Fire and herbivory are the main disturbances shaping the structure of savannas. In these ecosystems, the key strategies by which woody plants escape fire are either early height growth (the lanky strategy) or early bark growth (the corky strategy). We hypothesize that the dominance of each strategy in different savannas depends on the prevailing disturbance regimes. Given the importance of herbivory in afrotropical savanna, we expect woody plants in these savannas to be taller and have thinner barks (the lanky strategy) than plants in neotropical savanna where fire tends to be more intense (the corky strategy).
We compiled data on bark thickness and stem height in relation to stem diameter for afrotropical and neotropical savanna woody species and tested for differences in the allometric relationship between these two savannas with a general linear mixed model (GLMM).
Fire intensities were higher in neotropical than in afrotropical savannas. Afrotropical savanna plants were taller and had thinner barks for a given diameter than neotropical savanna plants – supporting our hypothesis that because of the contrasting disturbance regimes, the lanky strategy is more adaptive in afrotropical savannas, whereas the corky strategy is more adaptive in neotropical savannas.
Synthesis. While the lanky strategy is more associated with heavily browsed and fuel-controlled savannas, the corky strategy is associated with lightly browsed savannas that experience more intense fires. Because the relative role of disturbances varies across the globe, we suggest that the height-bark-diameter scheme is a powerful framework for understanding the ecology of many savannas.
The savanna is a major terrestrial biome, covering one-sixth of the world's land surface (Grace et al. 2006). Fire plays a fundamental role in the maintenance of savannas by regulating the stability of tree-grass coexistence (Higgins, Bond & Trollope 2000; Bond, Woodward & Midgley 2005; Bond 2008; Murphy & Bowman 2012). In these systems, fire affects the woody vegetation by topkilling juvenile trees before they become fire resistant and thus preventing them from growing to larger size classes (Bond & van Wilgen 1996; Gignoux, Clobert & Menaut 1997; Higgins et al. 2007; Hoffmann et al. 2009). This effect generates a demographic bottleneck that increases the dominance of short and juvenile woody plants (the ‘Gulliver’ syndrome; Bond & van Wilgen 1996; Higgins et al. 2007). The transition from below to above this ‘fire trap’ is a major constriction on tropical savannas (Bond, Cook & Williams 2012; Hoffmann et al. 2012) and is largely dependent on fire regimes (Gill & Ashton 1968; Gignoux, Clobert & Menaut 1997; Bond 2008; Hoffmann et al. 2012; Murphy & Bowman 2012).
Despite the generality of fire control on savannas, herbivory and climate may also have important direct and indirect effects (Bond 2008; Asner et al. 2009; Midgley, Lawes & Chamaillé-Jammes 2010; Murphy & Bowman 2012). Herbivory may generate demographic bottlenecks in a similar way to fire (Asner et al. 2009; Staver et al. 2009; Midgley, Lawes & Chamaillé-Jammes 2010; Moncrieff et al. 2011). For instance, browsers and grazers may slow the growth of trees by feeding on juvenile plants (Bond 2008; Scogings, Mamashela & Zobolo 2012) and foraging in recently burned areas (Fuhlendorf et al. 2008). In fact, the synergistic effect of fire and browsing plays an important demographic role in the structure of heavily browsed savannas (Staver et al. 2009; Moncrieff et al. 2011). The effects of grazers on woody plants may also be indirect because they reduce fuel loads by consuming biomass and favouring less-productive grazing-adapted grasses (Archibald et al. 2005; Savadogo et al. 2007; Bond, Smythe & Balfour 2008). This fuel reduction results in fires of lower intensity than in lightly grazed savannas (Archibald et al. 2005; Savadogo et al. 2007). In addition, the rainfall regime strongly controls C4 grass productivity (Fay et al. 2003; Sankaran et al. 2005) and is an important factor affecting the flammability of savannas (Higgins, Bond & Trollope 2000). This variability of factors and the wide distribution of the savanna biome results in savannas around the globe being subject to different disturbance regimes and hosting woody floras with distinct fire-escape strategies (Bond & van Wilgen 1996; Archibald & Bond 2003; Hoffmann et al. 2009; Lawes et al. 2011a; Higgins et al. 2012).
There are two main strategies by which woody plants escape from fire in savannas (Gignoux, Clobert & Menaut 1997). In the first strategy, plants escape by quickly growing in height to locate the canopy above the flame zone before the next fire (hide-and-resprout according to Gignoux, Clobert & Menaut 1997; Wakeling, Staver & Bond 2011). These plants tend to have a pole-like or lanky architecture (the lanky strategy hereafter; see Fig. 9b in Archibald & Bond 2003 for an example) and failure to grow enough during the inter-fire period results in top-killing followed by basal resprouting. Because building a tall canopy is also adaptive against large browsers, it should also be particularly important in heavily browsed savannas (Bond, Cook & Williams 2012). The second strategy includes plants allocating resources to thick heat-insulating barks that enable plants to stay and resist fire (Gignoux, Clobert & Menaut 1997; Fig. 1; the corky strategy hereafter). By investing early in insulating barks, these species protect the meristem from heat and are able to resprout epicormically after fires (Fig. 1; Medeiros & Miranda 2008; Hoffmann et al. 2009; Lawes et al. 2011a; Dantas et al. 2013; Dantas, Batalha & Pausas 2013). Plants presenting the first strategy are predicted to be taller and have thinner barks for a given diameter than plants presenting the second strategy (Gignoux, Clobert & Menaut 1997). Because of the contrasting allocation patterns, lanky plants have been hypothesized to be associated with relatively less intense fires than corky plants (Vines 1968; Gignoux, Clobert & Menaut 1997). However, while previous studies have demonstrated the importance of each strategy to fire-escape (Hoffmann et al. 2009; Higgins et al. 2012; Lawes, Midgley & Clarke 2013), none have directly linked their success to the prevailing type of consumer control in savannas.
The savannas of Africa and South America are among the most consumer-controlled ecosystems in the world and are subject to the highest fire frequencies (fire intervals between 1 and 5 years; Bond & Keeley 2005; Bond, Woodward & Midgley 2005; Pausas & Ribeiro 2013). Despite some local variability, fire intensities tend to be lower in afrotropical than in neotropical savannas because the greater herbivore pressure and drier climate in the former limit C4 grass biomass (Table 1 and Table S1 in the Supporting Information). In these less intense fire regimes, the demographic pressure exerted by herbivores plays a fundamental role in community structure (Staver et al. 2009; Moncrieff et al. 2011), and the dominance of strategies that avoid both herbivory and fire negative effects is expected. Thus, in these savannas, allocating resources to raise the canopy above the reach of fire and herbivores (the lanky strategy) should be more beneficial than building a thick corky bark. In contrast, in neotropical savannas, where the herbivory pressure is lower (Marquis, Diniz & Morais 2001), the corky strategy with epicormic resprouting should be fundamental to protecting the stem from more intense fires. Hence, we hypothesized that because of the contrasting disturbance regimes, the lanky strategy is more adaptive in afrotropical savannas, while the corky strategy is more adaptive in neotropical savannas. This model will be supported if, for a given diameter, African savanna plants have thinner barks and are taller than plants of South American savannas. To evaluate our hypothesis, we began by testing the differences in fire activity and rainfall between neotropical and afrotropical savannas using remote-sensing fire data and global climate information. Then we compiled data on bark thickness, height and diameter for a wide range of woody species growing in several neotropical and afrotropical savannas.
Table 1. The main factors distinguishing afrotropical and neotropical savannas
Full reference sources for this information are presented in Table S1.
Mean annual rainfall (mm)
Mean grass biomass (kg ha−1)
Mean fuel load (kg ha−1)
Mean fireline intensity (kW m−1)
Predicted fire-escape strategy
Materials and methods
Fire and rainfall
To evaluate the fire intensity dissimilarity between afrotropical and neotropical savannas, we complied fire information obtained from the MODIS sensor on the Terra satellite for the period between January 2001 and December 2009 (the Climate Modeling Grid; Fire Information for Resource Management System, NASA). Specifically, we compiled the monthly mean fire radiative power (in megawatts at 0.5° spatial resolution) as a measure of fire intensity (Wooster, Zhukov & Oertel 2003) for the savanna biome in Africa and South America – following the ecoregion approach proposed by Pausas & Ribeiro (2013). For each region, the mean fire radiative power of each pixel was averaged across the whole period (2001–2009). We also assessed the differential rainfall between afrotropical and neotropical savannas by compiling average annual precipitation for the same two regions from WorldClim data (0.0083° spatial resolution; Hijmans et al. 2005). We finally compared the mean and frequency distributions of the fire radiative power and annual precipitation for the two savanna regions and computed bootstrap confidence intervals for each region.
Tree data compilation
We compiled data relating stem diameter with bark thickness and stem height from 21 published articles – 7 studies on neotropical savannas and 14 studies on afrotropical savannas (see Tables S2 and S3 for details). To extract data published as figures in the articles, we used Web Plot Digitizer (Rohatgi 2012). We complemented the data set with a field sampling in Emas National Park (ENP), central Brazil (see sampling below). Overall, we considered 153 woody species (81 from afrotropical and 72 from neotropical savannas), included in 37 taxonomic families (16 and 28, respectively; see Table S4). Most of the data were at individual level, but we also included some average values. The total sample size was 2294 and 1097 for African savannas (height-to-diameter and bark thickness-to-diameter, respectively) and 1633 and 731 for neotropical savannas (height-to-diameter and bark thickness-to-diameter, respectively).
We collected data at ENP (central Brazil: 17°49′–18°28′S, 52°39′–53°10′W), a neotropical savanna reserve located in central Brazil (Dantas, Batalha & Pausas 2013). The reserve has a tropical warm wet climate, with dry winters (Aw type following Köppen 1931). Although the park includes a wide range of physiognomies – from open grassland to closed forest (Dantas, Batalha & Pausas 2013) – we only considered data from savannas and open woodlands (the same criteria used in the compilation). We performed the field sampling in the rainy season of 2009/10 on 99 5 × 5 m plots, distributed among different fire ages, representing the current fire history mosaic in the park. We identified all individuals with basal diameter ≥3 cm to species and measured bark thickness, basal diameter and stem height for each individual. Bark thickness was measured by inserting a knife in the stem at five haphazardly selected points and measuring the depth of penetration with a digital calliper. Most measurements were taken on the trunk at 0.3–0.5 m from the ground and then averaged. In some plants, the main stem was very short, bifurcating below 0.3 m; and in these cases, bark thickness was measured halfway from the base to the first bifurcation of the main stem. This sampling approach enabled the inclusion of both rare and common species but still emphasizes the predominant pattern (dominant species) as in the compiled data set. See Dantas, Batalha & Pausas (2013) for further details.
At first we compared the frequency distribution of bark thickness, height and diameter values in the two ecozones including all the compiled data (afrotropical vs. neotropical) using a Wilcoxon rank test. Because the size of the plants differed between ecozones, the subsequent statistical tests were performed considering only the range of overlapping diameters and excluding outliers (as defined by the Tukey's test, that is, excluding values >1.5 times the interquartile range). Because bark thickness and plant height are strongly related to diameter (allometrically), we fitted a general linear mixed model (GLMM) for each escape trait in relation to diameter and with ‘species’ as random effect. Diameter, height and bark thickness were log-transformed before the analysis. We then included ecozone (afrotropical vs. neotropical) as a fixed factor and tested for significant differences between the two models with a likelihood ratio (LR) test. As different studies in our data set measured diameters at different heights along the stem, we also repeated the analyses using only studies that measured basal diameters – and given that results did not change, these analyses are not reported. All statistical analyses were performed using the ‘nlme’ and ‘stats’ libraries from the R software (R Development Core Team 2012).
The remotely sensed fire data (radiative fire power) showed that more intense fires tend to be more frequent in neotropical than in afrotropical savannas (Fig. 2a). Similarly, precipitation tended to be higher in neotropical (mean = 1318 mm, CI = 1112.6–1504.9) than in afrotropical savannas (mean = 832 mm, CI = 697.0–978.63; Fig. 2b). For any of these two variables, the confidence intervals between the two regions overlap (Fig. 2). These results support the tendency observed with the sparse data set compiled from the literature (Table 1).
The analysis of the overall compiled data showed that afrotropical savanna plants were larger in diameter and height than neotropical savanna plants (P < 0.001), but they both had similar bark thickness values (P = 0.61; Fig. 3). The allometric relationship between bark thickness and diameter significantly differed between ecozones considering either plants of similar sizes (Table 2a) or the entire data set (Table S5a). Similarly, the relationship between stem height and diameter differed significantly between the two ecozones with a significant interaction effect (Table 2b and Table S5b). Inspection of the predicted values indicated that the differences between ecozones are larger for smaller woody plants, but the regression lines do not cross each other. In fact, repeating the analysis for trees of <5 cm in diameter yields a significant difference between ecozones with no significant interaction (Table 2c).
Table 2. Summary of the general linear mixed model analyses testing the differences in the relationship between bark thickness (mm) and stem height (m) against diameter (m) between ecozones (afrotropical vs. neotropical; see Fig. 3)
The analyses correspond to plants with overlapping range sizes between afrotropical and neotropical savannas (see Fig. 4); for a comparison including all data compiled see Table S5). Species were considered as random factors. Degree of freedom (Df), AIC and the likelihood ratio test (LR) are presented. For each model, the rightmost column provides the estimated parameters for the fixed effects of the final significant model. Analysis includes only the data in which the range of tree diameters overlap (see 'Materials and methods').
0 [Afrotr], 0.25 [Neotr]
+log(diam) × E
0 [Afrotr], −0.02 [Neotr]
+log(diam) × E
0.12 log(diam) [Neotr]
Height (for diam <5 cm)
0 [Afrotr], −0.20 [Neotr]
+log(diam) × E
As predicted, for a given diameter, afrotropical savanna plants are taller, whereas neotropical savanna plants have thicker barks. These patterns are consistent with our hypothesis that most afrotropical savanna plants fit the lanky strategy of quickly placing the canopy above the flame zone or, otherwise, resprouting from basal buds. In contrast, most of the neotropical savanna woody plants fit the corky strategy and their stems resist fire thanks to a thick fire-protective bark. Because large mammal herbivory and fire are important factors in many parts of the afrotropical savanna (Fuhlendorf et al. 2008; Staver et al. 2009; Moncrieff et al. 2011), pole-like trunk architectures (sensu Archibald & Bond 2003) with early height growth are more advantageous than either a cage-like architecture (common in heavily browsed low fire frequency savannas; Archibald & Bond 2003) or the corky strategy (typical of savannas with recurrent high-intensity fires; Dantas, Batalha & Pausas 2013). Moreover, because fire and herbivory selective pressure are concentrated on the sapling stage (2–3-m-tall plants; Bond 2008; Wakeling, Staver & Bond 2011), differences in height–diameter allometries are especially evident in smaller plants (Table 2). In contrast, the higher annual rainfall (Table 1, Fig. 2) and lower abundance of large mammals in South American savannas (Marquis, Diniz & Morais 2001; Costa et al. 2008; Lehmann et al. 2011) enable C4 grasses and shrubs to quickly accumulate. In these savannas, the typical fuel gap between grasses and mature trees that is observed in most African savannas is often lacking (Dantas et al. 2013; Dantas, Batalha & Pausas 2013). This results in relatively more intense fires (Table 1 and Fig. 2) that affect thin-barked plants (Hoffmann et al. 2009). In such conditions, a thick corky bark is a valuable trait for stem protection and subsequent epicormic resprouting. There is regional to local variation in fire intensity and rainfall (Fig. 2), as well as herbivory, in both afrotropical and neotropical savannas. This is a likely reason that plants with corky and lanky strategies coexist locally (Gignoux, Clobert & Menaut 1997; Rossatto, Hoffmann & Franco 2009; Lawes et al. 2011b) despite the prevailing patterns we have shown. The recent loss of herbivores from some afrotropical savannas may also have led to more intense fire regimes that are similar to those of the neotropics, perhaps resulting in novel selection pressures.
The fact that the lanky strategy dominates in afrotropical savannas does not mean that bark thickness has no role in these ecosystems; in fact, a certain bark thickness is needed to protect the stem from low intensity fires, but the relative importance of bark thickness clearly differs between the two biogeographical regions because of the contrasting fire regimes. If we consider the nearby (less flammable) forest patches embedded in the cerrado mosaics, neotropical forest trees have thinner barks than neotropical and afrotropical savanna trees (Fig. S1), suggesting that differences are ecological rather than biogeographical. Evidence also points out to the coexistence of the two strategies in some afrotropical savannas (Gignoux, Clobert & Menaut 1997). Moreover, the striking differences in size between afrotropical and neotropical savanna trees (larger in the former; Fig. 3) are not due to biogeographical differences, as in the Neotropics some trees have the same size or are even larger, than those in the African savannas (Fig. S1). However, because fires tend to be more intense in neotropical savannas, the thin barks of the lanky strategy provide insufficient protection (Hoffmann et al. 2009). Thus, development to adult size classes on these plants is often confined to protected patches of closed-canopy forests that inhibit shade-intolerant grasses and thus benefit from a considerable reduction in flammability and fire intensities (Hoffmann et al. 2011, 2012).
Thick barks have traditionally been associated with trees living in ecosystems with understorey fire regimes (Keeley & Zedler 1998; He et al. 2012). Here, we provide evidence that thick bark is also a successful fire protection strategy in fire regimes where the entire plant is affected by fire (Fig 4; Dantas et al. 2013 and Dantas, Batalha & Pausas 2013). Compared with typical crown-fire ecosystems, fires in neotropical savannas are very frequent and of relatively low intensity. In such conditions, small differences in bark thickness may provide sufficient fitness benefit to be selected. The corky strategies of having a thick bark along the whole stem and branches coupled with epicormic resprouting also appear in some nontropical oak woodlands where fires are frequent (Zedler 1995; Pausas 1997; Catry et al. 2012).
We have focused on bark thickness as a stem insulation mechanism, despite the fact that there are other stem traits that may also be relevant in reducing heat transfer, such as bark density and porosity (Lawes et al. 2011b), bark moisture (Midgley, Lawes & Chamaillé-Jammes 2010; Lawes et al. 2011b; Brando et al. 2012; Higgins et al. 2012) and stem-specific density (Brando et al. 2012). In isolation, these traits rarely predict the probability of topkill in savanna plants (e.g. Higgins et al. 2012), and there is overwhelming evidence that bark thickness is the main protective bark trait in tropical ecosystems (Pinard & Huffman 1997; Hoffmann et al. 2009; Lawes et al. 2011a; Brando et al. 2012). Although the main path by which African savanna plants escape fire seems to be related to the lanky strategy (Lawes, Midgley & Clarke 2013), the interspecific variability of bark traits may also be important in predicting species abundances at local and regional scales (Higgins et al. 2012).
Differences between ecozones could have a phylogenetic basis as the two zones share very few clades (Tables S4 and S6). However, this is not expected, at least for bark thickness, as this trait is strongly convergent (Simon et al. 2009); that is, thick barks appear in many different lineages and closely related species may have differing bark thicknesses for a given diameter (Hoffmann, Orthen & Nascimento 2003). Plant height is more evolutionary conserved (Swenson & Enquist 2009; Silva & Batalha 2010; Yessoufou et al. 2013), although there is some evidence, at least for one shared clade (Fabaceae), that tree saplings grow faster in height in afrotropical savannas (from 11 to 25 cm year−1; Wakeling, Staver & Bond 2011) than in neotropical savannas (from 3.1 to 5.52 cm year−1; Braz, Kanegae & Franco 2000; Kanegae, Braz & Franco 2000; Azevedo 2006). In addition, the fact that neotropical closed-forest have thinner bark and are taller than the afrotropical savannas (Fig. S1) also points out towards an ecological rather than a biogeographical or historical process.
Australian savannas are also subjected to recurrent fires and, because of the very long history of plant–fire interaction (Bond & Midgley 2001), species have evolved a variety of bud-insulating mechanisms (Burrows et al. 2010). Specifically, Australian eucalypts have buds embedded within the trunk (Burrows 2002) and consequently achieve stem protection with thinner barks than noneucalypt species (Lawes et al. 2011a). Because of this mechanism, eucalypts allocate little to bark thickness, and this enables a relatively fast growth in height (Prior et al. 2006). Nevertheless, the corky strategy is fundamental in noneucalypts of Australian mesic savannas (Lawes et al. 2011a; Lawes, Midgley & Clarke 2013) supporting our rationale of the importance of this strategy in mesic savannas with low herbivore densities.
While there is plenty of information about the role of fire (e.g. Keeley et al. 2011) and herbivory (e.g. Díaz et al. 2007) in shaping plant traits, their interactive effects have received little attention (Pausas & Lavorel 2003; Midgley, Lawes & Chamaillé-Jammes 2010). Savannas are a prevalent example of consumer-controlled ecosystems (Bond & Keeley 2005; Bond, Woodward & Midgley 2005), and we propose that the relative importance of the different consumers (fire and herbivory) may determine the dominant fire-escape strategy and community structure. Given that the relative role of consumers varies across the globe, our model might help to explain the contrasting strategies of savanna plants and the structure of savanna ecosystems world-wide.
The authors are grateful to the Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP; process: 2010/01835-0); the Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES; process: 1019-11-2); and the Spanish Government (VIRRA and TREVOL projects, CGL2009-12048/BOS and CGL2012-39938-C02-00) for financial support and the scholarships granted to the authors; Instituto Brasileiro do Meio Ambiente e dos Recursos Naturais Renováveis (IBAMA) for research permission; Emas National Park staff for logistical assistance; M.A. Batalha, M.V. Cianciaruso, N. Escobar, P.P. Loiola, N. Rosatti, D.M. Silva, I.A. Silva for valuable help in field sampling; M.A. Batalha, M.V. Cianciaruso and I.A. Silva for helping with species identification; and E. Ribeiro for helping in the GIS analysis. The authors have no conflict of interest to declare.