• Australia;
  • fire interval;
  • fire regime;
  • hazard function;
  • Kakadu;
  • probability models;
  • savannah;
  • seasonality

Abstract LANDSAT Multi-Spectral Scanner imagery was used to determine aspects of the fire regimes of Kakadu National Park (in the wet-dry tropics of Australia) for the period 1980–1995. Three landscape types recognized in this Park were Plateau, Lowlands and Floodplain. Areas burned in early and late dry seasons each year were documented using a Geographical Information System. Regression analyses were used to examine time trends in the areas burned each year and the interrelationships between early and late dry season burning. The proportions of landscapes having different stand ages (years since fire), and the proportions having had different fire intervals, were compared with results expected from the simplest random model (i.e. one in which the probability of ignition at a point [PIP] burning annually was constant). Using overlays of successive stand-age maps, PIP could be calculated as a function of stand age. The Lowlands burned extensively each year; the areas burned by late dry season fires adding to those burned in the early dry season such that around 50–60% of the total area burned annually. Early dry season fires have lower intensities than late dry season fires, on average. Using a theoretically constant PIP and the mean proportion burned per year as the only input, predictions of areas burned as a function of stand age and fire interval were reasonable when compared with the empirical data, but best for the Lowlands landscape. PIP functions for Lowlands and Floodplains had negative slopes, an unexpected result. The nature of these PIP functions may reflect heterogeneity in fire proneness of the various vegetation types within landscapes. The scale of measurement, the scale of variation in vegetation types within a landscape, and the accuracy of the determination of burned areas, are constraints on the accuracy of fire-interval and seasonally determination perceived from an analysis of satellite data.