Modelling bushfire impact on hydrology: The implications of the fire modelling approach on the climate change impact

Abstract

In dry environments bushfire is an integral part of the dynamics of the environmental system. Thus any computational model of the environment (e.g. the effect of climate change on vegetation, hydrology, landform evolution, etc) needs to include a model of bushfire (e.g. Thonicke, et al 2001). Distributed models need to not only model the temporal distribution of bushfire but also their spatial distribution. The spatial properties of bushfires include the location of the initiation point of the fire and the spread in space from that initiation point. Thus the time history of bushfire at any specific location in space is a function of the rate of bushfires in the region surrounding that point, and the likelihood that any given fire will spread to the point under consideration. Thus the fire history at a point is a function of the both the temporal and spatial properties of bushfire. In the applications contemplated in this paper we are making predictions into the future so we need to be able to make predictions in a statistical, rather than a deterministic, sense. Accordingly we are looking for a minimalist model that involves replication of the statistical properties of bushfire rather than a deterministic model that might be used for the prediction of the behavior of any individual bushfire event. Deterministic models of bushfire propagation exist (e.g. FARSITE; Finney et al, 1998) but their use for statistical prediction is limited because they require a variety of environmental inputs (e.g. humidity, wind speed, fuel load) that themselves need to be independently predicted (Pastor et al, 2003). This paper explores a simpler, purely statistical, approach where bushfire is modeled directly by monte-carlo simulation. The model is calibrated to published data for bushfire and then some preliminary assessment of its usefulness for looking at the impact of bushfire on hydrology is examined. The idea of the model presented here is that fire is modeled as a random process on a grid. Fire is randomly initiated at a time and a location, and some area around that location is then burnt. The burn history of the landscape is then the history of all of the individual fires that have been randomly generated in space and time. A spatially distributed stochastic simulation bushfire model is presented and it is calibrated to remotely sensed fire data. It is shown that the assumption of how the frequency of a fire and its subsequent extent are related has a significant impact on the net impact of bushfire on hydrology. In particular, it highlights how important it is to distinguish between how often, on average, a bushfire burns a location, and how often, on average, a location has an ignition event. These two properties are different and that climate change predictions will need to provide information on both of these properties. The fire model presented is very simple and this is its first test against field data. There are a number of areas where it might be improved to gain better fit to field data. The most obvious is a relationship between the last time a node was burnt and where fires are initiated and/or propagated. This will likely be linked to rate of recovery of biomass and thus there might be some form of climate dependency. There is also a need for more thorough testing of the spatial statistics of the model using data from other sites (e.g. Haydon et al 2000) and more comprehensive longitudinal datasets using NDVI and Landsat datasets.