Introduction

Over the last two decades, widespread tree decline and mortality have been documented in forests of almost every bioregion of the world (Allen and others 2010, 2015; Breshears and others 2009; Van Mantgem and others 2009). Several global change drivers have been identified as potential causes, such as increasing drought frequency and severity, recurrent pest outbreaks, or the spread of exotic pathogens (for example, Axelson and others 2009; Carnicer and others 2011; Loo 2009). A common feature to these mortality processes is that they usually show certain level of host-specificity, some tree species being much more vulnerable to abiotic and biotic sources of stress than others. As a result, they have a large potential to induce selective species removals and changes in community composition (Allen and Breshears 1998; Collins and others 2011). For example, in semi-arid Pinus edulis-Juniperus monosperma woodlands of southwestern USA, recent severe droughts have induced a shift towards Juniperus-dominated communities due to the higher drought tolerance of this species (Koepke and others 2010; Mueller and others 2005). In coastal California forests, the emerging pathogen Phytophthora ramorum has caused a rapid decline of Notholithocarpus densiflorus and an increase in the relative dominance of coexisting Umbellularia californica or Sequoia sempervirens (Brown and Allen-Diaz 2009; Ramage and others 2012). In Australian forests, the largest susceptibility to both drought and pathogens of the Eucalyptus genus compared to the co-dominant Corymbia genus has lead to severe alterations in the relative dominance of coexisting tree species (Fensham and Holman 1999; Cahill and others 2003). However, despite this evidence for rapid changes in the relative abundance of the adult community, we still have limited information on trajectories of forest recovery after tree decline and mortality and whether they are likely to result in long-term vegetation shifts.

The probability of occurrence of permanent vegetation shifts is to a large extent determined by regeneration dynamics after tree dieback (Galiano and others 2013; Kayes and Tinker 2012; Redmond and Barger 2013; Suarez and Kitzberger 2008). Tree defoliation and mortality can induce a series of changes in local environmental conditions that alters the probability of establishment of new tree seedlings, inducing shifts of species abundance rankings in the seedling bank. For example, an increase in radiation levels and drought stress in the gaps opened after tree death could preclude the establishment of late-successional shade-tolerant species, indirectly favoring pioneer drought-tolerant species (Amoroso and others 2012; Diskin and others 2011; Ibáñez and others 2015). It has also been shown that trajectories of recovery after drought- or insect-driven tree mortality might depend not only on new seedling establishment, but also even more strongly on advance regeneration established prior to the disturbance (Collins and others 2011; Kayes and Tinker 2012; Redmond and Barger 2013). For example, tree dieback in mature forests could release suppressed saplings of shade-tolerant species, allowing late-successional species to keep dominating the stands and indirectly limiting the establishment of light-demanding pioneer species otherwise typical of disturbed sites (DeRose and Long 2010; Veblen and others 1991). These examples illustrate the complexities inherent to post-mortality regeneration dynamics, with seedlings and saplings of different ages likely responding in different ways (for example, Galiano and others 2013) and show the need for further research that helps to elucidate long-term changes in stand composition of disturbed forests.

In this study, we analyzed the spatial patterns of natural regeneration of the woody plant community in mixed oak forests of southwestern Spain affected by the decline of its dominant tree species, Quercus suber. Our main objective was to elucidate whether the abundance and performance of seedlings and saplings of coexistent shrub and tree species varied depending on the composition and health status of neighboring trees, and whether differential responses among species in the regeneration bank could eventually lead to changes in species dominance at the canopy level. The decline of Q. suber has been reported throughout the Mediterranean Basin since the early 1990s (Brasier 1992; Brasier 1996), where tree death has occurred progressively along time forming stands of dead trees. Several abiotic (for example, extreme droughts) and biotic (for example, insects and pathogens) factors are potentially involved in this decline (Tuset and Sánchez 2004). However, in the study area, oomycete soil-borne pathogens (Phytophthora cinnamomi and Pythium spiculum) have been isolated from symptomatic Q. suber trees and are suggested to be the main drivers of the species decline (Brasier 1996; Romero and others 2007; Sánchez and others 2006; Sánchez and others 2002). For the analysis of natural regeneration patterns, we used a spatially explicit neighborhood approach (Canham and Uriarte 2006; Gómez-Aparicio and others 2008a; Gómez-Aparicio and others 2008b) where the abundance, survival, and richness of the seedling and sapling bank were predicted as a function of the distribution and health status of the canopy tree and shrub community. Previous studies in the same study area have shown Q. suber to suffer from stronger recruitment limitation than coexistent trees or arborescent shrubs (Pérez-Ramos and Marañón 2012), mainly due to heavy post-dispersal seed predation and high seedling mortality due to summer drought (Gómez-Aparicio and others 2008b; Pérez-Ramos and Marañón 2008). Moreover, in a parallel experimental study, where seeds of Q. suber and Q. canariensis were sown and the recruits were monitored for 3 years, we found that defoliated and dead Q. suber trees generated unsuitable microsites for survival of conspecific seedlings likely due to drought stress and/or high pathogen abundance (Ibáñez and others 2015). Following this, we hypothesized that the process of Q. suber decline might cause the seedling/sapling bank to become even less Q. suber dominated, hampering the potential for self-replacement and favoring successional trajectories towards forests dominated by other coexisting woody species.

Materials and Methods

Study Sites and Species

The study was conducted in Los Alcornocales Natural Park, a 170 000 ha protected area in Southwestern Spain. The climate is Mediterranean type, with cold and humid winters and warm and dry summers. Mean annual temperature varies from 14.6 to 18.4°C, with a mean monthly maximum of 36°C (July) and a mean monthly minimum of 2.8°C (January). Mean annual precipitation is 975 mm (mean 1951–1999). The bedrock is dominated by Oligo-Miocene sandstone and originates acidic, nutrient-poor, sandy soils, frequently interspersed with clayish soils derived from layers of marl sediments. Our study was carried out during three consecutive years (2010, 2011 and 2012) with contrasted climatological conditions. The year 2010 was extremely wet in terms of precipitation (1346 mm annual rainfall, 40 mm summer rainfall); 2011 had an average precipitation (1037 mm annual rainfall, 16 mm summer rainfall); and 2012 was particularly dry (474 mm annual rainfall, 0 mm summer rainfall).

The flora in the Alcornocales Natural Park is dominated by mixed sclerophyll forests of Q. suber, located mainly on non-carbonated soils, at altitudes between 100 and 700 m a.s.l. Within the park, the structure of the forests and its diversity vary depending on orography and soil type. In low altitude sites with clayish soils, Q. suber forms mixed open woodlands with the evergreen drought-tolerant Olea europaea var. sylvestris. The shrub layer in these woodlands is usually dense and largely dominated by the evergreen Pistacia lentiscus L. and the deciduous Crataegus monogyna Jacq. In sandier, moister, and colder sites, Q. suber coexists with the deciduous, shade-tolerant Q. canariensis forming closed forests. The shrubby understory is diverse and dominated by arborescent shrubs (Arbutus unedo L., Phillyrea latifolia L.) and heath species (Erica arborea L., Erica scoparia L.) (Ojeda and others 1996). These forests are relatively well conserved since 1989 (declared as Natural Park) and the main management activities are the extraction of cork, game hunting, and recreation (Anonymous 2005).

Field Sampling

The field work was conducted in six study sites within the Natural Park. Three of the sites were located in open woodlands of Q. suber and O. europaea var. sylvestris (hereafter woodland sites) and three in closed forests of Q. suber and Q. canariensis (hereafter closed forest sites). The six sites covered a gradient of climate and soil conditions (see Table A in Supplementary Material). During winter 2009, we established a 70 × 70 m plot at each of the six study sites. Each plot was subdivided into 49 10 × 10 m subplots. Within each of the 49 subplots, a smaller 1 × 1 m quadrat was permanently set up for monitoring of natural regeneration (n = 147 sampling quadrats per forest type, 294 quadrats in total).

During early June in 2010, 2011, and 2012, we counted and marked all the seedlings (that is, individuals emerged that spring) and saplings (that is, >1-year-old individuals smaller than 50 cm height) of woody species in each of the 1 m2 sampling quadrats. We chose this sampling date to ensure that most seedlings had emerged (Pérez-Ramos and Marañón 2012). Additionally, seedlings and saplings of tree species were revisited in early October 2010 and October 2011 to record survival after the summer, the main mortality period in Mediterranean systems (Gómez-Aparicio 2008;Pérez-Ramos and others 2012). We also calculated species richness for each quadrat as the number of different woody plant species censured at seedling or sapling stage. Due to the low growth rates of oak species in Mediterranean systems and their resprouting ability, saplings smaller than 50 cm height can be as old as 30 years (Galiano and others 2013). Therefore, the sapling bank could represent the regeneration accumulated during the last decades and even before the first report of Q. suber decline in the area (Brasier 1992; Brasier 1996).

To characterize the local neighborhood of each plot, we identified and mapped all live and standing dead trees with a diameter at breast height (dbh) greater than 2 cm and all shrubs in the 70 × 70 m permanent plots, as well as in a buffer zone 15 m (for trees) or 5 m (for shrubs) wide around each plot. Tree neighborhoods of similar size have been shown to capture the most important aspects of neighborhood interactions in temperate forests (Coates and others 2009; Gómez-Aparicio and others 2008a). Although we did not have any reference to choose the maximum shrub neighborhood, we considered a radius of 5 m to be big enough based on the small size of most shrubs in these forests (height usually <3 m) (Gómez-Aparicio and others 2012). We measured the dbh of each of the trees mapped (n = 1341 trees). Due to its multi-stem growth form, shrub size was characterized measuring the two diameters of the elliptical projection of its crown (n = 3005 shrubs). In addition, we visually evaluated the crown health status of Quercus suber, with a standardized semi-quantitative scale used routinely for monitoring purposes of oak decline (García and others 2011): (1) healthy reference trees, (2) defoliated trees, and (3) dead trees. No other tree or shrub species in the study area showed symptoms of decline.

Data Analysis

Neighborhood Models of Seedling and Sapling Performance and Richness

We used likelihood methods and model selection for the analysis of our data (Canham and Uriarte 2006; Johnson and Omland 2004). Following the principles of likelihood estimation, we estimated model parameters that maximized the likelihood of observing the data measured in the field. We fit separate models for each combination of dependent variables (abundance, survival, and richness), life stage (seedling and sapling), forest type (woodland and closed forest), and woody species/functional group. The dominant tree species were analyzed separately (Q. suber, O. europaea and Q. canariensis), whereas the remaining woody species were divided into three groups according to their life form and dispersal syndrome: fleshy-fruited shrubs (with endozoochorous dispersal), dry-fruited shrubs (with dispersal syndromes other than endozoochory, mainly abiotic dispersal), and lianas (Table B in Supplementary Material).

Our full regeneration model predicted mean seedling/sapling performance in each quadrat as a function of three components: (1) the potential seedling/sapling performance at each of the three study sites for each forest type in the absence of specific effects of neighboring trees (that is, site effects); (2) the identity, size, health status, and spatial distribution of the trees in the neighborhood (that is, tree neighborhood effects); and (3) the size and spatial distribution of shrubs in the neighborhood (that is, shrub neighborhood effects). We tested and compared two different model frameworks (additive vs. multiplicative) to describe neighbor effects (for example, Baribault and Kobe 2011):

$${\text{Additive}}\,\,{\text{model}}\,\,\,\,\,\,\,\,\,\,\,Y = \,a_{{{\text{Site}}}} \, + \,b_{{{\text{Year}}}} \, * \,{\text{NI}}_{{{\text{Tree}}}} \, + \,c_{{{\text{Year}}}} \, * \,{\text{NI}}_{{{\text{Shrub}}}} \, + \,\varepsilon $$
(1)
$$ {\text{Multiplicative model}}\quad Y \, = \, a_{\text{Site}} * \exp \left( {b_{\text{Year}} * {\text{NI}}_{\text{Tree}} } \right) * \exp \left( {c_{\text{Year}} * {\text{NI}}_{\text{Shrub}} } \right) \, + \varepsilon $$
(2)

The first term in the models, aSite, represents the site effects. The second term accounts for the tree neighborhood effects, which are assumed to vary as a function of a neighborhood index (NITree). The parameter b defines the steepness of the variation in performance due to an increment in NITree and was allowed to vary between years and account for inter-annual differences in tree effects. The tree neighborhood index quantifies the net effect of j = 1,…,n neighboring trees of i = 1,…,s species on seedling/sapling abundance, survival or richness and was assumed to vary as a direct function of the size (dbh) and as an inverse function of the distance to neighbors:

$$ {\text{NI}}_{{{\text{Tree}}_{ij} }} = \sum\limits_{i = 1}^{n} {\sum\limits_{j = 1}^{n} {\lambda_{i} } \frac{{dbh_{ij}^{\alpha } }}{{dist_{ij}^{\beta } }}} $$
(3)

where dbh ij is diameter at breast height, dist ij is the distance to the sampling quadrat of the jth neighboring tree of the ith species, and α and β are estimated parameters that determine the shape of the effect of the dbh (α) and the distance to neighbors (β) on the target variable. We used NI standardized (0< NIi/NImax <1) to facilitate comparisons across different seedling/sapling species. Because we were particularly interested in exploring whether tree effects varied between individuals of different species or health status, we grouped tree neighbors in four categories: healthy Q. suber trees, defoliated Q. suber trees, dead Q. suber trees, and the coexisting tree species at each site (O. europaea or Q. canariensis). We initially considered two categories of defoliated trees (<50 % and >50 % defoliation, see Gómez-Aparicio and others 2012 for a similar approach), but because they had similar effects on regeneration, we decided to merge them in just one category for simplicity. We then multiplied the net effect of an individual tree by a per-capita coefficient (λ), one for each category (estimating therefore four different λ values), that ranged from −1 to 1 and allowed for differences between neighbors in their effects (negative or positive) on the target variable.

The third term in Equations. 1 and 2 accounts for shrub neighborhood effects and were calculated as a function of a shrub neighborhood index (NIShrub). The parameter c defines the steepness of the variation in performance due to an increment in NIShrub, and was allowed to vary between years and account for inter-annual differences in shrub effects. This index is a simplified version of the tree neighborhood index and quantifies the net effect of j = 1,…,n neighboring shrubs on seedling/sapling abundance, survival or richness following the form:

$$ {\text{NI}}_{{{\text{Shrub}}_{j} }} = \sum\limits_{j = 1}^{n} {{\text{area}}_{j} } $$
(4)

The NIShrub was assumed to vary just as a direct function of the size (crown area) of neighbor shrubs in a 5-m-radius neighborhood. We decided not to include distance in the calculation of the index given the already restricted area over which shrubs were mapped and to keep the number of parameters in the models manageable.

Parameter Estimation and Model Selection

Because our main interest was to compare the effect of trees of different species and health status on regeneration, we first fit a simple model where only Tree neighborhood effects were considered (“Only tree” model). We then compared this simple model with more complicated models that included Site, Year or Shrub effects, finally selecting the simplest model that was not a significantly worse fit than any more complicated model. Additive and multiplicative models with the same number of parameters were also compared among them. We used the Akaike Information Criterion corrected for small sample sizes (AICc) to select the best model, with lower AICc values indicating stronger empirical support for a model (Burnham and Anderson 2002). We solved for the maximum likelihood parameter values with simulated annealing (Goffe and others 1994), a global optimization procedure. The likelihood functions used in our models varied according to the data. We used a negative binomial function to model the abundance data (counts) accounting for over-dispersion. We also tested the use of a zero-inflated Poisson distribution, but found the negative binomial to be more appropriate to the error distribution of our data. The survival data were binomial in nature, so a binomial distribution was used for the likelihood function of the corresponding models. Finally, the species richness data (counts) were adequately modeled using a Poisson likelihood function. We used asymptotic 2-unit support intervals (as an approximation to the 95 % confidence interval) to assess the strength of evidence for individual maximum likelihood parameter estimates (Edwards 1992). Coefficients whose 2-unit support intervals did not include zero were considered statistically significant. The slope of the regression (with a zero intercept) of observed versus predicted was used to measure bias (with an unbiased model having a slope of 1). We explored model residuals [that is, lack of patterns when represented against predicted values and independent variables and lack of spatial autocorrelation (Mantel test, Legendre and Fortin 1989)] to finally validate the model. All analyses were performed using R (R Development Core Team, 2009) and the “likelihood” package (Murphy 2012).

Results

Seedling and Sapling Abundance

A total of 4514 seedlings and 1865 saplings of 21 woody species were counted during the three sampling years at the six study sites (Table B in Supplementary Material). Due to the scarcity of seedlings and saplings of dry-fruited shrubs, this group was only included in the analysis of seedling abundance in woodland sites. All models produced unbiased estimates of seedling and sapling abundance (slope~1, Table 1) and did not show spatial autocorrelation in their residuals with two exceptions (liana seedlings in woodlands and Q. suber saplings in closed forests, Table C in Supplementary Material). It is important to highlight that likelihood-based inference is very robust even in cases of lack of independence, with deviations from independence mostly resulting in underestimates of variance terms but not in biased parameters values. Both the additive and multiplicative approaches were used in best models of seedling abundance, whereas most of the best models of sapling abundance involved a multiplicative approach (Table 1). Most of the best models of seedling (7/9) and sapling (7/8) abundance showed differences among the three sites in regeneration abundance. All models detected tree neighborhood effects on seedling and sapling abundance, as indicated by significant λ values for one or more neighbor categories (Tables D and E in Supplementary Material). These effects did not vary among years (Table 1). Our models did not detect any effect of shrubs on seedling abundance and only in one species (Olea europaea) for sapling abundance (Table 1).

Table 1 Comparison of the Alternate Models for the Analysis of Seedling and Sapling Abundance, Survival, and Richness
Full size table

In general, seedling abundance was higher in healthy tree communities (as indicated by positive λ values) than in neighborhoods dominated by defoliated or dead trees (as indicated by neutral or negative λ values) (Table D in Supplementary Material, Figures 1A, C, E, G, and 2A, C, E, G). However, the magnitude of the neighborhood effect varied among species/functional groups, causing shifts in rankings of species abundance among neighborhood types. In woodlands, Olea was the dominant seedling species in healthy tree neighborhoods (both conspecific and heterospecific), whereas Q. suber dominated the seedling bank in defoliated neighborhoods and fleshy-fruited shrubs in dead neighborhoods (Figure 1). In closed forests, lianas and (to a lesser extent) fleshy-fruited shrubs were the dominant groups in all type of neighborhoods (Figure 2). Among tree species, Q. suber and Q. canariensis dominated their respective conspecific healthy neighborhoods, whereas the abundance of both species was virtually null in defoliated and dead Q. suber neighborhoods (Figure 2E, G). In most neighborhood models of seedling abundance (7/9), the exponent α of the Neighborhood Index (which controls how neighbor effects scale with tree size, Equation 3) took high values (>2), indicating that the effect of neighboring trees scaled roughly with biomass (Table D in Supplementary Material). Low values of the β parameter (which controls the effect of distance to neighbors, Equation 3) in most models indicated that neighbor effects on seedling abundance slightly decreased with distance (Table D in Supplementary Material).

Figure 1
figure 1

Predicted variation in the abundance of seedlings and saplings of different woody species (Quercus suber, Olea europaea, fleshly-fruited shrubs, dry-fruited shrubs, and lianas) as a function of the neighborhood index (NI) calculated for four different tree neighbor types [O. europaea (A, B), healthy Q. suber (C, D), defoliated Q. suber (E, F) and dead Q. suber (G, H)] in woodland sites. NI values of 0 represent neighborhoods without trees, whereas NI values of 1 represent, for each neighbor category, the neighborhoods with the highest observed value of NI (that is, the highest influence of that neighbor type). Predicted abundance of seedlings and saplings was calculated using the values of the parameters a, b, and λ in Equation 1 or 2, given in Supporting Information (Tables D and E in Supplementary Material).

Full size image
Figure 2
figure 2

Predicted variation in the abundance of seedlings and saplings of different woody species (Quercus suber, Quercus canariensis, fleshly-fruited shrubs, and lianas) as a function of the neighborhood index (NI) calculated for four different neighbor types [Q. canariensis (A, B), healthy Q. suber (C, D), defoliated Q. suber (E, F), and dead Q. suber (G, H)] in closed forest sites. NI values of 0 represent neighborhoods without trees, whereas NI values of 1 represent, for each neighbor category, the neighborhoods with the highest observed value of NI (that is, the highest influence of that neighbor type). Predicted abundance of seedlings and saplings was calculated using the values of the parameters a, b, and λ in Equation 1 or 2, given in Supporting Information (Tables D and E in Supplementary Material).

Full size image

Sapling abundance of tree species was also generally lower in defoliated and dead neighborhoods than in healthy ones (Table E in Supplementary Material; Figures 1B, D, F, H and 2B, D, F, H). However, abundance of fleshy-fruited shrub and liana saplings was not negatively affected by tree decline in most situations (as indicated by neutral or positive λ values; Table E in Supplementary Material). Shifts in rankings of species abundance among neighborhood types were also found in the sapling community. In woodland sites, whereas healthy neighborhoods of both Q. suber and O. europaea were dominated by conspecific saplings, the strong negative effects of defoliated and dead trees on Q. suber saplings (stronger than on any other species) caused these neighborhoods to be dominated by heterospecific O. europaea saplings (Figure 1f, h). In closed forests, the sapling bank kept being dominated by lianas and fleshy-fruited shrubs in all type of neighborhoods (with the exception of fleshy-fruited shrubs in Q. canariensis neighborhoods). Among tree species, Q. canariensis dominated the sapling bank in conspecific neighborhoods and was also slightly more abundant than Q. suber in healthy Q. suber neighborhoods. However, Q. suber dominated the sapling bank in defoliated and dead neighborhoods likely as a result of stronger negative effects of decline (more negative λ values) on Q. canariensis than on Q. suber sapling abundance (Table E in Supplementary Material). In models of sapling abundance, the value of the α parameter was highly variable among species (Table E in Supplementary Material), whereas the β parameter was generally close to 0 indicating that abundance was mostly proportional to neighbor density.

Seedling and Sapling Survival

All models produced unbiased estimates of tree seedling and sapling survival (slope ~1), except for O. europaea and Q. suber seedlings in woodlands and closed forests (respectively) where best models tended to overpredict (slope >1, Table 1). Model residuals did not show spatial autocorrelation, except for Q. canariensis saplings (Table C in Supplementary Material). In all cases, additive models were a better fit to the data than multiplicative models. Only 25 % (2/8) of the best survival models found differences among sites in seedling and sapling survival (Table 1). All models detected tree neighborhood effects on seedling and sapling survival, as indicated by significant λ values of one or more neighbor categories (Table F, Figures H and I in Supplementary Material). These tree effects remained mostly constant among years, with three exceptions (Q. suber seedlings and O. europaea saplings in woodlands, Q. suber saplings in closed forests; Table 1). Our models did not detect any effect of shrubs on seedling survival (Table 1).

In woodlands, neighbor effects on survival were of generally weak magnitude, as indicated by the low value of most significant λ (Table F in Supplementary Material). Moreover, the sign of the effects on Q. suber seedlings and Olea saplings varied among years (see sign of parameter b in Table F in Supplementary Material), with survival being higher in healthy than defoliated neighborhoods in 2010 (a very wet year) but lower in 2011 (a normal year in terms of rainfall). In closed forests, Q. canariensis neighborhoods had the most negative effect (lowest λ value, Table F in Supplementary Material) on seedling survival of the two oak species, followed by healthy Q. suber neighborhoods. Although seedling survival was lower in healthy than in declining Q. suber neighborhoods, sapling survival was higher (more positive λ values) in healthy neighborhoods. This was particularly true for Q. canariensis saplings, which suffered from strong negative effects of defoliated and dead neighbors (Table F in Supplementary Material). Values of parameters α and β in the Neighborhood Index (Equation 3) showed a very high variability among best models of seedling and sapling survival (Table F in Supplementary Material).

Seedling and Sapling Richness

The two richness models produced unbiased estimates of species richness (slope ~1), but showed some spatial autocorrelation in their residuals (Table C in Supplementary Material). In both cases, an additive approach offered a better fit to the data than a multiplicative approach (Table 1). In woodlands, the best model detected an effect of the tree community on richness values, but did not find any site, year, or shrub effects (Table 1). In this forest type, species richness was higher in healthy tree communities of Quercus and Olea (large positive λ values) than in neighborhoods dominated by defoliated and dead Q. suber trees (Table G, Figure J in Supplementary Material). In closed forests, the best model was the full model that considered the effect of all the factors tested. Thus, in this forest type, richness varied among sites was lower in Q. canariensis neighborhoods than in any Q. suber neighborhood (in all years but the dry 2012) and were positively affected by shrubs in the three study years (Table G, Figure J in Supplementary Material). Values of the α and β parameters in the Neighborhood Index of the two best richness models were roughly around 1, indicating that neighbor effects scaled with tree size (α) and declined as a simple inverse function of distance (β) (Table G in Supplementary Material).

Discussion

Our results indicated that the decline of Q. suber in Mediterranean forests had detectable effects on both new seedlings and older saplings of coexistent woody species belonging to very different functional groups (trees, shrubs and lianas). Thus, we found total and relative regeneration abundance (both within and among functional groups) to vary depending on the health status of the canopy cover. Our neighborhood approach provided a useful framework that integrates the myriad ways that canopy and seedling species interact under natural conditions helping in this way to improve our understanding of the consequences of tree decline and mortality for regeneration dynamics.

Patterns of Natural Regeneration in Healthy Tree Neighborhoods

Healthy neighborhoods of the three dominant tree species in the study forests (Q. suber, O. europaea, and Q. canariensis) showed a tree seedling and sapling bank composed largely by conspecifics. It is interesting to note that the positive relationship found between abundance of conspecifics in the canopy and the seedling/sapling bank was stronger for new seedlings than for saplings (Figure 1A, B for O. europaea, Figure 2A, B for Q. canariensis, Figure 2C, D for Q. suber). Stronger positive conspecific effects for seedlings than for saplings could indicate the existence of negative density-dependent processes from conspecifics acting at the seedling stage due to host-specialized antagonists (for example, Janzen-Connell effects; Connell 1971; Janzen 1970), intense competition for resources, or to the generation of unfavorable abiotic conditions (for example, Bonanomi and others 2008). Accordingly, we found negative effects of conspecific neighborhoods on seedling survival of the three tree species (negative λ values), in agreement with previous studies (Pérez-Ramos and Marañón 2012; Ibáñez and others 2015). Our results, however, suggest that the magnitude of these negative density-dependent processes on survival were not strong enough to counteract the high number of seedlings emerged in conspecific neighborhoods (Figures 1A, B, C, D and 2A, B, C, D). High emergence values under conspecific trees are likely the result of dispersal patterns, with most seeds not being biotically dispersed and arriving under mother trees in both Olea (Rey and Alcántara 2000) and Quercus species (Pulido and Díaz 2005). For tree species in mixed forests, the existence of a conspecific sapling bank under adult canopies might be considered as a guarantee of self-replacement that could favor species coexistence (Catovsky and Bazzaz 2002; Frelich and Reich 1995; Van Breemen and Finzi 1998; Wilson and Agnew 1992). Our results therefore suggest that, in the absence of disturbances, positive canopy-sapling feedbacks could lead to a rather stable stand structure. The process of Q. suber decline could, however, act as a destabilizing force of this structure through its impacts on the regeneration bank.

Effects of Quercus Suber Decline on Natural Regeneration

A clear consequence of Q. suber decline was a reduction in the absolute abundance of regeneration (both seedlings and saplings) in all tree species. The negative impact of Q. suber decline on tree species regeneration might explain the lower species richness found in declining neighborhoods at some sites (for example, in woodlands). However, the magnitude of this effect varied substantially among co-dominant tree species. In woodland sites, the decline of Q. suber had a stronger negative effect on conspecific regeneration than on O. europaea regeneration, causing a shift in sapling relative abundance towards the dominance of O. europaea. Such among-species difference in the magnitude of the negative effects of decline could be due to the high abundance of Phythopthora cinnamomi (a soil-borne pathogen with high preference for Q. suber) under declining trees in woodlands (Gómez-Aparicio and others 2012) or to the better capacity of drought-tolerant O. europaea seedlings to tolerate stressful abiotic conditions in gaps opened after tree death (for example, high light levels and evaporation demand, Fetcher and others 1985; Royer and others 2011). Abiotic conditions in gaps might be considered to be particularly stressful in woodlands compared to closed forests, since their open structure and smaller tree size determined the formation of larger gaps after the death of individual trees. In closed forests, on the contrary, the somewhat weaker negative effect that Q. suber decline had on conspecific regeneration compared to Q. canariensis regeneration caused a shift in sapling relative abundance toward a slight advantage in dominance for Q. suber. This result was quite unexpected since, in a parallel experimental study where seeds of Q. suber and Q. canariensis were sown and the recruits monitored, we did not detect clear differences in the response of new seedlings and young saplings (2–3 years old) of the two oak species to neighborhoods of contrasting health (Ibáñez and others 2015). Because the sapling bank in this study potentially included saplings much older than 3 years, it is possible that this result reflects the development of among-species differences with ontogeny, with older saplings of the shade-tolerant Q. canariensis being more impaired than those of Q. suber by the high light levels of the gaps open after tree death.

Whereas the decline of Q. suber had important negative effects on abundance and performance of tree seedlings and saplings, other functional groups such as fleshy-fruited shrubs and lianas were more independent of the decline process and even benefited from it in some situations. These findings could be related to the fact that seedling survival of both lianas and fleshy-fruited shrubs has been shown to benefit from high-light microsites in the forest understory (Pérez-Ramos and Marañón 2012), they are not reported to be affected by P. cinnamomi, and they are the dominant woody species recruiting in degraded forests (for example, post-fire shrublands and reforested stands, Mendoza and others 2009). Our results therefore suggest that the process of Q. suber decline could induce an increase in the relative abundance of functional forms other than trees in the woody community. This finding is in agreement with previous studies that have proposed a potential development of Mediterranean forests towards more open systems with lower tree cover and higher shrub dominance due to global environmental change (Acácio and others 2007; Matías and others 2012; Pérez-Ramos and Marañón 2012).

Implications for Successional Dynamics in Declining Forests

The effects of tree dieback on regeneration dynamics is probably one of the less studied aspects of forest die-off, particularly when compared with the abundance of studies exploring the physiological mechanisms underlying tree dieback or its impacts on ecosystem processes (McDowell and others 2013). Among the scarce regeneration studies available, a large majority has focused on regeneration patterns after insect-decline mortality in mountain forests (Battles and Fahey 2000; Collins and others 2011; DeRose and Long 2010; Kayes and Tinker 2012; Nigh and others 2008; Veblen and others 1991; Vyse and others 2009) and to a lesser extent after temperature- or drought-driven mortality in water-limited systems (Galiano and others 2013; Redmond and Barger 2013; Van Mantgem and Stephenson 2007). Here we show, for a complex system involving water-limited forests affected by aggressive exotic pathogens, that the decline of the dominant species (Q. suber) implies shifts in the ranking of seedling and sapling abundance among tree species, which could affect successional trajectories potentially leading to vegetation shifts (Kayes and Tinker 2012; but see Redmond and Barger 2013; Suarez and Kitzberger 2008).

Yet, predicting trajectories of forest recovery after tree dieback is not an easy task. For tree dieback to induce compositional shifts, it is necessary that the die-off species has a disadvantage in terms of propagule input or that the new conditions created are relatively more favorable for regeneration of coexisting species (Suarez and Kitzberger 2008). Here we found that in some forest types, like open woodlands, the die-off species Q. suber met both requisites, showing much lower seedling and sapling abundance than the co-dominant O. europaea in all sites (Table B in Supplementary Material) and a stronger negative response to declining neighborhoods (Figure 1F, H). Everything being equal, we could therefore expect these oak woodlands to change towards O. europaea- dominated systems, in agreement with our initial hypothesis of study. In closed forests, on the contrary, we found Q. suber to have much lower regeneration abundance than the co-dominant Q. canariensis at the site level (Table B in Supplementary Material), but to be somewhat less negatively affected by the process of adult decline. Therefore, the future dominance patterns in this forest type will likely depend on the extent to which the relative advantage of Q. suber over Q. canariensis saplings under defoliated and dead trees can compensate for the intrinsic lower regeneration capacity of the species and its increasing seed limitation as the process of adult decline advances.

Understanding recruitment dynamics in Q. suber forests is of major ecological and social importance given the prominent role that this tree species plays in the configuration of the landscape and the economy of the Mediterranean basin and the problems of regeneration and decline that Q. suber suffers throughout its distribution area (Aronson and others 2009; Urbieta and others 2008). Here we have shown that based on current patterns of woody plant regeneration in declining mixed Q. suber forests, we could expect these systems to suffer from important changes in relative abundance among tree species as well as among functional groups (trees, shrubs, lianas) in the near future. Because most of these changes point towards a loss of cover and dominance of Q. suber, management strategies are urgently needed in order to either attenuate adult mortality (for example, adequate phytosanitary measures) or promote its regeneration (for example, enrichment plantings). Only by looking for stabilizing mechanisms that improve the balance between mortality and recruitment (Lloret and others 2012) we will be able to counteract the negative effects of global change drivers (exotic pathogens, climate change) on Q. suber, and preserve the many ecosystem services provided by these valuable forests.