oristic composition of two Cerrado savanna communities on different substrates?

We surveyed two savanna sites, one on fl at terrain with deep soil (DS), and the other on hilly terrain with rocky outcrops and shallow soil (RS), before and after an accidental fi re. We found that the fi re did not cause any signifi cant changes in the species composition or diversity of either community, and did not result in fl oristic homogenization. However, we did record a reduction in the density of plants and in basal area in the DS savanna in comparison with the RS savanna, as well as a higher rate of basal sprouting, which indicates a trade-off between mortality and sprouting. We conclude that, whereas post-fi re changes in vegetation structure were more pronounced in the DS savanna than in the RS, the diff erence in the underlying substrate did not have a direct infl uence on the post-fi re composition of woody species. Th e greater grass biomass found in the DS savanna in comparison with the RS savanna appears to have been the principal modulator of the severity of the fi res in the two phytophysionogmies, and accounts for the distinct responses to fi re we observed in the two woody communities.


Introduction
Savanna-like ecosystems are widely distributed in the tropics, where they can be found in an ample variety of habitats, ranging from fl at plains to mountainous terrain, on a diversity of soil types (Collinson 1 988;Young & Solbrig 1993;Mews et al. 2014).Th e vegetation of tropical savannas is physiognomically very variable, but is characterized by the coexistence of well-established arboreal and grassy strata.Fire is one of the principal ecological drivers of savannas worldwide (Bond & Keel ey 2005), modulating the coexistence of grassland, savanna and forest formations (D'odorico e t al. 2006;Geiger et al . 2011), and shaping the diff erent physiognomies.Th rough its interaction with the climate and soil, fi re can modify the basal area of woody species in savanna habitats, with profound implications for carbon storage (Lehmann et al. 2014).Fire is thus the agent that determines the coexistence between trees and grasses (D'odorico e t al. 2006) and infl uences the occurrence of plants with distinct functional traits in terms of their resistance to fi re (Dantas et a l. 2013), such as thick bark (Dantas & Pa usas 2013) and basal sprouting capacity (Hoff mann 19 98).
In Brazil, most of savanna is known as the Cerrado, which originally covered an area of more than two million square kilometers.Th e savanna formations of the Cerrado are found primarily on deep soils (mainly Oxisols) on relatively fl at terrain (Reatto et a l. 2008).However, around 7 % of the Cerrado is made up of savannas growing on shallow soils associated with rocky outcrops (Entisols), generally on steep, hilly relief.Recent evidence indicates that these two types of savanna are highly similar in their woody species composition, but diverge considerably in the density of their populations (Mews et al. 2014).These savannas are also quite distinct in their grassy cover and the amount of biomass, which is the primary source of combustible material, and is reduced in savanna on rocky soils.A greater grass biomass intensifies the severity of fires (Miranda et al. 1993;2002;Felfili et al. 2000), which reduce tree density (Medeiros & Miranda 2005;Gomes et al. 2014), the biomass of the vegetation (Hoffmann 1996), and recruitment rates (Hoffmann 1996;1998).Fire also supports the selective exclusion of the more sensitive plant species (Hoffmann & Moreira 2002;Gomes et al. 2014), leading to the simplification of species composition and, consequently, the reduction of species diversity over time (Libano & Felfili 2006).More frequent fires increasingly damage the woody vegetation of savannas worldwide (Lehmann et al. 2014), including the Brazilian Cerrado (Moreira 2000), making the vegetation more open.Given the ongoing increase in the frequency of fires (Ramos-Neto & Pivello 2000;Pivello 2011) and the progressive conversion of Cerrado habitats into farmland (Sano et al. 2010), especially on areas of flatter terrain with deep soils (Klink & Machado 2005), it is increasingly important to understand the relationship between fire and these different types of substrate.
In the present study, we surveyed two physiognomies of the Cerrado sensu stricto (see Ribeiro & Walter 2008), located on different types of substrate before and after an accidental fire.This category (Cerrado sensu stricto) is the predominant type of vegetation (70 % of the total area) found in the Brazilian savanna, whose structure is characterized by a mixture of shrubs and small trees with twisted branches, and a herbaceous layer dominated by grasses (Ribeiro & Walter 2008).The Cerrado sensu stricto is subdivided into four typical savanna physiognomies differentiated by the structure of the vegetation (mainly in the density and height of the shrub-tree layer) (Ribeiro & Walter 2008) and the properties of the soil, in particular the fertility, depth and moisture (Ruggiero et al. 2002;Carvalho et al. 2014).These physiognomies are also maintained or modified by fire (Moreira 2000).Our objective was to understand the effects of the fire on floristic composition, species diversity, and the density and sprouting patterns of the woody plants.
In particular, we tested the prediction that fire results in greater changes in species composition and diversity, as well as a greater reduction in plant density and basal area in the savanna on deep soil on flat terrain (DS) in comparison with the savanna on hilly, rocky soil (RS).This prediction was supported by the fact that grasses form a well-defined and continuous stratum in the DS savanna, providing more fuel for fires in comparison with the RS savanna, where the grassy layer is discontinuous and occupies the gaps between the rocks (Oliveira-Filho & Ratter 2002;Ribeiro & Walter 2008).We discuss the implications of our findings for the conservation of the two physiognomies and for the regional diversity of plant species.

Study area
We conducted this study on a private property in the municipality of Nova Xavantina, in the state of Mato Grosso, Brazil (Fig. 1A, B).We sampled one savanna site on flat terrain with deep soil (DS) (14º48'S, 52º34'W) and a second site on hilly terrain with rocky outcrops and shallow soil (RS) (14º48'S, 52º35'W) approximately 1.5 km away.The region's climate is of Köppen's Aw type, with well-defined dry (May-September) and rainy (October-March) seasons (Silva et al. 2008).According to data from the Meteorological Station located in Nova Xavantina-MT (9th District of Meteorology, Ministry of Agriculture, INMET 83319-MT), the average annual rainfall calculated over a 15-year period was approximately 1,536 mm, with annual mean around 25 ºC (Marimon & Felfili 2006).According to the landowner, the two studied areas has been accidentally burned every two years since 2004, that is, in 2006, 2008, 2010, 2012, and 2014.The DS savanna is located at an altitude of 440 m on quartzitic sandy soils, while the RS savanna is located at an altitude of 340 m, on quartzitic soils.

Data collection
At each site, we established 10 plots of 20 × 50 m, with a total area of one hectare (Felfili et al. 2005) and a minimum distance of 50 m between plots.Each plot was subdivided into 10 subplots of 10 m × 10 m (Fig. 1C).We identified and measured the stem diameter of all live woody plants (including lianas and arborescent monocotyledons), with a base diameter, measured at 30 cm from the soil (Db 30 cm ) ≥ 5 cm.When possible plants were identified in the field, but sampled collections were compared with vouchers in the NX herbarium or sent to taxonomists for confirmation.We used the APG IV ( 2016) system of families and the species nomenclature was confirmed accessing the Brazilian Flora Checklist (Lista de Espécies da Flora do Brasil 2015).
The first vegetation inventory was in August 2014, one month before the occurrence of a fire, and again in May 2015, eight months after the fire.In October 2014, one month after the fire, we surveyed the two sites and discovered that all the DS plots had been burned, while seven RS subplots (7 % of total) were unburned (four subplots in plot four and three in subplot seven).During these inventories, we included all new plants that had reached the minimum size for inclusion in the sample (treated as recruits).Plants that were considered to be alive during the first inventory but dead (aboveground biomass) during the second survey were recorded as dead, even though the underground organs of Cerrado plants may often remain alive (Hoffmann & Solbrig 2003).Therefore, the presence of basal shoots in these plants was also verified.During the second inventory, we estimated the grassy cover in the DS and RS savannas to obtain an indirect measure of the severity of the fire, given that, in the Cerrado, the undergrowth stratum, which includes grasses, comprises the main source of combustible material during the dry season (Miranda et al. 1993).This substrate was estimated in a 0.5 m × 0.5 m grid subdivided into 25 quadrants of 0.1 m × 0.1 m.These samples were obtained at 14 points, with a distance of 5 m between adjacent points, distributed regularly within the plot (Fig. 1C).At each point, the grid was set up at a height of 0.5 m above the ground, and the grass or rocks found in each grid square were quantified.The cover was scored as 0 for the total absence of rocks or grasses, 0.5 for a cover of up to 50 %, and 1 for a cover of more than 51 %.To avoid bias in these estimates, we calculated the arithmetic mean of the values obtained by two different researchers to obtain a measure of the cover at each point.In this case, the minimum and maximum values were 0 (no cover) and 25 (all 25 quadrants with a cover of at least 50 %), respectively.We then calculated the arithmetic mean of the 14 estimates to provide a general estimate of the grass cover in the 10 plots, as well as the percentage of rocky cover in the RS.

Data analysis
We used Sørensen (qualitative) and Morisita (quantitative) indices (Brower & Zar 1984) to compare the similarity of the species composition between sites, and at the same site between years (before and after the fire).We applied a Principal Coordinates Analysis (PCoA) to the data on species composition and density, to order the plots, including both pre-and post-fire sampling.We then used an ANOSIM (Clarke & Warwick 1994) to test the significance of the groups formed in the PCoA -DS savanna before (DSBF) and after the fire (DSAF) and RS savannas before (RSBF) and after the fire (RSAF).
We produced rarefaction curves based on the adjustment

Does fire determine distinct floristic composition of two Cerrado savanna communities on different substrates?
of the number of plants to compare the estimates of species richness (Gotelli & Colwell 2001) within and between as communities before and after the fire.We used diversity profiles based on the Rényi exponential series (Tóthmérész 1995) to compare the diversity of species between sites and inventories.This method permits the integrated application of a whole family of diversity indices, ranging from those which give more weight to the rare species to those that prioritize the influence of abundant species.
We used a paired t test to compare the mean densities of plants and the basal area of the live plants found in each plot in each community between years.We also compared the mean grassy cover by using the Mann-Whitney test.We then compared the frequency of dead plants sprouting and not sprouting in the two communities using Chi-square with Yates' correction.We adopted a 5 % significance level for all analyses.

Results
We recorded 99 species in the two communities, both before and after the fire (Tab.1).Before the fire, the two communities had 45 species (46 % of the total) in common, and after the fire, 44 (45 %).There was thus no evidence of any change in the similarity of the composition of the DS and RS savannas following the fire (Tab.2).A certain amount of variation was also found in the similarity indices when considering the same community before and after the fire (Tab.2).In this case, the fire neither increased nor decreased the similarity of the two communities, and did not induce any major alteration in the species composition of either phytophysiognomy (Fig. 2).These results were also confirmed by the PCoA (Fig. 2), given that the two groups (DS and RS savannas) established before the fire presented minor alterations after the fire (axis 1= 53.69 %; axis 2= 10.09 %) and were confirmed by the ANOSIM (R= 0.593; p < 0.001).
Minor changes were also found in the densities of the plant species after the fire in the two communities.None of the most abundant species (represented by at least 20 plants) suffered a loss of more than 10% of their abundance in the RS savanna following the fire, and only Myrcia lanuginosa (reduction of 16.5 %), Heteropterys byrsnonimifolia (13.6 %) and Hymenaea stignocarpa (12.5 %) were severely affected in the DS savanna.No increase in abundance exceeding 5 % was recorded in any of the more abundant species.In the case of the least abundant species (n = 1 plant per site), Table 1.Species abundance of the woody vegetation of the adjacent deep soil (DS) and rocky soil (RS) savannas before (BF) and after (AF) the fire that occurred in September 2014, in the municipality of Nova Xavantina, Mato Grosso, Brazil.The species are sorted by the sum of columns (DSBF, DSAF, RSBF, RSAF).Erythroxylum engleri was recorded in the RS savanna only before the fire, and Myrcia multiflora only after the fire.In the DS savanna, one plant of Miconia albicans was recorded only after the fire.No other species represented by at least two individuals arose or disappeared between inventories.

Species
In the RS savanna, we recorded 76 species before fire and 75 after the fire, while in the DS savanna, there were 68 species before the fire, and 67 afterwards.The species richness of the RS savanna was consistently higher than that of the DS savanna, although there was no major change in either community following the fire (Fig. 3).Similarly, the RS savanna had higher species diversity than the DS savanna both before and after the fire (Fig. 4).However, no major change in diversity was found after the fire in either community (Fig. 4).
The post-fire mortality rate in the DS savanna was double that of the RS savannas, and the recruitment rate was two times lower in the DS savanna (Tab.3).Given this, there was a significant reduction (5.3 %) in the density of live plants in the DS savanna following the fire (t= 2.893; p= 0.018), but a much more discreet reduction (0.7 %) in the RS savanna (t= 0.595; p= 0.566).Similarly, while there was a significant loss of basal area in the DS savanna (t= 2.72; p= 0.023), no such tendency was found in the RS savanna (t= 0.351; p= 0.733).The frequency of dead plants with basal sprouting after the fire was significantly higher in the DS savanna in comparison with the RS savanna (χ 2 = 6.830, p= 0.009).Eight months after the fire, the grass cover of the DS savanna was significantly greater than that recorded in the RS savanna (U = 5.26; p < 0.0001 (Tab.3).

Discussion
Differences in the species composition of woody savanna communities on different types of substrate have also been recorded in other studies on local (Gomes et al. 2011;Abreu et al. 2012) and regional scales (Mews et al. 2014).However, we found no evidence of any homogenization or simplification of the species composition of the two communities, nor any shift in the species diversity of either community, following the fire.However, these findings should be considered with caution, as the high frequency of fires in the years prior to the period of the present study (five events -virtually one every two years -between 2004 and 2014 at the two sites) may have resulted in the loss of the most sensitive species that are locally rare.A similar situation has been observed in other fire-affected areas of the cerrado (Sato & Miranda 1996;Silva et al. 1996;Gomes et al. 2014), as well as the present study, in which the two species excluded after the fire, Miconia albicans in the DS and Erythroxylum engleri in the RS, were each represented by only a single specimen prior to the fire.
Following fires, minor changes been observed in the species richness and composition of savannas on shallow, rocky, and hilly substrates (Gomes et al. 2014) and also on deep soils and flat terrain (Silva et al. 1996;Lima et al. 2009;Lopes et al. 2009;Ribeiro et al. 2012) in Brazil, as well as in savanna communities in other parts of the world (Higgins et al. 2000;2007).However, fires may also cause considerable impacts on species richness and composition, Does fire determine distinct floristic composition of two Cerrado savanna communities on different substrates?as recorded by Mews et al. (2013).These inconsistencies are probably the result of the distinct history of wildfires in each community.The Cerrado studied by Mews et al. (2013) was protected from fire for more than a decade, while the two savannas studied here had been burned off every other year over the past 10 years.This may have determined the accumulation of fine-stemmed and inflammable biomass in the savanna studied by Mews et al. (2013), resulting in more intense fires (Pivello & Norton 1996;Pivello 2006;França et al. 2007;Higgins et al. 2007), which had a greater impact on the flora.In areas with a prolonged absence of fire, the establishment of more fire-sensitive plant species will tend to be more likely (Moreira 2000), making these habitats more vulnerable to fire-induced changes in composition.
The low mortality rates recorded here in both the DS (5.6 %) and the RS savannas (2.6 %) were much lower than those recorded in other studies of woody Cerrado savanna communities following fires, whether on deep soils, i.e. 41% in Hoffmann et al. (2009), 67.4 % in Mews et al. (2013), and 13-16 % in Sato & Miranda (1996), or shallow ones, i.e., 43.6 % in Gomes et al. (2014).These mortality rates caused a significant reduction in the abundance of trees and shrubs in the Cerrado, in contrast with the pattern observed in the present study (Sato & Miranda 1996;Fiedler et al. 2004;Lima et al. 2009;Mews et al. 2013).We believe that the low mortality rates recorded in the present study are related to the relatively high frequency of fires in the two studied savannas.In the Cerrado, fires are known to cause higher mortality rates in plants of smaller size (Hoffmann & Solbrig 2003;Gomes et al. 2014) and will decimate or eliminate the populations of the more sensitive species.In this case, recurring fires prior to the one observed in the present study may have filtered out most of the smaller individuals and the species most sensitive to the effects of fire.In this scenario, the remaining individuals in 2014 were likely to have been the least susceptible to fire.
More intense fires increase mortality rates in the Cerrado (Hoffmann & Solbrig 2003).In the present study, the more significant reduction in plant density observed in the DS savanna indicates that the effects of the fire were more severe in this environment than in the RS savanna.This appears to have been related to the larger quantity of biomass found in the grassy-herbaceous stratum of the DS savanna, given that these plants represent the main source of fine-grade combustible material that fuels wildfires (Miranda et al. 2002).In this case, the presence of rocky outcrops in the RS savanna limits the establishment of grasses, and thus restricts the accumulation of biomass, resulting in less severe fires in this community in comparison with the DS savanna.This conclusion is reinforced by the fact that the fire did not affect seven RS subplots in any way, due to the rocky substrate and the absence of combustible material, while in the DS savanna, all the subplots were burnt completely.
Fire is known to cause a major loss in the basal area of live plants in the savanna communities of the Cerrado sensu stricto growing on deep soils (Mews et al. 2011) and on shallow soils interspersed with rocky outcrops (Gomes et al. 2014).However, the present study is the first to compare the effects of fire on the mortality and live biomass of adjacent areas of DS and RS savanna.In the present study at two sites with a high frequency of fires over the preceding 10 years, we found lower mortality rates and a reduced loss of basal area at both sites, regardless the substrate, in comparison with other studies (Mews et al. 2011;Gomes et al. 2014).Based on this, we conclude that the effects of fire on the resistance of the woody community, also depends on the history (frequency) of fire at each site, a phenomenon that demands more attention in future studies.
The higher stem mortality rate (topkill) and the greater loss of aboveground basal area recorded in the DS savanna in comparison with the RS appear to have been compensated by a higher frequency of basal sprouting.This mechanism has been recorded amply in the woody savanna communities of the Cerrado and other savannas worldwide (Higgins et al. 2000;Hoffmann & Solbrig 2003).However, despite the much lower topkill recorded in the RS savanna, the capacity for basal sprouting in this vegetation was also reduced.We believe that the partial or total lack of soil in the RS savanna constitutes a limiting factor for the basal sprouting capacity of this community, particularly because, in these environments, the hydrological deficit tends to be extremely high during the dry season (Oliveira-Filho & Martins 1986).In this case, while the rocky outcrops of the RS savanna may contribute to a reduction of the severity of the effects of fires -and consequently, plant mortality -they may also restrict the basal sprouting capacity of the plants burned by the flames.Interestingly, after the fire, the recruitment of individuals was three times greater in the RS than in DS.On this, we believe that in the RS savanna the fire may have caused less mortality in individuals with potential for recruitment, i.e., those with a stem diameter lower than the minimum threshold (5 cm) adopted in the present study, as there are fewer grasses and, therefore, less availability of fine fuels, which can make fires less severe and harmful.
We conclude that fire did not have differential effects on the species composition and diversity of the woody savannas growing on different substrates, although it has an impact on both.Even so, the post-fire dynamics of the two studied areas indicate that vegetation structural changes are more pronounced in savannas with a large and more continuous grassy-herbaceous stratum.Thus, our findings suggest that savannas more threatened by the expansion of agriculture in Brazil (i.e., DS savannas; Mews et al. 2014) may have their vegetation more structurally impacted by fire, which constitutes another erosion factor of biodiversity in savannas of Cerrado in the Central Brazil.In addition, we also showed, for the first time, a probable trade-off between mortality and basal sprouting in the two communities, characterized by greater resistance and reduced resilience, in short time, in the community of woody plants of the rocky soil savanna, in contrast with low resistance and high resilience of the deep soil savanna.However, our conclusions are limited to the understanding of the immediate response of the vegetation (less than one year) and the findings do not provide direct insights into the behavior of fire in the two physiognomies.In this case, longer periods of monitoring (e.g., Gomes et al. 2014 in the RS) in a larger area of habitat will be necessary to better comprehend the pattern of fire impact and the resistance and resilience of the DS and RS to fire, and the consequences of fire for the conservation of the flora of these two physiognomies.

Figure 1 .
Figure 1.A-C.Location of study area in eastern Mato Grosso, Brazil, and the study plots before and after the occurrence of fire.A = Brazil, showing the state of Mato Grosso (MT) and the Amazon (dark gray) and Cerrado (light gray) biomes, according to the classification of Brazilian Institute of Geography and Statistics (IBGE); B = Municipality of Nova Xavantina, showing the study area.C = Schematic diagram of the ten subplots used to sample the shrub-tree vegetation at each savanna site, showing the points used for the collection of the samples of grass and rock cover in deep soil and rocky savannas, as well as the rockiness of the latter habitat.

Figure 2 .
Figure 2. Ordination plot of the PCoA for the species composition of the deep soil savanna before (DSBF, Δ) and after the fire (DSAF, ▲), and the rocky soil savanna before (RSBF, ○ ) and after (RSAF, •) the fire, in the municipality of Nova Xavantina, Mato Grosso, Brazil.

Figure 3 .
Figure 3.Estimated species richness the deep soil savanna before (DSBF) and after the fire (DSAF), and the rocky soil savanna before (RSBF) and after (RSAF) the fire, in the municipality of Nova Xavantina, Mato Grosso, Brazil.The 95 % confidence interval is shaded in gray and the vertical line at the end of each curve represents the point of comparison pre-and post-fire.

Figure 4 .
Figure 4. Profiles of species diversity of the deep soil savanna (dotted line) before (DSBF) and after the fire (DSAF), and the rocky soil savanna (solid line) before (RSBF) and after (RSAF) the fire, in the municipality of Nova Xavantina, Mato Grosso, Brazil.

Table 2 .
Similarity of the species composition (Sørensen in the upper diagonal and Morisita in the lower diagonal) of the adjacent DS and RS savannas before (BF) and (AF) after the fire, in the municipality of Nova Xavantina, Mato Grosso, Brazil.