Introduction

Over the last two decades there has been a resurgence in studies of positive interactions and facilitation in plant communities (Kareiva and Bertness 1997; Callaway 2007). There is now little doubt that both direct and indirect facilitation occur among species in biomes worldwide (Callaway 2007). Facilitation has been proposed to underlie species turnover to varying degrees during both primary and secondary succession (e.g., Odum 1969; Yarranton and Morrison 1974; Pickett et al. 1987). Nonetheless, Connell and Slatyer (1977) persuasively argued that inhibition (competition) was the pervasive mechanism driving secondary succession (see also Tilman 1985).

Yarranton and Morrison (1974) developed one of the most prominent models of primary succession (the nucleation model) where facilitation played the pivotal role on sand dunes in Ontario, Canada. Under the nucleation model, a few successful early woody recruits establish among a matrix of grasses and open sand, and these early recruits serve as “nuclei for the subsequent growth of patches of persistent species which eventually coalesce” (Yarranton and Morrison 1974). There is now ample empirical evidence that this model often, though not always, applies to early secondary succession in tropical habitats dominated by perennial grasses (Duncan and Chapman 1999; Toh et al. 1999; Holl et al. 2000; Slocum 2000, 2001; Carriere et al. 2002a, b; Holl 2002; Ferguson et al. 2003; Zahawi and Augspurger 2006; Schlawin and Zahawi 2008; reviewed by Reis et al. 2010). In these habitats, remnant trees present in the pasture in the first year of succession (or the earliest woody recruits) facilitate the establishment and survival of woody seedlings in their immediate proximity. Although the nucleation model holds much promise to explain patterns of succession (Peterson and Carson 2008; Reis et al. 2010; Albornoz et al. 2013), there have been few tests of this model of sufficient duration in tropical systems to fully evaluate its utility (Schlawin and Zahawi 2008). Because nucleation can occur around very early woody recruits (Vieira et al. 1994) in addition to actual remnant trees remaining from the primary vegetation, we henceforth use the more neutral term “isolated” rather than “remnant” to refer to trees that may provide recruitment foci for nucleation, thus concentrating the focus on the effects of a tree rather than on its time of origin.

It is important to point out that Yarranton and Morrison (1974) discuss nucleation exclusively in relation to facilitation by pioneer trees; thus application of the nucleation hypothesis has generally been restricted to evaluating whether isolated trees serve as recruitment foci (but see Peterson and Haines 2000; Slocum 2000; Reis et al. 2010). This has been the focus of the vast majority of 30 studies on tropical pasture succession that cite Yarranton and Morrison (our survey of the literature, see also Peterson and Carson 2008). Nonetheless, three recent studies (Reis et al. 2010; Carlucci et al. 2011; Albornoz et al. 2013) have focused on additional nuclei, such as rocks or logs; their attention to additional types of nuclei, as well as our results presented here, led us to develop the matrix discontinuity hypothesis. These studies notwithstanding, there is little doubt that the majority of past research has centered almost entirely on trees or shrubs, perhaps because this was the clear focus of Yarranton and Morrison (1974). This focus has led to a recent surge of publications dealing with “applied nucleation” using trees (e.g., Corbin and Holl 2012; Zahawi et al. 2013). In contrast, we introduce here the term “matrix discontinuity” to expand the nucleation concept to other types of microsites.

While the nucleating influences of isolated trees or shrubs are well established, there are other conditions or characteristics of early successional sites that may create highly favorable recruitment patches (sometimes termed microsites). These favorable patches may be created by legacies of land use (e.g., plow lines, see Myster 2004), important topographic features (e.g., steep slopes, treefall mounds, rocks, Carlucci et al. 2011), presence of coarse woody debris (Peterson and Haines 2000; Slocum 2000), or patches of less inimical herbaceous species (e.g., ferns, Slocum 2000). Each of the above patches or microsites creates a discontinuity or break in the dense background matrix of graminoids that otherwise might preclude woody seedling colonization. Such discontinuities are either elevated above the ground-layer vegetation and litter (e.g., logs) or are openings where the herbaceous vegetation is less dense or less inimical to woody species colonization (Peterson and Haines 2000; Slocum 2000; Hooper et al. 2005). In either case, competition for resources from non-woody vegetation, particularly grasses, and potentially inhibitory litter layers (e.g., Carson and Peterson 1990) are likely ameliorated, allowing germination and early survival of woody species. Consequently, an exclusive focus on isolated trees may minimize attention to other key habitat features, which may be equally or more important than isolated trees. Regardless, all of these habitat features create habitat discontinuities that function as recruitment nuclei. Oddly, these alternative habitat features have rarely been discussed (but see Peterson and Haines 2000; Reis et al. 2010; Carlucci et al. 2011; Albornoz et al. 2013 for rare exceptions). Thus, we propose an extension of the nucleation model, the matrix discontinuity hypothesis, which predicts that the rate and pattern of succession will be determined not just by isolated trees, but rather by distinct and observable habitat patches or areas that provide clear regeneration niches (sensu Grubb 1977). Such patches or microsites do not necessarily include any discontinuity, e.g., flat expanses of bare rock have little graminoid vegetation but are not favorable for woody seedling colonization. Thus we wish to focus on favorable microsites. Conceptually, our proposed hypothesis is also related to the idea of patch dynamics (Pickett and White 1985) in that early successional tropical pastures can be envisioned as made up of a mosaic of fine-scale microsites, some of which are particularly favorable for regeneration. Such a fine-scale perception of patches does not negate a coarser (e.g., landscape-scale) perception of patch dynamics; our proposed fine-scale view can fit comfortably within a hierarchy of different scales of patch dynamics.

Spatially, the impact of favorable nucleation sites may by overwhelmed by high rates of colonization near intact forest, or constrained at greater distances from intact forest, if dispersal limitation is severe. Regardless, the recruitment of woody species likely drops off sharply with distance from forest edges. This pattern has been found for numerous pastures and old-fields in temperate systems (e.g., Myster and Pickett 1992), as well as tropical habitats (Aide and Cavelier 1994; Ferguson et al. 2003; Myster 2003) but notably, in some cases does not occur (e.g., Duncan and Duncan 2000; Slocum 2001). Consequently, we evaluate the degree that recruitment declines with distance from the forest edge.

Overall, we test three hypotheses proposed to explain patterns of succession during the first decade of succession in five large pastures in Costa Rica. The hypotheses are not mutually exclusive and thus all may contribute to explaining the conversion of these pastures to closed canopy forest.

  1. 1.

    The classic nucleation hypothesis Recruitment during early succession is concentrated around isolated trees (which may be remnants from pre-agricultural vegetation, or very early colonists)

  2. 2.

    The matrix discontinuity hypothesis Recruitment during early succession is concentrated in distinct habitat discontinuities that are identifiable a priori within a given site. We suggest that these discontinuities serve to break up or mitigate the deleterious influence of a dense layer of remnant graminoids. These habitat discontinuities can occur at both fine scales (e.g., nurse logs) and at more coarse scales (e.g., steep slopes, rocky outcrops, etc.)

  3. 3.

    The distance from forest edge hypothesis Recruitment during early succession decrease with distances from forest edges and is leptokurtic

Methods

Study area

We conducted our research within 5 km of the Las Cruces Biological Station in southern Costa Rica (8°47′N, 82°57′W) in a landscape mosaic of forest fragments and cattle pastures that were created primarily in the 1960s and 1970s (Juarez 1994). Pre-settlement vegetation was tropical premontane wet forest (Hartshorn 1983). Elevation ranges from 800 to 1,500 m. The climate is slightly seasonal with average annual precipitation of 3,820 mm and average annual temperature of 20.7 °C; the wettest month is October (mean = 660 mm), while the driest is February (mean = 70 mm) (L. D. Gomez, unpublished data.). The soils are volcanically derived Lithic Dystropepts that are deficient in phosphorus (Jin et al. 2001) and widespread throughout the wet tropics (Wilding et al. 1983).

Selection of pastures

We chose five different cattle pastures (hereafter sites 1–5) each with an adjacent forest patch. Sites had been in pasture since at least 1981 and aerial photographs confirmed that each was an active pasture in 1992. The only land-use in these pastures had been cattle grazing (L.D. Gomez, pers. comm.) and all had relatively straight forest edges where fences had excluded cattle from the adjacent forest. These pastures varied in aspect and our regional surveys suggested that these pastures were representative of those in the region. Each was dominated by a dense layer of grasses, predominantly Cynodon nlemfuensis and Brachiaria spp., and all were grazed to a height of roughly 10–50 cm. After the cessation of grazing, these and other perennial graminoids commonly form dense swards in abandoned pastures in this region and throughout the Neotropics (Martínez-Garza and González-Montagut 1999; Carpenter et al. 2001; Zahawi and Holl 2009; Holl et al. 2011).

No small trees occurred in the pastures above the grass layer in 1996, although isolated larger trees (i.e., >10 m tall) were present in the pastures; some were remnants from the pre-settlement forest, and some had established naturally during the period the pastures were grazed or had been planted as living fenceposts (Peterson and Haines 2000). In June of 1996, an area of 0.26–0.62 ha within each pasture was fenced to exclude cattle (study sites were not always rectangular, but approximate x, y dimensions, respectively, were as follows: site 1—70 × 80 m; site 2—74 × 65 m; site 3—90 × 85 m; site 4—69 × 49 m; and site 5—80 × 43 m). Site 1 had two isolated trees, sites 2 and 3 had 12 and eight isolated trees, respectively, and sites 4 and 5 had four isolated trees each. These densities (mean 17.2 ha−1) are within the range typically reported for pastures in Central America (3.3–25 trees ha−1; Harvey and Haber 1998; Holl et al. 2000; Guevara et al. 2004). Characteristics of isolated trees are presented in Online Resource 1. Because there were no new woody recruits present above the grass layer, and the construction of fences ended grazing, we consider this at or near the very beginning of woody plant succession.

Documenting woody plant colonization

We censused the trees and shrubs in each pasture on eight occasions during the first 11 years of succession. Censuses were conducted in March 1998, December 2000, 2001, and 2002, May 2004, March 2005, and June 2006 and 2007, corresponding to 1.75, 4.5, 5.5, 6.5, 7.92, 8.75, 10.0, and 11.0 years, respectively, after cessation of grazing. During each census, we tagged, measured (diameter at 1.4 m, or DBH, and height), identified, recorded microsite or patch type, and mapped all trees and shrubs >1.5 m tall (we use the term “woody recruit” or just “recruit” hereafter to describe these individuals, although “colonists” would be equally valid). Substrate “microsite” was classified as soil, or rotting logs or stumps (palms and lianas were uncommon in the adjacent mature forest and had not colonized our sites). Rotting wood was generally well decomposed, spongy, and likely able to absorb and hold moisture for extended periods. All rotting logs encountered were >20 cm in diameter.

We mapped the spatial location of recruits and isolated trees in 1998 using a Topcon CTS-2 total station and confirmed these locations in 2003 by re-mapping all stems. Expert botanist Jose Gonzalez of the Instituto Nacional de Biodiversidad confirmed all field identifications in 2000 and 2004. Only 4 % of recruits remained unidentified to species but were typically identified to genus. We quantified coverage of rotting wood in 1998 by sampling at 200 points in each pasture (every 50 cm along two parallel, randomly placed 50 m transects). This sampling revealed log or stump cover of 4.5, 1.5, 4, 11.5, and 3 % at sites 1 through 5, respectively.

Statistical analysis

We present results of two types of test for nucleation; a strict test and a more general test. As a strict test for evidence of nucleation, we focused on colonization by recruits around isolated trees. We counted the total number of woody recruits (established after fencing was completed in 1996) within 4 m of each isolated tree. This distance approximated the lateral spread of the crown of most large trees in our sites. All of the isolated trees were >10 cm DBH and all were clearly present prior to the beginning of pasture succession. To characterize the mean and variability in recruit density in areas away from the isolated trees, we randomly selected 100 points throughout each pasture, subject to the constraints that points were (1) >4 m from forest edge; (2) >4 m from the surrounding fence; and (3) not overlapping any of the 4-m circles around isolated trees. We tallied recruit densities in 4-m-radius circles drawn around each of these random points. Mean and standard deviation of recruit density around the random points were calculated. Nucleation should cause recruit density around isolated trees to be significantly greater than recruit density around the 100 random points. We tested for such differences using non-parametric Mann–Whitney U tests (because of non-normality of the data and large difference in sample size between the treatment groups), for all sites pooled, as well as within each site.

We further conducted a more general test for nucleation that was not restricted to isolated trees. Once a number of recruits have established in an early successional site, nucleation should cause later recruits to be spatially associated with the earlier recruits. This more general nucleation is slightly different than that described above around isolated trees; it focuses on all earlier recruits to see if later recruits are likely to be spatially closer to the early recruits, or if alternatively the later recruits are spatially distributed independent of the early recruits. We tested for significant spatial association using Ripley’s K analyses within each of the five pasture sites (Dale 1999). The foundation of Ripley’s K analysis (in either the univariate [i.e., single types of points] or bivariate [two types of points] form) is that in a spatially random distribution of points, the number of neighboring points around an individual is K(t) = πt 2, where t is the distance within which all neighboring points are counted (in our case, points correspond to recruits). For spatially clumped univariate distributions, the number of neighboring points at small distances will be greater than expected; for spatially uniform univariate distributions, there will be fewer neighboring points than expected. The bivariate version of Ripley’s, called K 12(t), tests for spatial association between two sets of points (e.g., two species or two cohorts) versus spatial repulsion. The statistic is calculated for neighborhoods at each of numerous distances (e.g., 1-m intervals from 1 to 15 m) around every plant (more details in Peterson and Squires 1995). We calculated the bivariate Ripley’s K 12(t) to test for spatial association between two categories of recruits: all stems present in 2005, and a second group comprised of new recruits found in 2006 or 2007 (i.e., the final two surveys). We used the SpPack (Perry 2004) add-into Microsoft Excel for the calculations, and performed 99 randomizations to generate 99 % confidence envelopes; observed values of the K 12(t) statistic outside of the confidence envelopes are significantly non-random and indicate that stems from the two groups are spatially associated.

Our observations over many years suggested that steeply sloping areas within the pastures were foci of colonization. To rigorously evaluate this, we divided each pasture into 4 m × 4 m cells, and tallied density of recruits in each cell. Steepness of each cell was calculated from variation in elevation of each of the four corners of the cell, which were in turn calculated from spatial interpolation of local elevations of hundreds of points within each pasture. We then chose a subset of cells for further analyses, retaining only cells from odd-numbered rows and odd-numbered columns of the grid cell map within each site (a single 4 m × 4 m cell from any 8 m × 8 m group of four cells); this approach ensured that no cells used in analyses were immediate neighbors of any other cells and that a minimum of 4 m separated any cells analyzed, thus reducing spatial autocorrelation. We then tested how recruit density varied with slope, with t tests of the 20 steepest and 20 least sloping cells.

We had previously demonstrated that during early succession (<2 years) remnant decomposing logs were recruitment foci (Peterson and Haines 2000), but it remained unclear whether the higher-density establishment on logs would persist over the next decade or whether it was a transient response that would dissipate as the logs disintegrated. For the woody plant recruits in all sites, we calculated 1998–2007 survival and mortality. We tested for differences in survival between individuals on logs and on soil with G-tests of table homogeneity, for all species pooled; this was done within each site, as well as for all sites pooled.

Finally, we tested whether the distribution of all seedlings and saplings (including post-1998 recruits) was still significantly non-random across microsites (i.e., concentrated on rotting wood microsites) after 11 years of succession. To do this, we performed goodness-of-fit G-tests on microsite distributions of all woody stems present in 2007, using all preceding cohorts combined. The null hypothesis was that seedlings and saplings would be distributed randomly relative to soil versus rotting wood microsites; specifically the number of seedlings expected on rotting wood microsites would be proportional to the rotting wood coverage of the soil surface.

If recruitment decreases with distance from forest, then the distribution of distances from forest for all individuals will be leptokurtic versus if individuals established randomly within the pastures. Random positions were generated between zero and the maximum distance from forest for each study site. These random distributions of distances were then compared to the actual distances from forest of the colonizing plants. We used Kolmogorov–Smirnov two-sample tests, with continuous distances binned into 2-m intervals.

Results

With regard to our key variables of interest (e.g., recruits beneath isolated trees), we did not detect any significant site (i.e., pasture) effects whereby one or a subset of pastures had different patterns of woody species colonization than another. Thus, we combined the data across all pastures to test our hypotheses.

Were isolated trees foci for colonization?

The 30 isolated trees were composed of 16 different species within 11 different families (Online Resource 1). Number of recruits around isolated trees could not be normalized by transformation, therefore we tested for differences between isolated trees and random points using non-parametric methods. Isolated trees, on average, were not foci for establishment during the first 11 years of succession (pooled across sites, median number of recruits around isolated trees = 6.0 within a 4-m radius, versus random locations = 6.0 recruits within a 4-m radius) and five of the 30 isolated trees had no recruits near them (Fig. 1). Only three of 30 isolated trees had neighboring recruit densities more than one standard deviation above the mean of random pasture locations (Fig. 1). The number of recruits beneath isolated trees increased with tree DBH (Fig. 2). Trees that bore fleshy fruits (half of all isolated trees and ten different species, Appendix 1) did not have significantly higher density of recruits versus non-fleshy fruited individuals, although the mean was substantially higher (mean fleshy fruited = 10.5 ± 11.1, vs. non-fleshy fruited individuals, mean = 5.4 ± 3.5, Mann–Whitney U test, p = 0.22, n = 15). The density of recruits beneath introduced species (13 individuals) did not differ from that of native species (Mann–Whitney U test, p = 0.56).

Fig. 1
figure 1

Density of woody recruits within a 4-m radius of 100 randomly positioned points, and isolated trees. Data (100 points per site) were pooled across five early successional pasture sites in southern Costa Rica. For random points, the bar represents median recruit density; error bar indicates 75th percentile. For isolated trees, recruit density is shown as points

Full size image
Fig. 2
figure 2

The relationship between tree size (DBH = diameter at 1.3 m) and number of woody recruits within 4 m of isolated trees

Full size image

Were steep slopes and logs foci for colonization?

Steeply sloped areas had significantly more recruits than flatter areas (Fig. 3). Factorial ANOVA, crossing sites and slope steepness (20 steepest vs. 20 least steep cells), indicated significant site (p < 0.001) and slope (p < 0.001) effects, with greater colonist density in steeply sloped cells, but no interaction (p = 0.19). This trend remains if sites are pooled and the 20 cells with the steepest slopes are compared to the 20 cells with flattest slopes across sites: the cells with the steepest slopes had significantly greater recruit density (2.97 ± 3.75) than the cells with the flattest slopes (1.77 ± 3.72) (t test, n = 40, t = 5.44, p < 0.001). Even when the effect of distance to forest (see below) was considered, recruit density still increased significantly with slope (multiple linear regression, n = 1159, F = 52.02, p < 0.001).

Fig. 3
figure 3

Density of woody recruits in the 20 steepest and 20 flattest 4 m × 4 m grid cells in each of five early successional pasture sites in southern Costa Rica. Slope was calculated from difference in elevation among the four corners of a grid cell. Bars are means (±SD). Steepest and flattest grid cells are those with the greatest and least slopes, respectively. ANOVA results: site F = 13.41, d.f. = 4, p < 0.001; slope F = 10.95, d.f. = 1, p < 0.001; site × slope F = 1.51, d.f. = 4, p > 0.2

Full size image

More than 40 % of all woody species occurred on decomposing logs and stumps in 1998, even though these sites were <5 % of pasture area (Fig. 4; goodness-of-fit G test, G = 358.0, 1 d.f., p < 0.001). Colonization after 1998 was distributed between rotting wood and soil roughly proportional to the abundance of these microsites, so after 11 years of succession, about 15 % of all recruits occurred on logs and stumps (Fig. 4; goodness-of-fit G-tests, G = 116.0, 1 d.f., p < 0.001). This suggests, and our visual observations confirmed, that these sites appeared to quickly fill early in succession with a fairly dense concentration of new recruits and subsequently thinned thereafter. While mortality for the 1998 cohort through 2007 did not differ between log microsites versus soil (mortality on logs = 57.7 % vs. soil = 57.1 %; G-test, p > 0.10), tree size (DBH) was slightly larger on logs versus soil (mean on logs = 11.1 versus soil 10.0, Mann–Whitney U test, p = 0.003). These results suggest that the main function of decaying logs is to enhance colonization and establishment early in succession versus subsequent growth and survivorship thereafter.

Fig. 4
figure 4

Percentages of all woody recruits present in 1998 and 2007, and post-1998 recruits, observed growing on soil versus decaying wood microsites in five early successional pasture sites in southern Costa Rica. The horizontal line indicates area coverage of rotting wood and therefore the expected percentage of recruits on rotting wood

Full size image

Did increasing distance from forest reduce colonization?

Colonization was more abundant near forest and leptokurtic (Fig. 5; comparing observed recruit distances from forest, and distances of random points, Kolmogorov–Smirnov two-sample test D observed = 0.114, D critical for 0.001 = 0.058).

Fig. 5
figure 5

a Cumulative frequency distribution of distances from forest for all woody recruits, and for a similar number of randomly positioned points in five early successional pasture sites in southern Costa Rica. Concentration of recruits nearer the forest is shown by bars exceeding the random expectation. b Relationship between number of recruits within 4 m and distance from the forest edge

Full size image

Were later recruits spatially associated with earlier recruits?

Ripley’s K 12 analysis indicated that at fine scales (i.e., 0–5 m), 2006/2007 recruits were significantly (99 % confidence envelopes) associated with stems present in 2005 in every one of our study sites (Online Resource 1). The scale of the significant association differed among sites: 0–5 m at site 1; 0–20 m (the maximum distance tested) at site 2; 0–12 m at site 3; 0–13 m at site 4; and 0–10 m at site 5.

Discussion

To our knowledge, this is the longest direct, continuous, multi-site study of pasture succession in the Neotropics. We documented several lines of evidence among five pastures and through 11 years of succession that identifiable and distinct microsites within pastures will be foci for woody species colonization, thus extending the well-known nucleating effect of isolated trees. To describe this broader range of favorable microsites, we propose the “matrix discontinuity hypothesis”, which is a logical extension of the classical nucleation concept of Yarranton and Morrison (1974). We suggest that the microsites or patches that promoted recruitment likely ameliorated competition with a dense graminoid layer (see below), though this conclusion awaits rigorous experimental confirmation. However, there is strong demonstration that the dense graminoid vegetation is inhibitory to woody seedling establishment in these pastures. Peterson and Haines (2000) reported the result of a short-term experimental manipulation in which all graminoid vegetation and litter was removed from 2 m × 2 m patches within three of the five pastures in August 1997. The removed vegetation was dried and weighed to calculate aboveground biomass + litter, and light levels were measured in cleared plots and controls. Aboveground graminoid biomass + litter was 1,278 ± 479 g m−2, 1,128 ± 375 g m−2, and 1,021 ± 391 g m−2, in the three sites. Light levels (photosynthetic photon flux density) just above the ground surface were only 1.75 % of ambient (above-vegetation light levels) in the dense vegetation of the controls (Peterson and Haines 2000). Three months later, woody seedling density was fivefold to tenfold greater in the cleared areas than in adjacent controls; percent cover showed a similar pattern, and both of these variables differed significantly between treatments in all three pastures. Species richness was typically twice as high in the cleared plots as in the controls, and this difference was significant in two of the three pastures. Thus, the strong inhibition by grasses in these sites is well established. We cannot say conclusively that such inhibition results from competition for resources without further experimentation; and it should be noted that allelopathic effects could contribute to this phenomenon, since Brachiaria species in Brazilian agricultural and agroforestry contexts have shown an allelopathic effect (e.g., Lacerda et al. 2013).

Key recruitment patches identified to date include isolated trees, decaying logs, patches of fern, steep slopes, animal disturbances, rocks, and slash piles (Uhl et al. 1982; Metcalfe et al. 1998; Peterson and Haines 2000; Slocum 2000; Reis et al. 2010; Carlucci et al. 2011). We also demonstrate that over the first decade, the main function of decaying log microsites was to promote early woody species colonization and establishment rather than enhanced growth or survivorship. Overall, we found modest evidence that isolated trees were foci for establishment (the classical nucleation hypothesis) even though these trees were present at typical densities and represented a variety of species. We are not arguing that isolated trees are not important, rather we found that at our five study sites only trees above a threshold size, and those bearing fleshy fruits, fostered nucleation. Our results suggest that the role of tree size and dispersal mode be more fully evaluated in future studies (see below).

The results of Ripley’s spatial association analysis provide a slightly different test of the nucleation concept than our detailed examination of the colonization around isolated trees. The spatial association analysis asks, for all woody plants present in 2005, whether subsequent colonization in 2006 and 2007 was closer spatially than expected if the later recruits were indifferent to the locations of the pre-2005 recruits. The recruitment of new woody individuals (new in 2006 and 2007) was consistently and strongly concentrated near earlier recruits, confirming that nucleation was happening, regardless of whether the initial recruits had colonized logs, steep slopes, or near isolated trees. A few of the common taxa colonizing our sites (e.g., the genera Vismia, Piper, and Palicourea) are likely to sprout extensively, thereby vegetatively establishing clones of interconnected stems, which could influence the spatial associations (note that although many shrubby species of Miconia do sprout extensively, the species that is by far the most abundant in our study sites, Miconia theizans, showed no indication that it was establishing vegetatively). Moreover, the dense clusters of stems that are typical of clonal reproduction are more likely to affect univariate K(t) analyses (testing for clumping of single species), and not likely to greatly influence the bivariate K 12(t) analyses (testing for associations between stems of different cohorts). Nonetheless, we acknowledge that a (probably small) proportion of the clustering documented by the Ripley’s K 12(t) analysis could result from sprouting.

Our work also confirmed previous results that recruitment decreased with distance from the forest edge, verifying that dispersal limitation also likely constrains pasture succession. In total, our results suggest that the matrix discontinuity hypothesis may be broadly applicable because all of our pastures showed similar responses across many years, and these pastures are generally representative of land uses and initial vegetation region-wide.

Isolated trees were not consistent recruitment foci

We found that many of the isolated trees in our sites failed to increase woody species recruitment in their immediate neighborhood as predicted by the classic nucleation hypothesis, although such conclusions are based on a modest sample size. These trees were a mixture of true residual individuals from the pre-agriculture vegetation (e.g., Beilschmedia alliophylla, Lauraceae), species that colonize forest fragment edges and gaps (e.g., Miconia tonduzii, Melastomataceae), and ruderal species (e.g., Heliocarpus appendiculata, Tiliaceae) that apparently established during the period our pastures were being grazed (Online Resource 1). Spathodea campanulata is an African exotic that is widely planted as living fenceposts. In other studies, isolated trees typically enhance seed dispersal (Slocum and Horvitz 2000; Holl 2002) and other studies have unequivocally demonstrated enhanced recruitment around isolated trees (Duncan and Chapman 1999; Toh et al. 1999; Holl et al. 2000; Slocum 2000, 2001; Carriere et al. 2002a, b; Holl 2002; Ferguson et al. 2003; Zahawi and Augspurger 2006; Schlawin and Zahawi 2008). In agreement with recent insights from Reid and Holl (2013), we suggest these remnant or isolated trees will most likely act as recruitment foci when they increase local seed dispersal and reduce the negative impact of the herbaceous layer beneath. Both of these effects will be enhanced beneath large remnant trees where competition from these large adults may suppress the herbaceous layer (Slocum 2001) or lead to greater seed dispersal (Duncan and Chapman 1999). In fact, our findings combined with those of others, suggest that large or tall remnant trees (>40–50 cm DBH and taller than 20 m) are the most likely to be effective nucleation foci (Guevara et al. 1986; Carriere et al. 2002a, b; Ferguson et al. 2003; Guevara et al. 2004; Schlawin and Zahawi 2008, but see Slocum 2001). Only six of the 30 isolated trees in our study were >40 cm DBH and the three that had the most recruits were all >60 cm DBH.

Half of our isolated trees produced fleshy fruits (Online Resource 1), which should attract avian foragers (Slocum and Horvitz 2000; Slocum 2001). Such potential recruitment foci should result in especially strong nucleation, given that the most common colonists in our five sites are all animal-dispersed, fleshy-fruited taxa: Miconia tonduzii and M. theizens (Melastomataceae), Cecropia spp. (Cecropiaceae), Palicourea padifolia (Rubiaceae), and Viburnum costaricanum (Caprifoliaceae). Moreover, findings from Vieira et al. (1994) suggest that such enhanced attractiveness should occur regardless of size, since Vieira et al.’s focal species was the shrub Cordia multispicata. In our study, recruitment beneath isolated trees that produced fleshy fruits and those that did not was not significantly different, although the average beneath fleshy-fruited trees was almost twice that beneath non-fleshy-fruited trees. The lack of statistical significance is likely an artifact of limited sample size; we therefore interpret these findings as consistent with the idea that isolated trees with fleshy fruits are likely to serve as the best nuclei for subsequent recruitment, but caution that the fleshy-fruit versus non-fleshy-fruit contrast is not well resolved. Carriere et al. (2002a, b) and Toh et al. (1999) both found that their fleshy-fruited isolated trees had no more recruitment than non-fleshy-fruited trees; Slocum (2001) also reported that dry-fruited Cordia alliodora did not differ in nucleation from fleshy-fruited Cecropia isolated trees. We suggest that future work should attempt to characterize a variety of interindividual and interspecific differences in traits of isolated trees (e.g., size, shade cast, rooting depth, dispersal mode) that promote colonization beneath them, and to distinguish the influence of those traits on the various processes of dispersal, germination, growth, and survival of seedlings. Indeed, Slocum (2001) found that while Ficus isolated trees had denser and more diverse seedling recruits beneath them, these recruits grew quite slowly due to the deep shade cast by the Ficus crown. Such contrasting effects do not seem apparent in our data, which showed an especially high density of large and seemingly fast-growing recruits beneath the crown of a very large Ficus, in agreement with the patterns reported by Guevara et al. (2004). This particular isolated tree in fact supports our suggestion regarding limitations to classical nucleation simply because it has the best combination of characteristics: it is large, fleshy-fruited, and very near (~12 m) intact forest, and therefore is clearly promoting substantial nucleation.

Steep slopes, decaying logs, and the matrix discontinuity hypothesis

To date, few studies have shown that steep slopes enhance recruitment (but see Metcalfe et al. 1998). Such a pattern could be due to any of several causes. Although speculative, we observed that steep slopes early in succession had more small gaps in the graminoid vegetation, possibly due to repeated cattle foot traffic in these locations prior to cattle exclusion (Peterson, pers. obs.), or as a result of greater soil erosion that creates small openings in the grassy cover. Alternatively, steeply sloping areas might have lower soil compaction due to lesser cattle traffic compared to more level areas of a pasture. However, we find it difficult to envision that greater colonization on steep slopes results from increased lodging of small seeds, as proposed by Metcalfe et al. (1998). That two of the above three alternatives center around the post-abandonment effects of pre-abandonment cattle activity may suggest that cattle leave a disturbance legacy that aids colonization thereafter. Regardless, the mechanism that promotes colonization on steep slopes will require further study, and some alternatives could be easily distinguished if we had soil bulk density measurements. Finally, with regard to colonization on logs, our results demonstrate that these sites continue to support high densities of colonists even a decade into succession as a consequence of colonization events in the first few years.

Overall, our results lead us to suggest the matrix discontinuity hypothesis, which calls attention to any biotic or abiotic process that creates patches that mitigate the negative influence of the graminoid layer, because this layer is typically inimical to woody species colonization (Holl et al. 2000; Slocum 2000, 2001; Zanne and Chapman 2001; Ferguson et al. 2003; Hooper et al. 2005; Peterson and Carson 2008). We argue that decaying logs and steep slopes serve to reduce the thickness of this layer (both litter and live mass) or allow seedlings to establish above it or both. Logs and stumps may also provide elevated perches for birds and rodents thereby increasing seed dispersal in the same way isolated trees can augment seed dispersal (e.g., Slocum and Horvitz 2000; Carriere et al. 2002a, b; Holl 2002). On the other hand, it is not obvious how steep slopes would enhance dispersal, and we therefore tentatively suggest that the greater recruitment on slopes may result from features that enhance germination and establishment of the seeds that do arrive there. In some cases these favorable microsites may serve to mitigate the impact of seed and seedling predators (Peterson and Carson 2008).

The matrix discontinuity hypothesis is a tightly focused idea that can be seen as nested within broader concepts of patch dynamics and the regeneration niche. Matrix discontinuity may broadly explain patterns of recruitment among very different systems. Dense non-woody vegetation that is inimical to tree recruitment is common worldwide (Royo and Carson 2006). For example elevated structures such as logs or treefall root mounds reduce competition with herb-layer vegetation and enhance tree seedling establishment in temperate forests in both Japan and North America (Nakashizuka 1989; Harmon and Franklin 1989) as well as in temperate old-fields (Goldberg and Gross 1988). Thus, the matrix discontinuity hypothesis may link nucleation-type phenomena in tropical pastures to a variety of temperate systems.

Woody species colonization and distance from the forest edge

We confirm the results of other studies (Aide and Cavelier 1994; Mesquita et al. 2001; Ferguson et al. 2003; Myster 2003; Hooper et al. 2004; Holl 2007, but see Duncan and Duncan 2000) that colonization decreases with distance from the forest edge, probably due to dramatic declines in propagule supply within only a few meters of intact forest (Nepstad et al. 1996; Holl and Lulow 1997; Cubiña and Aide 2001; Dosch et al. 2007). However, forest edges may also ameliorate harsh conditions of high temperatures or low humidity, which may also increase woody species establishment near forest edges. Overall, woody plant establishment requires that a propagule reaches a patch and once there establishes within that patch (Reid and Holl 2013); particularly favorable microsites (i.e., safe sites) maximize probabilities of both arrival and establishment and may circumvent the distance limitation seen overall, i.e., the few large, fleshy-fruited isolated trees (such as the Ficus in site 4), or on logs (colonized abundantly regardless of distance from forest). We suggest that modeling approaches that predict colonization based primarily on dispersal traits of species (e.g., Nuttle and Haefner 2005) be linked to patch-based studies that quantify the abundance of favorable patches or microsites.

Conservation and restoration implications and directions for future research

Our findings suggest that one simple and economical way to speed the conversion of pastures to closed canopy forests would be to increase the abundance of coarse woody debris within pastures and thereby increase the supply of these key microsites (see also Slocum 2000). This would also provide a rigorous experimental test of one prediction of the matrix discontinuity hypothesis. This hypothesis also directs attention to a variety of nucleation sites in addition to remnant trees. Attention should be given to any process that creates patches with a less dense layer of graminoids, patches elevated above the herbaceous canopy, or patches that mitigate the impact of enemies. All of these are likely to promote more rapid woody species colonization.