Low tolerance of New Caledonian secondary forest species to savanna fires

Abstract

Questions

How do early secondary successional forest species that grow in savannas differ in their tolerance to surface fires? What are the consequences of these fire tolerances for savanna–forest dynamics and landscape management?

Location

Anthropogenic savannas in the New Caledonian biodiversity hotspot (SW Pacific).

Methods

We estimated the range of fire intensity in New Caledonian savannas using field survey of fuels and the BehavePlus fire behaviour model. Within the predicted range of fire line intensity, we assessed theoretical fire injury to the cambium and crown for 11 species: the dominant tree of New Caledonian savannas (Melaleuca quinquenervia) and early secondary successional forest species. Using empirical models, for each species we estimated cambium damage from depth of necrosis (as a function of fire line intensity and fire residence time) and bark thickness, and crown damage from scorch height (as a function of fire line intensity) and tree height. We compared bark thickness and tree height increment patterns among species as well as species potential fire tolerance.

Results

The 11 species had very contrasting capacity to avoid fire injury to the bole cambium due to differences in bark investment patterns, but were all very exposed to scorching and crown injury. Overall, most of sampled individuals are likely top-killed by low intensity fires (<1000 kW·m⁻¹), which are frequent according to our simulations.

Conclusions

The early secondary successional forest species growing in New Caledonian savannas are poorly adapted to fire, in comparison with literature on worldwide trees in savannas. As a result, their juveniles are unlikely to reach adult size in fire-prone areas. Restoration using the most fire-tolerant species and fire prevention may be complementary strategies to manage such tropical landscapes in order to conserve biodiversity and ecosystem services.

Nomenclature

FLORICAL

(Morat et al. 2012)

Introduction

Savannas are characterized by a continuous grass layer with a discontinuous tree cover (Scholes & Archer 1997). These grasses, which constitute fine and well-aerated fuel in a microclimate that promote rapid curing, make savannas highly flammable (Biddulph & Kellman 1998; Stott 2000; Hennenberg 2006) . Moreover, their rapid post-fire re-growth promotes high fire frequency (Biddulph & Kellman 1998; Hély & Alleaume 2006). As a result, surface grass fire is one of the main disturbances, together with grazing, that interact to limit trees in savannas, allowing trees and grass to co-exist (Bond & Keeley 2005; Bond 2008; Staver et al. 2011). In New Caledonia (Jaffré et al. 1998; Ibanez et al. 2012), as in others places in the tropics (e.g. Cavelier et al. 1998; Veldman & Putz 2011), forests exposed to anthropogenic and environmental changes may shift to savanna, causing large losses of biological, ecological and economic resources (Scheffer & Carpenter 2003; Folke et al. 2004). Assessing fire tolerance of woody species growing in savannas is critical to understand forest–savanna dynamics and to manage and conserve biodiversity and associated ecological services.

During surface grass fires, fuel combustion transfers heat to the trees and injures exposed roots, boles, stems and crowns (see physical processes described in Michaletz & Johnson 2007). The degree of these injuries is determined by both fire behaviour and tree characteristics (Whelan 1995). Fire line intensity and fire residence time – describing the rate of heat or energy release by the fire front and the time of heating, respectively – are important fire behaviour characteristics widely used to predict stem mortality (e.g. Ryan & Reinhardt 1988; Williams et al. 1999; Michaletz & Johnson 2008). Bark thickness and height of the tree crown are considered as the most important fire-related traits because both insulate critical tissues from heat and are used as predictors of tree mortality (e.g. Pinard & Huffman 1997; Higgins et al. 2000; Hoffmann & Solbrig 2003; Bond 2008). Indeed, thick bark insulates the vascular cambium (lateral meristem) that produces vascular tissue (Gashaw et al. 2002; van Mantgem & Schwartz 2003; Dickinson & Johnson 2004; Bauer et al. 2010). Similarly, a high crown base removes the vegetative buds (apical meristems) that produces branches, buds, foliage and reproductive organs (van Wagner 1973; Michaletz & Johnson 2006) that are beyond fire heat effects, while below-ground meristems are insulated by the soil.

Bark thickness and tree crown base height constitute defences against fire that are largely dependent on stem diameter, so that large (mature) individuals will be less injured by fire than small (juveniles) ones (e.g. Hoffmann et al. 2003, 2009). Thus, during a savanna fire most small individuals are top-killed (i.e. their above-ground biomass is killed, including the crown) while the root system often survives and allows most of them to persist through basal resprouting (Bond & Midgley 2001). Juveniles have to reach the so-called ‘escape size’ that allows them to avoid fire top-kill and recruit to the mature size; however, in savannas that can burn every year, they may be caught in a ‘fire trap’ if the interval between two fires is too short to reach the necessary size (Higgins et al. 2000; Schutz & Bond 2009).

Confronted with such a fire disturbance regime, trees can either invest in growth to reach their ‘escape size’ and resist fire, or invest in below-ground reserves to allow them to resprout after top-kill and be resilient to fire (Gignoux et al. 1997; Vesk 2006; Schutz & Bond 2009). Although resprouting may allow species to persist in a fire-prone ecosystem, the time required to reach escape size, thus potentially allowing a species to reach its mature size in fire-prone ecosystems, is a critical trait (Gignoux et al. 2009).

Allometric relationships between stem diameter and sizes of fire-related traits (i.e. bark thickness and tree crown base height) offer a good framework to characterize and analyse differences in patterns of defence investment among species (Jackson et al. 1999). Escape sizes are widely determined by post-fire conformation of fire-related traits, and top-kill or mortality via survival analysis (e.g. Hoffmann & Solbrig 2003; Bond 2008; Lawes et al. 2011a). However, in ecosystems that have high fire frequency, post-fire mortality data may be difficult to obtain on a wide range of tree species and on stem diameters, as repeated fires limit the number of trees. Moreover, as mentioned above, top-kill, or mortality probability, is a function of tree characteristics and fire behaviour, the last being unknown for most post-fire studies.

In this study, we explored the fire tolerance of 11 woody species growing in New Caledonian open savannas and encompassing the dominant savanna species (Melaleuca quinquenervia) as well as early secondary forest species. We used an original combination of allometric relationships and models of fire behaviour with fire injuries to assess how these species differ in their tolerance to surface fires. We hypothesized that the fire tolerance of the studied species is highly variable due to species-specific traits that allow species to resist or avoid fire effects, e.g. differences in bark or height investment. We discuss the impacts of these differences in fire tolerance for savanna–forest dynamics and landscape management.

Methods

Study site

New Caledonia (NC), located in the south Pacific just above the Tropic of Capricorn, 1500 km east of Australia and 2000 km north of New Zealand (20–23°S, 164–167°E; Fig. 1), is one of the smallest (ca. 19 000 km²) world biodiversity hotspot (Myers 1988; Mittermeier et al. 2004). Its flora is particularly rich in relation to the little size of the archipelago, with 3371 native vascular species and a high level of endemism, reaching 75% (Morat et al. 2012), supported by a specific geological history (Grandcolas et al. 2008) and a mosaic of ecosystems linked to differences in edaphic substrate, rainfall and human activity (Jaffré et al. 1998; Pillon et al. 2009). However, more than half of the original vegetation has already been destroyed (Jaffré et al. 1998). Rain forest, now largely replaced by anthropogenic formations such as savannas, is rare and fragmented at low and middle elevations, and is therefore restricted to high elevations or inaccessible areas (Jaffré & Veillon 1994; Jaffré et al. 1998). Bush fires maintain the savannas and erode the rain forest edge, and are one of the main threats to New Caledonian biodiversity (Pascal et al. 2008).

Figure 1. Location of study area. Sampling was performed on the western side of Aoupinié (1006 m a.s.l.) between the lands of the Gohapin tribe and Poya city in the middle of the New Caledonian main island. Grey areas represent savannas (from SGT 2008) and hatched areas represent the ultramafic substrate encompassing peridotite and serpentinite (from Fritsch 2011).

New Caledonian savannas are dominated by M. quinquenervia (Myrtaceae), locally known as niaouli, which cover ca. 30% of the drier west coast of the main island (Fig. 1). Sampling was performed in savannas located on the west side of Aoupinié Mountain in lands of the Gohapin tribe, which receive 1500–3000 mm rainfall per year (METEO-France 2007). The landscape is a mosaic of remnant forests and savannas (Ibanez et al. 2012). The climate of NC is tropical oceanic with a cool and a hot season, both overlapping with a characteristic dry season from August to November. The dry season, characterized by low rainfall, increasing temperature and alizé east trade winds blowing almost permanently, corresponds to the bush fire season.

Species selection

From a preliminary field survey and community analysis (Ibanez 2012), we selected 11 species (six endemics and five non-endemic but natives species; Table 1) occurring in NC savannas. They comprise the dominant savanna tree species (M. quinquenervia) and early secondary forest species, consisting of both tall and small tree species. The latter grow in open areas and initiate forest re-colonization in savannas.

Table 1. Studied species

Family	Species	Code	Statusa	Biological type
^a EE, endemic genus; E, endemic species; N, non-endemic native species.
Apocynaceae	Alstonia costata (G. Forster) R. Brown	AC	E	Tree
Apocynaceae	Tabernaemontana cerifera Pancher & Sebert	TC	E	Small tree
Cunoniaceae	Codia albicans Vieill.	CA	EE	Tree
Cunoniaceae	Geissois racemosa Labill.	GR	EE	Tree
Loganiaceae	Fagraea berteroana A. Gray	FB	N	Small tree/Epiphyte
Moraceae	Ficus habrophylla Bennett ex Seemann	FH	N	Small tree
Myrtaceae	Melaleuca quinquenervia (Cav.) S.T. Blake	MQ	N	Tree
Pittosporaceae	Pittosporum simsonii Montrouz.	PS	E	Tree
Rutaceae	Acronychia laevis Forster &G. Forster	AL	N	Small tree/Shrub
Sapindaceae	Elattostachys apetala (Labill.) Radlk.	EA	N	Tree
Sapindaceae	Guioa villosa Radlk.	GV	E	Small tree

Fire resistance traits and allometric relationships

We selected and measured four morphological traits that affect fire tolerance: bark thickness (BT) that affects cambium injury, tree height (TH) and crown base height (CBH) that both affect tree crown injury, and the associated diameter at breast height (DBH). Thicker bark and higher crown bases are often associated with higher bole diameter and should reduce fire injury (Michaletz & Johnson 2007).

We measured these morphological traits on individuals located in savannas or open early secondary successional forest that did not show recent fire damage. BT was estimated as the mean of two bark gauge measures (1 mm accuracy) at two random opposite points at breast height (1.3 m). For individuals with fissured bark, measurements were made at the ridges. TH and CBH (i.e. height of the lowest branch of the crown) were measured using an ultrasonic hypsometer (Vertex 4) with 0.1 m accuracy.

We computed allometric relationships between DBH and BT, TH or CBH to compare species investment in these morphological traits throughout their growth using the following equation:

• (1)

where β is a proportionality coefficient, α is the allometric coefficient and ε is the error. The allometric coefficient α can be interpreted as an allocation coefficient; i.e. if α > 1 (positive allometry) the investment in bark or height growth is disproportionally higher at large DBH, if α < 1 (negative allometry) the investment in bark or height growth is disproportionally higher at small DBH, and if α ≈ 1 investment in bark or height growth is proportional to the increment in DBH (Jackson et al. 1999). We used least-squares regression to fit allometric relationships on log-transformed data and to estimate α.

We used Spearman correlation tests on ranked data to test across species, the correlation between α_BT (which conveys the rapidity or slowness of bark production) and normalized bark thickness (NBT), i.e. BT as a percentage of stem radius (which conveys the importance of bark production).

Fire behaviour in New Caledonian savannas

Fire-induced injury is a function of both tree characteristics (morphological traits measured and described above) and fire behaviour. Given that fire behaviour in NC savannas has not been measured in the field, we chose to characterize it using the BehavePlus 4.0.1 fire behaviour model (Andrews et al. 2008). BehavePlus calculates fire behaviour variables such as fire line intensity (I), rate of spread (ROS) and flame length (FL) from fuels, weather and topographic conditions. Fuels were implemented with field data from 29 NC savanna stands (Table S1, from Hély, C., Tinquaut, F., Ibanez, T., Curt, T., Géraux, H. & Finney, M. in preparation) from different localities on the main NC island. Then, fire behaviour characteristics (I, ROS and FL) were calculated for the 29 savanna fuel models using wind speeds from 0 to 30 km·h⁻¹, slopes of 5% and 30% (ca. 90% of the NC savannas are present on 30% slopes or less) and dry and medium fuel moisture scenarios (Table S2).

Estimated fire injury

Potential fire injury to the bole cambium (depth of necrosis) and tree crown (scorch height) was estimated from fire line intensity (I) using empirical models. Previous experimental studies have shown that although inter-species variation in bark moisture, density, structure or composition may affect the insulating capacity of bark, bark thickness is the main parameter determining fire resistance (Gashaw et al. 2002; Dickinson & Johnson 2004; Bauer et al. 2010; Lawes et al. 2011b; Brando et al. 2012). Thus, the depth of necrosis (i.e. the depth where T > 60 °C) is widely estimated from data on the heat source (e.g. Peterson & Ryan 1986; Michaletz & Johnson 2008). We estimated the depth of necrosis (D_n in mm) using the model developed by Bova & Dickinson (2005) as it combines both I (in kW·m⁻¹) and fire residence time (R_t, in s) as follows:

• (2)

We estimated D_n and explored the tolerance of trees to fire with assumed R_t of 60 s (Fig. S1). This R_t is reasonable according to the few available data on R_t in savanna fires (Stocks et al. 1996; Savadogo et al. 2007).

The scorch height (H_sc, in m) was estimated using the van Wagner (1973) non-linear relationship between H_sc and I:

• (3)

where k is an empirical coefficient measured from I and H_sc data. As we had no data on scorch height from NC fires, H_sc was estimated with k = 0.154 (Fig. S1). This k parameter was approximated from the empirical relationship between H_sc and I (H_sc = 21.2–17.6 × e^{(−0.000287 × I)}) observed during fires in northern Australian savannas, which involve eucalypt trees (Kapalaga experiment, Williams et al. 1998).

Results

According to the BehavePlus simulations, savanna trees were likely exposed to a wide range of fire line intensity (Fig. S2). Indeed, simulated values of I ranged from <10 to ca. 2000 kW·m⁻¹ (95% CI) under the lowest wind speed condition, and from ca. 75.0 to ca. 13 500 kW·m⁻¹ under the highest wind speed, with a rate of spread ranging from 0.1 to 3.4 and 0.8–27.4 m·min⁻¹, respectively, and flame length from 0.2 to 2.5 m and 0.5–6.1 m, respectively.

Species and DBH explained fire-related trait variability

The morphological traits involved in fire tolerance (particularly bark thickness) varied significantly among species (Table 2). Indeed, species and DBH together explained ca. 75% of the variance observed in BT and TH, while they explained <50% of the variance in CBH. However, note that the species, which affects the architecture and development of the canopy, explained ca. 30% of the variance in tree CBH. Species also explained more of the observed variance in BT than did DBH (48.3% and 24.1%, respectively), while the opposite was found for variance in TH (20.1% and 53.9%, respectively).

Table 2. Covariance analyses on fire tolerance morphological traits based on tree species and diameter at breast height (DBH)

	df	Sum of Sq.	% Var.	f-value	P-value
*P < 0.05; **P < 0.01; ***P < 0.001.
Bark thickness
Species	10	15.2826	48.3	78.0235	***
ln(DBH)	1	7.6218	24.1	389.1228	***
Interaction	10	0.8343	2.6	4.2593	***
Residuals	405	7.9328	25.0	–	–
Tree height
Species	10	6.4702	20.1	67.8978	***
ln(DBH)	1	17.3099	53.9	1816.4864	***
Interaction	10	0.2664	0.8	2.7951	**
Residuals	848	8.0809	25.2	–	–
Tree crown base height
Species	10	6.0716	29.8	27.5298	***
ln(DBH)	1	3.0515	15.0	138.3615	***
Interaction	10	0.5159	2.5	2.3391	*
Residuals	488	10.7627	52.7	–	–

Bark thickness, investment strategy and bole cambium injury

The 11 species had very contrasting capacities to avoid potential fire injury to bole cambium due to differences in bark investment patterns. On average, the normalized bark thickness (NBT) was 15.8 ± 4.3% (±SD) and ranged from 8.1 ± 2.5% to 22.9 ± 4.3% for Alstonia costata and Tabernaemontana cerifera, respectively (Fig. 2). The NBT tended to decrease with the estimated bark allometric coefficient, however the correlation was not significant (Spearman's rank test, P = 0.32). Overall, Geissois racemosa, M. quinquenervia and T. cerifera showed high investment in bark (high NBT and low α_BT; Fig. 2) and thus potentially high capacity to avoid fire injury to the bole cambium, as compared to Achronychia laevis and Guioa villosa, which were likely the most vulnerable species. These two last species were the only two showing positive allometry (α_BT > 1, concave form; Fig. 3), followed closely by Elattostachys apetala and Ficus habrophylla that had isometric allometry (α_BT ≈ 1, linear form), while all other species had negative allometry (α_BT < 1, convex form).

Figure 2. Normalized bark thickness (BT as percentage of stem radius) against bark allometric coefficients (bars represent 1 SE). The two-letter codes refer to species (see Table 1).

Figure 3. Specific allometric relationships between bark thickness and stem diameter at breast height (DBH). The two-letter codes refer to species (see Table 1); grey horizontal lines represent estimated depth of necrosis for a fire of 60-s residence time and fire line intensity (I) of 1000, 2500 and 5000 kW·m⁻¹ (dotted, dashed and full lines, respectively); arrows show minimum DBH required to avoid bole cambium injury. All represented relationships are significant (P < 0.05), bark allometric coefficients (α_BT) are given ± 1 SD (*P < 0.05; **P < 0.01; ***P < 0. 001), different letters show significant difference (Student's t-test, P < 0.05).

In the range of the studied DBH, and for fire with a 60 s residence time, the concave and linear patterns highlighted for A. laevis, G. villosa and E. apetala were the only ones for which no individuals are expected to avoid injury from a fire of 1000 kW·m⁻¹ (Fig. 3) because the bark of these species was very thin (BT < 1.15 cm). This was due both to low bark investment (low NBT and high α_BT) and low DBH, although for E. apetala, we were unable to sample the full range of DBH possible for the species. The range of minimum DBH required for the other species to avoid cambium injury during such fires was quite large, ranging from about 11 cm for T. cerifera to 42 cm for A. costata. G. racemosa, M. quinquenervia and T. cerifera showed particularly high capacity to avoid bole cambium injury as they could quickly reach the minimum DBH required to avoid fire of 1000 kW·m⁻¹ (12.2, 13.9 and 11.1 cm, respectively), 2500 kW·m⁻¹ (17.6, 21.0 and 14.2 cm, respectively) or 5000 kW·m⁻¹ (23.3, ca. 30 and 17.2 cm, respectively). Codia albicans and F. habrophylla could also avoid these fires, however their minimum required DBH were noticeably higher (e.g. 18.6 and 20.4 cm for I = 1000 kW·m⁻¹).

Tree height and crown injury

Allometric coefficients for TH growth (α_TH) were significant for all studied species (P < 0.05; Fig. 4), but more surprisingly α_TH were all <1, and of the same order for all trees: from 0.46 ± 0.04 to 0.64 ± 0.03 for C. albicans and G. racemosa, respectively, considered as large trees, and from 0.40 ± 0.15 to 0.47 ± 0.09 for A. laevis and Fagraea berteoroana, respectively, considered as small trees. Moreover, relationships between the CBH and DBH were weaker (not shown) and only seven species had significant α_CBH (all large tree species but only two species among the six small trees). Therefore, small tree species, with low TH and low CBH growth, should be very exposed to scorching and associated crown injury, which was confirmed for even the lowest intensity fire tested (Fig. 4). Indeed, in the range of the studied DBH, most of the small trees were likely totally scorched by low-intensity fires (i.e. I = 1000 kW·m⁻¹, potential scorch height ≈ 6 m), while for large trees the minimum DBH required to preserve half of the tree from scorch was wide, from ca. 10.5 cm for A. costata to 25.5 cm for M. quinquenervia (Fig. 4). However, even these most potentially resistant species were likely totally scorched by very intense fires (i.e. I = 5000 kW·m⁻¹ and potential scorch height ≈ 17 m).

Figure 4. Specific allometric relationships between tree height (black circles and full lines) and tree canopy base height (white circles and dotted lines) against stem diameter at breast height (DBH). As in Fig. 3, letters code species; dotted, dashed and full grey lines represent estimated scorch height for each fire line intensity (1000, 2500 or 5000 kW·m⁻¹, respectively) and arrows show minimum DBH required to prevent half of the tree from scorching (for I = 1000 kW·m⁻¹). All represented relationships are significant (P < 0.05), tree height allometric coefficients (α_TH) are given ± 1 SD (*P < 0.05; **P < 0.01; ***P < 0. 001).

Discussion

Bark thickness, not tree height, differentiates fire tolerance of species

The comparison of fire-related traits from early secondary successional forest species growing in fire-prone savannas showed, as hypothesized, that the fire tolerance of a species is highly variable and that investment in bark thickness seems to be the prevailing trait that differentiates species in their tolerance to fire, as found in other tropical savanna–forest regions (e.g. Pinard & Huffman 1997; Hoffmann et al. 2003, 2012). Indeed, analysis of covariance and weakness of the allometric relationships showed that CBH, which represents the second strategy to tolerate fire (e.g. Higgins et al. 2000; Bond 2008), was weakly determined by the species and DBH (i.e. ontogeny). We suggest that CBH is rather determined by environmental factors (e.g. fires, grazing, competition with other trees or grasses). Although taller species were better able to avoid scorching, all species were potentially exposed to substantial crown injury because they did not present sufficient stem growth and canopy height to raise their canopy above flames and scorch height. This was true even for low-intensity fires, which are common in NC savannas. Conversely, differences in incremental bark thickness allowed some species to be clearly less exposed to bole injury.

These results therefore support the idea that inter-species variations in fire response are mostly explained by variations in bark thickness and stem diameter (Balfour & Midgley 2006), which are positively correlated and influence top-kill (e.g. Jackson et al. 1999; Hoffmann et al. 2003, 2009). Whelan (1995), and more recently Lawes et al. (2011a), suggest that growth in height is not advantageous if the cambium is not protected within the stem. Moreover, Balfour & Midgley (2006) experimentally showed for an African Acacia species that death of canopy buds is not sufficient to cause top-kill. Thus, optimizing bark thickness is likely the only strategy to avoid top-kill in fire prone savannas (Lawes et al. 2011a) such as those found in NC species.

Resprouting may be a trait related to vegetation resilience and an alternative to persistance after fire. This strategy is frequent in woody species and many fire-prone ecosystems (Bond & Midgley 2001) and likely allows trees and grasses to co-exist in savannas (Higgins et al. 2000; Vesk 2006). Regarding the studied species, even though there are few data on their resprouting capacity, most are also able to resprout from roots after fire (Jaffré et al. 1997; Bocquet et al. 2007). The resprouting of juveniles allows species to persist in fire-prone ecosystems and recruitment into mature size classes depends on rare opportunities to reach escape size (Higgins et al. 2000; Archibald & Bond 2003; Vesk 2006). Thus, the more the juveniles develop fire-related traits, such as thick bark, the faster they can resist fire and the better their opportunity to recruit into mature size classes.

New Caledonian secondary forest species are poorly adapted to savanna fire regimes

Our results indicate that early secondary successional forest species growing in NC savannas are relatively poorly adapted to fire. Although most of them invested disproportionally more in bark at small DBH, as expected for typical savanna tree species (Jackson et al. 1999), their investment in bark was lower than in species in others savannas worldwide. For instance, compared to Brazilian Cerrado species (Hoffmann et al. 2003), the average normalized BT of NC species (15.3%) was closer to that of forest species than to savannas species (10.1% and 28.5%, respectively). The same trend was found when compared to other species in the region, as NBT of NC species was less than half of that of savanna species of northern Australia (34%; Lawes et al. 2011a). Although few field measurements are available in the literature to validate our simulations, the simulated fire line intensities (from 1 to 28 558 kW·m⁻¹) were of the same order than those published for similar ecosystems (from 25 to 22 000 kW·m⁻¹) to comparable fuel loads (from 0.11 to 25.10 t·ha⁻¹ in this study vs 0.83–14.20 t·ha⁻¹ in the literature) in Africa (Shea et al. 1996; Stocks et al. 1996; Hely et al. 2003; Gambiza et al. 2005; Govender et al. 2006; Savadogo et al. 2007), Australia (Williams et al. 1998, 2003) and South America (Kauffman et al. 1994).

Functional explanations for the variation in BT are not clear (Paine et al. 2010). Given the important role of bark in protecting the stem from fire heat, it is however clear that frequent and sufficiently intense fire imposes a substantial selection pressure on BT (Stephens & Libby 2006; Midgley et al. 2010). For example, savanna species submitted to frequent and relatively intense fire have thicker bark than forest species submitted to lower fire frequency and intensity (Hoffmann et al. 2003).

In NC, savannas and maquis are the two main fire-prone ecosystems (Jaffré et al. 1998; McCoy et al. 1999). These contrasting ecosystems have both expanded widely at the expense of forest since settlement of humans on the island (starting at ca. 3000 yr BP) and the increase in fire frequency from anthropogenic fires (Jaffré et al. 1998; Stevenson 2004). However, according to paleoecological records, savannas were very scarce before humans settled in NC (Stevenson 2004), whereas maquis existed long before human settlement (Hope & Pask 1998; Stevenson & Hope 2005). Thus, Jaffré et al. (1997) suggested that species growing in maquis are well adapted to fire as a result of long-term adaptation. Therefore, unlike maquis, we hypothesize that 3000 yr of fire pressure in savannas is still too short for the evolution of forest species towards complete adaptation to fire.

Implications for rain forest–savanna dynamics and management

It is widely recognized that fire and fire regime diversity can support biodiversity (Driscoll et al. 2010), however socio-environmental changes in NC, such as a dramatic increase in invasive weeds and mammals (wild deer and pigs) or changes in fire use, likely modify the fire regime and threaten biodiversity and ecosystems services because they reduce forested areas and prevent forest expansion and recovery (Ibanez et al. 2012). In this context, active restoration and fire prevention may be two complementary methods to manage the landscape, through both promoting secondary forest succession and limiting forest erosion, and thus preserving a landscape mosaic.

Species that are resilient to fire and can avoid top-kill are clearly advantaged in fire-prone savanna (Lawes et al. 2011a). In most NC savannas M. quinquenervia is the only tree species. Indeed this species originated in swampy areas but is very tolerant of savanna fires, exhibiting morphological defences such as thick, multi-layer spongy bark, high resprouting capacity from both roots and shoots (from epicormic buds), and high fire-induced seed release (Serbesoff-King 2003). However, savannas can be colonized by early secondary successional forest species, which occur when fire frequency and/or intensity decreases (Ibanez 2012). Among these species, G. racemosa and T. cerifera have bark thickness and allometry similar to M. quinquenervia. As G. racemosa is frequently observed as isolated adult trees in savannas (Ibanez 2012), with fire scars on trunks, this suggests that it is actually more tolerant to fire.

Decrease in fire frequency (mainly through fire prevention) to allow species to develop defences, and decreased fire intensity (e.g. through decreasing available fuel by clearing or prescribed fire) to decrease fire severity and top-kill may be two possibilities to promote the recruitment of mature forest trees in savannas and protect surrounding forests from fire propagation. The fight against invasive weeds, e.g. Lantana camara and Melinis minutiflora, that often dominate the NC savanna ground layer and modify both fire regime and severity (Dantonio & Vitousek 1992; Brooks et al. 2004; Hoffmann et al. 2004) may be one method to manage the fire regime.

Acknowledgements

We thank the ANR BDIV-07-008 project INC and the CNRS for funding this research and the PhD scholarship of T. Ibanez. We are grateful to Hubert Géraux and the WWF for logistic support and to Denis Meandu-Poveu for enthusiastic help in the field. We also thank Jérôme Munzinger, William Hoffmann, Philippe Birnbaum and Vincent Bonhomme for useful comments on earlier versions of the manuscript.