wavelet analysis of climate variability and savanna land-atmosphere interactions...
TRANSCRIPT
Declaration
i
Wavelet Analysis of Climate Variability and
Savanna Land-Atmosphere Interactions during the
20th
Century
Emily Goodale
Bachelor of Science (Honours)
Thesis submitted by Emily Goodale, BSc. already held in partial fulfilment of the
requirements for the Degree of Bachelor of Science with Honours in the School of
Environment, Faculty of Engineering, Health, Science and the Environment, Charles Darwin
University.
Submitted November, 2014
Declaration
ii
Declaration
I declare that this thesis is my own work and has not been submitted in any form for another
degree or diploma at any university or other institute of tertiary education. Information
derived from the published and unpublished work of others has been acknowledged in the
text and a list of references given.
Emily S Goodale, November 2014
Dedication
iii
Dedication
To my supervisors, Lindsay and Jason, whom without which my year would not have been a
success.
Acknowledgements
iv
Acknowledgements
First and foremost, I wish to thank my supervisors, Jason and Lindsay, you have gone above
and beyond to ensure this project was a success, for which I am endlessly grateful. This thesis
would not have been possible without your guidance, support and encouragement along the
way.
To my previous educators, especially Dr Cäcilia Ewenz of Flinders University, Adelaide,
thank you for inviting me into the wonderful world of research. Your wealth of knowledge
and passion for meteorology is inspired; I hope I can be half as knowledgeable as you
someday. Vielen dank für alles.
I am indebted to Jason, Rhys Whitley (Macquarie University) and Vanessa Harverd (CSIRO)
for their incredible work in gathering modelled historical data for me, yet another example of
the willingness to help others, which I am forever observing within this small network of
Australian and international scientists. Your efforts are not forgotten.
The wavelet analysis presented in this thesis could not have been achieved without the
endless help of Liang He. Liang, your wealth of knowledge in this area is exceptional,
especially as an early researcher and I’m eternally grateful for all your help and advice.谢谢.
Thank you to the Hutley Lab, your smiles and words of encouragement along the way are
undoubtedly the reason I have finished this thesis with my sanity (mostly) intact. I have never
met an academic group more willing to offer their help and advice, I hope that someday I will
be able to help out a lowly honours student beginning their research career in the same way
you’ve facilitated me this year.
Finally, I would like to thank Sinclair Knight Merz for their generous contribution and
scholarship granted to me this year; your award has demonstrated my capabilities as a
research scientist and I hope that you find my project worthy of your support.
Table of Contents
v
Table of Contents
Declaration ............................................................................................................................................... i
Dedication .............................................................................................................................................. iii
Acknowledgements ................................................................................................................................ iv
Table of Contents .................................................................................................................................... v
List of Figures ...................................................................................................................................... viii
List of Tables ......................................................................................................................................... xi
List of Acronyms/ Abbreviations .......................................................................................................... xii
Abstract ................................................................................................................................................ xiv
Chapter 1: Introduction ........................................................................................................................... 1
1.1 Land-atmosphere exchanges ......................................................................................................... 2
1.2 Climate change in Australia .......................................................................................................... 4
Temperature and rainfall ................................................................................................................. 5
1.3 Climate change in savanna-dominated north Australia ................................................................ 5
1.4 Mathematical techniques for time-series analysis......................................................................... 7
Wavelet analysis ............................................................................................................................. 7
1.5 Aims and objectives ...................................................................................................................... 9
Chapter 2: Literature Review ................................................................................................................ 11
2.1 Drivers of tropical climatology ................................................................................................... 12
The North Australian Monsoon .................................................................................................... 12
El Nino Southern Oscillation ........................................................................................................ 13
Tasman Sea Index ......................................................................................................................... 15
Indonesia Index ............................................................................................................................. 15
2.2 Analysing climatic and biological cycles .................................................................................... 16
Fourier vs. wavelet analysis .......................................................................................................... 16
The wavelet approach ................................................................................................................... 17
Applications .................................................................................................................................. 18
Table of Contents
vi
2.3 Savanna – the dominant north Australian ecosystem ................................................................. 19
Savanna physiology ...................................................................................................................... 21
Savanna phenology ....................................................................................................................... 22
2.4 Land surface models ................................................................................................................... 23
2.5 Summary ..................................................................................................................................... 26
Chapter 3: Approach and Methodology ................................................................................................ 27
Variables of interest ...................................................................................................................... 28
Tools of analysis ........................................................................................................................... 30
3.1 Site description ............................................................................................................................ 30
3.2 Data processing methods ............................................................................................................ 31
Time series trends ......................................................................................................................... 32
Power spectrum densities and Wavelet transform coherence ....................................................... 32
Software package .......................................................................................................................... 34
3.3 Data Sources and modelling approach ........................................................................................ 34
Tower observations ....................................................................................................................... 34
AWAP ........................................................................................................................................... 34
Climate Indices data ...................................................................................................................... 35
Modelling ...................................................................................................................................... 36
Chapter 4: Results ................................................................................................................................. 38
4.1 Time Series Analysis .................................................................................................................. 38
AWAP meteorological inputs ....................................................................................................... 39
Biological responses ..................................................................................................................... 40
4.2 Power Spectrum density graphs .................................................................................................. 44
4.3 Wavelet Coherence ..................................................................................................................... 51
Gross primary productivity ........................................................................................................... 52
Net ecosystem exchange ............................................................................................................... 52
Ecosystem respiration ................................................................................................................... 53
Latent heat flux ............................................................................................................................. 54
Southern Oscillation Index............................................................................................................ 55
Table of Contents
vii
Tasman Sea Index ......................................................................................................................... 57
Indonesian Index ........................................................................................................................... 58
Chapter 5: Discussion ........................................................................................................................... 64
5.1 Trends and temporal patterns ...................................................................................................... 64
Temperature .................................................................................................................................. 64
Rainfall and latent heat ................................................................................................................. 65
VPD............................................................................................................................................... 67
Carbon fluxes ................................................................................................................................ 67
Trends in ecosystem physiology ................................................................................................... 69
Evidence for ecosystem coupling to climate ................................................................................. 70
Climate indices .............................................................................................................................. 71
Limitations to the study................................................................................................................. 76
Chapter 6: Conclusions ......................................................................................................................... 78
References ............................................................................................................................................. 80
Appendices ............................................................................................................................................ 98
Appendix A: Climate Indices with high correlation to annual total rainfall over Australian continent
.......................................................................................................................................................... 98
Appendix B: Howard Springs Flux Tower Equipment ..................................................................... 99
Appendix C: Slope changes of annual mean carbon fluxes early vs. late century .......................... 100
Appendix D: Time series (anomaly plots) of climate indices ......................................................... 101
Appendix E: Rainfall coherence spectra and ecosystem fluxes ...................................................... 102
List of Figures
viii
List of Figures
Figure 1.1: Representation of the main processes and variables in the coupled land surface-ABL dynamic
system. The carbon and water cycles are represented as flux components within the ecosystem. Taken from
Vila-Guerau de Arellano et al. (2012)……………………………………………………………………………..3
Figure 1.2: Wavelet transform coherence of red and blue signal, observing the different properties of phase
shift, frequency and amplitude between the two signals. Taken from Xu Cui (2014)-
http://www.alivelearn.net/category/matlab/page/3/ ………………………………………………………………8
Figure 2.1 Conceptual diagram linking ecosystem and climate from ecosystem to global scale. (Source:
Beringer et al. 2011, adapted from Chapin et al. 2005 and Bonan 2008). ………………………………………11
Figure 2.2: The geographical distribution of the Australian savanna biome as defined by Fox et al. (2001).
Located north of the Tropic of Capricorn, it also includes the NATT and observation sites (Source: Hutley et al.
2011) ……………………………………………………………………………………………………………..20
Figure 2.3: Schematic of increasing levels of detail being added into surface modelling approaches. (Source:
Pitman 2003) …………………………………………………………………………………………………….25
Figure 3.1: General approach flow chart. The neural network is central to this modelling exercise, bringing
together two sources of information to generate a flux response from 1900-2013 on which the study is centred.
……………………………………………………………………………………………………………………28
Figure 3.2: Diurnal latent heat flux measured at 30 minute intervals and averaged over the 2001-2013 period for
observations (blue) and ANN (green). X axis is hours from midnight and y is Fe flux in W m-2
. ……………...37
Figure 4.1: Time series of three environmental drivers of ecosystem fluxes at the study site (a) temperature
(Ta), (b) rainfall and (c) vapour pressure deficit (VPD). Temperature and rainfall are AWAP inputs
(observations from BoM’s Darwin station) and VPD is a composite of temperature and humidity, demonstrating
the evaporative power of the atmosphere. ……………………………………………………………………….39
Figure 4.2: Linear trend of three carbon fluxes of interest. (a) Net ecosystem exchange (Fc), (b) gross primary
productivity (GPP) and (c) ecosystem respiration (Fre). Plots are ANN outputs over the 100 year period of
interest 1913-2013. For Fc, GPP and Fre -ve values indicate the atmospheric loss of carbon to the ecosystem, i.e.
a sink. …………………………………………………………………………………………………………….41
Figure 4.3: Linear trend of latent heat flux (Fe). Plots are ANN outputs over the 100 year period of interest
1913-2013. ……………………………………………………………………………………………………….42
Figure 4.4: Time series of (a) water use efficiency (WUE) and (b) light use efficiency (LUE) for the tower
observation period (2001-2013). GPP is divided Fe *0.49 to generate a time series for WUE and using the
conversion factor of 0.49 (Kanniah et al. 2010) in LUE gives GPP divided by short wave radiation *0.49, WUE
and LUE are shown at a monthly scale, and are reflective of ecological coupling. ……………………………..43
List of Figures
ix
Figure 4.5: Wavelet power spectrum of (a) temperature, (c) rainfall and (e) vapour pressure deficits and
corresponding wavelet spectrum at Howard Springs (b, d and e respectively). The thick black contour in the
wavelet power spectrums and blue dash line in the wavelet spectrums indicate the 5% significance level against
red noise. Wavelet spectral analysis was fed daily data over the 113 year period. ……………………………...45
Figure 4.6: Wavelet power spectrum of (a) Fc, (c) GPP, (e) Fre and (g) Fe and corresponding wavelet spectrum
(b, d, f and h respectively) at Howard Springs. The thick black contour in the wavelet power spectrums and blue
dash line in the wavelet spectrums indicate the 5% significance level against red noise. Wavelet spectral analysis
was fed daily data over the 113 year period. …………………………………………………………………….47
Figure 4.7: Wavelet power spectrum of (a) Southern Oscillation Index (SOI), (c) Tasman Sea Index (TSI) and
(e) Indonesian Index (II) and corresponding wavelet spectrum (b, d and f respectively) at Howard Springs. The
think black contour in the wavelet power spectrums and blue dash line in the wavelet spectrums indicate the 5%
significance level against red noise. The very low resolution of these images is most likely due to the reduced
number of values (Indies are measured monthly over the 110 year period). ……………………………………50
Figure 4.8: Wavelet coherence of GPP and temperature GPP-temperature is over the whole study period (1900-
2013) using monthly data. The solid black line is the 5% significance level against red noise calculated from the
Monte Carlo simulation. Arrows are the relative phase angle (arrows pointing right are in-phase, arrows pointing
left are anti-phase). …………………………………………………………………………………………….52
Figure 4.9: Wavelet coherence of net ecosystem exchange and VPD. The solid black line is the 5% significance
level against red noise calculated from the Monte Carlo simulation. Arrows are the relative phase angle (arrows
pointing right are in-phase, arrows pointing left are anti-phase). ……………………………………………...53
Figure 4.10: Wavelet coherence of (a) respiration and temperature and (b) respiration and VPD. The solid black
line is the 5% significance level against red noise calculated from the Monte Carlo simulation. Arrows are the
relative phase angle (arrows pointing right are in-phase, arrows pointing left are anti-phase). ……………….53
Figure 4.11: Wavelet coherence of (a) latent heat and temperature and (b) latent heat and VPD. The solid black
line is the 5% significance level against red noise calculated from the Monte Carlo simulation. Arrows are the
relative phase angle (arrows pointing right are in-phase, arrows pointing left are anti-phase). ……………….54
Figure 4.12: Wavelet Coherence of SOI against the biological responses (a) ecosystem respiration, (b) gross
primary productivity, (c) latent heat flux and (d) net ecosystem exchange. The solid black line is the 5%
significance level against red noise calculated from the Monte Carlo simulation. Arrows are the relative phase
angle (arrows pointing right are in-phase, arrows pointing left are anti-phase). ………………………………55
Figure 4.13: Wavelet Coherence of TSI against the biological responses (a) net ecosystem exchange, (b) latent
heat flux, (c) ecosystem respiration and (d) gross primary productivity. The solid black line is the 5%
significance level against red noise calculated from the Monte Carlo simulation. Arrows are the relative phase
angle (arrows pointing right are in-phase, arrows pointing left are anti-phase). ………………………………57
List of Figures
x
Figure 4.14.: Wavelet Coherence of II against the biological responses (a) net ecosystem exchange, (b) latent
heat flux, (c) ecosystem respiration and (d) gross primary productivity. The solid black line is the 5%
significance level against red noise calculated from the Monte Carlo simulation. Arrows are the relative phase
angle (arrows pointing right are in-phase, arrows pointing left are anti-phase). ………………………………58
Figure 5.1: Variable temperature trends in the previous century at the Howard Springs site. (a) Temperature
trend 1913-1949 and (b) temperature trend 1950-2013. Annual average values are used to generate the time
series. …………………………………………………………………………………………………………….65
Figure 5.2: WTC analysis of rainfall and latent heat, demonstrating the driving environmental conditions at an
annual cycle, but little coherence at other periods, especially post 1970 where wetting of the ecosystem is
expected to increase latent heat and other biological drivers. …………………………………………………...66
Figure 5.3: Wavelet coherence of temperature and (left) Fe (middle) Fre and (right) GPP. Besides the annual
cycle, the strongest and most consistent cycle has a frequency greater than 10 years. ………………………….70
Figure 5.4: Wavelet coherence spectra of (a) temperature and (b) rainfall with SOI. Graphs show time series of
SOI (black) with a 10 year running mean (grey) of (top) temperature and (bottom) rainfall, displaying the
accuracy (or not) of the wavelet coherence analysis in picking up the decadal signals. ………………………..72
Figure 5.5: Variation in response to SOI conditions between carbon (Fc) and water (Fe) exchanges of savanna
ecosystem. 2.5 year periodicities of carbon exchanges respond to neutral conditions of SOI, water exchanges
respond to drier (El Nino) conditions. …………………………………………………………………………...74
Figure A.1: Australian map of climate indices with highest correlation with annual total rainfall from 1900-
2010. Resolution is 125km2. Rainfall at Howard Springs is shown to correlate with the Tasman Sea Index and
Indonesian Index. ……………………………………………………………………………………………….98
Figure A.2: Time series and linear trends of carbon flux variables showing the variation in rate of increase
between the first and second halves of the 20th
century. (a,b) gross primary productivity, (c,d) ecosystem
respiration and (e,f) net ecosystem exchange. ………………………………………………………………….100
Figure A.3: Time series of climate indices. SOI appears to have approximately equal distributions of positive
and negative (wet and dry) events while TSI and II show marked increasing in frequency of positive (warm)
SST anomalies during the last quarter of the time period. ……………………………………………………..101
Figure A.4: WTC plots demonstrating lack of coherence beyond the dominating annual cycle of rainfall and
ecosystem flux variables. (a) net ecosystem exchange, (b) latent heat flux, (c) ecosystem respiration and (d)
gross primary productivity. …………………………………………………………………………………….102
List of Tables
xi
List of Tables
Table 2.1: Summary of selected wavelet analysis studies on terrestrial ecosystems response to climate. Note the
recent emergence of the wavelet approach for land atmosphere exchanges and climate drivers. ……………....19
Table 3.1: Meteorological and flux response variables used for time series and wavelet coherence analysis. ...28
Table 3.2: Climate indices of interest in this study. Chosen for their influence on Australian rainfall and
climate……………………………………………………………………………………………………………29
Table 4.1: Summary table of climate and biological variables following spectral analysis. Observed periodicity
and associated statistically significant time periods are included, calculated as the number of years a periodicity
is observed over the total number of years in the analysis (113). Each variable is ranked by significance fraction,
to demonstrate the variability in periodicity across meteorological and biological variables. ………………….49
Table 4.2: Summary table of gross primary productivity responses to the climate indices, SOI, TSI and II,
following wavelet coherence spectral analysis. Observed periodicity and associated significant time periods and
phase angle are also listed. ………………………………………………………………………………………59
Table 4.3: Summary table of latent heat flux responses to the climate indices, SOI, TSI and II, following
wavelet coherence spectral analysis. Observed periodicity and associated significant time periods and phase
angle are also listed. ……………………………………………………………………………………………..60
Table 4.4: Summary table of ecosystem respiration responses to the climate indices, SOI, TSI and II, following
wavelet coherence spectral analysis. Observed periodicity and associated significant time periods and phase
angle are also listed. ……………………………………………………………………………………………..61
Table 4.5: Summary table of net ecosystem exchange responses to the climate indices, SOI, TSI and II,
following wavelet coherence spectral analysis. Observed periodicity and associated significant time periods and
phase angle are also listed. ………………………………………………………………………………………62
Table A.1. List of instruments and location along the tower at the Howard Springs Flux site, collecting eddy-
covariance data capturing the land-atmosphere exchanges at a savanna site. ………………………………….99
List of Acronyms/ Abbreviations
xii
List of Acronyms/ Abbreviations
ABL …………………………………….………………………………..Atmospheric Boundary Layer
ACEAS………………………………...…… Australian Centre for Ecological Analysis and Synthesis
ANN………………………………………………………………………… Artificial Neural Network
AR4………………………………………………………………… Fourth Assessment Report (IPCC)
AR5………..…………………………………………………………. Fifth Assessment Report (IPCC)
AWAP …………………………………………………………...Australian Water Availability Project
BoM………………………………………………………...……… Australian Bureau of Meteorology
BVG…………………………………………………………………………... Broad Vegetation Group
CABLE/BIOS2 ………...…………..……. Community Atmosphere Biosphere Land Exchange Model
CAWCR ………………………………………...Centre for Australian Weather and Climate Research
CSIRO……………………………… Commonwealth Scientific and Industrial Research Organisation
ENSO………………………….…………….………………………….. El Niño Southern Oscillation
FE…………………………………………………………………………………….. Latent Heat Flux
fPAR…………………………………………….….. Fraction of Photosynthetically Active Radiation
FT…………………………………………………………………………………… Fourier Transform
GCM …………………………………………………………………………….Global Climate Model
GPP……………………………………………………………………….. Gross Primary Productivity
HWS………………………………………………………………………………….. Howard Springs
IOD……………………………………………………………………………… Indian Ocean Dipole
IPCC……………………….………………………….. Intergovernmental Panel on Climate Change
LAI…………………………………………………………………………………… Leaf Area Index
LSM………………………………………………………………………………. Land Surface Model
LUE……………………………………………………………………………… Light Use Efficiency
List of Acronyms/ Abbreviations
xiii
MODIS……………………………………………… Moderate-resolution Imaging Spectroradiometer
MSLP………………………………………………………………………… Mean Sea Level Pressure
NATT………………………………………………………….. Northern Australian Tropical Transect
NCAR…………………………………………….……….. National Centre for Atmospheric Research
FC…………………………………………………….……………………… Net Ecosystem Exchange
OZFLUX………………………………... Australian and New Zealand Flux Research and Monitoring
FRE………………………………………………………………………... Net Ecosystem Respiration
RN……………………………………...………………………………………………. Solar Radiation
SOTT………………………………………………………………… 2014 State of the Tropics Report
SST…………………………………………………………………………. Sea Surface Temperature
TERN………………………………………..….. Australian Terrestrial Ecosystem Research Network
VPD…………………………………………………………………………… Vapour Pressure Deficit
WFT …………………………………………………………………….Windowed Fourier Transform
WTC……………………………………..…………………………….. Wavelet Transform Coherence
WUE……………………………………...……………………………………… Water Use Efficiency
Abstract
xiv
Abstract
Land-atmosphere interactions and feedbacks mirror a dynamically coupled system that
functions on timescales from fractions of seconds to millions of years, driven by terrestrial
ecosystem processes and modes of climate variability. Land-atmosphere exchanges involve
the transfer of mass, energy and momentum and changes to these fluxes can modify regional-
scale climate. This thesis aimed to examine the variability and temporal cycles between
temperature (Ta), rainfall and vapour-pressure deficit (VPD) as drivers and gross primary
productivity (GPP), net ecosystem exchange (Fc), ecosystem respiration (Fre) and latent heat
flux (Fe) as ecosystem flux responses, over the 20th
century. It also sought to analyse how
influential modes of climate variability were to describing variability in ecosystem responses.
Climate indices such as the Southern Oscillation Index (SOI), Tasman Sea Index (TSI) and
Indonesian Index (II) were chosen to analyse this influence.
The study site is located at Howard Springs, NT (12.5⁰S, 131.2⁰E), chosen for its long-term
flux tower and proximity to a Class 1 Bureau of Meteorology station (Darwin). The site is a
typical coastal savanna with high annual rainfall (1690mm y-1
) experiencing frequent climatic
and fire disturbance. Tower flux data trained an Artificial Neural Network (ANN) which was
driven by meteorological observations to hind-cast ecosystem responses over a 113 year
period. Using a wavelet coherence analysis, interactions between eddy covariance fluxes and
meteorological drivers are quantified at multiple time scales over the previous 113 years,
using modelled and observational data, in an effort to establish the relationship between
climate and the terrestrial carbon and water cycle within tropical north Australia.
Spectral and temporal analysis revealed significant shifts in broad climate from the mid-
1960’s onwards with increased temperatures at the Howard Springs site in line with trends
observed for Australia, namely a 0.92 oC increase in temperature over the last 113 years.
VPD also increased by 24% relative to 1910 and rainfall by 17%. In effect, the Howard
Springs site has become hotter and wetter, with an increase in evaporative demand as
demonstrated by the increase in VPD. The wavelet analysis suggested increased coupling
between the climate and ecosystem exchanges with increased temperature, rainfall and CO2
concentrations. Water fluxes (Fe) showed sensitivity to El Nino phases (drier conditions)
while carbon fluxes (GPP, Fre and Fc) consistently correlated with neutral conditions,
indicating disconnect between the carbon and water exchanges in response to climate change.
Abstract
xv
These results can be further developed by analysing additional sites to unpack the
mechanisms involved between savanna and climate. The novel approach and unique study
period length has contributed to our understandings of the influence of past and present
climate variability to savanna land-atmosphere exchanges, providing insight to the behaviour
of coastal savanna ecosystems, which covers tens of thousands of square kilometres of
tropical north Australia.
Chapter 1: Introduction
1
Chapter 1: Introduction
Quantifying the association between ecosystem flux and climate at multiple time scales is
essential for a broad understanding of the role of climate and its impact on ecosystem
function (Stoy et al. 2009). The Earth’s surface is covered by just 29% land; however,
terrestrial ecosystems store approximately 1600 giga-tonnes of organic carbon, more than
twice that of the atmosphere (Pielke et al. 1998). Annually, ecosystems sequester more than a
quarter of all anthropogenically-released carbon emissions (ACREAS 2013), an important
ecosystem service. The carbon cycle is also of critical importance to atmospheric processes
through alteration to the planetary energy balance (Gornall et al. 2009) with increased
emissions of CO2 over the last century. The temporal and spatial distribution and magnitude
of this terrestrial sink is uncertain, particularly in tropical regions (Pan et al. 2011). The
weather and climate at the Earth’s surface influences many social and economic activities
calling for a realistic simulation of the land surface within climate models (Bounoua et al.
2002).
Within the framework of climate change, the carbon cycle is coupled with the water balance
through assimilation and water-use of vegetation (Seneviratne & Stockli, 2008). Mass
(carbon, water) and energy (radiation, heat, and momentum) is exchanged between the land
surface and the lower atmosphere. These land-atmosphere exchanges occur on temporal
scales from seconds to centuries and these processes that influence the meteorology from
regional to planetary scales. Interactions and feedbacks between vegetation and climate are
complex, operating over a range of temporal scales (Pitman 2003). The climate system and
terrestrial ecosystems are closely coupled by the cycling of carbon between vegetation, soils
and the atmosphere. The carbon cycle is hypothesised to have been modified by changes in
the climate and atmospheric CO2 concentrations, with an enhanced terrestrial sink and
storage proposed (Poulter et al. 2014) however, direct evidence for this is limited. Ecosystem
carbon stocks have changed, caused by shifts between stable climate states however, the
dynamic responses of ecosystem carbon fluxes to a changing climate are still poorly
understood (Cao & Woodward 1998).
This thesis aims to investigate this potentially changing relationship between climate
variability and the exchanges of mass and energy for a tropical savanna ecosystem. The
Chapter 1: Introduction
2
savannas of north Australia occupy a third of the land surface and understanding the coupling
between climate and ecosystem processes such as carbon sequestration, water use and energy
balance is critical for meaningful predictions of ecosystem response to climate change.
1.1 Land-atmosphere exchanges
The importance of understanding and quantifying land-atmosphere interactions and their
influence on the global climate system is increasingly being recognised (Seneviratne, &
Stockli, 2008). Climate modelling studies show that terrestrial ecosystem dynamics are as
important as changes in atmospheric composition in determining the rate of land-atmosphere
exchanges (Pielke et al. 1998). The exchange of mass and energy occurs within the
atmospheric boundary layer, the lower 500-2000m of the atmosphere. Early climate models
(1980’s- 1990’s) ignored this feedback with land surfaces, focusing solely on oceanic
processes which influenced climate variability (Pitman 2003). The sophistication of climate
modelling over the past forty years and current models integrate multi-component
representations of the full climate system, with functional plant types (FPT’s) defined that
capture spatial and temporal characteristics of land surfaces (Verburg et al. 2011). Terrestrial
ecosystem feedbacks can now be represented as a dynamically-coupled system with the ABL.
Chapter 1: Introduction
3
Figure 1.1 illustrates both energy (latent heat, sensible heat) and carbon fluxes
(photosynthesis, respiration) as the most important components of the land-atmosphere
interactions.
Climate models are most commonly used to explore the impact of trepidations in climate due
to human activity, however, a significant failure remains in many climate models and their
ability to quantify the land surface and land cover change for climate perturbation (Pitman
2003). This is worrisome since humans have significantly altered a huge fraction of the
Earth’s surface through land cover change, which is likely to accelerate over the next century
as a result of direct impacts due to deforestation, reafforestation and agricultural expansion,
the latter of which is a huge threat to the tropical land surface. With the suggestion that rising
temperatures may decrease the carbon sink by terrestrial ecosystems as a result of rising CO2
levels (Donohue et al. 2013; Pitman 2003), there is a need for improvements to
Figure 1.1: Representation of the main processes and variables in the coupled land surface-ABL dynamic
system. The carbon and water cycles are represented as flux components within the ecosystem. Taken from
Vila-Guerau de Arellano et al. (2012).
Chapter 1: Introduction
4
understandings of land surface and terrestrial ecosystems and their feedbacks to the climate
system.
1.2 Climate change in Australia
Two leading Australian scientific organisations recently released the State of the Climate
Report for 2014 (BoM & CSIRO 2014) summarising the change in climate since the
beginning of the 20th
century. This includes a continent-wide warming of 0.8⁰ C-0.9⁰ C since
1910 which has altered the frequency of extreme weather with more heating events and fewer
cooling extremes. Rainfall has increased on average since 1900, largest in the northwest since
1970. The south-east and south-west has experienced declined rainfall in the same period
(1970) as a result of reduced winter precipitation. The fire season has lengthened across large
parts of Australia since the 1970’s and ocean heat storage has increased as the global mean
sea level has risen by 225mm since 1880.
Modes of climate variability that affect Australian weather, such as the El Nino Southern
Oscillation (ENSO) and Indian Ocean Dipole (IOD) have also experience changes since the
1970’s. Since the mid-1970’s ENSO behaviour has changed and a stud y by Timmermann
(2001) revealed that global warming has shifted the ENSO from stable oscillatory behaviour
to an unstable oscillation, coinciding with an abrupt change in simulated ENSO activity and
changed Pacific Ocean dynamics. Cai et al. (2009) noted the recent high frequency of
positive IOD occurrences, indicating warmer SST across the Indian Ocean, which has
impacts for Australian rainfall. Positive IOD events have increased from four every 30 years
in the early 20th
century to over 10 in the last 30 years, at the same time, cooler SST events
have shifted from 10 over the same period to two. There are now more positive IOD events
than there are negative IOD’s which is also consistent with the reduction in rainfall over
south-eastern Australia since the 1950’s. The warming of the Indian Ocean has also led to the
synchronisation of the IOD with ENSO through the Walker circulation (Cleverly, pers.
Comm. May 2014).These climate phenomena are quantified using climate indices such as the
Southern Oscillation Index which describes Pacific Ocean conditions; Indonesian Index, an
anomaly of SST’s in the Indonesian region and the Tasman Sea Index, used to define extra-
tropical climate.
Building on a large body of work undertaken for the Australian region, climate change
projections for Australia are based on international climate change research. These changes
Chapter 1: Introduction
5
have the potential to create major impacts on human and natural systems and are likely to
continue if emissions of greenhouse gases increase (Climate Institute Report 2014).
Temperature and rainfall
The IPCC Fourth Assessment Report (AR4) predicts that temperatures are likely to warm at
amplitudes larger than that of surrounding oceans (AR4, 2007) within Australia. This
warming is expected to be smaller in the south and most prominent in the winter; frequencies
of extreme high daily temperatures are predicted to increase in Australia whilst vice versa, a
decrease in extreme cold conditions are expected (AR4, 2007). These conclusions are
reiterated in the IPCC’s Fifth Assessment Report (AR5), which noted that the rate of
warming above the land surface was twice that over oceans since 1979 (AR5, 2013).
Minimum temperatures have warmed faster than maximum temperatures and Australia has
warmed ~0.8⁰C over the last century, consistent with global trends (Hughes et al. 2003).
Rainfall trends in Australia are highly spatially-variable, with increases in the north-west of
about 300mm since the 1970’s with corresponding decreases in the south-east (CSIRO, BoM
2014; AR5 2013). These variable trends seem to be responding to changes in the surrounding
oceans, especially in the case of declined rainfall and the terrestrial water balance of south-
west Western Australia. England et al. (2006) observed a characteristic dipole pattern of
Indian Ocean sea surface temperature (SST) anomalies during extreme rainfall years, with
peak amplitudes in the eastern Indian Ocean adjacent to the west coast of Australia. Tropical
and extra-tropical processes in the Indian Ocean generate SST and wind anomalies off the
south-west coast leading to moisture transport and rainfall extremes in the region, a
significant agricultural area. Rainfall is a driver of ecosystem processes and changes in
rainfall seasonality will influence future ecosystem services such as carbon storage however,
these changes are not expected to turn vegetation into a carbon source which suggests that
CO2 fertilisation is the main driver of vegetation change (Scheiter et al. 2014).
1.3 Climate change in savanna-dominated north Australia
Understanding the effects of climate change in North Australia is important given its
contribution to the terrestrial carbon balance. It is of critical importance to the global carbon
and water budget as it accounts for 12% of global savanna land cover. Climate phenomenon
such as ENSO and IOD affect Australian rainfall which is turn influences the carbon sink of
terrestrial ecosystems through gross primary productivity (GPP). The influence of the IOD is
Chapter 1: Introduction
6
strongest in coastal north Australia and changes to these modes of climate variability will
have significant impact for the terrestrial carbon budget.
The Australian savannas experience one of the most seasonal of the world’s savanna climates
(Bowman & Prior 2005) as a result of the annual movement by the Australian summer
monsoon. The strong wet-dry seasonality resulting from the north-south movement of the
monsoon are compounded by the state of near-constant disturbance by other climate
extremes, such as cyclones and fire weather. These disturbances occur over large areas
creating a complex spatial and temporal mosaic of carbon sources and sinks, based on
previous disturbances, fire regime and long-term climate (Hutley et al. 2013).
In biomes dominated by C4 grasses, tree cover is suppressed by periodic fires causing
mortality of aboveground organs (Bond 2008). In the Northern Territory’s savanna, grass-tree
ratio increases with distance from coastline, following a decrease in rainfall and fire
occurrence (Hutley et al. 2011). Fire, tree cover and C4 grass feedbacks are vulnerable to
disruption via changes in climate, atmospheric CO2 concentrations and fire regimes (Bond &
Midgley 2012; Moncrieff et al. 2013). Tropical land cover change through agricultural
expansion will disrupt biogeochemical cycles and community structure (Parr et al. 2012),
which is particularly significant for a biome such as the extensive savanna, which stores
billions of tonnes of aboveground carbon (Lewis et al. 2009). Increased atmospheric CO2
concentrations may increase or decrease terrestrial carbon storage, given potential variations
to processes that may be affected by climate change, such as productivity, ecosystem
respiration rate or fire regimes. When these alterations are applied at the landscape level, they
may impact on climate through local changes in circulation patterns, regional heating,
precipitation and monsoon circulation (Beringer et al. 2003).
Savanna and tropical forest cover most of the land surface are of the tropics, dominating the
exchanges between land and atmosphere in this region. Thus it is important to quantify these
interactions accurately within global climate models (GCM’s). Quantifying these exchanges
within the tropics is difficult given the uncertainty of the source or sink status of tropical
terrestrial ecosystems to atmospheric carbon dioxide concentrations.
Savannas are important to global circulation given their size (approximately 20% of Earth’s
land is covered by savannas), current clearing rate and annual burning, which shifts surface
albedo and net exchange (Knapp et al. 2001). Annual burning emits large amounts of CO2,
methane and N2O to the atmosphere (van der Werf et al. 2010). Given this fire and
Chapter 1: Introduction
7
consumption of fuel, there is uncertainty surrounding the magnitude of the tropical savanna
sink. Beringer et al. (2014) reported a weak source for north Australian savanna, whereas
many studies have found tropical ecosystems to be a significant carbon sink (e.g. Pan et al.
2011; Phillips & Lewis 2013; Wright 2013), though the amplitude of that sink is argued
(Wright 2013). These uncertainties are further compounded by a lack of knowledge
surrounding climate variability and climate change in tropical north Australia; it is unknown
how variability will change in the future.
Some of this uncertainty can be resolved by understanding the climatic variability (cycles)
and the response of ecosystem to these climate cycles, manifest through shifts in productivity,
respiration and surface energy balance through time. Understanding past climate change and
ecosystem response is critical to understanding the future impacts to savanna ecosystems and
global carbon and water budgeting.
1.4 Mathematical techniques for time-series analysis
Ecosystems and climate are coupled through time and a time series analysis provides a
method to assess the strength and change of this link between climate and vegetation and the
associated driver-response mechanism. This can be done using a frequency-domain analysis
that unpacks the complex interactions between the savanna ecosystem and tropical climate
variability. The analysis needs to understand the spectral power and cycling of individual
variables before two can be coupled to assess the driver-response relationship. It also needs to
unpack the lagged response and phase shifts between two variables to understand savanna
function as a response to environmental conditions. Understanding these physical processes
and how they influence biological exchanges is a fundamental goal of functional ecology.
Wavelet analysis
Frequency-domain methods such as the Fourier series approximate a function by
decomposing it into sums of sinusoidal functions, making it capable of detecting dominant
frequencies in a signal (Tran 2006). However, the Fourier transform is only localised in
frequency, whilst wavelet analysis is localised in both time and frequency (Chang & Glover
2010). The Fourier transform can evaluate the size of the component of frequency with
precision, but in doing so loses all control of spatial duration so the time that the signal is
sounded cannot be determined. This is the limiting case of the Uncertainty Principle:
precision on frequency but zero control on temporal spread. The wavelet transform takes
advantage of the intermediate cases of the Uncertainty Principle as each wavelet
Chapter 1: Introduction
8
measurement (corresponding to a fixed parameter) describes both the temporal extent of the
signal as well as the frequency spectrum of the signal. Frequency precision is sacrificed in
order to gain temporal information in the wavelet transform and also ignores the assumption
of stationarity (a time series that does not change through time), making the wavelet a
superior method for analysis of ecological and geophysical time series. This trait is a strong
indicator of the suitability of the wavelet analysis in this study as it represents the coupling
through time between ecosystems and climate. The power of the wavelet transform to
understanding land-atmosphere exchanges between savanna ecosystems and atmospheric
processes is further exhibited through its ability to demonstrate the lagged response and phase
shifting at various cycles, by examining the driver-response mechanism as a mathematical
relationship.
Figure 1.2 presents two varying signals, red and blue analysed using the wavelet transform
coherence function demonstrating the wavelet’s ability to recognise what frequencies are
present and where in time, observing the signal’s change with time (ignoring the assumption
of stationarity).
On the coherence diagram, the thick black line surrounding the high correlating periods is the
cone of influence (COI), beyond which edge effects may distort the statistical confidence of
the coherence. The colour bar to the right shows the statistical strength in the degree of
Figure 1.2: Wavelet transform coherence of red and blue signal, observing the different properties of phase shift, frequency
and amplitude between the two signals. Taken from Xu Cui (2014)- http://www.alivelearn.net/category/matlab/page/3/
Chapter 1: Introduction
9
coupling between two variables (significance level 0-1). The y axis is time frequency
(periodicity) and x is time. The black arrows demonstrate the phase shift, left and right
pointing arrows show anti-phase and in-phase (0⁰ and 180⁰ respectively) and down and up
pointing arrows show the blue signal to be lagging by 90⁰ and 360⁰ respectively. The time
series is shown at the bottom of the figure, correlating with the coherence plot.
This novel approach is an ideal method for this project given that the ecological and
geophysical variables to be used ignore the assumption of stationarity.
1.5 Aims and objectives
Given the known coupling between vegetation and climate and the uncertainty in the tropics,
this thesis will focus on one tropical savanna site in Howard Springs, NT. Rainfall variability
at this site correlates with climate indices such as the Southern Oscillation Index (SOI),
Tasman Sea Index (TSI) and Indonesian Index (II). It is also known to be influenced climate
extremes such as cyclones and annual fire regimes. The degree of coupling between cyclic
patterns of environmental conditions (temperature, rainfall) and climate variability (SOI, TSI,
II) with ecosystem responses will be investigated, focusing on the strength of cycling and
phase shifts throughout the 20th
century. This study is interested in the relative response and
coupling to determine coherence and relationships by savanna ecosystems to meteorological
drivers and climate indices, not the prediction of past fluxes.
In order to achieve this, a sophisticated time series analysis known as the wavelet transform
will elucidate the driver-response mechanisms between climate and ecosystem by quantifying
ecosystem sensitivity to climate variability and identify potential lag responses at various
time scales.
This project aims to significantly improve our understanding of temporal variability of key
climate indices and their impact on savanna ecosystem function, where function is captured
by variables describing carbon ( productivity, respiration) and energy fluxes (latent heat or
evapotranspiration). Specifically, the following questions are posed;
1) How much variability in savanna biotic responses (productivity, respiration and
evapotranspiration) is due to climatic changes at seasonal and longer (greater than
annual) time scales?
2) What is the nature of the temporal relationship between meteorological drivers
(temperature, precipitation, vapour pressure deficit) and these biological variables?
Chapter 1: Introduction
10
3) How influential is the role of modes of climate variability as integrated by climate
indices (e.g. SOI, TSI, and II) to describing variability in ecosystem responses?
4) What, if any, lags exist between drivers and biological responses and why?
This thesis is structured as follows. Chapter Two provides a literature review of recent work
on savanna land-atmosphere interactions in north Australia and probable climate drivers and
associated climate indices. Mathematical techniques to correlate biological and climate
indices are expanded in Chapter 3, which details approach and methods. Chapter 4 describes
results from the century long time series of modelling outputs describing climate and
biological variables. Both linear trends and power spectral density data are presented that
describe inherent temporal cycles. Wavelet coherence plots are also presented, where the
degree of coupling of climate and biological variables is analysed and changes over century
time scales are examined. In Chapter 5 these findings are discussed in light of previous
studies, with conclusions summarised and recommendations for future research in climate
and land surface modelling given in the context of tropical ecosystems detailed in Chapter 6.
Chapter 2: Literature Review
11
Chapter 2: Literature Review
This chapter provides an overview of the modes of climate variability influencing tropical
meteorology, noting the recent emergence of two significant climate indices and their
implications for land-atmosphere exchanges in tropical north Australia. Current
understanding of carbon dynamics as driven by climate is also reviewed, highlighting the
cyclical nature and coupling of climate and ecosystem processes. The wavelet transform is a
frequency-domain time series analysis and has been established as a useful technique for
assessing coherence between ecological and geophysical time series (Hong & Kim 2011).
The ecological significance of the savanna biome is also examined, given its unknown
vulnerability to climate change. The literature review concludes with a short summary on the
history of land surface models.
A gap exists in the understanding of climate change impacts on a wet/ dry tropical savanna
ecosystem such as Howard Springs at various temporal scales. Given the importance of
climate for terrestrial ecosystems (e.g. Di Lorenzo et al. 2008; Eamus et al. 2013; Knapp et
al. 2008; Nemani et al. 2003; Overland et al. 2010; Peterson & Schwing 2003; Petheram et al.
2012), this uncertainty has major implications for the savanna biome and the terrestrial
carbon and water balance of tropical north Australia. Figure 2.1 demonstrates the linkages
between ecosystems and the atmosphere, as outlined by Beringer et al. (2011). Land-
atmosphere exchanges are modified by ecosystem properties such as structure, composition
and function and play a role in determining atmospheric conditions.
Figure 2.1 Conceptual diagram linking ecosystem and climate from ecosystem to global scale. (Source: Beringer et al. 2011, adapted from
Chapin et al. 2005 and Bonan 2008).
Chapter 2: Literature Review
12
Variability in ecosystem characteristics modifies land-atmosphere exchanges at the regional
scale, influencing the overlaying atmosphere and boundary layer. However, at the meso-
scale, this spatial variability can create disparities that drive mesoscale climate, such as the
Australian summer monsoon.
2.1 Drivers of tropical climatology
A notable publication is the State of the Tropics Report of 2014 (SOTT
http://stateofthetropics.org ), which examines predicted change within the tropics and
implications for the sub-tropical, temperate and polar regions (SOTT 2014). Key interests of
this report are the climate drivers of variability within the tropics, most notably, the South-
East Asia and Northern Australian monsoon and El Nino Southern Oscillation (ENSO).
Absent are predictions for change to the Madden Julian Oscillation, a little-known climatic
phenomena with potentially significant influence on the climatic and thus tropical ecosystems
due to interactions with the monsoon and ENSO systems (Hendon 2004; Kumar et al. 1999;
McBride & Wheeler 2004; Smith et al. 2008; Tang & Yu 2008). Within the tropics of
northern Australia, climatological research has concentrated predominantly on the Australian
summer monsoon (Bowman et al. 2010; Cook & Herdegen 2001).
The North Australian Monsoon
The wet-dry tropical climatology of North Australia is dominated by the presence of the north
Australian summer monsoon system via land-ocean pressure differences driving convection
(stormy weather) and low pressure systems, resulting in heavy rainfall and cyclones. The
summer monsoon provides most of the annual rainfall in the north Australian region,
dynamically similar to the system that affects the South Asia region, but with much less
strength and geographical extent (Suppiah 1992).
A number of papers on the wet season in north Australia have focused on predicting and
characterising the onset and duration of monsoonal rain events such as Cook & Heredegen
(2001) and Smith et al. (2008). The spatial variation in duration of wet seasons is critical to
understanding how climate variation affects ecosystems (Cook & Heredegen 2001), which
has driven many studies to characterise the monsoonal influence on the tropical biosphere.
There is however a lack of understanding in the temporal variation of climate, as driven by
the monsoon wet season and its effects on terrestrial ecosystems, over long periods of time
(Beringer et al. 2011).
Chapter 2: Literature Review
13
Monsoon onsets are not a recent interest, Troup (1961), Holland (1986), Hendon & Liebmann
(1990a) and Drosdowsky (1996) examined onset dates and described various methods for
defining the Australian summer monsoon. The most recent undertook a re-examination of
monsoonal activity utilising 35 wet seasons, from 1957-1992 via what is described by the
author as an easily applied objective definition of active and break phases, based solely on
zonal winds at the Darwin International Airport (Drosdowsky 1996). By discarding the
previously used term of ‘wet westerlies’ since there is no clear relationship between westerly
winds and rainfall (Drowsdowsky, 1996), the new onset dates are significantly related to the
Southern Oscillation index (SOI). The inter-annual variability in monsoon onset and
monsoon intensity correlates well with the El Nino Southern Oscillation and using cross-
spectral analysis, onset dates have shown a strong relationship to SOI in the spring prior to
the wet season (Drosdowsky, 1996). This suggests that an investigation into the lags between
climate indices and ecosystem function, especially with SOI is needed to better understand
the dynamical interactions between drivers (climate) and responses (ecosystem exchanges).
El Nino Southern Oscillation
The El Nino Southern Oscillation (ENSO) primarily influences rainfall in the spring, summer
and autumn over the Australian continent, due to the direct impacts of ENSO on the active
South Pacific Convergence Zone during these months (Folland et al. 2002; Franks 2004).
Additionally, a number of authors (e.g. Kiem et al. 2003; Franks 2004; Power et al. 1999)
have demonstrated the influence of the Inter-decadal Pacific Oscillation (IPO) on Australian
rainfall variability at decadal and multi-decadal timescales. Fewer studies exist to quantify
the relationship of ENSO to local rainfall variability in tropical north Australia, though its
influence on monsoonal systems is acknowledged (Wang et al. 2003).
Numerous studies have acknowledged patterns of sea surface temperature (SST) variability
occurring over the Indian Ocean (e.g. Ashok et al., 2003; Nicholls, 1989; Saji et al.1999;
Smith, 1994) that are consequently linked to rainfall variability in Australia, Asia and East
Africa, raising the possibility of a climate phenomenon separate to ENSO influencing
Australian rainfall. This is particularly evident given the absence of ENSO influences over
winter, even though these months display a high degree of variability.
The occurrence of an El Nino phase results in a fluctuation in tropical Pacific Ocean sea
surface temperatures (SST) whereby the equatorial surface waters warm considerably from
approximately 180oE/W to the western shoreline of South America (Stenseth et al. 2003).
Chapter 2: Literature Review
14
The atmospheric phenomenon coupled to the ENSO is called the Southern Oscillation,
quantified by the Southern Oscillation Index (SOI), in which exchanges of air occur between
the eastern and western hemispheres within tropical and sub-tropical latitudes.
Much of the climate literature in both the Southern and Northern Hemisphere has focused on
the ENSO phenomenon over the past several decades working on the pioneering papers by
Jacob Bjerknes in 1966. Before this, the term El Nino was used for centuries referencing the
annual incidence of warm, southward-flowing oceanic current (upwelling) off the coast of
Peru and Ecuador at Christmas-time (Trenberth & Hoar 1997). Qualitative listings of El Nino
events date back to 1726 by Quinn et al. (1987) based solely on this coastal phenomenon.
More recent literature defines the large-scale warming that occurs every few years, which
changes the local and regional ecological conditions, as ENSO (Strenseth et al. 2003). This
Pacific-basin-wide occurrence links with anomalous climate patterns around the world
(Chiew et al. 1998), including the Pacific North-American in the Northern Hemisphere and
the Pacific South American in the Southern Hemisphere (Strenseth 2003). ENSO forms two
phases, the cold, termed El Nino, and warm conditions are known as La Nina.
Within Australia, a relationship exists (though its strength varies with geographical proximity
to the Pacific) to rainfall, drought and streamflow (Chiew et al. 1998). The hydro climate of
Australia has captured the attention of many climate scientists given the spate of recent
extreme events, which can be coupled with the warming and cooling phase of ENSO.
Analyses demonstrate that dry conditions in Australia are typically associated with El Nino;
whilst rainfall and streamflow are statistically significant in most regions, ENSO is not
sufficiently strong enough to consistently predict such variables (Chiew et al. 1998). The
teleconnections vary in strength temporally, stronger in the latter half of the year, providing
an invaluable resource to land and water management in Australia where streamflow
variability is one of the highest in the world. Risbey et al. 2009 describe the ENSO as the
principle driver of climate in the tropics but concede that the Indian Ocean has some effect,
especially on rainfall variability. Recent work has highlighted the contribution of the Madden
Julian Oscillation (MJO) to tropical rainfall variability through its manifestation on the
Australian Summer Monsoon (Hendon & Liebmann 1990; Wheeler & Hendon 2004; Risbey
et al. 2009; Wheeler et al. 2009).
Chapter 2: Literature Review
15
Tasman Sea Index
The Tasman Sea Index (TSI; Murphy & Timbal 2008) is a measure of sea surface
temperature anomalies in the Tasman Sea at 30-40⁰S, 150-160⁰E. Murphy & Timbal (2008)
have established an association between TSI and autumn rainfall in south-eastern Australia,
where warm SST anomalies in the Tasman Sea are connected with increased rainfall within
the region. One reason for the focus on this region is the high population and agricultural
production density
Schepen et al. (2012) also examined TSI in the same paper analysing the lagged relationship
between II and rainfall over the northern tips and eastern Australia and established a similar
relationship. There was strong to very strong evidence supporting the one month lagged TSI
to forecast rainfall in the western half of Australia in the months October – January. However
since the relationship between seasonal rainfall and TSI has not been studied in this region,
the mechanism for this connection is not understood.
Indonesia Index
The Indonesia Index (II) was first classified by Nicholls (1989) as an index of SST
differences between the Indonesian region (0-10⁰S, 120-130⁰E) and the central Indian Ocean
which was found to be strongly associated with a broad band of rainfall variability stretching
from the northwest to the southeast corners of Australia and negatively correlated with east
coast rainfall. This is indicative of yet another factor that influences Australian rainfall,
largely independent of the well-known Southern Oscillation. Based on limited data,
O’Mahony (1961), Priestley (1964) and Priestley & Troup (1966) appear to be the first to
suggest the link between SST’s and Australian rainfall but it was not until the 1980’s that
studies identified the positive correlation of SST’s around Indonesia with Australian rainfall
(e.g. Kep 1984; Streten 1981; 1983; Whetton 1986).
More recently, a paper by Verdon & Franks (2005) investigating the aforementioned winter
rainfall variability over Australia explored the II in depth and found its anomalous SST’s to
be the most promising for potential practical application due to its ability to identify a
balanced and realistic frequency of wet and dry years. The SST’s over the Indonesian region
were also found to provide a good indication of winter rainfall over eastern Australia,
contradicting earlier studies. The predictive use of II on Australian rainfall is further
evidenced in Schepen et al. 2012 who identified the strong correlations between SST around
Indonesia and rainfall over the central northern tips and parts of eastern Australia. The II was
Chapter 2: Literature Review
16
lagged out by two months and used in a Bayesian joint probability modelling approach to
forecast spring rainfall with positive results.
Climate indices such as the SOI, TSI and II have seasonal and cyclic patterns, quantifying
complex and stochastic climate phenomena driving global exchanges of mass and energy
between the land surface and atmosphere. They are also known to have changed through
time, particularly during the last 100 years as a result of climate change and warming oceans.
A time series analysis is required to examine these indices and their coupling to ecosystem
exchanges in tropical north Australia.
2.2 Analysing climatic and biological cycles
Methods of time series analysis may be divided into two classes, time-domain methods and
frequency-domain methods. The latter method includes spectral analysis and the more recent
wavelet analysis. These techniques may be further categorised into parametric and non-
parametric methods, the former assumes that the underlying stationary stochastic process has
certain structures that can be described parametrically. Non-parametric approaches by
contrast explicitly estimate the spectrum of the process without assuming that the process has
particular structure (Lin et al. 2003). Spectral analyses examine cyclic behaviour, separating
components representing trend, seasonality, variation and cyclical irregularity in what is
known as decomposition of a time series.
Wavelet transforms in data analysis have been broadly formalised thanks to mathematicians,
physicists and engineering efforts (Morlet 1983), growing exponentially in popularity over
the past two decades since it represents a synthesis of old techniques coupled with robust
mathematic results and efficient computational algorithms (Daubechies et al. 1992). The
introduction of the time frequency decomposition to wavelet analysis allowed for its use in
ecological applications, where Fourier transforms were previously used (Cazelles et al. 2008).
These decomposition properties, particularly in its time-scale localisation, make it an
attractive tool within science and engineering for the analysis of non-stationary systems. It
can thus be used to examine the periodicity of climate and co-varying biological processes.
Fourier vs. wavelet analysis
Before the identification of the wavelet transform for use as a time series analysis, the FT was
a tool often used to study frequency distributions of geophysical time series. However, the FT
lacked in demonstrating the temporal locality of frequency distributions (Weng & Fau 1994)
which weakened its ability to identify changes through time. There was also a failure in the
Chapter 2: Literature Review
17
Fourier series by way of non-stationary signals that occurred during a short time interval
which could not be detected, thus there was a need for an algorithm that identified an
alteration of a signal in a small neighbourhood at a time locality that did not affect the entire
spectrum series (Weng & Fau 1994).
Whereas the Fourier transform (FT) is only localised in frequency, wavelets are localised in
both frequency and time. It ignores the assumption of stationarity, making it the ideal choice
for analysing time series of ecological processes (Cazelles et al. 2008). The wavelet
transformation is a powerful tool for multi-scale analysis and interactions that display
localisation in frequency and time through continuous and orthogonal functions (Weng &
Lau 1994). Given its capability to identify variations of both time and frequency, the wavelet
makes for an ideal analysis tool in this study, using time series of climate and biotic
responses which are cyclic in nature.
Gabor (1946) first identified the windowed Fourier transform (WFT) during his study of non-
stationary time series by localising signals in both time and frequency domains. The WFT
had limitations pertaining to detecting high-frequency signals rooted in low-frequency
phenomena and a fixed window for time frequency width series (Weng & Fau 1994). The
wavelet analysis is based off the WFT.
Early studies using the wavelet analysis identified faults, within the analysis itself, for the
misconception that it only produced qualitative results by way of ‘colourful pictures’ as it
transforms a one-dimensional time series to a diffuse two-dimensional time-frequency image
(Torrence & Compo 1998a). The lack of quantitative results from this method was
compounded by the lack of significant statistical tests and use of arbitrary normalisations.
The wavelet approach
The aim of the wavelet approach is to expand a signal into different waveforms with local
time-frequency properties which are well-adapted to the signal’s structure (Cazelles et al.
2008). The wavelet analysis is now a widely-used tool in medical research (most often in
neurological studies) and is gaining popularity within studies of geophysical time series
since, dissimilarly to the Fourier analysis, it analyses the signals with changing spectra and
thus the spectral characteristics of a time series is a function of time (Velda et al. 2012).
The wavelet method allows the analyser to view correlations that may otherwise not appear
from other methods. Frequencies of high correlations are shown in red, which can occur at
Chapter 2: Literature Review
18
different temporal scales and show common high power between two variables for a
particular periodicity. The wavelet analysis is appropriate for following gradual change in
forcing by exogenous variables, permitting an analysis of the relationships between two
signals; its versatility is exacerbated by its decomposition properties, especially its time-scale
localisation (Cazelles et al. 2008), making it a viable choice of methodology for this project.
Applications
The transform is notable for its significance in many neurological studies (Adeli et al. 2003;
Muthuswamy & Thakor 1998) and is becoming popular in geophysical studies, particularly in
its analysis of tropical convection by Weng & Lau (1994), El-Nino Southern Oscillation (Gu
& Philander 1995) and atmospheric cold fronts as analysed by Gamage & Blumen in 1993
(Torrence & Compo 1994). The wavelet analysis has been used on one other similar study
(Hong & Kim 2011) investigating climate and ecosystem response however, the period of
examination went over just three years using eddy covariance observations. This thesis is the
first of its kind using wavelets as a time series analysis over a 113 year period.
The relationships between environmental and ecological time series, as well as their specific
cyclic nature, gives rise to the attractiveness of wavelet analysis and Cazelles et al (2008)
provides abundant examples of such studies dating back to 1954, when spectral analysis in
ecology was first described. Table 2.1 lists recent studies that have used a wavelet approach
in the field of ecology, which gained popularity in the early 21st century. It has found a niche
for many other time series studies such as in Grenfell et al. (2001), who found synchrony
patterns of measles in the UK using wavelet analysis; Nezlin & Li (2003) used wavelets to
compare features of chlorophyll and environmental time series in a qualitative study; an
association between North American porcupine dynamics, local climate and the solar cycle
was demonstrated in 2004 by Klvana et al.; Cazelles et al. (2005) used the approach to
validate the highly significant and non-stationary association between El Nino, precipitation
and dengue epidemics in Thailand (Cazelles et al. 2008).
Table 2.1: Summary of selected wavelet analysis studies on terrestrial ecosystems response to climate. Note the
recent emergence of the wavelet approach for land atmosphere exchanges and climate drivers.
Reference Topic
Baldocchi et al. (2001) Spectral analysis of biosphere-atmosphere trace gas densities
Hong and Kim (2011) Impact of Asian Monsoon on ecosystem carbon and water fluxes
Katul et al. (2001) Multiscale analysis of vegetation surface fluxes
Stoy et al. (2006) Separating effects of climate and vegetation on evapotranspiration
Chapter 2: Literature Review
19
Stoy et al. (2007) Coupling of ecosystem carbon inputs and outputs on short
timescales
Stoy et al. (2009) Biosphere-atmosphere exchange of CO2 in relation to climate
Vargas et al. (2010) Multi-scale analysis of temporal variability of soil CO2
productions
Vargas et al. (2011) Multi-temporal correlation between photosynthesis and soil CO2
efflux
Vargas et al. (2013) Drought influence on accuracy of simulated ecosystem fluxes
Post (2013) and Klvana et al. (2008) recognise the variety of ecosystems that are driven by
large-scale climatic oscillations (Cazelles et al. 2008) such as the ENSO and MJO, however
little literature exists to demonstrate the ecological impacts of climate variability globally, but
especially in Australia, due to a lack of data supporting such hypotheses, until now. Recent
advances in flux data management have been accelerated by the efforts of Beringer via the
development of data gap filling and analysis tools and this project is the first of a new
generation of analysis that can now been undertaken given the advanced state of data
availability, both climate and flux time series that describe biological processes (e.g. net
ecosystem exchange, gross primary productivity, evapotranspiration, light-use efficiency,
water-use efficiency).
2.3 Savanna – the dominant north Australian ecosystem
This project is focused on understanding climate drivers of productivity and
evapotranspiration of a high rainfall (1690 mm y-1
) tropical savanna. The tropical savannas of
north Australia occurs across the wet-dry tropics of north Australia, stretching from the
Kimberley region, across NT and into central Queensland above the 600mm isohyet as
defined by Fox et al. (2001), with an area of some two million km2 (Figure 2.2). Globally,
tropical savanna occupies ~ 27.6 million km2 of Australia, South America, Asia and Africa
(Hutley & Setterfield 2008), consisting of a mixture of competitive C3 (trees) and C4
(grasses) vegetation types, both distinct from grasslands and closed forests. The competitive
balance creates highly variable spatial and temporal landscape dynamics (House et al. 2003;
Hutley & Beringer 2011). Australian savanna is one of the most ecologically intact in the
world which is attributable to the indigenous lands and conservation reserves which occupy
about 40% (Hutley & Beringer 2011), and experiences one of the most seasonal climates in
the world (Prior et al. 2009).
Chapter 2: Literature Review
20
The savanna biome provides a unique platform for research into ecology, ecosystem
functionality and climate change in both Australia and tropical regions around the world.
Given the size of the savanna biome globally and in Australia, well-calibrated land surface
models (LSM) and dynamic global vegetation models (DVM) are essential to predict
distribution and vegetation-atmosphere feedbacks at regional and global scale (van der Werf
2006). Despite this need, savanna is poorly characterised in current models, resulting in
difficulty in predicting future vegetation distributions (Scheiter & Higgins 2009). Current
research efforts are concentrated on improving model prediction of functional attributes
(gross primary productivity, net ecosystem exchange, water-use efficiency, light-use
efficiency). Ma et al. (2013) described the spatial patterns and temporal dynamics of savanna
phenology across north Australian savanna, a key parameter in climate and biogeochemical
cycle models, given the importance of leaf area.
There are no studies that have examined tropical land atmosphere exchanges and climatic
forcings at such an extensive temporal scale. Just as the literature on climate and ecology is
biased toward global (large-scale) climate patterns such as ENSO, this project is deliberately
focused on ecological processes such as gross primary productivity, evapotranspiration,
water-use efficiency and light-use efficiency. Stenseth et al. 2002; 2005 suggested that
biological effects to climate may be more strongly correlated to global indices than to any
single local climate variable.
Figure 2.2: The geographical distribution of the Australian savanna biome as defined by Fox et al. (2001).
Located north of the Tropic of Capricorn, it also includes the NATT and observation sites (Source: Hutley et al.
2011)
Chapter 2: Literature Review
21
Savanna physiology
Physiological characteristics of savanna vegetation vary greatly within the context Australian
savannas due to the multiple plant functional types found in the tree-shrub-grass layers (Ma et
al. 2013). Scholes & Archer (1997) also identify the significantly variable leaf area index
(LAI) and tree-grass ratios among different savanna classes which contribute to physiological
variability. The multi-canopy strata have differing phenological and functional responses and
interactions to seasonal rainfall, temperature, vapour pressure deficits (VPD) and other
environmental variables (Scholes & Archer 1997; Whitley et al. 2013).
The biophysical and biogeochemical processes are altered significantly as a consequence of
fire events within the Australian savanna biome, though its current structure, composition and
distribution have coevolved with fire (Beringer et al. 2014). Savannas with a well-developed
grassy understory are at risk of fire during the dry season, given the high fuel load content
combined with hot, dry and often windy conditions from April to September. Fire events are
known to immediately affect the radiative balance through understory consumption and
blackening of the surface and leaf death following scorching (Beringer et al. 2003). The
intensity of the fire causes variable effects; low intensity fires cause little canopy damage and
surface energy balance alteration with a slight increase in Bowen ratio (ratio of latent to
sensible heat partitioning); moderate fires result in comprehensive damage in the following
weeks after the event such as canopy scorch and almost complete leaf drop. Thermal energy
partitioning is driven by land surface characteristics, such that following a fire, the shutdown
of most leaves within the canopy reduces transpiration and shifts net carbon exchange from a
sink to carbon source (to the atmosphere) through elevated respiration and low levels of
canopy photosynthesis (Beringer et al. 2003).
Savannas typically occur on a range of soil types given their biogeographic range, such as
oxisols, ultisols, entisols and asfisols (Hutley & Setterfield 2008). These soils are generally
highly weathered and low in organic matter, tending toward types of sands to sandy loams.
Deep and well-drained, they lack the capacity to hold moisture for long periods of time and
are considered nutrient-limited. Thus the interactions between C3 trees and C4 grasses in the
savanna context tend to be dominated by competition for water and nutrients, rather than light
or growing space (Hutley & Setterfield 2008).
Chapter 2: Literature Review
22
Savanna phenology
Phenology is the analysis of the life cycles of flora and fauna as influenced by environmental
drivers such as climate and other seasonal influences (Ma et al. 2013). Landscape phenology
is a key parameter in climate and biogeochemical cycle models and despite their coverage of
approximately one eighth of the global land surface and sensitivity to climate change
(Beringer et al. 2012; Ma et al. 2013); little is known about the phenological triggers of
savanna vegetation (Prior & Bowman 2005). Much of the phenological research has been
carried out in northern mid-to-high latitude temperate regions (e.g. Chen et al. 2002; Zhang et
al. 2004), where the timing and duration of growing season is tightly coupled with
temperature. Within the savanna biome, phenological responses to seasonal climate are
complicated due to the heterogeneous assemblage of multiple tree, shrub and grass vegetation
strata, all of which vary independently to environmental factors (Ma et al. 2013). Within the
northern Australian savanna, key phenology transition dates such as the start, peak, end and
length of seasonal greening periods are subjugated by the Inter-Tropical Convergence Zone
(ITCZ) monsoon climate system, characterised by biogeographical patterns.
At a community scale, savanna phenology has been investigated comprehensively near
Darwin in the Northern Territory. Ground observations demonstrated that evergreen species
continue to flush and produce leaves throughout the dry season whilst brevi- and semi-
deciduous species commenced leafing just prior to the arrival of the rainy season (October
through March) (Williams et al. 1997). Distinct from these patterns, fully deciduous species
exhibit a variety of behaviours, flushing before and after rainfall (Williams et al. 1997).
Eucalyptus-dominated savannas observe an earlier greening period of enhanced vegetation
activity than their southern Acacia counterparts, resulting in a relatively stable length of
greening periods (Ma et al. 2013). This is due in part to the extreme rainfall variability
experienced in the north (strong wet-dry seasonality), but also due to the C3/ C4 grass-forb
understory which provides potential for extended greening periods. In contrast, their
responsiveness to hydro-climatic variability is significantly reduced. These phenological
characteristics are suggestive of a strong coupling to rainfall events. Predictions of increased
inter-annual variability and extreme climate events are likely to impact on savanna phenology
(Easterling et al. 2000; Seager et al. 2013).
Characterisation of biogeographical patterns in the phenological cycles of savanna is lacking,
despite their global importance and vulnerability to climate change, especially at landscape
Chapter 2: Literature Review
23
and regional scales (Ma et al. 2013). These limitations means effective understandings of
future climate impacts on functionality, composition and fire behaviour are absent in the face
of mounting concern and evidence over the savanna biome’s sensitivity to climate change
(Scholes & Archer 1997; Scholes & Parsons 1997; Bowman et al. 2010; Bowman et al. 2010;
Hill & Hanan 2013).
Across the globe, the phenology of the savanna biome has been investigated at regional
scales using remote sensing (Archibald & Scholes 2007; Ferreira & Huete 2013; Higgens et
al. 2011; Ratana et al. 2005), investigating the range of characteristics from Africa to South
America. However, many papers such as Zhang et al. (2003) have identified challenges
within the scope of remote sensing for its inaccuracy to calculate phenology metrics. Process-
based models have sought to work in conjunction with remote sensing capabilities to enhance
understandings of the land surface and feedbacks with the atmosphere.
2.4 Land surface models
The land surface is a key component of climate models as it controls the partitioning of
available energy and water into sensible and latent heat and evaporation and runoff
respectively. Many papers published in the 1990’s addressed the issue of weather and climate
being influenced by the state of the land surface such as Avissar & Verstraete (1990), Betts et
al. (1996) and Pielke et al. (1998). Sellers et al. (1992) and Pielke et al. (1998) identified
strong evidence for this at timescales ranging from seconds to millions of years. The role of
the land surface varies between climate and climate models and micrometeorology and
mesoscale meteorology due to spatial differences i.e. the climate as modelled through climate
models is not defragmented by small-scale processes over short-temporal scales (Avissar &
Pielke 1991; Pitman 2003). Modelling the climate is a mature discipline driven by the need
for valid predictions in the wake of recognition that human activities are strongly modifying
atmospheric conditions.
The development of land surface models (LSM’s) came from a need to satisfy gaps in
numerical weather prediction and climate modelling by feeding information about land
exchange processes such as the transport of momentum, thermal energy, carbon and water
(Kowalczyk et al. 2013). The first LSM was applied by Manabe (1969) into a climate model
that intentionally included a simple and idealised distribution of ocean and continent and
ignored the diurnal and seasonal cycles (Pitman 2003). Despite these simplifications, which
Manabe defended on the basis that he was exploring climate within a simplified model, this
Chapter 2: Literature Review
24
was a key first step to identifying the importance of land surface processes within the
framework of climate modelling.
Legg & Long (1975) and Thom (1972) idealised improvements to the basic LSM by
introducing the representation of heat and moisture transfer from the canopy which was built
heavily on the philosophy of Monteith and Szeicz (1962) for evaporation. Second generation
models included another fundamental step, first described in 1978 by Deardorff who
presented a method for simulating parameters of soil, such as temperature and moisture in
two layers, and vegetation as a single layer. Deardorff proposed that the single vegetative
layer protected a fraction of ground from solar radiation. LSM development provided a key
opportunity for a multi-disciplinary approach since processes were treated explicitly and
mathematically, allowing a new generation of micrometeorologists to contribute
significantly.
Third generation LSM’s identified the major advance and limitation in second-generation
LSM’s which modelled canopy conductance empirically. They accounted for plant and
environmental conditions but used this conductance to only model transpiration. A
supplementary explicit canopy conductance was identified in the 1980’s as a means to
improve the simulation of the evapotranspiration pathway and also address the issue of
carbon uptake by vegetation (Pitman 2003), a major revolution in modelling capabilities. The
support of the plant physiology community was integral to the addition of carbon to LSM’s.
It had been established that leaf assimilation of carbon was limited by the efficiency of
photosynthetic activity, the amount of photosynthetically-active radiation (PAR) and the
capacity of the leaf to utilise these products. Major fieldwork campaigns such as the HAPEX-
MOBILHY (Andre et al. 1988), FIFE (Sellers et al. 1992a), HARPEX-Sahel (Goutourbe et
al. 1997) and ABRACOS (Gash et al. 1996) brought together the plant physiologist,
micrometeorologist and hydrologist communities and led to the integration of photosynthesis
capabilities into LSM’s.
Over the previous decade, land surface models have evolved to improve the simulation of
canopy processes, plant physiology, and most importantly, the uptake of carbon at the land’s
surface. These improvements have led to some LSM’s having a complete and interactive
terrestrial carbon cycle with the incorporation of biogeochemical components and vegetation
dynamics (Pitman 2003; Kowalczyk et al. 2013). A schematic of the stages of increasing
development into LSM’s to its current model is shown in Figure 2.3.
Chapter 2: Literature Review
25
Calculating surface fluxes from given meteorological forcings, LSM’s can operate as ‘stand-
alone’ models with applications at single sites (as demonstrated in this thesis) as well as over
continental-scale and global regions. Comparisons of LSM’s have occurred in stand-alone
mode (e.g. Abramowitz et al. 2008) and establishment of benchmarking schemes are
continuing (Abramowitz 2012; Kumar et al. 2012; Luo et al. 2012). Despite these
developments, comparisons of LSM’s within single atmospheric models are less common due
to the coupling work involved, complicating assessments (Kowalczyk et al. 2013). Projects
such as the Land Information System which provides software frameworks for surface
modelling and data assimilation are addressing these issues.
The sophistication of land surface models has seen marked improvements to our
understandings of surface behaviour and exchanges with the atmosphere, satisfying previous
gaps in climate modelling.
Figure 2.3: Schematic of increasing levels of detail being added into surface modelling approaches. (Source: Pitman
2003)
Chapter 2: Literature Review
26
2.5 Summary
Given the highly seasonal dynamics of the tropical climate found in northern Australia,
changing seasonal duration has the potential to alter the current surface energy, water and
carbon balance through changes in the local meteorology driven by variations to meso-
climate circulations. This study aims to improve understandings of changing carbon and
water fluxes in response to climate by focusing on the temporal variability of response to
drivers at a century scale. This approach enables correlation between climate drivers and
biological responses at diurnal, seasonal, inter-annual and decadal scales. Savanna is the
dominant biome of North Australia with significant economic, biodiversity and cultural
heritage assets (Woinarski et al. 2007), but whose sensitivity to climate change is not yet
fully understood.
Improvements to the wavelet analysis over the past two decades have seen its validation in
statistical and quantitative studies, but only one study has been published that used this
approach. Hong & Kim (2011) analysed the influence of the monsoon and associated
environmental drivers (temperature, rainfall, vapour pressure deficit) on ecosystem processes
gross primary productivity (GPP), ecosystem respiration (Fre), net ecosystem exchange (Fc)
and latent heat flux (Fe). The wavelet analysis demonstrated the seasonal influence of the
monsoon on ecosystem function and showed the coupling present between environmental
drivers and ecosystem exchanges of carbon and water.
ENSO and SOI have been the focus of literature relating to their influence on temperature
and rainfall regimes in Australia. However, few studies have used the TSI and II to examine
temperature and rainfall patterns in Australia (Verdon & Franks, 2005; Murphy & Timbal,
2008), with no studies using these indices in wavelet analyses.
Chapter 3: Approach and Methodology
27
Chapter 3: Approach and Methodology
This thesis brings together two sources of data to train and drive an artificial neural network
(ANN) to hind-cast 100 years (1900-2000) of flux responses at the Howard Springs (HWS)
site (Figure 3.1). These hind-cast responses were analysed as a time series, along with the
driving data to examine the variable change through the 20th
century.
Twelve years of eddy covariance measurements collected at the HWS flux tower site were
used to train the ANN. This site is the longest-running data set representing coastal savanna
in tropical north Australia. Its extensive data set makes it a superior site for training the
neural network to detect seasonality, responses to climate extremes and fire disturbances, a
critical feature of savanna environments.
Meteorological data spanning the 20th
century from the Australian Water Availability Project
(AWAP) drove the ANN outputs of flux responses. These measurements are a combination
of observations and modelled predictions. Given the site’s proximity to the Bureau of
Meteorology’s class 1 Darwin site, the driving meteorology is highly accurate.
Tower derived
flux data
Meteorological
data
Artificial neural
network model
Flux response
outputs
Figure 3.1: General approach flow chart. The neural network is central to this modelling exercise, bringing
together two sources of information (tower derived flux data, GPP, Fc, Fre, Fe) and gridded meteorology
(AWAP data) to generate flux response variables hindcast from 1900-2013 on which the study is centred.
Chapter 3: Approach and Methodology
28
Variables of interest
In order to examine how climate and vegetation are coupled, ecosystem exchanges of carbon
and water are fluxes of interest, along with ecosystem physiological variables of water-use
efficiency (WUE) and light-use efficiency (LUE) (Table 3.1). Carbon exchanges are summed
as net ecosystem exchange (Fc), a net representation of carbon uptake through gross primary
productivity (GPP) and carbon loss as ecosystem respiration (Fre). Water exchange by the
ecosystem is defined as latent heat flux (Fe). These exchanges are driven by environmental
conditions such as temperature (Ta), rainfall and vapour pressure deficit (VPD) which in turn
are controlled by modes of climate variability (Table 3.2). In the tropics these modes may be
quantified using climate indices such as the Southern Oscillation Index (SOI), Tasman Sea
Index (TSI) and Indonesian Index (II).
Table 3.1: Meteorological and flux response variables used for time series and wavelet coherence analysis.
Variable Units Comments
FLUX RESPONSE VARIABLES
Gross Primary Productivity
(GPP)
t C ha-1
y-1
(unless otherwise
specified)
Total amount of carbon
materials created from
carbon sources such as
carbon dioxide
Ecosystem Respiration (Fre) t C ha-1
y-1
(unless otherwise
specified)
Release of carbon dioxide by
ecosystem to atmosphere
after oxidation process
Net Ecosystem Exchange
(Fc)
t C ha-1
y-1
(unless otherwise
specified)
Measures how much carbon
is entering and leaving the
ecosystem
Latent Heat Flux (Fe) W m-2
Energy release by ecosystem
during constant-temperature
process (change state of
matter known as
evapotranspiration)
Water Use Efficiency (WUE) GPP/(Fe*0.49) = umol C/
umol H2O
Ratio of carbon uptake per
water unit loss
Light Use Efficiency (LUE) GPP/ (shortwave
radiation*0.49) = umol/umol
Ratio of carbon uptake per
unit of light (radiation)
Chapter 3: Approach and Methodology
29
METEOROLOGICAL DRIVER VARIABLES
Temperature (Ta) oC Ambient temperature
Rainfall mm Rainfall at HWS grid cell
Vapour Pressure Deficit
(VPD)
kPa Difference between actual
water vapour pressure and
saturation water vapour
pressure at x temperature
Radiation (Rn) W m-2
Net incoming solar radiation
These variables quantify ecosystem function through carbon and water exchanges and
ecological coupling through water and light use efficiencies.
Table 3.2: Climate indices of interest in this study. , chosen for their influence on Australian rainfall and
climate.
Name Climate
Phenomenon
Description Time Scale
Southern Oscillation
Index (SOI)
ENSO Mean sea level
pressure difference
between Darwin and
Tahiti
Inter-annual to
decadal
Tasman Sea Index
(TSI)
Extra-tropical
Climate
Sea surface
temperature anomaly
(30⁰S-40⁰S, 150⁰E-
160⁰E)
Inter-annual to
decadal
Indonesian Index (II) Indonesia SST Sea surface
temperature anomaly
(0⁰-10⁰S, 120⁰E-
130⁰E)
Inter-annual to
decadal
The climate indices reflect climate phenomena that adjust sea surface temperatures (SST’s)
and drive Australian meteorology, especially rainfall. Their correlation with ecosystem flux
variables will provide further understandings of the degree of coupling present between
climate and vegetation processes in tropical north Australia. Indices used in this study
Chapter 3: Approach and Methodology
30
evolved from their influence on Australian rainfall. Appendix A demonstrates that rainfall at
Howard Springs correlates with the Tasman Sea Index, Indonesian Index and Southern
Oscillation Index
Tools of analysis
Linear trends demonstrate the change with time that a variable experiences. By analysing the
time series of each flux response and meteorological driver, trends and variability due to
climate change can be observed, beyond the seasonal cycle and over century time scales.
Correlating trends suggest tight coupling between environmental conditions and ecosystem
function, describing the temporal relationship and driver-response mechanism.
Power spectra density graphs detect the inherent cycling within a singular variable and its
changes in periodicity with time. This analysis is integral to the study and understandings of
common periodicities between two variables, which will come from the wavelet transform
coherence graphs (WTC). The WTC establishes the degree of coupling between the flux
variable and meteorological driver or climate index, analysing the strength of influence of
modes of climate variability on ecosystem function. A defining feature of the WTC is its
ability to determine lags through phase angles, demonstrating the temporal response of
ecosystem to environment.
Howard Springs is an ideal site, representative of coastal tropical north Australian savanna
that experiences influences by modes of climate variability. It is one of the longest flux sites
in the world with twelve years of observations, valid as a training tool for modelling past flux
responses.
3.1 Site description
The Howard Springs Flux Station (12⁰29.712’ S, 131⁰09.003’ E), located adjacent to the
Black Jungle Conservation Reserve in the Northern Territory (Beringer et al. 2007; Hutley &
Beringer 2011) has been operating since 2001 and is one of the longest flux records in
Australia. The instrument mast is 23m in height and using methods of open-path eddy
covariance, it measures heat, water vapour and CO2 fluxes above the 18m canopy (Hutley et
al. 2011). Meteorological parameters include temperature, humidity, wind speed/ direction,
precipitation, incoming and outgoing shortwave radiation and net radiation are also measured
in this region. Howard Springs receives approximately 1690mm annually, 95% of which falls
during the wet season (October to April). Maximum temperatures range seasonally by three
degrees Celsius, from 33.2⁰C in summer to 30.4⁰C in winter. Minimum temperatures range
Chapter 3: Approach and Methodology
31
slightly larger at 25.4⁰C in summer and 19.3⁰C in winter (TERN, 2013). Given its location
near the coast, temperatures are modulated by oceanic heat capacity, which also brings about
the presence of a sea breeze.
Average tree height is fourteen to sixteen metres, classified as open woodland savanna
dominated by Eucalyptus miniata and Eucalyptus tetrodonta overstory. The extreme tropical
location and high rainfall of Howard Springs results in a savanna characterised by a woody
overstory of evergreen Eucalyptus and Corymbia species and perennial C4 grasses and other
forb and shrub species making up the understory (Egan & Williams 1996; Hutley et al. 2011).
The location, geology and soil setting and domination of two species has allowed the site to
be classified into a Broad Vegetation Group (BVG) of Monsoon woodlands dominated by
Eucalyptus tectifica and Corymbia species, or BVG 4 (Fox et al. 2001). The highly-
weathered soil is low is nutrient and water-holding capacity, resulting in competition between
the C3 trees and C4 grasses for water and nutrients, rather than light and space. The
functional response of the C4 grass layer present in the northern Australian savanna is
primarily associated with the strongly seasonal presence of the monsoon system.
Unlike South America and Africa, Australia’s savannas are exhibit low levels of
fragmentation and are ecologically intact (Beringer et al. 2011; Woinarski et al. 2007)
providing us with greater capacity to understand the environmental controls of their growth
and functionality. This ‘living laboratory’ creates an ideal investigation of savanna phenology
and biogeographical variations (Hutley et al. 2011; Williams et al. 1996). Many studies have
utilised this region for investigations in biogeography, soils, ecophysiology, meteorology and
land-atmosphere processes (see review by Hutley et al. 2011).
3.2 Data processing methods
Now and hereafter, annual records of climate, meteorology or flux variables are organised as
July-June annual values, in order to capture the whole seasonal distribution of rainfall (a
significant ecosystem driver) within one time-step, as experienced in tropical Australia.
Initial data processing and sorting proved impossible with almost two and a half billion
values for 374 variables at 30 minute time steps from 1st January 1900 to 2014 (Beringer et
al., unpublished data). This was overcome by reducing temporal resolution to daily averages
before capturing the variables of interest (meteorological drivers, tower observations and
ANN outputs). Arbitrary classifications for wet and dry months were made in order to
compare seasonal patterns of influence, drives and lags. They are: May, June, July, August,
Chapter 3: Approach and Methodology
32
September and October (Dry); and November, December, January, February, March and
April (Wet).
We have assumed that the climate phenomena (i.e. ENSO, extra-tropical climate, SST’s in
Indonesian region) are the drivers of change in environmental conditions which in turn
influence these variables. GPP, Fre and Fc are carbon exchanges which have links to the
thermal regime while Fe is a constituent of the water cycle and as such is influenced by
moisture conditions in the environment, among other drivers of radiation (Rn) and vapour
pressure deficits (VPD). When analysing water and light use efficiencies (WUE and LUE
respectively), this could only be done for the tower data period (2001-2013). This is because
WUE and LUE are a measure of ecological coupling whereas GPP, Fre, Fc and Fe are
coupled physiologically. Thus, the ANN predicts GPP and Fe statistically and doesn’t know
about the coupling.
Time series trends
Time series were created to demonstrate the various trends and cycles of the meteorological
drivers and their biological counterparts, which respond to different drivers, such as
atmospheric inputs of solar radiation (evapotranspiration), precipitation (gross primary
productivity, ecosystem respiration) and air temperature, soil temperature and moisture also
have a significant influence on many flux responses (Pielke et al. 1998). Many responses are
driven by a combination of meteorological forcings, which will be demonstrated through
analysis of climate indices such as SOI, TSI and II, which influence precipitation and water
availability and radiation through various processes. Given the tropical scope of this project,
understanding the climate variability is of particular importance, due to the sensitivity of
tropical regions to climate change and the alterations that are expected to occur over the next
century with regards to temperature and precipitation extremes.
Anomalies of the time series were also plotted over the 100 year period (1913-2013) looking
at the annual variation. This was done to confirm modelled data suitability in further
correlative analyses with climate indices.
Power spectrum densities and Wavelet transform coherence
Wavelets provide time-frequency localisation, done by integrating the whole domain of time t
from -∞ to ∞, to find the frequency behaviour over a particular time period (Nason &
Silverman 1995), as is observed through the following steps of wavelet analysis.
Chapter 3: Approach and Methodology
33
The power spectrum density graphs detect inherent cycling within individual variables and
how these change with time (eq. 1). Variations throughout the 20th
century see changes in
periodicity length and strength of cycling. The wavelet transform coherence similarly detects
cyclic behaviour in the time series, however the algorithms (eq. 2-4) are fed two variables
and the common periodicities are found, along with their strength in coupling and the phase
angles, demonstrating the lagged effect by the response to the driver.
Initial analysis began with power spectrum density graphs displaying the global power
spectrum of meteorological drivers and ecosystem responses, using the Morlet wavelet as a
basis function in this study, following successful use in the same research context by Hong &
Kim (2011), defining the wavelet power spectrum (S) of Xn as
Sn (s) = |Wxn(s)|
2 eq. 1
Where s is the wavelet scale and Wn is a complex number with real and imaginary parts after
transform of Xn. The continuous wavelet transform (CWT) of Xn is calculated as (Torrence &
Compo 1998)
𝑊𝑛(𝑠) = ∑ �̂�𝑘ψ̂∗(𝑠𝜔𝑘)𝑒𝑖𝜔𝑘𝑛𝛿𝑡𝑁−1
𝑘=0 eq. 2
k is the scale index and N is the number of time series. x̂k is the Fourier transform of Xn, ψ̂∗
is the Fourier transform of the conjugate of the wavelet function; s is the scale; ωk is angular
frequency, δt is the time step, as documented in Collineau & Brunet (1994); Grinsted et al.
(2004) and Hong & Kim (2011). The cross wavelet transform the two time series Xn and Yn
are further analysed in preparation for the final wavelet coherence (WTC) spectrum, at scale
s, defined as
WnXY
(s) = WnX (s)Wn
Y* (s) eq. 3
Where (*) is the complex conjugate (a binomial formed by negating the second term of the
binomial) of wavelet coefficient WnY and the wavelet coherence becomes
𝑅𝑛2(s) =
|⟨𝑠−1𝑊𝑛𝑋𝑌(𝑠)⟩|
2
⟨𝑠−1|𝑊𝑛𝑋(𝑠)|
2⟩⟨𝑠−1|𝑊𝑛
𝑌(𝑠)|2
⟩ eq. 4
Where smoothing occurs in both scale and time and the factor s-1
is used to convert energy to
power spectral density using definitions0 ≤ 𝑅𝑛2(s) ≤ 1 (Hong & Kim 2011).
Chapter 3: Approach and Methodology
34
In the wavelet power spectra graphs and coherence plots, the x axis is time in years, from
1900-2013 and y axis is time scale and frequency ranging from six months to ten years.
The colour intensity exhibits the spectral power with dark red indicative of the highest
spectrum. The global power spectrum graphs specify the significant time-frequency variation
for each variable; where the x axis is time and y axis is spectrum.
Software package
Statistica (StatSoft Inc. Ver. 12) was used in conjunction with MATLAB (MathWorks Inc.
R2014a) to organise, summarise and conduct wavelet analyses. The wavelet analysis module
is a component of the WTC MATLAB program (Grinsted 2004) (MATLAB codes can be
downloaded from the website http://noc.ac.uk/using-science/crosswavelet-wavelet-
coherence) while the wavelet program was written by Christopher Torrence and Gilbert
Compo (Torrence & Compo 1998).
3.3 Data Sources and modelling approach
Tower observations
Twelve years of tower observations are used to train the ANN at the study site. Tower
observations at Howard were collected and calculated through eddy covariance, following an
open-path method, using the LI-7500A and LI-7700 (CH4) gas analysers (Burba & Anderson
2005). A full list of measurements and their descriptions can be found in Appendix B. Final
data output consists of an estimated ten per-cent error of which a number of issues can cause.
In this tropical region, the errors most likely result from spatial separation between sonic and
open-path analyser and precipitation events interfering with sensors (Burba & Anderson
2005).
AWAP
Feeding the ANN is AWAP meteorological data, prescribing the climatic conditions for the
site from 1900- present. The local meteorology used by the tower would usually be best here;
however it only goes back ten years.
The Australian Water Availability Project (AWAP) was created by CSIRO as a 0.05⁰ x 0.05⁰
gridded dataset to monitor the state and trend of the terrestrial water balance over the
Australian continent, using model-data fusion methods combining both observational data
and modelling (CSIRO 2014). However, observations haven’t existed in many parts of the
country since its inception (1900) and have had to use modelled data, creating a climatology,
Chapter 3: Approach and Methodology
35
or repeated trend for the period in which no data was taken. We acknowledge radiation as a
significant driver of ecosystem function however, given that records didn’t exist before 1990,
it would be futile to discuss the results of radiation as a driver in a century-long analysis.
Climate Indices data
Climate data was downloaded via online sources, at a monthly resolution provided by the
Centre for Australian Weather and Climate Research (CAWCR) and National Oceanic and
Atmospheric Administration (NOAA).
The Southern Oscillation Index (SOI; Troup 1965) is calculated as a standardised anomaly of
values of mean sea level pressure (MSLP) differences between Darwin and Tahiti from 1933-
1992. Sources use the Troup method as shown here:
SOI = 10(𝑃𝑑𝑖𝑓𝑓−𝑃𝑑𝑖𝑓𝑓𝑎𝑣
𝑆𝐷(𝑃𝑑𝑖𝑓𝑓)) eq. 5
Where Pdiff is the average Tahiti MSLP for the month minus average Darwin MSLP for the
month, Pdiffav is the long term average of Pdiff for the month in question and SD(Pdiff) is
the long-term standard deviation of Pdiff for the month in question (BoM 2014). The SOI is
related to the atmospheric response associated with ENSO; the Centre for Australian Weather
and Climate Research (CAWCR) believes that this may prove to be a better predictor for
Australian rainfall than indices of sea surface temperature, though the spatial variability of
this has not been explicitly examined.
The TSI is a measure of SST anomalies off the east coast of southern Australia (Table 3.2).
Its use as a quantifier of extra-tropical climate has only emerged recently (e.g. Murphy &
Timbal 2008; Schepen et al. 2012). The data used in the TSI analysis was extracted from the
Extended Reconstruction Sea Surface Temperature version 3b (ERSST.v3b) via the National
Oceanic and Atmospheric Administration (NOAA) – National Operational Model Archive
and Distribution System (NOMADS) Live Access Server (LAS)
(http://nomads.ncdc.noaa.gov/las/getUI.do).
The Indonesia Index characterises anomalies of sea surface temperatures in the Indonesian
region (Table 3.2). Australian winter rainfall was shown to display a high correlation to a
function of II when it was subtracted from SST’s over the central Indian Ocean (Nicholls
1989). This relationship was furthered by Verdon & Franks (2005) and Schepen et al. (2012)
Chapter 3: Approach and Methodology
36
used lagged II to demonstrate the correlation between II and rainfall over the Northern
Territory. II monthly anomalies were also extracted from NOAA’s distribution system.
By initiating a multi-disciplinary approach (eddy covariance flux measurements and
modelled data) with a multi-scaled methodology, a method advocated by Pittman (2003), it is
hoped that this study and series of experiments will minimise some of the knowledge gaps
that remain in understanding the processes that control the temporal variability of fluxes.
Modelling
A statistical-based model, known as an artificial neural network (ANN) was trained using the
flux observations from the Howard Springs Tower to hind-cast ecosystem flux responses
from 1900 onwards. Driving the ANN were meteorological variables, a combination of
observations and modelled predictions from AWAP. The ANN output was a 113 year trend
of ecosystem fluxes such as GPP, Fre, Fc and Fe.
This statistical approach has been used in space of a process-based model, given its ability to
accurately capture ecosystem responses, such as Fe (Figure 3.2). The diurnal dynamics of the
savanna water flux is represented in the ANN output, validating its use in this study.
ANN’s are highly empirical statistical models, but mechanistic constraints have been
imposed in this study, via selected meteorological drivers. Fractional photosynthetically
Figure 3.2: Diurnal latent heat flux measured at 30 minute intervals and averaged over the 2001-2013
period for observations (blue) and ANN (green). X axis is hours from midnight and y is Fe flux in W
m-2
.
Chapter 3: Approach and Methodology
37
active radiation (fPAR) has been used as a driver, which is significant as it allows the ANN to
capture structural components of the ecosystem such as leaf are index (LAI) permitting the
ANN to capture seasonal variability. The training data (tower observations) have enough
incidences of meteorological extremes, allowing the ANN to provide robust predictions.
Each time series analysis (linear trends, power spectral densities and wavelet transform
coherence) has been used to answer the research questions regarding savanna response
variability, cyclic patterns, and degree of coupling and lagged phase angles. The variables
used to characterise ecosystem function capture the exchange of carbon and water with the
atmosphere over the 20th
century, as hind-cast by the neural network.
Chapter 4: Results
38
Chapter 4: Results
Understanding and predicting the nature of ecosystem flux responses to climate change
begins with a basic trend line analysis of the meteorological and ecosystem variables, as
detailed in Table 3.1. The trend lines demonstrate the change experienced at Howard Springs
during the 20th
century, identifying similarities and differences between the drivers and
responses, as expected per the literature. Variability is also examined determining the
dynamics of change over time. A spectral analysis is then conducted on all variables,
including the climate indices (Table 3.2) demonstrating the inherent cycling present in
individual variables. These results contribute to our interpretation of the wavelet coherence
plots. These plots establish the temporal relationship and degree of coupling between
meteorological drivers and climate indices to flux responses, by demonstrating the coherence
and lagged effects between driver and response at common periodicities.
4.1 Time Series Analysis
In order to understand the temporal dynamics at Howard Springs, time series of each
meteorological and flux variable are shown (Figure 4.1, 4.2 and 4.3). Meteorological
variables were chosen for their known influence on ecosystem exchanges while flux variables
were chosen as representatives of carbon and water exchanges. It is useful to illustrate these
variable trends over time as it provides insight to how the environment at Howard Springs, a
representative of tropical north Australian mesic savanna ecosystems, has changed over time.
The linear trend analysis was conducted over the 1913-2013 time period, capturing 100 years
of mean annual change and variability. The first thirteen years was excluded from this
analysis due to the uncertain validity of AWAP inputs. All variables show an increasing trend
over the last 100 years but vary in rate of increase. Generally, the first half of the century
observed little to no change in slope while the second half of the century, particularly post
1970’s saw intensified warming and wetting, and increased rates of carbon and water
exchange.
Since tropical north Australia is dominated by the Australian monsoon system, annual
variability is expected as the wet season from year to year displays variable rainfall volumes
and distributions, thus confirming the variability seen in the meteorological time series.
Chapter 4: Results
39
AWAP meteorological inputs
The linear trends indicate that all meteorological model inputs have increased over the
previous century at various rates. Temperature at the Howard Springs site has increased on
average by almost one degree Celsius (0.92⁰C) over the 100 year period whilst rainfall has
increased by 300mm over the previous century. Inter-annual rainfall variability appears to
have increased during the second half of the 20th
century; however, no significant trend was
found in the coefficient of variability plot (result not shown). There appears to be an artefact
in VPD in the early 1940’s, at around the same period as the Darwin meteorological station
y = 0.0056x + 1.8391
1.5
1.7
1.9
2.1
2.3
2.5
191
3
191
6
191
9
192
2
192
5
192
8
193
1
193
4
193
7
194
0
194
3
194
6
194
9
195
2
195
5
195
8
196
1
196
4
196
7
197
0
197
3
197
6
197
9
198
2
198
5
198
8
199
1
199
4
199
7
200
0
200
3
200
6
200
9
201
2
VP
D (
kP
a)
c)
Figure 4.1: Time series of three environmental drivers of ecosystem fluxes at the study site (a) temperature (Ta),
(b) rainfall and (c) vapour pressure deficit (VPD). Temperature and rainfall are AWAP inputs (observations
from BoM’s Darwin station) and VPD is a composite of temperature and humidity, demonstrating the
evaporative power of the atmosphere.
Chapter 4: Results
40
was moved from the post-office (CBD) to the airport, less than 10 km away. The increase in
VPD is probably a response to temperature, but is also varies with changes in absolute
moisture content, a variable that is not analysed in this study.
Atmospheric CO2 concentrations and radiation are significant drivers of ecosystem flux
however, their trends are not shown here for separate reasons. Firstly, CO2 concentrations are
not displayed as the trend is clearly established in the literature and secondly, radiation was
not measured until 1990 resulting in most of the 20th
century representing a climatology
(averaged 1990-2010 conditions). The trend has low variability so its use as a driver in the
wavelet plots is not suitable or valid, as the algorithms cannot handle the extreme change in
variability between averaged and observed periods.
Biological responses
Following their meteorological drivers, time series of the flux responses of carbon and water
have uniformly increased over the previous century, mostly due to increased ambient
temperature and rainfall.
ANN outputs of ecosystem flux responses (Figure 4.2 and 4.3) demonstrate change in both
carbon and water exchanges. Negative carbon fluxes indicate a loss of carbon by the
atmosphere to the ecosystem, i.e. a carbon sink. Gross primary productivity (GPP) and
ecosystem respiration (Fre) have both increased at Howard Springs, by over one tonne per
hectare and half a tonne per hectare over the previous 100 years respectively. The rate of
increase in net ecosystem exchange (Fc) is 0.01 t C ha-1
y-1
over the last century, with rates
intensifying after the mid-1960’s. This is a reflection of high rates of increase in GPP and less
increase in corresponding Fre. The net result is enhanced carbon sink shown as an increasing
trend of Fc. Latent heat (Fe) has also increased during the 20th
century, by an average 1 W m-
2 each year.
Chapter 4: Results
41
Figure 4.2: Linear trend of three carbon fluxes of interest. (a) Net ecosystem exchange (Fc), (b) gross primary
productivity (GPP) and (c) ecosystem respiration (Fre). Plots are ANN outputs over the 100 year period of
interest 1913-2013. For Fc, GPP and Fre -ve values indicate the atmospheric loss of carbon to the ecosystem, i.e.
a sink.
GPP has increased noticeably over the previous 100 years, possibly driven by increased
rainfall. The ANN has suggested an increased carbon uptake of almost one and a half tonnes
per year over the previous century in a consistently sinking trend. Respiration has a less
distinctive increase (less than 0.4 t C ha-1
y-1
), reflective of significant rainfall and
temperature patterns, which are driving variables of ecosystem respiration.
The net sink has increased by about 1 tonne per hectare per year, from -5.5 t ha-1 y-1 to -6.5 t
ha-1 y-1over the previous century, which is at odds with observations. Beringer et al. (2004)
found Fc to average 4 tonnes per hectare per year for the 2001-2006 period, significantly
Chapter 4: Results
42
lower than the ANN-predicted Fc. The increase in carbon sink (GPP) means that
photosynthesis has also increased over the previous century and as a result of the warming
and wetting observed in the temperature and rainfall trends, transpiration has also risen as a
by-product of increased growth and biomass. A slope change analysis has been done for the
carbon flux variables, to examine the dependence of these to environmental conditions. These
results are shown in Appendix C.
Between the two arbitrary periods (1913-1949 and 1950-2013), increases were observed in
all variables except 1913-1950 respiration, which had a decrease of about 0.007 t C ha-1
y-1
,
which is negligible, as was early century GPP increase with a slope of just 0.002 t C ha-1
y-1
.
The largest increase came in late century net ecosystem exchange (0.015 t C ha-1
y-1
), as a
result of a GPP increase of 0.01 t C ha-1
y-1
and minimal increase in respiration (just 0.0038 t
C ha-1
y-1
).
Figure 4.3: Linear trend of latent heat flux (Fe). Plots are ANN outputs over the 100 year period of interest
1913-2013.
Fe has increased consistently through the 20th
century, but it is beyond the scope of this thesis
to attribute this trend to a specific driver, especially since representation of all potential
drivers is limited in this study (only three meteorological drivers are shown). This is
particularly evident in Fe analysis since increased Fe would suggest a rise in radiation which
would imply reduced cloud cover and rainfall, which is not observed at this site.
y = 0.1042x + 77.3
70
75
80
85
90
95
19
13
19
17
19
21
19
25
19
29
19
33
19
37
19
41
19
45
19
49
19
53
19
57
19
61
19
65
19
69
19
73
19
77
19
81
19
85
19
89
19
93
19
97
20
01
20
05
20
09
20
13
Fe
(W
m-2
)
a)
Chapter 4: Results
43
Figure 4.4: Time series of (a) water use efficiency (WUE) and (b) light use efficiency (LUE) for the tower observation period (2001-2013). GPP is divided Fe *0.49 to
generate a time series for WUE and using the conversion factor of 0.49 (Kanniah et al. 2010) in LUE gives GPP divided by short wave radiation *0.49, WUE and LUE are
shown at a monthly scale, and are reflective of ecological coupling.
-0.18
-0.13
-0.08
12001
72001
12002
72002
12003
72003
12004
72004
12005
72005
12006
72006
12007
72007
12008
72008
12009
72009
12010
72010
12011
72011
12012
72012
12013
WU
E (
um
ol /u
mol )
a)
-0.11
-0.09
-0.07
-0.05
-0.03
-0.01
12001
72001
12002
72002
12003
72003
12004
72004
12005
72005
12006
72006
12007
72007
12008
72008
12009
72009
12010
72010
12011
72011
12012
72012
12013
LU
E (
um
ol / um
ol )
b)
Chapter 4: Results
44
Water use efficiency (WUE) and light use efficiency (LUE) are ratios of carbon uptake (GPP)
and water loss and radiation use respectively. WUE and LUE are examples of ecologically-
coupled variables (Figure 4.4) which show no significant trend over the 13 year observational
period. Unlike WUE, LUE displayed a distinct seasonal pattern, with maximal LUE
occurring during the peak of the wet season (January-February) and minimal LUE occurring
during the late dry season (August-September).
The linear trend analysis displayed increases in all meteorological and ecosystem variables,
suggesting a dynamically-changing environment at Howard Springs. Whether this correlation
is a driver-response mechanism is still unknown.
4.2 Power Spectrum density graphs
The spectral analysis demonstrates the inherent cycling of singular variables which aids our
interpretation of the wavelet transform coherence graphs that follow. The following results
demonstrate the wavelet power spectra and global wavelet spectrum for the meteorological
and biological variables. The spectral analyses of meteorological drivers (Figure 4.5) feature
a distinctive annual periodicity due to the wet/ dry cycle. Ta also shows a clear six-month
periodicity as well as a 5 and 10 year cycle, though activity appears inconsistent across the
century. VPD exhibits similar patterns as temperature in terms of the larger-scale frequencies
however, the second quarter of the century ( ̴1930-1950) appears to show cycles from 2-10
year frequencies.
The power spectrum and wavelet analyses are conducted over the 1900-2013 period; early
century inconsistencies are partially removed from the plots as a result of a wavelet function
which detects where the data may be distorted due to edge effects.
Chapter 4: Results
45
The distinct annual cycle of rainfall is driven by monsoonal processes common of tropical
north Australia dominating annual dynamics, whereas temperature is more continuous. These
patterns of periods of high spectral power that are variable through time is repeated for all
Figure 4.5: Wavelet power spectrum of (a) temperature, (c) rainfall and (e) vapour pressure deficits and corresponding
wavelet spectrum at Howard Springs (b, d and e respectively). The thick black contour in the wavelet power spectrums
and blue dash line in the wavelet spectrums indicate the 5% significance level against red noise. Wavelet spectral
analysis was fed daily data over the 113 year period.
(c) Tair Wavelet Power Spectrum (d) Global Wavelet Spectrum
(a) Rainfall Wavelet Power Spectrum (b) Global Wavelet Spectrum
(e) VPD Wavelet Power Spectrum (f) Global Wavelet Spectrum
Chapter 4: Results
46
variables with periods of high and low power and significance evident. These temporal
dynamics have been summarised in Table 4.1 below for each meteorological and biological
variable, with an interpretation of results.
The biological variables (Figure 4.6) displayed similar annual patterns observed in the
meteorological spectral analysis. Similar patterns are emerging between carbon (Fc) and
water (Fe) variables. By contrast, GPP and Fre displayed less spectral strength and
consistency of larger cycles (2-10 years) across the 113 year period. Spectral analysis of Fc
stood out for its strong decadal signal present across the century as well as inconsistent 2-5
year cycles, which appeared again in the Fe analysis.
Chapter 4: Results
47
Figure 4.6: Wavelet power spectrum of (a) Fc, (c) GPP, (e) Fre and (g) Fe and corresponding wavelet spectrum (b, d, f and
h respectively) at Howard Springs. The thick black contour in the wavelet power spectrums and blue dash line in the
wavelet spectrums indicate the 5% significance level against red noise. Wavelet spectral analysis was fed daily data over
the 113 year period.
(c) GPP Wavelet Power Spectrum (d) Global Wavelet
(a) Fc Wavelet Power Spectrum (b) Global Wavelet
(e) Fre Wavelet Power Spectrum (f) Global Wavelet Spectrum
(g) Fe Wavelet Power Spectrum (h) Global Wavelet Spectrum
Chapter 4: Results
48
A summary of the periodicity length and occurrence of four environmental conditions and
four biological responses (Table 4.1) demonstrates the highly variable temporal dynamics and
cycling of individual variables, which will be assessed further as common periodicities
between the meteorological drivers and ecosystem flux responses.
Table 4.1: Summary table of climate and biological variables following spectral analysis. Observed periodicity
and associated statistically significant time periods are included, calculated as the number of years a periodicity
is observed over the total number of years in the analysis (113). Each variable is ranked by significance fraction,
to demonstrate the variability in periodicity across meteorological and biological variables.
Variable Periodicity (y) Time Periods Significance
fraction
Ta 0.5 1900-2013 0.95
5 1917-1952 0.35
10 1973-1983 0.09
VPD 0.5 1900-2013 (intermittent) 0.08
2-10 1940-1955 0.13
10 1925-1994 0.63
Rainfall 0.5 1900-2013 (intermittent) 0.1
2 1905-1909 0.03
Fc 0.5 1900-2013 (intermittent) 0.44
2-5 1910-1915, 1921-1937,
1938-1951, 1997-2003
0.36
10 1915-2001 0.78
Fe 0.5 1900-2013 (intermittent) 0.12
2-5 1922-1946, 1974-1983 0.3
10 1915-1955, 1965-2000 0.68
GPP 5 1927-1936 0.08
10 1915-1947, 1966-1999 0.59
Fre 10 1940-1975 0.31
Temperature and Fc showed the highest fraction of periodicity across the century in their
respective categories (meteorological and ecosystem fluxes) with over half of the century
observing multiple cycles from six months to ten years. This is expected as Fre is only driven
Chapter 4: Results
49
by Ta in the ANN. The longest periodicity observed in the power spectral analysis is that of
the six month cycle present in temperature, which was consistent throughout 95% of the last
113 years, with some notable exceptions during the early 1930’s and again in the mid to late
1940’s. Such a strong and consistent cycle was not observed in any other variable nor
periodicity. The strongest decadal cycle was observed in Fc with 78% of the last century
having some decadal frequency present in the net ecosystem exchange of carbon, suggesting
a significant cycle present at 10 years. This cycle is reiterated in other biological variables
such as Fe (68%), GPP (59%) and Fre (31%). Of the meteorological variables, only VPD
showed a significant 10 year cycle (occurring in 63% of the last century).
The climate indices were also put through the same analysis however, due to the reduced
number of values (indices are monthly values, Figures 4.5 and 4.6 used daily data), the
analyses are produced in a lower resolution. This could be improved in future studies by
adjusting (reducing) the colour spectrum that the spectral analysis uses. Whilst the above
tabular display of results cannot be recreated for the following results, inferences can be
made about the inter-annual and decadal cycles present in the Southern Ocean Index (SOI),
Tasman Sea Index (TSI) and Indonesian Index (II) gathered from Figure 4.7 below.
Chapter 4: Results
50
Whilst spectral analysis of the climate indices are not highly-resolved, they help to form an
understanding of the inherent cycling within the index and associated climate phenomena,
and how these might drive ecosystem responses at similar periodicities in coherence plots.
Figure 4.7: Wavelet power spectrum of (a) Southern Oscillation Index (SOI), (c) Tasman Sea Index (TSI) and (e)
Indonesian Index (II) and corresponding wavelet spectrum (b, d and f respectively) at Howard Springs. The think black
contour in the wavelet power spectrums and blue dash line in the wavelet spectrums indicate the 5% significance level
against red noise. The very low resolution of these images is most likely due to the reduced number of values (Indies are
measured monthly over the 110 year period).
(a) SOI Wavelet Power Spectrum (b) Global Wavelet Spectrum
(c) TSII Wavelet Power Spectrum (d) Global Wavelet Spectrum
(e) II Wavelet Power Spectrum (f) Global Wavelet Spectrum
Chapter 4: Results
51
The strongest pattern emerges from the SOI spectral analysis at around 5-7 year cycles across
the century, which is reflective of the literature (e.g. AchutaRao & Sperber 2002; Lau & Sheu
1988). In TSI and II, decadal cycles appear to emerge, however this is difficult to confirm
given the low resolution.
Significant 0.5 to 10 year temporal patterns have appeared within the variables of interest
using the power spectral analysis above. Various time frequencies have emerged from both
the meteorological and biological variables but these alone do not confirm the driver-
response mechanism through correlating cycles. A wavelet coherence analysis will allow
correlating patterns to emerge between the drivers (meteorological variables and climate
indices) and biological responses.
4.3 Wavelet Coherence
The following WTC analyses explores the different environmental driving factors of
biological responses such as GPP, Fe, Fre and Fc, through meteorological drivers and the
climate indices. It varies from the power spectrum analysis in that is shows the common high
powers and cycling of two variables, building on the individual temporal analysis in sections
4.1 and 4.2.
As shown and discussed in Figure 1.2 (conceptual diagram of the wavelet analysis and its
functionality) the arrows indicate phase angles, which demonstrate the lagged characteristics
of ecosystem responses to their meteorological and climate drivers. Following the previous
spectral analysis, only the significant meteorological and biological coherences have been
included here, categorised by biological responses.
Chapter 4: Results
52
Gross primary productivity
The only significant coupling to occur between the meteorological drivers and GPP (besides
annual cycles) was that of Ta, which showed anti-phased coupling in periodicity at a scale
larger than ten years. There are also two periods of 5 year cycling throughout the century,
between 1920-1930 and 1970-1990. Interestingly, rainfall showed little correlation with GPP,
suggesting no coherence in cycles between the two variables.
Net ecosystem exchange
An intermittent six month cycle is observed over about 25% of the century, from 1900-2000,
between Fc and VPD (Figure 4.9). This coherence appears to be anti-phased. At scales larger
than annual, very little coherence is observed, however patchy coupling appears at cycles of
2.5 to 10 years. Between 1915 and 1930 a 3-5 year cycle appears with the response (Fc) out
of phase by 180⁰. This is not reflected in later decades of the same time frequency with mid
1970’s coherence out of phase by 360⁰ and again post 1988 which appears in phase. A weak
ten year cycle appears between 1930 and 1955 which again appears out of phase by 180⁰.
Figure 4.8: Wavelet coherence of GPP and temperature GPP-temperature is over the whole study period (1900-2013) using
daily data. The solid black line is the 5% significance level against red noise calculated from the Monte Carlo simulation.
Arrows are the relative phase angle (arrows pointing right are in-phase, arrows pointing left are anti-phase).
Chapter 4: Results
53
Ecosystem respiration
Ecosystem respiration appears to correlate inconsistently at six month cycles with both Ta
and VPD. However this coupling occurs at all phase angles, exacerbating the inconsistency.
Fre also correlates with Ta and VPD at cycles larger than 2.5 years. A 5 year cycle exists
between Fre and Ta between 1915-1935 that appears in-phase, as does the 5-10 year cycles
between 1965 and 2000. A lengthy coherence exists between 1925 and 1980 at a scale larger
than 10 years, which is between 360⁰ and 0⁰ in phase, and appears weakened and at 180⁰ out
of phase with VPD. Coherences are also observed between Fre and VPD at smaller cycles (3-
5 years) in the latter half of the century, between 1960 and 1990 that is 90⁰ out of phase.
Figure 4.9: Wavelet coherence of net ecosystem exchange and VPD. The solid black line is the 5% significance
level against red noise calculated from the Monte Carlo simulation. Arrows are the relative phase angle (arrows
pointing right are in-phase, arrows pointing left are anti-phase).
Figure 4.10: Wavelet coherence of (a) respiration and temperature and (b) respiration and VPD. The solid black
line is the 5% significance level against red noise calculated from the Monte Carlo simulation. Arrows are the
relative phase angle (arrows pointing right are in-phase, arrows pointing left are anti-phase).
Chapter 4: Results
54
Latent heat flux
The six month cycle again appears inconsistently in coherences of Fe with temperature and
VPD. From 1915-1935, a 3-5 year cycle appears in both driving variables of latent heat flux
which is in phase. A similar cycle occurs again from 1965-1980 which is out of phase by 90⁰
followed by a weakened coherence from 1985-2000 that has a phase angle of 180⁰, for VPD.
Fe and temperature appears to couple at much larger periodicities, from 1930-1990 which is
mostly in phase with a larger than 10 year cycle and again from 1970-2000 with a 5-10 year
cycle, also in phase.
Climate indices quantify climate phenomenon at the meso-scale, integrating influences of
temperature, rainfall and VPD through various processes, including ocean-atmosphere
interactions. The indices that have been chosen for this analysis represent atmospheric
processes in the South Pacific Ocean region (SOI); extra-tropical sea surface temperatures
(SST) off the south-east coast of Australia (TSI) and SST’s in the Indonesian region (II).
Time series of the climate indices are shown in Appendix D observing the change and trend
over the 20th
century. Frequency of SOI positive and negative phases appears even through
time whilst positive SST anomalies increase in frequency toward the end of the century in
both the Tasman Sea and Indonesian region.
Figure 4.11: Wavelet coherence of (a) latent heat and temperature and (b) latent heat and VPD. The solid black line
is the 5% significance level against red noise calculated from the Monte Carlo simulation. Arrows are the relative
phase angle (arrows pointing right are in-phase, arrows pointing left are anti-phase).
Chapter 4: Results
55
Southern Oscillation Index
The SOI is calculated via the Troup method (eq. 5) which establishes a monthly averaged
value describing conditions of sea surface temperature and mean sea level pressure
differences between Darwin, Australia and Tahiti.
Figure 4.12: Wavelet Coherence of SOI against the biological responses (a) ecosystem respiration, (b) gross
primary productivity, (c) latent heat flux and (d) net ecosystem exchange. The solid black line is the 5%
significance level against red noise calculated from the Monte Carlo simulation. Arrows are the relative phase
angle (arrows pointing right are in-phase, arrows pointing left are anti-phase).
Chapter 4: Results
56
The standout coherence that exists in all four biological fluxes is the 5-7 year cycle that
appears toward the end of the century (1980’s and onwards), though varies in length and
strength of the relationship between the four measures of carbon and water exchange. Also of
significance are the intermittent patches of coherence that occur at periods of 2.5-5 years
throughout the century across all four signals, correlating with much of the current literature
that suggests a similar cycle that is emulated through the El Nino Southern Oscillation
phenomenon. This will be discussed further in the following chapter.
Tables 4.2, 4.3, 4.4 and 4.5 at the end of section 3.3 offer a summary of the results from the
WTC analysis of SOI, TSI and II on biological responses.
Chapter 4: Results
57
Tasman Sea Index
The TSI is a measure of SST’s off the east coast of southern Australia, with inter-annual to
decadal time scales, similar to that of SOI and II. TSI is bound by coordinates 30⁰-40⁰S and
150⁰-160⁰E. Similar to SOI is the long-term cycle is found at the end of the century (post
1980) in biological responses of GPP, Le and Fc. A one year intermittent cycle has also been
observed throughout the century across all four variables.
Again the emerging pattern is the ten year cycle present throughout the century, especially
with regards to ecosystem respiration, which is known to respond to moisture availability.
Figure 4.13: Wavelet Coherence of TSI against the biological responses (a) net ecosystem exchange, (b) latent heat flux, (c) ecosystem
respiration and (d) gross primary productivity. The solid black line is the 5% significance level against red noise calculated from the
Monte Carlo simulation. Arrows are the relative phase angle (arrows pointing right are in-phase, arrows pointing left are anti-phase).
Chapter 4: Results
58
Indonesian Index
The II measures the average SST anomaly in a region north-west of Australia in the south-
east tropical Indian Ocean (0⁰-10⁰S, 120⁰-130⁰E). Coherences between II and the biological
variables appear inconsistent across time frequencies of one to ten years; however the ten
year cycle at the end of the century seen in SOI and TSI has again emerged, as has the
century-long ten year cycle in ecosystem respiration.
Tables 4.2-4.5 (gross primary productivity, latent heat flux, ecosystem respiration and net
ecosystem exchange respectively) below will summarise and compare indices from the WTC
spectral analysis by biological response.
Figure 4.14.: Wavelet Coherence of II against the biological responses (a) net ecosystem exchange, (b) latent heat flux, (c)
ecosystem respiration and (d) gross primary productivity. The solid black line is the 5% significance level against red noise
calculated from the Monte Carlo simulation. Arrows are the relative phase angle (arrows pointing right are in-phase,
arrows pointing left are anti-phase).
Chapter 4: Results
59
Table 4.2: Summary table of gross primary productivity responses to the climate indices, SOI, TSI and II,
following wavelet coherence spectral analysis. Observed periodicity and associated significant time periods and
phase angle are also listed.
Climate
Index
Periodicity
(y)
Time Periods Phase Angle Significance
Fraction
SOI
1
1908-1912, 1924-
1926, 1940,1970,
1999-2002
180⁰ 0.1
2.5
1929-1935, 1940-
1950, 1977-1988,
2004-2006
0⁰ 0.26
5 1983-2000 180⁰ 0.15
TSI
0.5
1907, 1925, 1936,
1960, 1970, 1990,
2000
180⁰ 0.06
1 1900-2013
(intermittent) 0⁰-180⁰ 0.19
2.5 1918-1921, 1965-
1970 0⁰ 0.07
10 1974-1999 180⁰ 0.23
II
0.5 1910-1980
(intermittent) 0⁰ 0.07
1
1909-1911, 1930-
1933, 1959-1961,
1975, 1992, 2001-
2005
180⁰ 0.12
2.5-5
1917-1931, 1945-
1952, 1969-1988,
1990-2001
180⁰ 0.46
10 1967-1998 180⁰ 0.28
Table 4.2 above suggests that at smaller time scales (1-2.5 year periodicity), SOI has the most
influence on driving GPP at Howard Springs whilst the II drives GPP cycles from 2.5-5 to 10
years, lagged by 180⁰. At scales smaller than 1 year, little significant coherence is observed
Chapter 4: Results
60
between climate indices and ecosystem responses, which are expected given that these
indices are most active at inter-annual to decadal time cycles.
Table 4.3: Summary table of latent heat flux responses to the climate indices, SOI, TSI and II, following
wavelet coherence spectral analysis. Observed periodicity and associated significant time periods and phase
angle are also listed.
Climate
Index
Periodicity (y) Time Period Phase Angle Significance
Fraction
SOI
0.5 1900-1980
(Intermittent) 0⁰-180⁰ 0.11
1
1908-1911, 1925-
1927, 1941, 1972-
1974, 2000-2002
0⁰ 0.09
2.5 1932-1936, 1943-
1951, 1980-1987 180⁰ 0.17
3 1940-1949 90⁰ 0.08
5 1988-2000 0⁰ 0.11
TSI
0.5 1900-1975
(intermittent) 0⁰-180⁰ 0.08
1 1900-2013
(intermittent) 0⁰-360⁰ 0.15
10 1983-1997 0⁰ 0.13
II
0.5 1900-2013
(intermittent) 180⁰-360⁰ 0.07
1 1900-2013
(intermittent) 0⁰ 0.08
2.5-5 1917-1938, 1943-
1951, 1968-1998 360⁰-0⁰ 0.63
10 1974-1997 0⁰ 0.21
The climate index responsible for driving latent heat flux at larger time scales is again the II,
with a significance fraction of 0.63 at cycles of 2.5-5 years. Whilst the SOI does not stand out
at any particular periodicity, it does cohere with over half of the century at various cycles and
time periods, suggesting that SOI may be the overall driver of latent heat responses and the
water cycle with the 2.5, 3 and 5 year cycles moving from anti-phase to in phase with
increased periodicity.
Chapter 4: Results
61
Table 4.4: Summary table of ecosystem respiration responses to the climate indices, SOI, TSI and II, following
wavelet coherence spectral analysis. Observed periodicity and associated significant time periods and phase
angle are also listed.
Climate
Index
Periodicity (y) Time Period Phase Angle Significance
Fraction
SOI
0.5 1900-1975
(intermittent) 360⁰-0⁰ 0.06
1 1900-2013
(intermittent) 0⁰ 0.07
2.5-3
1908-1912, 1936-
1939, 1941-1952,
1956-1969, 1979-
2000
180⁰ 0.47
10 1995-2000 0⁰ 0.05
TSI
0.5 1900-2013
(intermittent) 0⁰ 0.07
1 1900-2013
(intermittent) 360⁰-0⁰ 0.19
2.5 1963-1967 360⁰ 0.04
10 1924-1996 0⁰-90⁰ 0.65
II
0.5 1900-2013
(intermittent) 0⁰ 0.05
1 1900-2013
(intermittent) 0⁰ 0.1
2.5-5
1917-1921, 1925-
1935, 1942-1951,
1964-1984, 1990-
2001
360⁰-0⁰ 0.57
10 1924-1998 0⁰-90⁰ 0.67
Again at cycles 10 years and larger, the II appears to be the most significant driver of
ecosystem respiration, closely followed by TSI. At cycles between 2.5 years and 5 years, the
II and SOI drive respiration at different periods and different phase angles during the century.
This suggests that the exhalation of carbon by the ecosystem to the atmosphere is driven by
various indices at various periodicities over the century.
Chapter 4: Results
62
Table 4.5: Summary table of net ecosystem exchange responses to the climate indices, SOI, TSI and II,
following wavelet coherence spectral analysis. Observed periodicity and associated significant time periods and
phase angle are also listed.
Climate
Index
Periodicity (y) Time Period Phase Angle Significance
Fraction
SOI
0.5
1910,1919, 1932,
1937, 1940, 1965,
1970, 1993
0⁰
180⁰
0.07
1
1908-1910, 1926-
1927, 1930, 1938-
1942, 1972, 1999-
2001, 2005
90⁰-360⁰ 0.01
2.5
1923-1928, 1932-
1936, 1945-1952,
1980-1987
360⁰-0⁰ 0.21
3 1940-1955 360⁰ 0.14
5-10 1976-2000 180⁰ 0.22
TSI
0.5 1910-1975
(intermittent) 180⁰-0⁰ 0.07
1 1920-2000
(intermittent) 0⁰ 0.15
2.5-5 1920-1927, 1965-
1980 0⁰-90⁰ 0.2
10 1980-1995 90⁰-180⁰ 0.14
II
0.5 1915-1990
(intermittent) 0⁰-180⁰ 0.07
1 1910-2000
(intermittent) 180⁰ 0.11
2.5-5
1921-1930, 1937-
1945, 1947-1950,
1978-1990, 1991-
1996
90⁰-180⁰ 0.34
10 1972-1999 180⁰ 0.25
Chapter 4: Results
63
Of the four ecosystem responses, net exchange by the ecosystem appears to correlate with the
climate indices the least. Of the three indices, SOI and II correlate at ten year cycles with
some significance (greater than 20%), as do the 2.5-5 year periodicities. At these
periodicities, TSI also has a significance fraction of 0.2 and is the only index at this cycle to
lead Fc mostly in-phase (0⁰-90⁰).
Overall, TSI appears to have little coherence with ecosystem responses at small scales and
the only significant decadal cycle occurs with ecosystem respiration. The SOI seems to have
a greater influence on the carbon cycle, dictated through Fc, GPP and Fre, than the water
cycle (Fe) which correlates well with the II, especially at mid-size scales (2.5-5 years).
Carbon and water appear to respond to different phases of SOI, suggesting dissimilar drivers
of ecosystem carbon and water fluxes. Given the known cycles of the SOI (2.5-7 years), it
was expected that these periodicities would have displayed more coupling with ecosystem
fluxes, but instead it has appeared as a stronger driver of small scales, less than 2.5 years. At
the same time as increases in temperature, rainfall and VPD are observed in the linear trends,
coherence between climate indices and ecosystem responses are also observed, post 1970.
Chapter 5: Discussion
64
Chapter 5: Discussion
This thesis examined the temporal dynamics of climate and ecosystem fluxes at a typical
mesic savanna site in Howard Springs. Key climate variables (precipitation, air temperature
and VPD) increased from the mid-20th
century and exhibited spectral shifts and coherence
with modelled biological variables over the last four decades, suggesting an enhanced
coupling between climate and ecosystem processes. This shift in enhanced coherence is
reiterated in the climate indices plots, particularly at 10-12 year cycles post 1970. These
results suggest that SOI, TSI and II influence ecosystem function at inter-annual cycles
however, the source of this coupling (whether it is mechanistic or as a result of the statistical
methodology) is beyond the scope of this thesis.
5.1 Trends and temporal patterns
Meteorological time series at Howard Springs, NT correlate well with global and regional
trends of increasing temperature and vapour pressure deficit. By contrast, tropical rainfall has
decreased since the 1970’s (AR5, 2013) and increased over land surfaces above 30⁰ N/S.
Rainfall over Australia contradicts these patterns with increased rainfall in the tropics since
the 1970’s (BoM 2014), possibly due to increased SST’s in the Indian Ocean which has also
decreased rainfall significantly in the south and east (Pitman 2003).
Temperature
Temperature has increased globally by almost 1⁰C over the previous century (AR4 2007;
AR5, 2013; Karl & Knight 1998; New et al. 2001), in Australia by 0.9⁰C (BoM 2014) and
0.92⁰C at Howard Springs. Temperature across Australia has warmed significantly since
1950, a trend also reflected at Howard Springs, which has increased by 0.44⁰C from 1950-
2013 versus a decrease of 0.02⁰C from 1913-1950. These contrasting trends within the
century are shown in Figure 5.1.
Chapter 5: Discussion
65
Temperature impacts biological responses such as photosynthesis, respiration and
decomposition (e.g. Atkin & Tjoelker 2003; Kirschbaum 1995; Singh & Gupta 1977; Yi et al.
2010) however, many studies suggest that precipitation regimes hold more consequence for
ecosystem function (e.g. Huxman et al. 2004; Knapp et al. 2008), especially in arid and semi-
arid regions.
Rainfall and latent heat
Australian climate appears to have changed significantly since the early 1970’s (Nicholls et
al. 1998; CSIRO & BoM 2014) as was the finding at Howard Springs. The State of the
Climate Report (CSIRO & BoM 2014) observed higher maximum temperatures relative to
rainfall since this period, contributing to a climate where temperatures are warmer but rainfall
does not alter significantly, or even decline in some regions, such as south-west WA (Pitman
et al. 2004). These summations came from a continent-wide study and contrast our north
Australian site where both rainfall and temperature are increased.
Compared to consistent patterns of temperature, the complexity, interactions and scope of
global-scale climate processes (Knapp et al. 2008) has resulted in difficulty in predicting
potential changes in precipitation regimes (Lim & Roderick 2009). Many GCM’s project a
modest increase in rainfall at a global scale with large discrepancies at regional and local
scales.
Rainfall has had a net increase across Australia (1910-2013) (BoM 2014); however it has
declined since 1970 in the southwest via reduced winter rainfall (Cai et al. 2009). Long-term
predictions for Australia suggest this trend will continue in Southern Australia but increased
rainfall is forecast to increase over most other parts of Australia, (BoM, 2014). This reaffirms
Figure 5.1: Variable temperature trends in the previous century at the Howard Springs site. (a) Temperature trend
1913-1949 and (b) temperature trend 1950-2013. Annual average values are used to generate the time series.
y = -0.0009x + 26.527 25.5
26.5
27.5
1913 1918 1923 1928 1933 1938 1943 1948
Ta (
oC
)
a)
y = 0.0071x + 26.823
25.5
26.5
27.5
19
50
19
55
19
60
19
65
19
70
19
75
19
80
19
85
19
90
19
95
20
00
20
05
20
10
Ta (
oC
)
b)
Chapter 5: Discussion
66
the results of this study, which showed increases in the latter half of the century in the
Howard Springs grid (Figure 4.1).
The impact of increased rainfall on tropical ecosystems is not well understood, with most
studies focusing on subtropical or temperate regions (e.g. Ciais et al. 2005; Reichstein et al.
2007; Knapp et al. 2008), with an emphasis on the effects of changes in rainfall amount and
seasonality (e.g. Beier et al. 2004). The intensification of the hydrological cycle as a
consequence of global warming is known and expressed in the form of increased cloudiness,
latent heat fluxes and frequent climate extremes (AR4 2007). Further, less frequent but more
intense precipitation events may alter latent heat fluxes and generate greater runoff (Fay et al.
2003; MacCracken et al. 2003). The relationship is indirect, increased rainfall suggests
increased cloud cover and reduced radiation, decreasing latent heat (evapotranspiration) so
greater runoff results.
This relationship is not evident at the Howard Springs site through increased rainfall and Fe
over the 20th
century. There is little coherence observed between rainfall and latent heat
beyond the annual cycle (Figure 5.2); despite finding coherences between Fe and climate
indices intensifying post 1970. This is possibly due to the strongly seasonal wet-dry
characteristics of the region and the vegetative properties of savanna which do not appear
water-stressed, even during the dry season (O’Grady et al. 1999).
Figure 5.2: WTC analysis of rainfall and latent heat, demonstrating the driving environmental conditions at an
annual cycle, but little coherence at other periods, especially post 1970 where wetting of the ecosystem is
expected to increase latent heat and other biological drivers.
Chapter 5: Discussion
67
The coherence at the annual scale is expected as annual rainfall provides a store for dry
season Fe. These ecosystems are likely not water-limited, resulting in more runoff as rainfall
increases with no response from Fe. Also unknown are the ecological implications of more
extreme rainfall regimes (Jentsch et al. 2007), a significant knowledge gap given the impacts
of enhanced rainfall variability on terrestrial ecosystems are predicted to rival the ecological
impacts of other global-scale changes such as increasing CO2 concentrations (Easterling et al.
2000; AR4 2007). Analysis of model results from the Howard Springs site suggests that
while there is a strong increasing trend in Fe over the century (Figure 4.3), coherence with
rainfall was weak.
VPD
VPD has increased since the mid 1940’s (Figure 4.1), probably as a result of increased
temperature which will have substantial eco-physiological impacts. VPD is a significant
driver of leaf (Prior et al 1997) and canopy scale (Eamus et al. 2001) conductance and is
particularly significant in seasonal environments. In general, as VPD increases beyond a
threshold we see a feedback to stomatal conductance which declines (Atwell et al. 1999)
limiting Fe and ultimately Fc. VPD responses in north Australian savanna have been
extensively examined at leaf, tree and canopy scales (Drake et al. 2014; Eamus et al. 2001).
During the wet season there is limited impact on gas exchange with increasing sensitivity as
the dry season progresses. Beyond temperature and VPD thresholds, gas exchange will be
limited reducing carbon sequestration. Future carbon sink by the savanna ecosystem will be
restricted by the warming environment with declines in production as temperatures warm
throughout the century. A thermal regime may be a dominant driver of ecosystem exchanges
in the future.
Carbon fluxes
The trend line analysis from the ANN suggests an increasing sink as indicated by GPP and Fc
over the previous century (Figure 4.2). Over the last 50 years, GPP has increased at a greater
rate than Fre because GPP (Figure 4.2) responds to increased temperatures and rainfall as
well as increased atmospheric CO2 (Scheiter et al. 2014). As a consequence Fc has an
increasing trend (more –ve and a sink), despite increased respiratory losses from the
ecosystem.
This sink is consistent with global tropical forest trends, which has been large (1.2 Pg C y-1
in
1990-2007) in recent decades (Pan et al. 2011); however, the terrestrial carbon sink in the
Chapter 5: Discussion
68
tropics has had the most uncertainty in literature, with many papers citing source, sink or
neutrality of tropical ecosystems. For example, Sabine et al. (2004) claims that tropical
forests are the greatest carbon sink among terrestrial systems, locking up large quantities in
both biomass and soil, conflicting with the argument that deforestation and burning reduce
carbon storage and increase the CO2 release to the atmosphere in the tropics (e.g. DeFries et
al. 2002; Houghton 1991; 2004).
Whether or not tropical ecosystems are a source or sink of atmospheric carbon dioxide is a
contentious issue within the literature with the scientific community debating the results of
studies and the use of methods to fuel arguments. Recently, Wright (2013) challenged the
results of a global study by Pan et al. (2011) who quantified the carbon sink of intact tropical
forests using region-specific estimates of carbon density or change in carbon density times,
stating that their estimated sink of 1.2 petagrams of carbon per year between 1990-2007 was
too large. Wright argued that sources used for forest area and carbon stocks were also
inaccurate (there is no global agreement on defining forest area) and that the study
overlooked or ignored a source for carbon stocks in their global study. Defending the Pan et
al. study was Phillips and Lewis (2013) who in a return letter to Wright contested that he only
focused on one aspect of the paper and ignored the global results, making unsubstantiated
(Phillips & Lewis 2013) claims regarding the mechanisms of carbon dioxide fertilisation
which were largely beyond the scope of the study.
Much of the uncertainty involving terrestrial sources or sinks of carbon in Australia is due to
modelling inconsistencies and predictions (Roxburgh et al. 2004). Models vary in approach
and driving inputs resulting in estimates that range 5-fold with most contradictions occurring
in the arid zone of Australia. Haverd et al. (2013) attempted to solve some of these
discrepancies using the Community Atmosphere Biosphere Land Exchange (CABLE/BIOS2)
model, which suggests savanna is a weak sink in Australia. However, climate and fire
disturbances experienced by the coastal and sub-coastal north Australian savannas create a
spatial and temporal mosaic of carbon sources and sinks and this study did not include
additional disturbance. Beringer et al. (2014) found that when disturbances are added, the
ecosystem becomes a weak carbon source.
The results from this study contribute to the uncertainty in mechanisms present driving the
terrestrial carbon cycle. We have observed increased GPP and rainfall at the Howard Springs
site, which would suggest the greening occurred as a result of increased precipitation
Chapter 5: Discussion
69
regimes. However, Donohue et al. (2013) observed the same increased trend in greening
(enhanced photosynthesis) even in areas of reduced rainfall using satellite analysis. These
regions (arid environments) are predominantly water-limited, contrasting the characteristics
of coastal savanna. This would suggest another driver at play, such as increasing atmospheric
CO2 concentration. The effect on photosynthesis resulting from rising CO2 is yet to be
established however, the CO2 fertilisation effect appears as a significant driver in these warm
and arid environments, reflecting GPP and photosynthesis at Howard Springs which
measured an identical trend based on observations and expectations given the meteorology of
the last 100 years, driven by warming, wetting and CO2 perturbations.
Trends in ecosystem physiology
Analysis of ecosystem physiology provided this study with an overview of ecosystem
function. Despite increased VPD and rainfall, simulated ecosystem WUE remained
unchanged throughout the observation period (2001-2013). WUE is a ratio of GPP and Fe,
both of which increased at similar rates over the 20th
century, resulting in the unchanged
trend. WUE was reduced during the wet season in response to ample rainfall and low VPD,
however, the dry season also displayed periods of reduced WUE, often around July or August
of any year. Water use was most efficient following the end of the dry season, peaking
around May or June each year, followed by a second, stronger peak during the late dry. This
is likely due to disturbances such as fire, resulting in very dry atmospheric conditions,
maximising VPD. Studies are seeing a global increase in WUE due to increased atmospheric
CO2 concentrations (Fuhrer 2003), a trend that is not established at the Howard Springs site,
possibly due to the small time period (2001-2013). Both light use and water use efficiency
show high intra-annual variability, as would be expected in a region with strong seasonal
dynamics. This is especially apparent for LUE, as it responds to less atmospheric interference
with solar radiation (i.e. cloudiness) which decreases LUE as the dry season persists.
Analysis of savanna biotic responses and physiological trends demonstrated the variability
and change over the 20th
century resulting from increased temperature, rainfall, VPD and CO2
concentrations. The temporal relationships between these variables (meteorological drivers
and flux responses) appear strong with persistent cycling of ecosystem variables driven by
the meteorology at Howard Springs.
Chapter 5: Discussion
70
Evidence for ecosystem coupling to climate
WTC identified temporal correlations and time lags between climate drivers and ecosystem
processes, with coherence absent in the early 20th
century for many coupled variables, but
strengthened significantly by the mid 1970’s. Climate-ecosystem coupling strengthened at the
same time as the rate of climate change increased during the second half of the 20th
century.
Such relationships are evident in recent studies using similar analyses (e.g. Hong & Kim
2011) but no investigation has examined either the savanna biome or ecosystems of north
Australia.
Ecosystem variables GPP, Fre, Fc and Fe showed little coherence beyond the expected
annual cycles. This is surprising given that the global spectrum analyses suggested longer
cycling present in climate indices. It was hypothesised that environmental variables would
show little cycling beyond annual frequencies, whereas spectral analysis suggested weak 5-10
year cyclic patterns (Figure 4.3). Following these observations, one would expect to see
coherences at similar or longer temporal frequencies when correlated with meteorological
drivers. This was the case for temperature and Fe, Fre and GPP with strong decadal cycles
approximately mid to late century.
Fe was shown to have a significant and consistent coupling with temperature at frequencies
greater than 10 years. It also appears that throughout this coupling (beginning mid 1920’s) a
0⁰ lag is observed, meaning Fe responds immediately to temperature as a driver; although this
shifts to approximately 360⁰ (or 12 months) out of phase later on in the century and as the
periodicity got larger. Absent from coherence with temperature is Fc, interesting since the
other two carbon variables displayed strong coherence at a decadal cycle.
Figure 5.3: Wavelet coherence of temperature and (left) Fe (middle) Fre and (right) GPP. Besides the annual
cycle, the strongest and most consistent cycle has a frequency greater than 10 years.
Chapter 5: Discussion
71
There are few studies linking between increasing temperature and Fc of tropical ecosystems.
Zhou et al. (2012) observed increased Fc as a result of increased soil temperature, as tested in
an urban, temperate environment although the study had no temporal component. Similarly,
Hong & Kim (2011) observed a strong six-month cycle between Fc and rainfall in a
monsoon-driven temperate deciduous forest in Korea. In contrast to this study where there
was no observed coherence between these two variables (Appendix E) at annual scales. This
in dissimilar to Beringer et al. (2007) who observed that over the short-term (six annual
cycles), there was a strong relationship between rainfall and Fc and GPP. It is unknown why
this expected relationship showed no significant coherence using longer time steps.
Climate indices
Time series of the climate indices are illustrated in Appendix D to compare changing
conditions to ecosystem flux variables. The temporal cycles of the three indices were
evaluated throughout the 113 year period to analyse their cyclic patterns (Figure 4.7) which
in turn may influence environmental drivers of temperature, rainfall and VPD. SOI showed
the highest power spectrum at a seven year cycle across the century (Figure 4.7). As an index
of ENSO, it is thought of as a principal tropical influence (Allan 1988; Nicholls et al. 1997;
Wang & Hendon 2007). The 2.5-5 year cycles that appear in similar studies are not observed
in the power spectrum analysis but it does appear in the wavelet coherence investigation as a
coupling between SOI and ecosystem responses of carbon and water.
There have been numerous studies examining the influences of SOI on Australian rainfall and
tropical weather (e.g. Allan et al. 1991; Nicholls et al. 1996; Power et al. 1998; Risbey et al.
2009; Suppiah 2004). These studies in general contradict the inconsistent patterns in the
coherence spectra seen in these results. Studies such as Nicholls et al. (1996) suggest good
correlation between Australian rainfall, temperature and SOI between seasons using principal
component analyses. This is not the case at the Howard Springs site; rainfall appears mostly
independent from SOI prior to 1970, in contrast, temperature loses coherence with SOI by
1950. In the second half of the 20th
century, significant changes are observed in the cyclic
patterns between SOI and temperature and rainfall at decadal cycles. A similar decadal cycle
is observed in temperature and SOI earlier in the century.
Chapter 5: Discussion
72
(a) Temperature-SOI
-60
-40
-20
0
20
40
26.3
26.8
27.3
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
SOI
Tem
pe
ratu
re (
oC
) SOI and Temperature as a 10 year running mean 1900-2013
(b) Rainfall-SOI
-60
-40
-20
0
20
40
120
140
160
180
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
SOI
Rai
nfa
ll (m
m)
SOI and Rainfall as a 10 year running mean 1900-2013
Figure 5.4: Wavelet coherence spectra of (a) temperature and (b) rainfall with SOI. Graphs show time series
of SOI (black) with a 10 year running mean (grey) of (top) temperature and (bottom) rainfall, displaying the
accuracy (or not) of the wavelet coherence analysis in picking up the decadal signals.
Per
iod
1 Month
6 Months
1 Year
2.5 Years
5Year
10 Years
Per
iod
1 Month
6 Months
1 Year
2.5 Years
5Year
10 Years
Chapter 5: Discussion
73
The decadal cycle present in the wavelet coherence between SOI and temperature in the early
century is not easily identifiable in the time series (Figure 5.4.b), but there is a very
noticeable increase in the 10 year running mean of temperature between 1950 and 1960. This
suggests a period of warming which was maintained throughout the second half of the
century, which has been widely reported in the literature. The same decadal cycle observed in
the spectral analysis of rainfall and SOI post 1980’s appears very clearly in the time series
analysis, with less than a 90⁰ phase angle, thus confirming the acknowledged change in the
relationship between SOI and rainfall after the 1970’s.
Evidence of the changed relationship between these variables and SOI has been described in
many climatological studies (e.g. Fredericksen & Balgovind 1994; Nicholls et al. 1996); with
many suggesting these changes are results of spatially varied distribution of the global sea
surface temperature anomalies since the early 1970’s.
There is a recurring 2.5 year frequency in the coherence spectra of SOI with biological
variables which are distinctive, as is the late-century coherence at the 5-7 year (GPP, Fe, and
Fc) and 10 year (Fre) periodicities (Figure 4.12). The earliest 5-7 year coherence occurs in
the mid to late 1970’s coupling with Fc whilst the 10 year cycle doesn’t appear with Fre until
the mid-1990’s which is then not significant.
The 2.5 year cycle appears patchy in each analysis of GPP, Fc, Fre and Fe. GPP and Fc have
similar coherences to SOI, occurring mostly early to mid-century, coinciding with relatively
neutral SOI conditions (i.e. within 1 standard deviation, standardised to 10). The exception to
this carbon flux trend was Fre which correlated with a La Nina (wet) phase followed by the
persistent El Nino in the mid 1990’s. Fe cycling at the 2.5 year periodicity appears to
correlate to periods of El Nino (dry conditions over western Pacific) which are strictly anti-
phased. These occur around 1932, mid 1940’s and 1980. Figure 5.5 demonstrates the variable
response between carbon and water exchanges to SOI cycles.
Chapter 5: Discussion
74
These analyses suggest the carbon cycle correlates with neutral conditions of SOI whilst the
water use (Fe) responds to drier periods of significant El Nino phases, advocating the
correlation between ENSO and tropical Australia’s water balance. Phase angles also change
with SOI. In the five year periodicities, GPP and Fc lag SOI by 180⁰ while Fe is in phase (0⁰)
with SOI. This suggests that water exchanges respond at immediate time scales with drier
conditions whilst carbon exchanges lag by six months to neutral conditions of SOI. The
significance of these responses are currently unknown.
The use of TSI to quantify extra-tropical SST’s has only recently emerged in the literature
(e.g. Murphy and Timbal 2008 and Schepen et al. 2012) hence there are not many studies that
include this index.
Anthropogenic climate change has most likely caused the increase in ocean temperature over
the previous century, which has seen positive increases within the TSI (and II) since 1900,
reflecting the IPCC’s AR4 finding of increased oceanic temperatures between 1961 and
2003 (Bindoff et al. 2007). Another feature of the TSI is its inter-annual and decadal cycles,
which appear weak in the spectral analysis (probably due to the low-resolution), but emerge
clearly within the coherence analysis, confirming its influence on environmental conditions
Figure 5.5: Variation in response to SOI conditions between carbon (Fc) and water (Fe) exchanges of savanna ecosystem. 2.5 year
periodicities of carbon exchanges respond to neutral conditions of SOI, water exchanges respond to drier (El Nino) conditions.
-40
-20
0
20
40
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
MSL
P S
tan
dar
d D
evia
tio
n
SOI
Chapter 5: Discussion
75
such as the increased wetting. These changes to moisture availability (e.g. through enhanced
evaporation off a warmer ocean surface) drive increases in ecosystem fluxes such as
respiration, which may eventually see a shift from sinking Fc to a carbon source in the
tropics.
II correlates well with Australian winter rainfall, which is sensitive to SST variations in the
Indonesian ocean region. It has no observed seasonal cycle, but a previous honours thesis (see
Rogers 2012, Monash University) has suggested it is one of the three highest correlated
climate indices with rainfall distribution over the study site. The II also exhibits inter-annual
and decadal cycles which, like the TSI, emerge within the coherence plots. The spectral
analysis demonstrated the cyclic patterns within the individual variables, the knowledge of
which helps to understand the patterns of ecosystem coupling to the changing climate within
the wavelet coherence analysis.
The Tasman Sea Index (TSI) and Indonesian Index show little consistency in coupling to the
ecosystem fluxes prior to the mid-late 1970’s. The exception is Fre suggesting a correlation
to TSI and II at periodicities larger than 10 years (Figure 4.6, 4.7). It appears that as positive
anomalies of TSI and II (i.e. warmer SST in Tasman Sea and Indonesian region) become
more frequent (Appendix D), the coupling weakens between respiration and climate indices.
It is possible that this is a mathematical coincidence as the spectral density of Fre weakens at
the same time (mid-1970’s) and increased moisture availability is a known driver of
respiration (Chen et al. 2002). GPP, Fc and Fe appears strongly coupled with both indices
post 1970 with carbon exchanges anti-phased and water exchanges in-phase with ten year
cycles of TSI and II.
Autotrophic and heterotrophic respiration is largely driven by temperature and the simulated
increase in Fre occurred since the 1970’s (Figure 4.2) and is likely to be driven by the
temperature increase of the 1970’s, since the model uses this driver-response mechanism. If
predictive modelling were used in a future investigation, the relationship of ecosystem
respiration and SST’s in Indonesian waters and the Tasman Sea could prove to be correlated
as oceanic warming increases.
Previous studies have identified multi-decadal warming and cooling trends over the Pacific
and Indian oceans (Allan et al. 1995; Franks 2002; Power et al. 1999). Power et al. 1999
identified a switch to a warming phase in the mid-1970’s over the Pacific Ocean and it
appears that a similar switch in the Indian Ocean to a warming phase may also have occurred
Chapter 5: Discussion
76
around the same time, in particular the central and western Indian Ocean (Verdon & Franks
2005). This shift may help to explain the wetting observed at Howard Springs during this
time as well as changes to coupling between environmental drivers and ecosystem fluxes.
This is reflected across WTC analyses using climate indices and is likely to be driven by
observed global warming and increasing ocean temperatures. Responding to these changes is
the increased Fre.
Limitations to the study
There are limitations of the wavelet coherence analysis via its inability to pick up
coincidental (mathematical) patterns of coupling. It is a mathematical analysis running off a
statistical model, rather than a mechanistic model, which has allowed the wavelet to pick up
mathematical coincidences. Another limitation to the modelling is the exclusion of ecological
disturbances such as fire and cyclones which have a decadal to 50 year impact (Hutley et al.
2013). This is a significant factor for Howard Springs as it is likely to have been impacted by
Cyclone Tracy in December 1974.
Neural networks are highly empirical statistical models which creates a possible limitation to
this study’s methodology and results. In this case, the ANN was driven solely by meteorology
whereas many process-based models (such as CABLE) utilise variables defining ecosystem
structure in a demographic approach. The ANN is a mathematical tool but has had
mechanistic constraints imposed via selected meteorological drivers a priori such as
temperature of the air and soil, rainfall, wind speed and VPD. The use of fractional
photosynthetically active radiation (fPAR) would have allowed the ANN to capture structural
components of the ecosystem, such as LAI which would tolerate seasonal variability to be
picked up by the model. This means that the ANN assumes static relationships that are
extrapolated back in time, which these results have demonstrated is not the case. Given the
neural network is trained to predict fluxes within a certain variable space or range, if we
move out of that variable space e.g. via increased temperature over a century, there is no
certainty the model will provide meaningful projections for changed conditions. However,
the training (observations) data have enough incidences of high temperatures to provide
robust predictions thus, if climate change simply shifts the frequency of wet, dry, hot or cold
conditions, these trends are all within the domain of the training data. Therefore the ANN is
not necessarily a limitation of this study if used conservatively as an analysis tool, given its
performance can be superior to a mechanistic model.
Chapter 5: Discussion
77
Fire is also not specifically dealt with in this study, though the ANN was trained using fire-
affected variables, so it has not been excluded altogether. Beringer et al. 2003, 2007
examined fire impacts on albedo, Fe and Fc and demonstrated that at a canopy scale, fire had
significant impact on the surface energy balance and annual sequestration rate with sever fire
consuming almost 50% of the annual sink (Beringer et al. 2007). Fire periods were not
screened and will have introduced error as the ecosystem would appear in dis-equilibrium
relative to driving meteorology post-burn (Beringer et al. 2003, 2007) during the
observational period. This means that some error was introduced during ANN training as the
model is reliant on fPAR to determine post-fire fluxes. In addition, Fe recovers at a very
different rate to Fc. Beringer found recover in Fe to pre-fire rates after a month or less
following a severe fire event (80% canopy scorch), whereas Fc took three months and the
onset of the wet season to shift the ecosystem from the strong post-fire source to a net carbon
sink (Beringer et al. 2007). In effect, fire events are integrated into the vegetation structure
and function, so individual fire events are actually built into the ANN when looking at 100
year trends. Of course what hasn’t been tested is the extent to which the neural networks
perform during fire-affected periods. Increased CO2 impact on savanna fluxes are not
directly addressed in this study, though we know the influence of CO2 fertilisation will be a
major factor driving the greening, or photosynthesis of terrestrial ecosystems (Donohue et al.
2013).
All ecosystem carbon and water variables have demonstrated increased coupling to climate
phenomena such as ENSO and SST increases in tropical (II) and extra-tropical (TSI) regions,
with cycles of 5 years and larger, occurring since the mid 1970’s. The consistency of this
pattern is significant and suggests that with increasing temperature, rainfall and VPD (as
driven by these climate phenomena) in the Howard Springs region, ecosystem variables
across the spectrum of carbon and water (i.e. GPP, respiration, net exchange and latent heat)
have also increased, as a coupled response to warming and wetting of the tropical land
surface. This is further validated through the increased coupling observed between SOI and
temperature and rainfall at the Howard Springs site, and reflects global tropical patterns and
predictions in future climate scenarios.
Chapter 6: Conclusions
78
Chapter 6: Conclusions
This investigation is the first of its kind to use wavelet spectral analysis over a century-long
period in such a context. There is evidence of increasing climatic influence on savanna
ecosystems in north Australia since the 1970’s. This has repercussions for our ability to
confirm both the results and compare to other studies across tropical terrestrial surfaces and
other savanna ecosystems. This complicates validation of the modelled data and limits the
conclusions that can be made from the results. These limitations are mostly derived from the
lack of observational data and the issue of feeding the wavelet algorithms modelled data.
The major conclusions are:
There is increased coupling of climate indices and ecosystem responses at decadal
cycles following a warming and wetting environment post 1970;
The role of SOI on rainfall, but not so much temperature, is influential in driving
ecosystem function, even though temperature appears as a stronger driver than
rainfall;
Both TSI and II influence ecosystem function at inter-annual cycles however, whether
this has a mechanistic basis is uncertain, given the statistical nature of the approach
used in this study;
Consistent with sensitivities to Fe to high and low rainfall years, latent heat constantly
correlates with El Nino phases of SOI, while the carbon fluxes correlate at neutral
phases, suggesting dissimilar climate drivers for carbon and water balances in the
tropics;
There is a disconnect between the carbon and water exchanges in response to
temperature and rainfall. Lags exist between biological responses and environmental
drivers which vary by 180⁰ (six months) between carbon and water fluxes.
These results can be further developed by analysing additional sites to unpack the
mechanisms involved between savanna and climate. A biome-scale assessment of
climate impacts on savanna ecosystems should integrate a spatial component and field
work validation.
Chapter 6: Conclusions
79
The results from this study suggest that the ecosystem is adjusting to the changing climate,
which contradicts Moncreiff et al. (2014) who observed that future dynamic global vegetation
models predict narrow climate space for savannas and grasslands.
This study has provided insight to the behaviour of coastal savanna ecosystems. These
ecosystems cover tens of thousands square kilometres of the tropics and experience high
rainfall and frequent fire, storm and cycle disturbance (Hutley et al. 2013), modifying land-
atmosphere interactions for months after disturbance. The project made use of one of the
longest savanna flux towers in the world, situated 35 kilometres from a Class 1 (highly
accurate, long-term observations) Bureau of Meteorology station (Darwin) which contributes
to the validity of these results.
This unique study has improved our understanding of the influence of past and present
climate variability to savanna land-atmosphere exchanges using a novel time series analysis.
It has established the changing climate as a dominant driver of terrestrial ecosystem processes
in tropical north Australia, a region of uncertainty in the literature.
References
80
References
ABRAMOWITZ ET AL. (2008) ‘Evaluating the performance of land surface models’ J.
Climate Volume 21 Issue 21 pp. 5468-5481
- (2011) ‘Diagnosing errors in a land surface model (CABLE) in the time and
frequency domains’ J. Geophysical Research: Biogeosciences Volume 116 Issue G1
doi:10.1029/2010JG001385
ACHUTARAO & SPERBER (2002) ‘Simulation of the El Nino Southern Oscillation: Results
from the coupled model inter-comparison project’ J. Climate Dynamics Volume 19 Issue 3-4
pp. 191-209
ADELI ET AL. (2003) ‘Analysis of EEG records in an epileptic patient using wavelet
transform’ J. Neuroscience Methods Volume 123 Issue 1 pp. 69-87
ALLAN ET AL. (1995) ‘Multi-decadal variability in the climate system over the Indian
Ocean region during the Austral summer’ J. Climate Volume 8 Issue 7 pp. 1853-1873
-ET AL. (1988) ‘El Nino Southern Oscillation influences in the Australasian region’
J. Progress in Physical Geography Volume 12 Issue 3 pp. 313-348
ARCHIBALD & SCHOLES (2007) ‘Leaf green-up in a semi-arid African savanna
separating tree and grass responses to environmental cues’ J. Vegetation Science Volume 18
Issue 4 pp. 583-594
- ET AL. (2009) ‘Drivers of inter-annual variability in Net Ecosystem Exchange in a
semi-arid savanna ecosystem, South Africa’ Biogeosciences Vol 6 pp. 251-266
ANDRE ET AL. (1988) ‘Evaporation over land surfaces: First results from HAPEX-
MOBILHY special observing period’ J. Annales Geophysicae Volume 6 pp. 477-492
ASHOK ET AL. (2012) ‘Impact of the Indian Ocean dipole on the relationship between the
Indian monsoon rainfall and ENSO’ Geophysical Research Letters Volume 28 Issue 23 pp.
4499-4502
ATKIN & TJOELKER (2003) ‘Thermal acclimation and the dynamic response of plant
respiration to temperature’ Trends in Plant Science Volume 8 Issue 7 pp. 343-351
References
81
ARYA, (2005) ‘Micrometeorology and Atmospheric Boundary Layer’ Birkhauser Publishing.
Journal of Pure and Applied Geophysics. Volume 162, pp. 1721-1745
AVISSAR & VERSTRAETE (1990) ‘The representation of continental surface processes in
atmospheric models’ Reviews of Geophysics Volume 28 Issue 1 pp. 35-52
BALDOCCHI (2003) ‘Assessing the Eddy Covariance Technique for evaluating carbon
dioxide exchange rates of ecosystems: Past, Present and Future’ Review Article, Journal of
Global Change Biology Volume 9, pp. 479-492
- (2008) ‘Breathing of the terrestrial biosphere: lessons learned from a global
network of carbon dioxide flux measurement systems’ Turner Review No. 15. CSIRO
Publishing. Aust. J. of Botany. Volume 56, pp.1-26
BEER ET AL. (2010) ‘Terrestrial gross carbon dioxide uptake: Distribution and covariation
with climate’ Science Volume 329 No. 5993 pp. 834-838
BEIER ET AL. (2004) ‘Effects of climate and ecosystem disturbances on biogeochemical
cycling in a semi-natural terrestrial ecosystem’ Biogeochemical Investigations of Terrestrial,
Freshwater and Wetland Ecosystems across the Globe. Springer: Netherlands Pp. 191-206
BERINGER ET AL. (2011) ‘SPECIAL – Savanna Patterns of Energy and Carbon Integrated
across the Landscape’ Bulletin of the American Meteorological Society (BAMS) November
1, 2011 pp. 1467-1486
- ET AL. (2003) ‘Fire impacts on surface heat, moisture and carbon fluxes from a
tropical savanna in northern Australia’ Int. J. Wildland Fire Volume 12 Issue 4 pp. 333-340
- ET AL. (2011) ‘Savanna patterns of energy and carbon integrated across the
landscapes’ J. American Meteorological Society Nov. 2011 pp. 1467-1485
- ET AL. (2014) ‘Fire in Australian savannas: From leaf to landscape’ Research
Review. J. Global Change Biology doi:10.1111/gcb12686
BETTS ET AL. (1996) ‘The land surface-atmosphere interaction: A review based on
observational and global modelling perspectives’ J. Geophysical Research: Atmospheres
Volume 101 Issue D3 pp. 7209-7225
References
82
BINDOFF ET AL. (2007) ‘Oceanic change and sea level’ The Physical Science (AR4) Ch. 2
pp. 105-107
BoM & CSIRO (2014) ‘State of the Climate Report 2014’ (www.bom.gov.au/state-of-the-
climate/documents/state-of-the-climate-2014_low-res.pdf?ref=bottom ) Accessed September
2014
BOND (2008) ‘What limits trees in C4 grasslands and savannas?’ J. Ecology, Evolution and
Systematics Volume 39 pp. 641-659
- & MIDGLEY (2012) ‘Carbon dioxide and the uneasy interactions of trees and
savanna grasses’ Phil. Trans R. Soc B. Volume 367 No. 1588 pp. 601-612
BOUNOUA ET AL. (2002) ‘Effects of land cover conversion on surface climate’ J. Climate
Change Volume 52 Issue 1-2 pp. 29-64
BOWLER ET AL. (2001) ‘Variations of the northwest Australian summer monsoon over the
last 300 000 years: The paleo-hydrological record of the Gregory (Mulan) Lakes System’ J.
Quaternary International Volume 83-85 pp. 63-80
BOWMAN ET AL. (2010) ‘Biogeography of the Australian monsoon tropics’ J.
Biogeography Volume 37 Issue 2 pp. 201-216
- & PRIOR (2005) ‘Why do evergreen trees dominate the Australian seasonal
tropics?’ Turner Review No. 10 Aust. J. Botany Volume 53 Issue 5 pp. 379-399
BURBA & ANDERSON (2005) ‘A brief practical guide to eddy covariance flux
measurements: principles and workflow examples for scientific and industrial applications’
American Geophysical Union, Fall Meeting 2006
CAI ET AL. (2009) ‘Recent unprecedented skewness towards positive Indian Ocean Dipole
occurrences and its impacts on Australian rainfall’ Geophysical Research Letters Volume
36, Issue 11 pp. 1-4
CAO & WOODWARD (1998) ‘Dynamic responses of terrestrial ecosystem carbon cycling
to global climate change’ Letter. Nature Volume 393 pp. 249-252
CAZELLES ET AL. (2008) ‘Wavelet Analysis of Ecological Time Series’ Journal of
Population Ecology Volume 156, pp.287-304. Springer Publication
References
83
CIAIS ET AL. (2005) ‘European-wide reduction in primary productivity caused by the heat
and drought in 2003’ Letter. Nature Volume 437 pp. 529-533
CHANG & GLOVER (2010) ‘Time-frequency dynamics of resting-state brain connectivity
measured with fMRI’ J. Neurolmage Volume 50 Issue 1 pp. 81-98
CHAVES & PEREIRA (1992) ‘Water stress, CO2 and Climate Change’ J. Experimental
Botany Volume 43 Issue 8 pp.1131-1139
CHEN ET AL. (2002) ‘Seasonal patterns of soil carbon dioxide efflux from a wet-dry
tropical savanna of northern Australia’ Aust. J. of Botany Volume 50 Issue 3 pp. 373-373
- ET AL. (2003) ‘Carbon balance of a tropical savanna of northern Australia’ J.
Ecosystem Ecology Volume 137 pp. 405-416
CHIEW ET AL. (1998) ‘El Nino Southern Oscillation and Australian rainfall, streamflow
and drought: Links and potential for forecasting’ J. Hydrology Volume 204 Issue 1-4 pp.
138-149
CLEUGH ET AL. (2012) ‘2012 OzFlux Workshop Presentation’ Terrestrial Ecosystem
Research Network (TERN). CSIRO Marine and Atmospheric Research Centre.
<www.ozflux.org.au/meetings/july2012/Wednesday/2_HelenCleugh.pdf > PowerPoint
Accessed 20/03/2014
COOK. & HEEREDEGEN (2001) ‘Spatial Variation in the Duration of the Rainy Season in
Monsoonal Australia’ CSIRO Sustainable Ecosystems, Darwin, Australia. International
Journal of Climatology Volume 21, pp. 1723-1732
CRAMER ET AL. (2001) ‘Global response of terrestrial ecosystem structure and function to
CO2 and climate change: results from six dynamic global vegetation models’ J. Global
Change Biology. Volume 7, Is. 4, pp. 357-373
DRAKE ET AL. (2014) ‘The capacity to cope with climate warming declines from temperate
to tropical latitudes in two widely distributed Eucalyptus species’ J. Global Change Biology
doi:10.1111/gcb.12729
DEARDORFF (1978) ‘Efficient prediction of ground surface temperature and moisture with
inclusion of a layer of vegetation’ J. Geophysical Research Volume 83 pp. 1889-1903
References
84
DEFRIES ET AL. (2002) ‘Carbon emissions from tropical deforestation and regrowth based
on satellite observations for the 1980’s and 1990’s’ Proceedings of the National Academy of
Sciences of the United States of America. Volume 99 Issue 22, pp. 14256-14261
DEUBECHIES ET AL. (1992) ‘Ten lectures on wavelets’ Book. CBMS-NSF Regional
Conference Series in Applied Mathematics. doi:
http://dx.doi.org/10.1137/1.9781611970104.fm
DI LORENZO ET AL. (2008) ‘North Pacific Gyre Oscillation links ocean climate and
ecosystem change’ Geophysical Research Letters Volume 35 L08607
doi:10.1029/2007GL02838
DONOHUE ET AL. (2013) ‘Impact of CO2 fertilisation on maximum foliage cover across
the globe’s warm, arid environments’ Geophysical Research Letters Volume 40 pp. 3031-
3035
- ET AL. (2014) ‘Evaluation of the remote-sensing-based DIFFUSE model for
estimating photosynthesis of vegetation’ J. Remote sensing of Environment (Article in press)
DROSDOWSKY (1996) ‘Variability of the Australian Summer Monsoon at Darwin: 1957-
1992’ Bureau of Meteorology Research Centre, Melbourne. Journal of Climate. Volume 9,
Issue 1 pp. 85-96
EAMUS & DUFF (1997a) ‘Seasonal and diurnal patterns of carbon assimilation, stomatal
conductance and leaf water potential in Eucalyptus tetrodonta samplings in a wet-dry
savanna in northern Australia. Aust. J. Botany Volume 45 pp. 241-258
- ET AL. (2001) ‘Daily and seasonal patterns of carbon and water fluxes above a
north Australian savanna’ J. Tree Physiology Volume 21 Issue 12-13 pp. 977-988
- ET AL. (2013) ‘Global change-type drought-induced tree mortality: vapour
pressure deficit is more important than temperature per se in causing decline in tree health’
J. Ecology and Evolution. doi:10.1002/ece3.664
EASTERLING ET AL. (2000) ‘Climate extremes: Observations, modelling, and impacts’
Science Volume 289 pp. 2068-2074
References
85
EGAN & WILLIAMS (1996) ‘Life-form distributions of woodland plant species along a
moisture availability gradient in Australia’s monsoonal tropics’ J. Australian Systematic
Botany Volume 9 Issue 2 pp. 205-217
ENGLAND ET AL. (2006) ‘Inter-annual rainfall extremes over Southwest Western Australia
linked to Indian Ocean climate variability’ J. Climate Volume 19 Issue 10 pp. 1948-1969
FAY ET AL. (2003) ‘Productivity responses to altered rainfall patterns in a C4 dominated
grassland’ J. Oecologia Volume 137 Issue 2 pp. 245-251
FERREIRA & HUETE (2013) ‘Assessing the seasonal dynamics of the Brazilian Cerrado
vegetation through the use of spectral vegetation indices’ Int. J. Remote Sensing Volume 25
Issue 10 pp. 1837-1860
FOLLAND ET AL. (2002) ‘Relative influences of the Inter-decadal Pacific Oscillation and
ENSO on the South Pacific Convergence Zone’ Geophysical Research Letters Volume 29
Issue 13 pp. 211-214
FOX ET AL. (2001) ‘The Vegetation of the Australian Tropical Savannas’ Environmental
Protection Agency, Brisbane
FRANKS (2002) ‘Assessing hydrological change: Deterministic general circulation models
or spurious solar correlation?’ J. Hydrological Processes Volume 16 Issue 2 pp. 559-564
- (2004) ‘Multi-decadal variability of drought risk, eastern Australia’ J. Hydrological
Processes Volume 18 Issue 18 pp. 2039-2050
FREDERIKSEN & BALGOVIND (1994) ‘Multi-decadal simulations of global climate
trends and variability’ Proceedings of the International Scientific Conference on the Tropical
Ocean Global Atmosphere Programme, Melbourne, Australia, April 1995.
FROST ET AL. (1986) ‘Responses of savannas to stress and disturbance’ Biology
International, Volume 10, pp. 1-82
FUHRER (2003) ‘Agro-ecosystem responses to combinations of elevated CO2, ozone and
global climate change’ J. Agriculture, Ecosystems & Environment Volume 97 Issue 1-3
pp.1-20
GABOR (1946) ‘Theory of communication’ J. Electrical Engineering Volume 93 pp. 429-457
References
86
GASH ET AL. (1996) ‘Amazonian Deforestation and Climate’ Wiley: Chichester
GRENFELL ET AL. (2001) ‘Travelling waves and spatial hierarchies in measles epidemics’
Nature. Volume 414 pp. 716-723
GRINSTED ET AL. (2004) ‘Application of the cross wavelet transform and wavelet
coherence to geophysical time series’ J. Nonlinear processes in Geography Volume 11 pp.
561-566
GORNALL ET AL. (2009) ‘The influence of land use change, CO2 fertilisation and climate
change on the European carbon budget’ Climate Change: Global Risks, Challenges and
Decisions. IOP Conf. Series: Earth and Environmental Science Volume 6
GOTTSCHALCK ET AL. (2010) ‘Overview of the Madden-Julian Oscillation’ National
Oceanic and Atmospheric Administration. Reviewed Online HARTMANN ET AL. (2013)
‘Observations: Atmosphere and Surface’ IPCC 5th
Assessment Report, Ch. 2. AR5
Contribution of Working Group 1 to the Fifth Assessment Report of the Intergovernmental
Panel of Climate Change 2013 Cambridge University Press
GU & PHILANDER (1995) ‘Secular changes of annual and inter-annual variability in the
tropics during the past century’ J. Climate Volume 8 pp. 864-876
HENDON (2004) ‘The Australian Summer Monsoon’ Bureau of Meteorology Research
Centre, Melbourne. Sourced from Academia.edu with contributions by McBride, J. &
Wheeler, M. <https://www.academia.edu/2605553/The_Australian_Summer_Monsoon>
Accessed 18/03/2014
- & LIEBMANN (1990a) ‘A Composite study of onset of the Australian summer
monsoon’ J. Atmospheric Sciences Volume 47 Issue 18 pp. 2227-2240
HIGGENS ET AL. (2011) ‘Is there a temporal niche separation in the leaf phenology of
savanna trees and grasses?’ J. Biogeography Volume 38 Issue 1 pp. 2165-2175
HOLLAND ET AL. (1986) ‘The BMRC Australian monsoon experiment: AMEX’ American
Meteorological Society Volume 67 Issue 12
HONG & KIM (2011) ‘Impact of the Asian monsoon climate on ecosystem carbon and water
exchanges: a wavelet analysis and its ecosystem modelling implications’ J. Global Change
Biology Volume 17 pp. 1900-1916
References
87
HOUSE & HALL (2001) ‘Productivity of Tropical Savannas and Grasslands’ In Roy, J.,
Saugier, B. & Mooney, H., editors, Terrestrial Global Productivity, San Diego, CA:
Academic Press, pp. 363-400
HUGHEN ET AL. (2004) ‘Abrupt Tropical Vegetation Response to Rapid Climate Changes’
Reports, Journal of Science Volume 304 pp. 1955-1958
HUGHES ET AL. (2003) ‘Climate change, human impacts and the resilience of coral reefs’
Science Volume 15 No. 5635 pp. 929-933
HUTLEY & BERINGER (2011) ‘Disturbance and Climatic Drivers of Carbon Dynamics of
a North Australian Tropical Savanna’ Ecosystem Function in Savannas, Measurement and
Modeling at Landscape to Global Scales. CRC Press
- ET AL. (2011) ‘A sub-continental scale living laboratory: Spatial patterns of
savanna vegetation over a rainfall gradient in northern Australia’ J. Agricultural and Forest
Meteorology Volume 151 Issue 11 pp. 1417-1428
- ET AL. (2013) ‘Impacts of an extreme cyclone event on landscape-scale savanna
fire, productivity and greenhouse gas emissions’ Environmental Research Letters Volume 8
pp. 1-12
- & SETTERFIELD (2008) ‘Savanna’ Ecosystems Volume 4 of Encyclopedia of
Ecology Volume 1-5 pp. 3143-Oxford: Elsevier
HUXMAN ET AL. (2004) ‘Precipitation pulses and carbon fluxes in semiarid and arid
ecosystems’ J. Oecologia Volume 141 pp. 254-268
KANNIAH ET AL. (2010) ‘The comparative role of key environmental factors in
determining savanna productivity and carbon fluxes: A review, with special reference to
northern Australia’ Progress in Physical Geography, Volume 34, No. 4, pp. 459-490
- (2011) ‘Environmental Controls on the Spatial Variability of Savanna Productivity
in the Northern Territory, Australia’ Agricultural and Forest Meteorology,
doi:10.1016/j.agrfomet.2011.06.009
- (2012) ‘Control of Atmospheric Particles on Diffuse Radiation and Terrestrial Plant
Productivity: A Review’ Progress in Physical Geography, Volume 36, No. 2, pp. 209-237
References
88
- (2013) ‘Exploring the link between clouds, radiation and canopy productivity of
tropical savannas’ Agricultural and Forest Meteorology, Volume 182, pp. 304-313.
KARL & KNIGHT (1998) ‘Secular trends of precipitation amount, frequency and intensity
in the united states’ Bull. American Meteorological Society Volume 79 Issue 2 pp. 231-241
KELLEY & HARRISON (2014) ‘Enhanced Australian carbon sink despite increased
wildfire during the 21st century’ Environmental Research Letters Volume 9
doi:10.1088/1748-9326/9/10/104015
KEP (1984) ‘A climatology of Cyclogenesis. Cyclone tracks and cyclolysis in the Southern
Hemisphere for the period 1972-1981’ Publication No. 25 Dept. of Meteorology, University
of Melbourne
KIEM ET AL. (2003) ‘Multi-decadal variability of flood risk’ Geophysical Research Letters
Volume 30 Issue 2 doi:10.1029/2002GL015992
KIRSCHBAUM (1995) ‘The temperature dependence of soil organic matter decomposition,
and the effect of global warming on soil organic C storage’ J. Soil Biology and Biochemistry
Volume 27 Issue 6 pp. 753-760
KLVANA ET AL. (2008) ‘Czech and Slovak bird migration atlas’ Prague: Aventinum J. eds
KNAPP ET AL. (2008) ‘Consequences of more extreme precipitation regimes for terrestrial
ecosystems’ J. BioScience Volume 58 Issue 9 pp. 811-821
KOWALCZYK ET AL. (2013) ‘The land surface model component of ACCESS: Description
and impact on the simulated surface climatology’ Aust. J. Meteorology and Oceanography
Volume 82 pp. 63-65
KUMAR ET AL. (1999) ‘On the weakening relationship between the Indian Monsoon and
ENSO’ Science Volume 284 No. 5423 pp. 2156-2159
LAU & SHEU (1988) ‘Annual cycle, quasi-biennial oscillation and southern oscillation in
global precipitation’ J. Geophysical Research: Atmospheres Volume 93 Issue D9 pp. 10975-
10988
LEGG & LONG (1975) ‘Turbulent diffusion within a wheat canopy, II: Results and
interpretation’ Quart. J. Royal Meteorological Society Volume 101 pp. 611-628
References
89
LEHMANN ET AL. (2014) ‘Savanna Vegetation-Fire-Climate Relationships Differ Among
Continents’ Science Volume 343 pp. 548-552
LEWIS ET AL. (2009) ‘Increasing carbon storage in intact African tropical forests’ Letter.
Nature Volume 457 pp. 1003-1006
LIM & RODERICK (2009) ‘An atlas of the global water cycle: Based on the IPCC AR4
climate models’ ANU E Press, Canberra Aus.
LYNCH ET AL. (2007) ‘Influence of savanna fire on Australian monsoon season
precipitation and circulation as simulated using a distributed computing environment’
Geophysical Research Letters Volume 34 Issue 20 pp. 1-5
MA ET AL. (2013) ‘Spatial patterns and temporal dynamics in savanna vegetation
phenology across the North Australian Tropical Transect’ J. Remote Sensing of the
Environment Volume 130 pp. 97-115
MAC CRACKEN ET AL. (2003) ‘Climate change scenarios for the U.S. National
Assessment’ Bulletin American Meteorological Society Volume 84 pp. 1711-1723
MCBRIDE &WHEELER (2004) ‘How strong is the relationship between the MJO and the
onset of the Australian Monsoon?’ Bureau of Meteorology Research Centre, Melbourne.
UCLA Tropical Meteorology Newsletter No. 64
MCPHERSON ET AL. (1994) ‘Chicago’s urban forest ecosystem: Results of the Chicago
Urban Forest Climate Project’ Gen. Tech. Rep. NE-186, Radnor PA. U.S. Dept. of
Agriculture, Forest Service, North-eastern Forest Experiment Station: 201 p.
MEEHL ET AL. (2000) ‘Trends in Extreme Weather and Climate Events: Issues Related to
Modeling Extremes in Projections of Future Climate Change’ Bulletin of the American
Meteorological Society. Volume 81, No. 3, pp. 427-436
MONCRIEFF ET AL. (2013) ‘Increasing atmospheric CO2 overrides the historical legacy of
multiple stable biome states in Africa’ J. New Phytologist Volume 201 pp. 908-915
MONTEITH & SZEICZ (1962) ‘Radiative temperature in the heat balance of natural
surfaces’ Quart. J. Royal Meteorological Society Volume 88 pp. 496-507
References
90
MOORCROFT (2003) ‘Recent advances in ecosystem-atmosphere interactions: An
ecological perspective’ Review Paper, Department of Organismic and Evolutionary Biology,
Harvard University, MA. The Royal Society Volume 270, pp. 1215-1227
MORLET (1983) ‘Sampling theory and wave propagation’ Issues in Acoustic Signal/ Image
Processing and Recognition. Springer-Verlag Berlin
MURPHY & TIMBAL (2008) ‘A review of recent climate variability and climate change in
south-eastern Australia’ Int. J. Climatology Volume 28 pp. 859-879
MUTHUSWAMY & THAKOR (1998) ‘Spectral analysis methods for neurological signals’
J. Neuroscience Methods Volume 83 Issue 1 pp. 1-14
MYERS ET AL. 1997 ‘Seasonal Variation in Water Relations of Trees of Differing Leaf
Phenology in a Wet-Dry Tropical Savanna near Darwin, Northern Australia’ Aust. J. of
Botany Volume 45 Issue 2 pp. 225-240
NASON & SILVERMAN (1995) ‘The stationary wavelet transform and some statistical
applications’ J. Lecture notes in Statistics Volume 103 pp. 281-299
NEMANI ET AL. (2003) ‘Climate-driven increases in global terrestrial Net Primary
Productivity from 1982-1999’ Science Volume 300 pp. 1560-1563
NEW 2001) ‘The weather and climate of southern Africa’ J. Progress in Physical Geography
Volume 25 Issue 4 pp. 586
NICHOLLS ET AL. (1982) ‘On Predicting the Onset of the Australian Wet Season at
Darwin’ Australian Numerical Meteorology Research Centre, Melbourne. Monthly Weather
Review, American Meteorology Society. Volume 110, pp. 14-17
- (1989) ‘Sea surface temperatures and Australian winter rainfall’ J. Climate Volume
2 pp. 965-973
- ET AL. (1996) ‘Recent apparent changes in relationships between the El Nino
Southern Oscillation and Australian rainfall and temperature’ Geophysical Research Letters
Volume 23 Issue 23 pp. 3357-3360
References
91
O’GRADY ET AL. (1999) ‘Transpiration increases during the dry season: Patterns of tree
water use in eucalypt open-forests of northern Australia’ J. Tree Physiology Volume 19 Issue
9 pp. 591-597
O’MAHONY (1961) ‘Time series analysis of some Australian rainfall data’ Bureau of
Meteorology
OVERLAND ET AL. (2010) ‘Climate controls on marine ecosystems and fish populations’
J. Marine Systems Volume 79 Issue 3-4 pp. 305-315
QUINN ET AL. (1987) ‘El Nino occurrences over the past four and a half centuries’ J.
Geophysical Research: Oceans Volume 92 Issue C13 pp. 14449-14461
PAN ET AL. (2011) ‘A large and persistent carbon sink in the world’s forests’ Science.
Volume 333, 988
PARR ET AL. (2012) ‘Cascading biodiversity and functional consequences of a global
change-induced biome switch’ J. Diversity and Distributions Volume 18 Issue 5 pp. 493-503
PETERSON & SCHWING (2003) ‘A new climate regime in northeast Pacific ecosystems’
Geophysical Research Letters Volume 30 Issue 17 doi:10.1029/2003GL017528
PETHERAM ET AL.(2012) ‘Estimating the impact of projected climate change on runoff
across the tropical savannas and semiarid rangelands of northern Australia.’ American
Meteorological Society doi:10.1175/JHM-D-11-062.1
PHILLIPS & LEWIS (2013) ‘Evaluating the tropical forest carbon sink’ J. Global Change
Biology Volume 20 Issue 7 pp. 2039-2041
PIELKE ET AL. (1998) ‘Interactions between the atmosphere and terrestrial ecosystems:
influence on weather and climate’ J. Global Change Biology Volume 4 Issue 5 pp. 461-475
PITMAN (2003) ‘The evolution of, and revolution in, land surface schemes designed for
climate models’ Review. Int. J. Climatology Volume 23 pp. 479-510
PITMAN ET AL. (2004) ‘Impact of land cover change on the climate of southwest Western
Australia’ J. Geophysical Research: Atmospheres Volume 109 Issue D18 pp. 1-12
POST (2013) ‘Ecology of climate change: The importance of biotic interactions’ Book.
Princeton University Press. Science
References
92
POULTER ET AL. (2010) ‘Net biome production of the Amazon Basin in the 21st century’ J.
Global Change Biology Volume 16 pp. 2062-2075
- ET AL. (2014) ‘Contribution of semi-arid ecosystems to interannual variability of
the global carbon cycle’ Letter. Nature. doi: 10.1038/nature13376
POWER ET AL. (1999) ‘Decadal climate variability in Australia during the 20th
century’
Int. J. Climatology Volume 19 pp. 169-184
PRIESTLEY (1964) ‘Rainfall-sea surface temperature associations on the New South Wales
coast’ Australian Meteorological Magazine Issue 49 pp. 15-25
- & TROUP (1966) ‘Droughts and wet periods and their association with sea-surface
temperatures’ Aust. J. Science Volume 29 pp. 56-57
PRIOR ET AL. (1997) ‘Seasonal and Diurnal Patterns of Carbon Assimilation, Stomatal
Conductance and Leaf Water Potential in Eucalyptus tetrodonta Saplings in a Wet-Dry
Savanna in Northern Australia’ Aust. J. of Botany Volume 45 Issue 2 pp. 241-258
RATANA ET AL. (2005) ‘Analysis of cerrado physiognomies and conversion in the MODIS
seasonal-temporal domain’ J. Earth Interactions Volume 9 pp. 1-22
REICHSTEIN ET AL. (2007) ‘Ecosystem-level water-use efficiency inferred from eddy
covariance data: definitions, patterns and spatial up-scaling’ J. AGU Fall Meeting
Abstracts’ Volume 1 pp. 1-4
REMMERT (1991) ‘Theory of Complex Functions: Readings in Mathematics’ pp. 29
Springer, 1991. Accessed 17 July 2014.
RISBEY ET AL. (2009) ‘On the Remote Drivers of Rainfall Variability in Australia’ The
Centre for Australian Weather and Climate Research. Monthly Weather Review, American
Meteorological Society. Volume 137, Issue 10, pp. 3233-3253
ROSBURGH ET AL. (2004) ‘A critical overview of model estimates of net primary
productivity for the Australian continent’ J. Functional Plant Biology Volume 31 pp. 1043-
1059
References
93
RUNNING & HUNT (1993) ‘Generalisation of a Forest Ecosystem Process Model for other
biomes, BIOME-BGC, and an Application for Global-Scale Models’ Scaling Physiological
Processes Ch. 8 pp. 141-159
SAJI ET AL. (1999) ‘A dipole mode in the tropical Indian Ocean’ Nature Volume 401 pp.
360-363
SANKARAN ET AL. (2004) ‘Tree-grass Coexistence in Savannas Revisited: Insights from
an examination of assumptions and mechanisms invoked in existing models’ Ecology Letters,
Volume 7, pp. 480-490
- (2005) ‘Determinants of Woody Cover in African Savannas’ Journal of Nature,
Volume 438, pp. 846-849
- (2008) ‘Woody Cover in African Savannas, the role of resources, fire and herbivory’
Global Ecology and Biogeography, Volume 17, pp. 236-245
SARMIENTO (1996) ‘Biodiversity and Water Relations in tropical savannas’ In Solbrieg,
O., Medina, E. & Silva, J., editors. Biodiversity and Savanna Ecosystem Processes: A Global
Perspective. Ecological Studies 121, Berlin: Springer, 31-41
SCHEITER ET AL. (2014) ‘Climate change and long-term fire management impacts on
Australian savannas’ J. New Phytologist doi:10.1111/nph.13130
- & HIGGINS (2009) ‘Impacts of climate change on the vegetation of Africa: an
adaptive dynamic vegetation modelling approach’ J. Global Change Biology Volume 15 pp.
2224-2246
SCHEPEN ET AL. (2012) ‘Evidence for using lagged climate indices to forecast Australian
seasonal rainfall’ J. Climate Volume 25 Issue 4 pp. 1230-1246
SCHOLES & ARCHER (1997) ‘Tree-grass interactions in savannas’ Annual Review of
Ecology and Systematics, Volume 28, pp. 517-544
- & HALL (1996) ‘The Carbon Budget of Tropical Savannas, Woodlands and
Grasslands’ SCOPE Volume 56, pp. 69-100
- & WALKER (1993) ‘An African savanna – synthesis of the Nylsvley Study’
Cambridge University Press, Volume 306
References
94
SEAGER ET AL. (2013) ‘Does global warming cause intensified inter-annual hydro-climate
variability?’ J. Climate Volume 25 Issue 9 pp. 3355-3372
SELLERS ET AL. (1992) ‘An overview of the First International Satellite Land Surface
Climatology Project (ISLSCP) Field Experiment (FIFE)’ J. Geophysical Research:
Atmospheres Volume 97 Issue D17 pp. 18345-18371
SENEVIRATNE & STÖCKLI (2008) ‘The role of land-atmosphere interactions for climate
variability in Europe’ Climate Variability and Extremes during the past 100 years (Book) pp.
179-193 Springer 2008
SHORROCKS (2007) ‘The biology of African Savannahs’ Oxford University Press, Volume
268
SINGH & GUPTA (1977) ‘Plant decomposition and soil respiration in terrestrial
ecosystems’ The Botanical Review Volume 43 Issue 4 pp. 449-528
SMITH ET AL. (2002) ‘Representation of vegetation dynamics in the modelling of terrestrial
ecosystems: comparing two contrasting approaches within European climate space’ J.
Global Ecology and Biogeography Volume 10 Issue 6 pp. 621-637
- ET AL. (2008) ‘Characteristics of the Northern Australian Rainy Season’ CSIRO
Marine and Atmospheric Research. J. of Climate, Volume 21, pp. 4298-4311
SOLBRIEG. (1990) ‘Savanna modelling for global change’ Biology International Special
Issue 24. Paris International Union of Biological Sciences
- ET AL. (1996) ‘Determinants of Tropical Savannas’ In Solbrieg, O., Medina, E. &
Silva, J., editors. Biodiversity and Savanna Ecosystem Processes: A Global Perspective.
Ecological Studies 121, Berlin: Springer, 31-41
SOLOMON ET AL. (2007) ‘Observations: Atmospheric surface and climate change’ Ch. 3
AR4 Contribution of Working Group 1 to the Fourth Assessment Report of the
Intergovernmental Panel of Climate Change 2007 Cambridge University Press
STAHLE ET AL. (1993) ‘Southern Oscillation extremes reconstructed from tree rings of the
Sierra Madre Occidental and Southern Great Plains’ J. Climate Volume 6 pp. 129-140
References
95
STENSETH ET AL. (2003) ‘Studying climate effects on ecology through the use of climate
indices: The North Atlantic Oscillation, El Nino Southern Oscillation and beyond’ Review
Article. Proc. R. Soc. Lond B. Volume 270 No. 1529 pp. 2087-2096
STOY ET AL. (2009) ‘Biosphere-atmosphere exchange of CO2 in relation to climate: a
cross-biome analysis across multiple time scales’ J. Biogeosciences Volume 6 Issue 2 pp.
4095-4141
STRETEN (1981) ‘Southern hemisphere sea surface temperature variability and apparent
associations with Australian rainfall’ J. Geophysical Research: Oceans Volume 86 Issue C1
pp. 485-497
- (1983) ‘Extreme distributions of Australian annual rainfall in relation to sea surface
temperature’ J. Climatology Volume 3 Issue 2 pp. 143-153
SUPPIAH (1992) ‘The Australian summer monsoon: A review’ J. Progress in Physical
Geography Volume 16 No. 3 pp. 283-318
TANG & YU (2008) ‘MJO and its relationship to ENSO’ J. Geophysical Research:
Atmospheres Volume 113 Issue D14 doi: 10.1029/2007JD009230
THOM (1972) ‘Momentum, mass and heat exchange of vegetation’ Quart. J. Royal
Meteorological Society Volume 98 pp. 124-134
TIMMERMANN (2001) ‘Changes of ENSO stability due to greenhouse warming’
Geophysical Research Letters Volume 28 Issue 10 pp. 2061-2064
TORRENCE & COMPO (1998a) ‘A practical guide to wavelet analysis’ Bull. American
Meteorological Society Volume 79 pp. 61-78
TRAN (2005) ‘From Fourier transforms to wavelet analysis: Mathematical concepts and
examples’ Whitman, unpublished.
TRENBERTH & HOAR (1997) ‘El Nino and climate change’ Geophysical Research Letters
Volume 24 Issue 23 pp. 3057-3060
TREWIN ET AL. (2014) ‘State of the Tropics (SOTT) 2014 Report’ James Cook University
<http://stateofthetropics.org > Accessed August 2014
References
96
TROUP (1961) ‘Variations in the upper tropospheric flow associated with the onset of the
Australian summer monsoon’ Ind. J. Meteorology and Geophysics Volume 12 Issue 2 pp.
217-230
- (1965) ‘The Southern Oscillation’ Quart. J. Royal Meteorological Society Volume
91 Issue 390 pp. 490-506
VAN DER WERF ET AL. (2006) ‘Inter-annual variability in global biomass burning
emissions from 1997-2004’ J. Atmospheric Chemistry and Physics Volume 6 pp. 3423-3441
- ET AL. (2010) ‘Global fire emissions and the contribution of deforestation, savanna,
forest, agricultural and peat fires (1997-2009)’ J. Atmospheric Chemistry and Physics
Volume 10 pp. 11707-11735
VARGAS ET AL. (2010) ‘Multiscale analysis of temporal variability of soil CO2 production
as influenced by weather and vegetation’ J. Global Change Biology Volume 16 Issue 5 pp.
1589-1605
- ET AL. (2011) ‘On the multi-temporal correlation between photosynthesis and soil
CO2 efflux: reconciling lags and observations’ J. New Phytologist Volume 191 Issue 4 pp.
1007-1017
VERBURG ET AL. (2011) ‘Challenges in using land use and land cover data for global
change studies’ J. Global Change Biology Volume 17 Issue 2 pp. 974-989
VERDON & FRANKS (2005) ‘Indian Ocean sea surface temperature variability and winter
rainfall: Eastern Australia’ J. Water Resources Research Volume 41 Issue 9
doi:10.1029/20044WR003845
VILA-GUERAU DE ARELLANO ET AL. (2012) ‘Modelled suppression of boundary-layer
clouds by plants in a CO2-rich atmosphere’ Letter. J. Nature Geoscience Volume 5 pp. 701-
704
WALKER & MENAUT (1988) ‘Research Procedure and experimental design for Savanna
Ecology and management’ Report of a meeting of an IUBS Working Group Decade of the
Tropics Programme on Tropical Savanna Ecosystems.
WANG ET AL. (2003) ‘Atmosphere-warm ocean interaction and its impacts on Asian-
Australian monsoon variation’ J. Climate Volume 16 Issue 8 pp. 1195-1211
References
97
WENDT ET AL. (2007) ‘Local boundary layer development over burnt and unburnt tropical
savanna: an observational study’ J. Boundary Layer Meteorology Volume 124 pp. 291-304
WHETTON (1988) ‘A synoptic climatological analysis of rainfall variability in south-eastern
Australia’ J. Climatology Volume 8 Issue 2 pp. 155-177
WHITLEY ET AL. (2011) ‘Modelling productivity and water-use across five years in a
mixed C3 and C4 savanna using a soil-plant-atmosphere model: GPP is light limited not
water limited’ J. Global Change Biology Volume 17 pp. 3130-3149
WILL ET AL. (2013) ‘Increased vapour pressure deficit due to higher temperature leads to
greater transpiration and faster mortality during drought for tree seedlings common to the
forest-grassland ecotone’ Oklahoma State University. New Phytologist Volume 200, Issue 2,
pp. 366-374
WILLIAMS ET AL. (1997) ‘Leaf phenology of woody species in a north Australian tropical
savanna’ J. Ecology Volume 78 pp. 2542-2558
WOINARSKI ET AL. (2007) ‘The nature of northern Australia: natural values, ecological
processes and future prospects’ ANU E Press
WRIGHT (2013) ‘The carbon sink in intact tropical forests’ Letter. J. Global Change
Biology Volume 19 pp. 337-339
YI ET AL. (2010) ‘Climate control of terrestrial carbon exchange across biomes and
continents’ Environmental Research Letters Volume 5 Issue 3 L34007
ZHANG (2005) ‘Madden-Julian Oscillation’ Rosentiel School of Marine and Atmospheric
Science, University of Miami. Review of Geophysics Volume 43, Is. 2, pp. 1-36
ZHANG ET AL. (2003) ‘Monitoring vegetation phenology using MODIS’ Department of
Geography and Center for Remote Sensing, BU. ELSEVIER. Remote Sensing of
Environment. Volume 84, pp.471-475
Appendices
98
Appendices
Appendix A: Climate Indices with high correlation to annual total rainfall over Australian
continent
Figure A.1: Australian map of climate indices with highest correlation with annual total rainfall from 1900-2010. Resolution is
125km2. Rainfall at Howard Springs is shown to correlate with the Tasman Sea Index and Indonesian Index. (Source:
Cassandra Rogers 2012)
Appendices
99
Appendix B: Howard Springs Flux Tower Equipment
Table A.1. List of instruments and location along the tower at the Howard Springs Flux site, collecting eddy-
covariance data capturing the land-atmosphere exchanges at a savanna site.
INSTRUMENT TYPE Make Model Situation
Open path CO2, H2O LI-COR LI-7500 23m
Sonic anemometer
- wind velocities (u,v,w)
- sonic temperature
Campbell Scientific CSAT-3 23m
Solar Radiation (incoming) Kipp and Zonen CM-7B 23m
Solar Radiation (reflected) Kipp and Zonen CG-2 23m
Atmospheric Pressure Li-Cor LI-7500 23m
Soil Heat Flux (2 replicates) Campbell Scientific HFT3 Ground, -15 cm
Soil moisture Campbell Scientific CS616 10-100cm
Temperature and Relative Humidity Vaisala HMP45A 2m, 23m
Rain Gauge Campbell Scientific TB3 Ground
Data Loggers Campbell Scientific CR-3000
Power Supply
- 12V DC EC flux station Kyocera Solar panels (3) (BP) Ground
(Source:
http://www.ozflux.org.au/monitoringsites/howardsprings/howardsprings_measurements.html)
Appendices
100
Appendix C: Slope changes of annual mean carbon fluxes early vs. late century
y = 0.0021x + 19.775 18
19
20
21
22
19
13
19
16
19
19
19
22
19
25
19
28
19
31
19
34
19
37
19
40
19
43
19
46
19
49
GP
P (
t C
ha-1
y-1
)
a)
y = 0.014x + 20.193
18
19
20
21
22
19
51
19
56
19
61
19
66
19
71
19
76
19
81
19
86
19
91
19
96
20
01
20
06
20
11
GP
P (
t C
ha-1
y-1
)
b)
y = -0.0075x + 16.831
15.5
16.0
16.5
17.0
17.5
19
13
19
16
19
19
19
22
19
25
19
28
19
31
19
34
19
37
19
40
19
43
19
46
19
49
Fre
(t
C h
a-1 y
-1)
c)
y = 0.0038x + 16.807 16.0
16.5
17.0
17.5
18.0
19
51
19
56
19
61
19
66
19
71
19
76
19
81
19
86
19
91
19
96
20
01
20
06
20
11
Fre
(t
C h
a-1 y
-1)
d)
y = 0.0125x + 5.6466 4.0
5.0
6.0
7.0
8.0
19
13
19
16
19
19
19
22
19
25
19
28
19
31
19
34
19
37
19
40
19
43
19
46
19
49
Fc (
t C
ha
-1 y
-1)
e)
y = 0.0153x + 6.0846 5.0
6.0
7.0
8.0
9.0
19
51
19
56
19
61
19
66
19
71
19
76
19
81
19
86
19
91
19
96
20
01
20
06
20
11
Fc (
t C
ha-1
y-1
)
f)
Figure A.2: Time series and linear trends of carbon flux variables showing the variation in rate of increase between the first and
second halves of the 20th
century. (a,b) gross primary productivity, (c,d) ecosystem respiration and (e,f) net ecosystem exchange.
Appendices
101
Appendix D: Time series (anomaly plots) of climate indices
-40
-20
0
20
40
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
MSL
P S
tan
dar
d D
evia
tio
n
SOI
-0.0175
-0.0075
0.0025
0.0125
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
SST
An
om
aly
TSI
-0.0175
-0.0075
0.0025
0.0125
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010
SST
An
om
aly
II
Figure A.3: Time series of climate indices. SOI appears to have approximately equal distributions of positive and negative (wet
and dry) events while TSI and II show marked increasing in frequency of positive (warm) SST anomalies during the last quarter of
the time period.
Appendices
102
Appendix E: Rainfall coherence spectra and ecosystem fluxes
Figure A.4: WTC plots demonstrating lack of coherence beyond the dominating annual cycle of rainfall and
ecosystem flux variables. (a) net ecosystem exchange, (b) latent heat flux, (c) ecosystem respiration and (d) gross
primary productivity.