1. Edaphic specialization among species may lead to greater productivity and resource use efficiency across heterogenous landscapes than could be achieved in the absence of specialization. Although this idea has been tested conceptually and in garden experiments, it has rarely been examined in undisturbed forests.
2. To address this gap in our knowledge, we measured aboveground net primary productivity (aboveground biomass increment plus litterfall; ANPP) and phosphorus use efficiency (ANPP/P uptake) for stands on infertile schist soil and fertile basalt soil on the Atherton Tablelands, Australia. We also measured aboveground biomass production and estimated P use efficiency (PUE) for 52 tree species within these stands. Soil P fractions and radiation use efficiency (ANPP/percent intercepted radiation) were also measured.
3. Phosphorus use efficiency was markedly variable (CV = 44%) among species across soil types. Phosphorus use efficiency of obligate specialists on infertile soil was twice as high as species common on both soil types. Plastic responses within species were also significant, with trees on infertile soil having 45% greater PUE than trees on fertile soil. At the ecosystem level, genotypic and phenotypic traits accounted for 49% and 29% of the total PUE variance. Phosphorus-efficient trees (PUE >8 kg biomass g−1 P uptake) on schist soils contributed more to stand-level species richness (schist = 73%, basalt = 20%), basal area (schist = 86%, basalt = 18%) and production (schist = 82%, basalt = 10%) than did P-efficient species on basalt soils.
4. Forest on schist soils achieved higher PUE than forest on basalt soils by partitioning more P to leaves rather than wood and by retaining P for longer periods of time before losing it via tissue senescence. These PUE traits enabled forest on schist to achieve similar ANPP and radiation use efficiency (i.e. PUE was not traded for radiation use efficiency). It is possible that opportunity costs of high PUE may exist among other life-history traits.
5. These results suggest that plasticity in traits that confer P conservation is significant, but limited, and that maximum P conservation at the landscape level must be achieved via genetic differences between species. Although this highlights the importance of genetic conservation in forests, it also demonstrates that high P conservation is possible with relatively few, but markedly plastic species.
No species can maximize growth, reproduction and competitive ability in all environments (Funk & Vitousek 2007). As a result, plants trade fitness in one environment for fitness in another environment. Such trade-offs are likely to engender plant specialization (e.g. edaphic specialization and species–soil associations) and the coexistence of species (Tilman 2000). Species–soil associations in heterogeneous landscapes are common in temperate (Aerts 1999; Lusk & Matus 2000; Read 2001) and tropical (Baltzer et al. 2005; Paoli et al. 2006; John et al. 2007) forests. Although species–soil associations are widespread, there appears to be no general mechanism driving the divergence of soil specialists. However, Aerts (1999) suggests that a trade-off between growth rate and nutrient retention may partially explain the divergence of species favouring sites of low fertility. This is because traits that lead to high nutrient retention, such as slow nutrient turnover and low tissue nutrient concentrations, can also contribute to low growth rate (Grime 1997; Aerts 1999). Thus, the trade-off between nutrient conservation and growth rate may be an important source of tree diversity (Reich et al. 1992; Aerts 1999) in forested ecosystems.
Edaphic specialization can lead to higher ecosystem-level productivity (Tilman et al. 1997), but plasticity of nutrient use within species can result in similar effects on ecosystem productivity. For example, in a Metrosideros polymorpha Gaud forest along a nutrient gradient, plastic resource use efficiency increased productivity because resources were used more efficiently where they were more limiting (Herbert & Fownes 1999). In this study, we ask how much genotypic (intrinsic) and phenotypic differences in resource use contribute to ecosystem-level productivity in heterogeneous landscapes. This is an important question because the inability of species to adapt phenotypically to edaphic variation has an opportunity cost – reduced distribution on unsuitable substrates. Clearly, ranges of soil edaphic variation, phenotypic plasticity and intrinsic differences among species contribute to ecosystem-level functioning.
We measured ANPP ([litterfall + biomass increment]/time) and P uptake for trees, species and stands across a P availability contrast in a tropical rainforest in Queensland, Australia. We asked the following questions. (i) Do P use traits differ between edaphic specialists and edaphic generalists on the same substrate? (ii) How plastic are P use traits among edaphic generalists growing on contrasting soils? (iii) How much of the total variance in P use traits among generalists is attributable to phenotypic plasticity or intrinsic differences between species? (iv) How do these differences in P use manifest at the ecosystem level?
Materials and methods
Site description and species
Research sites were established within Wooroonooran National Park (17°22′S and 145°43′E) at a location c. 32 km inland from the coast and between 700 and 800 m elevation. The evergreen rainforest in this area of the park is classified as complex mesophyll vine forest (Tracey 1982) and receives c. 3·5 m of annual rainfall, with c. 70% falling between the months of November and April. Species richness is high with >1000 tree species occurring within the regional area of north-east Queensland (Hyland et al. 2002) and c. 100 occurring within a 0·25-ha plot. The research sites are located in forest occurring across two soil types (schist and basalt), which are often directly adjacent to one another at the same elevation, slope and aspect. Soil fidelity among species in these forests has been compared using the procedure of Dufrêne & Legendre (1997) (S.M. Gleason, J. Read, A. Ares & D.J. Metcalfe, unpublished data). This method considers both the relative abundance and the relative frequency of species as they occur on each substrate type and is independent of the relative abundances of other species. All schist specialist (six species) favoured schist soil (P < 0·001). In this paper, we analyse two main groups of species arising from these analyses: (i) species occurring exclusively on schist soils (specialists) (n = 6) and (ii) species occurring on both schist and basalt soils (generalists) (n = 36). There are no obligate basalt specialists in the study area. Although species categorized as specialists are obligate schist specialists on our sites, they do occur on other P-poor substrates (e.g. weathered granite) outside the study area.
Plot establishment and fertilization
In September 2005, 24 forest production plots were located on two ridges (c. 300 m apart) that crossed the schist-basalt geological contrast described. Within each ridge, 12 circular 100-m2 plots were established, six plots on schist and six plots on basalt. Plots were placed c. 50 m apart, but distance varied somewhat as large gaps were avoided to homogenize basal area among the plots. The diameter at breast height (d.b.h.), species, height, and distance and azimuth from plot centre were recorded for every tree >5 cm d.b.h. (c. 25 trees/plot). Trees with a d.b.h. >10 cm were fitted with a stainless steel dendrometer band (Keeland & Young 2005). To determine whether forest productivity was limited by either N or P, half of the plots were fertilized with 100 kg ha−1 year−1 of P (as superphosphate) and N (as urea). Fifty percent of this fertilizer was applied at the start of the experiment (October 2005) and the remaining fertilizer was applied midway though the experiment (June 2006).
Soil description and analysis
Soils are derived from basalt (Red Ferrosol – Maalan series) and schist parent materials (Red Dermosol – Galmara series) (Malcolm et al. 1999). Whereas both soils have low pH, high organic C, moderate cation exchange capacities and high Al saturation, basalt soils have markedly higher P contents (labile and occluded pools) than schist soils (Table 1). Phosphorus limitation of plant growth on the schist soil has been previously demonstrated in a glasshouse experiment (Kerridge et al. 1972), in planted stands (Keenan et al. 1998; Webb et al. 2000), and is supported by plant stoichiometry, nutrient use traits and the distribution of schist specialists in these forests (Gleason et al., unpublished data).
Table 1. Sequential soil P fractionations and other soil variables for basalt and schist substrates
Values in parentheses represent 1 SE of the mean (n = 12). Variables for which no SE is given represent data from Malcolm et al. (1999) for the Maalan and Galmara soil series.
Mineral soil was collected (0–25 cm depth) on four different occasions during both wet (two collections) and dry (two collections) seasons from five random locations within each plot and combined into one bulked plot sample for each collection period. Soil P was fractionated using methods modified from Tiessen & Moir (1993). Although the most labile P fraction (NaHCO3-extractable P) varied seasonally, seasonal variability was similar between soil types and we present only dry season data here (Table 1). Additional details of these methods are given in Appendix S1 (Supporting Information).
Cation exchange capacity and Al saturation were determined using the ammonium acetate method (Malcolm et al. 1999). Total organic carbon and N were determined using an induction furnace and Kjeldahl methods, respectively (Laffan 1988). Soil pH was measured in a 1:2 soil to water (mass) mixture after allowing the solution to stand for 1 h.
Tropical Cyclone Larry struck Wooroonooran National Park on 20 March 2006, c. 6 months after plot establishment. The cyclone was downgraded from category 4 to category 3 by the time it passed within 17 km of the study plots, all of which were damaged to some degree. Damage included stripped leaves from branches, broken branches, uprooted trees and stem breakage. Post-cyclone mortality was 11·9% on basalt plots and 7·1% on schist plots (Gleason et al. 2008). Damage to surviving stems was usually slight (less than 20% crown damage), although occasionally moderate to severe damage was recorded. Damage was also species-specific (Gleason et al. 2008; Metcalfe et al. 2008) and more severe on basalt soils than schist soils (Gleason et al. 2008). Although production declined significantly immediately following the cyclone, recovery was relatively rapid, with leaf area index (LAI) increasing to 84% of pre-cyclone values within 8 months (Gleason et al. 2008).
Samples for soil and tissue nutrient analysis were collected prior to the cyclone and represent pre-disturbance conditions. Biomass production was independently assessed for two periods: pre-cyclone production (6 months) and post-cyclone production (18 months). Data presented herein represent the entire 24-month production period (pre-cyclone + post-cyclone), unless hypothesis tests differed using pre-cyclone and entire period data sets. In these cases only pre-cyclone production values are given, but we note where we have done so.
Aboveground net primary productivity (ANPP)
Litterfall was collected every 2 weeks from 12 530 cm2, self-draining baskets within each plot. Litterfall baskets were staked 20 cm above the forest floor in a rectangular grid. Because of logistic constraints, litterfall was occasionally (five times) collected monthly. Decomposition during these longer collection periods may have resulted in underestimates of litter production. However, because the objective of this study was to compare soil-specific production, small decomposition losses are unlikely to affect the hypotheses tests. Litter was dried at 65 °C in a fan-forced oven to constant mass. Species-specific litterfall mass was calculated for dry season, wet season and wet + dry season collections.
Aboveground biomass was calculated using a general allometric equation developed for lowland rainforest in West Kalimantan, Indonesia (Yamakura et al. 1986).
where WS, WB and WL represent dry mass (kg) of trunk, branches and leaves, respectively. D represents the d.b.h. (cm) and H represents height (m). These equations have been used to estimate biomass in old-world tropical rainforests (Yamakura et al. 1986; Yamada 1997; Kitayama & Aiba 2002), as well as in a temperate forest in Japan (Aiba & Kitayama 1999). We evaluated other general allometric equations (Brown 1997; Chave et al. 2005), including equations developed for Australian forests, but selected the Yamakura equation because it best represented the floristics and structure (basal area, height) of our forest and included both leaf and shoot components of aboveground biomass.
Errors incurred from using general allometric equations instead of species-specific functions generally fall within the range of error reported for individual regressions (Pastor et al. 1984). Although site-specific and species-specific allometric differences are not uncommon (Keith et al. 2000; Montagu et al. 2005), within-site general allometric equations used for wet tropical forests in Australia are not likely to introduce significant amounts of error (>10% difference) for trees <60 cm d.b.h. (Keith et al. 2000).
Stand-level ANPP was calculated by summing litterfall + biomass increment (oven-dry mass) for the period before the cyclone (6 months), the period after the cyclone (18 months) and the entire production period (24 months). Productivity of individual species was estimated using only the biomass increment (litter production was not included). In addition to stand-level production calculations, relative growth rates (RGR) were also calculated for individual stems following a multivariate method modified from Maguire et al. (2006) (details available in Appendix S1).
Leaf area index and tissue analysis
Leaf area index values were estimated using hemispherical photos and Hemiview® software (Dynamax Inc., Houston, TX). This software uses an inversion model (Campbell 1986) to back-calculate LAI through Beer’s Law (Delta-T Devices Ltd 1999). Photographs were taken 1 m above the forest floor using a Nikon Coolpix® (Nikon Inc., Melville, NY, USA) digital camera at five fixed locations within each plot, once before the cyclone and on four occasions after the cyclone. Pictures were taken early in the morning, before the sun rose to a position where it was visible in the photographs.
We determined how much basal area was represented by each tree species in every plot. Species were then chosen for tissue analysis so that at least 90% of the basal area for each plot included species for which leaf and wood nutrient data were determined. From this analysis, 54% of the species occurring on basalt (36 of 67) and 57% of species occurring on schist (37 of 65) were chosen for leaf and wood nutrient analysis. For the remainder of the species within each plot (<10% of BA), average leaf and wood nutrient data were used for plot-level calculations [i.e. nutrient uptake and P use efficiency (PUE)].
Green leaves were collected from branches that had fallen within the plots 1 day after Cyclone Larry passed the study sites. Although every effort was made to sample from as many trees as possible from each soil type, occasionally (for six species) leaves were collected from fewer than three trees per soil type. Senesced leaves were collected daily from suspended shade cloth in non-fertilized plots. As such, these samples contained leaves from an unknown number of individuals within (or near) the plot. Leaves for each species were bulked into two samples (green and senesced leaves), dried at 65 °C to constant mass, ground in a ball mill and analysed for total C, N, P, K, Ca, Mg, Na, Mn, Zn, Mo, Cu, Fe and Al. Nutrients other than N and C, which were analysed with a CHN auto analyser, were analysed using inductively coupled plasma emission spectroscopy following microwave-assisted digestion with concentrated nitric acid. To determine how well our bulked samples estimated mean species nutrient concentrations, for each of six species, four bulked leaf samples were collected and analysed separately.
Wood samples were removed from the outer 10 cm of sapwood using a cordless drill and a 1-cm diameter drill bit. Between five and eight trees of each species from each soil type were sampled. Wood samples for each species were bulked into one sample, placed in a paper envelope, dried to constant mass at 65 °C and ground in a ball mill. Nutrient analysis was performed using the same methods described above for leaf samples except that a combination of nitric acid + hydrogen peroxide solution was used during microwave digestion. To determine how well our bulked samples estimated mean species nutrient concentrations, for each of four species, four bulked wood samples were collected and analysed separately.
Measures of PUE AND RUE
Stand-level PUE (PUEstand) (Table 2) is most appropriately measured as biomass production per unit P taken up from the soil (Berendse & Aerts 1987). Phosphorus uptake was calculated as the sum of P mass in litter, new wood and new leaves. Reproductive structures were not analysed separately, but considering they contributed c. 6% of the total litterfall and c. 1·5% of ANPP, this is likely to have little effect on our PUE measurements. We calculated PUEstand following Vitousek (2004):
Table 2. Phosphorus use efficiency (PUE) indices
Calculation, description and use of PUE indices in this study
Stand-level production (wood increment + green leaf increment + litterfall) (kg) during the course of the experiment (2 years) divided by the P content (g) in wood increment + leaf increment + litterfall (i.e. P taken up from the soil). This index reflects how much biomass and litter was produced during the experiment per unit P taken up from the soil during the same time period. This index is used to compare PUE between basalt and schist stands. Although this is the most appropriate index of PUE at the stand level, the litterfall component cannot be calculated for individual stems and species in highly diverse forests without complex sampling designs (e.g. Osada et al. 2003). Therefore, PUE measurements for individual species were calculated without the litterfall component (PUEspecies)
Biomass increment (wood increment + green leaf increment) (kg) of individual trees divided by the P content (g) in the biomass increment. Biomass increment includes both wood and green leaf increment, but does not include litter production. This index is used to compare PUE between species and PUE plasticity within species
Initial P content (g) in the canopy divided by the total P content in litterfall (g) during the pre-cyclone period (0·5 year). Because the mass units cancel out after division, the formula yields the turnover time of canopy P in years. Although this equation can be used to calculate Pres in the entire plant (canopy + shoot + root), it is often applied only to P in the canopy. Assuming steady state conditions, this measure reflects how long a stand can use a unit of canopy P before it is lost via tissue senescence. This index is used to compare P residence times between schist and basalt stands and is not calculated for individual species
Stand-level production (wood increment + green leaf increment + litterfall) (kg) during the pre-cyclone period (0·5 year) divided by the initial P content (g) in the canopy. This index provides a measure how efficiently P in the canopy is used to produce biomass and litter. As such, it does not consider P pools in woody biomass or the partitioning of P between storage organs and leaves. This index is used to compare P productivity between schist and basalt stands and is not calculated for individual species
The product of Pres and Pprod. By combining Pres and Pprod, the formula simplifies to ANPP divided by the P content in litterfall. If it is reasonable to assume that P required for production is equal to P lost via litterfall (i.e. the stand is at steady state), CPUE is then equivalent to PUEstand. Thus, by breaking CPUE into its component parts (Pprod and Pres), it can be determined whether higher P efficiency is achieved by longer residence time (Pres) or greater canopy productivity (Pprod). This index is used to compare canopy PUE between schist and basalt stands and is not calculated for individual species
Additionally, we split PUE into two separate parts: (i) P productivity (ANPP/P in biomass) and (ii) P residence time (P in biomass/annual P requirement) (Berendse & Aerts 1987). We applied this equation to canopy P (i.e. ANPP/P in canopy) because canopies are where C assimilation takes place (Harrington et al. 2001) (Table 2). Canopy P pools were estimated using the previously described general allometric equation (leaf mass as a function of d.b.h.) and measured leaf P. We denote canopy P productivity as ‘Pprod’ and canopy P residence time as ‘Pres’. Integrated canopy PUE (CPUE) was calculated as the product of Pprod and Pres (Paoli et al. 2005) and represents ANPP produced per unit P uptake (canopy P), assuming steady-state conditions (inputs = outputs) (Table 2). Breaking PUE into its components allowed for a more detailed interpretation of contrasting PUE measures – did PUE differ because canopy Pprod differed or because Pres differed?
Phosphorus use efficiency for individual species is denoted as ‘PUEspecies’ (Table 2). Because litterfall rates for individual trees and species were not measured, PUEspecies was calculated after previous studies (Grundon 1972; Shaver & Melillo 1984).
Biomass increment in this equation includes new wood and new green leaves, but not litterfall. We feel this is a better plant-level measure of P conservation than leaf resorption or C:P ratios in litterfall. More than 90% of aboveground P in the canopy trees is sequestered in wood tissues, not leaves. It is likely that much of this P is mobile (Sauter & van Cleve 1994; Rosecrance et al. 1998), but is not being used for photosynthesis (i.e. it is not being used efficiently). Thus, large P pools in wood tissues are likely to reflect low plant-level PUE. Supporting this argument, there is good agreement between PUEspecies and PUEstand (r2 = 0·80) at the stand level.
Productivity per percent intercepted radiation (RUE) was used as a proxy measure for stand-level radiation conversion efficiency (ε). Percent intercepted radiation was obtained from pre-cyclone hemispherical photographs within stands as discussed above. When comparing RUE with PUEstand only the pre-cyclone data set was used. Because hemispherical photographs yield stand-level light interception, RUE values could only be calculated for stands, not individual trees or species. Comparisons of RUE are analogous to comparisons of ε, providing that compared stands receive identical levels of photosynthetically active radiation. This was a reasonable assumption because the stands were at similar elevation and not more than 500 m apart.
Traits (leaf P, wood P, biomass P, RGR and PUEspecies) were compared between soil specialists and soil generalists (including only measurements from schist forest) using a completely randomized anova and species averages as replicates. This analysis answers the question: do P use traits differ between edaphic specialists and edaphic generalists on the same substrate?
Trait (leaf P, wood P, biomass P, RGR and PUEspecies) plasticity across soil generalists was compared using a randomized complete block (RCB) anova, blocking by species. This analysis is equivalent to a paired t-test (i.e. species means on basalt vs. species means on schist) and answers the question: how plastic are P use traits among edaphic generalists growing on contrasting soils? Variance components among species (intrinsic differences) and within species (phenotypic differences) were then calculated for this anova (Sokal & Rohlf 1969). This analysis answers the question: how much of the total variance in P use traits among generalists is attributable to phenotypic plasticity or intrinsic differences between species? Because we used species means in this analysis, variance attributable to intrinsic and plastic components is likely to be overestimated if the unknown within species error was large. However, our analysis of variance within species (done for six species) revealed that standard deviations between bulked samples within species were insignificant compared to average plastic and intrinsic differences. Thus, our bulked tissue samples (collecting tissue from c. 6 individuals for each species on each soil type) provided low-error means.
Soil comparisons (basalt vs. schist) were made using average plot values (PUEstand, ANPP, P uptake, RUE, LAI, basal area and soil characteristics) in an RCB anova, blocking by ridge. This analysis addresses how species-level attributes (PUEspecies, RGR and basal area) manifest at the ecosystem level.
Intrinsic and plastic components of PUE
Schist specialists had higher PUEspecies, higher % P resorption, lower RGR and lower concentrations of P in leaves, wood and biomass than did generalists occurring on schist soils (Fig. 1). Additionally, schist specialists had 27% greater (P = 0·039) leaf P:wood P ratios than generalists occurring on schist soils, i.e. schist specialists allocated more P to leaf tissue relative to wood tissue than did soil generalists.
Phosphorus availability appeared markedly lower on schist soils than on basalt soils. Soil generalists occurring on schist had significantly higher PUEspecies and lower concentrations of P in green leaves, wood and biomass than did generalists growing on basalt (Fig. 1). Contrary to this pattern, generalists occurring on basalt had higher P resorption than generalists occurring on schist. Although tissue P concentrations for most species were plastic, the majority of tissue P variance was explained by intrinsic differences between species (Table 3). Wood P and PUEspecies (largely a function of wood P) were especially plastic within species (Table 3). Variance in green leaf P and % P resorption (withdrawing P from green leaves during senescence) was not well explained by either intrinsic differences between species or plastic responses within species.
Table 3. Variance components, species plasticity, and species means for nutrient and relative growth rate (RGR) measures. Variance components were calculated following an RCB anova, blocking by species (paired t-test), and using species averages for generalists occurring on each soil type. Thus, plastic and intrinsic values represent the percent variance attributable to either differences within species (plastic) or differences between species (intrinsic) across the basalt–schist soil contrast. ‘Within species plasticity’ displays minimum, maximum and mean plasticity within species. For example, ‘maximum plasticity’ denotes the difference between basalt and schist values for the most plastic species. ‘Species means’ displays the minimum and maximum differences among species. For example, ‘minimum’ PUEsp displays the average value (across soils) for the species with the lowest PUEspecies
% P res.
PUEsp, species-level P use efficiency; gl P, green leaf P; sl P, senesced leaf P; % P res, % P resorbed from green leaves prior to abscission.
Variance between soils
Within species plasticity
Nesting taxonomic levels (Pagel & Harvey 1988) revealed that species differences explained over 70% of the variance among nutrient measures and RGR. Wood P and % P resorption variables were exceptions, with 34% of the variance associated with families and 42% of the variance associated with genera respectively. Thus, species was the most appropriate level for taxonomic comparisons.
Scaling up from species to stand-level functioning
Aboveground net primary productivity was not significantly affected on either soil type by fertilizing with superphosphate and urea. Following the cyclone, some species showed a decrease in RGR in fertilized plots, which may have been associated with species-specific P-sensitivity, common in Australian plants adapted to P-poor soils (Handreck 1997; Lambers et al. 2008).
Although stand-level species richness was almost identical on basalt and schist soils, the majority of species on schist soils had high PUEspecies (>8 kg g−1), whereas the majority of species on basalt had low PUEspecies (<8 kg g−1) (Fig. 2a). These common P-efficient species on schist represented the largest proportion of stand-level basal area, whereas the majority of basal area on basalt was attributable to species with low PUEspecies values (Fig. 2b). Scaling these attributes up to stand level (Fig. 3) demonstrates the importance of high PUEspecies species on schist – species with high PUEspecies on schist take up markedly less P than species on basalt (Fig. 2d), yet produce similar amounts of biomass (Fig. 3).
Bicarbonate extractable P was a poor predictor of P uptake and PUEstand, whereas other soil P fractions (NaOH, 10 m HCl, total inorganic P, total organic P, total P) were relatively good predictors. Total inorganic P explained more of the P uptake, ANPP, and PUEstand variance than other P fractions, so we analysed plant response to this fraction in more detail. Phosphorus uptake increased asymptotically as total inorganic P increased on schist soil, but not on basalt soil (Fig. 4a). Phosphorus use efficiency (PUEstand) increased as total inorganic P decreased on schist soils and basalt soils, with a precipitous increase in PUEstand below c. 150 μg g−1 of inorganic P (Fig. 4b). This result suggests a marked reduction in PUEstand benefit as total inorganic P increases beyond 150 μg g−1. Whereas PUEstand increased as P uptake decreased, RUE decreased linearly with decreasing P in biomass, but only on schist soils (Fig. 4c).
Although RUE was similar between schist and basalt stands (Fig. 3), RUE was closely coupled with biomass P on schist soils, but not on basalt soils (Fig. 4c). This result suggests that stands on schist are able to leverage their limited P to enhance photosynthesis to a greater extent than stands on basalt soils, probably by allocating a greater percentage of total plant P to leaves. Additionally, stands on schist have longer Pres and higher CPUE (Fig. 4e,f). Thus, these stands not only achieve higher RUE per unit P uptake than stands on basalt, but they require less P and keep this P in photosynthetic tissues for longer time periods before losing it through leaf senescence. Contrary to our expectations, RUE and PUEstand were not inversely correlated (r2 = 0·01, P = 0·67, d.f. = 22), i.e. RUE was not traded for higher PUEstand. This result was evident irrespective of whether pre-cyclone or post-cyclone data sets were used.
Intrinsic and plastic components of PUE
Both intrinsic differences among species and plastic differences within species contributed significantly to ecosystem-level (across soils) PUE and ANPP. Although intrinsic differences contributed more to ecosystem-level PUEspecies, plasticity of wood P was also important (Table 3). Plasticity in wood P likely reflects luxury P uptake on basalt into wood storage cells, which can be significant in tree species (Rosecrance et al. 1998). Intrinsic and plastic components of PUEspecies responded similarly to changes in soil type with one notable exception. Phosphorus resorption from plant leaves prior to senescence was significantly higher for schist specialists than generalists, but P resorption actually increased on basalt soils within generalists. Although nutrient resorption is not a general mechanism of nutrient conservation in forests (Killingbeck 2004; Ares & Gleason 2007), we have no explanation for why P resorption increased on the P-poor soil among species, but not within species.
Although the range in PUE traits among species is quite wide, the plasticity within species is also relatively wide (Table 3). This shows that many species have similar ranges of PUEspecies and that there is much PUEspecies redundancy in these forests, i.e. many species can achieve the same PUEspecies either because they are intrinsically similar or through plastic response. Thus, the productivity of these forests, with regard to P, may be relatively resilient to species loss, assuming that PUE traits among species affect stand-level ANPP on P-poor soils. This is because as one species becomes extinct, other species with similar PUEspecies may serve as adequate replacements.
Regardless of the level of PUEspecies redundancy among species, plastic and intrinsic components of PUE resulted in stand-level PUE that was over 100% greater on schist soils than basalt soils. Even within a more plastic species (M. polymorpha), stands in Hawaii on P-limited soils could achieve only a 29% increase in PUE over stands that were not P-limited (Harrington et al. 2001). Metrosideros may reflect the limit of plasticity, whereas the Queensland species may demonstrate the ecosystem-level benefit of genetic divergence.
Scaling up from species to stand-level functioning
Lack of response to fertilization on the P-poor schist suggests that forest on schist is not P-limited. P-limitation should be viewed in the context of species sorting and composition across soil types. It is likely that the high density of P-efficient species (with low P requirements) on schist results in stands that are not P-limited. However, this implies that at some point during stand development, less efficient species were excluded on these sites, possibly because they were P-limited. Thus, secondary succession on schist is likely to be P-limited at some stage and among some species, even if P-limitation was not evident among schist specialists during the time this study was undertaken. We also note that P sensitivity (evident for some of our species) may have contributed to the lack of fertilization response, as RGR values were significantly reduced across all species in fertilized plots. Phosphorus sensitivity is not uncommon in Australian plants (Handreck 1997; Lambers et al. 2008) and is thought to be related to inadequate down-regulation of P uptake (Shane et al. 2004).
Stands growing on P-poor schist soils achieve similar rates of ANPP as stands growing on basalt soils. This is accomplished by three main mechanisms/conditions. (i) Lower concentrations of P in plant tissues, but mainly in wood, reduce the requirement for P uptake per unit biomass produced (PUEspecies). (ii) P-efficient trees (via plasticity or intrinsic difference) allocate more P to leaf biomass than to wood biomass than do P-inefficient trees, thus maintaining relatively high concentrations of P in photosynthetic tissues. (iii) P-efficient trees are able to assimilate C with this leaf P over longer periods of time than P-inefficient trees (i.e. they have longer Pres).
Trade-offs and edaphic specialization
Productivity relates not only to P use, but also to co-limiting resources and the trade-offs between them. Certainly, radiation limits production in most closed-canopy forest ecosystems (Pacala et al. 1996). Although within-plant P conservation is likely to increase fitness on P-limited soils (Aerts 1999), high plant-level RGR and stand-level ANPP must be accomplished by leveraging this P to assimilate C (i.e. plants must have high RUE and Pprod). It is likely that forest on P-poor schist accomplish this by partitioning a greater proportion of their total P to photosynthetic processes (rather than to storage), as well as using this P for longer periods of time prior to senescence (longer Pres). Supporting this conclusion, total P content in biomass was closely coupled to RUE (dry matter produced per unit radiation intercepted) on schist soils, but not on basalt soils (Fig. 4c). Additionally, LAI and Pprod were similar on both soil types (Fig. 4d), indicating that the productivity of leaves was similar between schist and basalt. This is a meaningful result if we consider that forest on schist has less than half the total P content in biomass than forest on basalt. It is not easy for stands to increase PUEstand without sacrificing RUE because PUEstand and RUE are likely to be inversely related (Herbert & Fownes 1999), although this trade-off was not evident in this study. Rather, forest on schist had similar RUE values, whereas their PUEstand was more than twice that of forest on basalt (RUE was not traded for PUE). Although it is likely that P partitioning (favouring canopy P), high Pprod and long Pres are traits that facilitate the avoidance of this trade-off, it is not known what opportunity costs might be associated with these traits (if not reduced RUE).
Not only are most P-efficient species more common on schist soils, but six of the most efficient species do not occur on basalt at all. Furthermore, this distribution pattern is not unique to this forest. Infertile soil specialists in an Indonesian lowland rainforest were also absent from high-P soils (Paoli et al. 2006), indicating that obligatory specialization on infertile soils may be a general phenomenon in tropical rainforests (Gary Paoli, personal communication). Could there be an opportunity cost to high PUEspecies and RUE that restricts the competitive ability and distribution of P-efficient species on fertile soils? We suggest that reduced survival (seeds, seedlings and mature trees) and reduced fecundity (seed mass, quantity and dispersal) offer possible opportunity costs to high nutrient use efficiency on infertile soils (McGraw & Chapin 1989; Stevens et al. 2004; Thomson & Leishman 2004).
Species that partition high percentages of plant P to leaves are more likely to lose vigour after canopy loss (e.g. following cyclone disturbances) than species that do not. Relatively high P partitioning to photosynthetic apparatus may also result in reduced P partitioning to other P sinks, such as reproductive structures, herbivore defence and storage. Seed size has been shown to be related to survival in shaded environments (Foster & Janson 1985), survival in nutrient-poor environments (Milberg & Lamont 1997) and survival of desiccation, litter burial and predation may also be linked to seed size (Metcalfe & Grubb 1997). Additionally, masting events in rainforests require considerable plant investment of P (c. 21% of annual leaf litter P) (Green & Newbery 2002). Although other trade-offs are likely to be important, reduced fecundity and increased mortality are likely opportunity costs of high PUE in tropical rainforests.
We thank Kumi Gleason and Laura Williams for their help in the field and Andrew Ford for his help with tree identification and rewarding discussions. We are grateful to Patrick Baker and Dennis O’Dowd for their critique of the overall study plan, to Aiden Sudbury and Doug Maguire for statistical advice, and to Dan Binkley, Paula Campanello, Christian Giardina, Darrell Herbert, David Newbery, Gary Paoli and an anonymous reviewer for providing comments on earlier versions of this manuscript. Two grants provided funding for this research: The Holsworth Wildlife Research Endowment, ANZ Charitable Trust, Australia, and the Monash Small Grant Scheme, Monash University, Australia. The Monash Silver Jubilee Scholarship and an Australian International Postgraduate Research Scholarship provided additional support for Sean Gleason. This study was completed under Queensland EPA permit no. WITK03219805.