Interannual growth rate variations of atmospheric CO2 and its δ13C, H2, CH4, and CO between 1992 and 1999 linked to biomass burning



[1] High-precision, multispecies measurements of flask air samples since 1992 from CSIRO's global sampling network reveal strong correlation among interannual growth rate variations of CO2 and its δ13C, H2, CH4, and CO. We show that a major fraction of the variability is consistent with two emission pulses coinciding with large biomass burning events in 1994/1995 and 1997/1998 in tropical and boreal regions, and observations of unusually high levels of combustion products in the overlying troposphere at these times. Implied pulse strengths and multispecies emission ratios are not consistent with any other single process, but do not exclude possible contributions from covarying processes that are linked through climatic forcing. Comparison of CO2 with its δ13C indicates that most of the CO2 variation is from terrestrial exchange, but does not distinguish forcing by biomass burning from imbalance in photosynthesis/respiration of terrestrial ecosystems. Partitioning of terrestrial CO2 fluxes is constrained by H2, CH4, and CO, all of which are products of biomass burning but which have no direct link to net respiration of CO2. While CO is a strong indicator of biomass burning, its short lifetime prevents it from usefully constraining the magnitude of CO2 emissions. If the H2 and CH4 variations were dominated by biomass burning, they would imply associated carbon emissions in excess of mean annual levels of other years, of 0.6–3.5 and 0.8–3.7 Pg C for 1994/1995 and 1997/1998, respectively. The large range in emission estimates mainly reflects uncertainty in H2/CO2 and CH4/CO2 emission ratios of fires in these years.

1. Introduction

[2] Atmospheric concentrations of greenhouse gases are rising, leading to a positive radiative forcing of climate and an expected warming of surface temperatures. The largest individual contribution to the change in forcing due to human activities is from carbon dioxide (CO2). Systematic monitoring of its atmospheric levels commenced at Mauna Loa, Hawaii, in 1958 [Keeling et al., 1995] and has since been extended to a large number of globally distributed sites [see also Conway et al., 1994; Nakazawa et al., 1997; Francey et al., 1996]. An increasing long-term trend in CO2 and a decreasing trend in its 13C/12C isotopic ratio is well established [e.g., Etheridge et al., 1996; Francey et al., 1999a] and can be attributed to combustion of fossil fuels [Marland and Rotty, 1984; Marland et al., 2000] and a significant net flux of CO2 to the atmosphere as a consequence of land use change [Houghton, 1999]. However, the rate of CO2 increase is not steady, with prominent growth rate variations observed on timescales of 2–5 years. Year-to-year variations in fossil fuel related emissions are relatively small and cannot account for this behavior. The variations in rates of land use change are also too small, although accounting procedures are not as precise as for fossil fuel emissions so that such carbon flux estimates carry larger uncertainty.

[3] Atmospheric methods to attribute CO2 exchange between oceanic and terrestrial reservoirs have often employed the observed covariation of CO2 with either its δ13C [Keeling et al., 1989] or the oxygen/nitrogen (O2/N2) ratio [Keeling et al., 1996]. Both tracers preferentially reflect exchange with the terrestrial biosphere, δ13C by way of the much stronger isotopic fractionation associated with terrestrial photosynthesis compared to air-sea exchange, and O2/N2 by way of the low solubility of O2 in the oceans compared to that of CO2. Early studies of interannual variations (IAV) in CO2 using δ13C drew conflicting conclusions [e.g., Keeling et al., 1995; Francey et al., 1995; Nakazawa et al., 1997]. Francey et al. [2001] now report a consistent relationship between CO2 and δ13C, indicative of mainly terrestrial IAV forcing, over 2 decades, with the possible exception of the 1991–1993 period. Results from direct monitoring of O2/N2 between 1991 and 1999 also suggest most of the IAV in carbon fluxes is due to terrestrial exchange, with highest net carbon release centred on 1994 and 1998 [Battle et al., 2000; Manning, 2001]. While both δ13C and O2/N2 are subject to second-order interactions that potentially complicate interpretation on interannual timescales, both tracers indicate similar estimates of terrestrial IAV forcing over the last decade. Using an inverse model constrained by the spatiotemporal distribution of CO2 alone, Bousquet et al. [2000] concluded that IAV in terrestrial exchange was twice as large as that of oceanic exchange.

[4] Variations of CO2 growth rate have been linked to variations in climatic parameters associated with the El Niño Southern Oscillation (ENSO) [e.g., Dettinger and Ghil, 1998; Rayner et al., 1999a]. Early studies noted the potential for ENSO-related oceanic circulation perturbations to influence atmospheric CO2 [e.g., Bacastow, 1976]. However, recent assessment of ocean processes via CO2 partial pressure differences (ΔpCO2) suggests that air-sea fluxes can account for only a minor fraction of the observed variability [Lee et al, 1998; Feely et al., 1999; Le Quéré et al., 2000]. On the other hand, climate-driven ecosystem models seem able to accommodate much of the CO2 IAV through an ENSO effect on photosynthesis/respiration in terrestrial ecosystems [Kindermann et al., 1996; Tian et al., 1999; Gerard et al., 1999; Knorr, 2000; Yang and Wang, 2000] with El Niño periods favoring net CO2 release and La Niña periods favoring net uptake in tropical regions.

[5] Here we examine the IAV (defined here as perturbations sustained over longer than seasonal timescales, but less than the 8-year period of our data set; see also Figure 1 caption) of CO2 between 1992 and 1999 using measurements of flask air samples from the Commonwealth Scientific and Industrial Research Organisation's (CSIRO) Global Atmospheric Sampling Laboratory (GASLAB). Our analysis includes records of δ13C of CO2, and of other trace gas species, hydrogen (H2), methane (CH4), and carbon monoxide (CO), obtained from the same air samples. Compared to CO2, interpretation of variations in the other trace gases is complicated by greater uncertainty in global budgets involving a larger number of source/sink processes. Nevertheless, there are common processes, and thus a multispecies analysis can provide additional information for interpretation of IAV.

Figure 1.

Trace gas records from flask sampling at Cape Grim. Data are classified as retained (diamonds) or rejected (crosses). The plotted curves are obtained using the fitting routines described by Thoning et al. [1989]. First, a four-term harmonic and quadratic function is fit to the data. Residuals from the fit are smoothed by a low-pass filter using two cutoff periods, 80 days and 1.8 years. The residual curve with 80-day smoothing is added to the harmonic/quadratic function to generate a “smooth curve” (solid line). The residual curve with 1.8-year smoothing is added to the quadratic function to generate the “trend curve” (dashed line).

[6] The monitoring of atmospheric H2, CH4, and CO has not been as prolonged or as extensive as is the case for CO2. IAV in these species has been noted before, but generally related to anomalous events rather than any general climate factor such as ENSO over the whole period of records. For example, Dlugokencky et al. [1994] suggested that observed anticorrelation of IAV in CH4 growth rate between Northern (NH) and Southern Hemispheres (SH) in 1987–1990 might be explained by variations in the rate of interhemispheric exchange. High CH4 growth rates observed in later years were attributed to other factors, in 1991/1992 to a reduced hydroxyl (OH) sink resulting from the Mount Pinatubo volcanic eruption [Dlugokencky et al., 1996] and in 1998 to increased emissions from wetlands and boreal biomass burning [Dlugokencky et al., 2001]. In the latter case, the increased wetland emissions were linked to climatic variations involving positive temperature and precipitation anomalies. Brunke et al. [1990] reported IAV in CO at Cape Point, South Africa, over the period 1978–1987 and speculated about a link to ENSO through affected source/sink or atmospheric circulation processes. Novelli et al. [1998] suggested that IAV in CO between 1990 and 1995 included contributions from perturbations to OH due to the Pinatubo eruption and also from biomass burning. IAV in H2 was observed over a similar period but its causes were not addressed [Novelli et al., 1999].

2. Methods

[7] Equipment and techniques used for CSIRO GASLAB sample collection and analysis have been documented elsewhere [Francey et al., 1996; Allison and Francey, 1999; Langenfelds et al., 2001a, and references therein] and are therefore described only briefly here. Discussion focuses on the extent to which changes in experimental procedure and instrument calibration may produce spurious IAV in atmospheric records. Some differences in sampling methodology exist among sampling sites and minor variations have been implemented since 1992, involving upgraded pump units and changes in flask type. Extensive testing has established a high level of agreement among collection techniques and characterized trace gas modification due to storage of air in different flask types. Furthermore, these changes were phased in over a period of several years so that any errors would not appear as concerted IAV at all sites and their significance would therefore be diluted in terms of global mean observations. There have been no major changes to analytical equipment or techniques since 1992. Measurements are supported by rigorous calibration strategies that permit quantification of uncertainties in IAV. Thus, we consider that changes in sampling methodology and uncertainties in calibration are of second-order significance compared to the true atmospheric variability.

2.1. Sampling Techniques

[8] Sampling units currently used at most sites consist of a diaphragm pump (Model N05STI.9, KNF Neuberger, Princeton, New Jersey) which draws air through a chemical drying agent, anhydrous magnesium perchlorate [Mg(ClO4)2]. They were introduced at network sites between 1992 and 1997, superseding older units constructed around metal bellows pumps also with Mg(ClO4)2 drying. Parallel operation of old and new pump units at Cape Grim (CGA) in 1994/1995 demonstrated close agreement between the techniques. Similar units, specifically configured for aircraft-based applications have been in service and unchanged since 1992. The two Canadian sites, Alert (ALC) and Estevan Point (EPC), are operated by the Meteorological Service of Canada (formerly Atmospheric Environment Service) and employ a different sampling technique. Air is drawn through a diaphragm pump and dried in a cold trap held below −60°C. Samples are then sent to CSIRO for analysis.

[9] Air is generally collected in glass flasks, though a small number of samples are also collected in electropolished, stainless steel flasks at Cape Grim. The glass flasks are sealed with one of three types of O-ring material, PFA (perfluoroalkoxy), PTFE (polytetrafluoroethene) or Viton. Results of parallel sampling in multiple flask/O-ring combinations at CGA between 1994 and 1999 were used to evaluate data consistency among different flask types. Laboratory tests have examined stability of sample composition in different flask types for varying periods of up to about 1 year for 0.5-L flasks and more than 2 years for 5-L flasks (see Cooper et al. [1999] for preliminary results). Measurable drifts were observed in some flask types for CO2, δ13C, and CO. Relative rates of change among glass flasks were linked to O-ring material and flask volume. Where drift rates were modest and consistent among flasks of the same type, storage corrections were established and routinely applied to all network data. CO and δ13C data were rejected from 0.5-L glass flasks with Viton O-rings and CO data from metal flasks, due to drifts being too large or variable to permit reliable correction. Effectiveness of our storage correction scheme has been tested by parallel use of different flask types at South Pole (SPU) and Macquarie Island (MQA) where any storage-related errors are accentuated by long storage times of up to 1 year.

[10] If our corrections do not precisely capture actual flask storage drifts or if systematic errors are introduced by use of different pumps or flask types, errors can be manifested as false IAV. By using mean differences in measurements obtained from parallel pump unit operation at CGA and flask type comparisons at CGA, SPU, and MQA as a constraint of experimental uncertainty, it is possible to estimate uncertainties in our determination of changes in atmospheric composition and hence in the fluxes driving global mean IAV. This calculation accounts for actual experimental changes (equipment and timing) implemented at each individual site and makes allowance for mean flask storage times particular to each site. Uncertainties are calculated in a way that is consistent with our treatment of atmospheric variations (see below). They refer to changes in mixing ratio (or isotopic composition), integrated over a time window of 1.8 years and are calculated to be (1σ) ± 0.04 ppm for CO2, ±0.001‰ (δ13C), ±0.1 ppb (H2, CO and CH4). These numbers are within measurement precision for individual samples and suggest that uncertainties introduced by changes in sampling techniques have negligible impact on our characterization of IAV in these species.

2.2. Analysis

[11] Mixing ratios of CO2, CH4, H2, and CO were determined by gas chromatography (GC). CO2 and CH4 were measured on a Carle Series 400 instrument with flame ionization detection (FID), where the separated CO2 is first converted to CH4 by a heated (400°C) nickel catalyst in the presence of H2. CO and H2 were measured on a Trace Analytical Reduction Gas Analyzer, where heated mercuric oxide (HgO) is reduced to Hg vapour for detection by UV absorption. The isotopic composition of CO2 was determined using a Finnigan MAT252 isotope ratio mass spectrometer (MS), equipped with a Finnigan MT Box C for cryogenic separation of CO2 from air. The same instruments have been in service since before 1992.

[12] Measurements of all species are supported by calibration strategies that monitor long-term stability of the reference scale, nonlinearity in instrument response and sample handling procedures. Calibration of GC measurements is centred on regular analysis of suites of high-pressure cylinders with mixing ratios spanning the range of background atmospheric variations. Calibration of δ13C involves regular analysis of both whole air and pure CO2 reference gases for characterization of variations due to CO2 extraction and mass spectrometric influences [Allison and Francey, 1999]. Isotopic data are reported in the VPDB-CO2 scale, CO2 in the WMO mole fraction scale, and CH4 in the CSIRO94 CH4 scale, which is derived from and almost identical to the CH4 scale maintained at National Oceanic and Atmospheric Administration's Climate Monitoring and Diagnostics Laboratory (NOAA/CMDL). The CSIRO CO scale is linked to the scale of NOAA/CMDL by way of a single high-pressure cylinder standard near 200 ppb; however, measurements are not closely aligned at lower mixing ratios. H2 data are referenced to a scale established at CSIRO by “bootstrapping” to a gravimetrically derived CH4 scale. Comprehensive descriptions are provided elsewhere of CSIRO GC measurement programs [Langenfelds et al., 2001a, and references therein] and intercomparison with measurements of these species made by NOAA/CMDL [Masarie et al., 2001].

[13] Uncertainties due to instrument calibration can be estimated from measurements of regularly analyzed, natural air standards that are not actively used for routine calibration and thus provide an independent cross check of overall instrument performance. Residuals from the long-term mean of each standard were compiled into a single time series per species, and processed in a similar way to that described above for changes in sampling techniques to give uncertainties consistent with our treatment of atmospheric data. Calculated uncertainties (1σ) due to instrument calibration are ±0.04 ppm for CO2, ±0.015‰ (δ13C), ±0.4 ppb (H2), ±1.6 ppb (CO), and ±0.2 ppb (CH4).

3. Atmospheric Observations

3.1. Global Sampling Network

[14] Table 1 lists the CSIRO global flask network sites used in this study [see also Francey et al., 1996, 1998]. They are distributed across latitudes from 90°S to 82°N and, under suitable meteorological conditions, provide access to “background” air that, for our purposes, is representative of zonal mean atmospheric composition. With the exception of MLU and SPU, all surface sites are coastally situated and encounter air masses characterized by long marine back-trajectories that are far removed from anthropogenic sources and exchange with terrestrial vegetation. Air mass histories of the inland sites MLU and SPU are of similar character, due to little or no surrounding vegetation. Air is routinely sampled under background conditions at all sites. Since 1992, approximately monthly light aircraft flights have sampled the troposphere above Cape Grim to altitudes of 6–8 km, also under background conditions. Data from the 4- to 8-km altitude range are used here to complement the surface records. Examples of trace gas records for one site, CGA, are shown in Figure 1. Points represent measurements from individual flask samples. Two curves are fitted to the data and used in subsequent data analysis: a “smooth” curve that tracks short term variations including the seasonal cycle and a “trend” curve that is smoothed over a longer period, thus highlighting IAV. Growth rate curves used for subsequent analysis are the first derivative of the trend curves.

Table 1. CSIRO Global Flask Sampling Network Sites
CodeSiteaLatitude, °NLongitude, °EAltitude, mMean Sampling Frequency (samples yr−1/days yr−1)b
  • a

    Sites were selected for continuous records over most or all of the period 1992–1999. Two long-running network sites have been excluded, Cape Rama in India and Charles Point in northern Australia, as they are heavily affected by local continental influences that obscure the smaller IAV signal. Aircraft data are from approximately monthly flights over Cape Grim. Only data from the 4- to 8-km altitude range are considered here.

  • b

    Mean sampling frequency refers both to the number of individual air samples collected and the number of days in which air is sampled, averaged over 1992–1999.

SISShetland Islands60.2−1.23025/25
EPCEstevan Point49.4−126.53924/12
MLUMauna Loa19.5−155.6339741/21
CFACape Ferguson−19.3147.1247/17
CGACape Grim−40.7144.794170/48
MQAMacquarie Island−54.5159.01251/25
SPUSouth Pole−90.0−24.8281030/21

[15] Ability to detect large-scale IAV varies significantly among sites. There are two main factors accounting for this, sampling frequency (see Table 1) and short term atmospheric variability (Table 2). The combination of these factors favours higher signal-to-noise characteristics of high Southern Hemisphere (HSH) records (CGA and sites to its south). CSIRO sampling frequency is comparatively low at NH sites and northern air masses are subject to larger short-term variability due to closer proximity to continental regions and stronger latitudinal gradients [Harris et al., 1992]. The contrasting sensitivity of respective hemispheres to detection of global IAV is illustrated by comparison of precision statistics at CGA and MLU in Table 2. Similar experimental techniques are employed at both sites but data scatter is larger at MLU, by up to a factor of 7 for CO. This reflects higher short-term atmospheric variability at MLU and hence better resolution of global IAV at CGA, at least for long-lived species.

Table 2. Experimental Uncertainties for CSIRO Measurements From Selected Sites in Either Hemisphere
SpeciesMeasurement PrecisionaExternal PrecisionbRSD (Cape Grim)cRSD (Mauna Loa)c
  • a

    Measurement precision is defined here as the mean (over 7 years) standard deviation of repeat aliquots taken from high pressure cylinders.

  • b

    External precision is the mean standard deviation in flask pair differences from Cape Grim.

  • c

    RSD is the residual standard deviation of retained flask data with respect to the smooth curve (see Figure 1).

δ13CO2, ‰
CO2, ppm0.
H2, ppb1.
CO, ppb0.
CH4, ppb2.

[16] In Figure 2, 1992–1999 growth rates are plotted for all trace gas species, as functions of time and latitude. A striking feature is strong correlation in the main component of temporal variations of all species (see also Figure 3, discussed further below). All show maxima in 1994/1995 and 1997/1998 and minima in 1995/1996 (note that δ13C growth rate is inverted in this plot). This feature is somewhat unexpected given that correlations among these species on a global scale have not been previously reported and also because their global budgets are not dominated by the same exchange processes. The link between CO2 and δ13C need not come as a surprise, though quantitatively the relationship is important in that it constrains relative contributions of terrestrial and oceanic exchange.

Figure 2.

A multispecies contour plot of growth rates showing variations with time and latitude. Each plot is constructed using growth rate curves for the nine surface sites (Table 1; but excluding aircraft data), and by linear interpolation between adjacent sites. There is no smoothing across latitudes. Data from ALC have been extrapolated to 90°N. The δ13CO2 scale is inverted to show positive correlation with CO2 through biological exchange.

Figure 3.

Comparison of global mean growth rate curves of multiple trace gas species and the SOI (plotted against the right-hand axis, which is inverted). Each growth rate curve is constructed by averaging smooth curves from the nine surface sites (with seasonal cycles removed) and reapplying the curve-fitting procedure. To facilitate direct comparison of the multispecies correlation, mean growth rates (1992.0–2000.0) are first subtracted from each curve. CO2 is scaled by a factor 10 and δ13CO2 by a factor of −160 to make their growth rate variations visible on the same scale and to highlight forcing by terrestrial exchange. Units are H2, CH4, and CO (ppb yr−1); CO2 (ppm yr−1 × 10); and δ13C (‰ yr−1 × −160). Perfect correlation between CO2 and δ13C in this plot would imply CO2 exchange characterized by δ13C discrimination of −17‰ with allowance made for modification of δ13C by isotopic disequilibrium fluxes over the smoothing time.

[17] Differences among sites in the intensity and phase of growth rate variations may hold information about the origins of the IAV forcing but should be viewed with more caution. This is especially the case for the NH where uncertainties are larger. Our expectation of better resolution of global IAV in the SH is supported by greater uniformity of growth rate variations among the sites south of 40°S in Figure 2. We suggest that for this study, interpretation of growth rate variations in terms of genuine and significant atmospheric features requires that they be consistent at two or more neighboring sites.

[18] There is some evidence of latitudinal dependence in the phase of growth rate maxima. A leading tropical signal is suggested for δ13C, CO2, and CO in 1997/1998 while there is no clear dependence in 1994/1995. One point that can be made is that the main component of the variability for all species and through both periods is a global phenomenon and is not isolated to limited sites or latitude bands. It cannot be explained by anomalies in meridional transport, which would induce simultaneous growth rate variations of opposite sign at different latitudes.

[19] There are some instances of growth rate anomalies that are not common to all five tracers. They indicate that even if the main component of IAV for all species is due to a common or closely linked forcing, there are other contributions to IAV from unrelated processes. One prominent example is an increase in CH4 growth rate during 1996 at MLU and sites to its south that is not mirrored by the other species. This might be due to an anomaly in either a CH4 source that is independent of the other species or in the rate of interhemispheric exchange. The latter possibility would require an anomalously large injection of CH4-rich NH air into the SH in 1996 and for this to have occurred at a time of the year when the interhemispheric CH4 gradient is large by comparison to the other species (relative to experimental precision). A preliminary assessment of MLU/CGA differences suggests that such conditions might exist at some times of the year, however a more thorough investigation of this hypothesis would require a transport model and possibly additional data from more suitably located sites.

[20] The multispecies correlation can also be displayed in the form of global mean growth rate variations, obtained by reapplying the curve fit to a mean “smooth curve” averaged over all nine surface sites (Figure 3) after the average seasonal cycle at each site is removed. The phase of both 1994/1995 and 1997/1998 maxima is closely consistent among all species. CO shows the largest departure with earlier maxima in both years, most notably in 1994 where its maximum precedes that of CO2 by 7 months. Some of this difference is due to the short lifetime of CO. More specifically, in this study we use the “adjustment time” (global average about 3 months for CO), which relates to the rate of decay of perturbations in atmospheric mixing ratio rather than the turnover time of total atmospheric burden [Prather et al., 1996]. A simple test using a single box model (described in more detail below), and employing the same algorithms used to generate the growth rate curves of Figures 2 and 3, shows that CO growth rate maxima inferred from the curve fitting procedure lead actual fluxes by about 3 months. This artifact is a consequence of the smoothing time used by the curve fit being much longer than the CO adjustment time. Growth rates of CO2 and δ13C in Figure 3 are scaled so as to highlight forcing by terrestrial exchange. Over the 8-year timeframe, similar amplitude of both curves implies that a major fraction of the variation involves terrestrial forcing. Also plotted in Figure 3 is the (inverted) SOI. Maximum trace gas growth rates in 1994 and 1997/1998 coincide with the latter part of El Niño events. This relationship resembles that observed for CO2 and the SOI in earlier decades [e.g., Keeling et al., 1989; Rayner et al., 1999a, 1999b], although the absence of elevated CO2 growth rates during the early part of the extended 1991–1994 event appears as an anomaly in more than 40 years of continuous monitoring of both parameters.

3.2. Aircraft

[21] Aircraft-based vertical sampling of the troposphere above Cape Grim has revealed large seasonal variations in trace gas composition between 4 and 8 km altitude. Between July and November, vertical profiles frequently show distinct plumes in limited altitude bands that are enriched in CO2, CO, CH4, and H2 and depleted in δ13C of CO2 with respect to background surface air. Through analysis of interspecies relationships and air mass histories for 1992–1997 data, Pak [2000] attributed the observed trace gas enrichment to emissions from biomass burning. Plume back-trajectories were traced to the SH tropics where they crossed landmasses subject to extensive biomass burning during the dry season at this time of year. Transit times to Cape Grim were typically 8 and 13 days from southern Africa and South America, respectively. Longer times apply to transport from northern Australia and southeast Asia.

[22] The vertical profile data exhibit IAV in the magnitude of trace gas enrichment of the midtropospheric plumes. This is most dramatically illustrated by CO (Figure 4) for which the phase of IAV was similar to that observed globally at the surface but with amplitude substantially larger than at Cape Grim and other SH sites. There was marked enrichment in midtropospheric CO with respect to Cape Grim surface data, centered on 1994 and 1997. High values were also observed in 1999 but appeared during a period of low sampling frequency and were dominated by data from a single profile. Growth rate curves for other species were generally well correlated with surface data, however uncertainties are larger because of the relatively low sampling frequency (mean 11 days yr−1) and the episodic nature of enriched plumes. The aircraft data suggest that the SH midtroposphere saw emissions from the same source(s) forcing the global IAV but with greater intensity, at least for the short-lived species CO.

Figure 4.

The difference in midtropospheric CO with respect to the surface at Cape Grim. Points are from aircraft-based flask sampling in the 4- to 8-km altitude range above Cape Grim, with the Cape Grim smooth curve subtracted. The residuals are fitted with a “trend curve.”

4. Source Attribution

4.1. Global Trace Gas Budgets

[23] A central issue for this paper is whether all species are responding to the same biogeochemical process or perhaps a combination of processes closely linked to a common climatic forcing (e.g., ENSO). Consider the sources (Table 3). Substantial amounts of CO2 are released to the atmosphere by combustion of fossil fuels and biomass burning. Oceans and terrestrial ecosystems may be either net sources or sinks, reflecting imbalance in large, opposing fluxes. Sources of H2 and CO are closely linked; major sources are photochemical production through oxidation of CH4 and nonmethane hydrocarbons (NMHCs) by OH and combustion of both fossil fuels and biomass. Oceans are a minor source of both gases and a small amount of CO is emitted directly from vegetation. The global CH4 budget comprises a wider range of sources falling within the broader categories of microbial (natural wetlands, rice paddies, oceans, ruminant animals, and landfills), fossil fuel (natural gas leakage, coal mining) and biomass burning. Other than biomass burning, the only sources common to all species are fossil fuel use and oceans, both of which can be excluded as sole forcing mechanisms on the grounds that magnitudes of implied flux variations are unrealistically high for at least some of the species. A primary oceanic forcing of CO2 is further contradicted by the δ13C data.

Table 3. Trace Gas Fluxes by Source/Sink Typea
 CO2, Pg C yr−1H2, Tg yr−1CO, Tg yr−1CH4, Tg yr−1
Fossil fuels
   Fugitive emissions   50–80
   Vegetation/soils60–120 50–160 
   Biomass burning1–315–20370–125025–70
   Landfills   20–40
   Rice paddies   40–280
   Natural wetlands   40–150
   Termites   20–150
   Ruminant animals   60–160
   Oxidation of CH4 7–29600–950 
   Oxidation of NMHCs 14–25290–1000 
   Soils 56–90180–39010–30
   Tropospheric OH 6–192000–3200380–490
   Stratosphere  80–14040–70

[24] Similar arguments rule out changes in sink processes. The main sink of CO and CH4 is oxidation by OH. Although H2 is itself destroyed by OH, it is also a by-product of CH4 destruction and a net product of OH photochemistry [Novelli et al., 1999]. CO2 is the end-product of both CO and CH4 (via CO) oxidation. Thus variations in OH concentration would produce anticorrelation in growth rates of CO and CH4 (net loss due to higher OH) against those of H2 and CO2 (net gain). Soils are a major sink of H2 but only minor sinks of CH4 and CO, and are a source of CO2. Thus, if the observed IAV were to be explained by a single exchange process, the current state of knowledge of the global budgets of these trace gases dictates that biomass burning is the only plausible candidate. While biomass burning has previously been associated with IAV of CO and CH4, it has not been identified as a major contributor to IAV in CO2 or H2. However, it is known that biomass burning is a significant source of these and other gases [e.g., Crutzen and Andreae, 1990; Andreae and Merlet, 2001] and that the extent of emissions can vary dramatically from year to year, at least on regional scales.

4.2. Other Midtropospheric Observations

[25] We have already shown that CO in the midtroposphere above Cape Grim exhibits IAV of similar phase to that at the surface but with larger amplitude. Concordant IAV has previously been reported for products of biomass burning in the free troposphere. Intense, seasonal CO maxima were observed in 1994, 1995 and 1997 above Lauder, New Zealand, and linked to seasonal maxima in aerosol [Jones et al., 2001]. Regular flask sampling from aircraft in the 8.5- to 13-km-altitude range above the western Pacific showed prominent CO maxima in late 1994 and late 1997 [Matsueda et al., 1999]. Positive ozone (O3) anomalies were found in spectroscopic and ozonesonde data over Indonesia at the same times [Fujiwara et al., 1999, 2000] and from satellite-based measurements above southeast Asia in 1997 [Thompson et al., 2001]. Spectroscopic measurements of CO, hydrogen cyanide (HCN) and ethane (C2H6) above Mauna Loa between 1995 and 1998 showed correlated enhancement in late 1997 [Rinsland et al., 1999]. Using similar techniques, Rinsland et al. [2000] further reported elevated CO, HCN, and C2H6 above Jungfraujoch, Switzerland, peaking in early to middle 1998. In each case the anomalies were linked to major biomass burning events in southeast Asia in 1994 and 1997/1998. The phase lag in anomaly maxima between Mauna Loa and Jungfraujoch in 1997/1998 may be consistent with either tropical emissions being transported northward and/or a contribution from northern, extratropical emissions in 1998. Spaceborne measurements from three 10-day space shuttle missions showed dramatically elevated CO over the tropics in October 1994 by comparison to both April 1994 and October 1984 [Connors et al., 1999], with largest enhancements over southeast Asia and tropical South America. The total atmospheric CO burden in October 1994 was calculated to exceed that of October 1984 by 90 Tg. This value can be used to approximate emissions of other trace gases. Assuming the entire difference was due to biomass burning with a molar CO/CO2 emission ratio of 0.1 [Pak, 2000] and negligible long-term CO trend, a lower limit for 1994 CO2 emissions is 0.4 Pg C. Total CO2 emissions integrated over the whole burning event would likely be higher if allowance is made for emissions after the October observations and for the short adjustment time of CO. It is likely that a significant fraction of the CO emitted earlier in the burning event would already have been oxidized by OH.

4.3. Biomass Burning

[26] Major tropical biomass burning events, linked to ENSO dry periods, were reported for both 1994/1995 and 1997/1998. However, the magnitude of emissions is not well known as there are large uncertainties in the parameters that determine the total trace gas emissions, for example area burned, biomass loading per unit area and combustion efficiency. Southeast Asia was one affected region with areas of 50,000 km2 or more burned in each of 1994 [Nichol, 1997; Folkins et al., 1997] and 1997/1998 [Liew et al., 1998; Levine, 1999; Legg and Laumonier, 1999; Siegert and Hoffmann, 2000]. One estimate of area burned in Indonesia alone in 1997/98 amounted to 90,000 km2 [Page et al., 2000, and references therein]. Most of the trace gas emissions from these fires were attributed to combustion of peat. A lower limit of 0.2 and a range of 0.4–0.9 Pg C were estimated for 1997/1998 carbon emissions by Levine [1999] and Page et al. [2000], respectively.

[27] Large fires also occurred in South America in 1997/1998, although the total areal extent is not well known. The Brazilian state of Roraima alone accounted for a burned area of 15,000 km2 [Nepstad et al., 1999; Andreae et al., 2001], contributing toward a possible doubling of net emissions from deforestation for the Brazilian Amazon relative to other years [Nepstad et al., 1999]. This would imply an additional release of about 0.2 Pg C following the estimates of Houghton et al. [2000]. Potter et al. [2001] suggest that in the early 1990s, emissions from the same region varied by up to 1 Pg C yr−1 with highest emissions occurring during El Niño years. IAV in African biomass burning emissions was reported by Barbosa et al. [1999] for the earlier period 1981–1991, with midrange estimates of annual emissions varying by up to 0.4 Pg C. However, maximum emissions were not directly correlated with El Niño as was the case for southeast Asia and South America. Interannual variability in biomass burning has been reported not only for the tropics, but also for boreal regions [Kasischke et al., 1999]. More than 110,000 km2 of forest covering both North America and Eurasia were burned in the severe fire year of 1998, and both 1994 and 1995 were also major fire years, at least in North America. A best estimate of 1997/1998 carbon emissions of 0.3 Pg C is implied by Kasischke et al. [1999], assuming consumption of 5000 tons km−2 of biomass containing 50% carbon.

4.4. Covariations

[28] While biomass burning appears to be the only process that by itself might plausibly satisfy the observed multispecies IAV, this does not exclude the possibility that climate driven covariations might exist among otherwise independent sources and sinks. Covariations with biomass burning are likely. One could contemplate many permutations involving sources and sinks of different species and external forcing factors; however, an exhaustive analysis is not attempted here. Rather, discussion is limited to covariations that might involve fluxes of sufficient magnitude to account for a major fraction of the observed IAV, and which could be linked to ENSO.

[29] Covariations of CO2 with ENSO have previously been identified and may include both oceanic and terrestrial components. We can use δ13C to investigate this partitioning (see below). The terrestrial component may involve fluxes due to both biomass burning and imbalance in photosynthesis/respiration. Perturbations favoring net respiration are predicted by some ecosystem models for warm/dry conditions associated with El Niño [Kindermann et al., 1996; Tian et al., 1999; Gerard et al., 1999; Knorr, 2000; Yang and Wang, 2000], so that a positive correlation with biomass burning is possible. It seems unlikely that photosynthesis/respiration could account for all of the CO2 variability because positive correlations with all three species, H2, CO, and CH4 are not expected. The absence of a direct link among emission of H2, CO, and CH4 and respiration of CO2 is supported by results of ground-based flask sampling in terrestrial ecosystems (Figure 5 [Lloyd et al., 1996; Miranda et al., 1997; Lloyd et al., 2001]). The samples were collected over several years from 13 individual campaigns on five different continents by (1) the Australian National University (ANU), Canberra, Australia, (2) The University of Edinburgh, Edinburgh, Scotland, and (3) the Max Planck Institute (MPI) for Biogeochemistry, Jena, Germany. All samples were analyzed by CSIRO. The highly elevated CO2 values result from a combination of net respiration and stable boundary layer conditions. H2 is strongly anticorrelated with CO2 due to uptake by soils while CO and CH4 show little dependence. Thus, when natural variations in terrestrial gas exchange are the prime source of variability, the relationship of CO2 exchange with that of H2, CO, and CH4 is very different from the large-scale relationships characterizing the observed global IAV (or the multispecies, biomass burning signals identified by Pak [2000]).

Figure 5.

Observed relationship of changes in H2, CO, and CH4 with changes in CO2 in natural, terrestrial ecosystems. Data are from near-surface flask sampling of air from terrestrial ecosystems in five different countries, Brazil (red; five individual campaigns distinguished by symbol type, includes sampling in regions of Jaru Forest and Cerrado), Siberia, Russia (blue; five campaigns), Cameroon (green; one campaign), Wagga, Australia (pink, one campaign), and Saskatchewan, Canada (magenta, one campaign). Solid lines represent mean emission ratios implied by our atmospheric observations for the two periods, 1994/1995 and 1997/1998 (see section 5) and are forced through approximate “background” values of 360 ppm (CO2), 500 ppb (H2), 120 ppb (CO), and 1800 ppb (CH4). Dashed lines show the full range of uncertainty covering both periods. Anomalously high ratios observed during one Siberian campaign (blue circles) were due to local biomass burning. CH4 in Siberia is also known to be significantly elevated by emission from bogs.

[30] Uptake by soils is the largest term in the global H2 budget and might be influenced by climate, though the nature of any covariation with CO2 fluxes is not known. CO global budget terms for direct emission from vegetation and consumption by soils are too small. Plants are known to release some NMHCs (isoprene and terpenes) that are oxidized in the atmosphere and produce H2 and CO via formaldehyde. This process represents a substantial global source of both species. The factors controlling NMHC emissions by plants are complex, but include climatic variables such as temperature and humidity [Zimmerman et al., 1978; Guenther et al., 1993, 1995; Goldstein et al., 1998; Schade et al., 1999; Pétron et al., 2001]. It is conceivable that increased NMHC production could be linked to net respiration of CO2. However, even if this link did exist, the resulting CO/H2 ratio (2.8 as calculated from Novelli et al. [1999]) would be lower than indicated by our IAV observations (best estimates 5–8, see below), though not excluded when all uncertainties are factored in. Microbial production of CH4 from wetlands and rice paddies is a major term in the global CH4 budget. Global emissions might be significantly perturbed by ENSO-related climatic variations, with emissions promoted by higher temperatures and soil moisture (via precipitation). Dlugokencky et al. [2001] used a global, process-based model to account for most of the 1998 CH4 growth rate anomaly through increased wetland emissions, responding to positive temperature and precipitation anomalies.

4.5. Influences on δ13C

[31] The main use of atmospheric δ13C in carbon cycle studies is as a tracer of net CO2 exchange with the terrestrial biosphere. However, it is also possible that IAV in δ13C occurs in the absence of net CO2 exchange. Changes in the mixing ratio and δ13C of atmospheric CO2 are described by separate mass budgets for 12C (≈[Ca]) and 13C (≈[Caδa]) [Enting et al., 1993],

equation image
equation image

where F, Nlb, and No are net CO2 fluxes from fossil fuels, exchange with the land biosphere, and oceans; ε terms represent isotopic discrimination with respect to atmospheric δ13C; Glb and Go are the gross fluxes of CO2 through land biosphere and oceans, and δalb and δao are the values of atmospheric δ13C that would be in equilibrium with the land biosphere and ocean reservoirs [Tans et al., 1993].

[32] Here we focus on the terms at the right-hand side of equations (1a) and (1b) that might exhibit significant IAV. It has been established elsewhere that peak-to-peak IAV of F [Marland et al., 2000] and No [Le Quéré et al., 2000] in the 1990s was ≤0.7 Pg C yr−1, implying that most of the IAV in d[Ca]/dt of up to several Pg C yr−1 was of terrestrial origin. Francey et al. [2001] use a revised 20-year record of δa to reach a similar conclusion, which implies little contribution to IAV from the No and G terms at the right-hand side of equation (1b). Some authors [e.g., Kaplan, 2001; Thompson and Randerson, 1999] have suggested that IAV in the εlb and Glbalb − δa) terms might be important if δ13C is used for partitioning. It is easy to demonstrate that possible variations in global εlb of 1–2‰ due to climatic influences [Kaplan, 2001] have a small (<10%) influence on the Nlb term on interannual timescales [e.g., Enting et al., 1993]. The G(δ − δa) terms represent an isotopic disequilibrium flux (isoflux), and combined are the largest term at the right-hand side of equation (1b). The terrestrial isoflux component represents about one third of the total isoflux [Trudinger, 2000]. Thompson and Randerson [1999] have estimated IAV of <10 Pg C‰ yr−1 in the terrestrial isoflux due to atmospheric δ13C variations. If this component is misinterpreted as a δ13C variation due to net terrestrial exchange with discrimination of −17‰ (exchange with predominantly C3 vegetation), the error is <0.6 Pg C yr−1, again small compared to the total net terrestrial IAV. Succession of C3 photosynthesizing plants by C4 species in tropical terrestrial regions following ENSO-induced fires has been raised as another possible source of IAV in the terrestrial isoflux [Ciais et al., 1999]. However, this effect will only be significant if the average lifetime of respired detritus in tropical regions is long compared to the 1.8-year lifetime of this study. In terms of a perturbation pulse model discussed below for the 1994/1995 and 1997/1998 high CO2 growth rates, it is pertinent that the midtropospheric CO2 and δ13C anomalies reported by Pak [2000], and attributed to tropical biomass burning, are consistent with the release of carbon from C3 vegetation (−0.048‰ ppm−1, characteristic of photosynthetic discrimination of about −17‰), and give no indication of significant isoflux or C4 vegetation effects. Here we assume that other possible sources of IAV in isoflux, including an oceanic component about which little is known, are small in the 1.8-year smoothed global average.

5. Modeling

[33] A simple one-box model, representing a well-mixed global atmosphere, is used to quantify flux anomalies forcing the observed IAV. This approach is designed to test the hypothesis that a significant fraction of the IAV is due to the major biomass burning events of 1994/1995 and 1997/1998. Specifically, we ask (1) can the observed IAV be reproduced by a release of pulses at these times and (2) are the implied emission ratios consistent with biomass burning? For CH4, CO, and H2, the model tracks time variation in the composition of the pulses by taking account of atmospheric adjustment times and incorporating the main interactions involving destruction/formation by OH photochemistry.

[34] Modeled mixing ratios take the form

equation image

obtained by adding a set of pulses (with each pulse defined by a response function R(t) and weighting coefficient b to “base curves” (Co + at) that are defined to have zero IAV, but may have constant, nonzero growth rate (as defined by the coefficient a and determined by the optimization procedure described below). The base curves represent the sum of all other processes and include a contribution from biomass burning that is equivalent to the mean seasonal emissions of other years.

[35] The evolution of CH4, H2, and CO pulse mixing ratios are described by the following set of equations, which are solved analytically:

equation image
equation image
equation image

[36] Because we are modeling the evolution of perturbations to the background atmosphere, it is appropriate to use atmospheric adjustment times rather than lifetimes [Prather et al., 1996, Prather, 1996]. Adjustment times (τ) are taken to be 12.2, 0.25, and 2.3 years for CH4, CO, and H2 [Warneck, 1988; Prather et al., 1996; Novelli et al., 1999]. We take a1 = 0.85, assuming OH destruction in the troposphere represents 85% of the CH4 sink and all CH4 is oxidized to CO, and a2 = 0.58 (0.68 × 0.85), assuming 0.68 moles of H2 are produced per mole of CH4 destroyed [Warneck, 1988]. Production of CO and H2 from OH destruction of NMHCs emitted by biomass burning are not explicitly treated in the model because (1) there are limited observational constraints of NMHC emissions for these events and (2) the lifetimes of NMHCs are generally very short with respect to transport times from emission sources to our measurement sites and the smoothing time employed by our analysis. Thus the fluxes inferred from our analysis include the CO and H2 both emitted directly by fires and produced later as by-products of NMHCs emitted by the fires.

[37] Response functions for pulses of CO2 and δ13C cannot be described in terms of a single lifetime because equilibration with different compartments of oceanic and land biosphere reservoirs involves various timescales. We use CO2 and δ13C response curves for small perturbations to the contemporary atmosphere obtained from the box-diffusion model of Enting and Lassey [1993] [see also Trudinger et al., 1999; Trudinger, 2000]. The model is calibrated against 14C and assessed against the Law Dome δ13C record of Francey et al. [1999a], and includes enhancement of net primary production (NPP) in terrestrial ecosystems due to CO2 fertilization [Trudinger, 2000]. On the timescales of interest to us (i.e., <6 years), the main influences are equilibration with the surface ocean and the “young” terrestrial biosphere. Response curves are fitted by functions comprising a sum of exponential terms, which for CO2,

equation image

(approximately) represent distinct “pools” of CO2 that are removed from the atmosphere on different timescales [e.g., Wigley, 1991] and where t is in years,

equation image

[38] These equations represent the evolution of a pulse of CO2 per unit emission and of unit change in δ13C in a well-mixed atmosphere. We take advantage of the fact that for small perturbations, the shape of the response is not strongly dependent on pulse strength. For δ13C, the response functions for injections of terrestrial and oceanic CO2 pulses are almost identical [Trudinger, 2000]. Uncertainties are estimated here by allowing for ±20% variation in the time constants, reflecting approximate uncertainty in gross flux (G) terms.

[39] It is important to note that signal strengths detected at our measurement sites will be modified from their initial values, especially for short-lived species. Figure 6a shows the decay in airborne fraction of all species as a fraction of an initial pulse at t = 0, here ignoring production from OH-destruction of precursor species. There is dramatic depletion of CO over the timescales of interest to us. Most is destroyed within a few months so that only a small fraction of a CO pulse is captured by the 1.8-year smoothing time. This situation is exacerbated by the fact that some depletion will occur between the time of emission and arrival of the signal at our measurement sites. Assuming a mean transport time of 2 months (equivalent to 50% response at Cape Grim of a signal released at the equator (P. J. Rayner, personal communication, 2000)), a CO pulse would already have decayed to just half its initial value. Depletion of other species over these timescales is not as dramatic, but is taken into account for quantification of emissions and emission ratios.

Figure 6.

(a) Airborne fraction as a function of time after emission, according to pulse response functions used in the model. For δ13C, the curve represents the fractional change in a perturbation to δ13C in a well-mixed atmosphere. The dashed line at 2 months represents the assumed mean transport time between source and measurement sites. (b) Response of H2 and CO to an emission pulse of 1 ppb CH4, taking into account their respective adjustment times and production of H2 and CO through destruction of CH4 by OH.

[40] By making allowance for adjustment time and OH-related interactions that include production from CH4 oxidation, it is possible to calculate the response of H2 and CO to an initial pulse of 1 ppb CH4 (Figure 6b). Maximum response is <0.08 and <0.02 ppb for H2 and CO, respectively, confirming that observed IAV of H2 and CO is largely independent of this source. For CO this represents a negligible fraction of observed IAV, while for H2 it accounts for <10%. Similarly, oxidation of CO and CH4 (via CO) can account for only a small fraction of CO2 IAV. Our model explicitly accounts for these interactions in the treatment of CH4, H2, and CO. The major contributions to CO2 and δ13C through CO occur within the first few months after release (Figure 6a) and due to the 1.8-year smoothing time are not treated explicitly but rather are counted as part of the initial CO2 and δ13C pulses.

[41] Modeled mixing ratio (and δ) curves are constructed by taking the first year (1992) of observations, appending “base curves” (1993–1999) with zero IAV but nonzero trend, and adding two sets of emission pulses each of 1-year duration. The first year of observations are used in the model to minimize errors from not resolving growth rate variations in 1994/1995 from those of 1992/1993. Each set of pulses is divided into equal, monthly parts centered on August 1994 and February 1998, thus matching the mean times of observed growth rate maxima. The 1-year duration of pulses approximates the duration of anomalous biomass burning activity. This period was well defined in 1997/1998 with major fires starting in August 1997 in the tropics and continuing through to September 1998 in boreal forests. The same duration is assumed for 1994/1995, although the time distribution is not as clearly defined in these years and may involve above average biomass burning emissions from different regions over a period of more than 1 year. Our objective, for each species, is to find the magnitude of both sets of pulses to best fit observed growth rate variations while maintaining integrity in the long-term trend. An iterative procedure optimizes two variables, pulse strength and growth rate of the base curve (i.e. coefficients b and a in equation (2)). Pulse strengths are adjusted to match the area beneath modeled and observed growth rate curves with respect to the growth rate of the base curve. The base curve growth rate is adjusted to maintain consistency of observed and modeled mixing ratios (and δ) before and after each period of modeled pulses (1993.0 and 1996.5 for 1994/1995 pulses, 1996.5 and 2000.0 for 1997/1998 pulses). Calculations for the two sets of pulses are thus independent except for the response to the 1994/1995 emissions which continues through 1997/1998. This procedure does not require modeled and observed curves to be perfectly in phase.

5.1. Results

[42] Modeling results are listed in Table 4 and plotted in Figure 7. The model successfully reproduces the main features of the observed global mean growth rate curves. Largest discrepancies appear for CO2 and H2 in 1994/1995. The shape of the 1994/1995 H2 growth rate peak is well reproduced by the model but leads the observations by 3 months. The same feature is manifested as an offset between observed and modeled mixing ratios during the period of modeled pulses. This discrepancy appears to be related to a HNH perturbation that was specific to H2. Peak growth rates in 1994/1995 at the three northernmost sites showed a phase lag with respect to other sites (Figure 2). A secondary maximum/minimum cycle was observed at ALC and SIS in 1993/1994, suggesting a contribution to IAV from another process (the discrepancy in Figure 7 is almost entirely removed when the observed global mean growth rate curve is calculated without data from the three northernmost sites).

Figure 7.

Global mean records (left-hand side) represented by observed “trend” (dark blue), observed “smooth” (light blue; averaged smooth curves from all sites with seasonal cycle removed), and modeled (red) curves. The right-hand side compares observed (blue) and modeled (red) growth rates and their difference (observed-modeled; black).

Table 4. Modeled Pulse Strengths and Emission Ratios for 1994/1995 and 1997/1998 Events
  • a

    Calculated from the difference in growth rate between records of individual sites and the global mean by integrating over a time window of 1.8 years.

  • b

    See section 2 for derivation of experimental uncertainty.

  • c

    Growth rates that are independent of modeled emission pulses, as determined by the iterative optimization procedure.

  • d

    The range of uncertainty in modeled emission pulses due to uncertainty in pulse response functions and in mean transport time between source and measurement sites (2 ± 1 months).

  • e

    Modeled emission pulses, expressed as a change in mixing ratio (or δ) in a well-mixed atmosphere. Uncertainties are calculated by adding in quadrature the contributions from signal resolution, calibration, and response/transport times.

Global mean growth rate 1992–19991.46 ppm yr−1−0.016‰ yr−11.4 ppb yr−1−2.6 ppb yr−14.4 ppb yr−1
Resolution of interannual variations among sites integrated over 1.8 years (1σ)a0.12 ppm0.009‰1.4 ppb13 ppb1.3 ppb
Uncertainty due to instrument calibration and changes in sampling procedure (1σ)b0.07 ppm0.015‰0.5 ppb1.6 ppb0.3 ppb
Model Input
Atmospheric adjustment time, years  2.3 ± 0.50.25 ± 0.112.2 ± 3
Model Output
Base curve growth rate 1993.0–1996.5c1.27 ppm yr−1−0.016‰ yr−11.0 ppb yr−1−0.6 ppb yr−12.5 ppb yr−1
Base curve growth rate 1996.5–2000.0c1.51 ppm yr−1−0.023‰ yr−10.8 ppb yr−1−0.8 ppb yr−14.1 ppb yr−1
Emission error factord0.95–1.060.90–1.140.90–1.170.47–4.40.98–1.04
1994/1995 emissione1.30−0.16+0.17 ppm (2.8 Pg C)−0.077−0.021+0.01910.8−1.9+2.4 ppb (3.8 Tg H2)57−33+194 ppb (283 Tg CO)9.2−1.3+1.3 ppb (26 Tg CH4)
1997/1998 emissione2.02−0.18+0.19 ppm (4.3 Pg C)−0.082−0.021+0.01914.9−2.1+3.0 ppb (5.3 Tg H2)117−63+398 ppb (581 Tg CO)11.7−1.3+1.3 ppb (33 Tg CH4)
1994/1995 emission ratio, ppm−1 CO2 −0.059−0.018+0.0168.3−1.7+2.2 ppb44−26+149 ppb7.1 ppb
1997/1998 emission ratio, ppm−1 CO2 −0.041−0.011+0.0107.4−1.2+1.7 ppb58−32+197 ppb5.8−0.9+0.9 ppb

[43] For CO2, the model does not adequately reproduce that part of the growth rate peak that extends into 1995 and which is not mirrored by CO or CH4 (H2 is inconclusive). For this reason, the results shown in Figure 7 and Table 4 were obtained by restricting integration of CO2 growth rate peaks to the 12-month period of modeled pulses. The δ13C was treated in the same way to maintain integrity between these species for partitioning of terrestrial and oceanic fluxes. This modified procedure provides a better fit to observations at the time of multispecies correlation in growth rate in 1994 but neglects that part of the CO2 IAV occurring in 1995. The difference between observed and modeled curves implies an unaccounted net CO2 release of 1.5 Pg C in 1995. There is a corresponding δ13C residual that, at face value, implies the 1995 CO2 release is approximately equally due to terrestrial and oceanic exchange, though this source allocation must be viewed with caution as the implied fluxes are smaller than their uncertainties.

[44] The discrepancies for CO2 (and other species) preceding the 1994 maximum are likely due in part to the assumption of no other contributions to IAV between 1993 and mid-1996. Growth rate perturbations observed for CO2, H2, CO, and CH4 during 1991–93 were linked, at least for CO and CH4 to the 1991 Mount Pinatubo volcanic eruption [Conway et al., 1994; Dlugokencky et al., 1996; Novelli et al., 1998, 1999]. These variations may not be fully resolved from the 1994/1995 event, especially in light of the 1.8-year smoothing time used in our analysis.

[45] A high level of agreement between observed and modeled curves was obtained for all species in 1997/1998. One feature worthy of mention is the 1998 spike in observed CO mixing ratio indicated by the light blue curve in Figure 7. Maximum values were recorded in late 1998, mainly reflecting a HNH signal that was distinct from CO variations at other latitudes (Figure 2). Growth rate maxima in the HNH occurred substantially later than in the tropics but were more intense. This suggests anomalously high CO emissions occurred in more than one latitude band. It is consistent with emissions peaking in the tropics in late 1997 and in higher latitudes of the NH in mid-1998, a distribution that corresponds to times of major biomass burning activity in both tropical (at least in southeast Asia) and boreal regions.

5.2. Uncertainties

[46] Pulse strength estimates in Table 4 allow for experimental uncertainty and for signal-to-noise in detection of IAV, as determined by consistency among sites and illustrated by the light blue curves at the left-hand side of Figure 7. They partly allow for assumptions relating to trace gas adjustment times and transport times between source regions and measurement sites. Of the species considered here, modeled CO emissions carry the largest uncertainty, owing to the short CO adjustment time. They are strongly sensitive to transport time, which will vary with latitude, and the season when emissions occur. The assumed transport time of 2 ± 1 months approximates an equatorial source but becomes less satisfactory for higher latitude emissions. Because OH is not uniform through the troposphere, our simplistic treatment of CO destruction (which assumes constant global OH) also contributes uncertainty. Actual adjustment times are shorter in the tropics and longer at high latitudes and in the upper troposphere. These factors have far weaker impact on modeled emissions of other species.

[47] There are other potentially large uncertainties that cannot be quantified and are thus not explicitly included in our calculations. They relate to the assumption that the IAV can be largely attributed to two sets of emission pulses. It is likely that some contributions to IAV forcing result from other processes with a different time distribution. For example, the CH4 growth rate anomaly commencing in 1996 at Mauna Loa and sites to its south is inconsistent with modeled 1997/1998 pulses. One way to gauge the possible significance of interfering processes is by comparison of the model-derived base curve growth rates for the two periods before and after mid-1996 (Table 4). If all of the IAV were fully explained by our modeled pulses, these terms should be equal. Significant differences do emerge for CO2 (i.e., 1.27 and 1.51 ppm yr−1), δ13C, and CH4 which translate to errors in modeled pulse strengths of as much as 30%.

5.3. Postburning Fluxes

[48] An additional factor not included in our model is the response of terrestrial ecosystems to a burning event. Decomposition of the dead plant matter that remains after a fire is a source of CO2 to the atmosphere and is opposed by CO2 uptake through regrowth [Ward et al., 1992; Poth et al., 1995; Houghton et al., 2000]. Of greatest significance here is the net response in the first 1–2 years because this quantity is not distinguished from the initial pulse emissions of our model. Over longer timescales, the response to fires in earlier years forms part of the modeled base curves. The net response is governed by the combustion factor (the fraction of dead vegetation burned), the rate of decomposition of the remaining dead vegetation, and the rate of regrowth. These factors can depend heavily on the nature of the fire, for example, the type of biomass fuel and the intensity. The net response to fires in forests (e.g., variable combustion factor dependent on fire intensity; rate of regrowth dependent on mortality rate of trees), peat swamps (effectively slow rate of decomposition and maybe slow regrowth), and savanna (high combustion factors and rapid regrowth) is likely to vary significantly. The potential impact on our results was explored by constructing a range of response scenarios using possible ranges of each parameter. Uncertainties are large and even the sign of the response in early years is uncertain, although an initial net release of CO2 to the atmosphere is favored. The scenario representing the upper extreme for net respiration of CO2 suggests that within the first year, this source could even exceed the initial release through combustion. However, a response as large as this is not supported by Figure 7, where both a phase lag in growth rate maxima and tailing of the CO2 and δ13C growth rate peaks would be expected relative to H2, CH4, and CO. The sustained CO2 and δ13C flux anomalies in 1995 are consistent with a smaller contribution from net respiration while there is no evidence for postburning fluxes in 1997/1998 that are not captured by our model.

6. Flux Estimates

6.1. Partitioning of Terrestrial and Oceanic CO2 Fluxes

[49] Model results for CO2 alone give optimal pulse strengths of 2.8 and 4.3 PgC in 1994/1995 and 1997/1998, respectively. The corresponding signal in δ13C can be used to separate terrestrial and oceanic components, using estimates for εlb and εo of −17 ± 2 and −1.8 ± 0.3‰, respectively. Significant net terrestrial release of 3.5 ± 1.1 and 3.7 ± 1.1 PgC is indicated for the two periods (Table 5). Implied oceanic releases of −0.8 ± 1.1 and 0.6 ± 1.1 PgC are smaller and within uncertainty. The assumed value for εlb of −17‰ represents CO2 exchange with predominantly C3 vegetation. Mean global values for photosynthetic discrimination, weighted by gross primary productivity (GPP), were estimated by Lloyd and Farquhar [1994] to be −17.8 and −3.6‰ for C3 and C4 plants, respectively. A heavy bias toward C3 exchange in 1994/1995 and 1997/1998 flux anomalies is expected if biomass burning played a significant role. The major fires in these years occurred in forests of Southeast Asia, South America and boreal regions, all of which are dominated by C3 vegetation [Lloyd and Farquhar, 1994]. Although large areas of savanna with a higher proportion of C4 plants are burned seasonally, especially in Africa and to a lesser extent in tropical America [Hao and Liu, 1994], it is only perturbations from mean seasonal biomass burning activity that are relevant here. In any case, Pak [2000] detected no clear influence of C4 photosynthetic discrimination in the 8 years of midtroposphere sampling of tropical biomass burning plumes (although the isotopic data are sparse and scattered). Furthermore, C3 plants account for the major fraction of global GPP (e.g., 79% as determined by Lloyd and Farquhar [1994]) so that any flux anomalies due to imbalance in photosynthesis/respiration might also be expected to favour C3 exchange [see also Kindermann et al., 1996; Tian et al., 1999].

Table 5. Partitioning of CO2 Pulses
CO2 SourceConstraintBiomass Burning Emission Ratio, mol mol−11994/1995,a PgC1997/1998,a PgC
  • a

    Positive pulses are into the atmosphere.

  • b

    This includes both biomass burning and imbalance in photosynthesis/respiration.

  • c

    Allowance is made for the range of literature estimates of emission ratios for biomass burning and for uncertainties in CH4, H2, and CO pulses as given in Table 4.

  • d

    Only the “best estimate” is quoted due to weakly constrained uncertainty in the biomass burning emission ratio.

  • e

    This is calculated assuming a biomass burning CO/CO2 emission ratio range of 0.06–0.16.

Oceansδ13C −0.8 ± 1.10.6 ± 1.1
Terrestrial biosphereb  3.5 ± 1.13.7 ± 1.1
Biomass burningCH4/CO2 (Pak [2000]; all fires)0.006–0.0161.1–3.8c1.4–4.7c
H2/CO2 (Pak [2000]; all fires)0.008–0.0340.6–3.5c0.8–4.8c
CO/CO2 (Pak [2000]; all fires)0.06–0.160.3–8.9c0.7–18.3c
CH4/CO2 (Matsueda et al. [1999]; Japan-Australia aircraft transects at 10 km altitude)0.0089 ∼2.8d
CO/CO2 (Matsueda et al. [1999]; Japan-Australia aircraft transects at 10km altitude)0.077 ∼3.2d
CH4/CO2 (Matsueda and Inoue [1999]; aircraft sampling above Singapore, October 1997)0.0045 ∼5.5d
H2/CO (Sawa et al. [1999]; aircraft sampling above Indonesia, October 1997)0.06–0.10 1.7–10.6e

6.2. Multispecies Constraints on Biomass Burning CO2 Fluxes

[50] If we assume that the modeled H2, CH4, and CO emission pulses are entirely due to biomass burning, we can use them with literature estimates of biomass burning emission ratios to directly constrain emissions of CO2 (Table 5). The main limitation to this approach is that literature values show large variations due to dependence of emissions on vegetation type and combustion efficiency [e.g., Crutzen and Andreae, 1990]. Although the high CO ratios with respect to other species are a strong indicator of biomass burning, estimated CO pulse strengths are too uncertain to provide a firm quantitative constraint of CO2 emissions. Our best estimates of CO/CO2 ratios in Table 4 are consistent with upwards of 10% of the terrestrial CO2 pulses being from biomass burning, based on a molar emission ratio range of 0.06–0.16 from the literature survey of Pak [2000].

[51] The longer-lived species H2 and CH4 are better suited to this application but uncertainty in actual source emission ratios becomes a serious limitation. Molar emission ratios for different fire types are mostly in the range 0.006–0.016 and 0.008–0.034 for CH4/CO2 and H2/CO2, respectively [Pak, 2000]. Generally, low values are associated with savanna fires where combustion efficiency is high. Values closer to the upper end of the range might be expected for the 1994/1995 and 1997/1998 events if emissions came mostly from fires in peat swamp forests with a higher proportion of smoldering combustion. Data for emissions from peat combustion are sparse but an identical range of 0.006–0.016 for CH4/CO2 was obtained by Yokelson et al. [1997] from laboratory experiments. The pulse strength ratios derived from our model are near the lower end of the literature range. Actual source ratios at the upper end of the range would imply minimum biomass burning contributions to terrestrial CO2 pulses of 17% in 1994/1995 and 22% in 1997/1998 (Table 5).

[52] For CH4, there is also a question of possible interference from covariations involving microbial CH4 production. The assumption made here, that most or all of the CH4 IAV can be ascribed to biomass burning, conflicts with the findings of Dlugokencky et al. [2001], who attribute most of the 1998 IAV to wetland emissions and a minor fraction to mostly boreal biomass burning. There are two points to be made about the Dlugokencky et al. [2001] analysis. First, it focused on 1998 and did not address the positive growth rate anomaly that commenced in 1996 in the tropics and SH and continued through 1997, in tandem with CO2, H2, and CO (see Figure 2). Second, it may have underestimated global biomass burning flux anomalies during the 1997/1998 period. A CH4 emission of 5.7 Tg was determined for boreal fires but only 1.8 Tg was assumed for fires elsewhere based on the estimate of Levine [1999] for Indonesia. Possible contributions from fires in other regions, for example, South America, were not included. It is possible that both biomass burning and wetland emissions were responsible for significant and overlapping IAV but, with available information, there is no definitive way of establishing their relative contributions.

[53] Observations of changes in the isotopic composition of atmospheric CH4 may help to clarify this situation. For example, CH4 from biomass burning and microbial production carries distinctive isotopic signatures in the 13C/12C ratio (δ13C ≈ −25 and −60‰) as compared to background atmospheric values of about −47‰. If the 1994/1995 CH4 perturbation were entirely due to biomass burning, it would have been accompanied by a shift in δ13C of +0.12‰. A concordant, positive shift is favored by the record of Lowe et al. [1997] from Baring Head, New Zealand, the Cape Grim record of Francey et al. [1999b], and the SH record of Quay et al. [1999]; however, a shift of this magnitude is not well resolved relative to overall precision by any of these records. It remains to be seen whether δ13C measurements through 1997/1998 are able to distinguish biomass burning from microbial sourced CH4 or constrain overlapping contributions from both processes.

6.3. Emission Ratios From Cape Grim Vertical Profiles

[54] Interspecies relationships of the seasonal source influencing the midtropospheric plumes above Cape Grim were obtained by comparing data from individual samples against background concentrations [Pak, 2000]. Calculated “emission ratios” with respect to CO2, averaged over 1992–1997, were 0.0086 ± 0.0008‰ mol mol−1 for H2, 0.0216 ± 0.0020‰ mol mol−1 for CO, 0.0073 ± 0.0007‰ mol mol−1 for CH4, and −0.048 ± 0.005‰ ppm−1 for δ13C. These values are consistent with literature estimates for biomass burning emissions, allowing for reduction of the CO/CO2 ratio by a factor of 2 or more due to rapid removal of CO by OH. In some respects, the vertical profile data are well suited to quantification of biomass burning emission ratios applicable to our study of global IAV. Sampled plumes are sufficiently young to retain high signal-to-noise characteristics, yet are sufficiently old to reduce uncertainties associated with sampling of fresh plumes, in relation to heterogeneity of emissions (e.g., due to source material and burning efficiency) and modification of aging plumes by chemistry involving short-lived species such as NMHCs [Mauzerall et al., 1998; Holzinger et al., 1999]. However, determination of emission ratios directly relevant to this study would require that the excess 1994/1995 and 1997/1998 components be resolved from mean seasonal signals. These ratios are likely to differ. Regular, seasonal emissions may include a substantial contribution from savanna fires, especially in Africa, which mostly involve flaming combustion and are characterized by low emission of H2, CO, and CH4. The large 1994 and 1997/1998 tropical fires were predominantly in peat swamps (in Southeast Asia) and forests [Nichol, 1997; Levine, 1999; Nepstad et al., 1999]. There is limited knowledge of emission ratios from peat combustion [Yokelson et al., 1997], but forest fires generally emit more H2, CO, and CH4 due to a higher proportion of smoldering combustion [e.g., Laursen et al., 1992]. The sparsity of vertical profile data prevents resolution of 1994/1995 and 1997/1998 emission ratios to a degree that would provide a firm quantitative constraint of global IAV.

6.4. Emission Ratios From 1997 Tropical Fires

[55] Some measurements were made of the plumes above the large Southeast Asian fires of 1997. Derived emission ratios have the advantage that they directly relate to fires we suspect of contributing to much of the IAV in CO2, but spatial and temporal sampling resolution is low. Matsueda and Inoue [1999] reported large enhancement of CO, CO2, and CH4 over Singapore in October 1997 from aircraft-based flask sampling. Measured emission ratios were 0.089 and 0.051 for CO/CO2 and CH4/CO, implying a CH4/CO2 ratio of 0.0045. Aircraft-based, in situ measurements of H2 and CO over Indonesia in October 1997 yielded H2/CO ratios of 0.06–0.10 [Sawa et al., 1999]. This compares with the global best estimate of 0.33 of Andreae et al. [1996] and a range of 0.06–0.52 implied by the survey of Pak [2000]. These results give low CH4/CO2 and H2/CO2 ratios that support a dominant biomass burning contribution to CO2 IAV (Table 5). It must be accepted, however, that their representativeness of large-scale behavior may not be reliable.

[56] A more reliable indication of the large-scale emission characteristics of the 1997 tropical fires is available from regular aircraft-based flask sampling of the midtroposphere (8.5–13 km altitude) between Japan and Australia. Matsueda et al. [1999] reported elevated CH4 and CO peaking in October to November 1997, where the enhanced CH4/CO ratio was consistent with a dominant biomass burning influence. The enhanced CH4/CO2 and CO/CO2 ratios from the same samples were 0.0089 and 0.077, but peak CO2 enhancement occurred 1–2 months after CH4 and CO (H. Matsueda, personal communication, 2000). These ratios are also consistent with a dominant biomass burning contribution to CO2 IAV in 1997/1998. However, the decoupling of CO2 from CH4 and CO demands an explanation. It would be hard to imagine the CO2 emissions being independent of CH4 and CO, as a phase difference of 1–2 months, while not trivial, seems too short for independent and coincidental sources. The CH4/CO observations alone imply that CO2 emissions from biomass burning are expected, and of magnitude similar to that observed, albeit by integrating over several months. A more likely explanation is that the phase difference reflects either heterogeneity in emission characteristics of fires during this period and/or some delayed CO2 emissions due to decomposition of dead vegetation following the combustion phase. It remains unclear as to how representative these observations are of all 1997/1998 tropical fires, integrated over all affected regions and over their full duration. Furthermore, no account is taken here of the contribution from 1998 boreal fires. Thus the accuracy of these emission ratios for linking to global IAV-related fluxes is uncertain.

6.5. Summary

[57] Our ability to precisely quantify biomass burning carbon fluxes is limited by knowledge of actual source H2/CO2 and CH4/CO2 emission ratios for the 1994/1995 and 1997/1998 events. However, we can define a range that is consistent with the observations. Upper limits of literature emission ratios imply lower emission limits for CO2 of 0.6 and 0.8 PgC for 1994/1995 and 1997/1998. Terrestrial pulses indicated by our model are 3.5 and 3.7 PgC, which we adopt as likely upper limits for biomass burning. Intermediate values would be consistent with some fraction of the observed terrestrial signal being due to imbalance in photosynthesis/respiration. Higher values for total, terrestrial release would violate agreement in observed and modeled growth rate peak shapes and/or the constraint imposed by δ13C but cannot be dismissed, especially in 1997/1998 where an oceanic release of 0.6 PgC is indicated by the δ13C-based budget. A net oceanic flux of this magnitude is somewhat surprising, especially as ΔpCO2-based estimates of air-sea flux favour a net flux of opposite sign during El Niño events [Feely et al., 1999; Le Quéré et al., 2000]. Any budgeting error in overestimation of the oceanic release would translate to underestimation of the terrestrial release. This is plausible, within the ±1.1 PgC uncertainty, and would require an error in the δ13C-based partitioning of fluxes, involving either random (e.g., signal-to-noise) or systematic influences (e.g., underestimated isotopic disequilibrium flux or actual εlb of source material >−17‰, for example, through a larger fraction of C4 exchange).

6.6. Comparison With Inventory-Based Estimates

[58] The magnitude of IAV in biomass burning carbon fluxes accommodated by our results is generally higher than allowed for by previous IAV/budgeting studies. This may reflect underestimation of fluxes in previous studies, together with the fact that while anomalously high emissions have been reported from time to time for particular regions, they have generally not been evaluated in terms of combined global fluxes. The available, inventory-based estimates of 1997/1998 fires for different regions are not inconsistent with results from our atmospheric, multispecies approach. The more conservative estimates of a 0.2 PgC release from each of Southeast Asia [Levine, 1999] and South America [Houghton et al., 2000] together with a best estimate of 0.3 PgC from boreal fires [Kasischke et al., 1999] gives a total of 0.7 PgC, similar to the lower bound of our estimated range. Upper bounds of 0.9 PgC for Southeast Asia [Page et al., 2000] and about 1 PgC for South America (Potter et al. [2001], if conclusions relating to the early 1990s are extended to the 1997/1998 event) would imply a global release of about 2.2 PgC.

[59] It is arguable whether the parameters used in the inventory-based estimates are sufficiently constrained to exclude even larger global fluxes. For example, the Levine [1999] calculations assumed conservative estimates of area burned that were based on satellite imagery for Kalimantan and Sumatra only. Fires on other Indonesian islands and in neighboring countries were not included. There appears to be substantial uncertainty in values adopted for biomass loading and combustion efficiency. For the 1997/98 Indonesian fires, Levine [1999] used conservative values for combustion efficiency of 20% and above-ground biomass loading of 10,000 tons km−2, though noting a mean value of 23,000 for tropical forests in Southeast Asia. These values compare with ≤50% and 20,000 tons km−2 used by Page et al. [2000]. Effective biomass loading of peat was taken as 97,500 and 40,000–100,000 tons km−2 by Levine [1999] and Page et al. [2000], respectively, partly reflecting differences in assumed burning depth (1.5 and 0.4–1.0 m) and combustion efficiency (50 and 100%).

[60] The thickness of peat layers consumed by these fires might be a key parameter. The depth of Indonesian peat ranges from 0.5 to more than 10 m, with about half its total area exceeding 2 m. Field surveys in central Kalimantan following the 1997/1998 fires revealed losses of between 0.2 and 1.5 m of the peat surface [Page et al., 2000]. Most peat forests in Malaysia are characterized by a greater thickness, perhaps averaging about 6 m, but with some peat layers being as deep as 16 m [Anderson, 1964; Richards, 1996]. Importantly, the peat forests of Sumatra and Kalimantan are considered to be true ombrogenous mires, comparable to the raised bogs of Eurasia [Richards, 1996]. Studies of the effects of fires on such bogs in Russia have revealed that complete burning of the organic part of the peat soil often occurs down to its mineral bottom, which in that study was at a depth of about 1 m [Zaidel'man et al., 1999]. Although the discussion of uncertainties here has focused on the Southeast Asian fires, similar considerations probably apply to the South American and boreal forest fires.

6.7. Pre-1993 IAV

[61] This study has focused on the period 1992–1999, owing to the availability of high-precision, multispecies data obtained from the same air samples. In light of the conclusion that biomass burning was a major influence on IAV during this period, we can further examine earlier multispecies IAV information and links to El Niño or biomass burning events that occurred before 1992. Concurrent, high precision records of δ13C in CO2, CH4, and CO extend back to the early 1980s, allowing comparison over this period with the longer CO2 records starting in 1958. Measurements of H2 over 3.5 years between 1985 and 1989 were reported by Khalil and Rasmussen [1990] but the records are too short to clearly resolve any IAV.

[62] High growth rates of CH4 and CO following the 1991 Pinatubo eruption were linked to perturbations in OH [Dlugokencky et al., 1996; Novelli et al., 1998] but were not accompanied by elevated growth rates of CO2. The largest positive global CO2 growth rate anomalies since 1980 were observed in 1983 and 1987 [Conway et al., 1994] (see also NOAA/CMDL website, coinciding with the largest CO growth rate anomalies of the decade observed at the SH sites, Cape Point [Brunke et al., 1990] and Cape Grim [Fraser et al., 1994; Langenfelds et al., 2001b]. Reevaluation of the CO213C relationship by Francey et al. [2001] shows these CO2 variations to be largely from terrestrial forcing. Both 1983 and 1987 anomalies occurred during El Niño events that led to major fires in Indonesia and elsewhere in the Western Pacific region [Malingreau et al., 1985; Page et al., 2000, and references therein]. Extensive fires were also reported for Amazonia [Setzer and Pereira, 1991] and Eurasia in 1987 [Cahoon et al., 1994; Kasischke et al., 1999]. Thus, there is a substantial body of evidence to indicate a biomass burning influence on IAV of CO2 and CO during the 1980s. However, CH4 IAV did not show as strong a correlation with the other species as was observed after 1992. Markedly different behavior was observed in the Northern and Southern Hemispheres between 1983 and 1990 [Dlugokencky et al., 1994] (see also Positive growth rate anomalies were observed in the SH in 1983 but not in 1987 [Dlugokencky et al., 1994; Fraser et al., 1994]. This suggests that processes other than biomass burning probably contributed to forcing of CH4 IAV during this period.

7. Conclusions

[63] Strong correlation in interannual growth rate variations of CO2, its 13C/12C isotopic ratio, H2, CH4, and CO was observed at a network of globally distributed surface sites between 1992 and 1999. The relationship of CO2 with ENSO was broadly consistent with observations from earlier decades. Growth rate variations were global in extent with a latitude/phase relationship suggesting partly tropical forcing with an additional contribution from high northern latitudes in 1998, at least for CO. Similar variations were observed in vertical profiles of the troposphere above Cape Grim, where plumes enriched in products of biomass burning are observed seasonally between July and November. Strong enrichment was observed in years coinciding with major biomass burning events in Southeast Asia and South America in 1994 and 1997/1998, and independent observations of elevated levels of biomass burning products such as CO, O3, C2H6, and HCN in the free troposphere at these times. Larger areas of boreal forest were burned in 1994, 1995, and 1998 than in neighboring years.

[64] Our analysis shows that the observed global IAV is largely consistent with two emission pulses of 1-year duration centred on August 1994 and February 1998 and that the chemical signature is consistent with a significant, perhaps major contribution from biomass burning. However, it does not prove this scenario. Our modeling involves an assumed time distribution of perturbations forcing IAV and thus does not attempt to provide a unique solution. Use of a 1-box model and 1.8-year smoothing examines large-scale behavior but does not use all available information on spatiotemporal trace gas variations. More detailed studies may be possible using 3D chemical transport models constrained by higher resolution data sets that also incorporate measurements from other laboratories and perhaps other chemical and isotopic tracers.

[65] Using our approach, estimation of the relative magnitudes of globally integrated fluxes for each species provides a means of assessing possible contributions from different source/sink processes. The relationship between CO2 and δ13C implies the major fraction of CO2 variability is from terrestrial exchange. Multispecies signatures involving H2, CH4, and CO are consistent with biomass burning and eliminate other sources and sinks as sole forcing mechanisms. However, covariations of other sources/sinks with biomass burning and/or ENSO cannot be excluded. Foremost among these are net respiration of CO2 from terrestrial ecosystems and CH4 production from wetlands and rice paddies.

[66] Best estimates of anomalous H2, CO, and CH4 emissions in 1994/1995 and 1997/1998 are 3.8 and 5.3 Tg H2, 280 and 580 Tg CO, and 26 and 33 Tg CH4. If it is assumed that biomass burning is the dominant forcing of IAV in H2, CH4, and CO, these species provide a direct, quantitative constraint of CO2 emissions from biomass burning and distinguish them from any net respiration signal. CO-based budgets are not well constrained due to the short CO adjustment time. Calculations based on H2 and CH4 rely on independent knowledge of source H2/CO2 and CH4/CO2 emission ratios, for which literature estimates show a large range of variation and globally integrated 1994/1995 and 1997/1998 values are poorly constrained. Consequently, uncertainties in derived carbon fluxes are large and would be reduced if better definition of these ratios became available. Total carbon (effectively CO2 + CO) pulses from biomass burning of 0.6–3.5 and 0.8–3.7 PgC are implied for 1994/1995 and 1997/1998, respectively. These pulses represent biomass burning emissions in excess of mean levels of other years. The lower bounds are constrained by the upper limit of literature H2/CO2 and CH4/CO2 emission ratios while upper bounds represent the full terrestrial component of each pulse, as constrained by δ13C.


[67] We are grateful for the continued, able assistance of personnel at our field sampling stations. Special thanks are due to Steven Boyle of the Australian Institute of Marine Science in Townsville, Harry Tait, formerly of Lerwick Observatory in Shetland, and Tom Murray of the Institute of Terrestrial Ecology, Edinburgh, each of whom provided much of their personal time to help initiate and maintain trace gas records at sites that are not represented by permanent air monitoring stations. Staff of CSIRO GASLAB, including Lisa Cooper, Darren Spencer, Paul Krummel, Marco Lucarelli, Emily Welch, Sovatha Chea, Kate Broadhurst, Bronwyn Dunse, and Fred de Silva, contributed to trace gas analyses and management of measurement programs. Valuable comments were provided by Rachel Law, Peter Rayner, Ian Enting, and an anonymous reviewer. Data in Figure 5 come from flasks filled with the help of John Grace, Antonio Miranda, Bart Kruijt, Mark Rayment, Julie Styles, and Lins Vellen.