Influence of savanna fire on Australian monsoon season precipitation
and circulation as simulated using a distributed computing
Amanda H. Lynch,1David Abramson,2Klaus Go ¨rgen,3Jason Beringer,1
and Petteri Uotila1
Received 3 June 2007; revised 23 July 2007; accepted 17 August 2007; published 16 October 2007.
to affect monsoon evolution, but the hypothesis is
controversial and the effects have not been quantified. A
distributed computing approach allows the development of
a challenging experimental design that permits simultaneous
variation of all fire attributes. The climate model
simulations are distributed around multiple independent
computer clusters in six countries, an approach that has
potential for a range of other large simulation applications in
the earth sciences. The experiment clarifies that savanna
burning can shape the monsoon through two mechanisms.
Boundary-layer circulation and large-scale convergence is
intensified monotonically through increasing fire intensity
and area burned. However, thresholds of fire timing and
area are evident in the consequent influence on monsoon
rainfall. In the optimal band of late, high intensity fires with
a somewhat limited extent, it is possible for the wet season
to be significantly enhanced. Citation: Lynch, A. H., D.
Abramson, K. Go ¨rgen, J. Beringer, and P. Uotila (2007),
Influence of savanna fire on Australian monsoon season
precipitation and circulation as simulated using a distributed
computing environment, Geophys. Res. Lett., 34, L20801,
Fires in the Australian savanna have been hypothesized
 Dry season fires are one of the largest natural and
anthropogenic disturbances in the tropical savannas. Be-
tween 40 to 75% of global savannas were burned annually
between 1975 and 1980 [Hao et al., 1990]. Savanna fires
consume three times as much dry matter annually as the
burning of tropical forests (e.g., 3690 Tg dm yr?1for
savanna versus 1260 Tg dm yr?1for tropical forests
[Andreae, 1991]) although recovery of the biomass is an
order of magnitude more rapid. The Australian tropical
savanna is one of the world’s largest, with vast tracts
affected by fire each year, particularly in deliberate burning
by pastoralists, Aboriginal land holders and conservation
managers – an estimated 244,000 km2of the total area of
northern Australia was affected in 1997 [Russell-Smith et
 Fires across this landscape have impacts on the
regional water, energy and carbon dioxide exchanges
[Beringer et al., 2003] through altered vegetation composi-
tion and reduced surface albedo, increased available energy
for partitioning into the convective fluxes, and increased
substrate heat flux into the soil. In addition, the aerody-
namic and biological properties of the ecosystem can
change, affecting surface-atmosphere interaction [Beringer
et al., 2007]. Local tethered balloon profile measurements
[Wendt et al., 2007] and idealized computer sensitivity
studies [Go ¨rgen et al., 2006; Pitman and Hesse, 2006] have
demonstrated intensified boundary layer processes in fire-
affected areas and have led some to suggest impacts
extending to the regional scale [Miller et al., 2005]. How-
ever, these impacts, and their potential for the modification
of the Australian monsoon regime, have not been system-
atically quantified. Here we address that quantification
using a comprehensive experimental design made possible
through the use of software that allows the implementation
of legacy code on a global computational grid.
 Quantifying the impacts of fire in savanna regions is
challenging because fire and its effects span many temporal
and spatial scales. Moreover, there is high interannual
variability. Hence, this question requires the application of
a high resolution climate model which allows the represen-
tation of the effects of savanna fire across a range of
manifestations through multiple realizations of the model.
 We represented savanna fires in the Conformal-Cubic
Atmospheric Model (C-CAM) from the Australian Com-
monwealth Scientific and Industrial Research Organization
(CSIRO), which is a global hydrostatic atmospheric model
coupled to a land surface model with a variable-resolution
conformal-cubic grid [McGregor and Dix, 2001]. This grid
provides a resolution of approximately 65 km over Australia
extending to about 800 km on the far side of the globe. Far-
field atmospheric nudging is achieved using the National
Center for Environmental Prediction/National Center for
Atmospheric Research (NCEP/NCAR) reanalysis data
[Kalnay et al., 1996] with an e-folding time of 96 hours,
part of the model domain [Go ¨rgen et al., 2006].
 A fire/re-growth scheme has been implemented in the
existing C-CAM land surface model. The surface properties
modified by fire and the subsequent vegetation re-growth
GEOPHYSICAL RESEARCH LETTERS, VOL. 34, L20801, doi:10.1029/2007GL030879, 2007
1Climate Program, Monash Sustainability Institute and School of
Geography and Environmental Science, Monash University, Clayton,
2Clayton School of Information Technology, Monash University,
Clayton, Victoria, Australia.
3De ´partement Environnement et Agro-Biotechnologies, Centre de
Recherche Public–Gabriel Lippmann, Belvaux, Luxembourg.
Copyright 2007 by the American Geophysical Union.
1 of 5
are defined in the model by four parameters which are based
on the observed fire regime [Russell-Smith et al., 2003].
First, the intensity of the fire in any one grid cell ranges
from 10% to 100% of the maximum possible biomass that
can be destroyed, on the basis of fire intensities associated
with an energy release between 103and 104kWm?1
[Chafer et al., 2004]. Secondly, the spatial distribution of
fire-affected grid cells is defined within a range from 10% to
100% of the maximum possible area, based on the observed
area burnt between 1997 and 2001. Thirdly, the temporal
range for the timing of the fire event lies between 1 May
and 30 November, which is followed a few weeks after by
the average monsoon onset date. Finally, the length of the re-
growth period is allowed to vary between 35 and 140 days.
 In order to quantify the relative importance of each of
the four forcing parameters (area, intensity, timing and
length of regrowth), a multifactorial approach that allows
the analysis of both the impacts of individual forcings and
the interactions between forcings is required. Rivers and
Lynch  developed just such a method using a factorial
experimental strategy based on the work of Henderson-
Sellers  for assessing the effects of the changing land
surface on early Holocene climate in Beringia using a series
of multi-year ensemble perturbation experiments that varied
all forcing parameters simultaneously. This approach has
been adopted here by varying all four forcing parameters
simultaneously using a Latin hypercube sampling to define
90 individual experiments, each of 21 years duration (1979–
1999) and each having a 5 year fire-free spin up period.
 Using a single sequential high end workstation would
require more than 10 years of wall clock time to complete
this experiment. Clearly this is unacceptable. However, the
individual realizations can be executed in parallel and it is
possible to reduce the execution time substantially using
this technique. Rather than using a single 90 processor
cluster (which was not available), we distributed the com-
putations across multiple independent clusters located in
Australia, Japan, Korea, Taiwan, Thailand and US. These
have been provided as part of a multi-country Grid com-
puting initiative called PRAGMA (the Pacific Rim Grid and
 Running such a large scale experiment over distrib-
uted resources is very time consuming and error prone
without adequate software support. In response, Abramson
et al.  have developed a software tool called Nimrod/
G, which automates the execution of parameter sweep and
search applications, such as the design developed here, over
global computational Grids. Nimrod/G manages the com-
plexity of running the experiment, handles network and
resource failures, and gives the appearance of a single
 Our experiment ran for an elapsed time of 170 days
– significantly less than 10 years – with an average of
37 processors in use across that period (Figure 1). Of the
90 original experiments, 85 executed successfully, and 5
were terminated due to instabilities in the model.
3.Response of the Monsoon
 The area-total monsoon precipitation over northern
Australia responds, with a statistically significant increase,
to increases in the fire intensity, the lateness of the timing
and the area burned in the preceding dry season. The
maximum increase of the simulations performed, due to
the combination of all four parameters, is 1.6 mm day?1
over an average precipitation rate of 5.1 mm day?1(a 31%
increase in average monsoon season precipitation)
(Figure 2a). This maximal response occurs in the simulation
that specifies a fire intensity of 72%, a fire area of 90%, and
a date of fire of the 5th November, around 6 weeks before
typical monsoon onset [Hendon and Liebmann, 1990]. The
length of the regrowth period has a limited impact on
monsoon precipitation, accounting for only around 9% of
the total variance in monsoon precipitation. In contrast, the
area accounts for almost 18% of the variance, and the fire
timing accounts for almost 58% of the variance in monsoon
Figure 1. Number of jobs executed by computational resources in each country participating in the experiment over the
PRAGMA grid (www.pragma-grid.org).
LYNCH ET AL.: GRID MODELING OF FIRE EFFECTS ON MONSOON
2 of 5
 Results from the 85 multiple realizations allowed the
development of a statistical model [Lynch et al., 2001] of
the response metrics (such as monsoon onset and intensity)
to variations in the four driving parameters. The statistical
model allows inter- and extra-polation from the simulations
that were successfully performed to postulate the maximal
response. This model suggests that the largest increase in
monsoon rainfall in response to previous dry season fires is
dependent upon a maximum possible fire intensity occur-
ring as late as possible during the dry season, but does not
support the largest possible fire area. This is due to the need
for sufficient vegetated area to remain to boost the moisture
available for convective activity. This limit occurs at around
90% of the total possible burned area.
 The interaction between monsoon season precipita-
tion, fire intensity and the timing of the fire is of particular
interest to decision-makers. High intensity fires that burn
both the over-storey and the undergrowth in the early dry
season (up to around mid-July) have limited impact on
monsoon precipitation because atmospheric moisture is
limited (Figure 2a). Conversely, late dry season fires have
a small influence on precipitation unless the fire is also very
intense. Fire area and timing also interact such that a large
fire area has effectively no impact during early dry season
fires, but fires of any area between around 40% and 90%
had a strong influence on monsoon precipitation if they
occurred during the late dry season. The relationship be-
tween monsoon precipitation and fire area shows less
variability than that between monsoon precipitation and fire
intensity. The variability in monsoon rainfall is due specif-
ically to variations in uplift intensity (see Figure 2b,
discussed below) rather than moisture availability.
 The impact of dry season fire intensity on monsoon
season precipitation is due to the interaction between the
generation of uplift due to surface heating and the avail-
ability of moisture. Fires result in decreased albedo, in-
creased net radiation, increased Bowen ratio (surface
heating) and enhanced vertical uplift. The area over which
enhanced vertical motion exceeds a threshold value of
1.5 Pa s?1(accumulated over three model levels in the upper
boundary layer) increases with both intensity (Figure 2b)
and area of fire. The influence of the timing of the fire on
vertical motion is rather small, as the intensification of
boundary layer processes is not driven by moist processes.
However, the timing of the fire becomes important in the
simulation of moist parameters, including latent heat fluxes,
precipitable water in the tropospheric column, and convec-
tive cloud (Figure 2c). There is insufficient available mois-
ture for fires to influence these variables in the early dry
season, but during monsoon buildup a combination of
Figure 2. Simulated differences (scenario minus reference
simulation) in response to variation in fire intensity and the
date of fire onset. Shown are differences in (a) average
precipitation [mm d?1], (b) vertically averaged boundary
layer uplift [Pa s?1], spatially averaged over those grid
elements that satisfy a threshold criterion (below 1.5 Pa s?1),
and (c) vertically integrated precipitable water [mm]. The
differences of the spatial averages over Australia’s land area
north of Tropic of Capricorn are averaged over the first
60% of the regrowth period over the 20 years of each
experiment. The crosses indicate the population of the
simulation space. The contoured response metric is then
Gaussian low pass filtered linear interpolations based on
LYNCH ET AL.: GRID MODELING OF FIRE EFFECTS ON MONSOON
3 of 5
enhanced convective activity with increased atmospheric
moisture leads to both an advancement of monsoon onset
and increased precipitation intensity, and hence an increase
in total monsoon season precipitation.
 Given the large scale influence of fire on surface
energy balance, albedo, vertical motion and moist process-
es, we hypothesized that the simulated changes may have an
impact on regional scale monsoon circulation patterns such
as that postulated by Johnson et al. . To test this we
used an indicator of large-scale circulation changes called
the Australian Monsoon Index (AUSMI) [Wang et al.,
2004], which is obtained by averaging the daily mean
850 hPa zonal wind speed from the equator to 10?S and
from 120?E to 150?E. Earlier studies [Go ¨rgen et al., 2006]
using single fire realizations have found that fires have a
limited impact, if any, on the large scale development of the
monsoon regime, with responses that are temporally and
spatially restricted to fire-affected areas. However, in this
more complete experimental design, we have found a
significant influence of fire intensity and area on the
AUSMI when consideration is restricted to late dry season
fires (Figure 3). For example, the maximum value of
AUSMI is 0.43 ms?1for a high intensity, late dry season
fire, compared to the average of all scenarios of 0.03 ms?1.
As for the local impacts, the moisture availability later in the
year is a requirement for a statistically significant perturba-
tion to the monsoon circulation. In contrast, fires that occur
before September have no discernable influence on the large
scale monsoon circulation. Thus, across the experiments
performed, fire timing accounts for almost 85% of the
variance in AUSMI. Nevertheless, changes in both fire
intensity and area burned account for 6% of the variance
respectively, and statistical modeling indicates that the area
threshold noted for maximum monsoon rainfall impact is
not evident in the influence on circulation.
 These responses are certainly model dependent, due
to variations in sensitivity amongst climate models, the
impacts of grid resolution and parameterization choice,
and the possible constraining effect of far field nudging.
However, the technical innovation that supports this exper-
imental design is highly scalable, and hence an extension of
this work will be to consider multi-model ensembles. In this
context, these experiments suggest that savanna burning can
shape monsoon response through two mechanisms. The
first, and primary, mechanism is the influence on the
intensity of surface heating, which has a direct effect on
the boundary layer circulation and thence large scale con-
vergence. The secondary mechanism is the influence of
these changes on convection and monsoon precipitation –
evident is a threshold of fire timing on the one hand and fire
area on the other. Under the appropriate condition of
sufficient available moisture combined with enhanced con-
vergence and uplift, the subsequent wet season is indeed
lian Research Council though grants FF0348550 and DP034474. The
support of the Cooperative Research Centre for Enterprise Distributed
Systems, the Department of Communications, Information Technology
and the Arts under a GranetNet grant, and the Australian Partnership for
Advanced Computing are also gratefully acknowledged. The support of the
reviewers to improve the manuscript is recognized and appreciated.
This research was supported by the Austra-
Abramson, D., J. Giddy, and L. Kotler (2000), High performance para-
metric modeling with Nimrod/G: Killer application for the global grid?
Paper presented at International Parallel and Distributed Processing Sym-
posium, Inst. of Electr. and Electr. Eng., Cancun, Mexico.
Andreae, M. O. (1991), Biomass burning: Its history, use and distribution
and its impact on environmental quality and global climate, in Global
Biomass Burning: Atmospheric, Climatic and Biospheric Implications,
edited by J. S. Levine, pp. 3–21, MIT Press, Cambridge, Mass.
Beringer, J., et al. (2003), Fire impacts on surface heat, moisture and carbon
fluxes from a tropical savanna in north Australia, Int. J. Wildland Fire,
Beringer, J., L. B. Hutley, N. J. Tapper, and L. J. Cernusak (2007), Savanna
fires and their impact on net ecosystem productivity in North Australia,
Global Change Biol., 13, 990–1004, doi:10.1111/j.1365-2486.
Chafer, C. J., M. Noonan, and E. Macnaught (2004), The post-fire measure-
ment of fire severity intensity in the Christmas 2001 Sydney wildfires,
Int. J. Wildland Fire, 13, 227–240.
Go ¨rgen, K., A. H. Lynch, A. G. Marshall, and J. Beringer (2006), The
impact of abrupt land cover changes by savanna fire on northern Aus-
tralian climate, J. Geophys. Res., 111, D19106, doi:10.1029/
Hao, W. M., M.-H. Liu, and P. J. Crutzen (1990), Estimates of annual and
regional releases of CO2, and other trace gases to the atmosphere from
fires in the tropics, based on FAO statistics for the period 1975–1980,
in Fire in the Tropical Biota, Ecol. Stud., vol. 84, edited by J. G.
Goldammer, pp. 440–462, Springer, New York.
Henderson-Sellers, A. (1993), A factorial assessment of the sensitivity of
the BATS land-surface parameterization scheme, J. Clim., 6, 227–247.
Hendon, H. H., and B. Liebmann (1990), A composite study of onset of the
Australian summer monsoon, J. Atmos. Sci., 47, 2227–2240.
Johnson, B. J., et al. (1999), 65,000 years of vegetation change in central
Australia and the Australian summer monsoon, Science, 284, 1150–1152.
Kalnay, E., et al. (1996), The NCEP/NCAR 40-year reanalysis project, Bull.
Am. Meteorol. Soc., 77, 437–472, doi:10.1175/1520-0477.
Lynch, A. H., S. McIlwaine, J. Beringer, and G. B. Bonan (2001), An
investigation of the sensitivity of a land surface model to climate change
using a reduced form model, Clim. Dyn., 17, 643–652.
Figure 3. Simulated differences (scenario minus reference
simulation) in response to variation in fire intensity and
burned area. Shown are differences in the mean zonal wind
speed [m s?1] at 850 hPa. Spatial averages of zonal winds
are calculated from 0? to 10?S and 120?E to 150?E, the zone
of monsoonal reversal of the flow in and out of Australia.
Only experiments with a timing of the fire-event later than
the Julian day 250 (7 September) are considered, as
indicated by crosses.
LYNCH ET AL.: GRID MODELING OF FIRE EFFECTS ON MONSOON
4 of 5
McGregor, J. L., and M. R. Dix (2001), The CSIRO conformal-cubic Download full-text
atmospheric GCM, in IUTAM Symposium on Advances in Mathematical
Modelling of Atmosphere and Ocean Dynamics, edited by P. F. Hodnett,
pp. 197–202, Kluwer, Dordrecht, Netherlands.
Miller, G., et al. (2005), Sensitivity of the Australian Monsoon to insolation
and vegetation: Implications for human impact on continental moisture
balance, Geology, 33, 65–68.
Pitman, A. J., and P. P. Hesse (2006), The significance of large-scale land
cover change on the Australian palaeomonsoon, Quat. Sci. Rev., 26,
Rivers, A. R., and A. H. Lynch (2004), The influence of land surface
transitions on the local climate of Beringia in the early Holocene, J.
Geophys. Res., 109, D21114, doi:10.1029/2003JD004213.
Russell-Smith, J., G. Allan, R. Thackway, R. Rosling, and R. Smith (2000),
Fire management and savanna landscapes in northern Australia, in Fire
and Sustainable Agricultural and Forestry Development in Eastern
Indonesia and Northern Australia, ACIAR Proc., vol. 91, edited by
J. Russell-Smith et al., pp. 95–101, Aust. Cent. for Int. Agric. Res.,
Canberra, A. C. T.
Russell-Smith, J., et al. (2003), Contemporary fire regimes of northern
Australia, 1997–2001: Change since Aboriginal occupancy, challenges
for sustainable management, Int. J. Wildland Fire, 12, 283–297.
Wang, B., I.-S. Kang, and J.-Y. Lee (2004), Ensemble simulations of Asian-
Australian-Monsoon variability by 11 AGCMs, J. Clim., 17, 803–818.
Wendt, C., J. Beringer, N. Tapper, and L. Hutley (2007), Local boundary-
layer development over burnt and unburnt tropical savanna: An observa-
tional study, Boundary Layer Meteorol., 124, 291–304.
? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ? ?
D. Abramson, Clayton School of Information Technology, Monash
University, Clayton, VIC 3800, Australia.
J. Beringer, A. H. Lynch, and P. Uotila, Climate Program, Monash
Sustainability Institute and School of Geography and Environmental
Science, Monash University, Clayton, VIC 3800, Australia. (amanda.
K. Go ¨rgen, De ´partement Environnement et Agro-Biotechnologies,
Centre de Recherche Public–Gabriel Lippmann, L-4422 Belvaux,
LYNCH ET AL.: GRID MODELING OF FIRE EFFECTS ON MONSOON
5 of 5