Demographic legacies of fire history in an African savanna

Authors


Summary

  1. Fire is a key determinant of woody vegetation structure in savanna ecosystems, acting both independently and synergistically through interactions with herbivores. Fire influences biodiversity and ecological functioning, but quantifying its effects on woody structure is challenging at both species and community scales.
  2. Deeper insight into fire effects, and fire–herbivore interactions, can be gained through the examination of species-specific demographic and dynamic changes occurring across areas with different fire regimes in the presence of large herbivores. We used the Carnegie Airborne Observatory (an integrated LiDAR and imaging spectroscopy system) to map woody tree structure, species and dynamics over a four-year interval across two adjacent savanna landscapes with contrasting fire histories in Kruger National Park, South Africa.
  3. A history of higher fire frequency was associated with reduced woody canopy cover (17% vs. 23%) and an increased overall rate of treefall (27% vs. 18%). The landscape with a history of higher fire frequency displayed a shift in woody canopy height distribution from a unimodal curve to a bimodal pattern at the community scale, with large reductions in height classes <7 m.
  4. Differences in tree height distributions and treefall rates across sites were underpinned by species-specific responses to fire frequency. Acacia nigrescens displayed the highest rates of treefall, most likely related to elephant activity, with losses exceeding 40% in the 6- to 9-m height classes.
  5. Synthesis. Our findings indicate that fire history imparts demographic legacies not only on vegetation structure, but also on current vegetation dynamics. Current treefall rates of certain tree species are exacerbated by a history of higher fire frequency. Species-specific and context-conscious investigations are critical for elucidating the driving mechanisms underlying broader community patterns.

Introduction

Fire and herbivory are powerful top-down drivers of savanna vegetation structure, modifying tree morphology and biomass through combustion, consumption and breakage (Williams, Woinarski & Andersen 2003; Bond & Keeley 2005; Higgins et al. 2007). The individual actions of these drivers frequently result in stem and canopy damage rather than direct mortality of trees, but structural weakening by either driver can render trees more susceptible to mortality from future disturbances (Shannon et al. 2011; Vanak et al. 2012). The effects of fire and large herbivores on vegetation structure are therefore highly synergistic (Midgley, Lawes & Chamaillé-Jammes 2010), and we need clearer understanding of how these interactions shape the composition and structure of vegetation communities to conserve biodiversity and ecological functioning in savanna landscapes.

Experimental manipulations of fire regimes have demonstrated extensive impacts of fire on savanna vegetation structure (Higgins et al. 2007; Furley et al. 2008; Smit et al. 2010), suggesting that savanna tree demography is regulated by a ‘fire-trap’ - a zone of c. 3 m above-ground level in which growing trees are subjected to top-kill by fire (Hoffmann et al. 2009). Complete mortality of established woody plants is rare, as savanna fires typically spread through the semi-continuous herbaceous layer at surface level and flames are seldom carried into the canopy of larger trees. Furthermore, most species invest heavily in thermally insulating bark and extensive subterranean root reserves (Mistry 1998; Hoffmann & Solbrig 2003). Most savanna fire manipulation experiments have not detected species compositional changes (Scott et al. 2012), suggesting that these systems are structurally responsive but compositionally resilient to fire. In African savannas, however, there is growing concern that long histories of altered fire regimes (van Wilgen, Everson & Trollope 1990) may have restricted the recruitment of certain keystone species and compositional changes may emerge over longer time periods (Helm & Witkowski 2013). Savanna trees need fire-free windows of sufficient duration to escape above the fire-trap and develop into mature adults. Tree growth rates vary considerably between species and across different environmental resource conditions (Prior, Eamus & Bowman 2004; Prior et al. 2006), so identifying the fire-free window required to achieve different biodiversity management objectives is challenging. Furthermore, these responses are likely to be species specific, as suggested by a recent landscape-scale analysis of fire effects on vegetation structure in different vegetation community settings (Levick, Asner & Smit 2012).

In contrast to fire which seldom topples established trees, elephants remove branches and break stems during the course of their foraging and display activities (Laws 1970), showing strong preferences for particular tree species (Levick & Rogers 2008; Helm et al. 2009; Shannon et al. 2011). Recent investigations across experimental exclosures in Kruger National Park (KNP), South Africa, revealed that elephants preferentially topple trees in the 5- to 9-m height classes (Asner & Levick 2012). This ‘elephant-trap’ represents an additional demographic bottleneck for savanna trees in Africa. However, most herbivore-browsing activities do not result in direct mortality, but rather render trees more susceptible to future disturbances (Shannon et al. 2011). For example, elephants frequently strip the bark off favoured tree species, such as Sclerocarya birrea and Acacia nigrescens, thus removing their thermal protection and exposing them to combustion by fire (Helm et al. 2009). Understanding the individual and interactive effects of fire and herbivory thus becomes critical for informed ecological understanding and management in savannas.

The architecture of an adult tree reflects its current growing conditions and its history, as it integrates all the environmental and disturbance factors that have influenced it from seedling to maturity (Archibald & Bond 2003). Given that fire and elephants primarily exert structural modification at separate height class intervals, we can use the canopy structure of different tree species to infer the primary drivers shaping populations and communities over the longer term (decades). Three-dimensional measurements of vegetation structure are typically lacking in savannas, especially at species levels, over time and over large spatial areas, due to the inherent constraints of field data collection. The integrated use of airborne LiDAR and imaging spectroscopy can overcome some of these constraints by providing high-resolution measurement of woody canopy structure and dynamics (when a time series is collected) at the species level.

We used the Carnegie Airborne Observatory (CAO, http://cao.ciw.edu), an integrated airborne LiDAR and imaging spectroscopy system, to map the species, canopy structure and canopy dynamics of individual savanna trees over a four-year period, across two adjacent savanna landscapes with equal herbivore densities but contrasting fire histories. We examined differences in vegetation structure and treefall rates of the whole community and of individual tree species within the two sites. We aimed to determine: (i) how long-term fire history influences the demography of different tree species and (ii) how long-term fire history influences current treefall rates of different tree species.

Materials and methods

Study Site

The Nwaswitshaka river catchment is located in the south-western corner of Kruger National Park (KNP), South Africa. The granitic landscape has weathered to deep sandy soils on hillslope crests, with the clay fractions increasing downslope (Gertenbach 1983). The region falls within the broad ‘Sabie and Crocodile river thickets’ landscape classification of Gertenbach (1983); however, the vegetation is too open and tall to be considered thicket. ‘Open savanna’ or ‘wooded grassland’ are more fitting descriptions, with Sclerocarya birrea, Combretum apiculatum and Acacia nigrescens dominating the woody vegetation community (Scholes et al. 2001).

Fire management in KNP has a long history, with a policy of fire suppression replaced by prescribed burning in the 1950s (van Wilgen et al. 2004). The fist active fire management policy was implemented in 1957 and called for early spring burns (October, after first rains) on a set block basis every 3 years (van Wilgen et al. 2014). This policy was replaced in 1992 by one that allowed only natural fires caused by lightning and changed again in 2001 to a policy targeted at achieving greater heterogeneity through annual burnt-area targets from point ignitions (van Wilgen et al. 2004, 2014). Our study sites were located on either side of the Nwaswitshaka tourist road, which also functions as a firebreak running north–south across the river catchment and separates two management blocks (Fig. 1a). The area west of the firebreak has a long-term fire return interval of 4·6 years, whilst that of the eastern side is 6·6 years (Table 1). Cumulatively, the western area has burned 14 times in the 70 years prior to 2008, whilst the eastern area burned 10 times (Fig. 1b). The fire history of these sites diverged most strongly in the 1970–1990 period, with fires occurring almost every 3 years in the western site. Both areas burned once during our study period in 2010. We selected 500 ha of topographically similar landscape in both the western (higher fire frequency) and eastern (lower fire frequency) areas for our vegetation structural analyses.

Table 1. Ecological characteristics and fire history of the two adjacent study landscapes
  Lower fire frequencyHigher fire frequency
  1. a

    Elephant density data derived from aerial surveys spanning 1985–2012 and calculated for a single 10 km2 block encompassing both landscapes (Smit & Ferreira 2010).

SiteGeologyGraniteGranite
Rainfall (mm yr−1)547547
Vegetation type Acacia, Combretum, Sclerocarya Acacia, Combretum, Sclerocarya
Mean elephant density (km−2)a0·31 ± 0·420·31 ± 0·42
Max elephant density (km−2)a1·751·75
Fire history (1957 to 2008)Historical return interval (yrs)6·64·6
Time since last burn for 2008 data (yrs)21
Community structure (500 ha)Mean woody cover (%)23·216·9
Mean canopy height (m)4·85·1
Figure 1.

Overview of study site and fire history. (a) Location of study sites within Kruger National Park, South Africa. Background image is a colour-infrared composite collected by the Carnegie Airborne Observatory in April 2008 – darker red colours indicate greater photosynthetic activity. White lines demarcate the two adjacent study sites separated by a gravel tourist road and firebreak (higher fire frequency site on the western side and lower fire frequency site on the eastern side). (b) The cumulative fire history of the two adjacent sites explored in this study. Fire history was complied from historical management maps and MODIS satellite imagery. Airborne LiDAR surveying was conducted in April 2008 and 2012.

Airborne Remote Sensing

We conducted a large campaign with the Carnegie Airborne Observatory Alpha system (CAO, http://cao.ciw.edu) in KNP in April/May 2008. The CAO Alpha system consisted of three integrated subsystems - LiDAR, hyperspectral (HiFIS) and navigation (GPS-IMU)(Asner et al. 2007). The CAO HiFIS subsystem provided spectroscopic images of the land surface using a push-broom imaging array with 1500 cross-track pixels, and sampling across the 367–1058 nm range at 9·4-nm spectral resolution. Radiance data from the imaging spectrometer were converted to surface reflectance using ACORN 5BatchLi (Imspec LLC, Palmdale, CA, USA) with a MODTRAN look-up table to compensate for Rayleigh scattering and aerosol optical thickness. The reflectance data were adjusted using a kernel-based BRDF model to correct for cross-track reflectance gradients due to differences in view and illumination angles (Colgan et al., 2012). The spectrometer subsystem was fully integrated with a waveform LiDAR subsystem having an adjustable laser pulse repetition rate of up to 100 kHz. The CAO LiDAR subsystem provided 3-D structural information on vegetation canopies and the underlying terrain surface. The GPS-IMU subsystem provided 3-D position and orientation data for the CAO sensors, allowing for highly precise and accurate projection of HiFIS and LiDAR observations on the ground.

We repeated the LiDAR survey in April 2012 with the CAO-2 AToMS sensor package (Asner et al. 2012) to explore changes in tree height over time and space. For both campaigns, the CAO data were collected from 2000 m above-ground level, providing airborne measurements at 1·12-m spatial resolution. The flights were conducted within 2·5 h of solar noon. HiFIS, LiDAR and GPS-IMU data were processed together to identify woody, herbaceous and bare soil based on their unique spectral and structural properties.

Data Analysis

Canopy cover and height distribution

The large number of LiDAR points collected at high pulse rates enabled the analysis of vegetation vertical structure through the rendering of pseudo-waveform profiles (Weishampel et al. 2000). The vertical distribution woody canopy was represented by binning LiDAR points into volumetric pixels (voxels) of 5 × 5-m spatial resolution and 1-m vertical resolution. The LiDAR derived ground elevation was used to standardize the vertical datum of each voxel. The height of each voxel within the vegetation canopy was defined relative to the ground at the horizontal centre of that voxel. The number of LiDAR points in each voxel was divided by the total number of LiDAR points in that column, yielding the percentage of LiDAR points that occurred in each voxel height layer. The mean number of LiDAR returns per voxel was calculated to represent the three-dimensional structure of the vegetation layer for the higher and lower fire frequency sites. The proportion of LiDAR ground returns to top-of-canopy returns was used to calculate percentage woody canopy cover across the study site.

Individual tree crown analysis and species classification

We derived a normalized canopy height model from the LiDAR data by differencing the top-of-canopy and ground return elevation models. Individual tree crowns were delineated in the 2008 canopy height model using the watershed segmentation algorithm implemented in SAGA-GIS (v2·1, http://saga-gis.org). A total of 33 313 crowns were delineated over the entire study area, 14 243 in higher fire frequency landscape and 18 968 in the lower fire frequency landscape. Crown area and height were calculated for each crown polygon from the 2008 canopy height model. The height of each crown was recorded as the 90th percentile height of the LiDAR returns within each polygon. We traced these same crown polygons through time and recalculated crown height in 2012. We measured tree height dynamics by differencing the height of crowns in 2012 and 2008. We considered a tree to have been completely toppled if its crown height decreased by more than 90% between 2008 and 2012.

We used the tree crown classification model described in Baldeck et al. (2014) to assign species identifications to each tree polygon. The model was constructed from the CAO-observed spectral signatures of 742 trees that were identified to species and geo-located in the field. Following atmospheric correction of the spectra, 54 of the CAO spectral bands were used ranging from 517 to 1017 nm. A classification model was built to identify crowns to one of 15 species: Acacia nigrescens/Acacia burkei, Acacia tortilis, Combretum apiculatum, Combretum collinum, Combretum hereoense, Combretum imberbe, Croton megalobotrys, Colophospermum mopane, Diospyros mespiliformis, Euclea divinorum, Philenoptera violacea, Spirostachys africana, Salvadora australis, Sclerocarya birrea/Lannea schweinfurthii and Terminalia sericea. Croton megalobotrys, C. mopane and S. australis are not present in south-western Kruger but were included in the original model so that it could be applied to other parts of KNP for future research. Another class, ‘other’, was included in the model to contain all additional species identified in the field so that the classification was not constrained to one of the 15 named species classes. The species classification model consisted of two stacked support vector machines (SVM) in which the first SVM classified each pixel based on its spectral signature and the second SVM integrated the output of the first SVM plus the crown height and area for a final classification of each crown polygon. Cross-validation tests performed at the crown level indicated that the SVM model classified crowns with ~76% accuracy.

Statistical analyses

Differences in canopy height distribution between areas of higher and lower fire frequency were compared with a bootstrapped Kolmogorov–Smirnov (K–S) test consisting of 1000 resampling iterations. For community-level analyses, we used the full 500 ha of canopy model height data available under each fire history scenario. Species-specific analysis of the difference in the crown height distribution between sites was conducted for three prominent species in the study area: Acacia nigrescens (22% of total mapped crowns), Sclerocarya birrea (14% of total mapped crowns) and Combretum apiculatum (14% of total mapped crowns). For each species, 1000 individuals were randomly selected from each fire scenario and a K-S test was performed (n = 6000).

Effect sizes were calculated for the difference in the proportion of voxels or individuals in each height class and were represented as loge ratios of the frequency/proportion of occurrences in that class under higher and lower fire frequency. We used two-way analysis of variance to examine the main and interactive effects of fire history and species on changes in tree height over time for the 1000 randomly selected individual crowns of A. nigrescens, S. birrea and C. apiculatum in each site (n = 6000). We used a paired t-test to examine differences between fire histories in the number of treefalls per height class for the same 1000 randomly selected individual crowns of A. nigrescens, S. birrea and C. apiculatum in each site. Nearest neighbour analysis (Clark & Evans 1954) was used to test for randomness in the spatial distribution of treefalls in both study sites. All statistical analyses were performed in r (http://www/R-project.org), and support vector machines were constructed with the ‘e1071’ r package (Dimitriadou et al. 2011).

Results

Fire History and the Demography of Woody Species and Communities

In 2008, the landscape with a higher fire frequency over the previous 70 years contained less woody canopy cover (17% cover) than that with a lower fire frequency (23% cover), and we found large disparities in the vertical structuring of that cover between the two sites (Fig. 2). The vertical distribution of woody canopy cover was unimodal under lower fire frequency, with a broad peak between 3–7 m (Fig. 2a). This distribution differed significantly from that under higher fire frequency (bootstrapped K–S test, < 0·001), where the distribution peaked in the 6–8 m range, and the pattern was more bimodal (Fig. 2a). A history of higher fire frequency resulted in a lower proportion of woody canopy in the 1- to 7-m height classes, but was associated with a higher proportion of cover in the 7- to 15-m height classes (Fig. 2b).

Figure 2.

Community-level structural differences derived from airborne LiDAR collected in 2008. (a) Normalized vertical distribution of canopy under higher and lower fire frequency. (b) Effect size of fire frequency on the proportion of LiDAR returns occurring in different height classes (the effect size represents the difference in the proportion of voxels in each height class as a loge ratio of the frequency of occurrences in that class under higher and lower fire frequency).

There were significant structural differences between the two different fire histories for all three species examined. A. nigrescens displayed subtle but significant structural differences between fire histories (D = 0·067, = 0·02246), and the tree height distributions were centred on the taller height classes of 9–10 m (Fig. 3a). We found pronounced structural differences in the height class distribution of S. birrea between the two fire histories (D = 0·248, < 0·0001), with fewer individuals in the 5- to 8-m height classes under higher fire frequency (Fig. 3b). Differences in C. apiculatum height distribution between fire histories were also significant (D = 0·227, < 0·0001) and differed from the other species in that there were a higher proportion of individuals in the 3- to 4-m height class under higher fire frequency (Fig. 3c). The effect size for C. apiculatum showed a more linear trend with increasing canopy height that differed from the complex patterns observed for A. nigrescens and S. birrea (Fig. 4).

Figure 3.

Species-specific structural responses to long-term fire history. Height class distribution of 1000 individuals of (a) Acacia nigrescens, (b) Sclerocarya birrea and (c) Combretum apiculatum in landscapes with different long-term fire histories. Black bars represent the landscape with a history of higher fire frequency; white bars represent the landscape with a history of lower fire frequency.

Figure 4.

Variation in effect size by tree height class for Acacia nigrescens, Sclerocarya birrea and Combretum apiculatum (the effect size represents the difference in the proportion of individuals in each height class as a loge ratio of the frequency of occurrences in that class under higher and lower fire frequency).

Long-Term Fire History, Current Structural Dynamics and Treefall

The mean change in tree height between 2008 and 2012, for the 6000 subsampled trees, was −1·7 m (SE ± 0·38). Two-way analysis of variance yielded a significant main effect for species (F(2,5997) = 147·0518, < 0·0001), but not for fire history (F(1,5998) = 0·3337, = 0·5635). However, the interaction effect of species and fire history was significant (F(2,5994) = 30·7917, < 0·0001), which indicates that fire history is an important driver of tree height dynamics for some species. In addition to changes in height, we also explored the dynamics of entire treefalls and found that 22% of the subsampled trees were completely toppled during the four years of our study. Acacia nigrescens experienced the highest rates of treefall, with 40–64% reductions in the number of individuals in the 6- to 9-m height classes. Treefall rates were significantly higher in the landscape with a history of more frequent fire for A. nigrescens (t(15) = −2·57, = 0·021, Fig. 5a) and S. birrea (t(15) = −2·20, = 0·044, Fig. 5b), especially in <9-m height classes. We observed no significant difference in treefall rates between sites for C. apiculatum (t(14) = 0·93, = 0·367, Fig. 5c). Spatial point pattern analysis of all fallen trees showed significant deviation from spatial randomness, with clumped distributions in both higher (= −19·5, < 0·0001) and lower (= −15·7, < 0·0001) fire frequency sites, indicating a spatially correlated cause.

Figure 5.

Rate of treefall per height class over four years for three common tree species in south-western Kruger National Park – (a) Acacia nigrescens, (b) Sclerocarya birrea and (c) Combretum apiculatum. Solid dots represent rates in the landscape with a history of higher fire frequency; open dots represent rates in the lower fire frequency landscape.

Discussion

Fire history markedly influenced woody structure in the semi-arid savanna landscape we investigated; and structural modifications were species specific. In addition to reduced woody cover and modified vertical profiles, the landscape with a history of higher fire frequency was also associated with higher rates of treefall under present conditions (when fire occurrence was constant). These findings illustrate how contemporary dynamics are influenced by historical context.

Long-Term Effects of Fire Frequency on Woody Community Structure

Savanna vegetation structure is well known to be heavily impacted by fire in the 1- to 3-m height classes (Bond & Keeley 2005; Higgins et al. 2007; Smit et al. 2010; Werner & Prior 2013). This so-called ‘fire-trap’ is the zone in which growing trees are still well within the flame zone during fire events and are thus particularly vulnerable to combustion and top-kill. Indeed, we found a substantially lower proportion of woody canopy in the 1- to 3-m height classes in the landscape with a history of more frequent fire. However, the impact of higher fire frequency on woody canopy was not restricted to this range but was present higher up the vertical profile to a maximum of 6 m (Fig. 2a,b). Trees in the 3- to 6-m height classes are unlikely to have been removed by fire, and this pattern likely rather stems from a time-lag effect. Frequent fires reduce canopy in the 1–3 m classes and prevent recruitment into the next stages. Therefore, the difference in canopy between higher and lower fire frequency sites may be propagated higher up the vertical profile over time.

Higher fire frequency did not only exert suppressive effects on woody canopy cover. At height classes >7 m, we found higher proportions of canopy cover under higher fire frequency (Fig. 2a,b). We expect rapid height gain to be selected for in areas of high fire frequency (Archibald & Bond 2003), as it increases the probability of shoots advancing above the fire-trap and developing into mature canopies. However, once safely out of the fire-trap, there is seemingly little benefit to be gained from continuing to invest in growing taller. In fact, trees investing too heavily in vertical growth may compromise on mechanical stability or hydraulic efficiency (Moncrieff et al. 2013). The promotion of taller canopy under higher fire frequency has been previously noted in this region though the mechanisms underlying these observed patterns still need to be explored (Levick, Asner & Smit 2012). Nonetheless, our species-level investigations help provide more insight into these community-level patterns.

Species-Specific Responses to Long-Term Fire Frequency

Species-level patterns indicated that the higher proportion of tall canopy under higher fire frequency is attributable in part to S. birrea, which contained a greater proportion of individuals in all height classes >9 m in the higher fire frequency area (Fig. 3b). Below 9 m, however, we found a much lower proportion of individuals in the higher fire frequency landscape (Fig. 3b). Population decline of S. birrea is of major conservation concern in KNP (Helm & Witkowski 2013), and a number of thorough field surveys have reported on the lack of individuals in recruiting height classes of 4–8 m (Helm et al. 2011; Helm & Witkowski 2012). This ‘missing height class’ problem has been attributed to the combined effects of fire and elephants (Shannon et al. 2011; Vanak et al. 2012). The presence of S. birrea in these height classes inside a long-term elephant exclosure in northern KNP (Jacobs & Biggs 2002a,b) has provided strong evidence in support of elephants being primarily responsible for this bottleneck. Here, we show that under equal herbivore/elephant pressure, a change in fire frequency can shift the population structure of S. birrea from a stable demography to one skewed toward large adults with a relative scarcity of medium-sized individuals (Fig. 3b). That is not to suggest that elephants are not important drivers of S. birrea population changes, but rather that fire is a key independent influence and that longer fire-free windows (>6·6 years as per our low fire frequency site) are needed for intermediate size classes to persist. However, it is important to note that the fire history of the two sites we explored has been very similar since the early 1990s, and it was in the period from 1970–1990 that the regime differed dramatically. The low fire frequency site only burned twice during this period, whilst the higher fire frequency site burned 6 times. The current scarcity of medium-sized S. birrea individuals is likely rooted in this period of frequent fire occurring every three years.

Acacia nigrescens appeared less structurally responsive to fire than S. birrea with fire effect size ranging from −0·5 to 0·5 vs. −1·0 to 1·5 (Fig. 4); however, we did not detect any individuals <6 m under either fire history. This pattern may stem to some extent from the inability of our combined LiDAR and imaging spectroscopy technique to map small individuals of this species (<3 m particularly), and targeted field-based studies should be implemented to assess the recruitment status of this species.

Structural patterns of C. apiculatum differed markedly from the other two species, whereby higher fire frequency was associated with a greater proportion of individuals in the 3- to 4-m height classes and a lower proportion in taller categories (Fig. 3c). Field studies at a nearby (~10 km) long-term fire experimental site suggest that part of this pattern may arise from poor regeneration of C. apiculatum under lower fire frequencies. Higgins et al. (2007) found that the Skukuza fire-exclusion plots lost more individuals of this species over time than plots that were burned; and hypothesized that C. apiculatum has difficulty regenerating in the absence of fire and that saplings are shade intolerant. If true, this would help to explain the patterns we observe in the 3- to 4-m height classes, but not the switch that occurs above 5 m where we find a greater proportion of individuals under lower fire frequency. Rather, we may be seeing an encroachment of C. apiculatum shrubs through repeated coppicing under more frequent fire, and it is only under lower fire frequency that individuals are able to reach the 5-m height classes and beyond.

Unfortunately, there are no long-term fire experiments in savanna that have also excluded elephants, but in coming decades, the fire-exclusion portions of the Nkuhlu and Letaba herbivore exclosure sites in KNP will help disentangle fire and elephant effects on vegetation structure.

The Influence of Fire History on Current Vegetation Dynamics

Levick & Asner (2013) reported treefall rates of 10·6% over two years in southern KNP (~5% yr−1). Here, we find that this average rate holds true over longer periods, with 22% of measured trees toppled in four years (~5% yr−1). Both sites burned once during the four years of our study in 2010; however, a single fire seldom topples adult trees with large single stems. The treefall patterns we see in the dynamics of A. nigrescens and S. birrea point towards interactions with elephants, which have been shown through a nearby exclusion study to dramatically increase treefall rates in the height classes between 5–9 m (Asner & Levick 2012). Elephants preferentially utilize A. nigrescens and S. birrea and repeated bark stripping makes even large adults susceptible to weakening by fire. Additionally, we might expect higher rates of treefall in landscapes with a history of higher fire frequency, as trees exposed to more fire are in turn more likely vulnerable to repeated damage by elephants. Acacia nigrescens and S. birrea were found to have higher rates of treefall in the landscape with a history of higher fire frequency, especially in the ‘elephant-trap’ height classes (Fig. 5). However, C. apiculatum showed similar patterns of change in both fire history settings and percentage losses were higher than we expected given that it is not a species particularly favoured by elephants. The fire occurrence in 2010 could partly account for the similar reductions in C. apiculatum in both landscapes. The multi-stemmed growth form of C. apiculatum renders it susceptible to large changes in canopy volume immediately after fires, though actual mortality rates are low due to strong resprouting abilities (Gandiwa & Kativu 2009). It is not uncommon to see mature individuals resprouting new ‘stems’ from a broken trunk after being completely knocked down.

Our findings suggest that increased fire frequency serves to magnify the effects of elephants on the keystone species of A. nigrescens and S. birrea. A growing body of research has emphasized the importance of spatial context in understanding vegetation dynamics in savannas (Asner et al. 2009; Levick & Rogers 2011; Lehmann et al. 2014), and our findings highlight the importance of temporal context as well. Long-term fire frequency imparts demographic legacies on vegetation communities, and current dynamics need to be considered in light of landscape history. The relatively ‘slow’ effects of fire on vegetation structure, which can take decades to shape communities through demographic bottlenecks, may soon be overshadowed by the ‘fast’ changes currently taking place by elephant mediated treefall in just a few years – especially if the current growth rate in protected area elephant populations is maintained. The type of species-specific monitoring that we have demonstrated here needs to be extended over larger areas to identify spatio-temporal refugia for key savanna tree species.

Acknowledgements

We thank the CAO team, particularly R. Martin, T. Kennedy-Bowdoin, D. Knapp and C. Anderson, for airborne data collection and processing. SANParks provided excellent logistical support, and we are grateful to I. Smit, N. Govender and S. MacFadyen for assistance with fire history and elephant density records. This study was supported by the Andrew Mellon Foundation. The Carnegie Airborne Observatory is made possible by the Avatar Alliance Foundation, Margaret A. Cargill Foundation, John D. and Catherine T. MacArthur Foundation, Grantham Foundation for the Protection of the Environment, W.M. Keck Foundation, Gordon and Betty Moore Foundation, Mary Anne Nyburg Baker and G. Leonard Baker Jr., and William R. Hearst III.

Data accessibility

Data for this paper are deposited in Dryad http://doi:10.5061/dryad.3v0p8 (Levick, Baldeck & Asner 2015).

Ancillary