Diverse functional responses to drought in a Mediterranean-type shrubland in South Africa


Author for correspondence:
Adam G. West
Tel: +27 21 6503628
Email: adam.west@uct.ac.za


  • Mediterranean-type ecosystems contain 20% of all vascular plant diversity on Earth and have been identified as being particularly threatened by future increases in drought. Of particular concern is the Cape Floral Region of South Africa, a global biodiversity hotspot, yet there are limited experimental data to validate predicted impacts on the flora. In a field rainout experiment, we tested whether rooting depth and degree of isohydry or anisohydry could aid in the functional classification of drought responses across diverse growth forms.
  • We imposed a 6-month summer drought, for 2 yr, in a mountain fynbos shrubland. We monitored a suite of parameters, from physiological traits to morphological outcomes, in seven species comprising the three dominant growth forms (deep-rooted proteoid shrubs, shallow-rooted ericoid shrubs and graminoid restioids).
  • There was considerable variation in drought response both between and within the growth forms. The shallow-rooted, anisohydric ericoid shrubs all suffered considerable reductions in growth and flowering and increased mortality. By contrast, the shallow-rooted, isohydric restioids and deep-rooted, isohydric proteoid shrubs were largely unaffected by the drought.
  • Rooting depth and degree of iso/anisohydry allow a first-order functional classification of drought response pathways in this flora. Consideration of additional traits would further refine this approach.


Mediterranean-type ecosystems are home to a substantial fraction of the world’s plant biodiversity. Located on five continents, yet covering only 2% of global land area, they contain 20% of all vascular plant diversity on Earth (Cowling et al., 1996; Klausmeyer & Shaw, 2009). These winter-wet, summer-dry ecosystems have been identified as being particularly threatened by future climate change, primarily due to a prediction of increased and regionally acute drought (IPCC, 2007). Of particular concern is the winter rainfall region of South Africa, home to the highly diverse and endemic Cape Floral Region (CFR; Goldblatt & Manning, 2002; Hannah et al., 2005). The CFR, comprising primarily the fynbos and succulent karoo biomes, contains over 9000 species of vascular plants of which c. 68% are endemic to the region (Goldblatt & Manning, 2002). Vascular plant diversity in this region is high as a result of recent taxonomic radiations, resulting in high species richness in relatively few clades (Linder, 2003; Klak et al., 2004). The large number of such radiations are thought to have coincided with the development of a winter-rainfall, summer-drought Mediterranean-type climate brought on by the establishment of the Benguela Upwelling System c. 10 Myr BP (Dupont et al., 2011). Subsequently, plant diversity in the CFR has been maintained by low extinction rates afforded by the mild climate during the glacial maximum (Dynesius & Jansson, 2000), reliable winter rains (Cowling et al., 2005) and many climate refugia that coincide with a topographically and geologically complex region (Verboom et al., 2009).

Future climate projections for the CFR indicate that the region will become warmer and drier, with more extreme events, leading to increased frequency and intensity of drought (Hewitson et al., 2005; Midgley et al., 2005; Hewitson & Crane, 2006; IPCC, 2007). This is of particular concern as the absence of extreme and prolonged drought in the CFR may have aided in maintaining its plant diversity. The temperature projections are largely supported by climate data for the region indicating that temperatures are on the increase (Midgley et al., 2005; Hoffman et al., 2009). There is, however, considerably more uncertainty in the simulations of future rainfall (IPCC, 2007) that make predictions of future climate in the CFR, and the consequences for biodiversity, problematic. Examinations of historical data for the Western Cape have yielded variable trends in rainfall across the region (Midgley et al., 2005; Hoffman et al., 2009, 2011). Nevertheless, a clearer understanding of the impact of drought in this summer-dry region is necessary.

The majority of work studying the effect of climate change on the CFR has taken the form of climate envelope modelling, both at the level of the biome (Midgley et al., 2002) as well as that of individual species (Midgley et al., 2003; Schurr et al., 2007). These studies have indicated the potential of severe consequences for the flora including local extinctions and range shifts of many of the dominant species (Hannah et al., 2005). As has been well documented in the literature, their many assumptions and simplifications compromise bioclimatic model assessments (Heikkinen et al., 2006), particularly in the mega-diverse CFR (Yates et al., 2010), indicating that current models are far from being able to provide robust predictions, rather serving as first-cut assessments that require testing using a variety of approaches (Yates et al., 2010). Considering the uncertainty in climate projections, together with the high levels of biodiversity, our limited understanding of species tolerance levels (Midgley et al., 2005) and the topographic and geological complexity of the region with the micro-refugia these might create (Dobrowski, 2011), the reality is likely to be far more complex.

How well can we predict plant response to drought in a biodiverse region such as the fynbos biome? While there is a long history of ecological work in the fynbos (Cowling, 1992; Stock et al., 1992; Rebelo et al., 2006), there has been relatively little work directly examining the response of species, in situ, to drought. Previous work has mainly focused on observations of seasonal water relationships (Miller et al., 1983, 1984; Moll & Sommerville, 1985; Jeffery et al., 1987; van der Heyden & Lewis, 1989, 1990; Richardson & Kruger, 1990), rooting depths (Higgins et al., 1987), or proxy studies of drought tolerance through examination of xylem anatomy and resistance to cavitation (February & Manders, 1999; Jacobsen et al., 2007, 2009). This body of work has pointed to the existence of distinct functional groups in the fynbos that are represented by large, taxon-specific groups. (e.g. tall, deep-rooted proteoid shrubs; fine-leaved, shallow-rooted, ericoid shrubs; and graminoid restioids) (see Stock et al., 1992) that are likely to respond differently to drought. However, experimental studies manipulating water availability in the field have been rare (Herppich et al., 1994; Agenbag, 2006), and there have been no direct tests of the mechanistic basis for using these functional groups to categorize drought response in this flora.

Recent progress in unifying the theory of plant response to drought argues for the central role of a plant’s ‘hydraulic strategy’ in determining the mechanisms of plant mortality under drought (McDowell et al., 2008, McDowell, 2011). The majority of this work has considered co-occurring species of a similar growth form (e.g. trees). In this context, hydraulic strategies can be viewed as a continuum between isohydry (regulation of plant water potential at the cost of gas exchange) and anisohydry (maintenance of gas exchange at the cost of declining plant water potential) (e.g. West et al., 2008). However, when considering the diversity of growth forms in any particular region, an evaluation of a plant’s hydraulic strategy would be incomplete without additional traits. At a first approximation, a crucial trait for comparison across growth forms is rooting depth. Rooting depth should determine the degree to which a plant experiences a given meteorological drought, with deep-rooted species being buffered from drought to a greater extent than shallow-rooted species (Nepstad et al., 1994, 2007). Can an evaluation of rooting depth and degree of iso/anisohydry help to predict drought responses across diverse growth forms? This is a central question of our study.

Here we describe the results from a multi-year experimental drought manipulation in a mountain fynbos shrubland, located on the Cape Peninsula, South Africa. Drought was experimentally imposed over the course of two summers, and a suite of physiological parameters monitored in seven species representing the three dominant growth forms. Our central hypothesis stated that drought impacts would be largely determined by rooting depth, augmented by water potential regulation strategy, and would cluster by growth form. We predicted that the greatest impacts would be seen in the shallow-rooted restioids and ericoid shrubs with the least impact occurring in the large, deep-rooted proteoid shrubs. We discuss how the evaluation of hydraulic strategy can inform, and improve, our categorization of plant functional responses to drought, ultimately helping to make more accurate predictions regarding plant response to future climate change.

Materials and Methods

Study sites

Study sites were located in the Silvermine Section of Table Mountain National Park, South Africa (S34.105, E18.444). Mean annual rainfall is 1187 mm with 75% occurring during the winter months (MJJASO). Two sites were selected – an 8-yr post-fire site with reproductively and structurally mature vegetation (labelled ‘Mature’ in figures) and a 2-yr post-fire site with mostly reproductively and structurally immature vegetation (labelled ‘Young’ in figures). The sites were located within 1.2 km of each other at a similar aspect and elevation (400 m asl). Soils at both sites were sands formed from the underlying quartz arenite (Table Mountain Sandstone) with a poorly developed organic horizon. Soil depth was uniformly > 1.5 m at the young site but was more variable at the mature site, with fractured sandstone rock being encountered between 0.5 and 1.5 m depth. The vegetation on both sites was classed as mountain sandstone fynbos (Rebelo et al., 2006). Vegetation height was c. 1.5 and 0.5 m at the mature and young sites, respectively. A plant survey recorded 37 and 68 species in 144 m2 at the mature and young sites, respectively. The mature site was dominated by the proteoid Leucadendron laureolum (Lam.) Fourc. and Erica subcapitata (N.E.Br.) E.G.H.Oliv. and Erica ericoides (L.) E.G.H.Oliv.. The most common restioid on the site was Staberoha cernua (L.f.) T.Durand & Schinz. The young site was dominated by the proteoids L. laureolum and Diastella divaricata (P.J.Bergius) Rourke., the restioid Hypodiscus aristatus (Thunb.) Krauss.. The most common Erica species were Erica pyxidiflora Salisb. and E. ericoides.

Experimental design

Nine 4 m × 4 m experimental plots were selected at each site so as to include sufficient individuals of the focal species (Table 1). Plots were clustered in groups of three, with each cluster consisting of two treatment plots and one control plot. Distance between plots within a cluster was c. 4 m and distance between clusters was c. 15 m. Over each of the experimental plots, a rain exclusion structure was constructed consisting of a steel framework with six overlapping, clear polycarbonate roof panels that could be tilted flat or at a 30° angle (Supporting Information Fig. S1). When the panels were flat, rain falling on the surface would run off into the gutter and be piped off site, resulting in 100% rain exclusion. When the panels were at 30°, rain falling on the panels would run onto the plot, resulting in 0% rain exclusion while still casting the same shade as for the flat configuration. The roofs were positioned slightly above the height of the tallest vegetation at the site. The perimeters of the plots were trenched to a depth of 50 cm and lined with thick plastic before being backfilled. Plants were not measured within 75 cm of the perimeter, providing a buffer zone from the effects of trenching and precipitation entering from the sides of the structures. Considerable effort was made to prevent trampling in the plots. Stepping-stones were placed in each plot to provide access to specific instrumentation. Plant measurements were conducted from a 5-m horizontal ladder supported on either side of the plot.

Table 1.   Details of the focus species in this study
SiteSpeciesCodeFunctional typeGrowth formPlant height (m)Regeneration strategy
YoungDiastella divaricataDiadivProteoidSprawling0.15Re-seeder
Leucadendron laureolumLeulau(s)ProteoidTall shrub0.35Re-seeder
Erica pyxidifloraEripyxEricoidShrub0.25Re-seeder
Erica ericoidesErieriEricoidShrub0.15Re-seeder
Hypodiscus aristatusHypariRestioidRestioid0.65Sprouter
MatureLeucadendron laureolumLeulau(a)ProteoidTall shrub1.2Re-seeder
Erica subcapitataErisubEricoidShrub0.7Re-seeder
Erica ericoidesErieriEricoidShrub0.3Re-seeder
Staberoha cernuaStacerRestioidRestioid0.45Re-seeder

Micrometeorological and soil moisture measurements

A fixed weather station was deployed at the young site, measuring temperature and relative humidity (Vaisala HMP-45C), rainfall (Texas Electronics TE-525 mm), windspeed and direction (03001 Young Wind Sentry Set; Campbell Scientific, Logan, UT, USA) and photosynthetically active radiation (QSO-S; Apogee Instruments, Logan, UT, USA). Sensors were connected to a CR10× data logger (Campbell Scientific). Photosynthetically active radiation was also measured under a rain exclusion structure. Standalone rainfall gauges (RG3-M; Onset Computer Corporation, Bourne, MA, USA) were deployed at both sites to confirm there were no significant differences in microclimate between the two sites. Temperature and relative humidity (EL-USB-2; Lascar Electronics, USA) were also monitored at both sites in the open, under treatment and under control rain-out-shelters. The rainout shelters caused slightly altered micro-climate (warmer, lower PAR) relative to ambient conditions (Fig. S2).

Soil moisture access tubes (5 cm ∅) were installed in each experimental plot 2 months before the commencement of the experiment. Using these access tubes soil moisture was measured every 0.1 m, for 1 m, using a Frequency Domain Reflectometry (FDR) soil water profile sensor (Diviner 2000; Sentek Sensor Technologies, Stepney, SA, Australia). These measurements were taken every 2 wk. In addition to the Diviner profiles, a continuously logging FDR soil moisture profile (Envirosmart SOLO; Sentek Sensor Technologies) was established at each site. Data were logged every 30 min at depths of 0.1, 0.2, 0.4, 0.6 and 1.0 m. The percentages of soil moisture measured with the FDR sensors were compared with gravimetric measurements of soil water for field calibration.

Gas exchange measurements

Assimilation (A) and stomatal conductance (gs) were measured using a LI-6400 infrared gas analyzer (Li-Cor, Lincoln, NE, USA). Measurements were made on clear, sunny days between the hours of 10:00 h and 15:00 h. From 12:00 h to 13:00 h, midday stem water potential measurements (ΨMD) were made on the same plants that were used for gas exchange using a Scholander Pressure Chamber (PMS Instruments, Corvallis, OR, USA). All plants within a site were measured on the same day to eliminate the effects of variable weather. Gas exchange measurements were usually completed within 120 s of clamping on to the leaf so as to minimize chamber acclimation. The CO2 concentration was maintained at 380 ppm, flow rate was 400 μmol s−1, PAR inside the chamber was 1500 μmol (LI-6400 LED light source) and chamber humidity was maintained slightly below ambient. Sample and reference cells were matched before every measurement and empty chamber measurements were taken regularly to correct for drift in the CO2 and H2O sample cell values. Measurements were made on the most healthy sunlit foliage on the north side of each plant. Immediately following the measurement, the measured leaf material was photographed underneath a LI-6400 gasket in the exact configuration that it had been in the chamber. Projected leaf area was calculated using ImageJ (http://rsbweb.nih.gov/ij/). This photographic method prevented excessive defoliation over the course of the experiment.

Isotopes and rooting depths

We measured hydrogen and oxygen stable isotope ratios of water as a proxy of rooting depths for the focus species in our study (Dawson, 1993). Suberized stems (proteoids and ericas), rhizomes (restioids) and soil cores were collected in the experimental plots during a summer sampling campaign (February, 2008/9) coincident with our gas exchange and water potential measurements. Samples were immediately placed in screw-topped glass vials and sealed with Parafilm™ (Pechiney Plastic Packaging, Chicago, IL, USA) to prevent any evaporation. Once in the laboratory, samples were kept frozen at − 20°C until extraction. Water was cryogenically extracted from the samples at the University of Cape Town following the methodology of West et al. (2006). Hydrogen isotope ratios (δ2H) were measured by injecting microlitre quantities of water into an H/Device coupled to a Delta Plus mass spectrometer (ThermoFinnigan, Bremen, Germany). Injected H2O was reduced to H2 gas in a hot chromium reactor and the 2H/H ratio of this gas was then analysed by mass spectrometry (Brand et al., 1996; Gehre et al., 1996). Oxygen isotope ratios (δ18O) were measured by CO2 equilibration (Socki et al., 1992). Water samples were left to equilibrate with a 0.2% CO2 headspace for 48 h at 21–23°C. Following equilibration, the vials were inserted into a GasBench II connected to a Delta Plus XL mass spectrometer (ThermoFinnigan). The GasBench II was modified with a 10-port injection valve, allowing a 0.2% CO2 reference injection to follow each sample CO2 injection. All isotope analyses were conducted at the Center of Isotope Biogeochemistry, University of California, Berkeley and are expressed in per mil notation (‰) relative to the standard V-SMOW, following the equation (Brand et al., 2009):

image(Eqn 1)

(n is the heavy isotope of element E, R is the ratio of the heavy to light isotope (2H/1H or 18O/16O)). Analytical precision on internal quality control standards was at least 1.5‰ (δ2H) and 0.15‰ (δ18O).

Growth, phenology and mortality measurements

In order to assess the impact of our drought treatment on the seven focal species, we made monthly measurements of growth, phenology and mortality. All measurements were made on six individuals per species, per treatment, per site. At each measurement period, phenological data were recorded as presence/absence of new growth, flowers, seeds or cones. Mortality data were scored as % mortality of the whole plant. Monthly shoot or culm growth was measured on four shoots/culms per individual that were tagged with coloured wire. For the Restionaceae (S. cernua and H. aristatus), culm growth was measured on healthy, newly emerged culms > 10 cm long. Tags were switched to new culms before they approached maximum length. For the Proteaceae and Ericaceae, distal tips of healthy shoots were targeted for measurement. In the case of shoot damage through herbivory or mechanical damage, the shoot was excluded from that month’s data and the tag was repositioned on another shoot. In the case of shoot death from drought, the shoot was measured and the data included. Following this the tag was repositioned on another healthy shoot on the plant. In this manner, measurements were biased towards actively growing parts of the plant. To account for canopy dieback, monthly shoot growth was scaled by the live canopy fraction. Growth data are presented as relative canopy growth rate (RCGR, mm mm−1 d−1):

image(Eqn 2)

(Lt is total shoot length measured at time t, Lt−1 is the total length of the shoot at the previous monthly measurement, D is the number of days between t and t−1, and CL is the live canopy fraction.

Effect size

In order to examine the significance of differences between treatment and control measurements (for soil moisture, growth and mortality), data were plotted as effect size (ES):

image(Eqn 3)

(MT and MC represent the mean values for treatment and control plots, respectively, and εT and εC are the standard errors of the mean for treatment and control plots, respectively.) Thus, ES is in units of standard errors. We considered times where −2 > ES > 2 to represent significant differences between treatment and control.

Response pathways

We used our analyses of traits and outcomes (described in the Results section) as evidence for describing the ‘response pathways’ species took in relation to the rainfall manipulation (Fig. 6). The relative position of each species at each tier (trait or outcome) was assigned qualitatively, based on data collected in this study, to compare responses and detect common pathways of response. The tiers are organized as a hierarchy from the traits measured to the outcomes of how the plants responded to our treatments in terms of mortality and flowering (fitness). For some of the species we were unable to obtain all of the data for every tier and in these cases the symbol and lines are simply missing from that tier.


Climate and soil moisture during the course of the experiment

Seasonal variations in temperature, vapour pressure deficit and rainfall at our sites reveal a cool, wet winter and warm, dry summer climate typical of the region (Fig. S3). Our rain exclusion structures significantly reduced soil moisture in the top 40 cm of soil in the treatment plots during the two summer drought periods, resulting in an intense, shallow soil drought (Fig. 1). The treatment effect was considerably greater in the first summer as rainfall was higher (Table S1) and more frequent (Table S2); thus, surface soil moisture in the control plots was regularly increased relative to the treatment plots. During the second summer, limited summer rainfall resulted in little difference between treatment and control plots, both being relatively dry.

Figure 1.

Soil moisture dynamics and treatment effect over the course of the study on functional responses to drought in a Mediterranean-type shrubland in South Africa. Left-hand panels: average gravimetric soil moisture (%, ± 1 SE) at multiple depths for treatment (closed symbols) and control (open symbols) plots at both sites. Right-hand panels: soil moisture treatment effect at various depths for each site, plotted as effect size (number of standard errors). Differences < ± 2 SEs from 0 are regarded as insignificant (open symbols). Differences > ± 2 SEs from 0 are regarded as significant (closed symbols). Differences that exceeded the scale are truncated and plotted as diamonds. For all panels, grey shaded bars represent the summer drought exclusion treatments (2008/9), and subsequent monitoring period (dashed lines) when no experimental drought was imposed (2010).

Effect of drought on gas exchange and water status

Changes in leaf level gas exchange with shoot ΨMD reveal different water potential regulation strategies for the species examined (Fig. 2). The patterns were consistent within a species for species present in both young and mature sites despite different plant age. The three Erica species –E. subcapitata, E. ericoides and E. pyxidiflora– exhibited a strongly anisohydric strategy, allowing ΨMD to fall to very negative values while still maintaining gas exchange, albeit at a very reduced rate. By contrast, the proteoids L. laureolum and D. divaricata maintained constant ΨMD despite decreasing gs and A to zero during the summer drought period. This is indicative of an isohydric regulation of Ψ at the cost of gas exchange. A similar isohydric strategy was observed for the restioid, S. cernua. The data from the restioid H. aristatus were suggestive of isohydry, but inconclusive. Gas exchange did not reach zero in this species and the water potential did not decline beyond the range seen for other isohydric species. However, it was clear that the experimental drought was insufficient to induce complete stomatal closure or highly negative water potentials in this species.

Figure 2.

Light-saturated photosynthetic assimilation (Amax) and stomatal conductance (gs max) vs midday shoot water potential (ΨMD) for seven fynbos species of varying growth form. Open symbols, control plants; closed symbols, treatment plants; circles, plants from the mature site; squares, plants from the young site. Species codes are listed in Table 1.

Isotopic evidence for rooting depths

In order to infer the relative dependence of plants on shallow or deeper water sources during summer, we measured the δ2H and δ18O of water from plant stems (proteoids and ericoids) and rhizomes (restioids) in February over 2 years (2008/9). From monthly measurements of rainwater over the course of the study, we constructed a local meteoric water line (LMWL; Fig. 3). This line was similar to that from a nearby long-term dataset (Harris et al., 2010). As expected, summer rainfall and cloud moisture collected over the month preceding our sampling effort fell on the LMWL. The amount-weighted rainfall (averaged over the study) was more negative than the summer precipitation owing to the predominance of winter rainfall inputs. We intended to use soil water isotope profiles to establish plant rooting depths. However, analytical problems with extracting water from the dry soils rendered these data unreliable. Despite this we were able to infer relative rooting depth (deep or shallow) by comparing the plant water data to the LMWL. Plant water plotted away from the amount-weighted annual rainfall along a line with a shallower slope than the LMWL (Fig. 3), known to represent evaporation (Sharp, 2007; Dawson & Simonin, 2011). An extensive literature on plant and water source isotope data (reviewed by Dawson et al., 2002) supports the interpretation that the greater the distance a sample occurs along an evaporation line from the amount-weighted annual rainfall, the shallower the soil the sample was derived from. Based on this interpretation, the most deep-rooted plants are the proteoids L. laureolum and D. divaricata. The Erica species and the restioids appeared to be more shallow-rooted. These general rooting patterns were confirmed with qualitative plant excavations on site.

Figure 3.

δ2H and δ18O isotopic composition of plant xylem water and precipitation in a Mediterranean-type shrubland in South Africa. GMWL, the Global Meteoric Water Line (= 8x + 10); LMWL, the local meteoric water line (= 5.7x + 6.9) defined from local precipitation over the course of the study. Summer rain and cloud are from samples collected a month before the plant sampling. Weighted rain is the amount weighted isotope composition of precipitation over the course of the study. Species codes are listed in Table 1. Plants inferred to be deep-rooted and shallow-rooted are circled.

Effect of drought on growth, canopy mortality and flowering

Ericas  The Erica species were the most severely impacted by the experimental drought, with droughted plants showing reduced growth, increased levels of mortality and reduced reproductive effort relative to the controls.

Our first rain exclusion induced very negative shoot water potentials in all three Erica species (Fig. 2) resulting in little to no shoot growth in the droughted plants (Fig. 4). At the same time, the relatively wet summer (Tables S1,S2) allowed considerable growth in the control plants. The result was a marked treatment effect, with significantly less growth occurring in the treatment plants for these species (Fig. 4). During this treatment, there was high mortality amongst the droughted E. pyxidiflora seedlings, with no mortality occurring amongst the control plants (Fig. 5). While there was no whole plant mortality for the adult Erica species, E. subcapitata suffered considerable leaf loss and canopy dieback following this period (Fig. 5). By contrast, E. ericoides managed to avoid extensive canopy dieback. When the drought period was relieved in May, previously droughted E. ericoides plants had increased growth relative to the control plants. Erica subcapitata also showed considerable shoot growth in the surviving shoots during this time; however, the high levels of canopy mortality meant that whole plant growth was indistinguishable from the control (Fig. 4). For E. pyxidiflora, there was no difference in shoot growth between previously droughted and control plants following the first drought period (Fig. 4).

Figure 4.

Shoot and culm growth rates as relative shoot growth (left-hand side panels) and effect size (right-hand panels) over the course of the study on functional responses to drought in a Mediterranean-type shrubland in South Africa. For relative shoot growth: closed symbols, treatment plants; open symbols, control plants. For effect size (plotted as number of standard errors), differences < ± 2 SEs from 0 are regarded as insignificant (open symbols). Differences > ± 2 SEs from 0 are regarded as significant (closed symbols). Shaded panels represent drought periods, as for Fig. 1.

Figure 5.

Mortality trends for all species over the course of the study on functional responses to drought in a Mediterranean-type shrubland in South Africa. Mortality is presented as % canopy mortality (closed symbols, treatment plants; open symbols, control plants) and cumulative plant mortality (black lines, treatment plants; white lines, control plants). Species codes are listed in Table 1. Shaded panels represent drought periods, as for Fig. 1.

The second experimental drought occurred during a much drier summer (Tables S1,S2), with control plants at the mature site having much lower growth than in the previous summer (Fig. 4). This resulted in a much smaller treatment effect over this drought period for E. subcapitata and E. ericoides. Nevertheless, the pattern of reduced growth in the droughted plants was repeated (Fig. 4). This second drought period resulted in some whole plant mortality for E. subcapitata (Fig. 5), as previously stressed plants with reduced live canopy area were unable to survive a second drought period. No control E. subcapitata plants died during this time. Plant mortality in E. ericoides was too low to interpret in terms of our experimental treatments (Fig. 5). At the young site, growth in control E. pyxidiflora plants was unaffected by the lower summer rainfall, most likely due to the wetter nature of this site (Fig. 1), and the same pattern of reduced growth in droughted plants was documented as for the first experimental drought. Interestingly, there was no E. pyxidiflora mortality associated with this second drought period, possibly as these plants had survived the first drought period presumably through being better established, and were thus able to survive this drought period too.

The experimental droughts had considerable impact on flowering in the Erica species. Following the first experimental treatment, none of the droughted E. subcapitata plants produced flowers, whereas 80% of control plants had considerable reproductive effort. Flowering over the second summer drought was more variable, but droughted plants had significantly less flowering effort than control plants (Table 2). For E. ericoides, flowering was less affected by the experimental treatments, and there was no significant difference between control and treatment over the course of the study (Table 2). However, what was evident was the month-long delay in onset of flowering in droughted E. ericoides following the first drought treatment (Fig. S4). Flowering was also somewhat reduced in droughted plants following the second experimental treatment. The effect of our experimental drought on flowering in E. pyxidiflora was the most striking. No droughted plants flowered over the course of the entire experiment (Fig. S4). By contrast, many of the control plants did flower, in increasing amounts as the young plants reached maturity.

Table 2.   Differences in flowering between treatment and control plants over the course of the study, as tested by paired t-test
SpeciesControl mean (% flowering)Treatment mean (% flowering)tdfP
  1. 1Presence of live female cones. Values in bold are significant at P < 0.05.

Diastella divaricata53.022.2−7.527< 0.0001
Erica ericoides16.112.1−1.1270.290
Erica pyxidiflora6.10.0−2.2270.039
Erica subcapitata25.412.1−2.3270.030
Hypodiscus aristatus97.597.60.06270.952
Staberoha cernua98.298.30.07270.943
Leucadendron laureolum195.295.20.02270.981

Proteoids  For both the adult and juvenile L. laureolum plants, there was very little treatment effect on growth (Fig. 4). The only significant differences for adult plants occurred in the first experimental drought when growth declined slightly earlier for droughted plants than control plants (Fig. 4). Interestingly, for the juvenile L. laureolum plants there were no significant differences between treatment and control plants during our experimental droughts (Fig. 4), suggesting that these plants were already able to access a stable water source from the shallow water table at this site. There was also very little canopy dieback and plant mortality for this species over the duration of the experiment (Fig. 5), and no difference in female cone production between treatment and control groups (Table 2).

In contrast to L. laureolum, D. divaricata was highly sensitive to the experimental drought. During both experimental drought periods, growth was significantly reduced in droughted D. divaricata plants (Fig. 4). There was also considerable plant mortality during the second experimental drought, with eight plants dying in treatment plots compared with only two in control plots (Fig. 5). Additionally, droughted D. divaricata plants flowered significantly less than control plants (Table 2), with this being most obvious during the late summer of our experimental treatments (Fig. S4). In the third summer, when there was no experimental drought, surviving D. divaricata plants had identical growth patterns in treatment and control plots, with no further mortality (Fig. 5). However, previously droughted plants had less of a reproductive effort than control plants, possibly a consequence of the previous two drought periods.

Restioids  There appeared to be little treatment effect on growth, mortality and flowering for the two restioid species studied (H. aristatus and S. cernua), with growth differing significantly between treatment and control plants only in a few instances (Fig. 4), and no significant differences occurring in the mortality (Fig. 5) or flowering (Table 2) data.


Our study has shown fynbos plants to have both considerable resilience and sensitivity to drought in this unique biodiversity hotspot. Far from being uniformly sensitive to our experimental 6-month droughts, we observed substantial variation in drought impacts for the study species. We had predicted that deep-rooted species would be buffered from our drought treatment and would show little impact, whereas shallow-rooted species would be negatively affected by the shallow soil drought. While this was the case for the deep-rooted L. laureolum (no impact) and shallow-rooted Erica species (moderate to high impact), it was not the case for the shallow-rooted restioids (no impact) and the deep-rooted D. divaricata (high impact). The variation in response to shallow soil drought highlights the complexity of predicting drought impacts in such a biodiverse system, and emphasizes the need for improved mechanistic understanding of plant susceptibility to drought (McDowell et al., 2011).

Hydraulic theory predicts that shallow-rooted, anisohydric species would be most severely impacted by our intense, shallow experimental drought as they would experience rapidly declining water potential, possibly to the point of hydraulic failure (McDowell et al., 2008). Hydraulic safety margins tend to increase with increasing degree of anisohydry (Meinzer et al., 2009), and thus anisohydric plants can continue to transpire into a drought, albeit with increased costs of xylem dysfunction (West et al., 2008). Survival of droughted anisohydric plants is thus tightly coupled to xylem vulnerability to cavitation (McDowell et al., 2008). All Erica species – adult or seedling – showed a strongly anisohydric response, with considerable declines in stem water potential being measured during the experimental droughts. Water potentials as low as − 10 MPa were measured, which is considerably lower than previously reported (e.g. Miller et al., 1983; Jacobsen et al., 2007). These low water potentials resulted in canopy dieback of up to 80% for E. subcapitata and considerable whole plant mortality for the E. pyxidiflora seedlings. Remarkably, there was little mortality or canopy dieback in E. ericoides, despite very low water potentials being measured. An independent survey of naturally droughted plants near our experimental plots (data not shown) showed that E. ericoides was able to withstand ΨMD of − 8 MPa over the summer and still retain a full canopy and subsequently flower in the wet season. However, plants that experienced more negative ΨMD were dead the following spring. As in other genera of the Ericaceae (e.g. Arctostaphylos in California), this high cavitation resistance may be related to xylem architecture (i.e. vasicentric tracheids; Carlquist, 1985). However, given a drought of sufficient intensity, such as those predicted to occur under climate change, Erica species are likely to suffer hydraulic failure and mortality.

In contrast to the anisohydric response described above, shallow-rooted isohydric species should have been less severely impacted by our experimental drought, as drought mortality in isohydric plants is only likely to manifest itself after prolonged drought (McDowell et al., 2008). While our rain exclusion was maintained for nearly 6 months, the period of drought experienced by plants was considerably shorter than this for two reasons. First, soil moisture took several months to decline following the imposition of our drought treatment (Fig. 1). Second, our rain-exclusion structures did not exclude regular periods of cloud moisture during the summer (Table S2), which may have periodically alleviated water stress for those species able to utilize it. Isohydric plants typically maintain high hydraulic conductivity through the regulation of water potential, allowing a greater ability to respond to pulses of soil moisture (West et al., 2007a,b). In our study, the shallow-rooted, isohydric restioids possibly avoided significant drought impacts through the avoidance of cavitation and the use of regular cloud moisture events. The potential for the restioid growth form to collect cloud moisture to alleviate drought has been recognized since the early 20th century (Marloth, 1903). Cloud moisture can improve plant water status by direct foliar uptake (Burgess & Dawson, 2004; Breshears et al., 2008; Limm et al., 2009; Simonin et al., 2009) or through uptake of cloud or fog drip from plant canopies into the shallow soil (Dawson, 1998). During our experiment, there was considerable evidence of cloud moisture collecting on restioid culms and being channeled to the base of these plants (Fig. S5). It is likely that the shallow-rooted, isohydric, restioids were exploiting regular cloud moisture events for brief periods of gas exchange during the summer drought, possibly alleviating a negative carbon balance in droughted plants. However, cloud moisture is a highly site-specific phenomenon in the Cape mountains, and restioids are not restricted to the cloud belt. Future research should explore the drought sensitivity of restioids without a cloud water subsidy to test the generality of our findings in this experiment.

The minimal impact of our experimental drought on both adult and seedling L. laureolum resulted from a combination of isohydric behavior and access to deeper water throughout the summer. Furthermore, hydraulic lift (Hawkins et al., 2009) may have mitigated the effects of shallow soil drought for this growth form. The lack of drought stress in the 2–3-yr-old L. laureolum supports the finding that taproot growth is rapid in this group (Manders & Smith, 1992), and is crucial to their survival over the summer drought. While established L. laureolum plants appear well buffered from drought, this growth form may still be vulnerable to intense summer drought in the post-fire establishment phase. In contrast to the woody, emergent L. laureolum, D. divaricata was more herbaceous, with a sprawling growth form constraining the plant to within the boundary layer created by surrounding vegetation. Why was this species so heavily impacted by our experimental drought? The strongly isohydric behavior of D. divaricata resulted in prolonged stomatal closure over the summer drought. Air temperatures below our rainout shelters, but above the vegetation boundary layer, reached 42°C over the summer drought. Stomatal closure, combined with the low windspeed in the vegetation boundary layer, would have resulted in limited opportunity for latent and sensible heat loss from leaves potentially leading to a negative carbon balance and/or lethal leaf temperatures in D. divaricata.

Despite 6 months of rain removal, there was little outright mortality in our experiment (Fig. 5). The relevance of nonlethal drought events in shaping vegetation structure is currently under debate (McDowell & Sevanto, 2010; McDowell et al., 2011). Yet, in a fire-prone flora, such as the fynbos, nonlethal droughts may be of great importance in determining the success of post-fire recruitment. Seeder life histories (i.e. nonsprouters) are common in the fynbos, possibly due to the reliable winter rainfall in this region (Ojeda, 1998; Cowling et al., 2005; Ojeda et al., 2005). While drought may not result in outright mortality of plants, compromised reproductive output may result in local extinction following fire. An example of this can be seen in E. pyxidiflora, where the drought delayed maturation of the seedlings, resulting in no flowers being produced (Table 2, Fig. S4). A fire in this system following the drought could result in local extinction of this species. For resprouters, the impacts of drought on post-fire recovery is less clear, but is likely to be tightly coupled to the plants carbon reserves (Verdaguer & Ojeda, 2002). Drought-induced carbon starvation is a controversial hypothesis (Sala et al., 2010); however, reduced photosynthesis and increased water stress during drought would likely result in a lower ability to store carbon, negatively impacting resprouting ability post-fire. Thus, nonlethal drought may have considerable impact on the vegetation of this frequently burnt flora, possibly shifting the dominance between seeders and re-sprouters. We suggest that in fire-prone systems, examination of the impacts of nonlethal drought on post-fire regeneration, particularly the interaction between hydraulics and carbon metabolism (Sala et al., 2010; McDowell et al., 2011), is an important avenue for future research.

The multiple lines of evidence collected during our experiment (from physiological traits to fitness outcomes) illustrate three key findings. First, in a broad sense, species representing tall proteoid shrubs (L. laureolum), small ericoid shrubs (Erica species) and restioids (S. cernua and H. aristatus) appeared to have distinct ‘response pathways’ (Fig. 6, inset) lending support to the use of these functional groups for predicting drought response. Of these three major growth forms, we suggest that shallow-rooted, anisohydric seeder species (e.g. Erica) will be the most sensitive to future increases in summer drought. Second, nested within these broad response groups is a considerable range of responses to drought (Fig. 6). In several instances, common physiological traits resulted in very different fitness outcomes (e.g. E. ericoides vs E. subcapitata). These apparent contradictions most likely arise from two main causes: (a) variables that appear as discrete are in fact a continuum, which we lack the power to resolve in this study. For example, the degree of isohydry and anisohydry is represented here as a somewhat artificial dichotomy (Fig. 6), rather than a continuum, mainly due to an inability to resolve the differences between species more finely, and (b) consideration of additional traits is necessary for determining species-specific drought responses. In the case of D. divaricata and L. laureolum, we hypothesize that growth form is a key trait influencing drought tolerance. In the case of Erica species, it is likely that variation in xylem vulnerability to cavitation is important in resolving their range of drought response. Third, and lastly, we must consider that microsite characteristics could strongly influence drought impacts in these major growth forms. Shallow water tables or high-altitude sites with summer cloud inputs may buffer certain elements of the flora (e.g. deep-rooted proteoids and shallow-rooted restioids, respectively) from increased summer drought. The extent to which drought may impact the flora in areas without these moisture resources is as yet unknown. Importantly, simple bioclimatic modelling approaches are unlikely to be able to capture these subtleties, and this approach needs careful amendment if it is to be applied in future. While our study has indicated that hydraulic strategy is a useful approach for catagorizing drought responses in a biodiverse area, it remains a significant challenge to scale-up from detailed physiological studies to regional predictions to enable suitable risk assessment in future.

Figure 6.

A graphical synthesis of the data presented in this study on functional responses to drought in a Mediterranean-type shrubland in South Africa. These ‘response pathways’ visually depict the variations in trait combinations and consequent fitness outcomes (mortality and flowering) for the species measured in this study. The position of each species in each tier was based on data presented in previous figures and is designed to reflect relative, rather than absolute, differences. Where symbols overlap, no difference in the value of the trait is inferred, and separation of symbols is simply for clarity. Inset: a simplified and conceptual view of the distinct pathways followed by the key functional groups in this study (P, proteoid; E, ericoid; R, restioid).


Funding was generously provided by the Andrew W. Mellon and National Research Foundation, South Africa. Site access was provided by SANParks and the South African Navy. We acknowledge dedicated assistance from C. Moseley, E. Kleyhans, R. Skeleton, U. Mutzek and numerous field assistants.