The study was carried out in the Catimbau National Park (8°24'00″–8°36'35″ S; 37°00'30″–37°10'40″ W, Fig. 1), a protected area covering 607 km² in rural Pernambuco State, northeastern Brazil (Rito et al., 2017). The climate type is semi-arid bsh’ with the transition to the tropical rainy As’ according to the Köppen classification system (Alvares et al., 2013). The annual precipitation ranges between 440–1100 mm with a significant spatiotemporal variation (Rito et al., 2017). The vegetation is predominantly composed of low-stature dry forests, with the families Leguminosae, Euphorbiaceae, Myrtaceae, and Burseraceae being the most species-rich and dominant families (Rito et al., 2017).

Fig. 1.

Location of the study area in (a) South America, (b) north-eastern Brazil and Pernambuco state (in grey), and (c) Catimbau National Park showing the distribution of the 18 0.1-ha plots (circles) along the chronic anthropogenic disturbance (CAD) and rainfall gradients.

(0.63MB).

Although the National Park was established in 2002, traditional farms (i.e., family smallholdings) still live within the area, and activities such as slash-and-burn agriculture, livestock farming (free-ranging goats), collection of living and dead wood, harvesting of non-timber forest products, and hunting are still common, leading to significant pressures on old growth forests (Rito et al., 2017).

Fieldwork was conducted between July 2018 and July 2020 in fifteen 0.1-ha (20 m × 50 m) plots for seed rain assessment and eighteen 0.1-ha plots for evaluating other ontogenetic stages. These permanent plots are part of the Catimbau ILTER project (https://www.peldcatimbau.org), which are distributed along independent gradients of chronic anthropogenic disturbance and annual rainfall (Rito et al., 2017; see also Fig. 1; more details about these gradients are presented below in the appropriate section.). All plots were situated on sandy soils with flat terrain and supported old-growth vegetation. The plots were separated by a minimum of 2 km and located within a total area of 21,430 ha (Rito et al., 2017).

Data on the level of chronic anthropogenic disturbance of each plot were obtained using the global multi-metric chronic disturbances index proposed by (Arnan et al., 2018). Concisely, this index considered the three most common sources of disturbance in the region: livestock pressure, wood extraction, and people pressure. These chronic disturbances indicators were measured using three distinct approaches: (1) Geographic distances based on remote sensing: using satellite imagery and ArcGIS 10.1 software, we measured the proximity of each plot to the nearest house and the nearest road. Since the level of disturbance is inversely related to distance, we used the inverse of distance as our metric; (2) Interviews with local inhabitants: the nearest village to each plot was identified using Geographical Information System and then informal and semi-structured interviews were conducted to assess the number of people in each village, which was weighted by distance from the plots; and, (3) Measures of disturbance in the field: within each plot, we conducted field assessments to measure the length of goat trails, the extraction of alive wood (stem cuts), and the extraction of coarse woody debris (litter). The values of these indicators of disturbance (i.e. proximity to the nearest house, proximity to the nearest road, number of people in each village, goat trail length, alive wood extraction, and coarse woody debris extraction) were standardized between 0 and 1, which makes disturbance metrics of different units comparable and easily to combine in the same index. The index ranged from 2 to 58 (from the lowest to the highest disturbance intensity) among the plots. For more details on the calculation of the disturbance index, please see Arnan et al. (2018).

Based on previous studies (Rito et al., 2017; Silva et al., 2019), mean annual rainfall data were extracted from the WorldClim global climate repository (Fick and Hijmans, 2017). The closest climate stations to the study sites were utilized to obtain these rainfall data, with a spatial resolution of 1 km, using the maptolls package in R 3.6.1 (Bivand and Lewin-Koh, 2020). This package enabled the calculation of the mean annual rainfall for each plot, which ranged from 510 mm to 940 mm.

As suggested by (Sfair et al., 2018), soil fertility for each plot was assessed by randomly selecting three soil samples measuring 10 cm × 10 cm from a depth of 0–30 cm. The soil samples were then sent for laboratory analysis of soil physicochemical properties related to nutrient availability (Al, H, S, P, CTC, Ca, pH, K), water content, organic matter, and acidity. These analyses followed the procedures recommended by the Brazilian Ministry of Agriculture for soil analyses (EMBRAPA, 1997). The soil fertility of each plot was determined using the formula SF = [Ca + Mg + K − log(1 + A1)] × OM + 5 (Lu et al., 2002), where SF = soil fertility, Ca = exchangeable calcium (cmolc dm−3), Mg = exchangeable magnesium (cmolc dm−3), K = exchangeable potassium (cmolc dm−3), Al = exchangeable aluminium (cmolc dm−3), and OM = organic matter (g kg−1). The soil fertility values ranged from 10.8–59.6 among the plots. More details on soil sampling and physical/chemical analyses can be found in Sfair et al. (2018).

To assess light availability in each plot, we estimated the leaf area index throughout 10 hemispherical photographs taken at a distance of 1.50 m from the ground level, at intervals of 5 meters, and either at dawn before sunrise or at dusk after sunset to prevent direct solar radiation in any part of the canopy (Whitmore et al., 1993). Photographs were taken again across the same plots and at the same time three times during the year 2016 within the seasons. We analyzed the hemispherical photographs with the Gap Light Analyser software (GLA, version 2.0) in order to obtain the leaf area index of each plot (Frazer et al., 1999), which ranged from 0.02 to 0.48.

Seed rain – Seed traps composed of a deep squared plastic tray with an effective trapping area were used to collect the seed rain. Holes were drilled through the tray to drain out rainwater. We randomly placed 5 seed traps of 1 m × 1 m at 1 m of the soil level in each plot. Seed rain was collected monthly between July 2018 and July 2020. The materials in the cloths were carefully screened and identified taxonomically in the laboratory. All visible and healthy seeds were counted and used for the analyses.

Soil seed bank – Forty soil samples were randomly collected in 20 cm × 20 cm × 5 cm squares in each plot, ten soil samples at the end of both dry and rainy seasons for two consecutive years (from July 2018 to July 2020). To assess seed viability, a tetrazolium test was conducted. Seeds were immersed in a 1% aqueous solution of 1% triphenyl tetrazolium chloride (TTC), then they were placed in Petri dishes with distilled water, covered with aluminum foil, and stored at a temperature of 35 °C for four hours. After the incubation, the seeds were washed in running water for observation of the embryo. Seeds were classified as viable if a red spot appeared on the seed tissue (indicating positive seed viability) or non-viable if no spot formation occurred (indicating negative seed viability) (França-Neto and Krzyzanowski, 2019).

Regenerating assemblages – In each plot, seedlings, saplings, and resprouts of woody species were sampled using standard forestry methods (Pancel and Köhl, 2016). Seedlings were defined as individuals with a height <1 m, while saplings were individuals with a height between 1 and 1.5 m and a diameter at soil base <3 cm (Ribeiro et al., 2015). Seedling communities were sampled in nine 2 m × 1 m subplots, and saplings in three 5 m × 5 m subplots, both located in the center of each 50 m × 20 m plot and separated by 10 m. All plants were identified at the species level according to the botanical nomenclature followed by the Angiosperms Phylogeny Group (APG IV, 2016) and Brazilian flora list. To determine if the seedlings and saplings were resprouts, a 30 cm hole was dug around each stem as needed. Individuals were classified as true seedlings/saplings if they were not connected to an adult tree, or as resprouts if they were connected to an adult individual that was not a resprout or a ramet according to Noutcheu et al. (2023).

Adult assemblages – In order to classify seeds and woody plant regenerating assemblages as either autochthonous or allochthonous based on the presence of adult individuals within the plot, we used data on adult assemblages from (Rito et al., 2017), who surveyed the permanent plots (i.e., the 18 50 m × 20 m plots) of the Catimbau ILTER project. Adults were defined as individuals with a diameter at soil level >3 cm and a height >1.5 m (Rito et al., 2017). Additionally, we used the adult assemblages to determine the natural regeneration capacity of the species based on the adult-to-offspring ratio (Pausas and Keeley, 2014).

To determine the spatial independence of plots, the species composition similarity among plots (Bray-Curtis index), and the inter-plots distance matrix (one per each regeneration mechanism), we ran a Mantel test. Mantel results indicated spatial independence among our plots for all regeneration mechanisms: seed rain (r = 0.104, p = 0.16), soil seed bank (r = 0.09, p = 0.115), woody plant regenerating assemblages (r = 0.07, p = 0.201), and adult (r = 0.05, p = 0.38).

In order to assess whether the species composition across regeneration mechanisms was recorded adequately, given that the sampling area of each regeneration mechanism was different, the sample coverage was performed (Chao and Jost, 2012) using the following equation:

where f1 and f2 represent the number of species with one and two individuals respectively, and n is the number of individuals. The mean ± standard deviation of the sample coverage was > 0.9% for each regeneration mechanism (seed rain 0.98 ± 0.03; soil seed bank 0.94 ± 0.02; woody plant regenerating assemblages 0.91 ± 0.03 and adults 0.99 ± 0.01), indicating adequate sampling coverage. To examine the taxonomic dissimilarity among the four ontogenetic stages, we calculated the Bray-Curtis dissimilarity matrix for the ontogenetic stages and compared them using the Mantel test.

To determine the structure of all regeneration mechanism assemblages in each plot, we calculated the density, and species richness of the seed rain, soil seed bank, seedlings, saplings, resprouts, and adults. Additionally, the status, sources of seed dispersal and the uses of some species (more abundant) were also informed. To compare species richness among regeneration mechanisms, we used the interpolation and extrapolation methods in the iNEXT package (Chao et al., 2014). Due to discrepancies in terms of abundance and species richness, we focused the statistical comparisons on the woody plant regenerating assemblages (seedlings, saplings, and resprouts), using Generalized Linear Models (GLM) followed by pairwise contrast analyses. Additionally, we used the adult-to-offspring ratio (Pausas and Keeley, 2014) to determine the natural regeneration capacity of the most abundant adult species.

To examine the isolated and interactive influences of the explanatory variables on the abundance and species richness of each regeneration mechanism (i.e., seed rain, soil seed bank, seedlings, saplings, and resprouts) on the adult-to-offspring ratio, and the number of allochthonous regeneration mechanisms, we used a model selection approach. We built generalized linear models (GLM) for the woody plant regenerating assemblages and adult-to-offspring ratio and generalized linear mixed models (GLMM) for the seed rain and soil seed bank. For all models, except the adult-to-offspring ratio, we used a Poisson distribution error and, in the case of overdispersion, a quasi-Poisson distribution error was applied. For the adult-to-offspring ratio, we used a binomial distribution (link = logit). Candidate models for each response variable (species richness and abundance) were chosen based on our explanatory variables and empirical support. Then, Akaike’s information criterion with a correction for small samples (AICc) was used to select the best models. In the case of overdispersion, qAICc was used instead of AICc to rank the models (Calcagno and Mazancourt, 2010). To select the best models, set models with AICc (or qAICc) differences lower than 2 were chosen (Burnham and Anderson, 2002). The best-supported models, which included important predictors beyond the intercept, were identified. The dredge function in the MuMin package in R was used for model selection (Δ ((Q)AICc) < 2). To assess the collinearity among the predictor variables, we calculated the variance inflation factor (VIF) of each predictor using the car package for R (version 3.0.1) (R, 2016). All VIF values were lower than 1.46, with soil fertility = 1.26, rainfall = 1.45, chronic disturbances = 1.24, and leaf area index = 1.11, indicating independence among the predictors (Jou et al., 2014) and considered suitable for further analyses.

The effects of chronic disturbance and environmental conditions on the species composition of each regeneration mechanism were assessed using non-metric multidimensional scaling (NMDS). Then, the envfit method from the vegan package was used to obtain the effect of each variable on the composition of each regeneration mechanism (Oksanen et al., 2019). Data that did not meet the homoscedasticity criteria were log(x) +1 transformed. Pairwise contrast analyses were conducted using the contrast function from the emmeans package (version 1.8.4−1); (Lenth, 2023; Searle et al., 1980)). The GLMM models were built using the lme4 package (version 1.1–7) (Bates et al., 2015), with plots as the random factor. In the case of overdispersion, the bblme package (version 1.0.23.1) (Bolker and R Development Core Team, 2020) was used to build the models. Additionally, a marginal R2 for each selected model was calculated with the method proposed by (Nakagawa and Schielzeth, 2013). All analyses were performed in R (R, 2016).

Over the course of two years, a total of 5,239 seeds were collected in the seed rain, along with 932 in the soil bank, representing 40 and 32 species, and 15 and 12 families, respectively (Table 1). Shannon diversity index ranged from 4.4 in saplings to 27.9 in adults (Table 1). Overall, the assemblages of all forest mechanisms were predominantly dominated by a small set of species (3–5), which were both frequent and accounted for most of the individuals (Table S1, Supplementary material) (Fig. 2a and e). In the seed rain, Pityrocapa moniliformis, Cnidoscolus bahianus, and Senegalia piauhiensis were the most widely distributed species, occurring in 12, 10, and 10 plots, respectively (Table S1, Supplementary material). In terms of abundance, Pityrocarpa moniliformis, Croton heliotropiifolius, and Guapira graciliflora accounted for over 66% of all seeds, while the 25 least abundant species had no more than 50 seeds each (Fig. 2). Regarding the soil seed bank, Pityrocapa moliliformis, Byrsonima gardneriana, and Commiphora leptophloeos were the most frequent species, occurring in 10, 9 and 9 plots, respectively (Table S1, Supplementary material). Pityrocapa moliliformis and Ditaxis desertorum accounted for more than 59% of all seeds, while the remaining 30 species had less than 50 seeds (Fig. 2). Among seedlings, Jatropha mutabilis and Pityrocapa moliliformis were the most widely distributed species, occurring in 11 and 4, plots, respectively (Table S1, Supplementary material). Three species alone accounted for 73.4% of all individuals (Jatropha mutabilis, Pityrocapa moliliformis, and Trischidium molle), while the 12 least abundant species (representing 70% of all species) had no more than 5 individuals each (Fig. 2). In the sapling assemblages, the genus Jatropha represented 83.5% of 85 individuals collected across all studied plots, and among the thirteen species, only three species had more than 5 individual each (Fig. 2). Jatropha mutabilis and Jatropha mollissima were the most distributed species, occurring in 10 and 4 plots, respectively (Table S1, Supplementary material). Among resprouts, Trischidium molle, Pityrocapa moliliformis, and Poincianela mycrophyla were the most frequently occurring species, found in 11, 9, and 8 plots, respectively (Table S1, Supplementary material). Pityrocapa moliliformis, Trischidium molle, and Senegalia piauhiensis accounted for more than 48% of all resprouting individuals, while the remaining 27 species had less than 35 individuals (Fig. 2). Additionally, a substantial proportion of species across the assemblages were rare in terms of both frequency and total abundance. Finally, the assemblages shared a small number of species on average (3.3%), particularly those considered dominant.

Fig. 2.

Species accumulation curves for the three size classes surveyed in this study: (a) seed rain, (b) soil seed bank, (c) true seedling, (d) true sapling, (e) resprout, and (f) adult, as functions of the number of individuals sampled in the 18 0.1-ha m2 plots in the Caatinga dry forest, Catimbau National Park, Northeast Brazil.

(0.42MB).

During the 25-month monitoring period, the accumulated seed rain ranged from 0.2 to 423 seeds per m2 across the stands (Table S2, Supplementary material), with an average species richness of 9.2 ± 0.73 species per m2 per stand (average ± SD). The soil seed bank exhibited variation in density, ranging from 0.07 to 35.5 seeds/m2, with an average species richness of 6.61 ± 0.61 species per m2 (Table S2, Supplementary material). Considering the woody plant regenerating assemblages, seedling density varied from 0 to 5.99 individuals per m2 across forest stands, while saplings varied from 0 to 0.69 individuals per m2, and resprout density varied from 0 to 1.28 individuals per m2 (Table S2, Supplementary material). Species richness exhibited a similar trend, ranging from 1.61 to 5.22 among these three regenerating mechanisms (seedling, sapling, and resprout). Despite such a low density and reduced species richness, these regenerating mechanisms differed significantly in terms of abundance and species richness attributes (Fig. 3. F = 37.72, df = 2, P < 0.001; F = 46.96, df = 2, P < 0.001, respectively). Resprouts supported the highest scores, followed by seedlings (Fig. 3); i.e., 21.6 % more resprouts as compared to seedlings and saplings. Resprouting from stems was the most frequent resprouting type (F = 220.61, df = 2; P < 0.001, Fig. S1 Supplementary material). In woody plant regenerating assemblages, resprouts accounted for 63% of all individuals and, among the most abundant species, the genus Jatropha was the only that did not exhibit any resprouts.

Fig. 3.

Number of individuals and species between resprouts, true seedlings, and true saplings in the 18 0.1-ha m2 plots in the Caatinga dry forest, Catimbau National Park, Northeast Brazil.

(0.21MB).

Regenerating assemblages did not correlate taxonomically with the adult assemblages, as follows: seed rain (r = 0.66, p = 0.001), soil seed bank (r = 0.52, p = 0.001), and regenerating assemblages (r = 0.55, p = 0.001). Accordingly, a substantial proportion of the species attending the regenerating assemblages was classified as allochthonous: 52% of seed rain, 50% of the soil seed bank, 22% of seedling, 31% of sapling, and 33% of resprout. The regeneration capacity, as measured by the adult-to-offspring ratio, varied among the most abundant species and correlated with the regeneration mechanisms, with seedlings achieving the highest scores (0.01−0.27), followed by resprouts (0−0.15), and saplings (0−0.07) (Table 2).

Overall, all regeneration mechanisms responded to the analyzed drivers, although responses were specific, and a significant proportion of the variation remained unexplained.

Briefly, the abundance of seeds (both seed rain and seed bank), seedlings, and resprouts were negatively affected by chronic disturbances (Table 3; Fig. 4a–d, respectively). Additionally, both seed rain and seedling abundances were also negatively affected by soil fertility (Table 3; Fig. 4e and f, respectively), while seedling abundance responded positively to rainfall (Table 3; Fig. 4g). Sapling abundance was negatively affected by the leaf area index across forest stands (Table 3; Fig. 4h). Furthermore, certain environmental factors had contingent effects depending on the level of chronic disturbance. For example, the negative impacts of soil fertility on seedling abundance and rainfall on resprout abundance were greater in more disturbed areas (Table 3; Fig. 5a and b, respectively).

Fig. 4.

Relationships between abundance of seed rain, soil seed bank, true seedling, true sapling and resprout with chronic disturbance, soil fertility, rainfall, and leaf area index in the Caatinga dry forest, Catimbau National Park, Northeast Brazil. Models were developed using log transformed data for some species.

(0.5MB).

Fig. 5.

Combined effect between chronic disturbance and soil fertility on (a) true seedling abundance and between chronic disturbance and rainfall on (b) resprouts abundances in the Caatinga dry forest, Catimbau National Park, Northeast Brazil.

(0.31MB).

Species richness exhibited less responsiveness, with seed rain positively affected by leaf area index, while the resprout assemblage responded negatively to chronic disturbance (Table 3; Fig. 6a and b, respectively). Resprout species richness also demonstrated interactive effects between chronic disturbances and rainfall (Table 3; Fig. 6c), with rainfall having a greater negative impact in more disturbed areas. The presence of allochthonous species in both seedlings and resprouts was positively related to rainfall, and neither chronic disturbance nor environmental factors were associated with theadult-to-offspring ratio (Table 3; Fig. 6d and e).

Fig. 6.

Relationships between richness of seed rain (a), resprout (b), woody plant regenerating assemblages (c), and leaf area index and chronic disturbance, and combined effect of chronic disturbance and rainfall on (d) resprout and (e) regenerating assemblages richness in the Caatinga dry forest, Catimbau National Park, Northeast Brazil.

(0.35MB).

Finally, the species composition across the regeneration mechanisms was influenced by the analyzed drivers as follows: sapling assemblages strongly responded to chronic disturbance (Fig. 7a; R2 = 0.39, P = 0.01), while resprouts responded to rainfall (Fig. 7b; R2 = 0.61, P < 0.01) and leaf area index (Fig. 7b; R2 = 0.35, P = 0.04).

Fig. 7.

Non-metric multidimensional scaling (NMDS) showing the association of environmental factors with species composition of (a) true sapling and (b) resprouts surveyed in the 18 0.1-ha m2 plots in the Caatinga dry forest, Catimbau National Park, Northeast Brazil. The acronyms stand for Aca = Acalypha brasiliensis, Bal = Balfourodendron molle, Bau = Bauhinia acuruana, Bou = Bouxera sp, Byr = Byrsonima gardneriana, Cha = Chamaecrista zygophylloides, CnB = Cnidoscoleus bahianus, CnP = Cnidoscoleus pubescens, com = Commiphora leptophloeos, Cos = Cospidaria agentia, CrH = Croton heliotropiifolius, CrN = Croton nepetifolius, Ery = Erythroxylum revolutum, Eug = Eugenia sp2, Mel = Melochia tomentosa, Neo = Neocalyptrocalyx longifolium, Oxa = Oxalis sp, pel = Peltogyne pauciflora, Pip = Piptadenia stipulacea, Pit = Pityrocarpa moniliformis, PoM = Poincianella microphylla, PoP = Poincianella pyramidalis, SeB = Senegalia bahiensis, SeP = Senegalia piauhiensis, Sen = Senna acuruensis, Sti = Stigmaphyllon paralias, Str = Strychnos rubiginesa, Tri = Trischidium molle, Var = Varonia leucoxyflora, Ziz = Ziziphus joazeiro.

(0.16MB).