Comparative fire ecology of tropical savanna and forest trees

Introduction

Predicting the global distribution of biomes under changing climate and disturbance regimes requires an understanding of the factors and processes determining the current locations of the ecotones between biomes. In the tropics, the savanna–forest boundary has eluded a clear and widely accepted explanation. The savanna–forest boundary is commonly abrupt, with savanna grading into forest over distances of a few meters. Climate, fire, hydrology, herbivory, as well as soil nutrients, texture and depth are all important in determining the location of savanna and forest (Furley 1992; Hopkins 1992; Tinley 1982), but it appears that multiple factors interact to determine the location of the boundary in many cases (Frost et al. 1986; Kellman 1984). Of these factors, only fire is primarily under human control, and is the most likely to cause observable changes within the time-frame of human observation.

Fire helps maintain a dynamic balance between savanna and forest (Hopkins 1992). Savanna vegetation is very flammable, and may burn at intervals of 1–3 years, whereas evergreen forest is typically less flammable due to the dense canopy that excludes grasses and maintains a more humid understorey. Most savanna fires do not penetrate far into undisturbed evergreen forest (Biddulph & Kellman 1998; Kellman & Meave 1997), thereby sharpening the savanna–forest boundary. However, under fire exclusion, forest commonly expands into savanna (Bowman & Fensham 1991; Hopkins 1992; King et al. 1997; Swaine et al. 1992).

The savanna–forest boundary is a discontinuity not only in tree density, but also in species composition, with few species common to savanna and forest (Adejuwon & Adesina 1992; Felfili & Silva Junior 1992). Savanna species are much more fire-tolerant than forest species; in the Brazilian savannas (Cerrado), fire causes community-wide tree and shrub mortality rates of 5–19% (mean = 11%, n= 10 sites; Sato 1996; Sato & Miranda 1996; Silva 1999; Silva et al. 1996), while in Amazonian forests, fire causes mortality of 36–96% (mean = 62%, n= 9 sites; Cochrane & Schulze 1999; Holdsworth & Uhl 1997; Kauffman 1991; Peres 1999; Uhl & Buschbacher 1985), though mortality rate in a burned Bolivian dry forest was only 21% (Pinard et al. 1999). Due to a lack of comparative studies (Longman & Jeník 1992), we know little about the physiological, morphological, and ecological differences between savanna and forest species that explain this difference in fire tolerance.

The present study compares fire-related traits of savanna and gallery forest trees of the Cerrado, a region of predominantly savanna vegetation occupying approximately 2 × 10⁶ km² in south-central Brazil. Evergreen gallery forests occur as narrow corridors along streams, giving them a large perimeter of contact between savanna and forest. This narrow, dissected shape makes them particularly susceptible to the high rates of anthropogenic burning in the adjacent savannas and grasslands (Kellman & Meave 1997), and appears to have created ample opportunities for reproductive isolation and speciation, as evidenced by the high degree of adaptive radiation among woody plants. In the cerrado, the radiation of woody plants into savanna and forest species has occurred independently in a large number of genera. For example, at one site where the flora has been particularly well described, 45 of the 82 genera of savanna trees and shrubs also contain forest species occurring in gallery forests in the same ecological reserve (Pereira, Silva & Mendonça 1992), and many of the remaining genera contain forest species native to other regions. For most genera containing both savanna and forest species, we have no information on whether the forest or savanna form is ancestral, though good information does exist for the Chrysobalanaceae (Prance 1992). In this family, savanna species are most commonly derived from forest species and the divergence of savanna and forest species occurred independently in at least 14 lineages within this family (Prance 1992). We used this large number of independent radiations to test the following hypotheses.

Methods

Ten savanna–forest species pairs (Table 1) were chosen based on availability at the study site. Each pair consists of one savanna tree species and one gallery forest tree species of the same genus. No two genera were chosen from the same family, and all species are evergreen, except for the Aspidosperma spp. All comparisons were based on these pairs to ensure phylogenetic independence.

For each species, at least 10 individuals in naturally occurring populations were located within the experimental farm Fazenda Água Limpa, of the University of Brasília, located within the Federal District, Brazil. Mean annual rainfall at the site is approximately 1500 mm, with a pronounced dry season from May to September. Fire is actively suppressed at the site, but occasional fires do enter the reserve from neighbouring areas. A detailed fire history is unavailable for all locations; populations with signs of recent burning (i.e. < 5 years) were avoided.

Stem diameter and bark thickness were measured at 30 cm height. Bark thickness was measured with a bark gauge (JIM-GEM) at two points along the circumference of the main stem. For species with fissured bark, measurements were made at the ridges rather than in the troughs between the ridges.

Four individuals of each species were sampled for stem and root carbohydrate in the peak of the rainy season (December 2002 and January 2003). In selecting these individuals, only apparently healthy individuals with a full crown of leaves were chosen. To the extent possible, shaded individuals were avoided. For forest species this meant selecting individuals at the forest edge or with the crown reaching the forest canopy. This was not possible only for M. chartacea and S. elliptica, two understorey trees occurring predominantly in moist sites along the stream in the interior of the forest.

Sapwood was sampled from the root and stem using an increment borer (4·5 mm diam.). The stem was sampled at 30 cm height and the root was sampled 15 cm below the soil surface. The samples were transported on ice from the field, and heated in a microwave oven to cease enzyme activity. Then the samples were dried to constant mass at 70 °C and finely ground. The ground material was extracted with water (20 mg dry mass mL⁻¹ distilled water) for 1 h at 100 °C. After centrifugation the supernatants were used for the determination of total non-structural soluble carbohydrates by the Anthrone assay (Scott & Melvin 1953). The insoluble residue was submitted to starch determination. After threefold extraction with 80% ethanol the starch was degraded by a heat-stable α-amylase (EC 3.2.1.1, Sigma) and amyloglucosidase (EC 3.2.1.2, Merck) into its monomer glucose which was determined enzymatically by the hexokinase-glucose-6-phosphate-dehydrogenase assay (Bergmeyer 1970). TNC content was expressed on structural dry mass (dry mass – TNC) basis.

To estimate the mean size of reproductive individuals, we relied on specimens from the herbarium of the University of Brasília. All specimens of the study species that presented fruit or flowers and were collected in the Federal District were included in the sample. If there were few such individuals for a species, then the sample was enlarged to include individuals collected in other parts of Brazil. The tree height estimated by the collector was recorded, with the assumption that collector bias was similar for savanna and tree species. The mean of 20 specimens per species were used, with the lowest number being eight (G. areolata).

Results

For every one of the species pairs, the savanna species had significantly thicker bark than the forest species (t > 2·6, P < 0·02, Fig. 1). Similarly, when compared over all species, savanna species had thicker bark (Paired t_9d.f. = 10·2, P << 0·001). On average, bark accounted for 28·5% of the stem radius of the savanna species but only 10·1% of the stem radius of the forest species, so forest species require considerably greater stem diameters to ensure stem survival during burns (Table 2). For four forest species, the fitted relationships between stem diameter and bark thickness indicate that even the largest individuals do not produce sufficiently thick bark to ensure 50% stem survival during high-intensity savanna fires. and for two of these species, G. areolata and M. chartacea, even low-intensity fires should top-kill more than 50% of the individuals in the largest size classes.

The allometric coefficient for the relationship between stem diameter and bark thickness was greater for forest species (mean = 1·27) than for savanna species (mean = 0·86; paired t_9d.f. = 2·08, P= 0·03; Table 2).

Overall, there was no difference between savanna and forest species in TNC content of roots (paired t_9d.f. = 0·62, P= 0·55) or stems (paired t_9d.f. = 1·24, P= 0·25). When examined individually, four genera exhibited differences between the savanna and forest species. For Miconia, the savanna species had greater TNC in both roots and shoots, while the savanna Guapira had greater stem TNC. In contrast, for Salacia and Ouratea, the forest species had greater root TNC than the congeneric savanna species (Fig. 2).

Over all species, TNC content was greater in roots than in stems (paired t₁₉ = 3·41, P = 0·003). Among savanna species, there was no difference between root TNC (14·1%) and stem TNC (12·7%; paired t_9d.f. = 1·42, P= 0·19), though for forest species, TNC was greater in root (12·7%) than in stems (10·1%) paired t_9d.f. = 3·95, P= 0·003).

Reproductive individuals of forest species were significantly taller than those of savanna species in eight of the 10 genera (t > 2·14, P < 0·05), and when tested over all genera (paired t_9d.f. = 3·58, P = 0·006). For forest species, the height of reproductive individuals averaged 8·0 m, compared to 3·6 m for savanna species (Fig. 3).

Discussion

There are large and consistent differences in tree height and bark thickness between savanna and forest species. The thicker bark of savanna species relative to forest species undoubtedly confers a greater ability to avoid top-kill during fire, but requires considerable investment of carbon. Investment in bark should be particularly costly for saplings in the forest environment where light is considered a limiting resource. As predicted by Jackson et al. (1999), savanna species were found here to make greater investments in bark early in development, while forest species defer this investment until later. More precisely, the allometric coefficient of the bark-diameter relationship averaged < 1 for savanna trees and > 1 for forest trees.

Despite the investment in thick bark, savanna species are unable to avoid top-kill altogether, especially for small individuals and in intense fires. Although resprouting is an essential strategy for surviving frequent savanna fire following top-kill, savanna species did not have higher TNC concentrations, contrary to predictions. Savanna species may nevertheless have a greater capacity to resprout, despite similar carbohydrate concentrations, because of greater investment in total below-ground biomass. As seedlings, the root : shoot ratio of savanna species averages 82% greater than forest species (Hoffmann & Franco 2003), probably explaining the greater fire tolerance of savanna species (Hoffmann & Moreira 2002). Field estimates of root : shoot ratios of tree-dominated cerrado have ranged from 1·03 to 2·9 (Abdala et al. 1998; Castro & Kauffman 1998), compared with 0·1–0·17 for Neotropical rainforests (Castro & Kauffman 1998; Jipp et al. 1998) and 0·42–0·84 for Neotropical dry forests (Castro & Kauffman 1998). We know of no similar data for gallery forest, but due to the relatively high incidence of fire (Kellman & Meave 1997), root : shoot ratios may be higher than for other evergreen forest types, but are not likely to be as high as in savanna. Therefore, despite similar root TNC concentrations, forest trees will have less total below-ground carbohydrate reserves available per unit of above-ground biomass, reducing resprout capacity.

The range of root TNC content (7–23%; Fig. 2) was similar to woody species with resprouting ability in other savannas (Miyanishi & Kellman 1986; Wan & Sosebee 1990) and fire-prone ecosystems (Bell et al. 1996; Bowen & Pate 1993; Canadell & López-Soria 1998), although herbaceous and subshrub species can have considerably higher concentrations (Medina & Bilbao 1991). The lowest TNC values were measured for the two Aspidosperma, the only deciduous species in the study. These low values may reflect the sampling several months after leaf flush, since deciduous species frequently exhibit considerable seasonality in TNC content (Loescher et al. 1990; Wan & Sosebee 1990).

Since bark thickness largely determines top-kill during a fire, forest tree species must be larger than savanna species to ensure stem survival. For adult individuals, this is partially compensated by the fact that reproductive individuals of forest species tend to be larger than reproductive individuals of savanna species. But when burning is frequent, the large size at reproduction presents a different problem. Given similar growth rates, species that reach maturity at small size are more likely to reach reproductive size in the period between burns (Hoffmann & Solbrig 2003), so frequent fire favours smaller growth forms (Hoffmann 1999; Hoffmann & Moreira 2002). For forest species, which have thin bark and large reproductive sizes, frequent fire may preclude the possibility of individuals reaching adulthood within the savanna environment. However, the time to reach a fire-tolerant size also depends on growth rate (Gignoux et al. 1997), so if forest species have higher above-ground growth rates, this would partially compensate for the larger size necessary to ensure fire resistance and maturity. Species with low root : shoot ratios tend to exhibit greater growth rates (Kitajima 1994), so forest species may indeed grow faster than savanna species, though this was not observed for seedlings (Hoffmann & Franco 2003).

We cannot overlook alternative explanations for the difference in tree height between savanna and forest species. Specifically, the greater competition for light in the forest environment should select for greater investment in height growth than in the savanna environment. The shorter stature of savanna species, combined with their shade-intolerant nature (Hoffmann & Franco 2003), probably explains their absence in forest.

Conclusion

Differences in species traits have important implications for the response of savanna and forest species to fire at the savanna–forest boundary. Occasional forest individuals extablishing within the savanna environment are more likely to be top-killed, may exhibit lower resprout rates, and provided they survive fire, should be less likely to reach reproductive size within the time between burns. All of these differences should place forest species at a strong disadvantage in comparison to savanna species under frequent burning, effectively restricting forest advance.

Fire can occasionally extend more than 1 km into evergreen forest (Cochrane & Laurance 2002), causing widespread mortality, and potentially shift forest retreat. With intensification of land use and increasing rural populations, this problem is likely to increase in the future.