Environmental signals from leaves – a physiognomic analysis of European vegetation

Authors

  • Christopher Traiser,

    Corresponding author
    1. Institut für Geowissenschaften, Eberhard Karls Universität Tübingen, Sigwartstrasse 10, D-72076 Tübingen, Germany;
      Author for correspondence: Christopher Traiser Tel: +49 7071 29 73560 Fax: +49 7071 29 5217 Email: christopher.traiser@uni-tuebingen.de
    Search for more papers by this author
  • Stefan Klotz,

    1. Institut für Geowissenschaften, Eberhard Karls Universität Tübingen, Sigwartstrasse 10, D-72076 Tübingen, Germany;
    2. Laboratoire PaléoEnvironnements et PaléobioSphère (UMR 5125), Université Cl.Bernard – Lyon 1, 27–43 boulevard du 11 Novembre, F 69622 Villeurbanne Cedex, France
    Search for more papers by this author
  • Dieter Uhl,

    1. Institut für Geowissenschaften, Eberhard Karls Universität Tübingen, Sigwartstrasse 10, D-72076 Tübingen, Germany;
    Search for more papers by this author
  • Volker Mosbrugger

    1. Institut für Geowissenschaften, Eberhard Karls Universität Tübingen, Sigwartstrasse 10, D-72076 Tübingen, Germany;
    Search for more papers by this author

Author for correspondence: Christopher Traiser Tel: +49 7071 29 73560 Fax: +49 7071 29 5217 Email: christopher.traiser@uni-tuebingen.de

Summary

  • • Leaf physiognomic traits vary predictably along climatic and environmental gradients. The relationships between leaf physiognomy and climate have been investigated on different continents, but so far an investigation based on European vegetation has been missing.
  • • A grid data set (0.5 ¥ 0.5 latitude/longitude) has been compiled in order to determine spatial patterns of leaf physiognomy across Europe. Based on distribution maps of native European hardwoods, synthetic chorologic flora lists were compiled for all grid cells. Every synthetic chorologic flora was characterised by 25 leaf physiognomic traits and correlated with 16 climatic parameters.
  • • Clear spatial patterns of leaf physiognomy have been observed, which are statistically significant related to certain, temperature-related climate parameters. Transfer functions for several climatic parameters have been established, based on the observed relationships.
  • • The study provides evidence that synthetically generated floras represent a powerful tool for analysing spatial patterns of leaf physiognomy and their relationships to climate. The transfer functions from the European data set indicate slightly different relationships of leaf physiognomy and environment compared with results obtained from other continents.

Introduction

A central problem in plant ecological research is to identify environmental factors that may be affecting different aspects of plant physiology and morphology, from an individual cell up to complex ecosystems, and to understand in which ways all these parameters may interact during the evolution of ecosystems. Among other factors, physiognomic characteristics of the vegetation are considered to be determined by environmental factors such as climatic and edaphic conditions, and thus to be strongly adapted to their corresponding habitat. For example, vegetation types from regions of similar environmental conditions exhibit distinct physiognomic similarities, even if the individual taxa which comprise these vegetation types have completely different phylogenetic histories. In contrast, there may be obvious differences in physiognomic patterns among vegetation types from different environments. Many environmental studies have analysed physiognomic characters of plants, for example growth and life form (e.g. Raunkiaer, 1907), tree architecture (e.g. Halléet al., 1978) and leaf physiognomy (e.g. Wolfe, 1993), or have combined different physiognomic characters to describe or classify vegetation (e.g. Webb, 1959; Vareschi, 1980).

Leaf physiognomy in particular can be regarded as an excellent tool for ecological studies because leaves, as the primary photosynthetic organs of a plant, have to be optimally adapted to environmental conditions, and thus react most sensitively to the environment. Besides the general botanic interest in leaf physiognomic studies resulting in a characterisation of vegetation types, leaf physiognomy also represents an excellent basis for palaeo-environmental and -climatic studies (e.g. Wolfe, 1979, 1993; Wilf, 1997; Jacobs, 1999, 2002). Although several attempts have been made in recent years to recognize the causal relationships between climate and leaf physiognomic parameters (Roth et al., 1995; Baker-Brosh & Peet, 1997; Sisóet al., 2001), they are not yet fully understood (e.g. Boyd, 1994; Jordan, 1997; Christophel & Gordon, 2004).

Obviously there is a great potential for further (quantitative) leaf physiognomic studies, which may not only be of importance for palaeo-climatic purposes, but also for our general understanding of ecological and evolutionary processes, from the specimen scale up to the landscape or even the continental scale. Such studies may not only reveal new information about the relationships between environmental parameters and selected leaf physiognomic traits of individual taxa, but also causal information about the functional advantages and disadvantages of individual physiognomic traits. Furthermore, quantitative leaf physiognomic approaches may represent a hitherto unexplored, objective methodology for the classification of different vegetation types. Here, we present an initial study on the overall leaf physiognomic composition of the European lowland vegetation (i.e. woody angiosperms) in relation to selected climatic parameters.

Known facts

The fact that external forces may influence certain leaf physiognomic traits has been known at least since the end of the 19th century (e.g. Stahl, 1880, 1883; Schimper, 1898). However, Bailey and Sinnott (1915, 1916) were the first try to quantify the relationship between a certain leaf physiognomic trait (i.e. leaf margin type) and such an external forcing factor (i.e. mean annual temperature (MAT)). These authors demonstrated that the type of leaf margin is strongly correlated with the MAT, with entire margined taxa predominating in the tropics, and taxa with nonentire (toothed) margined leaves predominating in cold-temperate regions, a fact that has been corroborated by many additional studies since then (e.g. Wolfe, 1979; Wilf, 1997; Greenwood et al., 2004). Besides these works on leaf margin types, various other studies have dealt with the analysis of a wide array of leaf physiognomic characters in relation to environmental parameters (for an overview, see Table 1).

So far, several statistical methods have been used to investigate the relationships between leaf physiognomy and environmental parameters. Many authors used simple and/or multiple linear regression methods to consider the relationship of an individual environmental parameter in relation to one or a few leaf physiognomic characters. An overview of such ‘straightforward’ approaches, recently introduced for palaeo-climatological purposes, has been given by Kowalski (2002) (see also Tables 6 and 8 later). Furthermore, different authors used multivariate ordination methods to analyse the relationships between several leaf characters and climatic parameters simultaneously (e.g. Wolfe, 1993, 1995; Kovach & Spicer, 1996; Wolfe & Spicer, 1999).

Table 6.  SLR-transfer functions of mean annual temperature (MAT). Functions 1–6 are based on the leaf character ‘margin entire’ as published by different authors; function 7 is based on the leaf character ‘base acute’; (b0) intercept; (b1) slope; (x) leaf physiognomic character; (r2) coefficient of determination; (se) standard error of estimate; (seR) ratio of standard error of estimate
 SLR transfer functionsApplication to the European data set
yestb0b1x (%)n dataregionr2seP-valueauthorr2seseR (%)
1MAT 2.2400.286(ma_enti)   9N, M, S America0.942.0< 0.0005Wilf (1997)0.602.014.5
2MAT 1.1410.306(ma_enti)  34East Asia0.980.8< 0.001Wing and Greenwood (1993)0.601.812.8
3MAT 0.5120.314(ma_enti)1835Europe0.601.7< 0.0001this study0.601.712.4
4MAT−0.0590.316(ma_enti)  14S-America0.891.6Gregory-Wodzicki (2000)0.601.813.0
5MAT−0.2660.291(ma_enti) 106N, M America, East Asia0.763.4< 0.0005Wilf (1997)0.602.215.8
6MAT−2.1200.270(ma_enti)  74Australia0.63< 0.0001Greenwood et al. (2004)0.604.129.6
7MAT−2.6890.421(ba_acut)1835Europe0.761.4< 0.0001this study0.761.4 9.7
Table 8.  MLR-transfer functions of mean annual temperature (MAT). (b0) intercept; (b1, b2, … ) slopes; (x) leaf physiognomic character; (*) raw data were transformed as described by the authors; (r2) coefficient of determination; (se) standard error of estimate; (seR) ratio of standard error of estimate; (cf. Table 7)
 MLR transfer functionsApplication to the European data set
yestb0b1x1 (%)b2x2 (%)b3x3 (%)b4x4 (%)b5x5 (%)b6x6 (%)n datar2seauthorr2seseR (%)
1MAT  2.53617.372(ma_enti) 2.896(ap_emar) –8.592(lw_1)  740.862.0Wing & Greenwood (1993)*0.676.546.5
2MAT  1.76816.656(ma_enti) 5.137(ba_acut) –5.594(lw_1) 4.879(ap_emar) –9.200(ls_lep2) 1060.921.5Gregory (1996)*0.672.316.4
3MAT−11.26223.258(ma_enti) 7.022(ba_acut)16.099(lw_1)10.282(le_lobe)12.210(ls_lep2)11.484(lw_2) 1060.892.3Gregory & MacIntosh (1996)*0.187.352.3
4MAT  9.865 0.207(ma_enti)0.058(ba_obtu) –0.202(lw_1) 1440.902.8Wiemann et al. (1998)0.721.611.4
5MAT  2.603 0.142(ma_enti) 0.210(ba_acut) –0.249(sh_obov) 0.140(lw_3)18350.890.9this study0.890.9 6.6

The analysis of relationships between certain leaf physiognomic traits and environmental parameters within an individual taxon is usually based on directly sampled material. This material has to be sampled from the field either on purpose for an individual study (e.g. Fonseca et al., 2000; Uhl & Walther, 2003), or it can be obtained from already existing herbarium collections (e.g. Uhl & Mosbrugger, 1999) or from greenhouse experiments under controlled environmental conditions (e.g. Boelhoff et al., 2001).

For analysing such relationships for entire floras, data sets are required which also contain information on the spatial distribution of leaf physiognomic compositions of these floras. For this, taxon lists have to be established and scored with regard to their leaf physiognomic characters. Different sampling strategies have been applied for this purpose so far, often involving direct sampling of leaves in the field, or using published floral lists or distribution maps. Leaf character scoring (e.g. relative abundance of entire margin in the sampled flora) is based on the use of the sampled material, floristic manuals or herbarium collections (Dolph & Dilcher, 1979; Wolfe, 1993; Wilf, 1997; Jacobs, 1999). All of these sampling concepts have their own advantages and disadvantages, depending on the availability, scale and structure of data sources in the regions investigated.

Direct sampling of leaves in the field offers the opportunity to reliably document the leaf–climate relationship at the local (micro) scale of the sample sites. Related studies have mainly focused on the Americas and on East Asia (e.g. Wolfe, 1993; Gregory-Wodzicki, 2000). Although various authors have applied this sampling strategy, this time-consuming procedure makes it almost impossible to obtain data sets with a high spatial resolution, covering large areas. Furthermore, in order to analyse the relationship between leaf characters and environmental parameters appropriately, the required environmental data must be derived near to the leaf sampling localities. In many regions worldwide, however, this cannot be realized.

Another strategy is to use published floral lists or distribution maps, respectively. In this context, several studies have focussed on sites mainly in America and Africa (e.g. Dolph & Dilcher, 1979, 1980; Wilf, 1997; Jacobs, 1999), but also in Australia (Greenwood et al., 2004). Major advantages of this indirect sampling strategy are concerned with its application to larger regions and the availability of a multitude of sampling sites, as well as the possibility to avoid time-consuming field trips in often remote areas. However, with this concept mostly regional (meso-scale) environmental conditions are suitably represented. Furthermore, small-scale intraspecific physiognomic variations in response to climate may not be resolved with such an approach (Wolfe & Uemura, 1999).

Problems in leaf physiognomic studies

Although many studies have contributed to the understanding of interactions between the environment and leaf physiognomy, many problems still remain unresolved. Some of these problems are outlined briefly in the following.

In order to overcome some of these drawbacks, we developed a high-spatial-resolution data set based on synthetically generated floras from potential and/or actual distribution maps of plants for Europe. On the basis of this data set, firstly we analysed the present-day distribution patterns of leaf-physiognomic parameters in woody angiosperm vegetation. Secondly, we analysed these data for coherences between leaf-physiognomic and environmental spatial distribution patterns, and in which way they are statistically correlated. Thirdly, we investigated which environmental parameters are (pre)dominantly controlling the leaf physiognomy of extant vegetation. Finally, we developed transfer functions which may help to estimate environmental conditions from leaf physiognomy.

Materials and Methods

Our study on the distribution pattern of leaf physiognomic characters in Europe and its correlation with different climate parameters is based on woody dicotyledonous species which serve as excellent proxies for a range of environmental conditions (Spicer, 1989; Wolfe & Spicer, 1999). The restriction to woody plants is reasonable because these taxa better reflect macroclimatic conditions than herbs, which are more closely dependent on microclimate settings (e.g. Bailey & Sinnott, 1915; Wolfe, 1993). In addition, woody angiosperms are long-lived over several decades and thus their physiognomy may better reflect long-term climatic conditions than herbs. Based on our calibration data set consisting of (1) the floristic composition, (2) the leaf physiognomic characters and (3) the environmental parameters available for every geographical coordinate in the 0.5° × 0.5° latitude/longitude grid over the study area, the relationship among patterns of leaf physiognomy and environmental parameters were statistically analysed. The study area in Europe ranges from 10° West to 45° East and from 30° North to 65° North, and is covered by a total number of 5166 terrestrial grid cells representing > 50% of land cover (Fig. 1c). The spatial resolution of 0.5° × 0.5° latitude/longitude corresponds to a grid cell size of approx. 35 km West–East and 55 km North–South in central Europe (Traiser, 2004).

Figure 1.

Grid data sets of Europe (0.5° × 0.5° resolution); extraction of the leaf physiognomic calibration data set (c) from elevation data (a) and species diversity (b). The grid cells used for the calibration data set include only those with elevation between sea level and 400 m and with at least 25 species.

Calibration data set

We restricted our data set to those grid cells where at least 25 species are available so that species richness would be sufficient to allow reliable statistical analyses (cf. Fig. 1b). In addition, we considered only those grid cells which show an average elevation below 400 m (referring to the GLOBE NOAA elevation model on a 1-km resolution; Hastings & Dunbar, 1998) in order to exclude the complex influence of elevation on leaf physiognomy (Wolfe, 1993; Tang & Ohsawa, 1999; Hovenden & Van der Schoor, 2003; Christophel & Gordon, 2004) (Fig. 1a). As a result of these restrictions, the calibration data set used for the analysis consists of 1835 data points (Fig. 1c). This calibration data set is named the European Leaf Physiognomic Approach (ELPA) (Traiser & Mosbrugger, 2004a,b).

Floristic composition  In contrast to most of the previous leaf physiognomic data sets derived from direct field sampling, our data set relies on the use of synthetic chorologic flora lists which are based on chorologic maps of the potential distribution of present-day plants, and on floristic manuals of vegetation. For compilation, maps of the present-day potential distribution of 108 native woody species, which represent approximately two-thirds of the woody angiosperm taxa in Europe (Schmucker, 1942; Meusel et al., 1964, 1978; Meusel & Jäger, 1992; Jalas & Suominen, 1972, 1973, 1976, 1979, 1980, 1983, 1986, 1989, 1991, 1994; Jalas et al., 1996, 1999) are digitised, and geo-referenced for the application in a Geographical Information System (GIS). Rare and endemic species, mainly from the Mediterranean, have not been taken into account. For every geographical coordinate in the study area, a synthetic chorologic flora is then determined by the list of those of the 108 plants whose distribution maps cover that location (cf. Klotz, 1999; Klotz et al., 2003; Traiser, 2004).

Leaf physiognomic composition  In our study, the generalised leaf physiognomy of each of the 108 species is characterised by 25 leaf physiognomic characters as derived from floristic manuals (Götz, 1975; Krüssmann, 1976, 1977, 1978; Godet, 1986). These characters are described in Table 2. Following procedures by Wolfe (1993), values ranging between 0 (character is absent) and 1 (character is present) are assigned to each physiognomic character; this is exemplarily described for some taxa in Table 3 (Traiser & Mosbrugger, 2004c). On account of this characterisation, the leaf physiognomic composition of a synthetic chorologic flora is determined by calculating the total percentage for each leaf physiognomic parameter from all species which participate in the synthetic chorologic flora.

Table 2.  Leaf physiognomic characters with minimum, maximum and mean percentages and ranges as they occur in the calibration data set of 1835 sites (ELPA (European Leaf Physiognomic Approach) data set). Further details of character definitions are given in Wolfe (1993); he spatial distribution of characters 3, 16 and a–f are presented in Figs 2, 3 and 4
 Leaf characterAbbr.Definitionmin (%)max (%)mean (%)range
1leaf simplele_simpleaf is not compound80.096.690.216.6
2leaf lobedle_lobeleaf with lobes 7.135.221.528.1
3leaf margin entirema_entimargin without teeth11.148.025.536.9
4leaf size: leptophyll 1ls_lep1< 20 mm2 0.0 2.0 0.0 2.0
5leaf size: leptophyll 2ls_lep220–80 mm2 0.0 4.3 0.6 4.3
6leaf size: microphyll 1ls_mic180–400 mm2 4.319.2 8.714.9
7leaf size: microphyll 2ls_mic2400–1400 mm228.050.036.822.0
8leaf size: microphyll 3ls_mic31400–3600 mm217.742.434.624.7
9leaf size: mesophyll 1ls_mes13600–6200 mm2 6.720.313.713.6
10leaf size: mesophyll 2ls_mes26200–10000 mm2 0.0 8.3 3.6 8.3
11leaf size: mesophyll 3ls_mes3> 10000 mm2 0.0 3.4 1.3 3.4
12apex of blade: obtuseap_obtua > 90°10.028.316.118.3
13apex of blade: acuteap_acuta < 90°66.388.080.721.7
14apex of blade: emarginateap_emarapex emarginated 0.0 8.0 3.2 8.0
15base of blade: obtuseba_obtua > 90°34.359.648.125.3
16base of blade: acuteba_acuta < 90°13.554.026.640.5
17base of blade: embayedba_embabase cordate 6.033.825.227.8
18blade: length : width < 1 : 1lw_1ratio l/w: < 1 : 1 1.924.116.222.2
19blade: length : width 1–2 : 1lw_2ratio l/w: 1–2 : 134.073.360.639.3
20blade: length : width 2–3 : 1lw_3ratio l/w: 2–3 : 1 6.030.012.024.0
21blade: length : width 3–4 : 1lw_4ratio l/w: 3–4 : 1 0.010.0 2.510.0
22blade: length : width > 4 : 1lw_5ratio l/w: > 4 : 1 0.022.0 8.722.0
23shape of blade: obovatesh_obovlargest width in upper 1/3 of lamina 7.129.620.022.5
24shape of blade: ellipticsh_ellilargest width in middle 1/3 of lamina38.260.749.322.5
25shape of blade: ovatesh_ovatlargest width in lower 1/3 of lamina22.051.130.829.1
asmall leavesls_smalls_lep1, -lep2, -mic1 4.323.1 9.418.8
bmedium-sized leavesls_medils_mic2, −359.380.671.421.3
clarge leavesls_largls_mes1, −2, −3 9.225.818.616.6
dnarrow leaveslw_narrl/w_4, −5 0.028.011.328.0
eleaves medium l/w ratiolw_medil/w_2, −360.085.572.525.5
fbroad leaveslw_broal/w_1 1.924.116.222.2
Table 3.  Leaf physiognomic characterisation of three common European woody species. Further details of the scoring procedure are given in Wolfe (1993)
 Leaf characterAbbr.Acer campestreLonicera xylosteumTilia cordata
1leaf simplele_simp111
2leaf lobedle_lobe100
3leaf margin entirema_enti110
4leaf size: leptophyll 1ls_lep1000
5leaf size: leptophyll 2ls_lep2000
6leaf size: microphyll 1ls_mic1000
7leaf size: microphyll 2ls_mic2010
8leaf size: microphyll 3ls_mic30.501
9leaf size: mesophyll 1ls_mes10.500
10leaf size: mesophyll 2ls_mes2000
11leaf size: mesophyll 3ls_mes3000
12apex of blade: obtuseap_obtu00.50
13apex of blade: acuteap_acut10.51
14apex of blade: emarginateap_emar000
15base of blade: obtuseba_obtu10.50
16base of blade: acuteba_acut00.50
17base of blade: embayedba_emba001
18blade: length : width < 1 : 1lw_10.500.5
19blade: length : width 1–2 : 1lw_20.510.5
20blade: length : width 2–3 : 1lw_3000
21blade: length : width 3–4 : 1lw_4000
22blade: length : width > 4 : 1lw_5000
23shape of blade: obovatesh_obov000
24shape of blade: ellipticsh_elli110.5
25shape of blade: ovatesh_ovat000.5

Environmental parameters  The climate data used in the study focus on temperature- and precipitation-related parameters, and on insolation and vapour pressure, all of which are extracted from the global climate set provided by New et al. (1999) in a 0.5° × 0.5° latitude/longitude resolution (period from 1961 to 1990) (Traiser & Mosbrugger, 2004b). All 16 climate parameters considered for the quantification of the relationship with leaf physiognomy are listed in Table 4. Explicitly, the spatial pattern of MAT and mean annual precipitation (MAP) is shown across the study area (Figs 6a and 7a, later); both parameters have been selected because they represent some of the major limiting factors for the distribution of plants in Europe (Woodward, 1987), and because they allow the direct comparison with the results of previous studies (e.g. Wing & Greenwood, 1993; Wilf, 1997; Wiemann et al., 1998). However, despite the large area covered by our data set in Europe, the range of individual climatic parameters is relatively small (MAT, 3.5–17.5°C; MAP, 309–2139 mm; cf. Table 4) compared with other regions in the world, such as North America (MAT, −2.0–27.2°C; MAP, 190–4050 mm). Moreover, there is no region in Europe with high MAT coinciding with high MAP as can be observed, for example, in Florida.

Table 4.  Environmental parameters with minimum, maximum and mean percentages and ranges as they occur in the calibration data set of 1835 sites (ELPA (European Leaf Physiognomic Approach) data set). The spatial distribution of climatic parameters 1 and 7 are presented in Figs 6 and 7
 Climatic parameterAbbr.Scale unitminmaxmeanrange
1mean annual temperatureMAT[°C]   3.5  17.5   8.5  14.0
2mean of warmest monthTmax[°C]  17.4  32.2  24.0  14.8
3mean of coldest monthTmin[°C] −15.5   8.5  −5.4  24.0
4max. daily temperature rangeDTR[°C]   5.8  14.9  10.9   9.1
5max. annual temperature rangeATR[°C]  15.2  41.2  29.4  26.0
6annual temperature sumTsum[°C]2208.96398.43380.84189.5
7mean annual precipitationMAP[mm] 3092139 6591830
8precip. wettest monthPmax[mm]  33 237  82 204
9precip. driest monthPmin[mm]   0 108  35 108
10precip. growing seasonPGS[mm] 1711209 3491038
11precip. 3 driest consec. monthsP3min[mm]  15 354 116 339
12ground frost days per yearFD[days]   2 178 111 176
13precipitation days per yearPD[days]  78 239 161 161
14growing season lengthGSL[month]   4.0  12.0   6.0   8.0
15mean vapour pressureVAP[hPa]   7.3  15.7   9.5   8.4
16mean global radiationRAD[W/m2]  85 181 117  95
Figure 6.

Distribution pattern of mean annual temperature (MAT). (a), actual MAT; (b), predicted MAT from the MLR transfer function, r2 = 0.89, se = 0.9°C; cf. Table 7; (c), residuals of predicted and actual MAT.

Figure 7.

Distribution pattern of mean annual precipitation (MAP). (a), actual MAP; (b), predicted MAP from the MLR transfer function, r2 = 0.26, se = 143.4 mm; cf. Table 7; (c), residuals of predicted and actual MAP.

Correlation between leaf physiognomy and climate

To analyse statistically the relationship between the spatial patterns of leaf physiognomic composition and climate parameters across the study area, simple linear regression (SLR) and multiple linear regression (MLR) have been applied. Transfer functions like this are widely used in palaeo-environmental reconstruction of fossil floras (see Tables 6 and 8, later). To select the most significant leaf physiognomic parameters in the case of MLR, the ‘forward stepwise regression’ technique was used, allowing identification of four independent variables. The quality of the transfer functions is estimated on the basis of three statistical coefficients: (1) the coefficient of determination (r2); (2) the standard error of estimation (se); and (3) the ratio of se to the total range of the environmental parameter in the calibration data set (seR) (cf. Tables 6, 7 and 8 later). The last two are defined by

Table 7.  MLR-transfer functions of different environmental parameters. P-values of all functions P < 0.00001; the spatial distribution of climatic parameters 1 and 14 are presented in Figs 6 and 7; (r2) coefficient of determination; (F) F-value; (se) standard error of estimate; (seR) ratio of standard error of estimate
 yestunitb0b1x1 (%)b2x2 (%)b3x3 (%)b4x4 (%)r2FseseR (%)
1MAT°C   2.60  0.21(ba_acut) –0.25(sh_obov)  0.14(ma_enti) 0.14(lw_3)0.893460  0.9 6.6
2FDd 148.99 −2.69(ba_acut) –2.42(ma_enti)  2.58(sh_obov) 2.71(lw_1)0.862629 16.0 9.1
3Tsum°C 987.24109.25(ba_acut)67.63(sh_obov)177.16(ls_lep2)52.85(ls_mes1)0.852377313.2 7.5
4Tmin°C −39.26  0.42(ma_enti)  0.49(ls_mic2)  0.20(sh_ovat)0.86(ls_mes3)0.842247  2.1 8.9
5VAPhPa   8.86  0.09(ba_acut)  0.07(ma_enti) –0.10(sh_obov)0.08(lw_1)0.832114  0.6 6.6
6GSLmonth   3.67  0.09(ba_acut)  0.33(ls_lep2) –0.08(sh_obov) 0.08(lw_3)0.721171  0.6 8.1
7RADW/m2  79.22  1.58(ba_acut)  4.41(ls_mes2) –1.95(sh_obov) 2.21(ls_mic1)0.701028  8.9 9.3
8ATR°C  54.51 −0.37(ma_enti)  1.32(ls_mes2) –0.35(le_lobe)0.35(ls_mic2)0.66 873  3.212.3
9Tmax°C  48.78 −0.29(ls_mic3) –0.38(sh_obov)  0.24(ba_acut)0.37(ls_mic2)0.55 546  1.711.8
10PDd 173.18 −2.36(lw_3)  3.48(sh_obov) –4.22(ls_mic1)4.59(ls_mes2)0.54 523 17.510.9
11Pminmm 101.01  1.12(lw_2) –1.38(ma_enti)  1.73(ls_mes1)0.68(sh_elli)0.45 368  8.9 8.3
12P3minmm  77.90  2.64(lw_2) –3.34(sh_elli)12.59(ls_mes3) 4.29(ls_mes1)0.44 355 28.4 8.4
13PGSmm −99.14  2.45(ba_acut)  8.81(sh_ovat) 19.62(ap_emar) 4.04(lw_3)0.31 208 78.0 7.5
14MAPmm1768.43  3.45(ma_enti)11.54(sh_elli)36.49(ls_mes3)7.22(ap_acut)0.26 161143.4 7.8
15DTR°C  11.90  0.36(ls_mes2) –0.11(sh_obov) –0.04(le_lobe) 0.07(ls_mic1)0.19 108  1.112.6
16Pmaxmm  96.41  1.98(lw_3)  2.36(ls_mes1) –1.12(sh_elli)0.72(le_lobe)0.19 109 17.5 8.6
image

and

image

where yest is the estimated value of a given environmental parameter, yreal is the real value of a given environmental parameter, yreal(max) is the maximum of the real value in the calibration data set, yreal(min) is the minimum of the real value in the calibration data set and n is the number of sites in the calibration data set.

Results

Spatial distribution patterns of leaf physiognomic characters

The approach used in this study determines specific patterns of the distribution and general spatial gradients of leaf physiognomic characters across Europe. Note that the analysis relies on the leaf physiognomic composition of those species which participate in the synthetic chorologic floras and not on actual floras. Here, we consider four leaf physiognomic characters: (1) leaf size; (2) leaf shape in terms of leaf length : width (L : W) ratio; (3) leaves with an acute base; and (4) leaves with an entire margin.

Leaf size  To analyse the pattern of leaf size distribution across the study area, three classes of leaf sizes were distinguished: (1) small leaves (leptophyll 1, 2 and microphyll 1; area A < 400 mm2); (2) medium-sized leaves (microphyll 2, 3; 400 mm2 < A < 3600 mm2); and (3) large leaves (mesophyll 1, 2, 3; A > 3600 mm2) (cf. Table 2). Medium-sized leaves (Fig. 2b) predominate within European vegetation (59.3–80.6%), whereas small leaves (4.3–23.1%) (Fig. 2a) and large leaves (9.2–25.8%) (Fig. 2c) are relatively rare. Furthermore, distinct spatial patterns related to the abundance of these leaf size classes are revealed (Fig. 2).

Figure 2.

Distribution patterns of three leaf size classes; the intervals for this and the other figures have been chosen after the ‘natural break-method’ in arcview (ArcView GIS, ESRI, New York, USA). This method identifies breakpoints between classes using a statistical formula (Jenks optimization), minimizing the sum of the variance within each of the classes (Jenks, 1963). (a) small leaves (< 400 mm2); (b), medium-sized leaves (400–3600 mm2); (c), large leaves (> 3600 mm2); cf. Table 2 (rows a, b, c).

Small leaves (Fig. 2a) show a bimodal distribution pattern across Europe, with a maximum in the Mediterranean (∼23%) and a secondary maximum in the North (∼16%). In contrast, medium-sized leaves (Fig. 2b) are particularly abundant in floras in the Western Central oceanic areas (∼81%), whereas adjacent regions to the North and South as well as the Mediterranean are characterised by significantly lower percentages (∼59%). A local maximum of medium-sized leaves occurs in the Pannonian basin. In most regions, the distribution pattern of large leaves (Fig. 2c) opposes that of medium-sized leaves. Large leaves are most abundant in the Eastern continental regions (∼26%), whereas in the Atlantic region, in the Pannonian basin as well as in the Mediterranean, large leaves are less abundant (∼9%).

Leaf L : W ratio  Leaf L : W ratio characters were grouped into three classes: (1) long narrow leaves with L : W ratio larger than 3 : 1; (2) medium L : W ratio leaves with ratio of 1 : 1–3 : 1, and (3) broad leaves with ratio of <1 : 1 (cf. Table 2). Leaves of medium L : W ratio dominate the data set, with proportions ranging from 60 to 85.5% (Fig. 3b), whereas long narrow leaves (0–28%) (Fig. 3a) and broad leaves (1.9–24.1%) (Fig. 3c) are less frequent.

Figure 3.

Distribution patterns of three L : W ratio classes. (a), narrow leaves; (b), medium L : W ratio leaves; (c), broad leaves; cf. Table 2 (rows d, e, f).

The spatial distribution of these classes relying on synthetic chorologic floras is shown in Fig. 3, revealing that long narrow leaves (Fig. 3a) are most abundant in the Mediterranean (up to 28%), but are absent or rare in Scandinavia. A second maximum of this leaf physiognomic character is observed in Northeastern Europe. The floras of Western oceanic regions show high percentages (up to 86%) of leaves with medium L : W ratios (Fig. 3b) compared with the continental region of Northeastern Europe with lowest percentages (∼60%); low percentages also prevail in most of the Mediterranean regions. Broad leaves (Fig. 3c) show a reversed distribution pattern compared with that of medium L : W ratio leaves. Most frequent, broad leaves (up to 24%) are found across the Northeastern continental areas, whereas Mediterranean and Western regions are characterised by low abundances (∼2%).

Leaves with an acute base  Leaves with an acute base (ba_acut) show a mean value of 26.6% (Table 2), and have the widest range of all leaf physiognomic characters, with 13.5% as its minimum value and 54% as its maximum value (Fig. 4a). A clear gradient from South to North is evident, with highest proportions in the Southern Mediterranean (∼39–54%) and lowest values in the Northeast (14–21%) (Fig. 4a).

Figure 4.

Distribution patterns of two leaf physiognomic characters. (a), leaves with an acute base (ba_acut); (b), leaves with an entire margin (ma_enti); cf. Table 2 (rows 16, 3).

Leaves with an entire margin  The character leaf margin entire (ma_enti) is not dominant across European vegetation, evidenced by a mean value of 25.5%, but it shows a relative wide range of ∼37% (11.1% minimum and 48.0% maximum) (Table 2). There is also a differentiated distribution pattern of this character, showing a gradient from the Southern Mediterranean and the Western Atlantic regions with high proportions (∼37–48%) to the Northeast with low proportions (∼11–18%) (Fig. 4b).

Correlation between leaf physiognomic parameters and climate parameters

The observed spatial distribution patterns of leaf physiognomic characters which are evidently not random raise the question about the major controlling factors of these patterns. In order to analyse the relevance of climate as a possible controlling factor, we calculated (1) the correlation matrix between leaf physiognomic and climatic parameters; and (2) transfer functions for SLR and MLR for several climatic parameters.

Correlation between leaf physiognomic and climate parameters  The correlation matrix (Table 5) reveals that only a few leaf physiognomic characters are relatively highly correlated (r2 = 0.6) with specific climatic parameters. The four leaf physiognomic characters – leaf margin entire (ma_enti), leaf base acute (ba_acut), leaf L : W ratio <1 : 1 (lw_1) and leaf L : W ratio 2–3 : 1 (lw_3) – are correlated mainly with temperature-related parameters such as MAT, temperature sum (Tsum), coldest month mean temperature (Tmin), the number of ground frost days (FD) as well as with vapour pressure (VAP), a parameter which also depends on temperature. The four leaf characters are characterised by relative long physiognomic gradients (ranges of > 22%) within the data set (cf. Table 2). In general, leaf physiognomy seems to be more affected by minimum temperature parameters (i.e. Tmin and FD) than by maximum temperature (Tmax). Precipitation-related parameters are largely not correlated with any leaf physiognomic parameter. The correlation matrix reveals four essential facts for the European data set: (1) in general, the correlation between individual leaf physiognomic characters and climatic parameters is not high; (2) only a few leaf physiognomic characters exhibit climatic ‘information’; (3) the character leaf base acute (ba_acut) is best correlated with climate, whereas leaf margin entire (ma_enti) is more weakly correlated with climate; and (4) in general, leaf physiognomic traits are best correlated with minimum temperatures.

Table 5.  Correlation matrix of leaf physiognomic characters and environmental parameters. The values represent the coefficients of determination (r2); the grey marked cells represent r2 = 0.6; bold values: P < 0.01; abbreviations see Tables 2 and 4
r2MATTmaxTminDTRATRTSUMMAPPMAXPMINPGSP3MINFDPDGSLVAPRAD
le_simp0.350.050.360.000.220.280.120.080.080.170.090.320.070.220.360.10
le_lobe0.020.050.100.020.160.000.010.010.070.000.050.030.000.010.030.02
ma_enti0.600.060.710.010.480.490.190.100.090.170.130.680.110.390.620.19
ls_lep10.030.020.010.000.000.030.000.010.020.000.020.020.030.030.010.05
ls_lep20.220.070.220.000.100.250.080.080.000.140.010.210.080.290.230.18
ls_mic10.120.180.040.010.000.210.010.060.030.070.010.060.180.250.100.22
ls_mic20.360.030.430.020.300.280.020.010.000.060.000.400.060.220.350.11
ls_mic30.370.240.230.020.050.430.030.090.030.160.010.260.280.430.300.39
ls_mes10.040.000.050.000.040.030.010.000.090.000.080.050.010.040.040.02
ls_mes20.020.060.120.100.200.010.030.000.070.000.090.080.050.000.050.03
ls_mes30.290.020.360.030.260.250.120.030.100.060.140.370.020.180.330.06
ap_obtu0.120.000.210.010.170.090.100.040.030.050.050.130.010.090.120.03
ap_acut0.110.010.170.010.120.110.090.060.010.060.030.120.020.120.110.05
ap_emar0.010.050.000.000.010.050.010.060.060.030.030.000.030.090.010.08
ba_obtu0.560.170.460.000.210.490.090.070.030.140.060.520.190.340.510.27
ba_acut0.760.240.610.000.260.740.100.100.010.210.030.730.240.580.700.42
ba_cord0.350.120.270.000.110.390.040.050.000.110.000.360.090.360.330.24
lw_10.630.140.570.010.290.590.120.080.040.170.080.670.170.460.630.30
lw_20.030.140.000.030.040.080.010.030.250.030.210.000.080.120.010.15
lw_30.530.220.370.000.130.540.040.110.040.160.010.420.300.500.460.42
lw_40.400.130.340.010.140.390.080.080.020.150.030.390.110.300.370.25
lw_50.010.050.060.020.120.000.040.000.150.000.130.020.000.020.010.02
sh_obov0.130.190.030.070.000.120.020.000.030.020.020.070.190.100.090.18
sh_elli0.090.000.190.010.190.070.140.040.160.080.180.160.010.040.120.00
sh_ovat0.330.100.290.010.130.280.070.030.070.180.090.330.070.190.320.15

Transfer functions based on simple (SLR) and multiple linear regression (MLR)

SLR  Since the early works of Bailey and Sinnott (1915, 1916), many SLR transfer functions linking MAT with the leaf physiognomic character leaf margin entire (ma_enti) have been established on the basis of various calibration data sets. These parameters are regarded as highly correlated; leaves with entire margins are frequent in the low latitudes and infrequent in high latitudes. Derived from our results, a transfer function for MAT is established and compared with five other transfer functions based on non-European data sets published by previous authors (Wing & Greenwood, 1993; Wilf, 1997; Gregory-Wodzicki, 2000; Greenwood et al., 2004) (Table 6, Fig. 5). Our transfer function based on the European data set for margin entire is defined as

Figure 5.

Simple linear regression transfer functions for mean annual temperature (MAT) based on different calibration data sets from different regions of the world. The small numbers of the transfer functions correspond to Table 6.

MAT = 0.512 + 0.314 × ma_enti

with r2 = 0.60, F-value F = 2797, P < 0.00001.

In contrast to the results of previous authors, we observe only a relatively weak correlation between MAT and the relative abundance of entire margined leaves in the European vegetation (Table 6, function 3). In Europe, the strongest correlation with MAT is observed for base acute (Table 6, function 7) (Figs 4a and 6a), for which we established a transfer function defined as

MAT = −2.689 + 0.421 × ba_acut

with r2 = 0.76, F = 5751, P < 0.00001.

MLR  For a refined analysis, the application of MLR based on 16 different climatic parameters allows for a more detailed view on the relationship between leaf physiognomy and climate (Table 7). The transfer functions with the highest r2 again concern temperature-related parameters, whereas precipitation-related parameters show weak correlation. Concerning MAT, r2 increases up to 0.89 using MLR (r2 = 0.6 for SLR). The factors ba_acut and ma_enti, which showed the highest correlations with MAT (cf. Table 5), also represent some of the main factors in MLR. Although these leaf physiognomic characters are autocorrelated (r = 0.74), they are used in the transfer functions because they represent independent acquired data and nothing is known about the causal relationship of their correlation.

In order to test the quality of MLR transfer functions, we have chosen as an example two climatic parameters, MAT and MAP, for which the predicted values (Figs 6b and 7b; cf. Table 7) are compared with the actual values (Figs 6a and 7a). Additionally, the residuals (predicted minus actual climatic values) are determined (Figs 6c and 7c). Concerning MAT, the transfer function is defined as (cf. Table 7)

MAT = 2.6 + 0.21 × ba_acut − 0.25 × sh_obov + 0.14 × ma_enti + 0.14 × lw_3.

The application of the transfer function on the European data set results in a high spatial correlation (r2 = 0.89) between predicted MAT (Fig. 6b) and actual MAT (Fig. 6a). The residual plot (Fig. 6c) reveals a spatial pattern representing regions with systematic underestimations and overestimations. Regions which are slightly overestimated (1–2°C) refer to the Middle Russian plate and Northern Germany, whereas significant overestimates (2–4°C) occur in the region North of the Marmara Sea. In contrast, slight underestimation of MAT (1–2°C) can be observed for the Pannonian basin, Southern Ukraine, Southwestern France and Ireland. In some Mediterranean coastal regions, parts of the Pannonian basin, and Southern Ukraine, even stronger underestimations of MAT (2–4°C) occur.

Concerning MAP, the transfer function is defined as (cf. Table 7)

MAP = 1768.43 + 3.45 × ma_enti − 11.54 × sh_elli − 36.49 × ls_mes3 − 7.22 × ap_acut.

In contrast to MAT, the present-day distribution pattern of MAP (Fig. 7a) shows no clear gradient. Thus MAP cannot be reliably reproduced with any of our transfer functions (SLR and MLR). Although some basic trends of distribution patterns may be reflected by MLR, the transfer function is not capable of reproducing the high local variability of precipitation within our study area. Precipitation of > 1000 mm is not predicted by the transfer function at all. In the residual plot (Fig. 7a) it is apparent that the underestimation of MAP is particularly high in Western and coastal regions which are characterised by high precipitation.

Discussion

This study is based on the use of synthetic chorologic floras, which allowed establishment for the first time a leaf physiognomic data set covering a large area with a high spatial resolution in Europe. The results, as inferred from different statistical approaches, demonstrate the existence of strong relationships between leaf physiognomy and climatic conditions on a continental scale. Due to the large number of samples (1835), a high statistical significance is obtained. The physiognomic character-space determined by the 108 taxa used in this study proved to be highly diverse across Europe, with specific physiognomic characters (e.g. ma_enti, ba_acut) clearly correlated with specific climatic parameters (e.g. MAT).

Concerning the methodology, the use of generalised leaf physiognomy and synthetic chorologic floras has several advantages when compared with the classical procedure of sampling actual forest vegetation, as follows.

  • 1It provides easy access to very large data sets and allows covering large areas, also avoiding time-consuming field sampling.
  • 2Following this procedure, the influence of local edaphic or microclimatic effects on the floristic composition and leaf physiognomy, which may confound macroclimatic effects, can largely be avoided (cf. Givnish, 1979; Turner, 1994). This technique reflects the generalised physiognomic character of the vegetation.
  • 3It considers the potential natural vegetation and is not influenced by anthropogenic biases, as in the case of sampling actual forest vegetation.
  • 4In contrast to the various techniques used so far (e.g. Dolph & Dilcher, 1979; Wolfe, 1993; Wilf, 1997; Jacobs, 1999) this procedure represents an objective and reproducible ‘sampling’ method.

With respect to possible drawbacks of the procedure, there may be a loss of information when using synthetic chorologic floras in form of grided data because they do not consider local floristic compositions as well as local microclimatic effects. However, these features are probably of minor importance in a more general (macro-scale) study on the correlation between leaf physiognomy and environment. It should be pointed out that in previous leaf physiognomic studies (e.g. Wolfe, 1993; Wilf, 1997; Gregory-Wodzicki, 2000) local climatic effects are also mostly not considered due to larger distances between sampling sites and the meteorological stations from which the climatic data were obtained (up to 40 km; Wolfe, 1993). Furthermore, our method can not determine small-scale changes in physiognomic characteristics as they may be found in actual vegetation; for instance, synthetic chorologic floras cannot resolve the physiognomic differences between riparian and hinterland vegetation (Burnham et al., 2001; Kowalski & Dilcher, 2003).

With respect to their relative abundance, three basic groups of leaf characters can be distinguished: (1) characters that are dominant in the European vegetation (relative abundance above 50%), e.g. le_simp, ap_acut, ls_medi, lw_medi; (2) characters which are generally very rare (with relative abundance below 10%), e.g. ap_emar; and (3) characters with an intermediate abundance and a relative wide range of occurrence, e.g. ma_enti, ba_acut, lw_1. Within the last group, leaf characters are most strongly related to temperature parameters (cf. Table 2). In this context, dominant leaf characters seem to provide close to optimal functional performance under the environmental conditions covered by our data set.

Spatial distribution patterns of leaf physiognomic characters

Regarding the relationship between climate and leaf physiognomy in the European data set, it is evident that specific temperature parameters correlate significantly with leaf physiognomy (Table 5). In contrast, the correlation with precipitation parameters is generally much weaker. The distribution maps of some leaf physiognomic characters, e.g. of small-sized leaves, show similar abundances in climatically different regions, thus underlining the complex influence of climate on leaf physiognomy (cf. Figs 2, 3 and 4).

It has long been recognised by ecologists that leaf size tends to change in relation to many environmental gradients such as MAT (e.g. Dolph, 1977; Givnish, 1984), MAP (e.g. Beard, 1945; Cowling & Campbell, 1980; Givnish, 1984; Stone & Bacon, 1995; McDonald et al., 2003; Uhl & Walther, 2003), soil moisture (Givnish, 1979), low soil nutrients (e.g. Beadle, 1954; Chapin, 1980; Dolph & Dilcher, 1980; Ashton & Hall, 1992; Cunningham et al., 1999; McDonald et al., 2003) and elevation (e.g. Givnish, 1984; Halloy & Mark, 1996; Cordell et al., 1998). On a global scale, there is a trend of increasing leaf size with increasing temperature and precipitation: the optimal leaf size should be greatest in the tropics, decreasing towards the subtropics, increasing towards the warm temperate forest, and decreasing towards the poles (Givnish, 1976). This general observation is in good agreement with the observation that medium-sized leaves are predominant (59–81%) in our European data set (Fig. 2b). Furthermore, regarding the distribution patterns within the three leaf size classes, large leaves (> 3600 mm2; Fig. 2c) are most abundant in cool climates, small leaves (< 400 mm2; Fig. 2a) are most abundant in warm climates, and medium-sized leaves (400–3600 mm2) are most abundant in intermediate climates. Hence, this trend is also in accordance with the studies by Givnish (1976).

In our data set the L : W ratio of leaves decreases with increasing latitude, corresponding to an increase in the percentage of broad leaves with increasing latitude. This pattern may be caused by factors linked to environmental gradients parallel to latitude. For example, the mean leaf L : W ratio of Fagus sylvatica, Acer campestre and Acer pseudoplatanus in Southwest Germany tends to increase with decreasing humidity and water availability (Uhl, 1999). Like leaf size, it is well known that the leaf L : W ratio may be influenced by several climatic parameters, and previous investigators found significant correlations between the mean L : W ratio of a flora and PGS (e.g. Wiemann et al., 1998; Wilf et al., 1998), as well as with log(MAP) (e.g. Jacobs, 1999). In the latter studies, L : W ratios of floras increased with increasing precipitation, a fact that is in contrast to the results obtained from individual taxa (e.g. Uhl, 1999). Summarising, these results may suggest a bimodal distribution with relatively large, broad leaves in areas with high water availability (i.e. tropical rainforests), and small, narrow leaves in areas with low water availability (i.e. the Mediterranean area) (e.g. McDonald et al., 2003). In general, it has been found that the overall L : W ratio is about 1 : 1 in deciduous forests and >3 : 1 in evergreen forests (Greenwood & Basinger, 1994).

The positive correlation between ba_acut and MAT (r2 = 0.76) that was observed in our data set corroborates the results by Wolfe (1993). Wolfe found a similar relationship on the basis of his Climate Leaf Analysis Multivariate Programme (CLAMP) data set covering 106 samples mostly from North America and East Asia, suggesting that ba_acut is particularly characteristic for mega-thermal and warm meso-thermal vegetation in humid to mesic environments. Subsequently, Wiemann et al. (1998) published a coefficient of correlation of r2 = 0.42 for the correlation between ba_acut and MAT for the slightly enlarged CLAMP 3B data set, which consists of 144 sites. In contrast to other studies where the character ma_enti is correlated most strongly with MAT (e.g. Bailey & Sinnott, 1915, 1916; Wolfe, 1979, 1993; Wilf, 1997; Wiemann et al., 1998), our study reveals a high coefficient of determination for this parameter (r2 = 0.6), but still lower than for the character ba_acut, reflecting the highest correlation. All in all, the reasons for the discrepancies in these results based on different leaf physiognomic data sets may be due to: (1) the use of 108 taxa (which only represent about two-thirds of the European hardwood taxa) in the European flora data set; (2) the insufficiency of previous data sets to suitably reflect the physiognomy or environmental variability in a study area (e.g. in the CLAMP data set, not all modern physiognomy-climate combinations are covered (e.g. Siberia); R.A. Spicer, pers. comm.); or (3) a different relationship between leaf physiognomy and environmental parameters in Europe compared with other regions (an interpretation in accordance with previous workers (e.g. Stranks & England, 1997; Gregory-Wodzicki, 2000)) which opted against a globally uniform relationship between leaf physiognomy and climate.

Climate estimates based on leaf physiognomy

Although the coefficient of correlation in the European data set is smaller than in previous studies (e.g. Wing & Greenwood, 1993; Wilf, 1997; cf. Tables 5 and 6), the resulting SLR transfer functions are very similar with respect to their slopes (slope b1 = 0.3), whereas the intercepts (b0) differ markedly. This suggests that there exist the same relationships between relative abundance leaf margin type and MAT in different regions of the world. Nevertheless, there are regionally variable starting points of the abundance of this character in the vegetation. The most striking difference is observed between the ‘Australian’ transfer function (function 6) and the others based on East Asian, European and American vegetation (functions 1–5): leaves with entire margins are much more frequent in Australian vegetation (for a given MAT) than in other continents. According to Greenwood et al. (2004), the lack of toothed taxa in the Australian vegetation may be caused by the absence of truly cold forest climate space on that continent today and possibly for all of the Cenozoic. On the other hand, Europe (function 3) and East Asia (function 2) are characterised by very similar portions of untoothed leaves in the vegetation at a given MAT. As a possible explanation, Greenwood et al. (2004) suggest a global evolutionary convergence of leaf physiognomy in response to the selection of temperature in mesic habitats. Our results derived from a data set which covers a totally different climatic space and a vegetation with a totally different systematic composition and vegetational history substantiates such an interpretation.

The complex relationships between climate and leaf physiognomy are significantly better reflected by MLR transfer functions than by SLR transfer functions (cf. Table 6 and Table 8). In comparison to other MLR transfer functions for MAT based on non-European data sets (Wing & Greenwood, 1993; Gregory & MacIntosh, 1996; Wiemann et al., 1998), similar r2, but se of about half can be observed within the European data set (Table 8). The application of the non-European transfer functions to our data set may lead to a considerable increase in se and a decrease in r2. This indicates that the application of transfer functions based on physiognomic data from outside Europe may therefore result in less accurate predictions for European floras. The results of MLR transfer functions also support the regional variability of the relationship of leaf physiognomic characters and environmental parameters. An interesting point is the fact that nearly all MLR transfer functions include similar leaf physiognomic characters such as ‘leaf margin entire’ and ‘base acute’ (cf. Table 8). This implies that these leaf characters contain a high portion of ‘climatic information’ not only on a regional but also on the global scale. The observed autocorrelations between some of these leaf physiognomic characters is ascribed to the existence of leaf physiognomic character syndromes in leaf morphology. However, each of these autocorrelated characters may contain different climatic information and consequently they are useful variables in transfer functions.

The MLR transfer function of MAT is suitable for predicting realistic values using leaf physiognomic parameters. This is demonstrated by the spatial distribution pattern of predicted values (Fig. 6b) showing a clear, more or less zonal organisation. In contrast, MAP (cf. Fig. 7b) and all other precipitation-related parameters (cf. Table 7) are not satisfyingly reproduced by MLR transfer functions within the European data set. This drawback may be explained by the very complex distribution pattern of precipitation within Europe (cf. Fig. 7a). Vegetation and thus leaf physiognomy seems not to reflect this sophisticated distribution patterns of precipitation parameters. It is likely that other parameters such as those related to soil characteristics (e.g. the water holding capacity or the ground water table) have a great influence on water availability to the vegetation and override any signal from precipitation alone. An example is the known physiognomic differences between riparian forests (high ground water table) and their corresponding hinterland vegetation (lower ground water table) (e.g. Burnham et al., 2001; Kowalski & Dilcher, 2003).

Finally, it should be emphasised that there is no indication in our study that climate and leaf physiognomy (and thus vegetation) are not in equilibrium within Europe. This is evidenced by the mostly deterministic patterns observed here. This interpretation is also supported by recent studies of Holocene vegetation dynamics suggesting very rapid migrations and vegetation reactions to abrupt climatic changes (Tinner & Lotter, 2001).

Conclusions

The study provides evidence that synthetic chorologic floras represent a powerful tool for analysing leaf physiognomic spatial patterns across Europe and their relationships to climate. For the first time, a European data set which is based on synthetic floras derived from potential distribution maps of plants has been compiled in order to determine the spatial patterns of leaf physiognomy. For this, 1835 data points representing synthetic floral lists were characterised by different leaf physiognomic parameters, and correlated with different climate parameters. This procedure overcomes the problem of using an anthropogenically influenced data set based on direct sampling of actual forest vegetation. As a result of the analysis, clear spatial patterns of leaf physiognomy have been observed which are related to certain climate parameters, e.g. temperature. Interestingly, contrary to the results of many previous studies, entire margin was not the character most strongly correlated with MAT; rather, it was acute base.

Based on the observed relationships, transfer functions for several climatic parameters have been established. These transfer functions differ slightly from previously published functions based on data sets from different regions of the world in respect of the slopes of the functions. Such differences may be caused by vegetation history and/or specific local edaphic, ecological or climatic conditions. Thus, the spatial restrictions of transfer functions should be taken into account when applying them to estimate palaeo-climatic conditions from fossil floras.

Acknowledgements

We thank B. Jacobs, L. Strickland, and B. Thompson for helpful discussions on leaf physiognomy. DU acknowledges financial support by the Deutsche Forschungsgemeinschaft (DFG grant UH 122/1–1) and SK thanks the Alexander von Humboldt-foundation for financial support.

Ancillary