High leaf mass per area of related species assemblages may reflect low rainfall and carbon isotope discrimination rather than low phosphorus and nitrogen concentrations


  • Byron B. Lamont,

    Corresponding author
    1. Department of Environmental Biology, Curtin University of Technology, PO Box U1987, Perth WA 6845, Australia,
    Search for more papers by this author
  • Philip K. Groom,

    1. Department of Environmental Biology, Curtin University of Technology, PO Box U1987, Perth WA 6845, Australia,
    2. Centre for Ecosystem Management, School of Natural Sciences, Edith Cowan University, Joondalup WA 6027, Australia, and
    Search for more papers by this author
  • R. M. Cowling

    1. Department of Environmental Biology, Curtin University of Technology, PO Box U1987, Perth WA 6845, Australia,
    2. Terrestrial Ecology Research Unit, Department of Botany, University of Port Elizabeth, PO Box 1600, Port Elizabeth, South Africa
    Search for more papers by this author

†Author to whom correspondence should be addressed. E-mail: rLamontB@curtin.edu.au


1. Leaf morphology at the site/species level should reflect environmental constraints on plant growth. One of the oldest controversies in ecology is the environmental basis for sclerophylly. The dominant view (Beadle’s theory) is that it has a nutritional, rather than a drought, basis, especially low phosphorus.

2. Using leaf mass per area (LMA) as an index of sclerophylly, we assessed its relationship with leaf phosphorus (P) and nitrogen (N) along extensive rainfall gradients in southwestern Australia and the Cape of South Africa. Leaf 13C/12C discrimination (Δ13C), as an index of intrinsic water-use efficiency, was also examined in the Cape. All Hakea species (Proteaceae) were sampled at 10 sites in Australia (96 species), and all Proteaceae at 14 sites in the Cape (82 species). All were evergreen shrubs with iso(bi)-lateral leaves.

3. In each region there was a strong (inverse) curvilinear relationship between mean LMA per site and mean annual rainfall and Δ13C, but none with mean P or N on a mass basis (although P and N on an area basis declined with rainfall). The Cape study was a particularly good test of Beadle’s theory, as P varied as much between sites as rainfall, and more between sites than within sites.

4. Leaf thickness and dry density were not as well correlated with rainfall as LMA, and leaf area and mass showed no relationship with rainfall. Area and mass had much greater variation within sites than between sites, limiting their value in plant–environment studies, while LMA was the most site-stable of the eight leaf attributes measured, except for Δ13C.

5. For all species considered individually in each region, there was a similar pattern as the site level, with LMA most strongly correlated (negatively) with rainfall and Δ13C and (positively) with leaf thickness, but no consistent relationship with P, N or density.

6. We conclude that when water and nutrient supply vary independently in the field, rainfall (as an index of water status) and Δ13C may be more closely correlated (inversely) with level of sclerophylly than nutrient status among evergreens, so that the role of sclerophylly as a drought adaptation warrants further consideration.


Many environmental factors affect leaf morphology, while only species with a certain leaf structure may be able to survive in certain habitats (Roderick, Berry & Noble 2000). As different environmental factors may affect leaf morphology in the same direction and act simultaneously, it may be difficult to identify the most important (limiting) ones (Fonseca et al. 2000). One of the oldest controversies in ecology involves Schimper’s (1903) assertion that species with tough, thick leaves (highly sclerophyllous) are the product of mediterranean climates with their long summer droughts. The interpretation of sclerophylly as an adaptive response to low water availability has subsequently received much support in the field and experimentally (Connor & Doley 1981; Fonseca et al. 2000; Groom & Lamont 1997; Oertli, Lips & Agami 1990; Poole & Miller 1975; Salleo, Nardini & Lo Gullo 1997; Specht & Specht 1989). However, the classic paper of Beadle (1966) resulted in a paradigm shift, with low soil phosphate (P) rather than low water supply held responsible for the distribution patterns of highly sclerophyllous vegetation. This interpretation has also received much support in the field and experimentally (Cowling & Campbell 1983; Fonseca et al. 2000; Loveless 1961; Specht & Rundel 1990; Stuebing & Alberdi 1973; Witkowski & Lamont 1991). Schimper’s view on the cause of sclerophylly became known as the ‘mediterranean myth’ (Johnson & Briggs 1975), and Beadle’s theory is now widely accepted. The most powerful support has come from non-mediterranean regions with nonseasonal rainfall but nutrient-impoverished soils and relatively sclerophyllous vegetation.

We chose to test Beadle’s theory along rainfall gradients (250–1200 mm per annum) in southwestern Australia and the Cape of South Africa, that is, the level of sclerophylly increases with decreasing P rather than decreasing rainfall (water availability). Two possible outcomes could be disproved (hypothetico-deduction) by this approach: P remained stable so sclerophylly was fixed, independent of rainfall; P and rainfall varied independently, but sclerophylly varied only with P. Principal components analysis (PCA) describes gradients through multivariate data and is an ideal tool for distinguishing these possibilities. There was the risk that P and rainfall would vary together, which would not test the theory. In this case, separate gra-dients would need to be sought (Fonseca et al. 2000) before using PCA. We included a comparison with nitrogen (N) as well as P, as this nutrient is also considered limiting in many systems and is often used in functional studies of leaf texture (e.g. Niinemets 1999). To minimize possible confounding due to taxonomic, leaf or life-form unrelatedness between sites, we restricted our study to co-occurring species in the genus Hakea (Proteaceae, subfamily Grevilleoideae) in Australia and Proteaceae, subfamily Proteoideae (mainly Leucadendron and Protea) in South Africa – all evergreen shrubs. Our index of sclerophylly was leaf mass per area (LMA) which includes the two recognized components of sclerophylly, leaf thickness and dry density, and is not confounded with nutrient content (Cowling & Campbell 1983; Edwards, Read & Sanson 2000; Groom & Lamont 1999; Witkowski & Lamont 1991). Beadle was unable to identify a suitable index of sclerophylly, instead separating hard-leaved, xeromorphic from soft-leaved, ‘rainforest’ species.

As the level of sclerophylly should be a direct response to internal levels of nutrients (Groom & Lamont 1999; Loveless 1961), we analysed leaf P and N levels rather than soil P and N, which are strongly correlated anyway, especially for plants of similar life form (Beadle 1966; Cunningham et al. 1999; Foulds 1993; Lamont 1995). It is clear that Beadle used extractable P as a surrogate for leaf P and noted ‘the leaves of species growing together have similar P contents’ and ‘the P content of the leaf (is) correlated with the P content of the soil’. Handley et al. (1999) even advocated leaf N content as a better index of fertility than soil analyses. In addition, all our species had proteoid root systems, giving them a similar mechanism for enhancing uptake of soil P in particular (Lamont 1993). Other problems with soil sampling, avoided by using leaf nutrients are (i) dependence on the extraction technique used – total and ‘available’ P are usually unrelated; (ii) the great variation expected within sites (Beadle 1966); (iii) effects of time of sampling in relation to rain and subsequent storage on observed values; and (d) inability to dig in rocky substrates (often encountered in our study) or sample those parts of the soil profile where most nutrients are absorbed.

Because net discrimination against the heavier isotope of carbon in photosynthesis (Δ13C) has been used as an index of climate control on stomatal conductance (Miller, Williams & Farquhar 2001; Panek & Waring 1997; Robinson et al. 2000), this was included in the Cape study (the Hakea material was unavailable). Δ13C is largely a function of the concentration of leaf CO2 (ci) which is controlled by both stomatal conductance and photosynthetic capacity. As intrinsic water-use efficiency (Wi) is a function of –ci (Ehleringer, Hall & Farquhar 1993; Feng 1999), it follows that Δ13C is a function of –Wi. Use of related species should reduce confounding due to possible differences in say, growing season and photosynthetic capacity. The design enabled possible nutrient effects on Δ13C (Radin 1984) to be assessed but not controlled. We used PCA to identify whether nutrients (P, N) or water-related constraints (rainfall, Δ13C) are most closely associated with variations in LMA at the local species assemblage level, the context of Beadle’s theory, as well as at the individual species level.

Materials and methods

Ten sites with Hakea species in abundance were selected along two gradients totalling 1300 km in southwestern Australia (Fig. 1). Fourteen sites with Proteaceae in abundance were selected along a less clearly defined gradient of 400 km in the western Cape of South Africa (Fig. 1). All site floras were evergreen and all species analysed were shrubs growing in the open. Sites were chosen to represent the range of annual rainfall in each region. In Australia, the average rainfall of the nearest town for 1985–94 was determined from records provided by the Bureau of Meteorology, Perth. In the Cape, rainfall data were derived from the Computing Centre for Water Research, University of Natal, as long-term (> 30-year) records from nearby stations. In those cases where stations were distant from our sampling sites, estimates were derived from interpolations of mean annual precipitation at the 1′ × 1′ scale (Dent, Lynch & Schulze 1989).

Figure 1.

Location of study sites in (a) southwestern Australia; (b) the Cape of South Africa.

At each site, all species present in the selected taxa were sampled, ranging from four to 16, with a mean of 10 in Australia and six in the Cape. Species with very hairy leaves (two in all) were avoided as this was considered an alternative strategy to sclerophylly in conserving water. Where the sexes of dioecious species had distinctive foliage (some Leucadendron, Aulax) they were treated as separate species. At most sites, the species sampled occurred over an area of 100 × 100 m, but in some flat areas (with negligible change in growing conditions) species were separated by up to 2 km. All sites were sampled in spring (September–November) from the previous season’s growth, that is, all leaves had passed through one summer’s drought. Flowering or fruiting branches were avoided, while only upper, sun-exposed branches were included. All leaves were iso(bi)lateral and rarely possessed protruding midribs or petioles (which were excluded if they were prominent). Five to 20 branches from a total of five representative, mature plants per species were collected and placed in sealed plastic bags.

Later that day, the thickness of the lamina at the longitudinal midpoint of one to two representative leaves per branch per plant were determined with vernier calipers. The transverse shape of needle leaves was noted for future correction of leaf volume. Avoiding mottled or damaged leaves, all previous season’s leaves were removed from their stems and bulked in equal proportions per plant. They were kept in rolled-up plastic bags for a further 1–2 days, then counted as they were placed diagonally on the conveyor belt (Li-Cor 3000, Lincoln, NK) or image plate (DIAS II, Delta-T Devices, Cambridge, UK) of an area meter. They were then placed in paper bags and dried at 60–70 °C for 2 days and bulked dry mass taken. However, area, mass and thickness for the hakea material were treated on an individual leaf basis. Leaf mass : area ratio (LMA) was determined, correcting for shape of the needle leaves (Witkowski & Lamont 1991). Dry leaf density (dry mass per volume) was determined from LMA divided by thickness (Witkowski & Lamont 1991).

Dried leaves were milled to pass through a 1 mm sieve and a 2 g subsample was assayed for total P (molybdenum blue colorimetry) and N (Kjeldahl digestion and titration) by CSBP Farmers Co., Perth or the Botany Department, University of Cape Town (methods after Grimshaw, Allen & Parkinson 1989). In addition, 0·5 mg of the Cape subsamples were analysed in duplicate for the stable isotopes 12C and 13C and the standardized ratio, δ13Cleaf, was determined (Farquhar, O’Leary & Berry 1982). This was converted to the simplified discrimination value, Δ13C, as δ13Cair (= −8) − δ13Cleaf. The effect was noted of allowing for altitude by subtracting 0·417 per 1000 m (Körner, Farquhar & Roksandic 1988) from this value. A mass spectrometer (Finnigan MATT 252) was used after tissue combustion in a Fisons CHN analyser under continuous flow and compared against the PDB standard (Ehleringer & Osmond 1989).

Mean LMA values per site were plotted against equivalent rainfall, P, N or Δ13C per region, and the best-fit curve (linear, exponential, power, log) was applied and its significance level noted. P and N were converted to leaf area (A) basis and plotted against rainfall in the same way. As the power function was the best fit between LMA and rainfall, A, mass (M), thickness, density and Δ13C, these were log-transformed for subsequent analyses. As the four sites along the south coast of the Cape had substantial summer rain (35–40% of annual) and had different patterns from the rest, they were kept separate in most analyses. To compare with the model of Specht & Specht (1989), LMA was calculated from evaporative coefficients resulting from simultaneous reductions in rainfall and increases in temperature for overstorey eucalypts in open-forest to open-scrub in mediterranean Australia (Specht & Specht 1995). To see if other meteorological variables might offer a better explanation than rainfall for variation in LMA, annual temperature and daily radiation (Bureau of Meteorology 1989), direct sunshine hours (Anonymous 1970), and evaporation (Lamont & Connell 1996), obtained by interpolation for the Australian sites, were all subjected to PCA.

Coefficient of variation (CV) per region was determined within each site for A, M, LMA, thickness, dry density, P, N and Δ13C, and their means taken. These were compared with CVs between site means, including rainfall, to identify cases where mean between-site CV exceeded within-site CV. The two data sets were submitted to PCA with vectors (Podani 1995). To test the working hypothesis, all mean LMA, rainfall, P and N (and Δ13C) per site per region were ordinated separately. LMA was ordinated with other leaf properties (A, M, thickness and density) as well as P and N to determine which showed the best relationship with rainfall. All attributes available per species (96 in Australia and 82 in the Cape) were subjected to separate PCAs per region to determine their overall relationship.


There was a highly significant (P < 0·001 for one independent variable) negative power function relationship between LMA and rainfall for both the Australian and Cape data (Fig. 2a). The four Cape sites with substantial summer rain had a lower LMA compared with the best-fit line for the rest of the data representing equivalent total rainfall. The two best-fit lines tended to converge at high rainfall, and to diverge at lower rainfall due to the greater (double) slope of the Australian line. LMA calculated from evaporative coefficients for five rainfall sites in mediterranean Australia had a similar trend to the Australian data, but the slope was slightly steeper (Fig. 2a). There was no significant relationship between LMA and leaf P (Fig. 2b), the best-fit line explaining 30·3% (P = 0·1, Australia) and 6·4% (Cape) of variance, and N (Fig. 2c), 10·0 and 3·2% of variance. There was a highly significant (P = 0·005) slightly curvilinear relationship (negative power function) between LMA and Δ13C (Fig. 2d). When the four southern Cape sites were included, the fit increased from 59·9 to 70·7% of variance. When corrected for altitude, Δ13C varied by up to 0·49% between sites. The best-fit line was parallel to the uncorrected one, and r2 was 67·1%. There was a significant power function relationship between P/A and N/A versus rainfall (Fig. 2e,f).

Figure 2.

LMA versus (a) rainfall; (b) leaf P; (c) leaf N; (d) leaf Δ13C. Rainfall versus (e) P per leaf area; (f) N per leaf area. •, Australian site data; ○, Cape data excluding south coastal sites (except d–f); □, south coastal sites. ▴, LMA for eucalypts versus rainfall calculated from evaporative coefficients at increasing temperatures in Specht & Specht (1995). Equations and coefficients of determination (r2) for the most significant curve fits are included; (d) includes the best-fit line (thinner) when Δ13C is corrected for the effect of altitude.

Coefficient of variation for area (A) and dry mass (M) varied more than rainfall between sites, especially in Australia (Table 1). P varied as much between sites as rainfall in the Cape, but much less in Australia. LMA was much less variable than rainfall in both regions, with Δ13C the most stable of all measured. Within-site CV for A and M varied much more than between-site CV, especially in the Cape (Table 1). Only density, P and Δ13C in the Cape varied less within sites than between sites. Within-site CV for LMA was the most stable of the variables measured, except for density and Δ13C in the Cape.

Table 1.  Coefficient of variation (%) at all sites for leaves of Hakea in southwestern Australia and of Proteaceae in the Cape
Coefficient of variationRegion (replicates)RainfallAreaMassLMAThicknessDensityPNΔ13C
Between-site meansAustralia (10)46·2 70·2 72·625·133·329·515·010·6
Cape (14)42·6 48·5 46·718·016·215·843·016·34·7
Within-site valuesAustralia (4–16)104·3 98·831·944·642·042·932·6
averaged across all sitesCape (4–11)126·8138·625·128·013·828·126·23·5

PCA of the Australian data accounted for 90% of variance and confirmed strong gradients in opposite directions for LMA and rainfall with a lesser (P) or negligible (N) relationship with nutrients (Fig. 3a). PCA of the Cape data accounted for 89% of variance and showed almost perfectly opposed gradients between LMA and rainfall, and Δ13C (the LMA and Δ13C vectors almost coincided) (Fig. 3b). There was a negligible relationship between LMA and P, and with N. Rainfall showed the strongest (negative) relationship with LMA, compared with the other four leaf attributes measured in both regions (Fig. 4). This was followed (positively) by thickness and, to a lesser extent, density, and least (negatively) by A and M. PCA of the 96 Hakea species, explaining 56% of variance in 2D, showed strong gradients of leaf mass, area, thickness and LMA through the data (Fig. 5a). The only significant relationships with LMA were rainfall (negative) and thickness (positive). PCA of the 82 Cape species, explaining 57% of variance, showed strong gradients of Δ13C, rainfall, A, M, LMA and thickness through the data (Fig. 5b). There were strong relationships between LMA and Δ13C and rainfall (negative), and thickness, M, A and density (positive), in that order. LMA was most closely correlated with log rainfall followed by direct sunshine (P = 0·003), weakly with radiation and evaporation, and negligibly with temperature (Fig. 6).

Figure 3.

PCA plus vectors for LMA, rainfall, P and N for (a) Australia; (b) Cape (plus Δ13C). The proportion of total variance accounted for by each component is shown, as well as r values for each variable with LMA.

Figure 4.

PCA plus vectors for five physical leaf attributes, P, N and rainfall for (a) Australia; (b) Cape (plus Δ13C). See Fig. 3.

Figure 5.

PCA plus vectors for (a) eight attributes for the 96 Australian species analysed; (b) nine attributes of the 82 Cape species analysed (including the four south coastal sites). See Fig. 3.

Figure 6.

PCA plus vectors for five meteorological variables and LMA for the Australian sites. See Fig. 3.


In both regions there was a much stronger (inverse) relationship between LMA and rainfall than between LMA and leaf P or N. In line with the mathematical relationship between LMA, leaf thickness and dry density (Witkowski & Lamont 1991), the data also show an empirical relationship, although of the three LMA was best correlated with rainfall. Thicker and/or denser tissues are common responses to reduced water availability at the individual species level (Abrams, Kubiske & Mostoller 1994; Gravano et al. 2000; Groom & Lamont 1997; Witkowski & Lamont 1991). It is therefore to be expected that species (at constant altitude; Körner 1989) with thicker and denser leaves (higher LMA) will also occur where rainfall is reduced (Abrams 1994; Midgley, van Wyk & Everard 1995; Reich et al. 1999). This was underscored by the four southern Cape (cooler) sites with significant summer rain, and more effective rain throughout the year (Deacon et al. 1992), where LMA was lower (forming a best-fit line with a lower slope) than the line fitted to sites with a more marked seasonal pattern.

In the Cape, Δ13C was closely correlated with rainfall at the site level, in the same direction as other gradient studies although with differently shaped slopes (Miller, Williams & Farquhar 2001; Schulze et al. 1998; Stewart et al. 1995). Δ13C was also inversely related to LMA (as in Schulze et al. 1998), suggesting a physiological relationship between leaf structure and rainfall. Four mechanisms that decrease discrimination against the heavier isotope through their effects on decreasing ci are possible: (i) with lower soil water availability, stomatal conductance will decrease; and, with thicker palisade and denser tissues, (ii) there is greater demand for carbon per unit leaf area, (iii) internal conductance will decrease (Hanba, Miyazawa & Terashima 1999) and (iv) resistance to water flow will increase, reducing turgor and stomatal conductance. Nativ et al. (1999) alluded to (i) and (ii) to explain decreasing Δ13C with increasing drought in the sclerophyll, Acacia saligna. The thickness component of LMA as it affected (ii) and (iii) was used by Hultine & Marshall (2000) to explain a similar correlation observed with changes in altitude. Our data show a negligible effect of altitude in explaining the results (Fig. 2d).

At an individual species level, LMA (r = −0·4284), thickness (r = −0·2706) and density (r = −0·2505) were correlated with Δ13C in our study (Fig. 5b). Density was closely correlated with Δ13C and LMA in Hakea psilorrhyncha in response to osmotically induced drought (Groom & Lamont 1997), consistent with the results here. Nutrition can also affect ci through its effects on photosynthetic capacity (Radin 1984). There was no relationship between Δ13C and P or N at the site level (confirming that water availability is the main environmental constraint), but at the individual species level P (r = −0·3107) but not N appeared to be limiting photosynthesis.

Within species, a reduction in cell size and increase in density are often associated with inadequate mineral nutrition (Atkinson & Davison 1972; Witkowski & Lamont 1991). But an increase in thickness as a response to low P or N is less certain. An associated reduction in thickness is more likely (Atkinson & Davison 1972; Beadle 1968; Lamont & Kelly 1988; Witkowski & Lamont 1991). Thus the two components of LMA should respond in opposite directions to a reduction in P and N availability, at least for the within-species level. This might explain why LMA shows such a poor relationship with P and N compared with rainfall and Δ13C. In fact there was no relationship between density or thickness with P or N at the site level (Fig. 4). At the species level, only thickness and N in Australia were correlated (r = −0·3238) but in the ‘wrong’ direction. Nevertheless, other intersite comparisons show strong correlations between leaf specific area, the inverse of LMA, and N concentration (Reich et al. 1999; Turner 1994) while Fonseca et al. (2000) recorded greater thicknesses at lower soil P. These studies omitted measurements of thickness and/or density, preventing further comparison with our results. With current interest in whether morphological responses to ‘stress’ are general or specific to the type of stress (Fonseca et al. 2000; Roderick et al. 2000), the relative responses of thickness in particular to water and nutrient gradients seems a worthy topic to pursue in more detail.

Nutrient concentrations on a leaf-area basis are usually more responsive to environmental gradients than on a mass basis (Körner 1989; G. Farquhar, personal communication). Our results showed a significant (negative) relationship between P/A and N/A and rainfall at the site level (Fig. 2e,f). N/A is equivalent to (N/M)LMA. So, even where N/M is independent of rainfall, as shown here and by Foulds (1993), N/A will increase with decrease in rainfall due to the LMA effect – more mass per unit surface area.

The Cape data were a particularly good test of the nutrient hypothesis, as leaf P varied as much between sites as rainfall (CV = 43%), and more between sites than within sites. Equally, P varied less in Australia and N less in both (11–16%), but LMA failed to be more stable or independent of rainfall. The Australian study would have benefited from a greater range of soil types, but a problem is the preference of Proteaceae for the most nutrient-impoverished soils (Lamont 1993). The quality of the answer to the question ‘which of rainfall and nutrients is more closely linked with sclerophylly in the field?’ depends on how representative the study sites are of the area in which the target group occurs. Our study covered the full geographic range in which the Proteaceae occur at sites where it is most abundant, so representativeness was not an issue.

Thus it appears that assemblages of related species in drier climates are likely to have a higher LMA, independent of their overall P and N status. If we accept LMA as a suitable index of sclerophylly (Groom & Lamont 1999; Edwards et al. 2000), then we have grounds for questioning the universality of the Beadle hypothesis. We have demonstrated that rainfall, as an index of water availability, can have much more value than nutrients in explaining varying levels of sclerophylly in two regions renowned for their impoverished soils. This was supported by the Δ13C results for the Cape, but this appeared to reflect both water availability and leaf morphology, so was not independent of sclerophylly.

This is not to say that LMA–nutrient gradients do not exist at other scales. The Australian plants had lower levels of leaf P and N than the Cape plants (Fig. 2b,c) which could contribute to their higher LMA at the regional level (but see below). Annuals, forbs and (summer) deciduous species (absent from the Proteaceae) are most likely to occur in more fertile soils, or to exploit only the more fertile (surface) parts of the profile (Aerts 1990; Specht 1963) and avoid the summer drought. Their leaves have higher P and N concentrations, lower levels of sclerophylly, and (by definition) shorter lifespans than evergreens (Körner 1989; Lamont 1995; Reich, Walters & Ellsworth 1997; Specht 1988). Thus comparisons that included different families and functional groups might instead highlight the role of nutrients in controlling LMA.

The decoupling of LMA from (inverse) nutrient levels at the site and species scales in our study indicates that the water-storage function of sclerophylly should be considered along with any nutrient-storage function. Beadle (1966) referred to the ‘water storage’ tissue of needle-leaved hakeas (Groom, Lamont & Markey 1997); and Lamont & Lamont (2000) showed how water storage is a function of leaf thickness. This, and the drought-resistance function of sclerophylly (Oertli et al. 1990), will be important only in evergreens, and it would be of interest to learn if LMA remains coupled with leaf lifespans exceeding 12 months, despite the decoupling with N (Reich 1993). In southwestern Australia, leaves of some Proteaceae may survive for up to 13 years (Witkowski et al. 1992).

Similar rainfall–LMA relationships have been reported in other regions without reference to possible accompanying N and/or P gradients (Abrams 1994; Midgley, van Wyk & Everard 1995; Reich et al. 1999). The application of Specht & Specht’s (1989) evaporative coefficient to equivalent sites in mediterranean Australia produced a similar trend with LMA as ours, except that the slope was steeper (Fig. 2a). This similarity is surprising as it was based on eucalypts (Myrtaceae), which are deep-rooted trees rather than shrubs, and the coefficient uses evaporation, conductance and canopy cover rather than rainfall. The reduced slope for our data could be due to temperature not rising with a fall in rainfall (Fig. 6) and/or the fact that we fitted a power rather than an exponential function to the data. Smith et al. (1997) obtained higher levels of sclerophylly (thicker leaves) in five south-western Australian communities with an increase in direct sunshine (rainfall not considered). Our data suggest that radiation is a secondary constraint to rainfall (Fig. 6), although it may interact with water availability in affecting LMA at the species level (Groom & Lamont 1997; Rôças, Scarano & Barros 2001).

We can also comment on the significance of total versus seasonal rainfall as a component of the ‘mediterranean myth’. LMA fell (and merged in the two regions) as total rainfall increased, implying that its seasonality declined as effective rain occurs in more months of the year (Lamont & Connell 1996). Further, the four southern Cape sites noted above, with less seasonal but the same total rainfall, had a lower LMA. It could even be argued that the divergence of the south-western Australian from the Cape LMA values with decreasing rainfall was an outcome of the less effective (shorter growing season) rain in the former, as the study sites ranged closer to the Equator at 28–35° S (Lamont & Connell 1996) compared with 32–35° S in the Cape. We conclude that high sclerophylly may also be an expression of strong seasonality. Our results support Schimper’s (1903) original interpretation, and we suggest that a reappraisal of the ‘mediterranean myth’ is warranted.

Ideally, an attribute used for plant–environment studies at the regional scale should vary more between sites than within sites. There is a strong tradition of using leaf-area classes (Raunkiaer 1934) in vegetation science. However, along with leaf mass, our study showed that leaf area varied much more within sites than any other variable measured. This was despite taxonomic and structural matching. LMA tended to be more stable than its structural components within sites, but was slightly less variable between sites. No attribute consistently varied more between sites, but LMA seemed the most satisfactory of the eight plant attributes assessed, apart from Δ13C, for which only one data set was available.

While the earlier propositions were directed at the community and species levels, making our approach appropriate, another test of the ideas might arise from studies of highly plastic species with wide environmental tolerances. With such high levels of landscape and regional diversity (Cowling & Lamont 1998), sufficient examples would be difficult to identify. Alternative studies of ‘sister’ species would also not be at the right scale, but might provide evolutionary insights. Adequate replication would also be a problem, as each member of the pair would need to be collected from sites over its full range for a single comparison.


This project was supported by the Australian Research Council (B.L. and P.G.) and the National Research Foundation of South Africa (B.L. and R.C.). The South African part of the work was undertaken in the Botany Department, University of Stellenbosch. The assistance of Heather Lamont, Neil Eccles, Karen Esler, Mike Cramer, Myke Scott and Willy Stock was invaluable. We thank Michael Roderick and another referee for their helpful comments on the manuscript.