The climatology of Australian tropical aerosol: Evidence for regional correlation



[1] Biomass burning aerosols from the tropical savanna of Northern Australia constitute a globally significant aerosol source, with impacts on regional climate and air quality. Knowledge of the seasonal cycle and spatial distribution of this aerosol is required for its realistic representation in models of global climate, and to help define the role of this region in the global carbon cycle. This paper presents a decadal climatology of these aerosols, based on Sun photometer records from three stations in the Australian tropics, over the period 1998–2012. The monthly time series shows enhanced aerosol emissions following prodigious wet seasons, two of which occurred during the study period. The monthly climatology shows the expected peak during the late dry season (September–November), when most burning takes place, with clear evidence of the dominant modulating effect of fine-particle smoke emission apparent from the annual cycle of the Ångström exponent, a proxy for particle size. The aerosol levels during the early dry season are higher at the northern “Top End” stations than at the south-westerly Kimberley station. The time variation of aerosol optical depth is highly correlated between all three station pairs, with a correlation coefficient r2> 0.75 at monthly resolution between all pairs. This high correlation between widely separated stations declines only gradually as the filtering interval is reduced, suggesting remarkably high coherence in the emission and transport of biomass burning aerosol across the entire region.

1 Introduction

[2] Aerosol emitted from the burning of tropical biomass constitutes a significant component of the global aerosol budget, with impacts on climate and regional air quality. The role of this aerosol in climate is summarized by 2007], but large uncertainties remain. Biomass burning generates a large flux of carbon to the atmosphere of order 2000 TgCyr−1[Ito and Penner, 2004; Schultz et al., 2008], and so, improved knowledge of these emissions is required to better constrain their role in the global carbon cycle as well as in modulating climate.

[3] The Australian tropics are largely vegetated with savanna open woodland/grassland covering approximately 1.5 million km2or 20% of the area of the continent. The climate is broadly defined by a wet season between December and March, and a dry season between May and October, with transitional conditions in April and November. Biomass burning is widespread during the dry season, with approximately 30% by area of the savanna regions lying within Western Australia and the Northern Territory being burnt each season [Meyer, 2011]. Although fires occur naturally through lightning strikes, the majority of burning is carried out deliberately in order to reduce woody undergrowth and promote subsequent grass growth for grazing. Burning regimes are timed to prevent the accumulation of large amounts of highly flammable material that pose a significant fire risk if left late in the dry season. Current reports of Australia's annual biomass burning emissions display large interannual variability; for example, Schultz et al. [2008] estimated Australia's emissions as 169TgC/yr with a range of 83–295TgC/yr, consistent with a more recent estimate of 127TgC/yr by Haverd et al. [2013]. Australia's biomass burning emissions comprise about 8% of the global total, ranking third by continent behind Africa (48%) and South America (27%) [Schultz et al., 2008].

[4] Australian tropical aerosol has received increasing attention over recent years, following the deployment of Sun photometer stations in the region beginning in the late 1990s both by the Commonwealth Scientific and Industrial Research Organization (CSIRO) and the Australian Bureau of Meteorology (hereafter abbreviated “Bureau”). Further aerosol measurements are being made via the large suite of atmospheric measurements at the Tropical Western Pacific Atmospheric Radiation Measurement (TWP/ARM) station at Darwin ( under the ægis of the U.S. Department of Energy. In addition, airborne campaigns complemented by surface in situ measurements have further assisted this research. In particular, O'Brien and Mitchell [2003] and Luhar et al. [2008] found significant radiative forcing caused by smoke aerosol emissions. The airborne measurements of Carr et al. [2005] and Allen et al. 2008] provided first-cut breakdown of the significant aerosol types in the dry and wet seasons, respectively. Allen et al. [2008] also identified two populations of biomass burning aerosol as fresh and aged particles, a demarcation independently inferred from the classification of Qin and Mitchell [2009] from the analysis of sky radiance inversions.

[5] More recently, Bouya et al. [2010] and Bouya and Box [2011] analyzed data from Bureau and TWP/ARM photometers to delineate the seasonal characteristics of the aerosol at Darwin, while Radhi et al. [2012] applied a similar technique to the data from the CSIRO stations at Lake Argyle and Jabiru. Both studies identified the large increase in smoke aerosol toward the end of the dry season (October) with evidence for coarse mode sea-salt aerosol during the wet season, more prominent at Darwin as expected given its coastal location.

[6] In this paper, we amalgamate Sun photometer data from CSIRO and Bureau stations across the Australian tropics over the last decade, with the primary aim of establishing a benchmark climatology of Australian tropical aerosol. In addition, we examine the relation between the aerosol time series at different stations and find evidence for a surprising degree of coherence in the emission of biomass burning aerosol across the region.

2 Observations

[7] Aerosol measurements in northern Australia were begun by CSIRO in October 1998 with the installation of a Sun photometer at Darwin for testing prior to deployment at Lake Argyle, in the Kimberley region of Western Australia, in May 1999. Subsequently, a Sun photometer was installed at Jabiru in 2000, and a further instrument deployed at Darwin in 2004. The location of the stations used in this study is shown in Figure 1, with deployment details listed in Table 1. These stations make up the tropical component of the CSIRO Aerosol Ground Station Network, which was federated with NASA's Aerosol Robotic Network (AERONET) in late 2001. Cimel Sun photometers at Darwin were managed by CSIRO between April 2004 and August 2010, after which management was transferred to the staff at the TWP/ARM installation. The use of Cimel CE318 Sun photometers at the CSIRO stations has enabled the federation of the Australian network with AERONET. The Cimel instruments acquire direct sun measurements in up to 10 bands at intervals not exceeding 15 min and also perform sky radiance measurements every hour, enabling inversion for a range of intrinsic aerosol properties. Further details of the operation of these instruments in AERONET can be found in Holben et al. [1998].

Table 1. Details of Sun Photometer Stations Included in This Study
Lake Argyle−16.1081128.7485CSIROMay 1999
Jabiru−12.6607132.8931CSIROMay 2000
Darwin−12.4240130.8915CSIROOctober 1998February 1999
Darwin−12.4240130.8915CSIROApril 2004August 2010
Darwin−12.4240130.8915ARMAugust 2010
Darwin−12.4240130.8915BureauApril 1999
Figure 1.

Location of Sun photometer stations in Australia, with the three stations discussed in this report indicated. The southern boundary of the savanna region can be seen in this pseudocolor image.

[8] The Bureau has operated Carter Scott SPO2 Sun photometers at Darwin airport since April 1999, as part of its solar radiation network. These instruments produce 1 min statistics of the direct solar beam in up to seven bands conforming to the World Meteorological Organization (WMO) Global Atmosphere Watch wavelength sets [WMO, 2005].

[9] Calibration methods for both Bureau and CSIRO Sun photometers were developed and demonstrated by Mitchell and Forgan [ 2003] to yield aerosol optical depth accuracy to 0.007 at the 95% confidence interval in the midvisible using collocated data from Alice Springs in central Australia. This method consists of (a) identification of “Langley” periods of highly stable atmospheric transmission, allowing determination of the exoatmospheric instrument response lnV0 at a reference wavelength, usually 870nm, and (b) iterative application of the “general method” [Forgan, 1994] at successively shorter reference wavelengths. This approach assumes stability of the aerosol size distribution but, unlike the standard Langley method, not the aerosol optical depth. In effect, it propagates a high precision estimate of lnV0obtained at 870nm toward shorter wavelengths. Typically, the uncertainty in lnV0 at 440nm obtained by this method is reduced by a factor of ∼2 over that obtained from direct Langley analysis.

[10] While calibration conditions in the tropics are less favorable than those in more arid regions, careful monitoring of instrument performance and selection of observation periods suitable for calibration derivation enable this technique to provide a standard uncertainty of better than 0.01 in instantaneous midvisible aerosol optical depth for the data presented below, as demonstrated later in this section. In addition, the use of this method ensures congruence of CSIRO and Bureau measurements and means that the entire CSIRO record from 1998 onward can be included: the AERONET record from these sites does not begin until late 2001. However, since Cimel operations at Darwin were transferred from CSIRO to ARM in August 2010, the Cimel-derived aerosol data for that site were obtained directly from the AERONET site (level 2.0) from April 2004 onward.

[11] Cloud screening methods differ between the Bureau and CSIRO processing suites, in part due to differences in the instrumental sampling regimes. The CSIRO screening closely follows the strategy defined by Smirnov et al. [2000], as applied to AERONET. For each designated observation time, the Cimel instrument acquires a “triplet” with each of the three direct sun measurements separated by 30s. These are filtered for short-term variations by requiring that the coefficient of variation within the triplet be <1%. Second, the coefficient of variation for all observations during a given day is calculated. If this is <1%, then no further tests are applied. If not, then data points more than three standard deviations from the daily mean are eliminated. The daily statistics are then recalculated, and the test is applied recursively. This latter step is useful in removing data affected by thin cirrus cloud that may pass the high frequency filtering imposed by the previous step, an issue that is of great importance in the tropics [Chew et al., 2011; Huang et al., 2012].

[12] Cloud screening of the Bureau data was carried out via the following steps. The atmospheric transmission data taken at 60s intervals throughout the day are filtered to remove transmission not significantly different from zero, and then the remaining transmissions are converted to pseudo-aerosol optical depth with known atmospheric extinction components (ozone, molecular) removed. The pseudo-aerosol optical depth data are then screened for cloud using an adaption of the method of Alexandrov et al. [2004] using a 15 min window. To match the CSIRO analysis, independent valid triplets were identified mirroring the CSIRO methodology and compared to the Bureau method. A comparison of aerosol optical depths derived from the two methods at Darwin showed minor absolute differences of less than 0.002 in the monthly means. While the CSIRO and Bureau cloud screening methods differ, the good agreement between these two time series boosts confidence in both approaches, particularly important given the high potential for contamination by subvisual cirrus in tropical regions noted above.

[13] The derived aerosol data were aggregated into daily and monthly means. Daily means were deemed available when the number of cloud-screened data points in a given day was ≥8. Monthly means were generated independently from all available measurements in a given calendar month. A mean was assigned to a given month provided that the number of discrete observations in the month Nm and the number of days represented in the month dmsatisfied one of two criteria: Nm≥300 and dm≥8, or Nm≥600 and dm≥6.

[14] The operation of different instruments at Darwin and the different calibration and processing systems at all sites allows quantification of intersystem linkages. Table 2 lists mean bias and mean absolute difference in monthly mean aerosol optical depth at 500nm between the three systems considered: Bureau (SPO2), CSIRO (Cimel), and Aeronet (Cimel). In this study, AERONET level 2.0 data were used. The mean absolute difference inline image is more relevant in this comparison than the RMS error since the differences are not expected to be normally distributed but are more likely to exhibit periods of relatively constant offset due to the different calibration methods. The table shows biases within ±0.007 and inline imagebetween 0.01 and 0.02, suggesting individual system accuracies of better than 0.01.

Table 2. Comparison Between Sun Photometer Systemsa
StationPeriodSystemsMonthsBiasinline image
  1. a

    Mean bias and mean absolute difference in monthly mean aerosol optical depth at 500nm between the three Sun photometer systems considered. System differences cover instrumentation, calibration methods and processing algorithms.

DarwinApr 2004 to Nov 2011Aeronet-Bureau270.00660.0136
DarwinApr 2004 to Aug 2010Aeronet-CSIRO27−0.00110.0195
Lake ArgyleApr 2002 to Mar 2012Aeronet-CSIRO90−0.00600.0108
JabiruDec 2001 to Oct 2011Aeronet-CSIRO65−0.00340.0145

[15] The discussion below focuses on the climatology of aerosol optical depth δ and the Ångström exponent α. The latter provides a first-order indication of particle size by assuming a power law dependence of aerosol optical on wavelength, τλα. For the CSIRO data, αwas defined across the wavelength pair (440nm, 870nm). For the Bureau data, the Ångström exponent was derived using the wavelength pair (500nm, 868nm) [WMO, 2005], and if that pair was not available, then the pair (500nm, 778nm) was used.

3 Results and Discussion

3.1 Aerosol Time Series

[16] The time series of monthly mean aerosol optical depth at 500nm for all stations is shown in Figure 2. The expected strong annual signal associated with biomass burning during the dry season is clearly evident, although with significant interannual variation both in amplitude and form. In particular, the time variation of aerosol optical depth during 2011 and into 2012 is anomalous being both low in amplitude and multipeaked. This is likely a consequence of the relatively high rainfall during the wet season of 2010–2011, when the totals over the months November 2010 to April 2011 at or near the three stations were—Lake Argyle: 1.3m, Darwin: 2.8m, and Jabiru: 2.2m—compared to the climatological means of 0.8m, 1.8m, and 1.7m, respectively. The extra rainfall likely promoted extra growth in flammable grasses leading to a strong maximum in aerosol optical depth in 2012. A similar explanation may apply to the large maxima observed in 2001 and 2002, with several consecutive relatively high wet season rainfalls recorded at all stations between 1999 and 2001. Both periods coincide with the La Niña phase of the ENSO cycle, extending a similar link found by Swap et al. [2003] over the African savanna.

Figure 2.

Time series of monthly mean aerosol optical depth at 500nm at three stations in Northern Australia.

[17] Figure 2 presents evidence for a close correspondence in the time variation of aerosol optical depth at the three stations. While this might be expected for the wet season where the background aerosol is likely to be controlled by synoptic meteorology, it is not expected during the dry season when the aerosol is largely sourced through biomass burning, particularly given the wide separation of the stations (e.g., Lake Argyle to Jabiru is ∼800km). This point is further explored below.

3.2 Aerosol Climatology

[18] The monthly climatology of aerosol optical depth at 500nm is shown in Figure 3. The climatology for Darwin was based exclusively on the Bureau data, in view of its excellent continuity and temporal coverage. The high coherence between stations noted in the time series above is evident in the climatology, with no significant differences at the 1-sigma level. The monthly mean aerosol optical depth at 500nm is seen to vary from below 0.1 in April–June, to a maximum of ∼0.28 in October. There is a systematic difference between the stations in the months June, July, and August, with the aerosol loading being lowest at Lake Argyle, intermediate at Jabiru, and highest at Darwin. This reflects a dependence on proximity to the coast and the associated maritime influence on the aerosol loading, as discussed by Bouya et al. [2010], Bouya and Box [2011], and Radhi et al. [2012].

Figure 3.

Climatology of aerosol optical depth at 500nm at three stations in Northern Australia. The filled circles indicate monthly means, while the vertical bars represent ± one standard deviation.

[19] The corresponding climatology of Ångström exponent is shown in Figure 4. Although the standard deviation of the monthly means is large, a clear annual cycle is evident, from a wet season minimum of ∼0.5 to a peak during the burning season of ∼1.4 in September–October. This reflects modulation of effective particle size due to seasonal variation in relative humidity and the influence of fine particles from biomass burning. This seasonal distribution is subject to change since a majority of the fires are deliberately lit by land managers, and there has been a tendency to burn earlier in the dry season to prevent the occurrence of large, high intensity fires toward the end of the dry season. Under this scenario, there would be an expectation of higher aerosol optical depth levels earlier in the dry season, with a corresponding earlier rise in the Ångström exponent in the later years of the study period. However, exploration of that issue is beyond the scope of the present study. The climatologies of the aerosol optical depth and Ångström exponent at the three sites are tabulated in the accompanying auxiliary material.

Figure 4.

Climatology of monthly mean Ångström exponent at three stations in Northern Australia.

[20] Comparable aerosol climatologies for the savanna regions of Africa and South America have been published by Queface et al. [2011] and Schafer et al. [2008], respectively. Maxima in monthly mean τ500from these studies are 0.62 (Zambia) and 0.83 (Amazonia), both occurring in September, compared with 0.28 from the present study, occurring in October. These results indicate marked differences in the factors affecting aerosol loading due to savanna burning over the three continents, including fuel loads, fire regimes, and aerosol transport.

3.3 Regional Correlation

[21] While the coherence between the time series of aerosol optical depth noted above at monthly time resolution might be expected due to simple temporal averaging of a region-wide increase in fire frequency (and hence, smoke emission) as the dry season progresses, it was not expected at shorter timescales where regional differences in fire management regimes were anticipated. To investigate this point, we filtered the daily aerosol optical depth records at all three sites with intervals ranging from 3 to 35 days. For filter widths between 3 and 10 days, a legitimate average was recorded only for periods containing data for all days in the window (100% occupancy), while for filter widths > 10 days, 80% occupancy was required. This relaxation was necessary to maintain adequate records to facilitate the subsequent matching of records between sites.

[22] The dependence of the correlation coefficient r2calculated from matching records between the three site pairs is shown as a function of filter width in Figure 5. This plot shows remarkable levels of correlation between all site pairs and persistence of this correlation to relatively short filter widths w. In particular, for w=15days, r2 varies from ∼0.72 (Lake Argyle:Jabiru) to ∼0.81 for Darwin:Jabiru and Darwin:Lake Argyle. This range implies that between 72% and 81% of the variance can be explained by a simple linear dependence between sites. The increased correlation between Darwin and Jabiru over that between Lake Argyle and Jabiru is expected since Jabiru lies only 250km east of Darwin, and with prevailing south-easterly winds, there is expectation of an overlap between the sampling regions for the two northerly stations. However, the high correlation between Darwin and Lake Argyle is surprising given their separation. While the fall in correlation for w<15days is expected, the values of r2 are still remarkable. For example, at w=5days, r2ranges from 0.58 (Lake Argyle:Jabiru) to 0.75 (Darwin:Jabiru), implying that linearity explains between 58% and 75% of the variance between the site pairs. This analysis is based on days when a daily mean aerosol optical depth is available, thus biasing the selection toward cloud-free days during the dry season when biomass burning is the dominant aerosol source. Two factors that would give rise to this high correlation are (1) strong similarity in the timing, frequency, and intensity of fires across the savanna region, and (2) regional-scale transport mechanisms that spatially integrate the heterogeneous emissions. While further analysis is required to investigate these factors, the present results provide clear constraints on the development and testing of models used to represent biomass burning and aerosol transport across the Australian savanna.

Figure 5.

Correlation of daily mean aerosol optical depth at 500nm between pairs of the three stations studied, as a function of the filter width in days.


[23] The CSIRO network was established with the support of the CSIRO Earth Observation Center, with valued contributions from Dean Graetz and Denis O'Brien. We also thank station managers at all sites, in particular Michael Byers (deceased) and Greg Smith at Lake Argyle, and Robert Thorn at Jabiru. For the Darwin measurements, we thank observers at the Bureau of Meteorology station, and the staff of the TWP/ARM station supported by the U.S. Department of Energy. The CSIRO component of this work is supported through the Australian Climate Change Science Program sponsored by the Australian Department of Climate Change, and the U.S. Department of Energy through the Atmospheric Radiation Measurement program.

[24] The Editor thanks two anonymous reviewers for their assistance in evaluating this paper.