Fire-created patchiness in Australian savannas

Models of ecosystem disturbances have been based on either points or patches. In this study, three types of fire-created patch were distinguished: the unburned 'islands' within a fire area; the patches with individual times-since-fire, those of the first year being the individual fire areas; and, patches with individual intervals between fires. Frequency distributions of patch areas were simulated using two randomly filled square grids (with a probability of 0.5, corresponding to 50% burnt each year) with two levels of aggregation - 'none' and 'clumped'. Fires were represented by clusters of filled cells on each single grid; years were represented by a set of similar, independent, grids. Results from the simulations were compared to those from Bradshaw Station in the savanna region of the northern part of the Northern Territory using a decade of Landsat MSS and Landsat TM imagery. Maps of times since fire (years) and intervals between fires (years) were constructed. Proportions of the maps with different times since fire and intervals between fires followed negative exponential curves. All frequency distributions of patch area, irrespective of patch type, were found to be log-log linear when data were logged and 'binned' (i.e. placed in categories). In both the simulations and at Bradshaw Station, the numbers of single cell patches first rose then declined as times since fire increased while the largest fire patches rapidly decreased in size. Between-fire interval patches decreased in size with increasing intervals but small-patch number did not increase as it did for times since fire. Errors in accuracy of Landsat imagery could heighten the apparent conformity between interpreted imagery and the simulations.