Shrubs in arid and semi-arid ecosystems are often associated with three distinct patch types: the shrub core, the shrub periphery and surrounding open patches. The distribution of herbaceous seeds in such a patchy system exhibits a well-documented spatial heterogeneity. However, the mechanisms that generate this heterogeneity are poorly understood, not least because of the difficulty of separating possible effects of the shrub on seed production (via the shrub's modification of resource distribution), seed dispersal and post-dispersal processes.
We used a well-studied system dominated by a common east Mediterranean shrub (Sarcopoterium spinosum) to directly test the effect of shrubs on herbaceous community seed dispersal. We clipped all potential herbaceous seed sources from plots with or without a shrub at their centre, designated ‘shrub’ or ‘open’, respectively. Seed rain was then sampled over a 7-month period, along four directions within a fine-scale radial sampling pattern from the shrub core to its periphery and in corresponding positions in the open plots. Seed predation was monitored likewise.
The overall abundance and species richness of herbaceous seeds were similar at all distances from the core in the open plots and at the shrub periphery, but lower under the shrub canopy. However, the observed patterns were clearly directional, with the highest seed abundance and species richness found on the upslope periphery of the shrub patch.
These patterns were observed after the elimination of all within-patch herbaceous seed sources, which suggests that the movement of seeds was mainly driven by gravity-related mechanisms whose effect was modified by the shrubs.
Synthesis. A shrub can function simultaneously as both a seed trap and a barrier to herbaceous seed flow, with the exact balance determined by location within the patch. Furthermore, whether the effect of the shrub on herbaceous seeds is regarded as facilitatory or competitive is scale-dependent. A mechanistic dissociation of seed dispersal from other processes modulated by the shrub in shaping the herbaceous community, as done in this study, is important for understanding the resilience of semi-arid and arid ecosystems to environmental changes, especially to the increasingly observed drought-induced mortality of shrubs associated with climate change.
Facilitation has been recognized as a major ecological and evolutionary force in plant communities (Callaway & Walker 1997; Flores & Jurado 2003; Maestre, Valladares & Reynolds 2005; Brooker et al. 2008; Kikvidze & Callaway 2009). Facilitation of herbaceous vegetation by woody plants is common in arid and semi-arid ecosystems (West 1990; Aguiar & Sala 1999; Tongway & Valentin 2001), where the landscape often consists of patches of woody vegetation (mainly shrubs) embedded in a bare soil matrix (the ‘open’ patch type) that is sparsely populated with herbaceous vegetation. Numerous studies have shown that the establishment, abundance and fecundity of herbaceous plants are higher in the shrub patch than in the surrounding matrix (Tielborger & Kadmon 1995; Guo 1998; Maestre & Cortina 2005; Badano & Cavieres 2006). Such facilitation is mainly attributed to the engineering effect of the shrub, which improves the physical, chemical and hydrological conditions for herbaceous vegetation growing in shrub patches, compared with conditions in open patches (Garcia-Moya & McKell 1970; Jones, Lawton & Shachak 1994; Pugnaire et al. 1996; Titus, Nowak & Smith 2002; Gilad et al. 2004; Pugnaire, Armas & Valladares 2004; Badano et al. 2006; Segoli, Ungar & Shachak 2008). In parallel to the clear differences in environmental conditions and resource levels, ‘shrub’ and ‘open’ patches also differed in the abundance, species richness and community composition of seeds, both in the seed rain (Marone, Rossi & Horno 1998; Giladi, Segoli & Ungar 2007) and in the seed bank (Aguiar & Sala 1997; Guo, Rundel & Goodall 1998; López-Pintor, Espigares & Rey Benayas 2003). The higher abundance and species richness of seeds in the shrub patch compared with the open patch could be the result of higher seed production, better seed retention and less granivory (Aguiar & Sala 1997; Moro et al. 1997b; Marone, Rossi & Horno 1998; Russell & Schupp 1998; Chambers 2000; Flores & Jurado 2003). In addition, the shrub patch may affect seed dispersal patterns and even function as a trap, which accumulates seeds from its surroundings (Russell & Schupp 1998; Bullock & Moy 2004).
Our foregoing description of the main landscape elements of shrub-facilitated systems is consistent with a dichotomy of shrub and open patches (Charley & West 1975; Aguiar & Sala 1999). However, more detailed investigations have refined this description, pointing to a fine-scale spatial pattern of environmental conditions and resource distribution within a shrub patch, which suggests a more complex landscape structure than a biphasic one (Moro et al. 1997b; Arnon et al. 2007; Caballero et al. 2008; Segoli, Ungar & Shachak 2012; Segoli et al. 2012). Like the abiotic environment, the seed rain and seed bank also exhibit fine-scale spatial patterns within the shrub patch: highest densities have been found at the shrub core (Guo, Rundel & Goodall 1998; Bullock & Moy 2004; Caballero et al. 2008) or at intermediate distances between the core and the periphery (Moro et al. 1997b). Although many studies have documented the spatial and temporal patterns of seed deposition and seed bank within and around shrubs, the mechanisms by which shrubs shape these patterns are poorly understood.
The effects of shrubs on the spatial distribution of herbaceous seeds may act via three interacting mechanisms: first, acting as ecosystem engineers, shrubs generate heterogeneity in resource distribution that might develop into a spatially heterogeneous pattern of seed production; secondly, the shrub might affect the spatial patterns of primary and secondary dispersal of seeds, either directly, for example, by blocking seed movement, or indirectly, for example, by modulating wind velocity and direction; thirdly, the shrub might affect seed survival, for example, by enhancing or reducing the risk of granivory. We focused our investigation on the contribution of the second of these mechanisms – seed dispersal – to fine-scale patterns in seed distribution. We particularly considered two seed dispersal mechanisms: wind dispersal and ground rolling (by gravity or run-off) and hypothesized that shrubs shape the distribution of herbaceous seeds by modifying them. Shrubs may alter wind velocity and wind direction through and around their canopies, thereby affecting dispersal by wind (Bullock & Moy 2004; Li 2008; Li et al. 2009; Wang et al. 2010). The accumulation of leaf litter and soil under the shrubs increases surface roughness, which in turn impedes run-off flow and traps seeds (Boeken & Shachak 1994; Chambers 2000; Boeken & Orenstein 2001). Therefore, we predicted that the distribution of seeds associated with the shrub patch will be different from that in an open patch and that the difference between shrub and open patterns will vary between wind-dispersed and ground-rolling seeds.
We tested the aforementioned hypothesis and predictions in a semi-arid shrubland in southern Israel where the dominant shrub is the dwarf cushion shrub Sarcopoterium spinosum (prickly burnet). In this study system, the abundance and species richness of herbaceous seeds in the seed rain at the shrub core are significantly lower than those in the seed rains of adjacent open patches (Giladi, Segoli & Ungar 2007), whereas the seed bank density in the shrub periphery is similar to that in the open patches (Holzapfel et al. 2006). These patterns at the two seed dispersal stages mirrored those in the standing herbaceous vegetation and could indicate very limited seed dispersal in conjunction with localized cycling between the seed and the herbaceous phases. However, these patterns could also result from the modification of seed flow by the dense S. spinosum canopy.
To test whether these patterns in seed distribution result from the modification of dispersal mechanisms by shrubs, we (i) clipped all potential herbaceous seed sources from patches of S. spinosum and from adjacent open patches, thus equalizing seed production between patch types; (ii) monitored seed predation in both patch types; (iii) conducted fine-scale sampling of the seed rain in both patch types and (iv) analysed separately the arrival patterns of wind-dispersed and ground-rolling species. The elimination of all seed production from shrub and open patches is predicted to result in inward diminution in the density and species richness of seeds in both patch types. If the shrubs impeded the movement of seeds, we predicted that the inward declines in the density and species richness of seeds in the shrub patch, between the shrub periphery and its core, will be steeper than the decline in the open patch. However, while wind-dispersed seeds are expected to accumulate mainly on the windward side of the shrub, ground-rolling seeds are expected to accumulate on the upslope side of the shrub. If the shrub does not modify seed dispersal, then the elimination of local seed sources is predicted to result in spatial patterns of seed rain that are indistinguishable between shrub and open patches.
Materials and methods
The study area is located in the Goral Hills in the northern Negev of Israel (31°21′52″ N, 34°49′46″ E), at the Lehavim Bedouin Demonstration Farm, which is part of the Israeli Long-Term Ecological Research (LTER) network. The site is located in the transition zone between the Mediterranean and the semi-arid climate zones, with an average annual precipitation of 305 mm and a range of 78–504 mm over the period 1953–95. The precipitation falls mostly from December to March. Average daily temperatures range from 10 °C in January to 25 °C in July (Baram 1996).
The landscape of the area slopes west–east at heights of 350–500 m.a.s.l. The soil is brown lithosol with aeolian loess and is based on Eocene limestone and chalk. The vegetation is comprised of various shrub species and a diverse community of herbaceous vegetation, mostly annuals, which persist for 3–5 months after the first rains, depending on the amount and temporal distribution of the precipitation. The dominant shrub species is S. spinosum (L.) Spach; it forms monospecific patches that may contain one or several individuals (Seligman & Henkin 2002; Reisman-Berman, Kadmon & Shachak 2006; Segoli, Ungar & Shachak 2008). Sarcopoterium spinosum (prickly burnet) is prevalent in many semi-arid shrublands of the eastern Mediterranean region (Litav & Orshan 1971). Compared with open patches, S. spinosum patches have higher soil moisture contents (Pariente 2002), higher soil nutrient concentrations, lower light intensity, buffered temperature extremes and lower wind velocities (Osem, Perevolotsky & Kigel 2002; Pariente 2002; Segoli, Ungar & Shachak 2008, 2012; Ungar et al. 2008; Segoli et al. 2012).
Total seasonal rainfall in the 2006/07 hydrological year (starting 1 October 2006), as measured at the nearby Lahav meteorological station, was 285 mm. The last major storm of the 2006/07 growing season was in mid-March 2007; it yielded 36 mm of rainfall over 4 days, bringing the seasonal total to 274 mm. Subsequent remaining rainfall of that season was scattered in small amounts over the following 2 months, and the first effective (>5 mm) rains of 2007/08 fell in mid-November 2007.
Ten experimental blocks were demarcated on a west-facing slope during the first week of April 2007, near the end of the 2006/07 growing season. Each block consisted of two (2 × 2 m) plots, one centred on a S. spinosum shrub (designated as ‘shrub’) and the other set in the nearest available intershrub open patch (designated as ‘open’) – usually at a distance of 4–5 m from the shrub patch – that was sufficiently large to include a plot and a 1-m-wide buffer zone. To prevent any local input of herbaceous seeds, all above-ground herbaceous vegetation, including that under the shrub canopy, was removed from all the plots, in the second week of April 2007.
Sampling of seed rain was initiated in mid-April 2007, which matches the onset of seed dispersal by most annual species at the site, and was terminated in late-December 2007, approximately 6 weeks after the first significant rain event of the following growing season. Nine seed traps were placed within each plot. In the shrub plots, one trap (designated ‘core’) was placed at the centre of the shrub patch, and two traps were placed along each of the four cardinal directions radiating from the shrub core, one centred under the canopy dripline at the shrub periphery (designated ‘periphery’) and the other midway between the shrub core and its periphery (designated ‘intermediate’, see Fig. 1 for a description of the experimental set-up). Seed traps in the open plots were placed in the same pattern. Each seed trap was a 9-cm-diameter Petri dish filled with fine sand and embedded in the soil with its rim at ground level; this method had previously been used successfully in studies conducted at this site and in similar landscapes (Rew, Froud-Williams & Boatman 1996; Bai & Romo 1997; Giladi, Segoli & Ungar 2007). Seeds were collected every 2 weeks from 7 May through 24 December 2007; this period spans the end of one growing season and the beginning of the next rainy season, when seed germination is initiated and thus terminates the possible dispersal period of most species. At each collection date, the sand from each trap was sifted through a 560-μm sieve, the seeds and dispersal units were bagged, and the sifted sand was returned immediately to the Petri dish. The efficiency of extracting seeds of various species from the seed traps was tested against a reference seed collection and, except for a very few species with extremely small seeds, was found to be very high (Giladi, Segoli & Ungar 2007).
The material collected from the seed traps was brought to a laboratory, the dispersal units were opened to count the seeds, which were identified under a microscope, and the number of seeds of each species was recorded. The data obtained from each seed trap were pooled across all sampling dates, and the number of species was calculated according to identified seeds only and excluding those of the shrub. Seed deposition rates (seeds m−2 day−1) were calculated separately for herbaceous species (including unidentified seeds) and S. spinosum. We used the morphology of the seed (or dispersal unit) to assign each of the identified species to one of three dispersal categories: wind dispersed (e.g. seeds with pappus), ground rolling (rounded or oval dispersal unit) or undetermined.
In the field, trapped seeds might have been exposed to granivory. If the intensity of granivory was to vary among patch types and trap positions, this effect would need to be taken into account in interpreting the observed patterns; therefore, to test for this potential effect, we spiked each trap with 10 millet seeds, which we counted and replaced with fresh ones at each sampling date.
We used linear mixed effect models (LMEs) to analyse the effects of patch type (shrub vs. open), distance from patch core (core, intermediate and periphery) and direction (W, S, E and N) on seed deposition rate (Pinheiro et al. 2012; nlme: linear and nonlinear mixed effects models. R package version 3.1-100). Since the core trap had no defined direction, a fully crossed design between direction and distance was not applicable; therefore, we used one set of statistical models, which included data from all seed traps, to test the effects of patch type and distance, and another set, in which the core trap was excluded from the analysis, for testing the combined effect of patch type and direction. Blocks were included as a random factor in all the analyses. Preliminary analysis of seed deposition rates indicated that assumptions of normality and homogeneity of variance were not met; therefore, the seed deposition data were log transformed prior to statistical analysis. Separate analyses of seed deposition rate were conducted for the whole seed community, wind-dispersed seeds and ground-rolling seeds. Because of the count nature of the species richness data and the inability of data transformations to correct for lack of normality and homogeneity of variance, we used generalized linear mixed models (GLMMs) (Bolker et al. 2009) to analyse the effects of patch type, direction and distance from the core on seed species richness and assumed a Poisson error distribution. As in the seed deposition analysis, we used separate sets of models for distance and direction. The GLMMs were fitted using the Laplace approximation in the ‘lme4’ package for the ‘r’ statistical program (R Development Core Team 2005; as recommended by Bolker et al. 2009). Although ‘traditional’ anova tables can be produced for linear mixed models, this is not recommended for testing hypotheses in the case of GLMMs (Bolker et al. 2009). Therefore, we followed a model selection approach that was based on comparing the Akaike information criteria (AICs) between full models and subsets that excluded one or more main factors or their interactions. Following model selection, we tested the significance of each fixed factor or interaction in the model.
A total of 5604 nonmillet seeds were collected, of which 83% were identifiable at the plant species level. Of the latter, 73% belonged to one of 50 herbaceous species, 26% were S. spinosum seeds, and four species of other dwarf shrubs accounted for the remaining 1%. Of the 50 herbaceous species, seeds of 31 were found in both ‘shrub’ and ‘open’ plots, 11 were collected in open plots only, and 8 were collected in shrub plots only. However, most (13) of the 19 herbaceous species that were unique to only one patch type were collected by a single seed trap, which suggests that the apparent uniqueness mainly reflects rarity. Sixteen of the identified species exhibited adaptations for wind dispersal, and 15 other species were adapted for ground rolling (having rounded or nearly rounded seeds or dispersal units). A total of 181 seeds of wind-dispersed species were trapped, and these were evenly divided between the two patch types (93 and 88 seeds in the open and shrub patches, respectively). In contrast, of the 133 ground-rolling seeds that were trapped, 88 were found in the open patch and only 45 in the shrub patch.
Most (92%) of the millet seeds that were added to the traps to test for granivory were subsequently retrieved, and the effects of patch type, distance to patch core and direction on this retrieval rate were all statistically non-significant.
The model that best accounted for the variance in herbaceous seed deposition rate (as indicated by yielding the lowest AIC – 346.0) included patch type, distance from the patch core and their interaction. Exclusion of the interaction term increased the AIC value to 349.2, suggesting that the interaction was somewhat weak. Models that included distance only or patch type only fitted the data poorly, with considerably higher AIC values of 357.5 and 370.3, respectively. Both the main factors and also their interaction had significant effects on seed deposition rates (distance: χ2 = 28.7, P < 0.001; patch type: = 11.6, P <0.001; distance × patch type: χ2 = 7.34, P = 0.025). The herbaceous seed deposition rate was higher in the open patch than in the shrub patch and higher at the patch periphery than nearer to its core, and the shrub patch effect displayed a periphery-to-core declining gradient (Fig. 2a).
The model for species richness that included patch type, distance and their interaction yielded the best fit to the data (AIC = 138.5). Exclusion of the interaction term caused a modest increase in AIC, to 142.9. Models that included patch type only (AIC = 153.5) or distance only (AIC = 145.1) accounted for the least variance. The effects of distance, patch type and their interaction on seed species richness were all statistically significant (distance: χ2 = 14.39, P <0.001; patch type: χ2 = 4.10, P = 0.04; distance × patch type: χ2 = 8.30, P = 0.02). Seed species richness was lower at the core and intermediate positions of the shrub patch than at its periphery or at any position in the open patch (Fig. 3).
Not surprisingly, the deposition rate of S. spinosum seeds was overwhelmingly affected by the presence of the shrub (Fig. 2b), although it is interesting to note that also the open patches received S. spinosum seeds (which are relatively heavy), despite their distance from shrub patches. The model that included patch type only was far superior to that based on distance only (AIC = 354.1 and 453.7, respectively), and the model fit was not improved by the addition of distance (AIC = 357.9) or the patch type × distance interaction (AIC = 360.6). The effect of patch type in the most adequate model was statistically significant (χ2 = 134, P <0.001).
Models that included both direction and patch type, either with or without their interaction, yielded the best fits to data for herbaceous species seed deposition rate (AIC = 314.1 and 313.9, respectively). The model that included only direction fitted the data almost as well (AIC = 316.3), whereas the model that included only patch type fitted the data poorly (AIC = 326.5). The effects of direction and patch type, but not their interaction, were significant (direction: F3,133 = 6.79, P <0.001; patch type: F1,133 = 4.31, P = 0.04; direction × patch type: F1,133 = 1.87, P = 0.14). The overall herbaceous species seed deposition rate was highest in the eastern side of the shrub patch, and no such pattern was observed in the open patch (Fig. 4a).
Variation in species richness of herbaceous seeds was best accounted for by models that included direction, either with or without patch type, which yielded AIC values of 131.0 and 131.6, respectively. Inclusion of the (direction × patch type) interaction, or use of a model that included patch type only, yielded poorer results (AIC = 136.3 and 136.1, respectively). The effect of direction on species richness was statistically significant (χ2 = 10.82, P = 0.01). Species richness of herbaceous seeds was highest in the eastern sides of both the shrub and the open patches, with equal, lower values in all other directions (Fig. 5). Species richness was positively correlated with abundance (r = 0.57, t = 8.85, d.f. = 166, P <0.001, after log transformation of both variables). However, this correlation was much tighter in the shrub patch (r = 0.72, t = 9.59, d.f. = 86, P <0.001) than that in the open patch (r = 0.28, t = 2.59, d.f. = 78, P = 0.01).
The models that best accounted for the variation in deposition rate of S. spinosum seeds, with AIC = 306.6, included both patch type and direction. Addition of the (patch type × direction) interaction raised the AIC to 310.1, the model that included only patch type was poorer, with AIC = 315.2, and the model that included only direction was worst of all, with AIC = 401.6. The deposition rate of S. spinosum seeds was affected by both patch type and direction (patch type: F1,133 = 133; P <0.001; direction: F3,133 = 3.98, P = 0.01), with many more seeds in the shrub patch than in the open patch, especially in the eastern side of the shrub patch (Fig. 4b).
Wind-dispersed seeds were evenly distributed within and between patch types. The deposition rates of wind-dispersed seeds were 0.90 ± 1.00, 0.55 ± 0.75 and 0.55 ± 0.69 (seed m−2 day−1, mean ± SD) in the periphery, intermediate and core positions of the shrub patch, and 0.60 ± 0.74, 0.57 ± 0.80 and 1.25 ± 2.49 in the corresponding positions of the open patch, respectively. The effects of patch type, distance from the core, direction and their interactions were all not statistically significant. In contrast, the spatial distribution of ground-rolling seeds showed a clear structure in the shrub patch, with most seeds accumulating at the shrub's periphery, but no clear trend was observed in the open patch. The deposition rates of ground-rolling seeds were 0.72 ± 1.22, 0.04 ± 0.17 and 0.15 ± 0.30 in the periphery, intermediate and core positions of the shrub patch, and 0.73 ± 1.74, 0.51 ± 0.84 and 1.20 ± 1.73 in the corresponding positions of the open patch, respectively. The effects of patch type, distance from the centre and their interactions on deposition rates of ground-rolling seeds were all significant ( = 10.57, P = 0.005; = 9.2, P =0.002; = 8.01, P =0.02; for distance, patch type and their interaction, respectively). The effect of direction on seed deposition rate of ground-rolling seeds was not significant.
The net effect of shrubs on the herbaceous vegetation in semi-arid ecosystems may be positive, negative or neutral, depending on the fine-scale heterogeneity in environmental conditions, resource availability and biotic interactions, beneath and around the shrubs (Moro et al. 1997b; Segoli et al. 2012). Numerous studies in arid and semi-arid ecosystems have found that seeds accumulate under shrubs (Moro et al. 1997b; Guo, Rundel & Goodall 1998; Flores & Jurado 2003; Caballero et al. 2008). However, because most studies examined the seed bank in the soil, the patterns they described represented the combined effect of various spatially heterogeneous processes that included seed production, primary and secondary seed dispersal, and post-dispersal processes such as predation. Studies of single processes that contributed to the fine-scale spatial patterns of seeds around shrubs have focused on seed production (Pugnaire & Lazaro 2000), primary and secondary dispersal (Aguiar & Sala 1997; Marone, Rossi & Horno 1998; Venable et al. 2008), seed retention (Moro et al. 1997a; Chambers 2000) and seed predation (Russell & Schupp 1998). However, it is important to note that most studies that focused on seed dispersal in this context recorded seed arrival (or seed deposition) patterns, with no explicit reference to the location and strength of putative seed sources. Therefore, these studies could not determine whether the observed patterns of seed arrival were the result of spatial heterogeneity in seed production or in other processes such as dispersal. Our study adds to only three studies we know of (Russell & Schupp (1998), Bullock & Moy (2004) and Venable et al. (2008)) in which the potential confounding effect of a shrub patch on the seed source was controlled for when determining its effect on seed deposition. Like Venable et al. (2008), we controlled for heterogeneous seed production by excluding all within-patch seed input, except for that of the shrub. Whereas previous studies focused on the distribution of one or two species, our study is the first to consider an entire herbaceous community while directly testing the effect of shrubs on seed deposition. We found that the shrub S. spinosum, a key ecosystem engineer in semi-arid landscapes, modified seed flow, especially of ground-rolling seeds, and thus contributed to fine-scale spatial patterns in the abundance and species richness of seeds. Combined with other engineering effects of the shrub, its effects on seed deposition patterns may determine the fine-scale spatial pattern of the plant community (Segoli et al. 2012). Our manipulation clearly showed that shrubs can modify seed arrival patterns. However, only long-term manipulative studies can determine the extent to which patterns in the herbaceous vegetation depends on seed arrival patterns (such as in Kadmon & Tielborger 1999), on the distribution of other resources (Segoli et al. 2012) and on local interactions.
The flow of herbaceous seeds from outside the shrub patch to its core was significantly hindered by the shrub, as indicated by the decrease in the seed deposition rate from the shrub periphery to its core (Fig. 2a). Such a periphery-to-core gradient in seed deposition rate could have resulted simply from the extremely limited dispersal that is typical of many plants in arid and semi-arid landscapes (Shmida & Ellner 1984; Venable et al. 2008), irrespective of any shrub effect. However, the lack of such a gradient in the manipulated open patch suggests that it was the actual presence of the shrub that generated this gradient. The low seed deposition rate at the shrub interior, relative to that at the shrub periphery and at all positions in the open patch, suggests that the shrub functioned as a barrier that prevented seeds from reaching its core (Fig. 2a). Furthermore, the differences between the spatial patterns of wind-dispersed and ground-rolling seeds in the shrub patches and the accumulation of seeds at the eastern periphery of the shrubs (Fig. 4a) indicated that the shrub also acted as a seed trap at its eastern side, mainly of seeds that travel on the ground.
The fine-scale spatial pattern of seed species richness followed the pattern we observed for seed deposition rates: seed species richness at the shrub periphery was higher than that in the interior and comparable with that in the open patch. Thus, the shrub effect included reductions in both seed abundance and diversity of seeds that reached its interior. The reduced species richness in the shrub interior might have been the result of a differential impediment of seed flow towards the core or a statistical effect of smaller numbers of seeds reaching the shrub interior. The positive correlation between species richness and abundance, which was much higher in the shrub patch, indicates that the reduced species richness in the shrub's interior can be attributed, to a large extent, to the reduced number of seeds arriving there. In that respect, it is important to note that about half (26) of the herbaceous species were collected in no more than three plots. In addition, the difference between shrub and open plots in the total number of seeds collected per species was five or fewer for most (41) of the 50 herbaceous species. These low values place a limit on our ability to detect differential impediment at the species level. The changes in seed deposition rate and species richness could cascade to the above-ground herbaceous community. Working in the same study system, Segoli et al. (2012) found that both biomass and species richness of herbaceous plants were lowest in the shrub core, highest at the shrub patch periphery and intermediate in the open patch. The outward decline in deposition rate of S. spinosum seeds (which were left intact on the shrub) from the shrub core (Fig. 2b) mostly reflects the dispersal kernel of the shrub.
Our prediction that the effect of shrubs on seed deposition patterns will vary with dispersal mode and would correspond to the directionality of the putative dispersal vectors (e.g. wind and gravity) was only partially supported. The dominant wind direction in the study region is north-west to south-east, and the study was conducted on a west-facing slope. Therefore, we predicted that if shrubs functioned as wind blockers, wind-dispersed seeds would accumulate at its windward (north-western) side and that if the shrubs modified ground rolling (i.e. by gravity or run-off), ground-rolling seeds would tend to travel downslope and accumulate at the upslope (eastern) side of the shrub. The lack of any discernible pattern of wind-dispersed seeds suggests that if the shrubs modify wind flow, this has little effect on seed deposition. However, the shrub did seem to impede the inward movement of seeds with adaptations for ground rolling, although we were unable to detect differential directionality in the effect. Nevertheless, when all herbaceous species were considered together, the eastern side of the shrub patch received more seeds and had higher species richness than other sides of the patch. Thus, the higher seed deposition and species richness that we observed at the eastern side of the shrub suggest that shrubs modified the movement of a wider range of seed species than those we were able to classify as ground rolling. This observation is in agreement with other studies that found run-off and ground rolling to be important seed dispersal mechanisms in such systems (e.g. Boeken & Shachak 1994). It is also consistent with the modification of run-off by S. spinosum that was observed in previous studies (Segoli et al. 2012).
Interestingly, the seeds of S. spinosum – which were most probably produced within the shrub patch – also accumulated on the upper (eastern) side of the shrub patch (Fig. 5). If gravity-related mechanisms were the main drivers of seed arrival, then this pattern could be the result of seeds arriving from other (upslope) shrubs. However, this pattern could also result from seed retention being higher in the upslope side of the shrub.
The seed deposition pattern that we found in our study is consistent with the findings of a few studies conducted in similar systems (Marone, Rossi & Horno 1998 (for some plant species); López-Pintor, Espigares & Rey Benayas 2003), but contradicts those of many others that found a positive effect of shrubs on seed deposition (e.g. Moro et al. 1997a; Guo 1998; Caballero et al. 2008; Venable et al. 2008; Wang et al. 2010). Such apparent discrepancies may have several explanations that complicate comparisons among studies: (i) different studies recorded seed distribution patterns at different stages – for example dispersal, seed deposition and seed bank – which entail differing sequences of processes; in particular, most studies did not eliminate, or otherwise control for local seed production, and therefore, they could not dissociate the effects of spatial heterogeneity of seed production on seed spatial distribution from those related to seed movement and seed retention; (ii) the effect of a shrub on seed flow may depend on its structural properties, which may vary among and within shrub species (Pugnaire & Lazaro 2000; Caballero et al. 2008); and (iii) seeds of different herbaceous species might respond differently to the effects of shrubs (Marone, Rossi & Horno 1998; Boeken et al. 2004; Wang et al. 2010). Careful comparison among studies, with particular attention to potential sources of differences, should significantly improve our understanding of the interactions between shrubs and herbaceous plants that are so common in arid and semi-arid ecosystems. It is important to recognize the processes that lead to observed patterns, to develop methods for studying them separately and to examine the sensitivity of these processes to environmental drivers, management practices and various disturbances.
The effect of shrubs on herbaceous vegetation in arid and semi-arid ecosystems is often regarded as facilitation. However, the sign and the magnitude of the shrub's net effect may vary on a fine spatial scale (Moro et al. 1997b; Arnon et al. 2007; Caballero et al. 2008). Specifically, a recent study by Segoli et al. (2012) demonstrated that S. spinosum simultaneously imposed both positive and negative effects on the herbaceous vegetation, owing to differing combinations of resources at various positions within and around the shrub. The fine-scale distribution of resources combination is generated by partial spatial overlaps between the shrubs various functions (e.g. shading, run-off retention and soil accumulation). Similarly, the present study showed that a shrub can simultaneously function as a seed trap that accumulates seeds at one segment of its periphery and as a seed barrier that hinders seed movement towards the shrub core. Therefore, the functionality of the shrub in relation to seed movement is a fine-scale variable.
One implication of global climate change for arid and semi-arid ecosystems is an increase in the frequency and temporal proximity of drought years that is expected to exacerbate drought-induced mortality of woody plants (McDowell et al. 2008; Hoffman et al. 2009; Sigal 2009; Saaroni et al. 2012). Such changes in the abundance and distribution of important ecosystem engineers could have dramatic and long-lasting effects on ecosystem functioning and services (Hastings et al. 2007). Although some engineering functions of shrub patches remain active long after the canopy has been removed, seed trapping is likely to be very much affected by the integrity of the canopy structure and is one of the first functions to be affected when shrubs shrink and later lose their above-ground structure. A mechanistic experimental dissociation of the various functions that shrubs exhibit in semi-arid ecosystems is paramount if we wish to understand the response of such ecosystems to the alterations they experience. Our experimental approach for dissociating shrub effects on seed dispersal from other engineering effects is an important step forward in that direction.
This research was supported by the James S. McDonnell Foundation, by grant 1077-03 from the Israel Science Foundation and by the Eshkol Program of the Israeli Ministry of Science, Culture and Sport. We thank Sonia Rozin and Merav Perry for technical assistance and Moshe Shachak for constructive criticism of an earlier version of the manuscript. We are grateful to Bertrand Boeken for fruitful discussions and for kindly providing us with laboratory space for seed sorting, storage and analysis. We thank two anonymous referees for constructive suggestions. We thank the Ramon Science Centre for providing the facilities for the seed experiment. This is publication No. 756 of the Marco and Louise Mitrani Department of Desert Ecology, of the Ben-Gurion University of the Negev. None of the authors has any conflict of interest with regard to this manuscript.