Are Australian savannas thickening?
We demonstrate that across a large area of mesic savanna in northern Australia there was little net change in woody biomass over a decade (1994–2004), in stark contrast to a number of recent reports of very high rates of woody biomass increase in the region's savannas (Chen et al., 2003; Beringer et al., 2007; Bowman et al., 2008). This absence of a strong thickening trend is despite the decade being the wettest on record, averaging 1399 mm annually across the region, compared to a long-term average of 1210 mm (Australian Bureau of Meteorology, 2012). The absence of a strong trend is also commensurate with the findings of Lehmann et al. (2009). Although they demonstrated a statistically significant increase in tree cover across the savannas of Kakadu between 1984 and 2004, they found that the magnitude of the change was very small, just +1.2% of ground area per decade. Based on a relationship between tree cover and basal area (R2 = 0.64) provided by Lehmann et al. (2009), we estimate that their observed increase in tree cover represents an increase in tree carbon stocks of just 0.07 t C ha−1 year−1, well within the 95% confidence interval of our estimate (0.0 ±0.11 t C ha−1 year−1). The conversion of savanna woody cover to biomass using allometric equations is a well-established practice (Asner et al., 2003; Fensham & Fairfax, 2003; Fensham et al., 2003; Suganuma et al., 2006). In the preceding 20-year period (1964–1984), the trend detected by Lehmann et al. (2009) was substantially greater, at +3.7%, equating to a change in carbon stocks of 0.22 t C ha−1 year−1.
The allometric approach that we used to estimate woody biomass is the standard approach for studying woody biomass dynamics (e.g. San José et al., 1998; Tilman et al., 2000; Burrows et al., 2002; Higgins et al., 2007; Lewis et al., 2009). However, there are a large number of potential sources of error in such biomass estimates that are rarely acknowledged, let alone quantified. In our case, the largest sources of error include: (1) the inherent error associated with the species-specific and generic allometric equations used (Nickless et al., 2011); and (2) the use of a ‘generic’ relationship where species-specific relationships are not available. Species-specific equations were only available for six species for aboveground tree biomass, and none for belowground tree biomass and shrub biomass. Systematic under- or overestimation of woody biomass carbon stocks would result in an equivalent under- or overestimation of absolute change. Unfortunately, we are unable to quantify the likely extent of such systematic error in our results. This is an important limitation of our study, but one shared with the vast majority of other examinations of woody biomass change over time (e.g. Tilman et al., 2000; Burrows et al., 2002; Higgins et al., 2007; Lewis et al., 2009).
The absence of a strong trend of increasing woody biomass carbon stocks in the savannas of our study area, inferred from both our results and those of Lehmann et al. (2009), contrasts strongly with some recent reports of dramatic thickening in the same region. Bowman et al. (2008) used historical aerial photography to show that savannas at a single location in Kakadu thickened between 1964 and 2004, with tree cover increasing from 48% to 65%, equivalent to an annual increase of about 0.4 t C ha−1 year−1 in tree carbon stocks. However, the findings of the more spatially extensive study of Lehmann et al. (2009) suggest that these high rates of thickening are atypical. Attribution of drivers of the local tree cover increase reported by Bowman et al. (2008) has been subject to debate, largely owing to the unique disturbance history of that study area. Petty et al. (2007) suggested that the tree cover change was related to the site's proximity to floodplains and high densities of feral Asian water buffalo (Bubalus bubalis) in the 1960s and 1970s, before their virtual extermination from Kakadu by the late 1980s. Work in Kakadu has demonstrated that buffalo promote the growth and survival of juvenile savanna trees, by reducing the abundance of grasses, which compete with juvenile trees and provide fuel for fires (Werner, 2005; Petty et al., 2007). However, Bowman et al. (2008) found virtually no relationship between rates of tree cover change and buffalo track densities (observed in historical aerial photographs), and attributed the thickening trend to larger-scale drivers such as increasing rainfall and [CO2], as well as their interactions with changing fire regimes. It is important to note that localized changes in tree cover and woody biomass do not preclude large-scale drivers. Indeed, high spatial variation in rates of woody thickening, reportedly associated with elevated [CO2], is consistent with several recent African studies (Higgins et al., 2007; Buitenwerf et al., 2012; Higgins & Scheiter, 2012).
Several localized studies near Darwin (Fig. 1) have reported very high rates of thickening (Chen et al., 2003; Beringer et al., 2007), although these rates are difficult to reconcile with our results and we suggest they are not indicative of a long-term or wider trend. Based on repeated woody biomass inventories, Chen et al. (2003) reported a dramatic increase in woody biomass carbon stocks, at a rate of 3.1 t C ha−1 year−1, equivalent to around 6% annually. However, the timeframe of that study was ≤ 2 years, and the study sites were not burnt during that time. Over longer time-scales, occasional severe fires would destroy at least some of the accumulated biomass. In the same area, Beringer et al. (2007) used flux-tower measurements to estimate a carbon sink of 2.0 t C ha−1 year−1 over 5 years and suggested that 1.2 t C ha−1 year−1 (60%) and 0.5 t C ha−1 year−1 (15%) of this was due to increases in tree and woody understorey biomass, respectively. In our study, only 3 of 136 plots experienced either positive or negative changes in woody biomass carbon stocks as large as these values. We suggest that the study sites of Chen et al. (2003) and Beringer et al. (2007) are thickening at an atypically high rate. Cook et al. (2005) and Hutley & Beringer (2010) attributed these high rates of biomass increase at least partly to the extensive tree damage suffered during severe Tropical Cyclone Tracy in 1974, with tree populations yet to fully recover. Thus, in examining rates of carbon sequestration it is important to use regional scale data collected over sufficiently long periods to capture the effects of periodic disturbance. Short-term, local data should only be used with extreme caution if the goal is to extrapolate sequestration rates to the wider region.
In north-eastern Australia, Burrows et al. (2002) reported that a 270,000 km2 area of savanna had thickened by 0.24 t C ha−1 year−1 in aboveground live tree biomass over the 1980s and 1990s. These authors attributed the thickening to land-use changes, namely the intensification of livestock grazing and coincident reductions in fire frequencies and intensities. Fensham et al. (2003) reported substantially smaller increases in aboveground live tree biomass over a much longer time period in north-eastern Australia (around 0.13 t C ha−1 year−1), emphasizing that thickening may reflect long periods of recovery from infrequent severe droughts that caused mass tree mortality. Lending support to this hypothesis, Fensham et al. (2009) found an equivalent decrease in woody cover over the period 1990–2002, coincident with the severe drought of the mid-1990s.
With some exceptions, when the results of the longer term and large-scale quantitative studies of Australian savannas are collectively examined there is a trend of increasing woody cover over the last half-century (Fig. 7). Across these studies – excluding the results of Burrows et al. (2002) and the present study, which did not measure cover directly and have strong spatio-temporal overlap with Fensham et al. (2003) and Lehmann et al. (2009), respectively – the mean increase reported is 0.08 ± 0.05% of ground area year−1 (± bootstrapped 95% CI), significantly greater than zero. Hence, we conclude that northern Australia's savannas have thickened over the latter half of the 20th century, and that the trend is generally consistent across northern Australia. However, it is important to note that the magnitude of the trend is very small, and this may explain why it was not detected using less precise historical descriptive records (e.g. Fensham, 2008). Significantly though, the trend of increasing woody cover was similar in the mesic savannas of Kakadu (Lehmann et al., 2009) and the more xeric savannas of north-eastern (Fensham et al., 2003) and north-western Australia (Fensham & Fairfax, 2003). Periodic severe droughts and extensive tree dieback are not known to occur in Kakadu, and thus the drought explanation provided by Fensham et al. (2009) cannot account for the weak thickening trend observed over 40 years in Kakadu by Lehmann et al. (2009).
Figure 7. Changes in remotely assessed woody cover in northern Australian savannas, derived from published studies: 1. Lehmann et al. (2009), Kakadu National Park; 2. This study, coastal Northern Territory; 3. Fensham et al. (2003), central Queensland; 4. Burrows et al. (2002), central Queensland; 5. Sharp & Bowman (2004a), north-western Australia; 6. Fensham et al. (2009), north-central Queensland; 7. Sharp & Whittaker (2003), north-western Australia; 8. Fensham & Fairfax (2003), north-western Australia; 9. Sharp & Bowman (2004b), north-western Australia. Note that the cover values of 4 (Burrows et al., 2002) were shifted upwards by 2% for the sake of clarity. The sizes of the circles are proportional to the natural logarithm of the size of the study area (× 100 km2). The asterisks denote cover estimates derived from tree basal area. Details on the derivation of the cover values are provided in Appendix S1 in Supporting Information.
Download figure to PowerPoint
It is important to acknowledge that there are potential sources of error in the aerial photographic studies included in our meta-analysis. Most notably, some studies did not describe how differences in the scale of photographs were accounted for, despite this being a substantial source of error. Fensham et al. (2002) point out that increasing the scale of photographs (e.g. from 1:100,000 to 1:25,000) leads to an apparent increase in cover. Specifically, in the study of Lehmann et al. (2009), photo-scale decreased from 1:16,000 to 1:25,000 between 1964 and 1984, which could have potentially dampened the thickening trend they observed. In the cases of Sharp & Whittaker (2003) and Sharp & Bowman (2004a,b), although not explicitly stated in those papers, the scale of the photographs used was consistently 1:50,000 (B.R. Sharp, New Zealand Ministry of Fisheries, Wellington, New Zealand, pers. comm.).
A frequently cited cause of thickening in savannas is overgrazing by domestic livestock (e.g. Archer et al., 1995; Sharp & Whittaker, 2003). By reducing grass biomass, overgrazing can reduce fire frequency and intensity and increase resource availability – changes which could be expected to favour the establishment of woody plants. Given that there are reports of intensification of grazing by domestic livestock in recent decades throughout northern Australia (Garnett & Crowley, 1995; Sharp & Whittaker, 2003), this seems a plausible explanation for the observed pattern of thickening. However, the general consistency of the pattern across a range of sites with different land-use histories (i.e. conservation versus pastoralism), albeit without replication, suggests that localized changes in land use are not solely responsible, and that processes operating at regional to continental scales are also important. Plausible explanations include: (1) increased [CO2] (Bond & Midgley, 2012; Higgins & Scheiter, 2012); (2) increased rainfall, with northern Australia subject to a trend of increasing rainfall over the latter half of the 20th century (Smith, 2004); and (3) increased soil moisture due to decreased evaporation (Roderick & Farquhar, 2004).
Does fire frequency affect woody biomass?
While woody biomass was relatively stable over the period 1994–2004 across our large, 24,000 km2 study area, changes in woody biomass were clearly related to fire frequency and severity. The strong decline in woody biomass in response to severe fires is consistent with the findings of two large-scale Australian fire experiments (Kapalga: Williams et al., 1999; Munmarlary: Russell-Smith et al., 2003), where annual late dry-season fires substantially reduced tree basal area, relative to unburnt areas (by c. 6.8% annually at Kapalga and 0.2–3.8% annually at Munmarlary, where fires were less intense). Our results are also commensurate with the natural experiment of Lehmann et al. (2009), who showed that annual fires, predominantly in the early dry season, reduced tree canopy cover in Kakadu by c. 0.6% annually, relative to unburnt areas. Our results also parallel those from Africa. Brookman-Amissah et al. (1980) found that following tree clearing in Ghana, tree basal area in unburnt areas increased at about 7 and 14 times the rate observed in areas subject to annual early and late dry-season fires, respectively. The 40-year Kruger fire experiment in South Africa produced similar results, with annual fires reducing woody biomass by about 1.5% annually, relative to triennially burnt areas (Higgins et al., 2007). A number of studies have also shown the large positive effect that complete fire exclusion has on woody biomass (e.g. Brookman-Amissah et al., 1980; San José et al., 1998; Tilman et al., 2000). In sum, our results show strong concordance with published results from savannas around the world. Low fire frequencies generally lead to an increase in tree biomass, while high fire frequencies, especially of severe fires, generally lead to a decrease in tree biomass.
An important finding of this study is that observed decreases in tree biomass following severe fires are not driven by mortality of individual trees, but primarily by decreases in the rates of biomass accumulation of surviving trees (Figs 5 & 6). The large negative effect of fire on individual tree growth rates in these savannas has been reported by Murphy et al. (2010), who attributed the effect to the carbon costs of rebuilding the canopy following fire. The effect of severe fires on tree growth is complemented by the partial or complete loss of individual stems, without whole-tree mortality, following severe fires (e.g. Williams et al., 1999), leading to large decreases in whole-tree biomass. Unfortunately, we cannot distinguish between the processes of individual stem (cf. whole-tree) recruitment, mortality and growth, because individual trees, rather than stems, were tagged. In relation to the effects of fire on growth of surviving trees, an important limitation of our study is that we cannot rule out that severe fires reduce the bark thickness of trees, leading to an underestimation of growth rates and biomass following a severe fire. Simple calculations based on the stand structures observed in our study suggest that the effect of severe fires on the biomass accumulation of surviving trees (shown in Fig. 4b) could be the result of a loss of 2 mm of bark thickness due to each severe fire. We suspect that substantially less bark than this would be lost, but are unable to provide evidence to support that conclusion.
The limited role that whole-tree recruitment and mortality play in the negative relationship between fire frequency and tree biomass was also demonstrated recently by Higgins et al. (2007) using the Kruger Park fire experiment in South Africa. Unexpectedly, they found that tree density was unresponsive to fire frequency (see also Buitenwerf et al., 2012), although tree biomass was strongly limited by frequent fires. They attributed this effect to high rates of fire-induced topkill (death of aboveground parts) of savanna trees, but low rates of whole-tree mortality. Higgins et al. (2007) concluded that the frequent experimental fires limited tree biomass not by inducing tree mortality or preventing recruitment, but by restricting trees to a suppressed, juvenile state. A similar effect has been found in a South American savanna system (Hoffmann et al., 2009). Like Higgins et al. (2007) and Hoffmann et al. (2009), we observed that frequent fires strongly limit the biomass of individual trees. However, these previous studies emphasized that fire primarily limits the transition between sapling and adult size classes, as did a recent northern Australian study (Prior et al., 2010). Our results suggest that even when individuals reach small adult size classes (≥ 5 cm d.b.h.), fire plays an important role in further preventing transition to larger size classes.