Responses of leaf traits to climatic gradients : adaptive variation versus compositional shifts

Dynamic global vegetation models (DGVMs) typically rely on plant functional types (PFTs), which are assigned distinct environmental tolerances and replace one another progressively along environmental gradients. Fixed values of traits are assigned to each PFT; modelled trait variation along gradients is thus driven by PFT replacement. But empirical studies have revealed “universal” scaling relationships (quantitative trait variations with climate that are similar within and between species, PFTs and communities); and continuous, adaptive trait variation has been proposed to replace PFTs as the basis for next-generation DGVMs. Here we analyse quantitative leaf-trait variation on long temperature and moisture gradients in China with a view to understanding the relative importance of PFT replacement vs. continuous adaptive variation within PFTs. Leaf area (LA), specific leaf area (SLA), leaf dry matter content (LDMC) and nitrogen content of dry matter were measured on all species at 80 sites ranging from temperate to tropical climates and from dense forests to deserts. Chlorophyll fluorescence traits and carbon, phosphorus and potassium contents were measured at 47 sites. Generalized linear models were used to relate log-transformed trait values to growingseason temperature and moisture indices, with or without PFT identity as a predictor, and to test for differences in trait responses among PFTs. Continuous trait variation was found to be ubiquitous. Responses to moisture availability were generally similar within and between PFTs, but biophysical traits (LA, SLA and LDMC) of forbs and grasses responded differently from woody plants. SLA and LDMC responses to temperature were dominated by the prevalence of evergreen PFTs with thick, dense leaves at the warm end of the gradient. Nutrient (N, P and K) responses to climate gradients were generally similar within all PFTs. Area-based nutrients generally declined with moisture; Narea and Karea declined with temperature, but Parea increased with temperature. Although the adaptive nature of many of these trait-climate relationships is understood qualitatively, a key challenge for Published by Copernicus Publications on behalf of the European Geosciences Union. 5340 T.-T. Meng et al.: Responses of leaf traits to climatic gradients modelling is to predict them quantitatively. Models must take into account that community-level responses to climatic gradients can be influenced by shifts in PFT composition, such as the replacement of deciduous by evergreen trees, which may run either parallel or counter to trait variation within PFTs. The importance of PFT shifts varies among traits, being important for biophysical traits but less so for physiological and chemical traits. Finally, models should take account of the diversity of trait values that is found in all sites and PFTs, representing the “pool” of variation that is locally available for the natural adaptation of ecosystem function to environmental change.


Introduction
The plant functional type (PFT) concept has been important in the development of dynamic global vegetation models (DGVMs), which combine vegetation dynamics 10 (changes in vegetation composition, expressed as abundances of PFTs) at the grid cell scale with hydrological and biogeochemical processes driven by the physical environment and modulated by PFT characteristics (Prentice et al., 2007;Prentice and Cowling, 2013). PFT classifications vary among models but nearly all include distinctions of life form (at least, woody vs. herbaceous plants), leaf habit (evergreen or deciduous) 15 and leaf form (broad or needle-leaves). Some models also distinguish climatic tolerance classes, related primarily to different overwintering mechanisms for woody plants (Harrison et al., 2010), and most distinguish C 4 plants. Usually a fixed set of properties (parameter values) is assigned to each PFT. This expedient simplifies modelling, but it is a potential weakness because it disregards continuous adaptive variation within Introduction  al., 1998;Fonseca et al., 2000;Niinemets, 2001;Wright and Westoby, 2002;Wright et al., 2004Wright et al., , 2005aSwenson and Enquist, 2007;Reich et al., 2007;Cornwell and Ackerly, 2009;Meng et al., 2009;Ordoñez et al., 2009Ordoñez et al., , 2010Albert et al., 2010;Prentice et al., 2011). Analyses of trait-environment relationships have been motivated partly by the objective of improving the representation of plant structural and functional 5 diversity in DGVMs (Woodward and Cramer, 1996;Díaz and Cabido, 1997;Lavorel et al., 2007;Kattge et al., 2011). In a new strand of DGVM development, modelling quantitative trait values rather than PFT abundances is the central objective (Kleidon et al., 2009;van Bodegom et al., 2012van Bodegom et al., , 2014Scheiter et al., 2013;Fyllas et al., 2014). An advantage of trait-based modelling is that it can take better advantage of the wealth 10 of georeferenced data now available on plant functional traits (Kattge et al., 2011). On the other hand, some leaf traits can have different relationships to climate depending on the PFT (e.g. Barboni et al., 2004;He et al., 2006;Meng et al., 2009). Moreover there are systematic leaf-trait differences between PFTs and these account for a substantial fraction of the total climatically related variation in leaf traits (e.g. Reich et al.,15 2007; Ordoñez et al., 2009Ordoñez et al., , 2010He et al., 2010). Thus it is not entirely clear from observational studies to what extent trait-environment relationships are universal; or conversely, to what extent differences in either trait values or trait-environment responses among PFTs are necessary to include in models to describe the totality of vegetation responses to environmental gradients -and, by extension, to directional environmental 20 change.
We address this question here with an analysis of variations in leaf traits in plant communities sampled on long gradients of temperature and moisture availability in China (Fig. 1). The data set consists of > 11 000 quantitative leaf trait determinations on all of the species present at 80 sites, with a wide geographic spread. We consider Introduction and potassium (K) contents, expressed on both an area and a mass basis. Thus we consider 12 traits in all. Area-based nutrient contents provide no independent information, as they are simply derived from mass-based nutrient contents and SLA, but they provide an alternative perspective on the regulation of leaf nutrient contents. LA, SLA, LDMC and N were measured at all sites; the other traits were measured at the 47 sites 5 in eastern China, which cover most of the climatic range of the full data set except for the driest climates in the west. Adopting a conventional PFT classification, we analyse variations of each trait with bioclimatic temperature and moisture indices (Harrison et al., 2010) within and across PFTs.

Sampling sites
The sites (Table 1) represent variation along the major gradients in temperature and moisture and include the major vegetation types in China apart from those unique to high elevations. Thirty-three sites in Xinjiang Autonomous Region in western China sample the extreme dry end of the moisture gradient, with annual rainfall between 12 15 and 468 mm (160 mm on average). Thirty-three sites on the Northeast China Transect (NECT: Ni and Wang, 2004) lie on an aridity gradient from closed forests with annual rainfall > 700 mm in the east, through grasslands to desert with annual rainfall of < 150 mm in the west. Fourteen sites located in forest reserves on the North-South Transect of Eastern China (NSTEC: Gao et al., 2003) have greater annual rainfall and 20 sample a range from temperate climates in the north to warm-temperate/subtropical climates in the south. The NSTEC sites are also differentiated in terms of rainfall, the sites in the east at any given latitude being wetter than those in the west. Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | tion structure were surveyed at each site. A checklist of vascular species at each site was created and field measurements were made on all the species for which sufficient material could be sampled.

Chlorophyll fluorescence measurements
F v /F m and QY were measured using a FluorPen FP100 (Photon Systems Instruments, 5 Czech Republic). F v /F m measures the potential rate of photosynthetic electron transport while QY measures the actual rate. QY is correlated with photosynthetic rate, although it also includes the diversion of electrons to non-photosynthetic activities such as the elimination of reactive oxygen species (Cavender-Bares and Bazzaz, 2004). 10 At least 10 g of leaves were collected for each species, except for a few species with very small leaves at the driest sites. Sunlit leaves of tree species were obtained with long-handled twig shears. The samples were subdivided for the measurement of specific leaf area (SLA), leaf dry matter content (LDMC) and C, N, P and K contents. The measurements used are averages of three replicates. Leaves were scanned with 15 a laser scanner; leaf areas were measured using Photoshop on the scanned images. Leaf fresh weight was measured in the field. Dry weight was obtained after air-drying for several days and then oven-drying at 75 • C for 48 h. Leaf C was measured by the potassium dichromate volumetry method and leaf N by the microkjeldahl method. Leaf P was analyzed colorimetrically (Shimadzu UV-2550). Leaf K was measured by Flame 20 Atomic Emission Spectrophotometry (PE 5100 PC).

Climate data and analysis
Mean monthly values of temperature, precipitation and fractional sunshine hours were obtained from 1814 meteorological stations (China Meteorological Administration, unpublished) and interpolated to a 10 km grid using ANUSPLIN 4.36 (Hutchinson and , 2006) with the help of a digital elevation model (Farr et al., 2007). Mean annual temperature (MAT) and precipitation (MAP), mean winter (P DJF ) and summer (P JJA ) precipitation and of precipitation seasonality and timing (defined as in Prentice et al., 2011) were calculated for each site. Bioclimatic variables were derived as in Gallego-Sala et al. (2010): mean temperature of the coldest month (MTCO) and 5 warmest month (MTWA), growing degree days above 0 • C (GDD 0 ), photosynthetically active radiation during the growing season (PAR 0 ), annual equilibrium evapotranspiration (EET), Moisture Index (MI = MAP/EET), annual actual evapotranspiration (AET) and the Cramer-Prentice α index of plant-available soil moisture (α = AET/EET) (Cramer and Prentice, 1988). Available water holding capacity (AWHC) values for the 10 calculation of α were assigned following Prentice et al. (2011), using sand, silt and clay fractions digitized from Shi et al. (2004). Principal components analysis was performed on standardized climate variables in SPSS. We analysed climate gradients for China as a whole, based on data from 89 623 10 km grid cells, and separately using just the 80 grid cells that included the sampling 15 sites.

Plant functional types (PFTs)
Plant species were classified as follows: trees (single-stemmed, maximum height > 2 m, subdivided as evergreen broad-leaved, evergreen needle-leaved and deciduous broad-leaved), shrubs (multi-stemmed with maximum height between 50 cm and 20 2 m, subdivided as evergreen and deciduous), erect dwarf shrubs (multi-stemmed with maximum height < 50 cm), lianas (woody climbing plants with perennial above-ground biomass), climbers (non-woody climbing plants with annual above-ground biomass), forbs, grasses, geophytes and ferns. Climbers and ferns were not included in the statistical analyses, however, as there were too few species of each. The optimum and 25 tolerance of each PFT in terms of α and GDD 0 , recommended by Harrison et al. (2010) as useful and globally applicable indices of effective moisture availability and warmth for plants, were calculated non-parametrically as follows (Fig. 2 variable was divided into bins, and average abundance values were calculated for the sites within each bin. The optimum was then calculated as the mean of the climate variable in the bins where the PFT was present, weighted by its average abundance in the bins. The tolerance range was calculated similarly, as the SD of the climate variable weighted by average abundance.

Generalized linear models
Generalized linear models (GLMs: Nelder and Wedderburn, 1972;Nelder and Baker, 2006) were used to quantify the relationships of trait values to climate variables (α and GDD 0 ), to avoid spurious bivariate relationships that can arise when (as here) the predictor variables are not perfectly independent. All traits were transformed to natu-10 ral logarithms (ln) to reduce skewness and linearize their relationships to the climate variables. This transformation has the property that regression coefficients represent fractional changes, which can be compared among traits measured in different units. The coefficients are expressed per unit of α (in other words, the change in ln trait value across the global range of α from 0 to 1) and per 10 −4 GDD 0 (equivalent to the 15 change in ln trait value across the global range from 0 to around 10 −4 GDD 0 ), so that their values are broadly comparable in magnitude between climate variables as well as between traits. We carried out three GLM analyses for each trait: (1) with climate variables (α and GDD 0 ) only as predictors, equivalent to ordinary multiple regression, (2)  A significance criterion of P < 0.01 was adopted for all regression coefficients in all three analyses. This is stringent enough to minimize the chance of "false positives" in 5 analyses (2) and (3). Results are presented as partial residual plots, using the visreg package in R. Partial residual plots are the multiple-regression analogue of simple x-y plots in ordinary regression. In plots showing the relationship of each trait to α, the y axis values of the data points are adjusted so as to remove the fitted effect of GDD 0 . Similarly, in plots showing the relationship of each trait to GDD 0 , the y axis values of 10 the data points are adjusted so as to remove the fitted effect of α.

Climate gradients
More than 80 % of the geographic variation in the climate of China can be summarized by variation on two principal axes (Table 2). Each principal axis is defined as a linear 15 combination of variables, and each variable is assigned a "loading" which represents the contribution of that variable to the combination. The first principal axis explains about 60 % of total variation and is related to temperature. MTCO, MAT, MAP, GDD 0 , and P DJF have large positive loadings. The positive loading for MAP reflects the general tendency for absolute amounts of precipitation to increase with temperature. The sec-20 ond axis explains a further 22 % of total variation and is related to moisture vs. aridity. MI, α, P JJA have positive loadings while PAR 0 and MTWA have negative loadings. The similar behaviour of PAR 0 and MTWA reflects an increasing period without clouds, and thus also higher temperatures in summer, as moisture availability decreases. A third axis relating to the timing and seasonality of precipitation accounts for only 9 % of total BGD 12,2015 Responses of leaf traits to climatic gradients A closely similar pattern emerged from analysis of climate data for the sampling sites (Table 2). This similarity confirms that the pattern of variation in climate across the sites reflects the general pattern of climate gradients across China, and that these gradients can be summarized using two variables, representing growing-season temperature and moisture availability respectively. For all further analysis we used the variables GDD 0 5 and α. The pattern of variation of GDD 0 and α across China is shown, with the site locations, in Fig. 1. Figure 1 also shows the frequency of different GDD 0 -α combinations among grid cells, and the site positions in this climate space.

Distribution of PFTs in climate space
The PFTs in our data set show distinct patterns of distribution in climate space (Fig. 2), 10 falling broadly into four groups. (1) Evergreen trees, evergreen shrubs and lianas favour the warmest and wettest climates, corresponding to the warm-temperate broad-leaved evergreen forests of southeastern China, with evergreen needle-leaved trees extending into cooler climates in the north. (2) Deciduous trees and deciduous shrubs favour cooler and drier climates, corresponding to the deciduous forests of central eastern 15 China. (3) Dwarf shrubs, grasses, forbs and geophytes favour still cooler and drier climates, corresponding to the grasslands, steppes and desert steppes of northern and northwestern China. (4) Ferns and climbers are prominent only in cooler and wetter regions of climate space; they occur more widely but not in any abundance, and they were not sampled elsewhere.

Trait-climate relationships: moisture effects
Significant community-level responses to growing-season moisture availability (α) were found for most traits (Fig. 3, Table 3). Dry climates generally favour small, thick, dense leaves (low LA, low SLA, high LDMC). Dry climates are also associated with slightly, or sometimes greatly, reduced potential and actual quantum yield. The steepest overall BGD 12,2015 Responses of leaf traits to climatic gradients  Inclusion of PFTs as predictors (Fig. S1 in the Supplement) shows that there are some differences among PFTs in the typical trait values found at any given α. This is most obvious for biophysical traits -LA, SLA and LDMC -and area-based nutri-5 ents. Needle-leaved evergreen trees stand out, having small, thick leaves, and high area-based nutrient contents, relative to other PFTs. The magnitudes of the regression coefficients against α for the different traits in this analysis are similar to those in Fig. 3, but now P area (in common with the other area-based nutrients) shows a significant negative effect of α. This relationship within PFTs is obscured in Fig. 3 by the abundance of 10 needle-leaved evergreen trees, with their very low SLA and therefore high P area values, towards the wet end of the gradient.
Where significant trait-PFT interactions in the response to α are found (Fig. S2), the responses are qualitatively (and usually, quantitatively) similar from one PFT to another. Regression coefficients for LA vs. α range from 3.8 to 6.1, with deciduous 15 shrubs and forbs showing significantly steeper responses than the rest. Regression coefficients for SLA range from 1.3 to 2.5 with forbs showing the steepest increases. Regression coefficients for LDMC range from -0.35 to -1.5 with forbs showing the steepest decreases. Different PFTs have different responses of QY to moisture, with geophytes responding most and forbs least. Neither area-nor mass-based nutrients 20 show any significant differences among PFTs.

Trait-climate relationships: temperature effects
Significant overall responses to growing-season warmth (GDD 0 ) were also found for most traits (Fig. 4, Table 3). Warm climates favour thick and dense leaves (low SLA and high LDMC). Warmer climates also show somewhat reduced potential and actual 25 quantum yield. The steepest overall relationship of any trait to GDD 0 is for SLA (-1.5) ( Including PFTs as predictors shows some differences among PFTs at any GDD 0 value, similar to those shown for α (Fig. S3). But the effects on the regression coefficients for GDD 0 are more profound. Most importantly, the within-PFT responses of the three biophysical traits -LA, SLA and LDMC -to temperature are non-significant. Thus, the overall responses of SLA and LDMC to GDD 0 shown in Fig. 4 are brought 5 about by PFT replacement, including the dominance of broad-leaved evergreen trees with low SLA and high LDMC at the warm end of the gradient. Within PFTs, N area and K area both decline with temperature, while P area increases. The lack of a significant relationship at the community level between N area and K area and temperature is due to PFT replacement along the gradient -again, most obviously, the prevalence of 10 broad-leaved evergreen trees with high N area and K area at the warm end of the gradient. Similarly, the steep overall declines in N mass and K mass with GDD 0 are mainly due to PFT replacement.
Relationships to GDD 0 fitted separately within PFTs (Fig. S4) showed fewer significant slopes, and less consistency among PFTs, than the corresponding relationships 15 to α. Individually significant PFT responses of SLA to GDD 0 could be increasing or decreasing (-0.57 to +1.3). Slopes of LDMC are negative (-1.6 to -3.0), with forbs and grasses showing the steepest declines. Area-and mass-based nutrients show few significant differences among PFTs in their responses to either GDD 0 ; however forbs show an increase in N mass and more steeply increasing P mass with GDD 0 compared to 20 other PFTs, and evergreen needleleaf trees show a steeper increase in P area .

Adaptive significance of trait responses to moisture availability
The observed continuous biophysical trait variations with moisture availability are consistent with previous studies and, qualitatively, reasonably well understood. The de- where soil moisture supplies are inadequate for transpirational cooling to be effective. High photosynthetic capacity coupled with high CO 2 drawdown, resulting in a low ratio of internal to ambient CO 2 concentration (c i : c a ), is also adaptive in dry environments (Wright et al., 2003;Prentice et al., 2014a) because of the high transpirational cost of keeping stomata open under conditions of high atmospheric aridity (vapour pressure 5 deficit). Increased photosynthetic capacity requires an increase in N area and a reduction in SLA. Low SLA of plants in arid environments may also allow leaves to avoid transient overheating when wind speeds fall (Leigh et al., 2012). The increase in LDMC with aridity is a key adaptation that allows leaves to maintain hydration even at low water potentials that may arise under drought conditions (Bartlett et al., 2012). 10 The reduction in QY with aridity points to drought-induced photoinhibition at the arid end of the gradient. Dry climates are characterized by high N area , consistent with a high photosynthetic capacity (compensating for low c i : c a ) as mentioned above. High K area in dry climates is consistent with the role of K in maintaining leaf function under waterlimited conditions (Sardans and Peñuelas, 2015). The regulation of leaf P is less well 15 understood, but the trend towards higher P area in dry climates is consistent with a relatively conservative N : P ratio within PFTs. Reduced mass-based N and P in arid climates are consistent with the increased allocation of carbon to leaf structural components in leaves with low SLA. 20 The observed tendency towards lower community-level SLA with increasing temperature may be linked to the well-known relationship between SLA and leaf longevity (Wright et al., 2004;Poorter et al., 2009). However, temperature-related trends in SLA within PFTs are mostly non-significant. The overall trend to lower SLA with increasing temperature is mainly driven by the shift from deciduous to evergreen PFTs, which 25 is to be expected given the clear advantage for evergreens in a subtropical climate that favours year-round photosynthesis and growth. Leaves also become more dense (higher LDMC) towards the warm end of the gradient, but within PFTs, the only signifi-7106 Introduction

Tables Figures
Back Close

Full Screen / Esc
Printer-friendly Version

Interactive Discussion
Discussion Paper | Discussion Paper | Discussion Paper | Discussion Paper | cant responses are for leaves to become less dense with increasing temperature. The community-level response of LDMC is thus driven by PFT replacement, with evergreen broad leaves characterized by high LDMC. Both potential and actual rates of electron transport in woody plants are reduced at the warm end of the temperature gradient. The effect is seen in both deciduous 5 and evergreen woody plants and is likely caused by heat stress resulting in a reduced efficiency of Photosystem II. The decrease in the potential rate implies that electrons are being diverted to protective mechanisms. The decrease in F v /F m is steeper than the decrease in QY.
The decline of both N area and N mass with temperature (after PFT differences have 10 been considered) is consistent with the declining N requirement to achieve a given catalytic activity of photosynthetic proteins as temperature increases (Reich and Oleksyn, 2004). The reasons for declining K area and K mass with temperature are unclear; possibly low temperatures in winter, towards the cold end of the gradient, create a K requirement similar to that caused by drought. The observed increases in both P area and P mass with 15 temperature are opposite to the general tendency of leaf N to increase allometrically with leaf P (e.g. Reich et al., 2010). These trends might reflect an increase in nonphotosynthetic electron transport processes that require a large supply of inorganic phosphate.

PFTs
Kattge et al. (2011) also examined trait variability within and between PFTs, in an analysis based on the TRY global plant trait data base. They showed differences in the fraction of total trait variance that could be attributed to PFTs vs. continuous variation within PFTs, with some traits predicted well by PFT identity. But for several traits, in- 25 cluding N area and SLA, they found that the largest fraction of the variance (as much as 75 %) was found within, not between PFTs. Our analysis extends that of Kattge et al. (2011)  shown contrasts in the responses of different traits to climate, and also contrasts in their responses to different aspects of climate. In most cases, nutrient traits show similar responses to climate within PFTs to those shown at the community level; and no significant differences were found between the responses within different PFTs. This is in agreement with the finding of Zhang et al. (2012) that climate is a more important 5 predictor of leaf element concentrations (except for S and SiO 2 ) than species identity. Variations of biophysical traits with respect to moisture availability are also similar within PFTs and at the community level. However, these same traits show patterns of response to temperature that are dominated by differences among PFTs. The differential responses of leaf N and P contents to moisture availability and temperature require 10 further investigation.

Implications for modelling
It is reasonable to expect that the performance of vegetation models would be improved by representing the values of phenotypically or genotypically plastic traits as state variables, rather than parameters (Prentice et al., 2007). This "adaptive" ap- 15 proach has been adopted explicitly in some recently developed models, e.g. Schymanski et al. (2009) and Scheiter et al. (2013). In the LPJ family of models descended from Sitch et al. (2003), leaf-level photosynthetic capacity (V cmax ) is allowed to vary adaptively within PFTs, based on an optimality hypothesis that predicts realistic responses of N area to light, temperature and CO 2 (Dewar, 1996;Haxeltine and Prentice, 1996). 20 On the other hand, the LPJ-family models treat SLA as a PFT-specific parameter and thus do not allow for covariation of SLA with N area , as has been demonstrated to occur, here and in other contexts (e.g. Lloyd et al., 2010;Prentice et al., 2011). Our findings suggest that vegetation models should retain the PFT concept and a minimal set of PFTs, because the distinctions between woody and herbaceous, de-25 ciduous and evergreen, and angiosperm and gymnosperm plant types systematically influence the values of key biophysical traits in ways that would not be predictable from assumed universal relationships. Moreover certain observed overall responses of trait 7108 Introduction values to climate, including the decline in SLA and increase of LDMC with increasing temperature in our study, appear to be driven principally by PFT replacement rather than by adaptive variation within PFTs. Nonetheless, the prevalence of continuous, consistent trait variation within and between PFTs for many traits and trait-environment relationships supports the conclusion that models should avoid prescribing fixed, PFT-5 specific values for most quantitative traits (e.g. Wright et al., 2005). Fixed, PFT-specific values could be replaced by universal adaptive functions of environmental variables: thus reducing the multiplicity of uncertain parameters, while simultaneously increasing the realism of next-generation DGVMs (Prentice et al., 2014b). To do so, however, requires that these functions be well specified and robust. Although some progress has 10 been made in developing trait-based models based on statistical trait-environment relationships, process-based model development requires these responses to be quantitatively predictable, based on explicit hypotheses about the adaptive significance of traits.
The Supplement related to this article is available online at BGD 12,2015 Responses of leaf traits to climatic gradients