How does remotely sensed degree of curing and fuel load vary in grasslands and effect modelled fire spread?

Abstract

Wildfire can become a catastrophic natural hazard. Grasslands are the most common fuel type in Australia with nearly 75% coverage. There are multiple factors, particularly weather and fuel conditions, which dictate the severity of grassland fire. While grassland fires are very responsive to weather conditions, availability and variability in fuels are also important drivers. Here, grassland fuel availability and variability are studied by examining the degree of curing (DOC) and fuel load. DOC is the percentage proportion of dead material in the fuel bed (100% DOC indicates a fully cured grassland). Satellite data, including recently developed microwave based vegetation optical depth (VOD) and normalised difference vegetation index (NDVI) along with field observed data, are used to calibrate and evaluate satellite estimated DOC and fuel load models over Australia. Gridded, 8 day estimated DOC is derived via a regression analysis of VOD and NDVI, while gridded, annual estimated fuel load is constructed from aboveground biomass carbon derived from VOD. Both DOC and fuel load have persistent spatial variability across Australia. DOC has a continental mean of 85.7% and spatial standard deviation of 20.4%, while grassland fuel load has a mean of 5.9 t ha-1 and standard deviation of 2.5 t ha-1 across all years. DOC shows strong temporal variation in southeast and southwest Australia, while fuel load shows strong temporal variation towards the east coast. Their continental mean temporal standard variations are 11.9% and 0.5 t ha-1, respectively. Both satellite based DOC and fuel load data are then used in fire spread models (Phoenix RapidFire and Spark) to assess the changes in modelled grassland rate of spread due to spatial variability of DOC and fuel load. Rate of spread experiments are conducted using: 1) idealised experiments with artificial DOC, fuel load, and weather; and 2) realistic experiments using satellite based DOC and fuel load along with weather station data based on past fire events. Results suggest that predicted rates of spread are more sensitive to changes in DOC than fuel load. Observed spatial variations in fuel characteristics can significantly alter the modelled rate of fire spread across a landscape. Phoenix and Spark models predict divergent rates of spread in high fuel load environments (over 15 t ha-1). With an estimated DOC as an appealing alternative to the currently available DOC product, a newly developed annual estimated fuel load over a continental scale, and direct comparison of fire spread simulation between a popular, classic and a recently developed, high potential fire spread models, this thesis pushes the boundary of grasslands fire fuel monitoring and sheds some light on the prediction results of current fire spread models. Grassland fire spread prediction is challenging and requires careful consideration of fuel related parameters and variabilities across space and time, such as DOC and fuel load, as well as selection of fire spread models.