INTRODUCTION
The population dynamics of plants are determined by vital rates of individuals, including reproduction, growth, recruitment and survival. These vital rates can respond to temporal environmental heterogeneity caused by natural events such as fire, hurricanes, low annual precipitation, high mean temperature, high seed predation and high herbivory (Elderd & Doak Reference ELDERD and DOAK2006, Evans et al. Reference EVANS, HOLSINGER and MENGES2008, Pascarella & Horvitz Reference PASCARELLA and HORVITZ1998, Ticktin Reference TICKTIN2003). Plant populations intensively influenced by natural disturbances have a high probability of changing vital rates or have high resilience following disturbance (Alfonso-Corrado et al. Reference ALFONSO-CORRADO, CLARK-TAPIA and MENDOZA2007, Jacquemyn et al. Reference JACQUEMYN, BRYS, HONNAY, HERMY and ROLDÁN-RUIZ2006, Sletvold & Rydgren Reference SLETVOLD and RYDGREN2007). These natural environmental disturbances can be deterministic, with regular periodic fluctuations, or stochastic, when the occurrence of the event has a probability distribution but cannot be predicted with certainty. Therefore, a model of population dynamics that includes temporal variation in environmental conditions is considered more realistic than a deterministic one (Caswell Reference CASWELL2001).
Stochastic events of herbivory can influence plant population dynamics. The effects are dependent on factors such as how herbivores influence vital rates; the frequency and intensity of herbivory events (Bastrenta et al. Reference BASTRENTA, LEBRETON and THOMPSON1995, Ehrlén Reference EHRLÉN1995); and the interaction between herbivores and abiotic factors (Elderd & Doak Reference ELDERD and DOAK2006, Endress et al. Reference ENDRESS, GORCHOV and NOBLE2004). The mortality of palm individuals due to mammals has been described many times. Examples include the predation of Podococcus barteri by porcupine (Bullock Reference BULLOCK1980); the apical meristem of juveniles of Oenocarpus mapora subsp. mapora by an unidentified mammal species (De Steven Reference DE STEVEN1989); individuals of Desmoncus orthacanthos by deer, pigs and rodents (Siebert Reference SIEBERT2000); and the apical meristems of reproductive individuals of Geonoma brevispatha by capuchin monkeys (Souza & Martins Reference SOUZA and MARTINS2006). However, the effects of these natural herbivory events on palm population dynamics and the processes of population maintenance following disturbances remain unknown.
Geonoma schottiana Mart. occurs in riparian forests in central Brazil, where juveniles of this palm species can be intensively preyed upon by mammals (Sampaio Reference SAMPAIO2006). The processes of resistance to disturbances include high proportion of individuals in the initial life stages, clonal growth, high fertility, abundant and permanent seed bank, and high seedling recruitment rates (Sampaio Reference SAMPAIO2006, Sampaio & Scariot Reference SAMPAIO and SCARIOT2008). We tested the hypotheses that the population dynamics of this palm is not influenced by stochastic events of herbivory by mammals. We expect that these processes of resistance could contribute to the long-term maintenance of a large population of this species in an environment subject to natural perturbations. To test the hypothesis we asked four questions: (1) Does the natural perturbation cause significant change in the demography of an abundant population of this palm species? (2) If yes, how does it occur? (3) What are the most important processes of population maintenance in a stochastic environment? and (4) What is the highest frequency of intensive disturbance events the population can tolerate? Answering these questions can contribute to understand the effects of temporal fluctuations on the population dynamics of an understorey palm species in the riparian tropical forest.
METHODS
Area and study species
The study was carried out in a riparian forest along the headwater of the Três Barras stream in the Chapada Contagem plateau in the National Park of Brasília, a well preserved reserve of 30 000 ha. The forest occurs in a narrow strip surrounded by shrub grassland, in the cerrado of the Central Brazil (15°35′–15°45′S, 47°53′–48°05′W; c. 1100 m asl). Cerrado is a savanna-like vegetation that originally covered 2 million km2 mostly in the central part of Brazil (Klink & Machado Reference KLINK and MACHADO2005). The study area substrate is composed of well-drained latosol and poorly drained organic soil. The climate has two well-defined seasons, Aw according to Köppen (Reference KÖPPEN1931). In the dry season, from April to October, polar air masses bring low temperatures and low relative humidity. More than 90% of the annual precipitation occurs in the rainy season, from November to March, due to cold fronts and tropical air masses. The mean annual temperature is 21 °C and the mean annual precipitation is c. 1600 mm (Nimer Reference NIMER1989).
Geonoma schottiana (Arecaceae) is monoecious and occurs in several regions of Brazil (Henderson et al. Reference HENDERSON, GALEANO and BERNAL1995); in the cerrado it occurs only in riparian forests. The palm stem, which is frequently bent, is 2–8 m long, 5–15 cm in diameter and has protuberant rings, resulting from leaf scars. During the life cycle, individuals undergo significant changes in morphology, so we classified individuals in six stages: seed, viable and dormant in the soil for at least 1 y; seedling, with underground stem and bifid leaves; infant, with underground stem and at least one leaf with more than two leaflets; juvenile, with underground stem and at least one completely pinnate leaf; immature, with aerial stem and no inflorescences or signs of past reproduction; reproductive, with aerial stem and inflorescences or signs of past reproduction. The palm has clonal growth and ramets of a genet remain connected throughout the life time of the ramets. Ramets of seedling, infant and juvenile stages can be produced by vegetative propagation of individuals of all stages but seedlings. Sexual reproduction contributes only to seed and seedling stages.
Data sampling
We surveyed the population of G. schottiana in 40 permanent plots of 20 × 10-m during four demographic intervals from 2000 to 2004. The plots were randomly chosen among 124 plots of a grid formed by 10 parallel transects 100 m apart and crossing the riparian forest from one edge to the other. Each plot had an internal 5 × 10-m subplot in which we sampled all individuals. In the remaining plot (15 × 10 m) we surveyed only immature and reproductive individuals. All individuals were marked and classified in stages. Each ramet of a genet was marked separately and considered an independent individual in the analysis.
Matrix construction
We constructed four Lefkovitch matrices (At), each one using the demographic parameters from the survey of one time interval. Each matrix was composed of the sum: A = T + F; where T was the transition matrix and F was the fecundity matrix (Caswell Reference CASWELL2001). The element tij of T was the probability of an individual of stage j at time t surviving to become a member of stage i at time t + 1. The transitions were stasis, growth or retrogression. Individuals with underground stems (infant and juveniles) could regress between stages because they could lose the leaf that characterises them. The t 54 transition of the 2000–2001 period was replaced by 0.0001, because no immature individuals recruited to the reproductive stage in this period.
The F matrix included the values of vegetative and sexual fecundity (fij), i.e. the expected number of offspring of stage i, produced by an individual of the stage j. Vegetative fecundity occurred when individuals produced new ramets from basal axillary buds. The sexual fecundity values were estimated from the results of previous studies of reproductive phenology, seed germination in the field, seed longevity and density in the soil seed bank (Sampaio Reference SAMPAIO2006, Sampaio & Scariot Reference SAMPAIO and SCARIOT2008) and demographic data.
The number of seedlings produced by sexual reproduction (f 26) was calculated by f 26 = pg, where p was the total number of seeds produced by a reproductive individual in 1 y and g was the probability of seed germination in the field in the first year following seed dispersal. The number of seeds produced by reproductive individuals that remained viable in the soil seed bank for at least 1 y (f 16) was f 16 = (p–f 26)sb; where sb was the proportion of viable seeds surviving at least 1 y in the soil after dispersal. The probability of seed germination in the soil seed bank (t 21) was:
where np was the number of recruited seedlings in the period, nr was the number of reproductive individuals in the population and nb was the mean number of viable seeds in the soil. The nb value did not change between years and seasons (Sampaio Reference SAMPAIO2006). The stasis of seeds in the soil seed bank (t 11) was:
Asymptotic model
We calculated the asymptotic growth rate (λ) of the four At matrices using the power method (Caswell Reference CASWELL2001). The stage-classified matrix model does not take age explicitly into account, but age is implicitly present in the model since the matrix parameters are estimated on an annual base (Barot et al. Reference BAROT, GIGNOUX and LEGENDRE2002, Cochran & Ellner Reference COCHRAN and ELLNER1992). To calculate the individual age parameters (mean age in the ith stage, yi; mean age of residence in the ith stage, Si; conditional remaining life span of individuals in the ith stage, Ωi; mean time to reach the ith stage from the seed stage, τseed,i; total conditional life span of individuals that have reached the ith stage, Λi) of the four At matrices we used the method proposed by Cochran & Ellner (Reference COCHRAN and ELLNER1992). The elasticity of λ to proportional changes in the matrix elements was estimated for each At matrix (Caswell Reference CASWELL2001, de Kroon et al. Reference DE KROON, PLAISIER, VAN GROENENDAEL and CASWELL1986). To evaluate the annual variation in λ, we calculated the confidence intervals for each λ from 10 000 bootstrap estimates from each At matrix (Caswell Reference CASWELL2001). Each estimate produced one Ak matrix whose elements aij were the mean of the vital rates of 40 plots randomly chosen with the same probability and with replacement. The dominant eigenvalue (λi) was calculated from each Ak matrix. The bootstrap population growth rate (λb) was the mean of the 10 000 λi and confidence intervals were estimated by the percentile method from 5–95% of the distribution of λi.
Life-table response experiment
Juvenile and reproductive mortality rates varied between years, being notably worse in one year then the other three. Vital rate data for this ‘bad’ year were considered separately from the ‘good’ years in a life-table response experiment (LTRE). To calculate the contribution of each matrix element (cij) to the observed differences in λ between the bad (A 3) and good (A 1, A 2 or A 4) years we used a fixed design LTRE: cij = Sij (aij − aij), where Sij was the sensitivities of the mean matrix among periods, and aij was an element of the mean matrix (Am) calculated as: Am = (A 3 + Agood)/2. Agood was one of the good year matrices (A 1, A 2 or A 4). Some elements of the transition matrix (At), such as retrogression, clonal growth and sexual reproduction, were the sum of multiple vital rates. To calculate the contribution of each vital rate we used the procedure described by Martorell (Reference MARTORELL2007). When Sij is positive, a negative value of the cij element indicates the value of that vital rate in the bad year is lower than in the good year, contributing to the observed decrease in λ.
Stochastic model
To evaluate the effect of environmental stochasticity on population dynamics, we estimated the stochastic growth rate [Log(λs)] using an analytical calculation (Caswell Reference CASWELL2001). We assumed that the four matrices At had demographic parameters influenced by environmental conditions that were independent and identically distributed across time. We then used the maximum likelihood estimator from Heyde & Cohen (Reference HEYDE and COHEN1985) with 10 000 simulations (T). For each simulation one of the four matrices At was selected with equal probability (0.25). The population size [N(t)] of each simulation (t) was calculated using: n(t + 1) = An(t) and N(t) = Σnj; where n was the vector column with the number of individuals in each stage j. The initial population size used was N = 1. The population growth rate in each time step (rt) was:
The stochastic population growth rate [Log(λs)] was:
The first 1000 simulations were disregarded to eliminate bias resulting from the initial distribution of individuals across stages. We estimated the confidence interval of log(λs) from the 5th and 95th percentiles of the distribution of the 9000 valid simulations (Caswell Reference CASWELL2001).
Throughout the 4-y study period we found only one bad year, so we could not estimate the true frequency of occurrence of bad years. Thus, we used stochastic models to simulate the effect in the long term of all possible frequencies of bad years on population dynamics. The same procedure described above was used to calculate [Log(λs)] and confidence intervals for each frequency of bad years. We varied the frequency of occurrence of the bad year (A3 matrix) from 0.00 to 1.00 at intervals of 0.01, thus we ran 101 procedures. For each procedure we used T = 2000 and the other matrices (A 1, A 2 and A 4) had equal probabilities of occurring, which varied according to the frequency of A 3. The probability of local extinction of the population was estimated for each frequency of A 3 as the proportion of the 2000 simulations in which population size declined to zero (Caswell Reference CASWELL2001). To each procedure the first 300 simulations were disregarded. We used the software MATLAB (version 6.5.1, The MathWorks, Inc., Natick, MA, USA), for the matrix analysis and a SAS/IML routine created by Sébastien Barot for age estimations.
RESULTS
The density of G. schottiana individuals was (mean ± SD) 14 034 ± 314 ramets ha−1 during the study period. The seedling stage was the most abundant (64–69%) in the population (Table 1). The mean number of seedlings produced annually by sexual reproduction (3525 ± 482 seedlings ha−1) was 17 times higher than the number of individuals (seedlings + infants + juveniles) produced annually by clonal growth (206 ± 33 ind. ha−1). The population had only 219 ± 36 clumps ha−1, with a mean of 3.1 ± 0.1 ramets per clump and a maximum of 15 ramets per clump, where juveniles had clonal growth higher than the other stages (Table 2).
Table 1. Projection matrix of Geonoma schottiana population during the study period (2000–2004) in a riparian forest of Central Brazil. nx = number of individuals sampled and qx = proportion of dead individuals.
*The transition probability was substituted with 0.0001 because no juvenile individuals were found recruiting to the immature stage in this period.
Table 2. Mean vegetative fecundity (± SE) of Geonoma schottiana population stages during the study period (2000–2004) in riparian forest, Central Brazil.
The population dynamics were at equilibrium throughout the study period, because the asymptotic population growth rates did not differ statistically among years and did not differ from unity (Figure 1). During the study period, the elasticity of λ to changes in the transitions was higher for stasis of individuals (82–97%) than growth (2–12%), and for all of the other transitions together (1–6%). The stasis of reproductive individuals was the demographic rate with highest elasticity, with the exception of the 2002–2003 period, when the stasis of the infant stage had the largest elasticity (Figure 2). The stasis of juvenile individuals has the second-highest elasticity in two (2000–2001 and 2003–2004) of the four study periods.
Figure 1. Population growth rate and confidence interval (95%) for a Geonoma schottiana population during the periods 1 = 2000–2001, 2 = 2001–2002, 3 = 2002–2003, 4 = 2003–2004 in a riparian forest, Central Brazil.
Figure 2. Elasticity values of λ to changes in the stasis of the seed, seedling, infant, juvenile, immature and reproductive stages in the periods: 2000–2001 (a), 2001–2002 (b), 2002–2003 (c), and 2003–2004 (d), in a Geonoma schottiana population, in a riparian forest, Central Brazil.
The majority of vital rates were similar between years (Table 1), so the λ values did not differ among years (Figure 1). However, the mortality of juvenile individuals (χ2 = 4.47–13.8, df = 1, P < 0.05) and reproductive individuals (χ2 = 5.19–7.64, df = 1, P < 0.05) was at least three times higher in the 2002–2003 period than in the other periods (Table 1 and Figure 3). We found many juveniles inside and outside of some plots with their apical meristems damaged by mammals in the 2002–2003 period. The event of herbivory occurred clustered in patches inside the study area. We could not identify precisely the mammal species because we found only tooth marks left on destroyed palm individuals and pig rooting in the soil around the individuals, suggesting that potential predators could be white-lipped peccary (Tayassu pecari) or collared peccary (Pecari tajacu), both occurring in the study area. Predated juvenile ramets died, but 16% of them produced a new clonal ramet from basal axillary buds after the event of predation. The apical meristems of individuals of other stages were not preyed upon and we were unable to identify the cause of high mortality of reproductive individuals in the 2002–2003 period (bad year). The bad year changed both the demographic rates with highest elasticities, stasis of reproductives and juvenile individuals. The vital rates seem to have returned to normal levels in the year after (2003–2004) the event of intense herbivory (Table 1 and Figure 1).
Figure 3. Mean mortality (± SE) of juvenile (a) and reproductive (b) individuals of Geonoma schottiana during the study periods 1 = 2000–2001, 2 = 2001–2002, 3 = 2002–2003 and 4 = 2003–2004, in a riparian forest, Central Brazil. Different letters represent statistical differences (P < 0.05) assessed by χ2.
The annual changes in vital rates caused a large variation in the age-based life-history parameters mainly to juvenile, immature and reproductive stages (Table 3). If herbivory events occur annually in the long term, individuals will become reproductive on average with 40 y of age, while if events of herbivory never happen, sexual maturity will occur late in the palm life-cycle (106–386 y). Also, herbivory reduced the mean age of residence and life span of juvenile, immature and reproductive stages (Table 3).
Table 3. Mean (±SD) of age-based parameters (Cochran & Ellner Reference COCHRAN and ELLNER1992) of Geonoma schottiana population during the study period (2000–2004) in a riparian forest, Central Brazil. Mean age in the ith stage, yi; mean age of residence in the ith stage, Si; conditional remaining life span of individuals in the ith stage, Ωi; mean time to reach the ith stage from the seed stage, τseed,i; total conditional life span of individuals that have reached the ith stage, Λi.
According to LTRE results, the low value of λ in the bad year was caused by lower stasis of individuals in the later life stages, mainly reproductive, juvenile, infant and immature individuals (Figure 4). The stasis of reproducers had at least five times higher elasticity than stasis of juveniles in good years, but the LTRE contribution of juvenile stasis was similar or even higher than the contribution of reproducers in the bad year (Figure 4). Conversely, the vital rates with highest contribution for population stability maintenance in the bad year were the growth of infant and juvenile individuals and clonal growth of juveniles. These patterns were generally consistent among years (Figure 4).
Figure 4. Life-table response experiment (LTRE) contributions of growth (a), stasis (b), clonal growth (c) and retrogression (d) to the difference in the annual population growth rate between bad (2002–2003) and good years (2000–2001, 2001–2002 and 2003–2004) for a Geonoma schottiana population in a riparian forest, Central Brazil. The stages are 1 = seeds, 2 = seedlings, 3 = infant, 4 = juvenile, 5 = immature and 6 = reproductive. Note that the scale of the y-axis differs between panels.
If the environmental conditions of the 4-y study period are representative of the long-term environment and independent and identically distributed, the population will be in equilibrium, because the stochastic growth rate [Log(λs)] was 0.015 (CI95% = −0.041 to 0.007), not statistically different from zero. Throughout the study period we found only one bad year, so we could not estimate the true frequency of occurrence of bad years. In spite of this, even if a bad year occurs once every 3 y (i.e. frequency of occurrence of A3 matrix = 0.33) the population will be in equilibrium (Figure 5a). Only if the occurrence of bad years is greater than once every 2 y is the population size expected to decrease significantly (frequency >0.50; Figure 5a). The local extinction probability of the population is negligible for frequencies of occurrence lower than 0.6, but if the frequency of bad years is higher than 0.6 for a long time, the probability of local extinction in 300 y increases greatly (Figure 5b). Considering that the observed frequency of bad years (0.25) occurs according to the binomial distribution, there is less than 15% of chance that the true frequency of bad years is higher than 0.6. However, there is a chance higher than 30% that the true frequency of bad years is between 0.10 and 0.45 (Figure 5c).
Figure 5. Effects of the frequency of occurrence of bad years (A3 matrix) on the stochastic growth rate (± confidence interval, dotted line) (a), extinction probability (b) and likelihood of the frequency of occurrence of bad years according to the binomial distribution (c) for a Geonoma schottiana population in a riparian forest, Central Brazil.
DISCUSSION
Geonoma schottiana have a high proportion of individuals in the initial life stages, clonal growth, high fertility, with seeds produced during nearly all months of the year (Sampaio & Scariot Reference SAMPAIO and SCARIOT2008), that can remain dormant for more than 4 y in an abundant soil seed bank (Sampaio Reference SAMPAIO2006), with a large number of seedlings emerging annually (3525 ± 482 seedlings ha−1). All of these processes could contribute to population maintenance in stochastic disturbances (Alfonso-Corrado et al. Reference ALFONSO-CORRADO, CLARK-TAPIA and MENDOZA2007, Jacquemyn et al. Reference JACQUEMYN, BRYS, HONNAY, HERMY and ROLDÁN-RUIZ2006, Sletvold & Rydgren Reference SLETVOLD and RYDGREN2007). Moreover, the population density is high in relation to other natural populations of Geonoma species (Table 4) and is similar to the density found for the abundant Geonoma orbignyana (Rodríguez Butiricá et al. Reference RODRÍGUEZ-BUTIRICÁ, ORJUELA and GALEANO2005).
Table 4. Density of individuals of Geonoma species in the understorey of different tropical forest types.
All of those processes of resistance are of minor importance to population dynamics of G. schottiana, as we can see in the results of the elasticity and LTRE analysis, and despite them, the growth rate of the abundant population can be reduced when bad years occur frequently. Additionally, the bad years can reduce the life span, mean age of residence and time to reach the reproductive stage. The cause for the occurrence of the intense herbivory event on juvenile individuals can be the low availability of resources to predators in some years. Despite the pattern of relative frequency of fruiting being annual for species of riparian forest (Funch et al. Reference FUNCH, FUNCH and BARROSO2002), the biomass of ripe fruits can be low in some years.
The herbivory of Geonoma brevispatha reproductive individuals by the capuchin monkey (Cebus apella) is associated with a dry period and reduces population growth of this palm species in a swamp forest (Souza & Martins Reference SOUZA and MARTINS2006). To the G. schottiana population, the demographic changes between years were mainly due to the event of herbivory by vertebrates. The highest contributor to the annual variation was the juvenile mortality. This vital rate had a negative contribution to the population growth rate in the bad year in relation to good years. However, a substantial population size decrease of G. schottiana will occur only if the stochastic environmental changes promote frequent reductions in the stasis of reproductive and juvenile individuals, the transitions for which the elasticity of λ is highest. Similarly, the growth rate of Chamaedorea radicalis populations are severely reduced when donkey browsing increases the mortality of juveniles and small reproductive individuals (Endress et al. Reference ENDRESS, GORCHOV and NOBLE2004). Otherwise, when bad years occur, the processes with the highest contribution to the maintenance of the G. schottiana population are the growth of infant and juvenile individuals and the clonal growth of juveniles.
The higher elasticity of λ to the stasis of reproductive individuals than the vital rates of the initial stages (mainly seed and seedling) found in G. schottiana is similar to the majority of long-lived plant species (Kwit et al. Reference KWIT, HORVITZ and PLATT2004, Silvertown et al. Reference SILVERTOWN, FRANCO, PISANTY and MENDOZA1993, Zuidema & Boot Reference ZUIDEMA and BOOT2002) and to the majority of palm species living in different environmental conditions and having different growth forms (Barot et al. Reference BAROT, GIGNOUX, VUATTOUX and LEGENDRE2000, Enright & Watson Reference ENRIGHT and WATSON1992, Kouassi et al. Reference KOUASSI, BAROT, GIGNOUX and ZORO BI2008, Olmsted & Alvarez-Buylla Reference OLMSTED and ALVAREZ-BUYLLA1995, Pinard Reference PINARD1993, Souza & Martins Reference SOUZA and MARTINS2006). This pattern contrasts with that of herbaceous species, for which seed survival and reproductive fecundity are the main factors responsible for population maintenance (Adams et al. Reference ADAMS, MARSH and KNOX2005, Menges & Quintana-Ascencio Reference MENGES and QUINTANA-ASCENCIO2004, Silvertown et al. Reference SILVERTOWN, FRANCO, PISANTY and MENDOZA1993, Sletvold & Rydgren Reference SLETVOLD and RYDGREN2007). This fact can explain the low contribution of the processes such as the high fertility, abundant and permanent seed bank, and high seedling recruitment rate to the population dynamics of G. schottiana. In turn, the clonal growth of G. schottiana and other palm species (Chazdon Reference CHAZDON1992, Souza & Martins Reference SOUZA and MARTINS2006, Svenning Reference SVENNING2000) contributes greatly to reduce the mortality of established genets in bad years, as proposed by De Steven & Putz (Reference DE STEVEN and PUTZ1985). This result contrasts with the expectations of Bullock (Reference BULLOCK1980), that clonal growth is only a form of individual propagation. Three aspects support this hypothesis: (1) at least 16% of the juveniles preyed on produced a new ramet by clonal growth in the bad year; (2) clonal growth was lower in good years than in the bad year (Table 2); and (3) the number of new individuals produced by sexual reproduction was at least 17 times higher than the number of individuals produced by clonal growth.
Despite the importance of clonal growth in bad years, G. schottina had a lower investment in vegetative propagation than other palm species. Pinanga coronata has up to 31 ramets per genet and approximately 9.5% of individuals occurring in clumps (Kimura & Simbolon Reference KIMURA and SIMBOLON2002). Geonoma congesta has 98% of the population in clumps and a maximum of 29 ramets per genet (Chazdon Reference CHAZDON1992). Geonoma schottiana probably had lower investment in clonal growth than in sexual reproduction because the population occurs in a riparian forest, which is a mesic environment. Thus, the microhabitat conditions are favourable to seed germination, while other species living in environments with conditions that restrict the establishment of new individuals such as deserts (Clark-Tapia et al. Reference CLARK-TAPIA, MANDUJANO, VALVERDE, MENDOZA and MOLINA-FREANER2005) or permanently flooded forests (Souza & Martins Reference SOUZA and MARTINS2004, Reference SOUZA and MARTINS2006) have poor seed germination and high clonal growth. Therefore, the clonal growth is essential to the recruitment of new ramets of the clonal palm species living in restrictive environments. In this case, ramet establishment is positively related to genet size. Additionally, clonal growth increases the probability of genet survival because pre-reproductive ramets behave as a meristem bank in these species (Chazdon Reference CHAZDON1992, Souza & Martins Reference SOUZA and MARTINS2006). In G. schottiana, clonal growth seems not to be important to genet survival and to ramet recruitment, having only a secondary importance to population growth rate in good years, but having a high importance for population resilience, mainly when stochastic events of intensive ramet mortality occur. The established genets of G. schottiana may invest more stored resources than young individuals into clonal growth to increase their own fitness, contributing to regeneration of the population following unfavourable habitat conditions, similarly to herbaceous species (Jacquemyn et al. Reference JACQUEMYN, BRYS, HONNAY, HERMY and ROLDÁN-RUIZ2006).
Geonoma schottiana has a large population in the riparian forest studied and many processes of persistence. The main vital rates responsible for long-term population viability are the stasis of reproductive individuals and clonal growth of juveniles. However, if stochastic disturbance events occur with frequency higher than once every 2 y over a long period of time, the population will have a negative growth rate and the probability of local extinction will be very high. The effects of intensive and frequent herbivory events can have even worse consequences for the persistence of smaller populations, not only of G. schottiana, common in the riparian forests of Central Brazil, but also of many other understorey palm species of tropical forests. Furthermore, this study showed that stochastic events of herbivory can cause most of the variation between years in the dynamics of plant populations.
ACKNOWLEDGEMENTS
We thank Carol Horvitz and two anonymous reviewers for their suggestions to improve the manuscript and Sébastien Barot for the SAS routine to age parameters estimation. Thank you to our field assistants Bernardo Bianchetti, Brunno S. de Andrade, Gustavo de O. Lopes, Hugo A. H. Schaedler, Ísis M. Medri, Juarez P. do Amaral, Leandro de S. Lima, Nilton F. Barbosa and Patrícia C. Bueno, and the staff of the National Park of Brasília and IBAMA. Lisa Mandle reviewed the English. CAPES and CNPq awarded a fellowship to the first author. Ecology and Conservation Laboratory of Embrapa Recursos Genéticos e Biotecnologia provided resources for fieldwork.
