas in some regions of northern Australia are burnt during the 7-month (May-November) dry season each year (Press 1988). Even in Kakadu National Park, a United Nation's World Heritage property, an average of 55% of lowland savanna habitats and 28% of habitats occupying both sandstone plateau and riverine landforms has been burnt each year (Russell-Smith et al. 1997). Depending on a variety of purposes of management and conservation, dry season fire is ignited to reduce the level of flammable fuel loads, to encourage 'green pick' for cattle and kangaroos, and to create patchy mosaics of burnt and unburnt country to help developing habitat diversity (Jacklyn and Russell-Smith 1998). However, the bulk of late dry season uncontrolled wildfire sweeping across the Northern Territory of Australia, are catastrophic for fire-sensitive flora and fauna to be the consequence of such contemporary fire regimes. To add the understanding of the savanna landscape structure and dynamics, fire scars mapping has been carried out using satellite images (Russell-Smith et al. 1997).
Acquired in various geometric, radiometric and spectral resolutions, satellite remotely sensed data, multi-temporal or single post-fire, have been processed successfully to map burnt scars, species affected and post-fire recovery (Chuvieco and Congalton 1988, Pereira et al. 1997). Various fire-mapping methods have been developed and applied, but most of these focus on the areas where vegetation distribution and topographic features are comparatively uniform. However, when mapping bush fire in tropical savannas, such as in northern Australia, spectral overlaps between fire scars and terrain shadows, water bodies, unburnt canopies, and vegetation regrowth in burnt area create substantial difficulties in separating and discriminating fire classes. Russell-Smith et al. (1997) utilized the hard copies of Landsat MSS data to define the fire scars. In this study, an interactive fire mapping method that improves the accuracy of bush fire mapping and helps overcome these problems of separation and discrimination was developed for this special landscape type in the Australian tropical savanna.
2. Study area
The study area is located on the tropical savannas with typical landscape features in the Top End of the Northern Territory, Australia, which is bounded by longitudes 131o48'E, 132o27'E and by latitudes 13o41'S, 14o17' (Figure 1).
Figure 1. Location of the Study Area (Yinberrie Hills, Northern Territory, Australia)
