1. Oak savannas are biodiversity-rich landscapes allowing sustainable livestock production throughout the world. The long-term persistence of these ecosystems critically depends on the regeneration of the tree layer. Nevertheless, studies addressing the mechanisms involved for conservation planning are of limited value because they tend to focus on single explanatory factors.
2. We evaluated the combined effect of biotic and abiotic factors on recruitment of holm oak Quercus ilex in the Mediterranean savannas of western Spain. Transition probabilities from flower to seed to the established seedling were estimated in grazed, shrub-encroached and cropped plus fenced habitats in two consecutive years.
3. Trees in cropped habitats produced more female flowers and larger acorn crops in both years. The physiological condition of trees was better in cropped habitats and worst in shrub-encroached plots. Overall, resource-mediated effects overrode the effects of biotic damage on tree fecundity in all habitats.
4. Acorn survival and seedling establishment were higher in cropped and shrub-encroached plots, though in cropped plots saplings are predictably destroyed by subsequent grazing and/or by mechanical treatment used to restart the cropping cycle.
5. Complete regeneration failure was found in 6 out of 24 possible management scenarios, mostly in the presence of large vertebrate herbivores. However, even low positive cumulative transition probabilities between life stages exceeded a safe threshold for early regeneration.
6.Synthesis and applications. Natural early recruitment of oak savannas can be achieved through various management regimes. These include cereal cropping in fenced plots (provided established saplings were not subsequently destroyed) or shrub encroachment in undergrazed or livestock-excluded plots. Among these, natural recruitment after encroachment is a cost-effective tool as compared to artificial plantation.
Among the most biodiversity-rich ecosystems threatened by global change, Mediterranean oak savannas are expected to suffer from both climate and land use changes (Blondel & Aronson 1999). Such savannas (or Spanish dehesas) have been historically shaped by humans through modification of existing evergreen forests (Moreno & Pulido 2009). In these ecosystems a two-layered structure results from the elimination of an intermediate layer of shrubs to enhance tree, grass or crop production (Moreno & Pulido 2009). Because shrub patches are dominated by early successional plant species, a temporary reduction in grazing intensity results in rapid shrub encroachment and a three-layered structure (Ramírez & Díaz 2008). At present, many ancient Mediterranean savannas have become treeless due to their prolonged use and lack of tree replacement (Pulido, Díaz & Hidalgo 2001; Plieninger, Pulido & Konold 2003).
Despite the general occurrence of regeneration failure in savannas, studies addressing the mechanisms involved tend to evaluate the role of a limited number of environmental factors judged a priori to be critical for tree regeneration (e.g. seed predation/dispersal or seedling mortality due to competition or abiotic stress). In the case of Mediterranean oak savannas, flower-to-sapling recruitment depends on the net effect of multiple interactions occurring at each stage and largely driven by human practices (Moreno & Pulido 2009). Therefore, to develop sound techniques for improving oak regeneration we must first analyse the effects of these interactions on the success of any given transition between stages of the regeneration cycle.
In this study we followed an integrative approach to study the effect of the habitat mosaic generated by land use on the regeneration of holm oak Quercus ilex L. On the basis of previous studies we formulate a number of predictions. Thus, we expected (i) a negative effect of shrub encroachment and a positive effect of cropping on tree physiological status and, consequently, on acorn production and viability (Moreno et al. 2007); (ii) a positive role of shrubby habitats in attracting acorn dispersers (Pulido & Díaz 2005; Pons & Pausas 2007; Muñoz, Bonal & Díaz 2009); and (iii) a facilitative effect of shrubs on seedling survival due to nurse effects associated with shade and protection from herbivores (Smit, Den Ouden & Díaz 2008). Results from the tests of such predictions were subsequently linked to ecosystem restoration by asking which combination of abiotic and biotic factors could ensure early regeneration of holm oak trees in Mediterranean savannas.
Materials and methods
The study was conducted in Cuatro Lugares County, within the province of Cáceres in western Spain (34·4 N, 6·13 W). In this region holm oak is the only tree species scattered (11–16 trees ha−1) over an extensive grassland dominated by annual species. Main soil types are Chromic Luvisols developed over Tertiary sediments (Moreno et al. 2007). Dehesas are primarily devoted to continuous grazing by cattle or sheep, with cereals being intercropped in long rotations on fenced plots in about 20% of the land. Cereals are sown in the interstitial spaces between oak trees and beneath the peripheral canopy area. Areas with reduced grazing intensity have been encroached with shrubs (Cistus ladanifer L., C. salviifolius L., Genista hirsuta Vahl. and holm oak resprouts).
Holm oak is a small- to medium-sized tree (usually 5–10 m in height). It is monoecious, with fertilization occurring in May–June after wind pollination. Developing fruits may fall before maturity either because of abortion, abnormal sap exudation (‘drippy’ fruits) or insect infestation by borer larvae of moths (Cydia spp.: Tortricidae) and weevils (Curculio elephas Gyll.: Curculionidae). From October to January fallen acorns are exposed to predation by vertebrates such as wild boar Sus scrofa L., rodents (Apodemus sylvaticus L. and Mus spretus Lat.), wood pigeons Columba palumbus L., and livestock. In the study area holm oak relies on scatter-hoarding by rodents for the dispersal of acorns (Muñoz, Bonal & Díaz 2009). Surviving acorns germinate in December–February and seedlings emerge between March and June (Pulido & Díaz 2005).
Study plots and general field design
We estimated the transition probabilities between stages of the life cycle, that is, the proportion of flowers producing fruits and the proportion of seeds producing seedlings. Probabilities were estimated in three different habitat types resulting from within-farm dehesa management: grazed, cropped, and shrub-encroached plots. In October cereal crops were sown after ploughing in cropped plots in between scattered holm oak trees. Crops were subsequently harvested in June the following year. Since cropped plots had to be fenced to prevent crop damage until harvesting, these plots were also useful to analyse the effect of the temporary exclusion of large mammals. Shrub-encroached plots resulted from shrub invasion due to low grazing intensity. Three replicated plots per habitat type, each on a different farm, were selected within an area of 10 × 10 km. Within each plot we randomly selected and marked seven mature holm oak trees (i.e. a total of 9 plots and 63 trees).
Precipitation and soil water availability
We monitored precipitation and soil moisture content through the study period. Monthly precipitation data were taken from the nearest weather station (Cáceres city; 39·47 N, 6·34 W). We monitored soil water content at monthly intervals from May 2002 to December 2004 in grazed and cropped-fenced plots, and from April 2003 to December 2004 in encroached plots. Soil water content was measured by means of the Time Domain Reflectometry method, with probes buried every 20 cm from the upper surface to 200 cm depth. Soil water content was measured in the three plots per habitat, each plot including four replicated soil water content profiles (Moreno et al. 2007).
Tree physiological status
We used maximum daily water potential (Ψl,), net leaf photosynthesis (A), and leaf nitrogen concentration (N) as the main indicators of the physiological conditions of adult trees during the limiting summer drought period. Leaves of six mature oaks per plot were harvested for nitrogen determination (mg kg−1) in August 2002 and 2003. Ψl was measured (in −MPa) just before dawn with a pressure chamber. For grazed plots values were calculated by averaging available data for summer 2002, 2003, and 2004; for cropped-fenced plots only data for 2003 were used, and for encroached ones only data for 2003 and 2004 were averaged. For each year per habitat combination, data were taken in two replicated plots and Ψl was estimated in 4–6 trees per plot. Net leaf photosynthesis was measured (in μmol CO2 m−2 s−1) at mid-morning in mature leaves of the same trees by means of a differential infrared gas analyzer and a broadleaf chamber (Moreno et al. 2007).
Acorn production and pre-dispersal acorn losses
We quantified tree fecundity and the incidence of pre-dispersal acorn losses by means of seed traps (four 0·12 m2 traps per tree) operating from May 2002 to January 2004. Flower or seed fall was quantified by means of monthly inspection of the traps, and the items found were classified as: non-fertilized flowers, aborted acorns, ‘drippy’ acorns, acorns infested by Cydia moths, acorns infested by Curculio weevils and sound acorns. We considered infested acorns to be killed by predators, as insect larvae destroy most internal seed reserves, seeds rarely germinate, and they produce less vigorous seedlings (Lombardo & McCarthy 2009). We estimated the percentage incidence of each source of pre-dispersal loss as the proportion of undeveloped or damaged flowers or fruits. In addition, the absolute number of flowers and fruits were calculated by extrapolating data from the traps to the whole projected canopy area (Pulido & Díaz 2005). The number of female flowers was computed as the sum of all types of propagules fallen into the traps (either non-fertilized, fertilized, infested or sound). From these data we calculated the transition probabilities from flower to fruit set (fertilization) and from undeveloped fruit to sound mature fruit.
Acorn availability and removal under the mother tree
After maturation we monitored acorn density on the ground beneath oak canopies. In October 2002 we selected 10 trees per plot in the three habitats (seven trees used in the pre-dispersal phase plus three trees added, i.e. 90 trees in nine plots). We placed two 0·2 m2 sampling circles under the canopy of each tree. One circle was open and the other excluded large vertebrates while allowing rodent activity by means of wire cages with a 51 × 51 mm mesh size. Density of acorns inside and outside the cages was recorded at monthly intervals from late October to the time of seedling emergence (April) in both 2002–2003 and 2003–2004. Predation rate by large vertebrates was computed as [1−(DO/DI)] × 100, where DO and DI are acorn densities outside and inside the cages, respectively. A rough estimate of rodent impact inside the exclosures was computed as [1−(DIA/DIO)] × 100, where DIA is the inside density of acorns in April and DIO is the inside density of acorns in October. This procedure renders a minimum estimate of rodent predation rates as acorns may continue to fall after the peak of acorn density (Pulido & Díaz 2005). The transition probability from sound acorn beneath the tree to unpredated (potentially germinating) acorn was obtained as 1 − PR, where PR is the predation rate exerted by large vertebrates or rodents.
Removal of experimentally dispersed acorns
From late October to May in both years we established one subplot away from adult trees in each plot to analyse the fate of experimentally dispersed acorns. On each subplot we placed 100 acorns in a 5 × 2 m grid by burying them at 2 cm depth. At monthly intervals acorns were classified as intact, removed by potential dispersers (rodents), trampled by livestock or wild boar, or partially eaten in situ by birds. Finally, we calculated a probability of transition as the proportion of intact dispersed acorns at the end of the experiment.
Emergence of seedlings in the field
We studied seedling emergence and spatial distribution under field conditions in each plot by means of non-permanent square sampling around the 10 focal trees per plot used in the study of predation of naturally dispersed acorns. For each tree we randomly established two sampling squares in each of three locations: under the canopy adjacent to the trunk, under the periphery of the tree canopy, and 10 m away from the tree in the open. In each square we counted the cumulative number of seedlings in June 2003 and 2004, that is, seedlings emerged from the 2002 and 2003 acorn crops respectively.
Seedling emergence and survival under controlled conditions
We studied emergence and survival of seedlings by sowing 100 acorns in one 1 × 1 m vertebrate-proof cage per plot in each habitat type. Cages were located in the interstitial spaces between trees since we were interested in the effects of the understorey management and these could be obscured by the effects of trees. Within the exclosures, acorns were arranged in a 5 × 5 grid, 7 cm apart from each other. We sowed them at 2 cm depth, covering the exclosure with a 2 mm plastic mesh to avoid interference by animals. Acorns were sown in the first week of December 2002 and 2003. Seedling emergence and survival were checked in June and October in 2003 and 2004 respectively. The term ‘seedling’ is used for emergence (June), whereas ‘established seedling’ implies that the plant survived the first critical summer drought (October). We calculated transition probabilities for emergence (seed to emerged seedling) and survival (emerged to established seedling).
Physical characterization and the physiological status of trees responded to incomplete designs (certain habitats were not studied in some years). Thus, two partial comparisons of means were conducted by General Linear Modeling (GLM) on an anova-type design for each parameter. In all these analyses, year and habitat were fixed factors and plot was a random factor nested in habitat. We tested for between-year differences in soil water content only in grazed plots. Between-habitat differences were tested as a function of habitat, year and soil depth. Between-year differences in leaf water potential and photosynthetic activity were tested for in grazed plots only. Between-habitat differences were tested for in year 2003 only. Finally, differences in leaf nitrogen content were tested using the complete design.
With respect to the demographic analysis, the probabilities of transition from any given stage to the next were used as response variables. Again, we used GLM with year and habitat as fixed factors and plot as a random factor nested in habitat. Generalized Linear Models (GLZ; StatSoft 2007) with the logit link function were used to test for effects on dichotomous variables. Independent observations were acorns and seedlings and they were coded as predated/intact or alive/dead, respectively. We used statistica 7.0 (StatSoft 2007) for all statistical analyses.
Water availability and tree physiological status
Rainfall monitoring revealed a first dry year followed by two average years (Fig. 1). After summer drought, soil re-watering occurred gradually through autumn and winter. Minimum soil water content values were similar in the 3 years with available data, though they occurred earlier in summer 2002 and 2004 (Fig. 1). On average, soil remained significantly wetter in summer 2003 than in summer 2004, and much wetter in 2004 than in 2002 (F =44·7, d.f. = 2,1454, P =0·002; Fig. 1). These differences matched the pattern found in summer leaf water potential and photosynthetic rate of adult oaks, which were significantly lower in 2002 and higher in 2003 (F =36·2, d.f. = 2,154, P =0·027 for leaf water potential, and F =44·3, d.f. = 2,329, P =0·022 for photosynthetic rate; Fig. 1).
Soil was significantly drier in encroached plots than in grazed and cropped-fenced plots (F =17·0, d.f. = 2,6, P =0·002; Fig. 1). These differences were highly significant at 50–100 cm and 100–200 cm depth (P <0·001), but they were not at 0–50 cm depth (P =0·651). Again, between habitat differences in soil water content paralleled those of leaf water potential and photosynthesis, with significantly lower values in encroached plots and higher in cropped-fenced as compared to grazed plots (F =11·0, d.f. = 2,138, P =0·041 for leaf water potential, and F =125·1, d.f. = 2,3, P <0·001 for photosynthetic rate; Fig. 1, Table 1). Finally, leaf nitrogen content also exhibited significantly lower values in encroached plots and higher in cropped-fenced plots than in grazed plots (F =18·0, d.f. = 2,6, P =0·003; Table 1), a result consistent between years.
Table 1. Mean values of leaf nitrogen content (N), pre-dawn water potential (ψ1), and net photosynthetic rate (A) in leaves of adult holm oak trees located in three different habitats
Different letters denote significant differences in Fisher post hoc LSD tests.
Nitrogen content (mg kg−1)
Predawn water potential (−MPa)
Net photosynthetic rate (μmol CO2 m−2 s−1)
Acorn production and pre-dispersal acorn losses
The absolute number of female flowers per tree was greater in year 2 (2003–2004) than in year 1 (2002–2003) but the differences were not statistically significant (Table 2). In addition, flower number was significantly larger in the cropped-fenced habitat (21 670 and 34 384; average values per tree for year 1 and year 2 respectively), which showed an approximately two-fold difference with respect to the other habitats in both study years (encroached: 9871 and 9987; grazed: 8948 and 16 961; Table 2). The differences among habitats peaked in the year with the larger flower crop. A significant effect of habitat was found on acorn production (Table 2). Cropping was associated with increased acorn production in both years (13 016 and 18 213), whereas fecundity of grazed plots (5592 and 9955) exceeded that of encroached plots (6147 and 5158) only in the second year.
Table 2. GLM analyses on nested anova design testing for the effect of year, habitat (fixed factors) and plot (random nested factor) on absolute fecundity (flowers and acorns) and on acorn viability of holm oak trees. For simplicity, propagule types were pooled into three main groups according to the source of damage: ‘abiotic’ damage (abortion plus sap exudation), ‘biotic’ damage (infestation by borer insects, square root-transformed) and sound acorns
% Abiotic damage
% Biotic damage
% Sound acorns
Figures in the table are F-values. *P <0·05; **P <0·01; ***P <0·001; ns, non-significant.
Year × Habitat (2,114)
Plot (Habitat) (2,114)
Pre-dispersal losses caused by abiotic factors (mostly abortion) were more important than those caused by biotic agents (weevil or moth infestation) irrespective of the habitat considered (Fig. 2). This was especially true for the second year, when the increased fecundity resulted in a higher proportion of acorns escaping from insect damage (Table 2, Fig. 2). Between-habitat differences in pre-dispersal losses were apparent for the percentage of sound acorns, the grazed habitat showing the highest values and the encroached one the lowest (Table 2, Fig. 2). Nevertheless, no consistent among-habitat differences were found in the impact of abiotic or biotic damages (Table 2, Fig. 2).
We found a significant negative effect of total flower production on the percentage of fertilized flowers (F =10·98, d.f. = 1,119, P =0·001), whereas flower production did not affect the percentage of sound acorns (F =0·38, d.f. = 1,119, P =0·536; covariate effects in ancova with habitat as factor). Finally, no effects of the size of the acorn crop on the percentage of drippy, weevil-infested or moth-infested acorns were found (P >0·389 in all cases).
Acorn dispersal and removal under trees
The density of acorns fallen under the canopy of trees was significantly higher in year 2, with the highest values recorded in cropped-fenced plots and the lowest in encroached plots (Fig. 3). A three-way anova showed no differences between the initial density of acorns inside and outside the cages (P =0·946), but significant effects of year (P <0·001), habitat (P =0·019), and their interaction (P =0·007) were found. From early autumn to the end of the winter period removal led to depletion (<0·26 acorns m−2) of unprotected acorns in grazed and encroached sites (Fig. 3). By contrast, in cropped-fenced plots unprotected acorns were depleted in year 1 but not in year 2 (mean density: 4·71 acorns m−2). Within cages, final acorn densities were close to depletion (<1·5 acorns m−2) in the three habitats analysed in year 1 but not in year 2 (Fig. 3). Large vertebrates depleted unprotected acorns in all plots in year 1. However, in year 2 acorn removal between habitats was marginally different (Table 3). Similarly, rates of removal by rodents were very high in almost all plots, with a similar decrease in the good acorn year 2 (Fig. 3). Removal by rodents was marginally affected by year but was not affected by habitat type (Table 3).
Table 3. GLM analysis on nested anova design testing for the effects of habitat, year (fixed effects) and plot (random nested effect) on predation rates by large vertebrates (wild boar, livestock and birds) and rodents (both variables after angular transformation)
Predation rate by large vertebrates
Removal rate by rodents
Year × Habitat
Removal of experimentally dispersed acorns
Large mammals (wild boar and livestock) and birds killed acorns through consumption or trampling, while removal can be attributed exclusively to rodents. As a result, the majority of acorns were manipulated in some way in both study years (Fig. 4). The survival of experimental acorns was significantly affected by year (P =0·006), habitat (P <0·001), and their interaction (P <0·001; GLZ analyses with survival of acorns as dichotomous response). In both years most acorns were removed by rodents in cropped-fenced and encroached, while in grazed plots most acorns were killed by trampling (Fig. 4). Again, predation rates in general decreased in the good acorn year.
Natural emergence of seedlings
Differences among years in the density of seedlings that had emerged by June from the preceding acorn crop were not significant (Table 4, Fig. 5). However, a significant effect of habitat type was noticed in both years, with seedling density increasing from grazed to encroached to cropped-fenced plots. Within any given plot, emergence was significantly concentrated under the peripheral canopy of trees (Fig. 5, Table 4). Almost no seedlings were found in the interstitial space between adult trees, and the few seedlings established in the open (necessarily derived from rodent dispersal) occurred in cropped-fenced or encroached plots (Fig. 5).
Table 4. GLM analysis on nested anova design testing for the effects of habitat, microhabitat, year (fixed effects), and plot (random nested effect) on the density of naturally emerged seedlings
Source of variation (d.f.)
Plot (Habitat) (2,3216)
Microhabitat × Habitat (4,3216)
Microhabitat × Year (2,3216)
Habitat × Year (2,3216)
Microhabitat × Habitat × Year (4,3216)
Emergence and survival of caged seedlings
Within cages the success of seedling emergence in year 1 was moderately high in grazed (61·0 ± 13·6%) and encroached plots (57·3 ± 25·2%) but it was lower in cropped-fenced plots (26·0 ± 8·5%). Emergence decreased in year 2 in all habitats, though the extremely low success rate in the grazed plots seemed to be the result of intense competition due to our failure to completely control herb growth within cages that year (Fig. 6). Overall, the significant effect of habitat seemed to reflect the amount of competing herb biomass within experimental cages, which was much greater in cropped-fenced plots (Table 5). Habitat type had a significant effect on seedling establishment, with cropping being associated with enhanced summer survival of seedlings (Table 5). Survival was intermediate in encroached plots and extremely unlikely in grazed plots in both years (Fig. 6).
Table 5. Results from Generalized Linear Models on a nested anova design testing for the effects of year, habitat, and plot on the emergence and survival of caged seedlings
Chi-square (LR test)
Chi-square (LR test)
Year × Habitat (2)
Plot (Habitat) (8)
Probabilities of transition among life stages and final recruitment
Besides evaluating the effects of year and habitat, we compared the probabilities of transition between life stages in a ‘Control’ scenario characterized by the presence of all herbivores and a ‘Exclusion’ scenario with no large vertebrate present (that is, within cages close to experimental trees). In addition, two dispersal regimes were considered by computing transition probabilities under the influence of adult trees or far away from them (see Appendix S1 for the rationale used in these calculations). We found zero recruitment in 6 out of 24 possible combinations (2 years × 2 treatments × 3 habitats × 2 dispersal regimes), which predominantly occurred in Control plots (five cases) or in grazed habitats (four cases). In all cases where recruitment rates were positive, we estimated from the number of female flowers produced that, on average across plots, recruitment would suffice to produce at least one seedling per adult tree each year (Appendix S1).
Figure 7 depicts propagule cumulative survival curves through the whole reproductive cycle for all year-treatment combinations. Demographic processes leading to early recruitment were sensitive to treatment effect (control vs large herbivore exclusion), habitat effect (cropped-fenced vs. grazed vs shrub-encroached), as well as to the dispersal scenario (under canopy vs. away from trees). In addition, the relative importance of these factors changed from one year to the next (Fig. 7).
The success of a given individual tree at the end of the pre-dispersal phase of the cycle is estimated by the absolute number of sound acorns produced. No consistent among-habitat differences were found in the impact of abiotic or biotic damage. Therefore between-habitat differences in fecundity must be attributed to disparities in the absolute number of female flowers rather than to differential success of the transitions from flower to fruit. Trees in cropped-fenced habitats produced more flowers and, consequently, larger acorn crops. In agreement with our expectations, the physiological status and fecundity of trees was better in the cropping situation, while it declined in shrub-encroached plots. Low flowering success in shrubby plots is presumably the consequence of their higher tree and shrub densities, which imposes a severe competitive regime, mainly for water (Moreno et al. 2007).
Pre-dispersal losses due to abiotic factors were greater than those caused by biotic agents in all habitats. However, previous studies with holm oaks have shown even higher rates of fruit abortion during the summer drought period (Pulido & Díaz 2005). The relatively low rate of fruit abortion is consistent with the high leaf water potential and photosynthetic rate measured in trees of the three habitats, whose values were much higher than those reported in denser stands (Savé, Castell & Terradas 1999; Moreno & Cubera 2008). These authors reported leaf water potentials and photosynthetic rates ranging −1·6 to −2·5 MPa and 2·6–2·7 μmol CO2 m−2 s−1, respectively, compared to ranges of −0·4 to −1·5 MPa and 8·7 to 11·3 μmol CO2 m−2 s−1 found in our study. In fact, a water plus nutrient addition experiment conducted in our study area resulted in no increase in female flower or acorn production (Moreno et al. 2007). Overall, these results suggest that some mechanisms could have operated to alleviate resource limitation and tree-understorey competition for soil resources, such as water uptake from deep soil layers (Moreno et al. 2005).
Our search for the main intrinsic predictors of pre-dispersal losses showed that, across habitat types, the initial flower load negatively influenced fertilization rates which, coupled with abortion, were the main determinants of the final seed output. Regarding the effect of acorn crop size on insect infestation, we found that the attack rates of weevils and moths were generally low (<20%) and density independent, as noted in previous studies (Pulido & Díaz 2005; Bonal, Muñoz & Díaz 2007). Taken together, these results indicate that resource-mediated effects on flower and fruit production override the effects of biotic interactions in the pre-dispersal stages. From an applied perspective, this implies that eventual seed limitation of recruitment could be more effectively reduced through decreasing plant–plant competition (tree and, preferably, shrub density) rather than by controlling fruit pests.
Post-dispersal fate of acorns
Vertebrate consumption of acorns resulted in crop depletion in most cases. Thus, acorns dispersed beneath isolated holm oak trees only survived in the absence of large vertebrates, i.e. in cropped-fenced plots, during a good acorn year. Similarly, the impact of acorn removal by rodents was also moderate to high in both years. However, in cropped-fenced plots, rodents were apparently satiated, as acorns were still available within cages at the end of the season. As previous studies have shown, acorn removal by rodents could be mediated by the effect of interference by large mammals (Focardi, Capizzi & Monetti 2000; Muñoz, Bonal & Díaz 2009). If this is the case, cropped-fenced plots would provide food-rich refuges for rodents facing both direct behavioural interactions and indirect competition for food from larger animals in grazed plots.
Removal of dispersed acorns was greater in cropped-fenced and encroached sites, whereas regular grazing resulted in the destruction of most acorns placed in open grasslands. Few acorns were removed by rodents in grazed habitats, whereas 40–80% of acorns were removed in the other habitats. Some acorns could have cached and left uneaten by rodents, thus producing new seedlings (Acàcio et al. 2007; Pons & Pausas 2007). On the other hand, the proportion of intact acorns was greater in the cropped-fenced plots in one of the years. Overall, our results partially support the predicted role of shrub patches in increasing acorn survival, since consumption and removal rates were comparable or even lower in the cropped-fenced habitat, presumably due to the reduced impact of large vertebrates, as also found by Muñoz, Bonal & Díaz (2009).
Habitat and microhabitat sensitivity of seedlings
All previous studies on the emergence of seedlings from acorns of Mediterranean oaks have shown that this process does not limit recruitment, as germination rates are usually moderate to high (Pulido & Díaz 2005; Acàcio et al. 2007). Though we did not conduct manipulative experiments, emergence rates seemed to depend in part on herb interference, as they were strongly reduced in cropped-fenced plots despite the favourable moisture conditions (Moreno et al. 2007). By contrast, emergence was relatively high when herbs were removed from inside the cages in experiments simulating grazing by herbivores (Weltzin & McPherson 1999). On the other hand, even under stronger inhibition by herbaceous plants, the density of naturally emerged seedlings was much higher in the cropped-fenced plots, followed by the shrub-encroached plots. This pattern is the result of the much lower survival of potentially germinating acorns in grazed sites.
The majority of seedlings were found beneath the peripheral tree canopy and close to the trunk, where the bulk of acorns were passively dispersed, as also noted by Weltzin & McPherson (1999). Almost no seedlings were found in the interstitial space between adult trees, and the few seedlings established in the open in cropped-fenced and encroached plots where there was some protection from large vertebrates.
The survival of seedlings from emergence to post-summer establishment followed the opposite pattern when compared with emergence. Thus, summer survival of seedlings increased in the croplands, whereas establishment was extremely unlikely in grazed plots. Assuming that by the onset of the summer drought period herbs were completely dry, their role would turn from competition to shade facilitation (Smit, Den Ouden & Díaz 2008). It must be noted, however, that, despite enhanced seed and seedling survival in cropped-fenced plots, in such plots crop harvest and the subsequent removal of fences allowed livestock grazing once again. In addition, soil cultivation of cropped habitat would destroy tree seedlings in the long term.
An integrated view of the recruitment process
Several novel findings result from our integrated approach to holm oak early regeneration in savannas. First, we found that whenever early recruitment rates were greater than zero (18 out of 24 possible combinations), the recruitment levels found would be enough for self-replacement of individual trees under the assumption of declining mortality rate with juvenile age (Gibbons et al. 2008). Secondly, multiple limitations to recruitment were found to act in holm oak savannas, as also shown for cork oak stands (Acàcio et al. 2007). Thus, complete regeneration failure was found in six cases, mostly in the Control treatment including large mammals and also in grazed sites (Fig. 7, Appendix S1). Interestingly, failure to recruit any seedlings was due to complete post-dispersal acorn depletion in the poor acorn year and to complete seedling mortality in the good acorn year. These results suggest that, at the interannual scale, acorn production could be an important driver of early tree regeneration, as also shown in our own 12-year data for holm oak ungrazed forests (F. Pulido & M. Díaz, unpublished data).
As an additional limitation, the importance of seed escape from post-dispersal predators is well illustrated by comparison of early recruitment rates under canopies vs. open sites. Only in two grazed sites (out of 12 possible cases) was recruitment failure found away from established trees, and this was due to acorn destruction by grazing animals (Appendix S1). Thus, contrary to most previous field studies (e.g. Pulido & Díaz 2005; Acàcio et al. 2007; Smit, Den Ouden & Díaz 2008), we found that in most sites a number of seedlings were able to tolerate full sun and water deficit in open microhabitats. Therefore, at least under certain conditions, rodent dispersal to open sites may not only lead to escape from seed predators but also to successful establishment in the short term.
Thirdly, by considering the 24 curves representing probabilities of transition among life stages under different conditions (Fig. 7 and Appendix S1), different trajectories leading to similar recruitment outputs were apparent. Thus, even though the values of final recruitment rates were within one order of magnitude, they were achieved through different values of the probabilities of transition among stages. These novel results point to the existence of several pathways for reaching an effective regeneration threshold.
Applying tree demography to the restoration of three-layered savannas
The lack of tree recruitment is a critical, long recognized threat for the persistence of many savannas worldwide (Manning, Fischer & Lindenmayer 2006; Fischer et al. 2009). In addition to some shared features, three-layered (shrub-encroached) savannas differ from simple wood pastures in several key traits for regeneration. First, trees behave as sinks for regeneration due to the extremely high mortality rates of seeds and seedlings under established adults (Pulido & Díaz 2005; and this study). Secondly, while seed dispersal to interstitial spaces is a requisite, it does not ensure tree recruitment due to high drought-induced mortality. Thirdly, unlike savannas in which safe sites for seedlings are generated by trees (Weltzin & McPherson 1999) or livestock foraging (Reid & Ellis 1995), recruitment in Mediterranean savannas has the potential to be assisted by a third layer of shrubs (Pulido & Díaz 2005; Smit, Den Ouden & Díaz 2008). As this study shows, decreased tree fecundity in shrubs patches could be more than compensated by reduced removal and increased dispersal of acorns, as well as by shade facilitation of seedling establishment. Therefore, the management of shrub patches should be viewed as a cost-effective tool to achieve oak regeneration under specific conditions, as also shown for tree islets in agricultural landscapes (Ramírez & Díaz 2008; Smit, Den Ouden & Díaz 2008; Rey-Benayas, Bullock & Newton 2010). We suggest that a mosaic landscape of wood pastures including regenerating islands of trees resulting from shrub encroachment, would ensure the long-term persistence of savannas habitats.
Financial support for this work came from projects SAFE (European Union QLK5-CT-2001-0560), and MODE (Spanish Research Program, AGL2006-09435). J.J.O. and E.G. were supported by grants from ANUIES (México). We are also grateful to Mario Díaz, Juli Pausas, and several reviewers for critically reading earlier versions of the manuscript, and to Fundación Global Nature and the owners of Sotillo, Cerrolobato and Dehesa Boyal de Talaván for permission to work at their states.