Land management agencies are increasingly using multispectral satellite imagery to monitor post-fire ground conditions to guide effective conservation and asset protection management strategies. Estimates derived from such data can span over multiple decades and can be analysed to identify spatiotemporal burnt area patterns. Interestingly, fire severity is rarely derived from multi-decadal remote sensing datasets. Accordingly, in this study we explored the utility of imagery captured with Landsat sensors (5 TM, 7 ETM+ and 8 OLI) to characterise fire severity of burnt areas over a 23 year period in a national park with a woodland and heath ecosystem. The multi-temporal differenced normalised burn ratio (dNBR) was used to estimate fire severity for individual burnt areas, which were then aggregated to identify spatiotemporal patterns. Accuracy assessment was achieved using ground truth data collected with the Geometrically Structured Composite Burn Index (GeoCBI), which incorporates the fraction of cover (FCOV) of vegetation over the total plot. Our results indicated that Landsat imagery was ideal for significantly estimating fire severity (accuracy = 72 %, kappa = 0.63, P = < 0.001) in sclerophyll woodland and heath ecosystems. Further, we had interesting insight into the patterns of increased fire severity within specific vegetation types and across the study site.
QCON14 Conference: Repositioning for a Sustainable Future, Cairns, Australia 7-10 October 2014