Spatio-temporal savanna vegetation dynamics in northern Australia

  • Caroline Elisabeth Lehmann

    Student thesis: Doctor of Philosophy (PhD) - CDU


    Savanna is characterised by a C4 grass understorey and a discontinuous overstorey of trees, and occurs across the globe in regions of high mean annual temperature, and intense seasonality in rainfall and productivity. An understanding of the patterns of tree biomass in mesic savannas has remained an ecological conundrum, largely due to the complexity and spatio-temporal heterogeneity of these environments. Yet, understanding the dynamics of these systems is of substantial applied and theoretical significance. I therefore aimed to answer the question: What determines the spatial pattern of tree biomass in Australian tropical mesic savannas? To answer this question, I focussed on a savanna system dominated by Eucalyptus tetrodonta, which occupies 450, 000 km2 of north Australia and accounts for 3% of global savanna lands. I combined use of remote sensing, landscape ecology, data syntheses and statistical modelling. Further, I aimed to investigate this question at four spatial scales: sub-continental (1000’s km); regional (100’s km); landscape (10’s km) and the level of the individual. Investigating this problem from different perspectives enabled to me to attempt to quantify the relative roles of rainfall, fire, mega-herbivory and resource-limitation in determining patterns of savanna tree biomass. 

    At the sub-continental scale, and from 700 -1700 mm mean annual rainfall (MAR), I found tree biomass increased with MAR; MAR limited potential savanna tree biomass yet, this potential was not often realised. Further, I found evidence consistent with the role of disturbance in generating spatial variability in savanna biomass. As an example, areas subjected to experimental fire treatments showed tree biomass could change by +/- 50% over 25 years under a variety of fire regimes. 

    At the regional scale, across a geographically restricted rainfall gradient (1200 - 1600 mm MAR) in Kakadu National Park, and over a 40 year period of observation, the greater variability in fire activity and the inherently higher tree cover in more mesic areas resulted in a greater dynamism of tree cover compared to the drier end of the rainfall gradient. Fire was an important process determining deviations in tree cover below a MAR determined potential. 

    At the landscape scale (10’s km), and over a 40 year period of observation, change in tree cover tracked inter-annual variation in rainfall. Controlling for rainfall, I showed that frequent fire led to declines in tree cover and biomass and large-mammal herbivory promoted an increase in tree cover, likely via the altered spatial and temporal patterning of fire due to changes in grass fuel loads. 

    Fire is a pervasive feature of these ecosystems, modifying vegetation structure and biodiversity. The most important way in which fire limits savanna tree cover is thought to be via its effects on sapling trees, which need several years without fire to grow large enough to enter the canopy. I found evidence for a recruitment bottleneck in four savanna tree species including E. tetrodonta. Further, I found a positive interaction between treetree competition and fire on the recruitment of saplings. This was evidence for the importance of fire in driving cycles of canopy growth and renewal. 

    Across a range of spatial and temporal scales tree cover and biomass in E. tetrodonta dominated savannas is highly spatially and temporally variable, primarily in response to rainfall, fire and their interaction. The historical and spatial perspective provided by this thesis shows the importance of fire in maintaining this savanna system. In the light of this knowledge I make a number of recommendations to help achieve fire management regimes that maintain heterogeneity and population viability of these systems.
    Date of AwardJul 2007
    Original languageEnglish
    SupervisorDavid Bowman (Supervisor) & Lynda Dorothy Prior (Supervisor)

    Cite this