Volume 109, Issue 9 p. 3233-3245
Research Article
Free Access

Demographic traits improve predictions of spatiotemporal changes in community resilience to drought

Maria Paniw

Corresponding Author

Maria Paniw

Center for Ecological Research and Forestry Applications (CREAF), Cerdanyola del Vallès, Spain

Department of Conservation Biology, Estación Biológica de Doñana (EBD-CSIC), Seville, Spain


Maria Paniw

Email: [email protected]

Search for more papers by this author
Enrique G. de la Riva

Enrique G. de la Riva

Department of Ecology, Brandenburg University of Technology, Cottbus, Germany

Search for more papers by this author
Francisco Lloret

Francisco Lloret

Center for Ecological Research and Forestry Applications (CREAF), Cerdanyola del Vallès, Spain

Unitat Ecologia, Dept. Biologia Animal, Biologia Vegetal i Ecologia, Universitat Autònoma Barcelona, Cerdanyola del Vallès, Spain

Search for more papers by this author
First published: 26 January 2021
Citations: 4

Handling Editor: Iain Stott



  1. Communities are increasingly threatened by extreme weather events. The cumulative effects of such events are typically investigated by assessing community resilience, that is, the extent to which affected communities can achieve pre-event states. However, a mechanistic understanding of the processes underlying resilience is frequently lacking and requires linking various measures of resilience to demographic responses within natural communities.
  2. Using 13 years of data from a shrub community that experienced a severe drought in 2005, we use generalized additive models to investigate temporal changes in three measures of resilience. We assess whether community-weighted, species-specific demographic traits such as longevity and reproductive output can predict changes in resilience and how the performance of these traits compares to more commonly used plant functional traits such as leaf area or root dry matter content.
  3. We find significant spatiotemporal variation in community dynamics after the drought. Overall, demographic traits are better at predicting absolute and relative resilience in total plant cover, but functional traits outperform demographic traits when resilience in community composition is assessed. All resilience measures show nonlinear and context-specific responses to demographic and functional traits; and these responses depend strongly on the severity of initial drought impact.
  4. Synthesis. Our work demonstrates that a full picture of the mechanisms underlying community responses to drought requires the assessment of numerous species-specific characteristics (including demographic and functional traits) and how these characteristics differentially affect complementary measures of community changes through time.



  1. Los episodios climáticos extremos tienen impactos importantes en las comunidades ecológicas. La resiliencia, considerada como la capacidad de las comunidades para recuperar su estado anterior, refleja el efecto acumulado de dichos eventos. Sin embargo, desconocemos con detalle los mecanismos demográficos que determinan dicha resiliencia.
  2. A partir de 13 años de datos de una comunidad de matorral que experimentó una sequía severa en 2005, hemos aplicado modelos aditivos generalizados para investigar cambios temporales en tres medidas de resiliencia de la comunidad. Hemos evaluado si la abundancia ponderada de rasgos demográficos, como longevidad y éxito reproductivo, predicen cambios en la resiliencia, y si esta capacidad explicativa difiere de la proporcionada por rasgos funcionales, como área foliar o contenido de materia seca radicular.
  3. Analizando la variación espacio-temporal de la dinámica post-sequía de la comunidad, encontramos que los rasgos demográficos predicen mejor la resiliencia absoluta y relativa de la cobertura vegetal, pero los rasgos funcionales superan a los demográficos al evaluar la resiliencia en la composición. Todas las medidas de resiliencia muestran respuestas no lineales y específicas para rasgos demográficos y funcionales, y están determinadas por la afectación inicial.
  4. Síntesis. El estudio demuestra la necesidad de considerar diferentes tipos de rasgos (demográficos y funcionales) para comprender las respuestas de la comunidad vegetales a episodios de sequía extrema, ya que tienen distintos efectos en las diferentes medidas que describen la dinámica de las comunidades.


Under global environmental change, extreme weather events—such as flooding, drought, heatwaves or windstorms, that are outside the normal range of weather variation in a system—are on the rise (Smith, 2011). These events, particularly when they occur simultaneously and repeatedly, can have devastating compound effects on natural plant communities (Smith, 2011; Zscheischler et al., 2018). Eventually, extreme events that cause elevated mortality, especially of the dominant vegetation, may lead to vegetation shifts altering community structure and function (Breshears et al., 2005; Miriti et al., 2007).

The cumulative effects of repeated extreme episodes or disturbances on community composition are typically assessed via changes in community resilience (MacGillivray & Grime, 1995). The concept of resilience broadly refers to the capacity of ecological systems to absorb disturbances and yet maintain structure and function (Walker et al., 2004). More specifically, resilience is often considered as the capacity to achieve pre-disturbance levels of structure and function, also called ‘engineering resilience’ (sensu Pimm, 1984; see Holling, 1996 and Gunderson, 2000; Scheffer et al., 2001). This estimation of resilience is commonly based on direct comparisons between pre- and post-disturbance levels of the system properties (hereafter, absolute resilience, e.g. Lloret et al., 2011) and simply shows post-disturbance levels irrespective of disturbance severity. However, the mechanisms underlying resilience are better addressed by weighting absolute resilience by the damage a system incurs during the extreme event (hereafter, relative resilience, sensu Lloret et al., 2011). An alternative concept of resilience, also called ‘ecological resilience’ (Holling, 1996; see also Angeler & Allen, 2016), is more focused on system dynamics and refers to the capacity of a system to absorb changes in state variables, remaining in a dynamical equilibrium until reaching a tipping point that leads to a new equilibrium state. For this approach, early warning signals advising on the proximity to tipping points represent a key concept for both theoretical and applied ecology (Scheffer et al., 2009). Although ‘engineering’ and ‘ecological’ resilience approaches are addressed unconnected in most studies, slow recovery rate after disturbance (critical slowing down) can be used as an indicator of how close a system is to a tipping point (Clements & Ozgul, 2018; Scheffer et al., 2009), thus representing a common indicator of resilience in a broad sense (Walker et al., 2004).

Community resilience is mediated by stabilizing processes which are ultimately determined by the demography of the interacting species (Lloret et al., 2012; McGill et al., 2006). That is, the effects of an extreme event on the entire community via increased mortality of individual components can be buffered via subsequent increases in recruitment or survival, thereby conferring resilience to a community (Bréda et al., 2006; Koepke et al., 2010). Compensatory effects therefore depend on population-level demographic traits, such as longevity or lifetime reproductive output, which reflect differential investment in survival and reproduction and characterize the life-history strategies of the interacting species under environmental variation (Paniw et al., 2018; Salguero-Gómez et al., 2016). For instance, communities dominated by long-lived species are generally resistant to or recover fast from extreme events because long-lived organisms have evolved strategies to buffer from environmental fluctuations (Morris et al., 2008) and may show high genetic diversity (Kelly et al., 2003) or phenotypic plasticity (Clark, 2010). Similarly, species with a high reproductive output can maximize population fitness by timing mass recruitment to follow disturbances; this may then allow quick recovery resulting from increased recruitment success (Metcalf et al., 2009). Given the importance of demographic traits in mediating community responses to extreme events, incorporating such traits in the analyses of resilience would provide important mechanistic insights.

While species-specific life-history or demographic information has increasingly been incorporated into early warning signal frameworks (reviewed in Clements & Ozgul, 2018), current approaches to predict engineering resilience rarely account for such information. However, they frequently incorporate functional traits, that is, plant-level structures and stoichiometric properties reflecting balances of resource allocation with consequences for plant fitness (de la Riva et al., 2017). Linking resilience to these functional traits assumes that such traits regulate the responses of plant communities to extreme events. Several studies have shown that functional traits linked to the economics spectrum such as tissue dry matter content or water use efficiency (Wright et al., 2004) may be important predictors of community stability particularly after extreme events related to water scarcity (de la Riva et al., 2017; Griffin-Nolan et al., 2019; Pérez-Ramos et al., 2017). Despite the unquestionable value of studies analysing the role of functional traits in promoting community stability, inference can be limited when functional attributes do not address the direct mechanisms allowing populations to recover from extreme environmental fluctuations (Lloret et al., 2016). For numerous species, functional traits are not well-linked to demographic performance (Bohner & Diez, 2020; Paine et al., 2015; Yang et al., 2018), which ultimately explains population and thereby community recovery after extreme events. It is not clear to what extent functional traits can be used in lieu of direct demographic information to predict resilience.

Apart from determining appropriate predictors of community resilience, another important challenge when analysing resilience is capturing the temporal nature of the phenomenon. Approaches often rely on time-invariant measures of resilience, comparing pre- and post-disturbance community composition at a single point in time across different sites (e.g. Breshears et al., 2005; Lloret et al., 2011). Although they have been performed for disturbances such as wildfires (Trichon et al., 2018), time-variant investigations of resilience are particularly rare for studies measuring the effects of extreme drought events (but see Trichon et al., 2018). However, community responses to extreme events operate at complex spatiotemporal scales (Lines et al., 2010) driven by inter-annual environmental fluctuations (including abiotic and biotic factors) and their consequences on demographic processes (Ne'eman et al., 2004; Zedler et al., 1983). In fact, declines in resilience over time may indicate an approaching collapse of the system (Scheffer et al., 2009). It is therefore crucial to account for temporal trends in community composition (Hollister et al., 2005).

Here, we investigate the demographic mechanisms underlying spatiotemporal changes in resilience of a shrubland community following an extreme drought event, which occurred in 2004–2005 in the Doñana National Park (Spain). We parameterized three measures of resilience analysing data collected along 12 years after the event. We calculated absolute and relative resilience in total vegetation cover and resilience in vegetation composition and linked these three metrics to four demographic traits weighted at the community level: age at first reproduction, longevity, minimum size at first reproduction and ratio of recruits/adults. We also assessed whether these demographic traits explained more variance in resilience than 10 community-weighted morphological and physiological traits (de la Riva et al., 2017; Table 1).

TABLE 1. Summary of demographic and functional traits used to model variation in shrub resilience after drought
Trait Abbreviation Unit Relation to resilience
Demographic traits
Age at first reproduction AgeFR years Approximates pace of life of an organism, which determines responses to environmental perturbations (Paniw et al., 2018)
Maximum longevity L years Approximates pace of life of an organism
Minimum size at first reproduction SizeFR cm Approximates capacity to delay reproduction to improve reproductive success or future reproduction (Sletvold, 2002)
Ratio recruits to adults R/A Describes the success of any given reproductive event
Functional traits
Specific leaf area SLA m2/kg Light capture and growth rate
Specific root area SRA m2/kg Efficiency in water and nutrient acquisition
Leaf dry matter content LDMC mg/g Physical resistance and stress tolerance
Root dry matter content RDMC mg/g Physical resistance and stress tolerance
Stem dry matter content SDMC mg/g Resistance to physical stress
Leaf nitrogen concentration LNC % Light capture and photosynthetic rate
Leaf chlorophyll LChl μg/g Light capture and photosynthetic rate
Isotopic carbon fraction δ13C Gas exchange and water use efficiency
Plant height Phg m Light capture, above-ground competition, colonization capability
Seed mass Smass mg Dispersal and colonization capability


2.1 Study area

The study area is located in Doñana National Park in southwest Spain. The area is characterized by a Mediterranean-type climate (for details on site-specific climate, see de la Riva et al., 2017) and is extremely vulnerable to predicted increases in temperature and rainfall scarcity (Sánchez et al., 2004). Vegetation in the study area is dominated by extensive shrublands, with locally dense stands of juniper (Juniperus phoenicea ssp. turbinata) mixed with stone pine (Pinus pinea). The study sites were located in a shrubland locally known as ‘monte blanco’, with dominant shrub species being Cistus libanotis, Halimium halimifolium, Helichrysum pichardii, Lavandula stoechas, Rosmarinus officinalis, Thymus mastichina and Ulex australis (García Murillo & Sousa Martín, 1999). As the biomass of herbaceous species is marginal, the latter were not considered in this study.

The shrublands experienced a drought in the 2004–2005 hydrological year, which, due to the statistical rarity of the event and the significant effect on shrub communities, was considered an extreme climatic event (Smith, 2011). In that year, total annual rainfall was 170 mm, which constituted the lowest record measured 1978–2019 (Figure S1). Low rainfall was combined with an episode of extreme cold, where minimum temperatures were lower than 95% of the values registered in the last 40 years (Figure S1). The combined effect of the rare drought and cold events caused a die-off of green tissue on the dominant shrub vegetation that reached 75% of the plant cover in some stands and changed species composition and the functioning of plant–soil interactions (Lloret et al., 2015). In the years following the drought, extreme weather conditions did not occur.

2.2 Quantifying community composition

In order to assess community composition before, during and after the drought, 18 permanent plots of 25 m2 (5 × 5 m) were established in November 2007 (2 years after the drought) on a gradient of drought impact (for details see Lloret et al., 2016; de la Riva et al., 2017). The plots were located at three sites (with six plots per site): Raposo (N 37°0′2″, W 6°30′20″; at 18 m a.s.l.), Marqués (N 37°0′45″, W 6°31′50″; 21 m a.s.l) and Ojillo (N 36°59′40″, W 6°30′50″; 30 m a.s.l.). To avoid spatial autocorrelation, all plots were separated by at least 50 m from each other.

Species plant cover was estimated from contacts with branches along transects within plots; these contacts were divided into two categories corresponding to living or dead canopy, as detailed in Lloret et al. (2016). Dry organs with signs of old decay (stumps, decomposed stems, branches without thin tips) were excluded. This procedure allowed us to calculate an estimator of plant cover suitable to compare canopy state before and after the drought episode. Thus, canopy prior to the episode was considered as the sum of living and dead (i.e. dry) plant canopy in 2007. Characterizing prior canopy cover in this way was supported because plants considered as dead were recently defoliated, as indicated by the presence of non-decomposed leaves on the ground beneath plant canopies. Also, the growth of new shoots and leaves between 2005 and 2007 should be considered as irrelevant to our sampling, which focused on branches. See Lloret et al. (2017) and de la Riva et al. (2017) for further discussion on the representativeness of this procedure to obtain a measure of the impact of the climatic episode on vegetation cover. Dead individuals were identified to the species level by their morphological traits (bark, branch ramification, dry leaves on the ground). Relative abundance of each species per plot in years after the extreme drought was calculated as the proportion of their contacts of living canopy relative to the sum of the contacts of living canopy of all species. Similarly, the total vegetation cover per plot was calculated as the summed contacts of living canopy of all species. Relative abundances previous to the drought were calculated as the sum of the living and dead canopy of 2007 of each species relative to the sum of the living and dead canopy of 2007 for all species.

2.3 Resilience in cover

We calculated absolute (Rs) and relative resilience (RRs) of total plant cover as described in Lloret et al. (2011). Values of Rs and RRs were calculated comparing all five post-drought sampling times t (2008, 2010, 2013, 2017 and 2019) to pre-drought (2005) and drought (2007) state, that is, Rst = (post-drought covert)/(pre-drought cover). Weighting this resilience by the damage occurred during the drought,

High values of relative resilience indicate a high capacity to recover from or compensate for the particular effects of the drought event (Sánchez-Pinillos et al., 2019).

2.4 Resilience in composition

In order to investigate whether the drought altered the relative abundance of species within each plot, we analysed changes in vegetation taxonomic composition in time using a multidimensional distance-based analysis. First, we summarized the general species composition by a non-metric multidimensional scaling (NMDS) ordination (Kruskal, 1964). We performed NMDS on a matrix with species relative abundance as rows and unique plot-year sampling units as columns in the r package vegan (Oksanen et al., 2013). We used the Bray–Curtis dissimilarity measure and default settings, except that the number of iterations was increased to 100 and three ordination dimensions, instead of two, were used (due to a lower stress score; 0.17 versus 0.25). To characterize resilience in composition, we then calculated the distance in ordination space from the coordinates of each post-drought sampling in each site to coordinates of the respective pre-drought vegetation composition, using the dist function in r. From this distance, we calculated relative resilience in composition change (RRc), a measure of the system's elasticity to change, as:
where Dpre−t is the distance in NMDS space between the post-drought year t sampling and the pre-drought vegetation estimation; and Dpre−dist is the distance in NMDS space between the disturbed state, which in our case corresponds to the 2007 sampling and the pre-drought vegetation estimation (Sánchez-Pinillos et al., 2019).

2.5 Demographic traits

We obtained four demographic traits for each plant species encountered in the sampling plots (Table 1). These traits are often measured when plant communities are surveyed, can potentially characterize a wide range of life histories, but are not typically included in analyses of resilience. The traits included age at first reproduction (AgeFR), maximum longevity (L), minimum size at first reproduction (SizeFR) and ratio of adults to recruits (R/A). To calculate the first three traits for each species, 50 plants growing near the plots were randomly selected in June, 2019, in vegetation patches without signs of drought-induced impact. These plants covered the whole range of sizes of the population. For each plant, we recorded size (height, mean canopy diameter) and reproductive stage (flowering versus. non-flowering). We estimated plant age from yearly growth scars in the main shoots. We calculated age at first reproduction from inverse prediction (Probability of not flowering = 0.9) obtained from the logistic nominal regression between reproduction stage and plant age following the standard procedure in JMP v. 10.0.0 (SAS Institute Inc., Cary, NC, 1989-2019). We obtained minimum size at first reproduction using receiver operating characteristic (ROC) curves, which plot true-positive against false-positive rates at various thresholds. Minimum size at flowering was then determined as the threshold that maximizes true-positive rates, that is, where the area under the ROC curve is highest (Fawcett, 2006). Lastly, we approximated maximum longevity as the 90th percentile of estimated age from yearly growth scales in the 50 plants growing near the plots in patches not impacted by drought; since scars disappear in older plants, 5–10 years' age intervals—depending on species' growth form and size—were added to the older scars in larger plants. For the large shrubs E. arborea and the small tree J. phoenicea subsp. turbinata, maximal age was approximated according to the estimated age of the shrubland turnover before burning or replacement by forest (Granados Corona, 1987). We note that we used the best available data for longevity estimates, but the accuracy of these estimates would be improved with the availability of long-term demographic data on the shrub species.

We calculated ratios of recruits to adults (R/A) from data collected in the years 2007, 2008, 2013 and 2019. Data included the total number of juvenile non-reproductive plants (recruits) and the total number of adults of each species found in plots with low drought impact (six plots). The number of plants of each type (recruits, adults) of these plots were merged, and the ratio between recruits and adults was calculated for each year. The final value of the ratio for each species was averaged from the ratio values obtained from the 4 years. The recruits/adult ratio for Erica scoparia was estimated from literature (Arévalo & Fernández-Palacios, 2003) because the number of plants (n = 4) was too low in the plots.

We calculated community-weighted means (Garnier et al., 2004) of each trait for each plot and year of sampling as ∑Pi × traiti, where Pi is the relative abundance of species i in the plot and traiti is the average trait value obtained for i.

2.6 Functional traits

To compare the performance of demographic traits to functional (morphological and physiological) traits in explaining spatiotemporal variance in resilience, we calculated community-weighted means of eight above- and two below-ground traits, which are among the most commonly used to describe plant functional types and have previously been obtained in the study plots in late spring 2013 (for details on these traits and how they were measured, see de la Riva et al., 2017; Table 1). Above-ground functional traits included plant height (Phg), leaf area per unit of leaf dry mass (SLA), leaf dry matter content (LDMC), stem dry matter content (SDMC), leaf nitrogen concentration (LNC), leaf chlorophyll concentration (LChl), isotopic carbon fraction (δ13C) and seed mass (Smass). Below-ground traits included specific root area (SRA) and root dry matter content (RDMC).

We calculated community-weighted means of functional traits as for demographic ones.

2.7 Principal component analysis

Due to inter-correlation among traits, we used a principal component analysis (PCA) to quantify main patterns in the variation of community-weighted demographic and functional traits among community sampling units (i.e. plot-year combinations). We performed the PCA on demographic and functional traits separately. Results from the PCA then served as predictors of community resilience, in terms of total plant cover and community composition, to drought. PCA reduces the dimensionality of data by quantifying the preponderant axes that are necessary to explain a given percentage of variation of the data. For the PCA, we rescaled the trait data to μ = 0 and SD = 1. We then implemented the PCA using the prcomp function in r and used the Kaiser criterion (Legendre & Legendre, 2012) to keep only principal component axes with associated eigenvalues >1. Lastly, we performed a variance maximizing rotation of the significant PCA axes using principal function in the r package psych (Revelle, 2015).

We note that some demographic and functional traits are strongly correlated (Figures S2 and S3), and because both types are related to plant fitness (Polley et al., 2013), their distinction can be difficult. For instance, seed mass, which is often included in plant functional studies (Díaz et al., 2016), also informs on important demographic parameters such as dispersal and seedling survival (Hulme, 1998). Similarly, demographic traits can be mediated by the taxonomic and functional diversity of the interacting species (Lloret et al., 2007). Therefore, to account for potential bias in categorizing traits as demographic or functional, we repeated all modelling using each of the 14 traits (Table 1) as predictors instead of PCA axes (Figures S5–S7; Table S3).

2.8 Statistical modelling

To gain a mechanistic understanding of the processes affecting community responses to extreme events, we analysed the effects of time and the main axes of trait variation (PCAs including demographic and/or functional traits) on the three measures of resilience, that is, absolute (Rs) and relative (RRs) resilience in total plant cover and relative resilience (RRc) in community composition. For Rs and RRs, we also considered the effect of composition change, as obtained from the distance measurements in NMDS space described in ‘Resilience in composition’. For RRc, we did not consider composition change, as the latter was used to calculate RRc. Instead, we considered initial drought impact, that is, percent cover loss for each sampling plot immediately following the drought, as surveyed in 2007, as an additional covariate. We modelled resilience using generalized additive models (GAMs) to account for nonlinear interactions (Wood, 2006). We considered additive effects of drivers as well as all possible two-way interactions. We included site as a random effect on the mean in all models. We ran models including PCA results where both demographic and functional traits were considered, or PCA results where only demographic or only functional traits were used (Figure 1). Selection of the most parsimonious GAM structure was based on Akaike's information criterion (Akaike, 1998). Specifically, we considered the most parsimonious GAM explaining variation in resilience to be the one with the lowest AIC value. If ΔAIC of two models was <2, we considered the model with fewer effective degrees of freedom to be the most parsimonious.

Details are in the caption following the image
Main axes of variation in demographic and functional traits among communities. To derive the main axes, a PCA was performed on demographic (a) or functional (b) traits in each plot-year combination (points, n = 90). Demographic traits include longevity (L), age at first reproduction (AgeFR), size at first flowering (SizeFR) and ratio of recruits to adults (R/A). Functional traits include plant height (Phg), specific leaf area (SLA), leaf dry matter content (LDMC), stem dry matter content (SDMC), leaf nitrogen concentration (LNC), leaf chlorophyll concentration (LChl), isotopic carbon fraction (δ13C), seed mass (Smass), specific root area (SRA) and root dry matter content (RDMC). Arrow lengths are proportional to the loadings of each trait onto the two axes. Point sizes are proportional to the initial drought impact (% cover loss). Yellow panels summarize main trait relationships, or trade-offs, represented by each axis (Repro. str.—reproductive strategy; LES—leaf economics spectrum)

In parameterizing the GAMs, we used tensor product smooth terms (te), with thin-plate regression splines as the marginal bases for the effects of all covariates with the exception of site. To model site effects, we used parametric terms (s) penalized by a ridge penalty as base. When applying smooths, we used three-knot locations for each covariate to avoid overfitting due to overly flexible smooths. We used the r package mgcv to fit GAMs (Wood, 2006), and the package MuMIn (Barton, 2009) for model selection.

We performed all statistical modelling in r version 4.0.0. Data and code to perform and plot the PCA (pca_plots.R) and obtain the three measures of resilience and model them as a function of PCA results (resilience_analysis_PCA.R) and individual traits (resilience_analysis_traits.R), as well as visualizing the relationships between individual traits and resilience measures (plot_resilience_traits.R) are available in the Supporting Material.


3.1 Patterns of demographic and functional traits

Variation in average demographic traits across communities and years was adequately captured by two main PCA axes (eigenvalues > 1) that together explained 87% of variation in community traits (Figure 1a; Table S1). PCA 1 captured variation in the community-averaged plant pace of life, with highest positive loadings of longevity (L) and age at first reproduction (AgeFR), suggesting a slow pace of life. PCA 2 captured variation in the reproductive strategy ranging from communities dominated by species with low-output strategies (high SizeFR and low R/A ratios indicating a conservative strategy to maximize offspring survival or future reproduction) to ones dominated by species with a high reproductive output (suggested by high R/A ratios) which can come at a cost for offspring survival or future reproduction if it occurs at small sizes (Sletvold, 2002) (Figure 1b; Table S2).

In the PCA performed with functional traits, three main axes with eigenvalues > 1 explained 79% of the trait variation (Figure 1b). Here, PCA 1 captured variation in plant resistance mechanisms and root resource uptake strategies. Meanwhile, PCA 2 captured the trade-off between resource acquisition and conservation at leaf level (leaf economics spectrum) with highest positive loadings of traits suggesting a conservative resource uptake use strategy (LDMC, δ13C) and high competitive ability (Phg, Smass); and negative loading of SLA suggesting an efficient leaf acquisitive strategy (SLA) (Figure 1b). PCA 3 captured variation along the leaf life span strategies from communities dominated by semi-deciduous species (high SLA) to communities dominated by species with higher resistance and longer leaf life span with higher metabolic secondary compounds (high N) (Figure S4; Table S3).

3.2 Predictors of resilience

Demographic traits, characterized by the two main axes in PCA space, outperformed functional traits as predictors of relative resilience (RRs) in total cover (Table 2). This was also true for absolute resilience (Rs), but here, including the PCA constructed from functional traits explained only slightly less variance (45% of model deviance) than the PCA from demographic traits (49% of deviance explained; Table 2). When considering single traits, SizeFR was included in the most parsimonious model of Rs, and all demographic traits were in the top models explaining RRs (Figures S5–S7; Table S3). Rs changed relatively little with time but was highest in plots that changed their composition little and were dominated by plants that had a fast to intermediate pace of life and showed a conservative reproductive strategy (i.e. marked by low output) (Figure 2). At the same time, plots dominated by plants with the aforementioned combinations of life-history strategies showed the overall lowest Rs when their composition change had been large (Figure 2). RRs changed more substantially with time and was highest in communities dominated by species with a slow pace of life and a conservative reproductive strategy (Figure 3).

TABLE 2. Most parsimonious GAM models for resilience. CompoChange—compositional change measured as distance in NMDS space; PCA—scores from PCA performed on demographic (D) or functional (F) traits; Impact—initial impact of drought (% cover loss); time—years; site—one of the three study sites (random effect in GAMs). Functions te(xdf) and s(xdf) are the tensor product (for fixed effects) and spline smoothing functions (for the random site effect) of x, respectively, with the given degrees of freedom df The df represents the amount of nonlinearity in the model component, where df = 1 indicates linear fit. In all models, sample size = 90. Dev shows the % deviance explained by each model
Resilience measure Best modela Dev (%)
Absolute resilience in total cover (Rs) te(CompoChange, PCAD 1df:3.9) + te(CompoChange, PCAD 2df:1.0) + te(PCAD 1, PCAD 2df:1.2) + s(sitedf:1.5) + te(timedf:1.8) 49.1
Relative resilience in total cover (RRs) te(PCAD 1df:1.8) + te(PCAD 2df:1.9) + te(timedf:1.9) + s(sitedf:1.8) 52.1
Relative resilience in composition change (RRc) te(Impact, PCAF 1df:6.5) + te(Impact, PCAF 2df:3.3) + te(Impact, PCAF 3df:1.7) + te(PCAF 2, PCAF 3df:2.4) + te(time, Impactdf:2.7) + te(time, PCAF 1df:2.1) + te(time, PCAF 2df:1.0) + te(time, PCAF 3df:1.0) + s(sitedf:1.9) 89.4
  • a The full models for the three measures of resilience included as fixed effects CompoChange (or Impact), all significant PCA axes, time and two-way interactions of these fixed effects; and site as a random effect.
Details are in the caption following the image
Absolute resilience (Rs) in total cover across time modelled as a function of composition change of the communities and two main PCA axes describing demographic-trait variation: pace of life (columns) and reproductive [rep.] strategy (rows). Lines and shaded areas show mean and SE, respectively, of predictions from the most parsimonious model describing variation in Rs. For better visualization, continuous distributions of PCA scores and compositional change were fixed at their 10th, 50th and 90th percentiles to obtain predictions
Details are in the caption following the image
Relative resilience (RRs) in total cover across time modelled as a function of two main PCA axes describing demographic-trait variation: pace of life (columns) and reproductive strategy (colours). Lines and shaded areas show mean and SE, respectively, of predictions from the most parsimonious model describing variation in RRs. For better visualization, continuous distributions of PCA scores were fixed at their 10th, 50th and 90th percentiles to obtain predictions

In contrast to resilience in total cover, the most parsimonious model of relative resilience in composition, RRc, included the first three main axes in PCA space characterizing the variation in functional traits (Table 2). Demographic traits, either jointly on the PCA or as single traits explained less variation in RRc (Figures S5–S7; Table S3). On the one hand, RRc increased slightly with time in plots with intermediate initial drought impact, especially where community-weighted below-ground resource acquisition strategies were least conservative, and physical resistance (i.e. high RDMC/SDMC and low SRA) of plants was high but plants showed a low competitive ability (i.e. low Phg/Smass). On the other hand, RRc decreased with time in plots with low initial drought impact, and was lowest in communities with a high relative abundance of plants with a conservative above-ground resource uptake strategy (Figure 4).

Details are in the caption following the image
Resilience in composition change (RRc) as a function of two main PCA axes that describe functional-trait variation across years and initial drought impact in the communities. The third significant axis (characterizing leaf life span) was held constant. The two axes depict the physical resistance strategy (columns) and the leaf economics spectrum (LES; rows), which approximates the resource acquisition strategy. Increasing values along the resistance axis indicate increasing functional resistance to drought, while increasing values along the LES axis indicate a conservative resource uptake strategy. Lines and shaded areas show mean and SE, respectively, of predictions from the most parsimonious model describing variation in RRc. For better visualization, continuous distributions of PCA scores and drought impact were fixed at their 10th, 50th and 90th percentiles to obtain predictions


Under a given environmental context, dynamics within a natural community are primarily determined by species-specific demographic strategies (i.e. relative rates of survival and reproduction) (Moritz & Agudo, 2013; Rüger et al., 2020). The relative frequency of demographic traits that characterize demographic strategies has however been rarely used to assess possible drivers of community resilience in studies of extreme drought events (Lloret et al., 2012). Instead, relative frequencies of species-specific functional traits within communities are typically used to predict resilience (Enright et al., 2014; Sakschewski et al., 2016), assuming that such traits inform on demographic responses (Salguero-Gómez et al., 2018). Our work demonstrates that different traits explain different resilience measures and that functional traits cannot substitute demographic traits when assessing resilience to drought in a Mediterranean shrubland. Instead, both demographic and functional traits explain different components and thus provide a complementary picture of changes in community resilience through time.

In our analyses, two PCA axes of community-weighted demographic traits explained more variation in the absolute and relative resilience of total plant cover than functional traits. This is likely because plant cover is a dynamic variable that responds fast to plant growth, particularly in early stages after disturbances when plants are released from competition and resources are less limiting. In turn, plant growth under these conditions is mediated by demographic processes through the establishment of new or the growth of surviving individuals (e.g. Hérault & Piponiot, 2018; Kayal et al., 2018). Highest resilience in plant cover can then be achieved if a low number of new species arrive into a community after drought (i.e. the community composition changes little), and these species invest into fast growth at the expense of reproduction (as represented by PCA 2 in Figure 1b). This is consistent with the well-known role of fast-growing seeder species in the resilience of shrublands after disturbances; these species are a key component of communities after wildfires in this Mediterranean region and they have also been reported to increase after drought-induced mortality in shrublands of the region (Del Cacho & Lloret, 2012). At the same time, when compositional change in a community is high, the abundance of species with a slow pace of life (as represented by PCA 1 in Figure 1a) may gain an increasing role in determining resilience. Long-lived shrub species, often corresponding to late-successional species, are stochastically expected to arrive when more new species get established in a given plot, and they tend to acquire large size, thus dominating community cover. In summary, resilience in plant cover can be achieved by a gradient of demography-defined species performance ranging from fast-growing species contributing to plant cover by many relatively small plants to long-living species contributing to plant cover by accumulating canopy belonging to fewer individuals (Belsky, 1986).

While demographic traits may improve predictions of resilience at the level of plant cover, they were outperformed in our analyses by functional traits when predicting resilience in community composition. Community composition changes smoothly through time as a result of assembly processes via successive filters. The first filter, determined by species colonization, can be considered finished at the spatiotemporal scale of our study, since the pool of colonizing species at landscape level is stable and similarly available in the different plots we surveyed. This pool of available species has been filtered by the historical climatic conditions of the location. For instance, low winter temperatures and dry summers are the most important climatic stressor of these Mediterranean communities (Pérez-Ramos et al., 2017), which reduces the range of functional traits to the set of species that contains specific morphological and physiological features to tolerate such abiotic constraints (de la Riva et al., 2018). Demographic traits, in contrast, are often highly context-dependent and may not reflect well historic conditions (Coutts et al., 2016). Therefore, changes in composition, and thereby resilience to such changes, are more easily correlated with plant functional traits.

Large differences in the resilience of composition among plots revealed the coexistence of different functional strategies in this community. Plots with highest resilience to changes in composition, in particular under high initial drought impact, were dominated by semi-deciduous species (e.g. Cistus libanotis or Rosmarinus officinalis) with a more acquisitive resource-use strategy. These semi-deciduous species have different demographic strategies but are functionally similar, that is, they are able to massively shed their leaves under environmental stress, remaining dormant or physiologically inactive during stressful drought periods (Grant et al., 2015; Zunzunegui et al., 2005). This strategy allows the individuals to remain alive in the plots, recovering faster when the environmental conditions are more auspicious, which may explain why communities dominated by these plants change little in composition (Ouédraogo et al., 2013). In contrast, species with traits related to a conservative resource-use strategy (i.e. higher values of tissue dry matter content and δ13C) and higher seed mass typically dominate shrub communities over time in the absence of severe disturbances (here, when drought impact is low). For instance, higher seed mass produces larger seedlings that are more robust and better able to form deeper and more extensive roots to capture more soil water (Pérez-Ramos et al., 2010; Quero et al., 2007); large seeds therefore improve the chances to successfully establish (Muller-Landau, 2010). However, under persisting adverse environmental conditions, green canopy in plots dominated by more conservative species may not fully be attained several years after the die-off episode, explaining the decreasing trend in the relative measure of resilience provided by RRc (Lloret et al., 2016; de la Riva et al., 2017).

In addition to the importance of analysing resilience on different community properties (cover and taxonomic composition) to gain a complementary picture of post-disturbance community dynamics, our results suggest that addressing the temporal scale of these dynamics is equally important. Because recovery after disturbance is not immediate, resilience is an inherently temporal phenomenon (Thurm et al., 2016). In addition to the changes associated with the trend to full recovery or alternatively to community shift, temporal variability in resilience may be explained by fluctuations in environmental conditions, including weather (e.g. Bêche & Resh, 2007) and biotic interactions such as competition (e.g. Vogel et al., 2012), herbivory or pests/pathogens (Schädler et al., 2003). In our case, drought conditions, though not as intense as in 2005, occurred in the 2007–2019 period (Figure S1). Persisting long, hot and dry summers promote photoinhibition and restrict carbon assimilation, limiting plant performance (Pérez-Ramos et al., 2017; Zunzunegui et al., 2005). This explains the small temporal changes of plant cover Rs, reflecting an overall loss of canopy 14 years after the disturbance in plots ranging from initially more to less impacted; overall, cover does not achieve pre-disturbance levels (Rs < 1). But interestingly, RRs increased with time (after 2008) indicating that the most impacted plots eventually tend to proportionally recover more than less impacted ones so that differences in cover tend to diminish. This reflects the capacity of dominant vegetation to grow from surviving structures such as stems and branches in less impacted plots. It also reflects longer lags to attain size in new plants that establish in more impacted plots, or, alternatively, the lower resprouting ability of more affected plants, as documented in studies about resprouting in Mediterranean species (Lloret et al., 2004). These findings highlight the necessity of long-term studies regularly surveying the community after extreme events, particularly droughts (Martínez-Vilalta & Lloret, 2016).

This study provides mechanistic interpretation of the response of woody communities to climate change, particularly to extreme weather episodes that are predicted to become more common and severe. Demographic properties of species, particularly described by the ‘pace of life’ and ‘reproductive strategy’ life-history axes, which have been shown to predict population responses to environmental variation (Paniw et al., 2018; Postuma et al., 2020), are revealed as key features to interpret community dynamics and their eventual resilience. This is in line with emerging evidence that using shifts in fitness-related biological traits—as body size for animals (Clements et al., 2017)—can improve assessments of early warning signals of population and community collapse in front of strong environmental alterations (Clements & Ozgul, 2018). We also demonstrate that assessing the relative severity of the extreme event as well as continuous community monitoring after the event are important to determine the levels of resilience or alternatively facilitate eventual shifts. Importantly, we emphasize that different metrics to assess resilience give a complementary picture of the post-disturbance dynamics and are sensitive to different community properties and species attributes.


The authors thank the staff in the Doñana Biological Reserve for their continued logistical support during field sampling. They specially thank Rafael Villar for his advice and support, particularly for his comments on an early version of the manuscript. They are also grateful to I. Granzow, J. Margalef, M. Angeles Pérez and T. Quirante for help during sampling. M.P. was supported by a Spanish Ministry of Economy and Competitiveness Juan de la Cierva-Formación grant FJCI-2017-32893. F.L. was supported by Spanish Ministry of Economy and Competitiveness grants CGL2006-01293/BOS, CGL2009-08101, CGL2012-32965, CGL2015-67419-R, AGAUR (Generalitat de Catalunya) grant (2017 SGR 1001) and ICTS RBD 2007/38, 2013/11, 2016/25 projects.


    M.P. and F.L. conceived the ideas and designed the methodology; F.L. and E.G.d.l.R. collected the data; M.P. analysed the data; M.P. led the writing of the manuscript. All the authors contributed critically to the drafts and gave final approval for publication.


    The peer review history for this article is available at https://publons.com/publon/10.1111/1365-2745.13597.


    Data and code available from the Dryad Digital Repository https://doi.org/10.5061/dryad.xwdbrv1cn (Paniw et al., 2021).