INTRODUCTION
Fire occurrence generally increases after large-scale droughts in the Amazon (Alencar et al. Reference ALENCAR, NEPSTAD and DIAZ2006, Reference ALENCAR, ASNER, KNAPP and ZARIN2011; Aragão et al. Reference ARAGÃO, MALHI, ROMAN-CUESTA, SAATCHI, ANDERSON and SHIMABUKURO2007, Bush et al. Reference BUSH, SILMAN, MCMICHAEL and SAATCHI2008, Yang et al. Reference YANG, RANDERSON, MORTON, DEFRIES, COLLATZ, KASIBHATLA, GIGLIO, YUFANG and MARLIER2011). In 1997–1998, the region suffered from the influence of the El Niño–Southern Oscillation (ENSO), which contributed to one of the most important mega-fires in the world with 3.9 million ha of burned forests and up to 0.329 Pg of biomass loss (Alencar et al. Reference ALENCAR, NEPSTAD and DIAZ2006). In the state of Roraima (northern Amazon), up to 1.4 million ha of primary forest were affected (Barbosa & Fearnside Reference BARBOSA and FEARNSIDE1999), which highlights the prominence of fires as a threat to these forests that should not normally burn (Cochrane & Laurance Reference COCHRANE and LAURANCE2002).
Uncontrolled fire incidents results in a wide diversity of ecological effects (Bowman et al. Reference BOWMAN, BALCH, ARTAXO, BOND, CARLSON, COCHRANE, D'ANTONIO, DEFRIES, DOYLE, HARRISON, JOHNSTON, KEELEY, KRAWCHUK, KULL, MARSTON, MORITZ, PRENTICE, ROOS, SCOTT, SWETNAM, VAN DER WERF and PYNE2009), altering forest structure and floristic composition. Amazonian studies have demonstrated the higher mortality of small-diameter trees in recently burned areas (Barlow et al. Reference BARLOW, PERES, LAGAN and HAUGAASEN2003, Cochrane & Schulze Reference COCHRANE and SCHULZE1999), whereas an increase in mortality of larger trees (≥ 50 cm dbh) has been reported 1–3 y after a fire (Barlow et al. Reference BARLOW, PERES, LAGAN and HAUGAASEN2003). A positive relationship was found between tree mortality and char height, which can be used as a proxy of fire intensity (Balch et al. Reference BALCH, NEPSTAD, CURRAN, BRANDO, PORTELA, GUILHERME, REUNING-SCHERER and CARVALHO2011, Barlow et al. Reference BARLOW, PERES, LAGAN and HAUGAASEN2003).
Fire can dramatically reduce living biomass in tropical forests (Haugaasen et al. Reference HAUGAASEN, BARLOW and PERES2003, Slik et al. Reference SLIK, BERNARD, VAN BEEK, BREMAN and EICHHORN2008), even to the same degree as logging activities (Gerwing Reference GERWING2002). In Amazonian forests, Cochrane & Schulze (Reference COCHRANE and SCHULZE1999) estimated average biomass losses of 80%, 50% and 10% in heavily, moderately and lightly burned areas, respectively. Three years after the fire, the biomass had recovered to 49% of that of a neighbouring unburned forest and the loss of 107 Mg ha−1 was attributed to the delay in mortality of trees ≥ 50 cm in stem diameter (Barlow et al. Reference BARLOW, PERES, LAGAN and HAUGAASEN2003). Balch et al. (Reference BALCH, NEPSTAD, CURRAN, BRANDO, PORTELA, GUILHERME, REUNING-SCHERER and CARVALHO2011) reported losses of 32% and 21% with respect to the initial above-ground live biomass in once- and thrice-burned forests, respectively, with trees < 30 cm dbh significantly contributing to the total dead biomass. Thus, fire may be a critical factor for determining carbon balance in tropical forests (Balch et al. Reference BALCH, NEPSTAD, BRANDO, CURRAN, PORTELA, DE CARVALHO and LEFEBVRE2008, Zarin et al. Reference ZARIN, DAVIDSON, BRONDÍZIO, VIEIRA, SA, FELDPAUSCH, SCHUUR, MESQUITA, MORAN, DELAMONICA, DUCEY, HURTT, SALIMON and DENICH2005).
Tree species composition may have a considerable impact on the response of a forest to fire, as the probability of mortality is species-dependent (Balch et al. Reference BALCH, NEPSTAD, CURRAN, BRANDO, PORTELA, GUILHERME, REUNING-SCHERER and CARVALHO2011, Brando et al. Reference BRANDO, NEPSTAD, BALCH, BOLKER, CHRISTMAN, COE and PUTZ2012, Slik et al. Reference SLIK, BREMAN, BERNARD, VAN BEEK, CANNON, EICHHORN and SIDIYASA2010). Fire tends to favour species that can tolerate heat stress and to increase the abundance of pioneer species (Barlow & Peres Reference BARLOW and PERES2008, Slik et al. Reference SLIK, BREMAN, BERNARD, VAN BEEK, CANNON, EICHHORN and SIDIYASA2010), at least in the short term (Cochrane & Schulze Reference COCHRANE and SCHULZE1999).
Ecosystems can normally recover from disturbances. However, frequent and severe fires can offset this balance. As a result, the ecosystem may lose the ability to recover, which could lead to its collapse with the loss of their major ecological functions. We hypothesize that fire disturbances of increasing frequency and severity, associated with shorter recovery times, may lead to declines in biomass and in species diversity. In this study, we examined the impact of single and repeated forest fires of different intensities on the above-ground live biomass.
STUDY SITE
This study was conducted in the Apiaú and Roxinho region in the central part of the state of Roraima, northern Brazilian Amazon (Figure 1). This region contains many rural land grants concentrated along the roads. Fires, which occur more frequently during the dry season, have historically been used for forest conversion and pasture maintenance, occasionally causing accidental fires in primary forests. Thus, in the study area, these fires are mainly understorey fires that escape from neighbouring pastures and reach the forest. The region was identified as the centre of the Great Fire of Roraima (Fearnside et al. Reference FEARNSIDE, BARBOSA and GRAÇA2007), which occurred in 1998 and was strongly influenced by ENSO.
Figure 1. Location of the study area in northern Brazilian Amazon (state of Roraima) and of the sample plots used in data analysis. The 50 inventoried plots are not all distinguishable due to their proximity.
The average annual precipitation is approximately 2000 mm (Barbosa & Fearnside Reference BARBOSA and FEARNSIDE1996), and the dry season extends from December to March. The topography is gently undulating to hilly predominantly 100–180 m asl (Fearnside et al. Reference FEARNSIDE, BARBOSA and GRAÇA2007). Soils are acidic and poor in nutrients (Barbosa & Fearnside Reference BARBOSA and FEARNSIDE1996). The natural vegetation mosaic is composed of tropical rain forests, tropical open forests and tropical semideciduous forests, which are located close to savannas (IBGE 1992).
METHODS
Field sampling and data collection
Field work was conducted in January and February of 2010. Plots (25 × 100 m) were set up in unburned primary forests and across different classes of fire-damaged primary forests (Table 1). The level of fire disturbance was defined in terms of the times of recurrence (fire frequency). The level of burn damage (severity) was also characterized for the areas that burned in 1998. Severity classes were defined based on historical information from remote-sensing data, locally obtained from landowner interviews and especially from results of IBAMA (1998) on post-fire mortality of trees, fuel quantity, scorch height and damage levels. Based on IBAMA (1998), 50% of the area where plots are located was affected only at the understorey level (lightly burned), whereas 30% was affected by stronger fires that reached the understorey and canopy levels (heavily burned). The rest of the area (20%) was not affected by fire. In addition, field work observations on tree density and the general aspect of understorey supported the classification of the areas affected in 1998 in terms of severity. There were no signs of logging activities or history of other degradation type in the inventoried area.
Table 1. Plot characterization and distribution across fire disturbance classes in northern Brazilian Amazon. N is the number of plots.
Within each georeferenced plot, dbh and height were obtained for all individuals (trees, palms, bamboos and pseudostems of Phenakospermum guyannense with dbh ≥ 10 cm. Tree heights were estimated visually by one of us (J. R. dos Santos) and then each tree height was modelled according to Gonçalves & Santos (Reference GONÇALVES and SANTOS2008). The identification of morphospecies was performed from dendrological characteristics. Species nomenclature and family classification were adjusted to the Tropicos database (cite URL Tropicos.org 2012).
Above-ground biomass estimation
Five allometric equations were used to estimate individual above-ground living dry biomass (Table 2), hereafter simply called biomass. For tree species (other than Cecropia spp.), we compared estimates from the allometric equations developed by Brown et al. (Reference BROWN, GILLESPIE and LUGO1989) and Chave et al. (Reference CHAVE, ANDALO, BROWN, CAIRNS, CHAMBERS, EAMUS, FÖLSTER, FROMARD, HIGUCHI, KIRA, LESCURE, NELSON, OGAWA, PUIG, RIÉRA and YAMAKURA2005), which are frequently employed in the Amazon Basin. Because there was no significant difference between them (ANOVA; α = 0.05), we decided to use the biomass values resulting from the allometric equation presented by Chave et al. (Reference CHAVE, ANDALO, BROWN, CAIRNS, CHAMBERS, EAMUS, FÖLSTER, FROMARD, HIGUCHI, KIRA, LESCURE, NELSON, OGAWA, PUIG, RIÉRA and YAMAKURA2005).
Table 2. Allometric equations used to estimate above-ground live biomass (Y) per individual (kg) in each species group in northern Brazilian Amazon. ρ is the specific gravity (g cm−3), Ht is total height (m) and dbh is diameter at breast height (cm).
Above-ground biomass (AGB) was defined as the cumulative tree biomass whereas the total above-ground biomass (TAGB) was defined as the sum of the biomass of all individual trees, palms and P. guyannense pseudostems (dbh ≥ 10 cm). Bamboos, which occurred in just one plot, were not included in the estimation of TAGB. Carbon content of biomass was assumed to be 50% (Fearnside Reference FEARNSIDE1997).
Analysis of fire impacts on forest above-ground live biomass, structure and diversity
Comparisons of the mean AGB and TAGB among the five fire disturbance classes were based on one-way analyses of variance (ANOVA) after checking each treatment for normality with Shapiro–Wilk's test (Shapiro & Wilk Reference SHAPIRO and WILK1965) and for homoscedasticity with Bartlett's test (Snedecor & Cochran Reference SNEDECOR and COCHRAN1989). The Tukey's multiple comparison test was then used to identify significant differences (α = 0.05) among the groups. This method was also used to assess changes in the contributions of each diameter size class to AGB, with or without considering individuals of the Cecropia spp. It is important to note that differences in AGB and TAGB between unburned and burned forests are indicative of direct reductions by fire as well as delayed mortality. Biomass gains incorporate growth of surviving trees in addition to the establishment of new individuals.
Floristic composition across plots was compared based on the species and families with the highest contributions to AGB. Shannon's diversity index (H’) and the importance value index (IVI), determined only for tree species, were assessed in relation to AGB. Finally, a principal coordinate analysis (PCO) was performed using the Bray–Curtis similarity measure and species abundance in order to identify floristic composition gradients within the classes.
RESULTS
General characteristics
A total of 5593 individuals were measured across all plots (12.5 ha). The sample was composed of 177 species (83 identified only to the generic level and 19 unidentified) distributed across 122 genera and 53 families. For the unburned forest plots, the mean (± SD) density of trees, mean (± SD) dbh and mean (± SD) basal area were 310 ±28 ind. ha−1, 23 ± 15 cm and 18 ± 5 m2 ha−1, respectively. A Shannon index (H’) of 4.02 was obtained for the UF plots considering only the tree species.
Effects of fire on above-ground live biomass
Significant differences in AGB were found between the fire disturbance classes, indicating a gradual effect of successive fires on biomass (Figure 2). The mean AGB ranged from 70 ± 32 Mg ha−1 in thrice-burned forest (BF3) to 188 ± 78 Mg ha−1 in unburned forest (UF). According to the Tukey's test, the mean AGB was not significantly different among the UF, the once lightly (BF1L) (163 ± 57 Mg ha−1) and the once heavily (BF1H) (119 ± 52 Mg ha−1) burned forest plots. UF plots had significantly higher levels of AGB than twice- (BF2) (108 ± 48 Mg ha−1) and thrice- (BF3) (70 ± 32 Mg ha−1) burned forest plots, which showed mean reductions of biomass stocks of 57% and 63%, respectively.
Figure 2. Boxplots showing the gradual effect of fire frequency and severity on mean above-ground living biomass (AGB in Mg ha−1) in northern Brazilian Amazon. UF = unburned forest; BF1L = once lightly burned forest; BF1H = once heavily burned forest; BF2 = twice-burned forest and BF3 = thrice-burned forest. Different letters (a, b and c) associated with the fire disturbance classes indicate significant differences according to the Tukey's test (α = 0.05).
The loss of biomass between UF and BF1H (37%) was about three times greater than the difference between UF and BF1L (13%). BF1H presented an intermediate degree of degradation, as its mean biomass was not significantly different from either the two less-disturbed classes (UF and BF1L) or the most disturbed classes (BF2 and BF3). The presence or absence of Cecropia spp. from the analysis did not change the results.
In UF plots, carbon storage in the living AGB was estimated at 94 Mg C ha−1. After a single fire event, this stock was reduced to 81 Mg C ha−1 in BF1L and to 59 Mg C ha−1 in BF1H. The latter value is close to the 54 Mg C ha−1 found in BF2. In BF3, 35 Mg C ha−1 remained stored in living AGB, indicating a difference of 59 Mg C ha−1 of carbon in relation to UF plots.
The mean contribution of palms and P. guyannense pseudostems to the TAGB reached approximately 5% (Table 3) considering the five classes, while the palms alone contributed 3.2% (4.2 Mg ha−1). The low biomass contributions of palms and P. guyannense explained the similar results of the ANOVA and Tukey's test between AGB and TAGB (Table 3). In UF plots, the mean contribution of palms to TAGB was estimated at 6.81 ± 6.94 Mg ha−1 or 3%. Palm contributions to TAGB were substantial in UF and BF1L (4%) plots, whereas P. guyannense pseudostems were most important in BF1H plots (4%).
Table 3. Variation in biomass values for fire disturbance classes (UF = unburned forest; BF1L = once lightly burned forest; BF1H = once heavily burned forest; BF2 = twice-burned forest and BF3 = thrice-burned forest) in northern Brazilian Amazon. AGB = above-ground live biomass (mean ± SD); TAGB = total above-ground live biomass (mean ± SD). The values of AGB and TAGB were not significantly different, considering each class, according to the Tukey's test (α = 0.05).
Effects of fire on biomass in relation to tree size distribution
The distribution of mean AGB values in dbh sizes ≤ 50 cm did not show a clear pattern with respect to fire disturbance classes, except for the BF3, in which the mean AGB decreased when compared with the other ones (Figure 3a). However, mean AGB was significantly lower in UF3 than in UF or BF1L. For the dbh class greater than 50 cm, the AGB differences were more evident and progressively decreased as the level of fire disturbance increased (Figure 3a). However, the only significant difference was observed between UF and BF3 plots. For these large trees, AGB from the BF1L to BF3 fire disturbance classes was reduced by 21%, 45%, 50% and 54% relatively to the UF. We also noted that the mean contribution of ≤ 50 cm dbh trees was similar to that of > 50 cm dbh trees in unburned plots. In these plots, differences in tree density of small (293 ind. ha−1) and large (17 ind. ha−1) trees highlighted the major contributions made by a few large trees to the AGB in primary undisturbed forests. The only gain in biomass observed in burned forests, in relation to UF, occurred in the BF1L among the < 50 cm dbh trees.
Figure 3. Fire effects on the above-ground biomass content by tree size classes in northern Brazilian Amazon. Mean values considering all tree species, merging large (> 50 dbh cm) individuals (a) and considering only small individuals (dhb ≤ 30 cm) other than the Cecropia spp. (b), by fire disturbance classes: UF = unburned forest; BF1L = once lightly burned forest; BF1H = once heavily burned forest; BF2 = twice-burned forest and BF3 = thrice-burned forest. Different letters (a, b and c) associated with the fire disturbance classes indicate significant differences by dbh class according to Tukey's test (α = 0.05). For the 40–50 dbh class, the data did not fulfil the requirements for ANOVA.
Because the individuals of Cecropia spp. (pioneer species that are common in burned areas) had generally less than 30 cm in diameter, the removal of this genus only altered the biomass in the smaller size classes (Figure 3b). For example, in the fire disturbance classes BF1H, BF2 and BF3, removal of these species from the 10–20-cm dbh class resulted in biomass losses of 32%, 36% and 34%, respectively. As a result, differences in biomass between these heavily disturbed forests and unburned or lightly burned forests were accentuated.
Species contributions to biomass
For the five species with the highest contributions to AGB, mean dbh generally decreased and mean density increased with fire disturbance (Appendix 1). These trends were partially driven by individuals of Cecropia leucocoma Miq., which made notable contributions to AGB in the BF1H as well as in the more heavily burned classes, also with increasing IVI. Similar results were found for Pouteria sp., whereas Protium sp. made major contributions to AGB in the UF, BF1L, BF1H and BF2 plots (approximately 9%) and it was not an important species in BF3 plots. These selected tree species were mainly classified as non-pioneer species, except in the thrice-burned forest. Vochysia sp. and Manilkara huberi (Ducke) A. Chev., two species of economic interest, had the highest AGB values in BF1L.
Concerning palm species, Oenocarpus bacaba Mart. was present with highest abundance in UF and BF1L plots, while Maximiliana martiana H. Karst. was more frequent in BF2 and BF3 plots. However, Maximiliana martiana made the highest contribution to biomass among all palms in every fire disturbance class except BF1L.
The contribution of Leguminosae to AGB decreased with increasing fire disturbance and reached 5% in BF3 (Appendix 1). The contribution of Vochysiaceae showed values close to 1% in BF1H, BF2 and BF3. The Sapotaceae made high contributions to AGB in all classes and no clear trend in relation to fire damage was detected. The Burseraceae, another predominant family, contributed to AGB up to BF3, where the contribution fell to 4%. The Apocynaceae made a substantial contribution to AGB in BF1H plots (16%) but only accounted for 2% in the other classes. The AGB contribution of Urticaceae, represented by two species of Cecropia, increased with fire damage. The contribution of Sapindaceae to AGB increased slightly in the most damaged plots (Appendix 1).
Relationships between plant diversity, biomass and fire disturbance
The highest mean diversity of tree species (H′) was found in the UF class (4.02). BF1L and BF1H plots, which only differed in fire severity, exhibited the greatest difference in diversity between two successive classes (3.95 and 3.14, respectively, or −22%). A significant reduction in diversity was observed in BF1H in relation to UF plots (−28%), even 12 y after the fire. Conversely, BF1L plots showed only minor reductions in diversity after the same recovery time. Similar and lower values of H′ were found for BF2 and BF3 (2.90 and 2.89, respectively).
H′ was positively related to AGB (Figure 4). An outlier plot was removed from the regression (in BF1H, AGB = 207 Mg ha−1; H′ = 1.81) because the abnormal biomass value could be mostly explained by two trees. For the BF1H and BF2, there was a positive relationship between AGB and H′. For the UF, BF1L and BF3 classes, no significant relationships were found because the diversity was constant across the biomass range.
Figure 4. Logarithmic relationship between AGB and H′ for the full set of plots inventoried in the northern Brazilian Amazon and their relationships with each fire disturbance class (UF = unburned forest; BF1L = once lightly burned forest; BF1H = once heavily burned forest; BF2 = twice-burned forest and BF3 = thrice-burned forest).
The first two ordination axes of the PCO explained 21.1% and 13.4% of the total variability of the floristic composition (Figure 5). The plots were gradually distributed along the first axis according to the fire-damage classes, with a clear discrimination between UF and BF3 plots. The second axis showed a detachment of a set of plots mainly associated with the BF1H class.
Figure 5. Scatter plot showing the first and second axis of the principal coordinate analysis (PCO) considering the 50 inventoried forest plots in northern Brazilian Amazon, based on Bray-Curtis similarity measure and floristic abundance data. UF = unburned forest; BF1L = once lightly burned forest; BF1H = once heavily burned forest; BF2 = twice-burned forest and BF3 = thrice-burned forest.
DISCUSSION
Above-ground biomass content of unburned primary forests in Apiaú region
The mean AGB recorded in unburned primary forests, 188 ± 78 Mg ha−1, was consistent with results of previous studies in the same region by Xaud (Reference XAUD1998) and Barbosa et al. (Reference BARBOSA, KEIZER, PINTO, Barbosa and Melo2010). Our results agreed with earlier findings showing that these Amazon forests located close to savannas had lower biomass than the dense rain forest of the central Amazon (Malhi et al. Reference MALHI, WOOD, BAKER, WRIGHT, PHILLIPS, COCHRANE, MEIR, CHAVE, ALMEIDA, ARROYO, HIGUCHI, KILLEEN, LAURANCE, LAURANCE, LEWIS, MONTEAGUDO, NEILL, NUNEZ VARGAS, PITMAN, QUESADA, SALOMAO, SILVA and LEZAMA2006, Saatchi et al. Reference SAATCHI, HOUGHTON, ALVALÁ, SOARES and YU2007). In these tropical forests, the effect of large trees on biomass was exemplified by individuals of Dinizia excelsa, which strongly increased the AGB in the UF and BF1H classes. Dinizia excelsa is an emergent species that occurs at low densities and has a high specific gravity (EMBRAPA 2004).
Losses and post-fire biomass recovery
As expected, our results showed a gradual reduction in biomass stocks after more frequent/severe fires on the studied time scale. By comparing BF1L and BF1H, the impact of fire severity to produce higher reductions in biomass was demonstrated, as also reported by Gerwing (Reference GERWING2002). Thus, the dynamics of biomass recovery was reduced by fire severity.
The range of AGB values in the BF3 class (36–114 Mg ha−1) revealed that some plots that experimented recurrent fire events reached similar levels of live above-ground biomass to those of secondary forests estimated in the same region by Xaud (Reference XAUD1998) and Araújo (Reference ARAÚJO1999) with values ranging from 33 to 57 Mg ha−1 and from 14 to 79 Mg ha−1, respectively. The gain in biomass observed for the BF1L in relation to UF class, regarding ≤ 50 cm dbh trees, showed forest recovery and establishment of intermediate-diameter and non-pioneer trees after 12 y from the fire. For the dbh class > 50 cm, differences in AGB were more pronounced revealing the impact of fire on the AGB of large trees.
Figure 6 summarizes some of the major post-fire biomass studies performed in the Amazon basin and illustrates the temporal dynamics of biomass levels after a fire event. Time intervals between fires, the number of fire occurrences, fire severity and remaining biomass estimates have been reported. The biomass estimates from the present study were associated with the longest post-fire time interval (12 y), complementing the results of the previous ones. The values formed a hump-shaped distribution, with an increase in biomass loss within the first 3 y after the fire.
Figure 6. Synthesis of results in terms of living above-ground biomass reduction (or difference) with respect to the time interval between the last fire and AGB data collection (y) based on nine studies in the Brazilian Amazon (Balch et al. Reference BALCH, NEPSTAD, CURRAN, BRANDO, PORTELA, GUILHERME, REUNING-SCHERER and CARVALHO2011, Barbosa & Fearnside Reference BARBOSA and FEARNSIDE1999, Barlow et al. Reference BARLOW, PERES, LAGAN and HAUGAASEN2003, Cochrane et al. Reference COCHRANE, ALENCAR, SCHULZE, SOUZA, NEPSTAD, LEFEBVRE and DAVIDSON1999, Fearnside et al. Reference FEARNSIDE, GRAÇA and RODRIGUES2001, Reference FEARNSIDE, BARBOSA and GRAÇA2007; Gerwing Reference GERWING2002, Graça et al. Reference GRAÇA, FEARNSIDE and CERRI1999, Haugaasen et al. Reference HAUGAASEN, BARLOW and PERES2003) as well as the present study.
Previous studies reported a reduction in biomass of approximately 20% immediately after a fire (Balch et al. Reference BALCH, NEPSTAD, CURRAN, BRANDO, PORTELA, GUILHERME, REUNING-SCHERER and CARVALHO2011, Fearnside et al. Reference FEARNSIDE, GRAÇA and RODRIGUES2001, Reference FEARNSIDE, BARBOSA and GRAÇA2007; Graça et al. Reference GRAÇA, FEARNSIDE and CERRI1999), with an average loss of approximately 50% 3 y later (Balch et al. Reference BALCH, NEPSTAD, CURRAN, BRANDO, PORTELA, GUILHERME, REUNING-SCHERER and CARVALHO2011, Barlow et al. Reference BARLOW, PERES, LAGAN and HAUGAASEN2003, Gerwing Reference GERWING2002). This delay in biomass loss was likely the result of the higher resistance of large trees to fire, which expanded the mortality distribution over several years (Barlow et al. Reference BARLOW, PERES, LAGAN and HAUGAASEN2003). The loss of biomass was less severe 7 and 12 y after the fire, similar to the biomass values estimated immediately after the fire. Although the precise point at which biomass stops decreasing after a fire remains unclear, our analysis suggested that it may be between 3 and 7 y after the fire, assuming no additional disturbances. Twelve years after the fire occurrence, the biomass was not significantly different from the initial level, illustrating the resilience of these forests in term of biomass recovery, at least for areas burned only once.
Our results confirmed that forest resilience in terms of biomass depends on the fire history, which includes not only the elapsed time after the fire but also the fire frequency and severity (Numata et al. Reference NUMATA, COCHRANE and GALVÃO2011). Both fire frequency and severity increased the loss of biomass and delayed its recovery. The exception was for the thrice-burned forests in which biomass was measured less than 1 y after the last fire, and because the intervals between the three fires were abnormally short (Balch et al. Reference BALCH, NEPSTAD, CURRAN, BRANDO, PORTELA, GUILHERME, REUNING-SCHERER and CARVALHO2011). Thrice-burned and heavily burned forests, observed 1 and 2 y after the last fire, respectively, had experienced the greatest loss of biomass. For these two cases, only one of the two critical variables (severity and frequency) was available.
Relationships between biomass, diversity and fire
The Shannon index (H′) of 4.02 obtained in our UF plots was slightly those obtained in the Roraima state by Alarcón & Peixoto (Reference ALARCÓN and PEIXOTO2007), 4.66, or in the National Forest of Tapajós (Pará state), by Gonçalves & Santos (Reference GONÇALVES and SANTOS2008), 4.22. In the central Amazon, high levels of diversity were found by Oliveira et al. (Reference OLIVEIRA, AMARAL, RAMOS, NOBRE, COUTO and SAHDO2008) (H′ of 5.1).
Tree diversity decreased with fire disturbance in the Amazon, as demonstrated by Barlow et al. (Reference BARLOW, PERES, LAGAN and HAUGAASEN2003) and Cochrane & Schulze (Reference COCHRANE and SCHULZE1999). In the case of BF1L, a post-fire recovery time of 12 y allowed tree diversity (H′) to return to initial levels with a floristic composition very similar to the unburned forests, as shown by our PCO analysis. However, severely, recently and/or recurrently burned areas (BF1H, BF2 and BF3) showed lower diversity than lightly burned or unburned forests (BF1L, UF). The difference between H′ values of BF1H and BF1L likely stems from the fact that this type of disturbance favoured the regeneration of P. guyannense pseudostems. This was confirmed by our PCO results, where the second axis clearly discriminated plots with high pseudostem densities, reaching 237 pseudostems ha−1 in plot 8. Thus, fire severity seems to be one of the most important factors for determining post-fire forest resilience, at least in terms of diversity.
In cases of increased fire recurrence and decreased recovery time (BF2), the positive relation between biomass and diversity illustrates the similar responses of the two proxies to fire disturbance. In plots exhibiting increased fire recurrence and reduced post-fire intervals (BF3), no relationship between biomass and diversity was observed. We hypothesize that in these recurrently disturbed forests, the time after each fire was not sufficient for the biomass to recover, as large tree mortality continues for several years after a fire event (Barlow et al. Reference BARLOW, PERES, LAGAN and HAUGAASEN2003). BF3 also had the lowest mean values for biomass but not for diversity. This pattern suggests that the characteristics of BF3 fires favoured the installation of evenly distributed pioneer species and that these changes in vegetation composition did not alter the Shannon diversity index. Such recurrent disturbances thereby negated the relationship between biomass and diversity. As neither biomass nor diversity values were estimated before the second or third fires, and as no long-term studies on post-fire biomass dynamics have been performed in this region, we were not able to dissociate the effects of fire recurrence, severity and elapsed time after fire on biomass and vegetation diversity.
When pioneer individuals of Cecropia spp. were removed from the analysis of tree size distribution, differences in AGB and the significance of these differences were enhanced only for small trees. This result can be explained by the observation that Cecropia spp. diameters were always < 40 cm in our sample. Indeed, Cochrane & Schulze (Reference COCHRANE and SCHULZE1999) and Barlow & Peres (Reference BARLOW and PERES2008) described the dense post-fire establishment of pioneer species, which compensates for the loss of Amazonian primary forest trees.
Despite a few reports of low mortality of Cecropia spp. after low or mild fire severity (d'Oliveira et al. Reference D'OLIVEIRA, ALVARADO, SANTOS and CARVALHO2011, Uhl et al. Reference UHL, CLARK, CLARK and MURPHY1981), Barlow & Peres (Reference BARLOW and PERES2008) found that the density of these pioneer species and their contributions to AGB decreased relative to other species after recurrent fires. In other words, these pioneers were more sensitive to fire recurrence. However, we found that pioneer IVI, density and contributions to AGB increased in sites where fires have occurred two or three times since 1998, especially for Cecropia spp. and Jacaranda copaia (Aubl.) D. Don. Except for lightly burned forests, Cecropia spp. made the highest contributions to AGB 3, 7 and 12 y after the fire incident. Cochrane & Schulze (Reference COCHRANE and SCHULZE1999) also reported an increase in pioneer density with increasing fire intensity and frequency. Moreover, not all individuals of Cecropia spp. in these forests were recently established, as their mean diameter was higher in the most disturbed plots. Therefore, the sensitivity of pioneer species to fire recurrence does not necessarily modify the vegetation structure through middle-term reductions in biomass or density contributions of such species, because non-pioneer species are progressively excluded from forest as fires occur. Fire recurrence might thus favour pioneer species renewal. The time elapsed between or after fires must be considered as a key factor in determining the biomass contribution patterns of pioneer species.
To some extent, our results agreed with those of Balch et al. (Reference BALCH, NEPSTAD, CURRAN, BRANDO, PORTELA, GUILHERME, REUNING-SCHERER and CARVALHO2011), who found that Protium guianense and Vochysia vismiifolia had high fire-induced mortality rates. Indeed, the biomass contribution of Protium sp. decreased with increasing fire frequency and for Vochysia sp. this contribution was the highest where the forest was only lightly burned (BF1L). The latter species was also recorded in UF plots. Pouteria spp. showed similar behaviours in the two studies, as its fire-induced mortality rate was low in Balch et al. (Reference BALCH, NEPSTAD, CURRAN, BRANDO, PORTELA, GUILHERME, REUNING-SCHERER and CARVALHO2011) and its biomass contribution increased with fire incidence. Fire-induced tree mortality may be influenced not only by cambium insulation but also by species traits that reduce the indirect effects of fire (Brando et al. Reference BRANDO, NEPSTAD, BALCH, BOLKER, CHRISTMAN, COE and PUTZ2012).
In the same region, Fearnside et al. (Reference FEARNSIDE, BARBOSA and GRAÇA2007) found a palm biomass reduction of 79% (2.09 Mg ha−1 to 0.44 Mg ha−1) immediately after a fire event. Less severe reductions were found in the present study (−65% in BF2, −56% in BF1H and −52% in BF3). Both results illustrate the major impact of fire on palm species, which represent less than 50% of the initial biomass, even 12 y after a fire. These studies showed greater reductions in palm biomass compared with results by Graça et al. (Reference GRAÇA, FEARNSIDE and CERRI1999) in open forests (8.6 to 7.1 Mg ha−1 or 17%), or by Fearnside et al. (Reference FEARNSIDE, GRAÇA and RODRIGUES2001), in palms of central Amazon (1.3 to 1.1 Mg ha−1 or 15%).
Carbon emissions
Assessing changes in carbon stocks of the AGB is difficult because they are related not only to immediate fire impacts but also to delayed mortality and biomass recovery. Therefore, this balance integrates many sources and sinks and carbon estimates become more difficult with increasing time after a fire. However, for thrice-burned areas measured 3 y after the last fire (maximum biomass losses according to findings of this work), we expected more precise estimations of carbon losses.
The need for long-term, precise surveys of burned sites in the Amazon is immediate, particularly given the present and future fire scenarios for the region. According to Cochrane & Barber (Reference COCHRANE and BARBER2009), the future fire regime will be a product of climate change and human land practices. For example, the predicted warmer and drier climate (Salazar et al. Reference SALAZAR, NOBRE and OYAMA2007) may result in an increase in hazards related to natural fire ignition and propagation. Fires may therefore become more frequent and intense, increasing the vulnerability of the forest and imparting high risks of biodiversity erosion and biomass loss.
In conclusion, the biomass stocks in the studied area were drastically affected by fire, and important shifts in tree size distribution and species contributions to biomass occurred after each additional fire event. Losses in biomass stocks were significant after fire recurrence. Twelve years after the fire, biomass stocks had recovered in once-burned forests, although differences in floristic composition were still observed in higher fire severity classes. For large trees (> 50 cm dbh), there was a notable reduction in biomass after one or two fire events, whereas this reduction was significant in thrice-burned areas. In terms of species contribution, we observed that Cecropia spp. dominated the biomass composition, even 12 y after a severe fire incident. Our results, in combination with those of previous studies, showed that forest recovery of biomass depends on fire history (e.g. elapsed time, fire frequency and severity) and that losses progressively occur until some point between 3 and 7 y after the last fire, assuming no additional disturbances. Both fire frequency and severity exacerbated biomass losses and delayed biomass recovery. Overall, the present study confirms the high vulnerability of these tropical forests (located close to savannas) to fire. It remains difficult to distinguish between the effects of fire frequency, fire severity and time since last fire artefact on vegetation. Long-term surveys are therefore essential for assessing the diversity of ecological impacts of fire on Amazon forests.
ACKNOWLEDGEMENTS
We acknowledge Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq) for the research grants. We also received financial and logistic support from the Instituto Nacional de Pesquisas Espaciais (INPE) and Empresa Brasileira de Agropecuária (EMBRAPA-Roraima). Field work assistance was provided by rural crews from Apiaú and Sr. Erly Pedroso (parabotanist). We also thank Daniel E. Silva for his helpful contributions. The authors are very grateful to the anonymous reviewers for the very useful suggestions.
Appendix 1. Summarized floristic statistics for fire disturbance classes (UF = unburned forest; BF1L = once lightly burned forest; BF1H = once heavily burned forest; BF2 = twice-burned forest and BF3 = thrice-burned forest) in northern Brazilian Amazon; Dens. = density of the listed species (ind. ha−1); dbh = species diameter at breast height (mean ± SD); IVI = importance value index (%) of species considering only tree species. An asterisk (*) indicates the pioneer species.
