Accuracy assessment of spectral indices & determining factors influencing burn severity of the 2021 Wooroloo bushfire

Author Identifier


Date of Award


Document Type



Edith Cowan University

Degree Name

Master of Science by Research


School of Science

First Supervisor

Eddie van Etten

Second Supervisor

David Blake

Third Supervisor

Pierre Horwitz


Fire regimes are changing throughout the world with climate change resulting in increasing global burned area and fire severity. Much of this change is occurring in recent years and is projected to increase further with rising temperatures. Increases in consecutive days that meteorological, climatic and biophysical variables combine to influence fire is projecting an increase in fire size and intensity. This prompts investigation into the severity of recent fires to gain a better understanding of the factors that influence fire severity. This work tested a suite of spectral indices which indicate fire severity against field-collected data using Random Forest (RF) to determine a spectral index that is the most accurate index measuring the severity of the Wooroloo Bushfire of 2021. Using the most accurate index, a geospatial analysis was undertaken to identify key variables that control the variability in fire severity of this fire. Random Forest classification was used to determine spectral index classification accuracy. Random Forest regression and General Additive Model (GAM) were utilised to determine variable importance in modelling fire severity and to create parsimonious models. The Relativised Burn Ratio (RBR) with a 64-day pre- and post-fire image collection period and a minimum pixel value was determined to be the most accurate spectral index. Elevation and the Soil Dryness Index (SDI) were the two most influential factors in explaining the variability of the fire. Whilst a good variable importance was achieved, the parsimonious Random Forest and GAM models performed poorly, managing to explain 14% of variance (RF) and 17% of deviance (GAM). This work outlined the need for localised spectral index testing to determine the most accurate fire severity index and supports reports that the Wooroloo Bushfire was severe, unpredictable, and difficult to control. Further testing into fire behaviour variable interaction and the influence of fire on fire severity should be conducted to determine important variables in modelling fire severity.



Access Note

Access to this thesis is embargoed until 7th May 2025.

Access to this thesis is restricted. Please see the Access Note below for access details.