Coordinated variation in leaf and root traits across multiple spatial scales in Chinese semi-arid and arid ecosystems


  • Guofang Liu,

    1. State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, Chinese Academy of Sciences, Beijing 100093, China
    2. Graduate University of Chinese Academy of Sciences, Beijing 100049, China
    Search for more papers by this author
  • Grégoire T. Freschet,

    1. Systems Ecology, Department of Ecological Science, VU University Amsterdam, De Boelelaan 1085, NL–1081 HV, Amsterdam, the Netherlands
    Search for more papers by this author
  • Xu Pan,

    1. State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, Chinese Academy of Sciences, Beijing 100093, China
    2. Graduate University of Chinese Academy of Sciences, Beijing 100049, China
    Search for more papers by this author
  • Johannes H. C. Cornelissen,

    1. Systems Ecology, Department of Ecological Science, VU University Amsterdam, De Boelelaan 1085, NL–1081 HV, Amsterdam, the Netherlands
    Search for more papers by this author
  • Yan Li,

    1. Xinjiang Institute of Ecology and Geography, Chinese Academy of Sciences, Urumqi 830011, China
    Search for more papers by this author
  • Ming Dong

    1. State Key Laboratory of Vegetation and Environmental Change, Institute of Botany, Chinese Academy of Sciences, Beijing 100093, China
    Search for more papers by this author

Author for correspondence:
Ming Dong
Tel: +86 10 82594676


  • Variation in plant functional traits is the product of evolutionary and environmental drivers operating at different scales. Little is known about whether, or how, this variation is coordinated between aboveground and belowground organs across and within spatial scales.
  • We address these questions using a hierarchically designed dataset of pairwise leaf and root traits related to carbon and nutrient economy of 64 species belonging to 14 plant communities in northern Chinese semi-arid and arid regions.
  • While both root and leaf traits showed most of their variance among (individuals and) species within communities, leaf trait variance tended to be relatively higher at coarser spatial scales than root trait variance. While leaf nitrogen (N) per area to root N per length ratio increased and specific leaf area to specific root length and leaf [N] to root [N] ratios decreased from semi-arid to arid environments owing to climatic/edaphic shifts, the matching pairs showed a strong pattern of positive correlation that was upheld across spatial scales and geographic areas.
  • Thus, trade-offs in plant resource investment across organs within individual vascular plants are constrained within a rather narrow range of variation. A new challenge will be to test whether and how such trait coordination is also seen within and across other biomes of the world.


Variation in plant functional traits (Chapin et al., 1993; Lavorel et al., 1997; Grime, 2001) is the product of evolutionary and environmental drivers that operate at a variety of different scales (Woodward & Diament, 1991; Keddy, 1992; Reich et al., 2003a). Variation in the values of functional traits along environmental gradients reflects variation in the relative importance of different plant adaptive mechanisms and their interactions with climatic, edaphic or topographic drivers (Diaz et al., 1998; Wright et al., 2004; Cornwell & Ackerly, 2009). Quantifying the variation pattern of key traits related to plant carbon, nutrient and water economy at each of multiple spatial scales could provide an opportunity to evaluate the relative importance of these drivers for species assembly (Ackerly & Cornwell, 2007). This, in turn, will help in predicting the responses of local-scale vegetation structure and ecosystem function to local, regional and global environmental changes (McGill et al., 2006).

Predictable relationships between values for traits related to plant carbon and nutrient economy with temperature or precipitation gradients have been significant but rather modest, but show major scatter around fitted regression curves (Wright et al., 2004, 2005b; Chave et al., 2006; Ordoñez et al., 2009). By contrast, at the scale of local sites and plant communities, major variation in plant traits relating to carbon or nutrient economy can be seen (Grime, 2001; Ackerly & Cornwell, 2007). Indeed, in a global synthesis of leaf economics traits, 36% and 38% of interspecific variation in specific leaf area (SLA, leaf area per mass) and mass-based leaf nitrogen (N) concentration (Nmass), respectively, was found within sites, with even much higher percentages for local variation in leaf lifespan, dark respiration and photosynthesis rates (Wright et al., 2004). Thus, a large part of plant trait variance related to carbon and nutrient economy is not explained by broad-scale climatic, edaphic or topographic influences (see also Cornwell et al., 2008; Freschet et al., 2010).

Spatial relationships for carbon and nutrient economy traits have been studied mostly for aboveground organs, with particular emphasis on leaf traits (Reich et al., 1998b, 1999; Wright et al., 2001). This paper aims to shift this imbalance by giving equal weight to root and leaf economics traits in an analysis of trait variation across spatial scales in arid and semi-arid northern China. A key innovation of this study is that it focuses on the scale-dependence of relationships between aboveground and belowground pairs of traits critical to nutrient and carbon economy. Evidence is growing that consistent trait syndromes exist across plant organs (Chapin et al., 1993; Wright & Westoby, 1999; Reich et al., 2003b; Tjoelker et al., 2005; Withington et al., 2006; Freschet et al., 2010), which means there must be coordination between aboveground and belowground parts with respect to the acquisition and allocation of limiting resources and adaptation to climatic stressors. However, environmental factors can also impact differentially on different traits of the same species (Reich et al., 1999; Wright et al., 2001). Thus, scaling relationships between pairs of traits can for example be negative within habitat but positive across habitats, as discussed in Reich et al. (2003b) for SLA vs leaf size.

Here we investigate whether consistent scaling relationships can be seen across spatial scales in three logically related aboveground-belowground trait pairs, and how robust such relationships are to the different scales at which diverse environmental drivers may operate at variable strengths. Specifically, along a transect of increasing aridity across northern China between sites (within and between the semi-arid and arid regions), we expect a coarse-scale reduction both in water availability and, consequently, in the time windows and quantities of available nutrients, particularly of nitrogen. This should result in coarse-scale variation both in water and nutrient (and related carbon) economy traits. At the local scale, at a broadly similar precipitation regime, we expect edaphically driven heterogeneity in nutrient availability to mostly control variation in nutrient economy traits between plant communities. Later, we formulate predictions of how interactions of these drivers operating at different scales may influence the scaling of three matching pairs of aboveground vs belowground traits. We therefore match SLA with specific root length (SRL, root length : root mass) because both relate to resource (light vs water and nutrient) acquisition vs resource conservation strategies in that they describe the amount of ‘harvesting’ tissue per unit mass invested vs the relative protection of both organs to desiccation and herbivory (Mooney & Dunn, 1970; Reich et al., 1998a; Craine et al., 2001; Schädler et al., 2003; Poorter et al., 2006; Withington et al., 2006). Leaf Nmass (N per unit leaf mass) and root Nmass both reflect the ability to capture resources through enzymatic processes and reflect plant nutritional quality with respect to defence against herbivory (Craine & Lee, 2003; Agrawal & Fishbein, 2006; Kerkhoff et al., 2006). Leaf Narea (N per unit leaf area) and root Nlength (N per unit root length) should both relate to soil N availability and express how the amounts of N in tissues relate to the capacity of organs to exchange nutrients, water and/or CO2 with their external environment, dependent on their morphology (Gutschick, 1999; Anten & Hirose, 2003; Tjoelker et al., 2005). While coordinated variation has been found for these pairs of traits (Craine & Lee, 2003 for altitudinal gradient; Craine et al., 2005; Tjoelker et al., 2005; Kerkhoff et al., 2006; Withington et al., 2006; Freschet et al., 2010), several studies have questioned its generality (Reich et al., 2003a; Craine et al., 2005; Tjoelker et al., 2005). Suggested causes for noncoordination may include differences in leaf–root trait relationships between climatic areas (Craine et al., 2005; Wright et al., 2005b) or divergence between plant types (Reich et al., 2003a; Craine et al., 2005). Craine et al. (2005) revealed the potential of soil freezing and type of nutrient limitation to determine leaf–root trait relationships. We need to know whether and to what extent other climatic factors, but also resource availability, stress or disturbance potentially influence these relationships, which are critical for better understanding plant assembly rules (Keddy, 1992) and linkages between the aboveground and belowground subsystems (Wardle et al., 2004).

As leaf–root trait pair relationships might be disrupted by a differential influence of climatic factors on aboveground and belowground plant parts (Craine et al., 2005), we test the hypotheses that, along a strong aridity gradient: leaf and root traits critical to plant carbon and nutrient economy will differ in their interspecific variance partitioning across spatial scales; interspecific variation in pairs of analogue leaf and root traits (SLA–SRL; leaf Nmass–root Nmass, leaf Narea–root Nlength) should generally be coordinated, but somewhat be offset across geographic areas and spatial scales. Specifically, we expect SLA and SRL to show largely coordinated variation, with SLA reduction in arid sites (Cunningham et al., 1999; Wright et al., 2001; Ackerly, 2004) at given SRL causing some mismatch at the regional scale. Similarly, while leaf Nmass and root Nmass should be coordinated across scales (Kerkhoff et al., 2006), species from the arid region should display strategies favouring higher belowground (water and nutrient) than aboveground (light) resource acquisition compared with species from the semi-arid region. Leaf Narea has been shown to be relatively high in species from more arid areas because their leaves need to increase photosynthetic rates per unit area in order to maintain strong internal CO2 gradients to compensate for longer stomatal closure (Wright & Westoby, 2002). Therefore, we expect higher leaf Narea at given root Nlength to cause some mismatch in coordination with aridity at the regional scale.

We tested these hypotheses and specific predictions across seven localities representing the predominant vegetation belts of arid and semi-arid northern China by measuring species for pairwise leaf and root traits in a five-level hierarchical design downscaling from regions to localities, communities, species and individuals.

Materials and Methods

Study area, spatial design and sampling

This study was conducted along a transect of increasing rainfall ranging from northwest to northeast China. Mean annual precipitation (MAP) ranges from 43 mm to 372 mm, mean annual temperature (MAT) from 12.1 to −0.2°C and aridity index (AI) from 23.8 to 2.6 (Fig. 1). Across this transect, most precipitation falls during the growing season, from April to August.

Figure 1.

 The seven localities of study and their environmental features. Lon, longitude; Lat, latitude; E, elevation; MAT, mean annual temperature; TJan,mean temperature in January; TJul, mean temperature in July; MAP, mean annual precipitation; GSP, growing season precipitation (April–August); PET = Penman potential evaporation (based on Penman–Monteith equation, Allen et al., 1998); AI, aridity index.

Within this transect, two distinct climatic areas were identified, referred to here as arid (annual precipitation < 200 mm) and semi-arid regions (> 200mm). Three of the seven study localities were in the arid region (Cele, Fukang and Shapotou in Taklamakan, Gurbantunggut and Tengger deserts, respectively) and four in the semi-arid region (Ordos, Otindag, Naiman and Hulunbeir in Mu Us, Otindag, Horqin and Hulunbeir sandy grasslands, respectively), ranging from northwest to northeast (Fig. 1). Within these localities, plant communities representative of the local area were selected with respect to their land cover and contrasting features (two communities per locality on average) and named after the single-most dominant species. Dominant species, representing roughly 80–90% of total vascular plant biomass of the ecosystem (Cornelissen et al., 2003), were then sampled in each community. For each species, several individuals were sampled separately to account for intraspecific variation (random sampling within 0.25 ha area). In this way, we obtained a spatial hierarchical design spanning five levels of observations, from regions to individuals.

To avoid effects of seasonal variation, all living leaves were collected while fully mature and before the onset of senescence, in early August 2008. Root sampling was carried out from mid-August to early September 2008. In total, 115 groups of individuals (only one species per group) belonging to 64 species (owing to redundant species across communities) were sampled for leaves and roots, with 2–14 dominant species per community (see the Supporting Information Table S1 for the detailed species list and trait values). Species nomenclature follows Flora of China (ECCAS 1974–1999). For each species, a minimum of 10 different plant individuals were used for leaf sampling and three for root sampling (to minimize labour and disturbance) in order to ensure the representativeness of the pool collected. For leaves, part of the collection was placed in a paper bag and air-dried for chemical analyses while the other was immediately placed in a closed plastic bag to be analysed for SLA within 6 h (for more details see Cornelissen et al., 2003). For root sampling, plant individuals were excavated and brought to the laboratory with soil still attached. Soil and alien material were washed off the root system before living, undamaged roots were collected. To ensure a fair comparison of root types, in terms of structure and function, only the finest root branch orders of each species were considered (absorptive roots < 1 mm) and pioneer roots (i.e. large scouting roots) were avoided. As a result, 90% of the total length of collected roots (all species taken together) belonged to the 0–0.5 mm diameter class while 10% belonged to the 0.5–1 mm diameter class. Large mycorrhizal rhizomorphs were brushed off the roots. For both organs, materials presenting obvious symptoms of damage, infection or herbivore activity were avoided. Petioles and rachides were included as part of the leaf for all analyses. For the leafless plant Calligonum mongolicum the plant part functionally analogous to leaf was collected.

Plant trait measurements

For SLA and SRL measurements (Cornelissen et al., 2003), 10 leaves and 3–10 root pieces (10 in most cases) were scanned individually at 300 dpi resolution for each plant species (BenQ 5550 scanner, BenQ Corp., Taipei, Taiwan, China), then oven-dried (60°C, 48 h) and weighed separately. Scanned leaves and root pictures were analysed using image j software ( and winrhizo software (, respectively. Specific leaf area was then expressed as the ratio between leaf area (mm2) and leaf dry mass (mg). Twenty leaves per individual were used for microphyllous species to reduce measurement error. Specific root length was expressed as the ratio between root length (m) and root dry mass (g). Leaf and fine root N concentration were measured with Kjeldahl determination (BUCHI AutoKjeldahl Unit K-370, BUCHI Laboratory Equipment, Flawil, Switzerland). For that purpose, air-dried subsamples of each material were first ground and subsequently oven-dried for 24 h at 60°C. Leaf and root Nmass (mg g−1), leaf Narea (g m−2) and root Nlength (mg m−1) were derived from these analyses following Cornelissen et al. (2003) and Tjoelker et al. (2005).

Data analysis

Nested ANOVAs, based on the decomposition of type I sums of squares, were performed in order to partition variance components of each leaf and root trait across multiple spatial scales (Nested Procedure, SAS version 8.0; SAS Institute Inc., Cary, NC, USA). All traits were log-transformed before analysis. Standardized major axis (SMA) regressions (Falster et al., 2006) were used to quantify allometric relationships of pairwise SLA–SRL, leaf Nmass–root Nmass and leaf Narea–root Nlength across and within spatial scales, but also across main plant families (Chenopodiaceae, Poaceae, Fabaceae and Asteraceae) and plant types (shrub, C3 grass, C4 grass, legume and forb). The Salsola laricifolia community with two species only was excluded from analyses at the community scale. The DOS-based smatr package used for SMA regressions allows testing both for homogeneity among SMA slopes via a permutation test and for differences in SMA elevation (intercept) via the SMA analogue of standard ANCOVA (Wright et al., 2005a). Intercept homogeneity comparisons were performed only when slopes were homogeneous. Within-scale comparisons were made between both regions, between all localities and between all communities. For all three pairwise relationships, SMA slopes, intercepts and correlation coefficients of all species assemblages (at community, locality and regional scales) were plotted against species assemblage size. The 95% confidence interval was produced using random species assemblage resampling. For each sample size (from 3 to 70), 1000 random species assemblages were drawn from the total species pool (transect). The slopes, intercepts and correlation coefficients of each assemblage were then calculated using the ‘smatr’ package for r (R Development Core Team, 2009) and the 95% confidence intervals were produced based on their (close to normal) distribution.


Variation of leaf and root traits across spatial scales

The contributions of the different scales to trait variance varied depending on the trait considered (Fig. 2). However, in our nested design, among-individual and among-species differences together consistently explained over 55% of the total trait variance, irrespective of the trait consid-ered. Contributions of among-community within-locality, among-locality within-region or among-region differences were smaller and much more variable across traits. The variances explained by regional and local differences were much higher for leaves than for roots.

Figure 2.

 Variance components of leaf traits (left half) and root traits (right half) based on nested ANOVA across spatial scales including region, locality, community, species and individual.

Relationships between pairs of leaf and root traits across spatial scales

When comparing pairwise leaf and root functional traits across multiple species from arid and semi-arid regions representing a broad spectrum of vascular plant taxa and growth forms, significant positive correlations were found in all cases (Table S2; see also Fig. 3(a,b,c), upper left panels). Thus, SLA and SRL (r = 0.57), leaf Nmass and root Nmass (r = 0.49) and leaf Narea and root Nlength (r = 0.65) were significantly correlated (P < 0.001 in all cases). When separating this dataset into arid and semi-arid regions, significant positive correlations were still found for all three leaf-root pairwise trait relationships, irrespective of the region considered (Table S2). Downscaling further into locality and community datasets, several correlations were lost, while some others remained significantly positively correlated and positive trends dominated the pattern.

Figure 3.

 Standardized major axis regressions (SMA) of pairwise trait combinations: (a) specific root length (SRL)–specific leaf area (SLA), (b) root nitrogen (N) per unit mass (Nmass)–leaf Nmass, (c) root N per unit length (Nlength)–leaf N per unit area (Narea), at different spatial scales. Data points represent species trait values averaged at the community level. Significant regressions are indicated by continuous lines, marginally significant regressions by dashed lines and nonsignificant regressions by dotted lines. Localities a (Cele), b (Fukang) and c (Shapotou) belong to the arid region, while d (Ordos), e (Otindag), f (Naiman) and g (Hulunbeir) belong to the semi-arid region. Communities (with main dominant species in parentheses) 1 (Calligonum mongolicum) and 2 (Caratoides latens) belong to locality a; 3 (Nitraria sibirica) and 4 (Seriphidium terrae-albae) belong to b; 5 (Artemisia ordosica) and 6 (N. sibirica) belong to c; 8 (Artemisia sphaerocephala) and 9 (A. ordosica) belong to d; 10 (Artemisia intramongolica) and 11 (Artemisia frigida) belong to e; 12 (Caragana microphylia) and 13 (Artemisia halodendron) belong to f; 14 (A. halodendron) belongs to g.

While the nonindependence of trait values did not allow testing for changes in SMA slopes or intercepts of paired leaf–root trait relationships across spatial scales, funnel plots permitted visual comparisons (Fig. 4). For all three pairs of traits, the scatters around the SMA slopes, intercepts and correlation r of transect relationships were consistent with decreasing variance with increasing sample size, yielding a series of ‘funnel’ plots. Irrespective of the spatial scale, most sites fell within the boundaries of the 95% confidence interval, suggesting that most potential differences in slopes, intercepts and correlation coefficients across scales could be caused by sampling variance only.

Figure 4.

 Funnel plots of slope, intercept and strength of leaf-root trait pair relationships across spatial scales in relation to sample size (n). Across scales, sites are displayed with different symbols: circles, community; triangles, locality; +, region. Solid horizontal lines indicate the standard major axis (SMA) slope, intercept or correlation r-values seen for the whole-dataset relationships (transect). Dotted lines delimit the 95% confidence intervals of SMA slopes, intercepts and correlation coefficients as obtained from a procedure of random site resampling (1000 iterations for each sample size) from the total transect dataset.

Relationships between pairs of leaf and root traits within spatial scales

Looking for consistency in leaf–root trait relationships within each scale we found distinct patterns for the different pairs of traits. For SLA–SRL relationships, within each spatial scale, all regressions shared the same SMA slope (P=0.491, 0.968 and 0.976 for among-region, among-locality and among-community comparisons, respectively), but were shifted in elevation (P < 0.001 in all cases; Table 1). Thus, for a given SRL, species from the arid region had lower SLA than species in the semi-arid region. This trend was consistent with the significant effect of MAP and MAT on SMA regression elevation, which decreased with decreasing rainfall and increasing temperature (Tables S4 and S5). Penman potential evaporation (PET) did not affect the relationship. Conversely, within each spatial scale, some regressions between leaf Nmass and root Nmass differed significantly or marginally in their SMA slope. Thus arid and semi-arid regions showed a marginally significant difference in SMA slope (P = 0.057); all localities but one showed a significantly different slope from at least one other locality (P = 0.019; see details of multiple comparisons in Table S3) and some communities displayed marginally significant differences with each other (P = 0.067). A significant positive effect of PET on leaf Nmass–root Nmass slope was detected (Table S5), consistent with the steeper slope found in arid than in semi-arid area. Conversely, MAP and MAT did not influence leaf Nmass–root Nmass slope. Similarly to SLA–SRL relationships, within each scale all SMA regressions between leaf Narea and root Nlength shared the same slope (P of 0.228, 0.235 and 0.471 for among-region, among-locality and among-community comparisons, respectively), but shifted in elevation (P < 0.05 in all cases, Table 1). For a given root Nlength, species from the arid region had therefore higher leaf Narea than species from the semi-arid region. While PET and MAT did not affect the leaf Narea–root Nlength relationship, MAP had a significant negative effect on its elevation (Table S5).

Table 1.   Standardized major axis regression (SMA) analyses of pairwise relationships within each spatial scale
Leaf-root trait pairsSpatial scaleSlope homogeneity (P)Shift in elevation (P)
  1. SRL, specific root length; SLA, specific leaf area; Nmass, nitrogen (N) per unit mass; Narea, N per unit area; Nlength, N per unit length.

SLA–SRLRegion0.491< 0.001
Locality0.968< 0.001
Community0.976< 0.001
Leaf Nmass–root NmassRegion0.0570.006
Leaf Narea–root NlengthRegion0.2280.007


Leaf and root traits differ in their interspecific variance partitioning across spatial scales

As hypothesized, we found diverging patterns of leaf and root trait variance partitioning across spatial scales. Indeed, while species nested within communities accounted for most of the variance for all traits, a relatively large part of leaf trait variance was explained by either among-region or among-locality differences, as opposed to the predominance of among-community differences for their root counterparts. This spatial mismatch in variation in leaf and root traits would imply probable shifts in the pairwise leaf–root trait relationships across scales, as discussed later. The heterogeneous partitioning of variance across both leaf and root traits is consistent with Wright et al. (2004), who found within-community variance ranging from 20% for leaf phosphorus (P) content to 67% for leaf photosynthetic capacity in a global dataset. However, our results show that root traits are also constrained by environmental filters acting at different intensities at distinct spatial scales. The large variance explained by within-community differences demonstrates that community scale is of particular relevance when studying mechanisms of species assembly.

Convergence of leaf and root traits

Relationships between belowground and aboveground traits have, to date, been poorly understood (Wardle et al., 2004). Several studies have suggested cross-species correspondence between leaf and root traits related to plant carbon and nutrient economy. Thus, among other traits, significant correlations were found for Nmass (Tjoelker et al., 2005; Kerkhoff et al., 2006), Pmass (Kerkhoff et al., 2006), mass-based respiration (Tjoelker et al., 2005), SLA–SRL (Reich et al., 1998a; Wright & Westoby, 1999; Withington et al., 2006; but see Tjoelker et al., 2005) and lignin and dry matter content (Freschet et al., 2010). Our study extends the pattern of coordinated variation of both SLA–SRL and leaf Nmass–root Nmass to arid and semi-arid areas and extends this syndrome to a complementary pair of traits: leaf Narea–root Nlength.

Across all scales studied and for all three pairs of traits, most of the variance in slopes, intercepts and strengths of the relationships could be explained by sample size alone, which suggests that the trade-off in plant resource investment across plant organs within individual vascular plants is constrained within a rather narrow range of variation, both in terms of nutritional and structural investments (Freschet et al., 2010). In other words, with respect to carbon and nutrient economy, plant strategies seem to be coordinated across belowground and aboveground organs (Grime, 2001). However, although a large part of across-scale variance might be attributed to sample size, significant differences in slope or intercept occur within spatial scales, as seen from the several sites lying outside the confidence interval.

Leaf–root trait relationships differ across spatially separate areas at similar scales

As predicted, the geographic area and associated environmental factors had an influence on pairwise trait relationships, whether in slope or in elevation. As previously shown in studies on the relations between economics traits of leaves, such as N and P content, SLA, respiration rate or photosynthetic capacity (Reich et al., 1999; Wright et al., 2001), trait scaling relationships can be influenced by geographic areas. Which factors can explain the similar shifts in aboveground–belowground trait relationships we found across geographical areas? First, while none of the paired relationships showed heterogeneous slopes between families (Chenopodiaceae, Poaceae, Fabaceae and Asteraceae) or plant types (shrub, C3 grass, C4 grass, legume and forb), shifts in elevation were found across both categories, supporting the hypothesis (Reich et al., 2003a) that the type of plant might be a factor influencing such aboveground–belowground trait relationships (Table S6). However, the rather even distribution of plant families and plant types across all communities, localities and regions (see Table S1) limits the potential influence of such factors on trait relationships within each of these scales. Second, in arid and semi-arid ecosystems,because of the low or pulse-wise nutrient and water availability, and the discontinuous vegetation cover, interspecific competition is rather limited (Grime, 1977). Trait variation across geographic areas of northern China seems therefore more related to abiotic factors such as climate, soil texture or topography.

Indeed, when downscaling from global to local scale, several abiotic and biotic factors create a selective pressure on the occurrence of particular trait values of species assemblages (Keddy, 1992; Weiher et al., 1998). These selective pressures should theoretically have different impacts on distinct traits and can therefore affect aboveground and belowground plant traits differently (Craine et al., 2005). Thus, the slope of leaf Nmass–root Nmass relationships was significantly (among-locality) or marginally (among-region, among-community) influenced by spatially separated sampling areas and contrasting evaporation conditions, while the slopes of SLA–SRL and leaf Narea–root Nlength relationships were highly consistent between geographic areas. Differences in elevation were nevertheless found for the two last relationships, partly owing to climatic factors such as rainfall and temperature.

The minimal difference in SRL between regions suggests that climatic and edaphic factors do not constrain plant water and nutrient uptake strategy to a large extent at this scale. Most of the variation occurs indeed at the intraspecific and interspecific levels within communities. The large range of SRL values (from 3 to 53 m g−1) found across the total dataset indicates that contrasting strategies generally occur within the same community and therefore climatic regime. In contrast, a much larger part of the variation in SLA is found between regions, which presumably reflects a strong influence of climate, soil texture or topography on the selection of SLA values of the regional species assemblages. The consistently lower SLA to SRL ratio found for the arid region compared with the semi-arid region can therefore be partly explained by a shift in SLA between arid and semi-arid environments compared with a rather stable trend in SRL, as predicted. Structural reinforcements reducing SLA in arid environments provide leaves with increased protection against desiccation (Mooney & Dunn, 1970; Ackerly et al., 2002), herbivory (Grubb, 1992; Schädler et al., 2003; Agrawal & Fishbein, 2006) and other physical damage (Wright & Cannon, 2001). Indeed, plants from nutrient-poor environments (Aerts, 1995) and water limited environments (Wright & Westoby, 2002) are usually more conservative and invest in longer-lived structures. By contrast, the similar pattern of SRL across arid and semi-arid environments might reflect the widely varying strategies used by plants of (semi-)arid environments to face water shortage, ranging from fast regrowth during water pulses (promoting high SRL; Eissenstat, 1991; Hodge, 2004) to investment toward long-lived roots (promoting low SRL; Gill & Jackson, 2000). This multidirectional constraint on root water acquisition strategy could therefore potentially lead to similar distributions of SRL across arid and semi-arid environments.

The slope heterogeneity of SMA regressions between leaf Nmass and root Nmass of arid and semi-arid regions contradicts our hypothesis that the leaf Nmass–root Nmass ratio should be lower at arid sites. While species with the lowest investments of N in their tissues invest less Nmass in their leaves compared with roots in arid than in semi-arid environments, species with the highest N content behave similarly in arid and semi-arid regions. Only conservative species of arid environments therefore seem to favour nutrient and water uptake efficiency by fine roots over light capture. While both positive and negative relationships have been proposed, it is now recognized that no general relationship exists between Nmass and aridity (Killingbeck & Whitford, 1996; Wright et al., 2001), which is confirmed by our findings. The slope heterogeneity of SMA regressions between leaf Nmass and root Nmass seems therefore more related to shifts in root Nmass. This trend may be explained by different strategies used by the most conservative and acquisitive species found in both arid and semi-arid regions. Plants favouring fast regrowth strategies are indeed generally found to be more acquisitive than species with year-round drought resistance strategies. While conservative species may grow more slowly and steadily, relying largely on water and nutrients accumulated during wet periods, species favouring drought avoidance use water and nutrient to grow during wet periods only. Therefore, only nutrient conservative species may have an interest in investing more N per unit mass in roots than in leaves.

The consistently higher leaf Narea in arid than in semi-arid environments at a given root Nlength supports the hypothesis that increased leaf Narea is beneficial for species in arid environments in order to maintain strong internal CO2 gradients to compensate for long duration of stomatal closure (Wright & Westoby, 2002). Another explanation for this trend could, nevertheless, be found in the lower root Nlength of species in arid than in semi-arid areas at a given leaf Narea leading to lower fine root respiration rates and therefore greater root longevities (Eissenstat et al., 2000). However, given the higher among-region variance in leaf Narea compared with root Nlength, the former seems the more dominant factor.

Our results are consistent with Wright et al. (2005b) who showed that climatic factors can affect both the slope and intercept of leaf trait relationships. Within the worldwide leaf economics spectrum (Wright et al., 2004) shifts in trade-off between covarying traits, as influenced by climatic or others factors, may therefore explain part of the scatter around the regression line for that spectrum. Similarly, the noncoordinated part of the leaf–root trait relationships in our study may indicate scale-related shifts in the trade-off between traits of aboveground and belowground organs (sensuTilman, 1982).


To summarize, our analyses of leaf and root trait variance across and within spatial scales in the semi-arid and arid regions of northern China have revealed consistent patterns. First, while both root and leaf traits showed most of their variance among (individuals and) species within communities, leaf trait variance tended to be relatively higher at coarser spatial scales than root trait variance. It will be interesting to test whether these differences can also be extended to other root and leaf traits, such as those with a known strong link to macroclimate. For example, individual leaf size, which is known to be important with respect to climatic factors (Körner et al., 1989; Cornelissen, 1999; McDonald et al., 2003; Meier & Leuschner, 2008), showed much (32.8%) of its total variance at the regional scale in this northern Chinese flora (data not shown, available from the authors on request). Second, matching pairs of leaf and root traits related to carbon and nutrient economy showed a strong pattern of positive correlation upheld at different spatial scales. While the slopes and intercepts of these relationships varied importantly between geographic areas (localities, regions) and between scales relative to climatic and edaphic influences, these differences did not conceal the main finding that the flora of northern China features strong coordination of root and leaf traits. This aboveground–belowground coordination is consistent with strong whole-plant adaptation for either a more conservative or a more acquisitive resource economic strategy. The new challenge will be to test whether and how such coordination is also seen in other biomes of the world.


We thank Fanjiang Zeng, Xueyong Zhao, Xiaoping Xin, Jianjiang Qiao, Qiang Zhang, Yu Chu, Xuehua Ye, Yingxin Huang, Yuan Sui, Qingguo Cui, Shuqin Gao, Juntao Zhu and Bin Jiang for assistance during field sampling. Thanks to the Chinese State Meteorological Administration for providing over 30 yr of nationwide meteorological data. This research received support from the Chinese Academy of Science Ecology Stations of Cele, Fukang, Shapotou, Ordos, Naiman, Inner Mongolia and Hulunbeir. The research was funded by CAS (KZCX2-YW-431-04), NSFC (30521005), State Key Laboratory of Vegetation and Environmental Change (Institute of Botany, Chinese Academy of Science) (VEWALNE – VEgetation-WAter Long-term Networked Experiment in Chinese Steppe Zone) and CAS-grant (Visiting Professorship for Senior International Scientists) to JHCC. We are particularly grateful to David Ackerly and three anonymous reviewers for their highly relevant comments on the manuscript.