CSIRO PUBLISHING International Journal of Wildland Fire 2008, 17, 768–781 www.publish.csiro.au/journals/ijwf Big fires and their ecological impacts in Australian savannas: size and frequency matters Cameron P. Yates A,B,C , Andrew C. Edwards A,B and Jeremy Russell-Smith A,B A Bushfires NT, PO Box 37346, Winnellie, NT 0821, Australia. B Tropical Savannas Cooperative Research Centre, Charles Darwin University, Building 31, Level 3, Darwin, NT 0909, Australia. C Corresponding author. Email: cameron.yates@nt.gov.au Abstract. Savannas are the most fire-prone of the earth’s major biomes.The availability of various broad-scale satellite- derived fire mapping and regional datasets provides a framework with which to examine the seasonality, extent and implications of large fires with particular reference to biodiversity values in the tropical savannas of northern Australia. We document the significance of savanna fires in the fire-prone ‘Top End’region of the Northern Territory, Australia, using 9 years (1997–2005) of National Oceanic andAtmosphericAdministration (NOAA)-Advanced Very High Resolution Radiometer (AVHRR)-, Landsat Thematic Mapper (TM)- and Enhanced Thematic Mapper (ETM+)-derived fire mapping. Fire (patch) sizes from both AVHRR- and Landsat-scale mapping increased through the calendar year associated with progressive curing of grass and litter fuels. Fire frequency data at both satellite sensor scales indicate that regional fire regimes in higher rainfall regions are dominated by large (>1000 km 2 ) fires occurring typically at short (∼2–3 years) fire return intervals. In discussion, we consider the ecological implications of these patch size distributions on regional fire-sensitive biota. Collectively, assembled data illustrate that many northernAustralian savanna flora, fauna and habitats embedded within the savanna matrix are vulnerable to extensive and frequent fires, especially longer-lived obligate seeder plant taxa and relatively immobile vertebrate fauna with small home ranges. Additional keywords: Arnhem Land, AVHRR, fire-affected area, Landsat, patch size, Top End. Introduction The significance of large fires can be understood in various contexts, including their absolute size and their size relative to various ecological (e.g. population distributions of rare taxa), spatial design (e.g. fragmented reserves) and social organisation (e.g. human settlements) landscape-scale features. The relative significance of these conditions obviously varies in different landscape settings. For example, although it is recognised that biomass burning in savanna systems globally contributes the single largest source of pyrogenic emissions (Andreae 1997), major differences in savanna fire activity and patterning are observed between continental regions with respect to differences in seasonality, landcover and fuel types, land use, and population density and distribution (Dwyer et al. 2000; Csiszar et al. 2005; Roy et al. 2005). Features distinguishing the fire-prone woodland savannas and grasslands of northern Australia include relatively very low human population densities and associated infrastructure devel- opment, a reliably long dry season, and mostly subdued terrain. Under such conditions, individual fires (although often devel- oped from multiple ignition sources) may burn over tens of thousands of square kilometres (Luke and McArthur 1978). Based on compilation of national fire extent statistics derived from Advanced Very High Resolution Radiometer (AVHRR) imagery, an annual mean of ∼19% of Australia’s 1.9 mil- lion km 2 tropical savannas region has been fire-affected over the period 1997–2005, making this the most fire-prone Australian landscape (AGO 2007; Russell-Smith and Yates 2007). The implications of contemporary fire regimes on north Australian savanna biodiversity, soils, greenhouse emissions, pastoral pro- duction, and broader social implications are increasingly well documented and understood (e.g. Townsend and Douglas 2000; Williams et al. 2002; Dyer et al. 2003; Russell-Smith et al. 2003a, 2006; Ellis et al. 2004; AGO 2007; Preece 2007). In recent decades, a considerable number of studies have taken advantage of satellite-based optical sensor sources (e.g. Landsat, Moderate-Resolution Imaging Spectroadiometer (MODIS), AVHRR) to describe major features of contemporary fire regimes in north Australian savannas, especially in regard to fire seasonality, extent, and frequency in different regional and habitat settings (e.g. Press 1988; Beringer et al. 1995; Craig et al. 2002; Fisher et al. 2003; Russell-Smith et al. 2003b; Felderhof and Gillieson 2006). Studies of fire (or, more usefully, ‘patch’)- size distributions (PSD) in north Australian savanna systems have included: comparison of PSDs between major habitat types in Kakadu National Park, derived from 15 years of Landsat-scale fire-affected area (FAA) mapping (Russell-Smith et al. 1997); comparison between PSDs derived from AVHRR- and Landsat- scale FAA mapping (Yates and Russell-Smith 2002); and, using 10 years of Landsat-scale FAA mapping for a 9000-km 2 prop- erty, exploration of the frequency distributions of three types of fire-created patch (unburnt ‘islands’; patches created with © IAWF 2008 10.1071/WF07150 1049-8001/08/060768