Assessing the capabilities of Landsat imagery for measuring fuel properties in Sydney Coastal Dry Sclerophyll Forest

Australia is one of the most fire-prone continents in the world (King et al, 2011). Fire can be a great threat to human safety and property (Bradstock & Gill, 2001) and so it is important that effective techniques are developed to be able to predict future fire events. The amount of fuel available to burn strongly determines the likelihood of a fire occurring and the nature of the fire (e.g. severity, intensity) (Bradstock, 2010). Traditionally researchers have attempted to predict future fire events by manually measuring fuel in the field. However such methods at large spatial scales are time-consuming, costly, require man-power and are limited for regions inaccessible to humans. For these reasons remote sensing methods are becoming increasingly popular to measure fuel (Frokling et al, 2009). However, only a few studies have been conducted in Australia to investigate the potential of remote sensing for measuring fuel. Australian studies have only tested a small number of spectral indices, have not tested the effects of the understorey layers on the spectral signal over long time-periods of fuel accumulation and have not taken into consideration dead fuel components in the forest layers. This research project attempted to fill these gaps in the research by assessing the ability of Landsat 5 TM in measuring fuel loads. Seven different indices were compared in their ability to measure fuel in Sydney Coastal Dry Sclerophyll Forest, a forest characterised by a prominent understorey and the presence of both live and dead fuel. Ground truth data extracted from a fuel database for 31 sites was regressed against spectral indice values calculated from the Landsat images. Two approaches were used: Approach 1 did not account for phenological variation; Approach 2 did account for phenological variation. Results found that overall moisture indices (NDIIb5 and NDIIb7) performed the best at measuring fuel cover, as they were not greatly affected by phenological variation or hindered by the presence of dead fuel. Greenness indices (SR, NDVI, GNDVI and SAVI) performed well at measuring fuel but only when phenological effects were accounted for and dead fuel was eliminated from the GT data. When indices performed well they all improved with the addition of the elevated fuel cover layer, suggesting that in this forest type, the understorey significantly contributes to the spectral signal. SATVI was affected by phenological variations, but when this was accounted for it was not as affected by the presence of dead vegetation. The main implication from this study was that, in multiple layered forests that contain dead fuel in the layers, indices that incorporate SWIR wavelengths may prove to be more accurate and successful to use for measuring fuel than the traditional greenness indices.