Study area

We surveyed 45 sites along an aridity gradient from central to southeast Spain (Supporting Information Fig. S1). Mean annual precipitation and temperature along this gradient ranged from 294 to 479 mm and from 12 to 18°C, respectively. Aridity (1 – aridity index; precipitation/potential evapotranspiration; Delgado-Baquerizo et al., 2013) values range from 0.57 to 0.76, and are strongly correlated to both annual mean precipitation (R² = 0.97) and temperature (R² = 0.89) in the studied sites. Climatic data were extracted from the WorldClim global database (Hijmans et al., 2005), while data to calculate the aridity index were obtained from Trabucco & Zomer (2009). All the studied sites were located on south-facing slopes, with slope values ranging from 1 to 22° (measured in situ with a clinometer), and had soils derived from limestone (Lithic Calciorthid; Soil Survey Staff, 1994). Vegetation at these sites was either a grassland dominated by Stipa tenacissima or a shrubland dominated by obligate-seeder shrubs such as Rosmarinus officinalis (hereafter nonsprouting shrubs, Fig. S2). Within grasslands and shrublands, we selected sites with and without tall sprouting shrubs (such as Quercus coccifera; Fig. S2; Table S1).

Vegetation sampling

We established a 30 × 30 m plot at each study site. Total plant cover within each plot was sampled by using four 30 m long transects located 8 m apart from each other, which were extended parallel to the slope. In each transect, the cover of every perennial species in 20 consecutive quadrats (1.5 × 1.5 m) was visually recorded. We focused on perennial plants as they represent most of the plant biomass in drylands (Whitford, 2002), and their cover is a good predictor of ecosystem functioning in these areas (Maestre & Escudero, 2009; García-Gómez & Maestre, 2011; Gaitán et al., 2014). Species abundance per site was calculated as the sum of the cover measured in the 80 quadrats.

At each plot, we measured the traits of all the perennial plant species that accounted for at least 80% of the total plot cover, in decreasing order of relative abundance. These measurements were conducted on 10 randomly selected individuals per species during the peak of the vegetation growth season (spring). We assigned to each species and plot the average value of the individuals measured in that plot. In the case of the species for which we did not have local trait values we used the average trait values observed in the three nearest sites. Ten traits were measured following standardized protocols (Cornelissen et al., 2003): plant architecture traits, including vegetative height (VH, cm), lateral spread (LS, cm²), branching density (BD, number of main stems) and ramification (Br, number of ramifications per stem) (these traits are related to plant water-use efficiency and/or competitive ability; Westoby et al., 2002); and leaf traits, including leaf area (LA, cm²), leaf length (LL, cm), leaf width (LW, cm) and leaf thickness (LT, mm), all reflecting light interception and water stress tolerance (Westoby et al., 2002), and SLA (cm² g⁻¹) and leaf dry matter content (LDMC, g g⁻¹), which correlate with plant relative growth rate and nutrient acquisition and utilization (Wright et al., 2004).

Soil sampling and analyses

Soil cores (0–7.5 cm depth) were sampled during the peak of the dry season (July–August) under the canopy of five randomly selected S. tenacissima and R. officinalis individuals, and five others in randomly selected open areas devoid of vascular vegetation. In those sites with sprouting shrubs, additional soil cores were sampled under the canopy of five randomly selected individuals of these shrubs. Hence, 10 or 15 soil samples, respectively, were collected per site.

Soil samples were sieved by a 2 mm mesh and air-dried for 1 month before laboratory analyses. For each soil sample, the following variables were quantified as described in Maestre et al. (2012a) and Delgado-Baquerizo et al. (2013): organic C, pentoses, hexoses, total nitrogen (N), total available N, amino acids, proteins, net potential mineralization rate, total phosphorus (P), available inorganic P, Olsen P (inorganic P – HCL 1M) and the activities of phosphatase and β-glucosidase. These variables constitute a good proxy for processes such as nutrient cycling, biological productivity, and build-up of nutrient pools, which are important determinants of ecosystem functioning in drylands (Whitford, 2002). Most of these processes are also considered to support ecosystem services, as other types of ecosystem services depend on them (MEA, 2005; Isbell et al., 2011).

Data management

Community trait distribution

For each measured trait, we calculated two complementary indices of functional structure: CWM and FD. CWM corresponds to the mean trait value of a community weighted by the relative abundance of each species, and reflects the trait values of the most dominant plant species in a given community. It was calculated with the following equation (Violle et al., 2007):

CWMj=∑inpijTij(1)

where p_ij is the abundance of the species i in the community j, and T_ij is the mean trait value of the species i in the community j.

Functional diversity quantifies the degree of trait dispersion within a community (adapted from Laliberté & Legendre, 2010). Calculated for each trait separately, FD is similar to the variance of the community trait distribution weighted by the relative abundance of each species within the community. It was calculated as:

FDj=∑inpijTij−CWMj∑inTij−CWMj(2)

where p_ij is the abundance of the species i in the community j, T_ij is the mean trait value of the species i in the community j, and CWM_j is the community-weighted trait of the community j. High FD values suggest higher complementarity in resource used between species within a given community (Maire et al., 2012).

Statistical analyses

Functional structure of studied communities

The CWM of the studied communities segregated along two PCA components, which accounted for 62% of the total variance found in the data (Fig. 2a; Tables S4a, S5a). The first component (36% of the variance) separated communities according to their leaf trait values (hereafter CWM-leaf trait), with SLA and Br being negatively correlated to LDMC, LL and LA. The first PCA component was negatively correlated with the abundance of S. tenacissima (r² = 0.82, P < 0.001; dot scale in Fig. 2), and discriminated grasslands from shrublands. The second PCA component discriminated communities according to plant size traits (hereafter CWM-size trait), with VH, LW and LS being negatively correlated to BD.

Similarly to what was observed with CWM, the FD of the studied communities was explained by the two first PCA components, which accounted for 55% of the total variance in the data (Fig. 2b; Tables S4b, S5b). The first component (31% of the variance) discriminated communities according to the FD of traits related to plant size (hereafter FD-size trait), such as FD-Br, FD-LS, FD-LW and FD-VH. The second PCA component (24% of the variance) segregated communities according to the FD of leaf traits (hereafter FD-leaf traits), such as FD-SLA, FD-LDMC, and FD-LT.

Community response traits to aridity and shrub encroachment

The abundance of nonsprouting shrubs largely determined CWM-leaf traits (73% of the explained variance; Table 1) and the communities dominated by these species had higher CWM-SLA and -Br, and lower CWM-LDMC, -LL, and -LA. These traits were also significantly impacted by aridity and the abundance of sprouting shrubs, although to a lesser extent (8 and 19% of the explained model variance, respectively; Table 1). A quadratic relationship was observed between aridity and CWM-leaf traits (Table 1; Fig. S4). By contrast, CWM-size traits were mostly driven by the abundance of sprouting shrubs (Table 1). Communities with high CWM-size traits were those dominated by tall sprouting shrubs.

The abundance of sprouting shrubs largely impacted FD-size traits (97% of the variance explained), whose values peaked at intermediate values of nonsprouting shrub abundance (Table 1). Finally, variations in FD-leaf traits were driven by the interplay of aridity and shrub abundance (Table 1). A positive quadratic relationship between aridity and FD-leaf traits (r² = 36%) indicated that the FD values of these traits peaked in low and high aridity conditions. Sprouting shrubs tended to have a negative impact on FD-leaf traits (r² = 20%), while nonsprouting shrubs increased FD-leaf traits (r² = 44%).

Linking community response traits to effect traits on multifunctionality

The model including the combined effects of CWM and FD (combined hypothesis) was the only model not rejected by the data (Fig. 3; Table S6). This model explained 62% of the variation in multifunctionality. Importantly, it highlighted that the effects of shrub encroachment on multifunctionality were mostly indirect via its effects on the functional structure of the plant community (Fig. 4).

While aridity had a direct effect on multifunctionality, it also had a large cascading effect by altering the functional structure of the studied communities. Aridity favored the abundance of nonsprouting shrubs, which resulted in higher values of CWM-leaf traits (Fig. 3). Shifting leaf trait values toward higher SLA had a strong adverse effect on multifunctionality. By contrast, the abundance of sprouting shrubs was independent of aridity. Increasing the abundance of these shrubs changed the value of CWM-size traits towards higher plant height. Such an increase did not directly impact multifunctionality, but had an indirect effect via the changes it promoted in FD (Fig. 3). Increasing the average size of the species in the community augmented the FD of size traits, although it decreased the FD of leaf traits, especially for intermediate values of CWM-size traits (quadratic relationship). It should be noted that communities showing a high variance in size traits were also characterized by high FD values of leaf traits. Increasing FD values of both leaf and size traits generally increased multifunctionality. However, a significant interaction between aridity and FD leaf traits was observed (Fig. 3). This indicates that the effect of these traits on multifunctionality shifted from positive to negative under high aridity conditions. Finally, sprouting and nonsprouting shrubs did not have a direct effect on multifunctionality (Fig. 4), suggesting that all their effects on multifunctionality were explained by the functional traits measured.

Model scenarios

In the sensitivity analyses of our final path model (Fig. 3), scenario 1 modeled the direct effect of aridity on multifunctionality as it had fixed zero abundance of both types of shrubs. In this case, multifunctionality directly decreased with increases in aridity (orange line, Fig. 5). In scenario 2, we modeled the effects of aridity on the abundance of nonsprouting shrubs (significant link in Fig. 3) and its consequences for multifunctionality. Increasing the abundance of nonsprouting shrubs augmented CWM-leaf traits, and strongly decreased multifunctionality, along the aridity gradient (green line, Fig. 5). Finally, in scenario 3 we fixed the abundance of sprouting shrubs to 30% to maintain high values of FD along the aridity gradient and to model its effects on multifunctionality. In this scenario, multifunctionality values remained high for most of the aridity gradient, declining only under high aridity conditions (red line, Fig. 5).

FD enhances multifunctionality in drylands

Functional diversity within dryland communities improved ecosystem multifunctionality, and accounted for a large fraction of the variation across communities (42% of the effect on multifunctionality; Fig. 4). This result contrasts with studies conducted in more mesic ecosystems, which highlighted the importance of CWM as a driver of ecosystem functioning (Garnier et al., 2004; Díaz et al., 2007; Mokany et al., 2008). However, most studies conducted so far addressing the relationship between FD and ecosystem functioning have considered single ecosystem functions (e.g. productivity (Garnier et al., 2004) or soil C (Laliberté & Tylianakis, 2012); for a review, see De Bello et al., 2010). Our results suggest that FD and the associated niche complementarity might be particularly important when considering multiple ecosystem processes simultaneously (Mouillot et al., 2011).

In temperate ecosystems, the effect of high FD on ecosystem functioning has generally been associated with higher resource acquisition rates (Van Ruijven & Berendse, 2005) and resource-use efficiency (Gross et al., 2007a), temporal niche variability (Maire et al., 2012) and plant soil feedbacks (Van Der Heijden et al., 2008). While future experiments are needed to identify the underlying mechanisms supporting the positive relationship between FD and ecosystem multifunctionality reported here, our results suggest that FD may improve multifunctionality in drylands via two distinct pathways. First, increasing the FD of size traits can lead to regular spatial distributions of plants according to their size (Gross et al., 2013), with tall individuals being regularly spaced between each other. Such spatial distributions, which are characteristic of dryland communities (Fowler, 1986), can limit runoff and maximize soil infiltration and heterogeneity (Valentin et al., 1999), thus enhancing species diversity (Soliveres et al., 2011) and maximizing plant growth and ecosystem functioning (Puigdefábregas et al., 1999). Secondly, high leaf trait diversity indicates the occurrence of contrasting leaf strategies (Westoby et al., 2002) commonly found in Mediterranean systems (e.g. stress avoidance vs tolerance; Ackerly et al., 2002; Freschet et al., 2011). Differences in the leaf strategy of co-occurring species may have strong positive effects on ecosystem processes such as productivity (Gross et al., 2007a), C cycling (Milcu et al., 2014) and litter decomposition rates (Bardgett & Shine, 1999; Cornwell et al., 2008). For instance, some studies have shown that increasing the FD of litter positively influences microbial communities (Zak et al., 2003) and litter decomposition rates (Vos et al., 2013), two potentially important factors for maintaining and improving dryland multifunctionality.

Multiple traits mediate the impact of mass ratio processes on multifunctionality

By considering multiple traits, our study showed that the outcomes of shrub encroachment can be explained by size-related and leaf traits. Shrub encroachment by sprouting shrubs species (such as Q. coccifera) had a positive cascading effect on multifunctionality, which was mediated by increasing CWM-size trait values (Fig. 3), in accordance with Maestre et al. (2009). Increasing plant size in dryland communities has been shown to be strongly associated with an increase in FD-size traits locally, and with a high spatial heterogeneity of plant biomass within communities (Gross et al., 2013), two features that can have potential positive effects on ecosystem functioning, as discussed earlier. Maestre et al. (2009) showed how large Quercus species can increase the availability of local soil resources under their canopy in semiarid S. tenacissima grasslands. The positive effects of these shrubs on local resources have been shown to increase species diversity of the whole community (Maestre et al., 2009; Soliveres et al., 2011), an important parameter reinforcing the positive effect of sprouting shrubs on multifunctionality (Quero et al., 2013). Our results complement previous findings by illustrating how sprouting shrubs can enhance FD within dryland communities, ultimately affecting multifunctionality.

The CWM-leaf trait values increased with an increase in the abundance of nonsprouting shrubs (Table 1; Fig. S4). This had a negative impact on multifunctionality, particularly in the most arid part of the gradient. The negative effect of fast-growing species on multifunctionality can be explained by a negative plant soil feedback, as suggested by Garnier et al. (2004). Negative relationships between SLA and soil nutrient contents have previously been found in Mediterranean French grasslands (Garnier et al., 2004) and along successional vegetation stages, where fast-growing species are replaced by slow-growing species (Berendse, 1990). Higher growth and nutrient acquisition rates may accelerate nutrient uptake from the soil (Lavorel & Garnier, 2002). At the same time, plants with higher SLA may produce litter with higher decomposition rates (Kazakou et al., 2006). Together with the reduction of litter accumulation per unit of soil surface, these effects may accelerate nutrient loss at the scale of the whole ecosystem (Garnier et al., 2004). This may be particularly true in the most arid part of the aridity gradient, where the typical characteristics of the semiarid Mediterranean climate are worsened. For instance, the high variability of interannual precipitation distribution promotes increases in water runoff during short periods (Martínez-Mena et al., 2001) and increases soil erosion that might accelerate nutrient loss (Martínez-Mena et al., 2002). In addition, the negative effect of fast-growing summer deciduous species on multifunctionality can be amplified via an effect on FD-size traits (e.g. the negative link between CWM-leaf traits and FD-size traits in Fig. 3). Summer deciduous species with a stress avoidance strategy can outcompete the more stress-tolerant grass and shrub species (Gross et al., 2013) by producing allelopathic compounds (as has been found for species such as Artemisia herba-alba; Escudero et al., 2000). Competition between fast- and slow-growing species may decrease the abundance of slow-growing sprouting shrubs and modify the size and spatial distribution of plant biomass within communities (Gross et al., 2013). This situation may decrease the positive effects of sprouting shrubs on FD, accelerating species loss and affecting the functioning of the whole ecosystem (Maestre et al., 2009).

Importance of FD for ecosystem resistance to increasing aridity

The sensitivity analysis allowed us to explore how aridity interacts with plant functional community structure to determine multifunctionality (Fig. 5). While aridity had a direct detrimental effect on multifunctionality (scenario 1, Fig. 5; Delgado-Baquerizo et al., 2013), this negative effect was further reinforced by the increase in abundance of nonsprouting shrubs, as favored by increasing aridity (scenario 2, Fig. 5). Moreover, we found an interactive effect of aridity and FD-leaf traits on multifunctionality (Fig. 3), suggesting that the effects FD-leaf traits shifted from positive to negative as aridity increased. At low aridity, high FD-leaf traits may reflect the coexistence between fast-growing species characterized by perennial leaves (e.g. Brachypodium retusum), and stress-tolerant shrub or grass species (Frenette-Dussault et al., 2012) that maximized ecosystem multifunctionality. By contrast, under high aridity conditions, the increase in FD-leaf traits observed reflected the increase in abundance of nonsprouting shrubs (see the selection effect in Loreau & Hector, 2001), characterized by the high value of leaf traits (i.e. fast-growing species with summer deciduous leaves; Gross et al., 2013) that may negatively affect ecosystem functioning.

An important result of our study was that high FD (enhanced by the occurrence of sprouting shrubs in grasslands) strongly delayed the collapse of multifunctionality under high aridity conditions. This was suggested by our sensitivity analysis (Fig. 5) where high FD-size traits were generally able to buffer the negative effects of aridity on multifunctionality, hence increasing the ecosystem resistance to aridity. Our results agree with previous experimental studies showing how higher species or FD can improve ecosystem resistance to global change drivers such as climate or land-use changes (Hooper et al., 2005; Isbell et al., 2011; Cardinale et al., 2012). Understanding how the attributes of biotic communities mediate the resistance of ecosystem structure and functioning to global change drivers is a major focus of current ecological research. By identifying how fundamental attributes of biotic community predict ecosystem multifunctionality, our findings can be particularly useful for developing mechanistic models aiming to predict ecosystem resistance to climate change in drylands, which will increase the degree of aridity experienced by these ecosystems worldwide (Feng & Fu, 2013).

We standardized our sampling design by selecting sites with similar soil, slopes, and aspect (south-facing slopes). Local variation in topo-edaphic conditions could, however, alter plant community structure (Fonseca et al., 2000; Gross et al., 2008) and multifunctionality. For instance, while we did not find any significant effect of slope on multifunctionality, other local factors such as slope, aspect, soil texture or bedrock type could affect water availability (Fonseca et al., 2000; Gómez-Plaza et al., 2001; Delgado-Baquerizo et al., 2013). Evaluating how local topo-edaphic factors interact with climatic/land use factors to determine the functional structure of dryland communities and their effect on multifunctionality is an important research objective for the future.

Concluding remarks

Our work suggests that the functional traits of dominant species and their diversity within communities modulate changes in multifunctionality in Mediterranean ecosystems along gradients of aridity and shrub encroachment. We showed that maintaining and enhancing FD (promoted by sprouting shrubs) in these ecosystems may help to buffer negative effects of climate change on multifunctionality. We also identified key traits that can accurately predict the outcome of shrub encroachment. Our results contribute to resolving the existing debate in the literature on the contrasting effects of shrub encroachment in drylands worldwide (Schlesinger et al., 1990; Maestre et al., 2009). On the one hand, traits related to the size of the plant species reflected the abundance of sprouting shrubs, which positively feed back on multifunctionality via their positive effect of FD. On the other hand, leaf traits such as SLA were related to the abundance of nonsprouting shrubs, which negatively impacted multifunctionality (particularly at the driest part of the aridity gradient studied). These results suggest that high values of SLA may typify those shrub species that are commonly associated with land degradation and desertification in drylands (Eldridge et al., 2011).

Our results can be used to develop specific trait-based management and restoration programs (Sandel et al., 2011; Laughlin, 2014) aiming to buffer the effects of climate change and shrub encroachment on multifunctionality. For instance, reintroducing/favoring the development of plants with low SLA and/or large size, such as sprouting shrubs, and enhancing local FD would reverse or limit the negative effects of increasing aridity and seasonal fast-growing summer deciduous plant species on multifunctionality.