Development of a synthetic record of fire probability and proportion of late fires from simulated growth of ground stratum and annual rainfall in the Australian tropical savanna zone

In this study, we sought to address the issue of how to derive an extended synthetic record of fire incidence and timing at regional scale that would be representative of a short remotely-sensed calibration record. We used annual rainfall and simulated annual ground stratum growth to develop multiple regression relationships for prediction of annual fire probability and proportion of late (August–November) fires from AVHRR NDVI fire footprint data across Australian tropical savannas. Relationships were examined using spatial averaging in moving windows varying from 3 × 3 to 61 × 61 pixels in size. Model fits as measured by R2 improved as window size increased, but output layers became smoother and less representative of natural heterogeneity. A 25 × 25 pixel window was selected as the best compromise between model fit and smoothing. A 113-year synthetic record of annual fire probability and proportion of late fires was generated using the spatially explicit layers of model coefficients. The statistical properties of the synthetic fire probabilities were compared with those derived from the available fire footprint record, using a simple vegetation classification based on ground stratum type for spatial stratification. The two data sets showed a strong correspondence for both burned area and fire probability; spatial variation in mean and coefficient of variation of fire probability was representative of that observed in the historical record. There was significant temporal variation in the synthetic annual fire probability for different vegetation zones across the tropical savanna region for the full 113-year length of record. This simple approach could readily be applied to other areas of the world provided rainfall data are available and annual ground stratum growth can be simulated with a suitable model or estimated with remote sensing.