摘要

We characterised fire frequency in western Arnhem Land, northern Australia (similar to 24 000 km(2)), during the period 1990-2008, using available satellite burnt area maps. We estimated fire mortality and fire survival distributions, and hazard functions by vegetation type. We tested the performance of three probability models to study fire interval distributions: continuous and discrete Weibull, and discrete lognormal. Over the 19 year study period the mean annual area burnt was 36%. Median fire intervals ranged from 1 to 4 years. The discrete lognormal model best fitted the data, yielding non-monotonic hazard functions that peak at 2 to 3 years, making it more appropriate for fire frequency analysis in fire-prone tropical savannas than the more popular Weibull model. Open forest showed the highest flammability dependence on fuel age, and closed forest the lowest. The probability of burning as a function of time since last fire reaches an early peak and subsequently declines, due to fuel dynamics in these flammable savanna systems. Age-specific fire incidence is much higher for older vegetation patches than was suggested by earlier analysis of fire interval distributions. Fitting an appropriate model is important to characterize the observed fire frequency patterns, and make inferences for unobserved, longer fire intervals.

  • 出版日期2013