Using innovative remote sensing techniques to improve the quality and accuracy of koala habitat mapping in eastern australia

Resource demands required by increasing human populations are accelerating; these demands are exacerbating world-wide species extinctions. Globally, 42,100 species are threatened with extinction (https://www.iucnredlist.org/about/background-history). Reducing the likelihood of species extinction requires addressing imbalance between competing demands, which often occur in the same locations. Unfortunately, the koala (Phascolarctos cinereus) provides an ideal case study of this conundrum, and is the subject of this thesis. The koala is an arboreal marsupial species primarily relying on eucalypt species for food and shelter. Koalas are widely distributed in eastern Australia, but distribution is patchy for several reasons. While many eucalypt species may be available in any particular area, koalas rely on a few highly-preferred species found on fertile soils which have been largely cleared for human needs, e.g., farming, infrastructure and housing. Allied to this, many studies show that koalas prefer taller trees, which are favoured by the logging industry. Finally, although hunting for the fur trade ceased in the 1930s, it is likely that many local koala populations have never recovered and are now extinct. The koala is now listed as endangered in the Australian Capital Territory, New South Wales, and Queensland (Department of Climate Change, Energy, the Environment and Water 2022), and, while researchers provide information and recommendations, it is incumbent upon Governments to incorporate these findings into policies aimed at preventing further declines. Early research in the 2000s highlighted the need to identify areas with higher koala habitat use, which could then be given greater protection within habitat management plans. At local government scales (e.g., >50,000 ha) this was, and still is, accomplished in two stages. Preferred tree species are first identified from plot data. This information is then incorporated into existing forest community maps which enable the production of habitat-quality maps suitable for management use. Determining the resource value of particular species, and species proportions within forest maps, has resulted in several different habitat map schemas, potentially producing conflict between different map users and interest groups. The fundamental problem with all habitat maps is widespread species heterogeneity within eucalypt forests, i.e., where local variation is not captured by habitat classification. This prompted my first research questions: what are the limitations of existing habitat maps, mapped at low resolution? I used four different habitat schemas to assess differences in mapped habitat quality. Three maps shared the same internal consistency (i.e., high to low values) but only one captured the degrees of difference required to highlight the differing habitat values between communities. This study also examined habitat quality variation between 44 plots within a 60-ha focal study area. Two map schemas correctly classified the entire community polygon within their respective schemas, but plot food tree percentages varied markedly, and only 52% - 64% of plots fell within the correct habitat class. This highlights the scale at which habitat assessment conflict occurs, and the need to address these shortcomings, which can be only reduced by methods which improve habitat mapping at the required spatial resolution. Problems associated with low spatial-resolution habitat mapping prompted my next research question: are there other methods we could use to highlight differences in habitat quality? Many studies have shown that koalas prefer larger/taller trees, so I next investigated tree height using LiDAR (Light Detection and Ranging). This study also had two components. Across southeast Queensland, I selected 238 “virtual” LiDAR plots within a forest community known to be used by koalas, and extracted the maximum canopy height within each 30 m x 30 m plot. Assessing this data, I concluded that canopy height varied markedly across the region, and concluded that the best way to capture height variation was to use a polygon-by-polygon approach. An expert panel examined several spatial clustering techniques which classified contiguous height classes within each polygon, and which would be suitable for map interpretation, and identified a suitable algorithm. This algorithm also successfully captured a forest ecotone dominated by one particular highly-preferred koala tree species, and so, potentially, has wider application to improve the resolution of both existing habitat and forest-type maps. For my third research question, I examined whether spatial clustering of canopy height might provide some insight adding to our knowledge of koala ecology. I obtained radio-tracking data for 135 koalas in southeast Queensland, and generated home ranges (95% kernel estimate) and core home ranges (50%). For individual forest types within home ranges, I used the previously-identified clustering algorithm, and determined that core home ranges, compared to the remaining home range portion, had 30% more of the highest canopy class. I concluded that areas of higher canopy were indeed an important factor in habitat use by these individual koalas. My final research question was: how important is the canopy height factor in comparison to other known habitat factors? I chose a 25,000-ha (25 km2) study site in the Strathbogie ranges in northeast Victoria, and derived factors known to influence habitat occupancy (forest cover, preferred species cover, etc), and used these as variables in a generalised linear mixed model. This study was limited by the smaller extent of LiDAR data and, likely because of this limitation, this factor was not important in the final model. However, I showed that habitat use was primarily influenced by terrain slope at the plot scale. Previous broad-scale studies had identified elevation as an important factor, but, at my study scale, slope provided a framework incorporating other factors known to be important to koalas, e.g., soil fertility and higher soil moisture have influence on both species composition and forest structure. In summary, my thesis firstly identified limitations of existing low-resolution habitat maps when used at higher-resolution management scales. Following a literature review showing koala preference for higher trees in some areas, I developed methods to further investigate this preference, and how this might be depicted in a map which might assist habitat management. I then demonstrated that areas with higher canopy height are, indeed, preferentially used by koalas. Unfortunately, data limitations prevented the full use of canopy height information in a generalised linear mixed model, but this model showed that terrain slope can be a major factor in habitat use. Both canopy height and slope can be derived from remotely-sensed LiDAR data, and conveniently depicted as a higherresolution layer for use in conjunction with current lower-resolution habitat maps. My research has potential for wider application beyond the field of koala ecology. Firstly, in Chapter 2, I have highlighted problems encountered by koala ecologists, and others, who are required to use low-resolution, general-purpose vegetation and forest-type maps in their work. This is an issue which likely applies to management of other fauna and flora species globally. Secondly, in Chapter 3, I assessed spatially-constrained methods to classify a raster canopy height. This approach has potential application for classification of other discrete raster data, and particularly in the field of satellite derived spectral imagery classification. In Chapter 4, I successfully demonstrated an application of these methods to assess resource use in koala home ranges, this approach could have wider application to other fauna and flora with defined home ranges.