School of Agriculture, Food and Ecosystem Sciences - Theses

Permanent URI for this collection

Search Results

Now showing 1 - 1 of 1
  • Item
    Thumbnail Image
    Quantifying fire-severity patterns using optical remote sensing data in temperate eucalypt forests of south-eastern Australia
    Tran, Bang Nguyen ( 2020)
    Wildfires have significant biophysical and ecological impacts on ecosystems worldwide from local to regional and national scales. The magnitude of such impacts is related to wildfire severity. Recent increases in wildfire occurrence have been associated with climate change, however whether there has also been a change in fire severity remains underexamined in many biomes. Better understanding of fire-severity patterns is required for effective wildfire management, particularly in the fire-prone landscapes of temperate south-eastern Australia, which support a diversity of forests varying in species composition, structure, and post-fire regeneration strategies. Thus, the overarching aims of my Thesis were to accurately quantify wildfire severity at landscape scales and to examine spatial and temporal variation in wildfire severity across a range of forest types in Victoria, south-eastern Australia. To meet the overarching aims, my Thesis involves: (1) identification of optimal optical spectral indices for mapping fire severity across the dominant and most fire-prone forest types in Victoria; (2) a comparison of the accuracy of two different fire-severity mapping approaches, namely single spectral indexing thresholding and machine learning; (3) using the acquired knowledge, the development of fire-severity maps for large (>1000 ha) wildfires occurring in Victoria between 1987 and 2017, and a retrospective analysis of changes in spatial patterns of high-severity fires over that period; and (4) an analysis of the relative importance of four groups of environmental variables (namely fire weather, fuel, topography and climate) as predictors of high-severity fire extent and landscape configuration. My evaluation of remote sensing based spectral indices indicated that the best-performing indices of fire severity varied with forest type and forest functional group, but that there is scope to group forests by structure and fire-regeneration strategy to simplify fire-severity classification in heterogeneous forest landscapes. Results from my comparative analysis confirmed that machine learning outperformed the spectral index thresholding approach for mapping fire severity in most cases, increasing overall accuracy by 11% on a forest-group basis, and 16% on an individual wildfire basis. My results also confirmed that the accuracy achieved with a reduced set of predictor variables that included the previously identified optimal indices of fire severity was not improved by adding more variables. Greater overall accuracies (by 12% on average) were achieved when in-situ data (rather than data from other fires) were used to train the machine-learning algorithm. As such, my study demonstrates the utility of machine-learning algorithms for streamlining a robust fire-severity mapping approach across heterogeneous forested landscapes. Analysis of spatial patterns highlighted that high-severity wildfires in temperate Australian forests have increased in extent and aggregation in recent decades. The total and proportional high-severity burned area increased through time from 1987 to 2017. While the number of high-severity patches per year remained unchanged in that period, the variability in high-severity patch size increased, and high-severity patches became more aggregated and more irregular in shape. Finally, key findings from my models on the relative importance of environmental drivers (climate, fire weather, fuel, and topography) were that fuel type and fire weather were the most important predictors of the extent and configuration of high-severity fires in Australian temperate forests. My Thesis presents one of the most comprehensive analyses of fire-severity patterns from remote sensing data in Australia. My research results support the reliable estimation of wildfire severity from optical images using machine-learning algorithms once optimal spectral indices are identified and when in-situ training data are available for individual fires. Importantly, the quantified shifts in fire regimes across Victoria’s forested landscapes may have critical consequences for ecosystem dynamics, as fire-adapted temperate forests are more likely to be burned at high severities relative to historical ranges, a trend that seems set to continue under projections of a hotter, drier climate in south-eastern Australia. It is therefore critical that forest scientists and land managers continue to acknowledge and quantify changing wildfire-severity patterns so that they are better informed to address the ecological consequences.