Journal of Biogeography (J. Biogeogr.) (2006) 33, 1750–1763

SPECIAL ISSUE

Soil nutritional factors improve models of plant species distribution: an illustration with Acer campestre (L.) in France Christophe Coudun*, Jean-Claude Ge´gout, Christian Piedallu and Jean-Claude Rameau 

LERFoB, UMR INRA-ENGREF 1092, Ecole Nationale du Ge´nie Rural, des Eaux et des Foreˆts, 14, Rue Girardet, CS 4216, 54042 Nancy Cedex, France

ABSTRACT

Aim To estimate the relative importance of climate and soil nutritional variables for predicting the distribution of Acer campestre (L.) in French forests. Location France. Methods We used presence/absence information for A. campestre in 3286 forest plots scattered all over France, coupled with climatic and edaphic data. More than 150 climatic variables (temperature, precipitation, solar radiation, evapotranspiration, water balance) were obtained using a digital elevation model (DEM) and a geographical information system (GIS). Six direct soil variables (pH, C/N ratio, base saturation rate, concentrations of calcium, magnesium and potassium) were available from EcoPlant, a phytoecological data base for French forests. Using a forward stepwise logistic regression technique, we derived two distinct predictive models for A. campestre; the first with climatic variables alone and the second with both climatic and edaphic variables. Results The distribution of A. campestre was poorly modelled when including only climatic variables. The inclusion of edaphic variables significantly improved the quality of predictions for this species, allowing prediction of patches of presence/absence within the study region.

 Deceased *Correspondence: Christophe Coudun, Centre for Terrestrial Carbon Dynamics (CTCD), Forest Research, Alice Holt Lodge, Farnham, Surrey GU10 4LH, UK. E-mail: [email protected]

Main conclusion Soil nutritional variables may improve the performance of fine-scale (grain) plant species distribution models. Keywords Climatic variables, EcoPlant, logistic regression, pH, soil nutritional variables, spatial autocorrelation, species distribution modelling.

INTRODUCTION Predictive modelling of species and habitat distribution is a rapidly evolving topic in ecology. Hundreds of studies have been conducted since the end of the 1990s (Franklin, 1995; Guisan & Zimmermann, 2000; Guisan et al., 2002; Lehmann et al., 2002b; Scott et al., 2002; Rushton et al., 2004; M.P. Austin, in review). In particular, terrestrial plant species have often been subject to distribution modelling (see e.g. Carpenter et al., 1993; Lenihan, 1993; Guisan et al., 1998, 1999; Zimmermann & Kienast, 1999; Hooten et al., 2003; Randin et al., 2006). Information on species and habitat distribution is needed for many purposes, for example to model the risk of invasion (Peterson et al., 2003; Peterson, 2003), to assess the consequences of global change (Iverson et al., 1999; Schwartz et al., 2001; Theurillat & Guisan, 2001; Bakkenes et al., 2002; 1750

www.blackwellpublishing.com/jbi doi:10.1111/j.1365-2699.2005.01443.x

Dirnbo¨ck et al., 2003; Skov & Svenning, 2004; Arau´jo et al., 2006), to test biogeographical hypotheses (Huntley et al., 2004; McPherson et al., 2004) or to deal with conservation planning problems (Margules & Stein, 1989; Arau´jo & Williams, 2000; Polasky et al., 2000; Arau´jo et al., 2004; Cabeza et al., 2004; Williams et al., 2005). The combination of particular plant species models may also help mapping of community and vegetation type (e.g. Cawsey et al., 2002; Elith et al., 2002). Existing models rely on the assumption of pseudo-equilibrium between a particular species and its environment (Leathwick, 1998; Guisan & Zimmermann, 2000; Arau´jo & Pearson, 2005), with all potential favourable habitats occupied by the species. Many statistical techniques have been developed to model the response of species to environmental conditions (Yee & Mitchell, 1991; Huisman et al., 1993; Elith, 2000; Vayssie`res et al., 2000; Gelfand et al., 2003; Robertson et al., ª 2006 The Authors Journal compilation ª 2006 Blackwell Publishing Ltd

Distribution of Acer campestre in France 2004). Among existing techniques, generalized linear models (GLMs; McCullagh & Nelder, 1997) and generalized additive models (GAMs; Hastie & Tibshirani, 1997) are the most commonly used (e.g. Bio et al., 1998; Bio, 2000; Austin, 2002; Ferrier et al., 2002; Luoto et al., 2006). Specific software has been developed for GLM/GAM modelling purposes (e.g. grasp; Lehmann et al., 2002c). The ecological response and the spatial distribution of plant species is often studied with regard to climatic factors alone. These are postulated to be the most important drivers of species distributions at large scales (Box et al., 1993; Shao & Halpin, 1995; Heegaard, 2002; Lehmann et al., 2002a). Indeed, with the development of geographical information systems (GIS), access to climatic data layers interpolated from meteorological station measurements is facilitated (McKenzie & Halpern, 1999; McKenzie et al., 2003a,b; Thuiller et al., 2003b, 2004). However, soil factors are also known to be important for plant species, although they are often disregarded because accurate data are often lacking. As a consequence, simultaneous study of the response of plant species to both climatic and soil nutritional factors has rarely been performed (see, however, Austin, 1971, 1972; Gignac et al., 1991; Gignac, 1992; Bragazza & Gerdol, 1996; Heegaard, 1997, 2001; Pinto & Ge´gout, 2005). The main objective of this study was to test whether the integration of soil nutritional factors improved the distribution models of plant species. Thus, we characterized the ecological response of a tree species to both climatic and soil nutritional factors in France using logistic regression models. We chose Acer campestre because this species is known to be sensitive to both climate and soil nutritional conditions. Acer campestre L. (field maple) is a 12–15 m tree that can be found in hedgerows and deciduous mixed forests almost anywhere in Europe (Rameau et al., 1989; Mills, 1996), but it is absent from the northern and Mediterranean areas (Tutin et al., 2001). Although the autecology of the species remains poorly known, there are indications that the species prefers temperate climates and rich soils (calcareous soils, limestone), and is often present on dry sites (Motel, 1995; Bendixen, 2001). In France, it is present in most parts of the country with the exceptions of many mountainous areas, as well as the Atlantic and the Mediterranean regions (Rameau et al., 1989). Since the 1990s, the economic interest in this species has increased in France, but it still remains largely unmanaged by foresters. MATERIALS AND METHODS Calibration data set To study the ecological response of A. campestre in French forests, we extracted 3286 phytoecological releve´s (i.e. samples) from the EcoPlant data base (Ge´gout, 2001; Ge´gout et al., 2005). This data base stores complete species lists with environmental information (climate- and soil-related variables measured in the field or chemically and physically analysed in laboratory) for more than 6400 forest plots scattered over

France (Ge´gout et al., 2005). Here, we selected 3286 plots with sufficient edaphic information, and found that A. campestre was present in 460 of these plots (Fig. 1). Most calibration plots had 400 m2 of area, consistent with the current phytoecological practice (Mueller-Dombois & Ellenberg, 1974). We used presence/absence information, where 1 was attributed to plots where A. campestre was present and 0 to plots where it was absent.

Climatic data The position of all 3286 selected plots was known with a precision of 1 km and 159 climatic variables and indices were available as layers with 1 km2 resolution in a GIS that covers the whole of France. The climatic variables were obtained from a digital elevation model (DEM), and from the meteorological model Aurelhy (Benichou & Le Breton, 1987) based on interpolated measurements (Table 1). We selected climatic variables that are postulated to have an influence on (1) plant growth (T, mean temperature; T05.09, mean temperature from May to September; GDD6, growing degree-days above 6 C; VP6, length of vegetation period above 6 C), (2) primary production (PET, potential evapotranspiration; AET, actual evapotranspiration; R, radiation), (3) frost conditions (Tn, minimum temperature; FD, number of frost days) and (4) drought (Tx, maximum temperature; P, mean precipitation; MoistRat, moisture index; SMD, soil moisture deficit; SMS, soil moisture surplus; DP, length of drought period; WB, water balance; see Table 1 for definitions). We used two different methods of computing potential evapotranspiration, based on the work of Thornthwaite & Mather (1957) and Turc (1961), respectively. Most of these climatic variables have been used in recent studies dealing with plant species distribution modelling (e.g. Leathwick & Rogers, 1996; Leathwick & Whitehead, 2001; Pearson et al., 2002; Thuiller, 2003). There was strong collinearity among the climatic variables (especially variables from succeeding months), and our aim was therefore to find a limited set of climatic variables best describing the current the distribution of A. campestre.

Soil data Many authors have promoted the use of direct variables in plant ecological studies (e.g. Austin & Meyers, 1996; Austin, 2002; Lawesson & Oksanen, 2002; Diekmann, 2003). In practice, such variables have rarely been integrated in models because access to site-specific measurements of the chemical properties of soils requires soil sampling in the field and further expensive analyses in the laboratory. Instead, in Europe the soil characteristics of plots are often computed with species lists and associated Ellenberg indicator values (Ellenberg et al., 1992), which represent subjective numerical assessments of the responses of species to environmental conditions. Ellenberg et al. (1992) indeed derived preference values for moisture, soil reaction (acidity and lime content), nutrient availability (soil nitrogen status), salinity (soil chloride concentration), light regime,

Journal of Biogeography 33, 1750–1763 ª 2006 The Authors. Journal compilation ª 2006 Blackwell Publishing Ltd

1751

C. Coudun et al.

Figure 1 Location of the 3286 investigated forest sites on the topographical map of France (calibration data set), with indication of the presence (460 black points) and absence (2826 white points) of A. campestre (from EcoPlant) (Ge´gout et al., 2005).

temperature and continentality for 2726 vascular plants in Central Europe. Ellenberg indicator values allow the estimation of bioindicated site variables when a complete floristic inventory is available (ter Braak & Barendregt, 1986). In this study, however, for the 3286 releve´s extracted from EcoPlant we had information on six measured soil nutritional variables, representing acidity (pH), nitrogen conditions (C/N ratio), base saturation (S/T ratio, where S ¼ [Ca] + [Mg] +[K] and T ¼ S + [Al] + [H]) and mineral nutrient concentrations (Ca, Mg and K in their logarithmic form; Table 1). Evaluation data set To evaluate our climate and soil/climate models, we had access to 88,004 additional phytosociological releve´s from the French Forest National Inventory (IFN), scattered over France, and for which we had information on the presence/absence of A. campestre. The sampling strategy of the French Forest National Inventory is based on a 500 · 500 m2 grid map of France, within which sampled plots are chosen randomly. The minimum distance between two evaluation plots is thus also 500 m. Climatic variables were obtained for these 88,004 plots by overlay with the same GIS layers used for the calibration data set. It was also possible to compute a bioindicated value, i.e. the weighted average of the ecological optima of all plant 1752

species present on each plot (ter Braak & Barendregt, 1986), for three of our selected edaphic variables (pH, C/N and S/T). The species optima were extracted from an unpublished catalogue of edaphic and climatic indicator values for 700 forest plant species in France (Ge´gout et al., 2002, 2003). Because of a lack of data, edaphic variables are often difficult to map across large regions and precise maps of soil nutritional characteristics are still under development. For this study, we created grid maps for three edaphic variables (pH, C/N and S/T) with 1 km2 resolution, using the inverse distance weighting (IDW) spatial interpolation method (Philip & Watson, 1982) for the bioindicated values from the evaluation data set. Logistic regression modelling We used forward stepwise logistic regression to model the response of A. campestre to climate and soil nutrient factors (ter Braak & Looman, 1986; McCullagh & Nelder, 1997). Logistic regression, a special case of a GLM with a logit link function and a binomial error distribution, has often been implemented to characterize species–environment relationships (e.g. Odland et al., 1995), and is easy to implement in most statistical packages. We built two different models for A. campestre: the first with climatic variables alone (159 candidate variables) and the second with both climatic and soil

Journal of Biogeography 33, 1750–1763 ª 2006 The Authors. Journal compilation ª 2006 Blackwell Publishing Ltd

Distribution of Acer campestre in France Table 1 List of available (a) climatic and (b) edaphic variables to model the distribution of Acer campestre in French forests, with minimum, mean, and maximum values Code

Variable

Unit

Minimum

Mean

Maximum

(a) Climatic variables T Annual temperature* T05.09 Temperature from May to September GDD6 Growing degree-days (above 6 C) VP6 Length of vegetation period (above 6 C) Tn Annual minimum temperature* FD Frost days* Tx Annual maximum temperature* DP Length of drought period R Annual radiation* P Annual rainfall* pAu Autumn rainfall (October, November and December)  PET Potential evapotranspiration*à AET Actual evapotranspiration*à AETTh Actual Thornthwaite evapotranspiration  MoistRat Moisture index (1 – AET/PET)*à SMD Annual soil moisture deficit (AET – PET)*à SMS Annual soil moisture surplus (P – PET)*à WB Water balance (yearly sum of monthly P – PET)*à

C C C days C days C days kJ. cm)2 mm mm mm mm mm – mm mm mm

4.8 10.8 754 163 )0.2 21 9.5 0 294 542 111 435 378 402 0.00 )399 90 )218

9.7 15.7 1748 246 5.0 77 14.4 129 429 990 280 673 509 502 0.23 )163 476 312

13.5 19.3 2743 365 9.2 180 18.8 223 590 2264 714 912 685 611 0.47 0 1685 1682

(b) Edaphic variables (A horizon of soil) pH pH(H2O) ST Base saturation rate CN C/N ratio lnCa Calcium content§ lnMg Magnesium content§ lnK Potassium content§

– % – mEq mEq mEq

3.3 0.0 7.9 )3.2 )4.2 )3.9

5.1 51.6 17.0 0.8 )0.7 )1.4

8.2 100.0 45.8 4.4 2.3 0.4

*Monthly values were also available, and were tested in the models.  pAu and AETTh have been added to this table since they contribute to explain the distribution of Acer campestre (see also Table 2). àTwo different methods were used to compute potential evapotranspiration, based on (1) Turc (1961) and (2) Thornthwaite & Mather (1957). By default, the figures presented in this table have been computed with Turc evapotranspiration. §Calcium, magnesium and potassium contents are expressed in mEq per 100 g dry soil, and are presented in their logarithmic form.

nutritional variables (165 candidate variables). For each model a forward stepwise selection procedure, based on the maximum decrease in residual deviance, was adopted to successively select the variables that were most relevant in explaining the ecological response of the tree species (McCullagh & Nelder, 1997; Bio, 2000). Variables with their quadratic form were tested at each step in order to account for bell-shaped response curves (ter Braak & Looman, 1986). When the incorporation of the quadratic form of the variable [tested with a residual deviance test (McCullagh & Nelder, 1997)] was not significant (at the 0.001 level), the variable was kept in its simple form, thus describing a monotonic response curve. The stepwise procedure was stopped when the incorporation of a supplementary variable (in its simple form or coupled with its quadratic form) did not lead to a significant reduction in residual deviance, at the 0.001 level.

absence data, an optimal threshold value of probability from a receiver operating characteristic (ROC) curve (Fielding & Bell, 1997), was selected. This threshold allowed transformation of the predicted probabilities of presence into predicted presence/ absence binary values. A ‘confusion’ matrix was then computed to compare predicted and observed presence/absence, for this threshold. Three synthetic measures to qualify the rate of correct classification were derived: (1) global success (S, correct classification rate of both presence and absence), (2) sensitivity (Sn, correct classification rate of presence), (3) specificity (Sp, correct classification rate of absence). The area under the ROC curve (AUC) was used to assess the predictive performance of our models for the evaluation set (Fielding & Bell, 1997; Elith & Burgman, 2002; Elith, 2002). All computations were performed with the S-Plus statistical package (MathSoft, Inc., 1999). Spatial autocorrelation

Quantitative model evaluation To compare the resulting probabilities of presence of A. campestre obtained from both models, with real presence/

Spatial autocorrelation can inflate the predictive performance of models (Legendre, 1993; Lennon, 2000; Hampe, 2004; Arau´jo et al., 2005) when calibration and evaluation plots are too close

Journal of Biogeography 33, 1750–1763 ª 2006 The Authors. Journal compilation ª 2006 Blackwell Publishing Ltd

1753

C. Coudun et al. to each other. To avoid this problem we tried to select plots for calibration of our models that were geographically and ecologically independent of each other (Coudun, 2005). We used the following rule: when two plots were separated by less than 500 m we kept both of them in the data only if their ecological conditions were different in terms of altitude (at least 50 m), aspect (at least 90) or soil pH (at least one unit). We then used indicator kriging (Journel, 1983; Isaaks & Srivastava, 1989), a specific geostatistical analysis for categorical variables such as presence/absence, that is based on ordinary kriging (see Bayliss et al., 2005). In this study, indicator kriging was used on the observed presence/absence data from the evaluation set (see also Miller & Franklin, 2002; Marinoni, 2003; Bayliss et al., 2005). A semi-variogram computed with ArcGis (ESRI, 2002), was then used to evaluate the spatial dependence (c) in relation to the distance between evaluation plots. We examined this relationship for short distances (less than 20 km) because other studies have indeed shown that the abundance of Acer species could be autocorrelated at a local scale (e.g. Clark et al., 1998; Schwarz et al., 2003). Indeed, the presence of one tree might inflate the probability of the presence of neighbouring trees within the seed dispersal distance of the species. However, no increase in the semi-variogram was detachable for very short distances (see Fig. 2 for the 88,004 evaluation plots) meaning that the species’ dispersal does not seem to have an important effect on short distance autocorrelation. To investigate further the possible influence of spatial autocorrelation on the predictive success of our models, the four evaluation measures (AUC, area under the ROC curve; S, global success; Sn, sensitivity; Sp, specificity) were computed with different subsets of the evaluation set. First, evaluation sample plots were selected that were separated from calibration plots by increasing distances, namely 500 m, 1, 5, 10 and 20 km. Then, evaluation sample plots were selected that were separated from calibration plots and from each other by the same increasing distances (Fig. 3). Finally, evaluation sample data sets with different sizes were selected randomly.

RESULTS Climatic response of Acer campestre The forward stepwise selection procedure led to a two-variable climatic model, including mean autumn precipitation (pAu, sum of mean October, November and December precipitation) and the mean actual evapotranspiration calculated with the Thornthwaite method (AETTh). The final equation linking the probability (p) of presence of A. campestre to climatic factors was:   p ln ¼ 6:331  0:01462pAu þ 0:01636AETTh ð1Þ 1p The response of A. campestre to both variables was monotonic, either decreasing in the case of autumn precipitation (pAu, preference for dry sites), or increasing in the case of actual Thornthwaite evapotranspiration (AETTh, preference for sites with maximal growth conditions; Fig. 4). The model’s reduction in deviance was 118 and 117 for pAu and AETTh, respectively (Table 2a), and the inclusion of additional variables in this model did not bring a significant reduction in deviance. The predicted distribution map resulting from the climatic model (Fig. 6c) reveals low probabilities of presence of A. campestre in the mountains (Fig. 6b: regions 1a, Vosges; 1b, Jura; 1c, Alps; 1d, Massif Central; 1e, Pyrenees) and hyperAtlantic (e.g. Fig. 6b: region 3a, Bretagne) regions. These are zones where values of autumn precipitation seem to be too high to allow the presence of A. campestre. The model also predicts low probabilities in the Mediterranean area (Fig. 6b: 2, Mediterranean region), where low pAu is favourable but where AETTh is too low. The climatic model predicts higher probabilities of presence of A. campestre in a south-west/northeast zone excluding mountainous areas. However, the poor quality (Swets, 1988) of this model is illustrated by a low value for the area under the ROC curve (AUC ¼ 0.72 for the calibration data set and AUC ¼ 0.64 for the evaluation data set; see Table 2a).

γ 0.215

Climatic and edaphic response of Acer campestre

0.172 0.1129 0.086 0.043 0

2.5

5

7.5

10 12.5 Distance (km.)

15

17.5

20

Figure 2 Semi-variogram computed with the indicator kriging technique, on A. campestre presence/absence, for the evaluation data set (88,004 plots). Each dot represent pairs of points that have a common distance (lag size) and direction between each other; 40 lags of 500 m are represented, showing relationships between plots for a distance up to 20 km. This figure is a standard output from the ArcGis software (ESRI, 2002).

1754

When edaphic factors were allowed to enter the model, they performed significantly better than any climatic variable in terms of the reduction of explained deviance in the first step of the forward stepwise selection procedure. The pH variable was kept in the final model, representing a complex environmental gradient, which is easy to measure in the field, and is often used in studies assessing species–environment relationships (see e.g. Coudun & Ge´gout, 2005). Two further steps led to the inclusion of climatic variables, with the two variables from the climatic model being again the best ones in our forward stepwise selection procedure (Table 2b). Autumn precipitation (pAu) and actual Thornthwaite evapotranspiration (AETTh) were therefore kept in our final climatic and edaphic model, whose logistic equation is:

Journal of Biogeography 33, 1750–1763 ª 2006 The Authors. Journal compilation ª 2006 Blackwell Publishing Ltd

Distribution of Acer campestre in France

Figure 3 Maps of successive evaluation sets with increasing distance (D) between evaluation plots and between evaluation and calibration plots. The numbers of evaluation plots for distances equal to 0, 0.5, 1, 5, 10 and 20 km are 81,554, 54,879, 8359, 2278 and 374, respectively. 0.7 0.6 0.5 0.4 0.3

500 450 400

Figure 4 Climatic response surface of A. campestre (contour map of probabilities of presence), with regard to mean autumn precipitation (pAu) and mean annual actual Thornthwaite evapotranspiration (AETTh), derived with logistic regression modelling based on 3286 forest plots (see Table 2a for details about the climatic model). Real climatic conditions in which A. campestre is either observed present (filled squares) or absent (small points) are also represented.

AETTh (mm)

550

600

0.8

0.2 100

0.1 200

300

400

500

600

700

pAu (mm)

Table 2 Summary of the selected (a) climatic model and (b) climatic and edaphic model derived through forward stepwise logistic regression modelling between presence/absence of A. campestre in French forests and available ecological variables. Information is given for each step of the computation on the variable included in quadratic form or otherwise. D2 is calculated as (null deviance ) model deviance)/ null deviance. The AUC value is the area under the ROC curve. S, Sn and Sp. (all in %) are derived from the optimal confusion matrix and represent the success rate (rate of correct prediction of presence and absence), the sensitivity (rate of correct prediction of presence) and the specificity (rate of correct prediction of absence), respectively (Fielding & Bell, 1997) Null deviance ¼ 2661 Step

Added variable

Quadratic (Y/N)

(a) Climatic model 1 pAu N 2 AETTh N (b) Climatic and edaphic model 1 pH Y 2 pAu N 3 AETTh N

Calibration (n ¼ 3286) 2

Evaluation (n ¼ 88,004)

Model deviance

Deviance reduction

D

AUC

S

Sn

Sp

AUC

S

Sn

Sp

2543 2427

118 235

0.044 0.088

0.64 0.72

54 55

69 84

52 50

0.56 0.64

38 54

81 73

30 50

1933 1835 1744

728 826 917

0.273 0.310 0.344

0.85 0.88 0.89

76 73 81

84 92 85

75 70 81

0.83 0.81 0.85

71 66 73

88 89 86

67 62 70

Journal of Biogeography 33, 1750–1763 ª 2006 The Authors. Journal compilation ª 2006 Blackwell Publishing Ltd

1755

0.6 0.4 0.0

0.2

Probability of presence

0.8

1.0

C. Coudun et al.

4

5

6

6

8

pH

Figure 5 Ecological response curve of A. campestre, to soil pH (pH), derived from logistic regression modelling based on 3286 forest plots (see Table 2b for details about the climatic and edaphic model). Observed presence/absence data of the species along the pH gradient are also represented by crosses.

  p ¼ 30:47 þ 7:56pH  0:536pH2  0:01235pAu ln 1p þ 0:01378AETTh

ð2Þ

The calcicolous response of A. campestre was apparent (optimum pH values between 6.5 and 7.5), when looking at the ecological response curve with regard to pH (Fig. 5; see also Smart et al., 2004 for a more elegant but manual and timeconsuming way of presenting such results). The climatic response of A. campestre remained stable when pH was also incorporated as an explanatory variable. The predicted distribution map resulting from the climatic and edaphic model (Fig. 6d) still excludes A. campestre from the mountainous and Mediterranean areas (Fig. 6b: regions 1a, Vosges; 1b, Jura; 1c, Alps; 1d, Massif Central; 1e, Pyrenees; 2, Mediterranean region) with very low expected probabilities of presence. This model also excludes the species from acidic regions, where it cannot grow and survive (Fig. 5). As a consequence, the prediction of the absence of the species is much better, leading to a stronger value for specificity (see Table 2b). Acer campestre is predicted to be absent from primary crystalline acidic zones (Fig. 6b: region 3a, Bretagne) and from Tertiary/Quaternary sandy acidic zones (Fig. 6b: region 3b, Landes; region 3c, Sologne). At local scales, very low probabilities of presence are predicted for, for example, the Fontainebleau (Fig. 6b: region 3d) or Haguenau (Fig. 6b: region 3e) acidic forests. With high correctly predicted probabilities of presence in Nord, Lorraine, Alsace, Bourgogne, Charente and Gascogne (Fig. 6b: regions A, B, C, D, E and F, respectively), the model is powerful in predicting the presence of A. campestre in France. The high quality (Swets, 1988) of this model is supported by the high value of the area under the 1756

ROC curve (AUC ¼ 0.89 for the calibration data set, AUC ¼ 0.85 for the evaluation data set; see Table 2b). The predictive success of the climatic and the combined climatic and soil nutritional model did not vary much when evaluation plots were selected further from calibration plots (Table 3a). More than 98% of the evaluation plots are at least 500 m from a calibration plot. With a global success of S ¼ 71%, the model was still successful when all evaluation points were situated at least 20 km from EcoPlant calibration points (Table 3a). This corresponds geographically for example to the Charente region (region E in Fig. 6b) where no calibration point was available for model calibration (Fig. 1), but where the predicted map fits the observation data reasonably well. However, when evaluation plots were selected further from calibration plots and from themselves, a decrease in the predictive success was noticeable (Table 3b). Finally, the predictive success of the model when evaluation plots are randomly selected is comparable to the predictive success computed with the whole evaluation data set (Table 3c,d). DISCUSSION Methodological remarks Logistic regression is a robust technique for predicting the distribution of A. campestre in French forests, in accordance with previous comparative studies (Brotons et al., 2004; Segurado & Arau´jo, 2004). The number of occurrences of the species in our data sets was sufficient to ensure reliable characterization of species–environment relationships (Ch. Coudun & J.-C. Ge´gout, in review). Even if this statistical technique constrains the shape of the response curves along each gradient to be either

Journal of Biogeography 33, 1750–1763 ª 2006 The Authors. Journal compilation ª 2006 Blackwell Publishing Ltd

Distribution of Acer campestre in France

a

b A B

3a 3d

3e 1a

3c

C

D 1b

E 1c

1d 3b 2

F 1e

c

d

Figure 6 Maps of actual presence/absence of A. campestre and predicted probabilities of presence. (a) Location of the 88,004 forest sites from the French Forest National Inventory (evaluation data set), with indication of the presence (green points) and absence (red points) of A. campestre. (b) Schematic view of France with indication of some natural regions or forests. (c) Grey-scale map of predicted probabilities of presence with the climatic model (see also equation 1, Fig. 4 and Table 2a). (d) Grey-scale map of predicted probabilities of presence with the climatic and edaphic model (see also equation 2, Fig. 5 and Table 2b). Blank areas in Fig. 6a,b represent French departments for which floristic information is not yet available, and thus for which we could not compute any edaphic value based on species composition.

flat, monotonic or bell-shaped unimodal, fair values for the AUC were found in the calibration and evaluation data sets (Swets, 1988), with a number of variables included in the models being chosen selectively. The characterization of the ecological response of A. campestre was also robust, since its climatic response seemed to be stable when an edaphic variable was introduced into the model (equations 1 and 2; Table 2). The stability of the predictive success of the model when it was tested with an evaluation data set spaced over varying distances supports its robustness.

Species and community modelling techniques are currently being improved, especially in the forest context (see Maggini et al., 2006). To improve model quality further, more ‘releve´s’ would be needed in the French EcoPlant data base, especially in the Mediterranean region; and to improve the species distribution maps, the data base provided by the French National Forest Inventory (IFN), should be completed in order to fill the gaps in the areas where there are no available data (Fig. 6). In any case, this study uses two large national data bases of forest ‘releve´s’ (Ge´gout et al., 2005), which together

Journal of Biogeography 33, 1750–1763 ª 2006 The Authors. Journal compilation ª 2006 Blackwell Publishing Ltd

1757

C. Coudun et al. Table 3 Predictive success of the distribution model of A. campestre with climatic and edaphic factors, computed with different evaluation data sets obtained: (a) by increasing the minimum distance between evaluation and calibration plots; (b) by increasing the minimum distance between evaluation plots and between evaluation and calibration plots; (c) by randomly selecting evaluation plots. The predictive success assessed with the whole evaluation data set is provided in (d). Nplots is the number of selected evaluation plots, Npres in the number of times Acer campestre is present in the Nplots, AUC is the area under the ROC curve and S, Sn and Sp are the global success, sensitivity and specificity (all in %), respectively Nplots

Npres

AUC

S

Sn

Sp

(a) Minimum distance between evaluation and calibration plots (km) 0.5 86,825 14,251 0.84 73 86 71 1 84,373 13,956 0.84 73 86 71 5 65,278 11,270 0.83 71 86 68 10 47,756 7610 0.82 70 85 67 20 13,025 2220 0.82 71 88 67 (b) Minimum distance between evaluation plots and between evaluation and calibration plots (km) 0.5 81,554 13,630 0.84 73 86 1 54,879 9696 0.83 71 88 5 8359 1463 0.80 67 88 10 2278 393 0.79 69 83 20 374 65 0.82 66 92

71 67 62 67 61

et al., 1998; Schwarz et al., 2003). The dispersal distance of A. campestre was not investigated, but the low height of this species (up to 15 m; Rameau et al., 1989) should limit its dispersion, as for other maple species (Guries & Nordheim, 1984). Patchiness of presence/absence at a regional scale is probably due to spatial autocorrelation in the predictor environmental variables: similar ecological conditions are likely to be found in nearby sample plots. Increasing the distance between evaluation plots and between evaluation and calibration plots led to a progressive elimination of evaluation plots within large forest patches in France (large forested regions such as north-east and south-west France or the mountains) and the conservation of plots in small forest patches only (mainly agricultural regions such as north-west France). With large minimum distances between plots (10 and 20 km; Table 3b), the evaluation plots were mostly selected in small separated forest patches, overrepresenting small forests with regard to the total forested area in France. This might explain the observed decrease in the model predictive success (Table 3b). This is also confirmed by the rather similar predictive success computed with the whole evaluation data set or with randomly selected subsets (Table 3c). Indeed, the random selection ensures a probability of sampling an evaluation plot that is proportional to the area of the forest patch on which the plot is present. The importance of soil nutritional variables

(c) Random selection of evaluation plots – 81,554 13,339 – 54,879 8919 – 8359 1459 – 2278 392 – 374 73 (d) Complete evaluation data set

88,004

14,373

0.85 0.85 0.84 0.85 0.84

73 73 75 70 69

86 86 84 93 95

71 70 73 65 62

0.85

73

86

70

provide data of higher quality than is usually available for distribution modelling of plant species. For example, these data are more suitable than the data provided by standard distribution atlases since point locality data and information on soil variables associated within each point are used (M.P. Austin, in review). Spatial autocorrelation Spatial autocorrelation might bias the results of niche-based models (Diniz-Filho et al., 2003; Munoz & Felicisimo, 2004; P. Segurado, M.B. Arau´jo & W.E. Kunin, in review). Spatial autocorrelation, due to life history characteristics of species, does not influence the success of predictions probably because the distance between plots (greater than 1 km for more than 95% of the plots) is superior to the extent of local patchiness in the pattern of A. campestre distribution. Studies on local patterns of abundance of maple species have shown that maximum dispersal distance and recruitment limitations are often lower than 0.5 km (Guries & Nordheim, 1984; Clark 1758

Few studies have addressed the need to incorporate edaphic variables when modelling species distributions over large spatial extents and most recently published papers have modelled the distribution of plant species with climatic variables alone (e.g. Vetaas, 2002). Even if some studies integrate soil-related/terrain variables (topography, lithology, etc.), they do not use ‘direct’ (sensu Austin, 2002) measured variables (Le Duc et al., 1992; Firbank et al., 1998; Franklin, 1998, 2002; Franklin et al., 2000; Fertig & Reiners, 2002), which have a physiological impact on plant growth and which have been promoted many times (see M.P. Austin, in review). Contrary to the findings of Huntley et al. (2004), this study shows that the influence of numerous climatic factors on the distribution of A. campestre in France is weaker than the influence of soil pH. This illustrates the potential interest of incorporating soil nutritional variables into regional or national distribution models of plant species. Further research could, for example, investigate the relative contribution of climate and soil factors to the explained deviance of species distribution in French forests, using a similar procedure as N.E. Zimmermann G.G. Moisen, T.C. Edwards Jr, T.S. Frescino & J.A. Blackard (in review). The choice to focus on one species (A. campestre) in France was to illustrate that, when good quality data are available, edaphic factors could play an important role in the modelling of plant distribution. We studied the ecological response of A. campestre in a representative part of its geographical range in Europe, since the range of ecological conditions found in France is wide (Table 1). The preference of A. campestre for

Journal of Biogeography 33, 1750–1763 ª 2006 The Authors. Journal compilation ª 2006 Blackwell Publishing Ltd

Distribution of Acer campestre in France high values of actual evapotranspiration limits the range of conditions in which it can survive to areas where it is neither too cold (mountainous climate) nor too dry (Mediterranean climate). Its high probability of occurrence in areas with low autumn precipitation is consistent with the idea that the species has a lack of resistance to wet climates; for example the species is absent from Scotland and Ireland, which both have high annual rainfall. Predicting the distribution of the species with climatic variables alone gave a broad estimate of the geographical range of the species at the national scale, but the incorporation of soil pH (consistent with observations of ecologists: Rameau et al., 1989; Ellenberg et al., 1992) acted as a local filter allowing models to further discriminate predictions, especially absences, at a finer resolution. Climate and soil probably act at different scales and some hierarchical frameworks have been proposed to deal with this scale issue (Cherrill et al., 1995; Collingham et al., 2000; Lipsett-Moore et al., 2003; Thuiller et al., 2003a; Pearson et al., 2004). The national model of the distribution of A. campestre was, for example, recently compared to a regional model derived in the French Alps (Py, 2005), stressing again the interest of working at different scales. Indeed, accurate up-to-date predicted distribution maps, valid at national and/or local scales, are of great interest for species management and conservation (Fortin et al., 2005), and could be derived for other natural plant species in the same way as presented in this study. ACKNOWLEDGEMENTS The authors wish to thank the French National Forest Inventory (IFN) for the permission to use their data as an evaluation data set in this study, and Ingrid Seynave, who computed the edaphic variables for the IFN evaluation data set. The authors also wish to thank Paul Taylor and Mari Bullock from CTCD, for their careful correction of the English language, the ‘Riederalp 2004’ group for stimulating discussions, Miguel B. Arau´jo for his editorial work, as well as Michael P. Austin, Jose´ Alexandre Felizola Diniz-Filho and another anonymous referee for their appropriate comments on earlier drafts of the manuscript. This study was financed through grants to Christophe Coudun by the French National Forest Office (ONF) and the Lorraine Regional Council (CR Lorraine). EcoPlant is a phytoecological data base financed by the French Institute of Agricultural, Forest and Environmental Engineering (ENGREF), the French Ministry of Agriculture (DERF), the French National Forest Office (ONF) and the French Agency for Environment and Energy Management (ADEME). REFERENCES Arau´jo, M.B. & Pearson, R.G. (2005) Equilibrium of species’ distributions with climate. Ecography, 28, 693–695. Arau´jo, M.B. & Williams, P.H. (2000) Selecting areas for species persistence using occurrence data. Biological Conservation, 96, 331–345.

Arau´jo, M.B., Cabeza, M., Thuiller, W., Hannah, L. & Williams, P.H. (2004) Would climate change drive species out of reserves? An assessment of existing reserve selection methods. Global Change Biology, 10, 1618–1626. Arau´jo, M.B., Pearson, R.G., Thuiller, W. & Erhard, M. (2005) Validation of species-climate impact models under climate change. Global Change Biology, 11, 1504–1513. Arau´jo, M.B., Thuiller, W. & Pearson, R.G. (2006) Climate warming and the decline of amphibians and reptiles in Europe. Journal of Biogeography, doi: 10.1111/j.1365-2699. 2006.01482.x. Austin, M.P. (1971) Role of regression analysis in plant ecology. Proceedings of the Ecological Society of Australia, 6, 63–75. Austin, M.P. (1972) Models and analysis of descriptive vegetation data. Mathematical models in ecology (ed. by J.N.R. Jeffers), 61–86. Blackwell Scientific, Oxford. Austin, M.P. (2002) Spatial prediction of species distribution: an interface between ecological theory and statistical modelling. Ecological Modelling, 157, 101–118. Austin, M.P. & Meyers, J.A. (1996) Current approaches to modelling the environmental niche of eucalypts: implication for management of forest biodiversity. Forest Ecology and Management, 85, 95–106. Bakkenes, M., Alkemade, J.R.M., Ihle, F., Leemans, R. & Latour, J.B. (2002) Assessing the effects of forecasted climate change on the diversity and distribution of European higher plants for 2050. Global Change Biology, 8, 390–407. Bayliss, J.L., Simonite, V. & Thompson, S. (2005) The use of probabilistic habitat suitability models for biodiversity action planning. Agriculture, Ecosystems and Environment, 108, 228–250. Bendixen, K. (2001) Zum Reproduktionssystem des Feldahorns (Acer campestre L.) – Blu¨hpha¨nologie und genetische Untersuchungen. PhD Dissertation, Institut fu¨r Forstgenetik und Forstpflanzenzu¨chtung, Universita¨t Go¨ttingen. Benichou, P. & Le Breton, O. (1987) Prise en compte de la topographie pour la cartographie des champs pluviome´triques statistiques. La Me´te´orologie, 7, 23–34. Bio, A.M.F. (2000) Does vegetation suit our models? Data and model assumptions and the assessment of species distribution in space. PhD Dissertation, University of Utrecht, Faculty of Geographical Science. Bio, A.M.F., Alkemade, J.R.M. & Barendregt, A. (1998) Determining alternative models for vegetation response analysis: a non parametric approach. Journal of Vegetation Science, 9, 5–16. Box, E.O., Crumpacker, D.W. & Hardin, E.D. (1993) A climatic model for location of plant species in Florida, U.S.A. Journal of Biogeography, 20, 629–644. ter Braak, C.J.F. & Barendregt, L.G. (1986) Weighted averaging of species indicator values: its efficiency in environmental calibration. Mathematical Biosciences, 78, 57–72. ter Braak, C.J.F. & Looman, C.W.N. (1986) Weighted averaging, logistic regression and the Gaussian response model. Vegetatio, 65, 3–11.

Journal of Biogeography 33, 1750–1763 ª 2006 The Authors. Journal compilation ª 2006 Blackwell Publishing Ltd

1759

C. Coudun et al. Bragazza, L. & Gerdol, R. (1996) Response surfaces of plant species along water-table depth and pH gradients in a poor mire on the southern Alps (Italy). Annales Botanici Fennici, 33, 11–20. Brotons, L., Thuiller, W., Arau´jo, M.B. & Hirzel, A.H. (2004) Presence–absence versus presence-only habitat suitability models: the role of species ecology and prevalence. Ecography, 27, 437–448. Cabeza, M., Arau´jo, M.B., Wilson, R.J., Thomas, C.D., Cowley, M.J.R. & Moilanen, A. (2004) Combining probabilities of occurrence with spatial reserve design. Journal of Applied Ecology, 41, 252–262. Carpenter, G., Gillison, A.N. & Winter, J. (1993) DOMAIN: a flexible modelling procedure for mapping potential distribution of plants and animals. Biodiversity and Conservation, 2, 667–680. Cawsey, E.M., Austin, M.P. & Baker, B.L. (2002) Regional vegetation mapping in Australia: a case study in the practical use of statistical modelling. Biodiversity and Conservation, 11, 2239–2274. Cherrill, A.J., McClean, C., Watson, P., Tucker, K., Rushton, S.P. & Sanderson, R. (1995) Predicting the distributions of plant species at the regional scale: a hierarchical matrix model. Landscape Ecology, 10, 197–207. Clark, J.S., Macklin, E. & Wood, L. (1998) Stages and spatial scales of recruitment limitation in southern Appalachian forests. Ecological Monographs, 68, 213–235. Collingham, Y.C., Wadsworth, R.A., Huntley, B. & Hulme, P.E. (2000) Predicting the spatial distribution of nonindigenous riparian weeds: issues of spatial scale and extent. Journal of Applied Ecology, 37(Suppl. 1), 13–27. Coudun, Ch. (2005) Approache quantitative de la re´ponse e´cologique des espe`ces ve´ge´tales forestie`res a` l’e´chelle de la France. PhD Dissertation, French Institute of Forestry, Agricultural and Environmental Engineering, ENGREF, Nancy, France. Coudun, Ch. & Ge´gout, J.-C. (2005) Ecological behaviour of herbaceous forest species along a pH gradient: a comparison between oceanic and semi-continental regions in northern France. Global Ecology and Biogeography, 14, 263–270. Diekmann, M. (2003) Species indicator values as an important tool in applied plant ecology: a review. Basic and Applied Ecology, 4, 493–506. Diniz-Filho, J.A.F., Bini, L.M. & Hawkins, B.A. (2003) Spatial autocorrelation and red herrings in geographical ecology. Global Ecology and Biogeography, 12, 53–64. Dirnbo¨ck, T., Dullinger, S. & Grabherr, G. (2003) A regional impact assessment of climate and land use change on alpine vegetation. Journal of Biogeography, 30, 1–17. Elith, J. (2000) Quantitative methods for modeling species habitat: comparative performance and an application to Australian plants. Quantitative methods in conservation biology (ed. by S. Ferson and M.A. Burgman), pp. 39–58. Springer, New York. Elith, J. (2002) Predicting the distribution of plants. PhD Dissertation, University of Melbourne, Australia. 1760

Elith, J. & Burgman, M.A. (2002) Predictions and their validation: rare plants in the Central Highlands, Victoria, Australia. Predicting species occurrences: issues of accuracy and scale (ed. by J.M. Scott, P.J. Heglund, F. Samson, J. Haufler, M. Morrison, M. Raphael and B. Wall), pp. 303–313. Island Press, Covelo, CA. Elith, J., Burgman, M.A. & Regan, H.M. (2002) Mapping epistemic uncertainties and vague concepts in predictions of species distribution. Ecological Modelling, 157, 313–329. Ellenberg, H., Weber, H.E., Du¨ll, R., Wirth, V., Werner, W. & Paulißen, D. (1992) Zeigerwerte von Pflanzen in Mitteleuropa. Scripta Geobotanica, 18, 1–248. ESRI (2002) ArcGIS 8.2. Environmental Systems Research Institutes, Redlands, CA. Ferrier, S., Watson, G., Pearce, J. & Drielsma, M. (2002) Extended statistical approaches to modelling spatial pattern in biodiversity in northeast New South Wales. I. Species-level modelling. Biodiversity and Conservation, 11, 2275–2307. Fertig, W. & Reiners, W.A. (2002) Predicting presence/absence of plant species for range mapping: a case study from Wyoming. Predicting species occurrences: issues of accuracy and scale (ed. by J.M. Scott, P.J. Heglund, F. Samson, J. Haufler, M. Morrison, M. Raphael and B. Wall), pp. 483– 489. Island Press, Covelo, CA. Fielding, A.H. & Bell, J.F. (1997) A review of methods for the assessment of prediction errors in conservation presence/absence models. Environmental Conservation, 24, 38– 49. Firbank, L.G., Ellis, N.E., Hill, M.O., Lockwood, A.J. & Swetnam, R.D. (1998) Mapping the distribution of weeds in Great Britain in relation to national survey data and to soil type. Weed Research, 38, 1–10. Fortin, M.-J., Keitt, T.H., Maurer, B.A., Taper, M.L., Kaufman, D.M. & Blackburn, T.M. (2005) Species’ geographic ranges and distributional limits: pattern analysis and statistical issues. Oikos, 108, 7–17. Franklin, J. (1995) Predictive vegetation mapping: geographic modelling of biospatial patterns in relation to environmental gradients. Progress in Physical Geography, 19, 474–499. Franklin, J. (1998) Predicting the distribution of shrub species in southern California from climate and terrain-derived variables. Journal of Vegetation Science, 9, 733–748. Franklin, J. (2002) Enhancing a regional vegetation map with predictive models of dominant plant species in chaparral. Applied Vegetation Science, 5, 135–146. Franklin, J., McCullough, P. & Gray, C. (2000) Terrain variables used for predictive mapping of vegetation communities in Southern California. Terrain analysis: principles and applications (ed. by J.P. Wilson and J.C. Gallant), 331–353. Wiley, New York. Ge´gout, J.-C. (2001) Cre´ation d’une base de donne´es phytoe´cologiques pour de´terminer l’aute´cologie des espe`ces de la Flore Forestie`re de France. Revue Forestie`re Franc¸aise, 53, 397–403.

Journal of Biogeography 33, 1750–1763 ª 2006 The Authors. Journal compilation ª 2006 Blackwell Publishing Ltd

Distribution of Acer campestre in France Ge´gout, J.-C., Coudun, Ch., Brisse, H. & Berge`s, L. (2002) Comportement e´cologique des espe`ces forestie`res vis-a`-vis du climat et du sol en France: application a` l’e´valuation des charges critiques d’acidite´ et d’azote. French Institute of Forestry, Agricultural and Environmental Engineering, ENGREF, Nancy, France. Ge´gout, J.-C., Herve´, J.-C., Houllier, F. & Pierrat, J.-C. (2003) Prediction of forest soil nutrient status using vegetation. Journal of Vegetation Science, 14, 55–62. Ge´gout, J.-C., Coudun, Ch., Bailly, G. & Jabiol, B. (2005) EcoPlant: a forest sites database to link floristic data with soil resources and climatic conditions. Journal of Vegetation Science, 16, 257–260. Gelfand, A.E., Silander, J.A., Jr, Wu, S., Latimer, A., Lewis, P.O., Rebelo, A.G. & Holder, M. (2003) Explaining species distribution patterns through hierarchical modelling. Bayesian Analysis, 1, 1–47. Gignac, L.D. (1992) Niche structure, resource partitioning and species interactions of mire bryophytes relative to climatic and ecological gradients in western Canada. The Bryologist, 95, 406–418. Gignac, L.D., Vitt, D.H. & Bayley, S.E. (1991) Bryophyte response surfaces along ecologic and climatic gradients. Vegetatio, 93, 29–45. Guisan, A. & Zimmermann, N.E. (2000) Predictive habitat distribution models in ecology. Ecological Modelling, 135, 147–186. Guisan, A., Theurillat, J.-P. & Kienast, F. (1998) Predicting the potential distribution of plant species in an alpine environment. Journal of Vegetation Science, 9, 65–74. Guisan, A., Weiss, S.B. & Weiss, A.D. (1999) GLM versus CCA spatial modeling of plant species distribution. Plant Ecology, 143, 107–122. Guisan, A., Edwards, T.C., Jr & Hastie, T.J. (2002) Generalized linear and generalized additive models in studies of species distributions: setting the scene. Ecological Modelling, 157, 89–100. Guries, R.P. & Nordheim, E.V. (1984) Flight characteristics and dispersal potential of Maple samaras. Forest Science, 30, 434–440. Hampe, A. (2004) Bioclimate envelope models: what they detect and what they hide. Global Ecology and Biogeography, 13, 469–471. Hastie, T.J. & Tibshirani, R. (1997) Generalized additive models. Chapman and Hall, London. Heegaard, E. (1997) Ecology of Andreaea in western Norway. Journal of Bryology, 19, 527–636. Heegaard, E. (2001) Environmental relationships of perichaetial and sporophyte production in Andreaea spp. in Western Norway. Journal of Bryology, 23, 97–108. Heegaard, E. (2002) A model of alpine species distribution in relation to snowmelt time and altitude. Journal of Vegetation Science, 13, 493–504. Hooten, M.B., Larsen, D.R. & Wikle, C.K. (2003) Predicting the spatial distribution of ground flora on large domains

using a hierarchical Bayesian model. Landscape Ecology, 18, 487–502. Huisman, J., Olff, H. & Fresco, L.F.M. (1993) A hierarchical set of models for species response analysis. Journal of Vegetation Science, 4, 37–46. Huntley, B., Green, R.E., Collingham, Y.C., Hill, J.K., Willis, S.G., Bartlein, P.J., Cramer, W., Hagemeijer, W.J.M. & Thomas, C.J. (2004) The performance of models relating species geographical distributions to climate is independent of trophic level. Ecology Letters, 7, 417–426. Isaaks, E.H. & Srivastava, R.M. (1989) An introduction to applied geostatistics. Oxford University Press, New York. Iverson, L.R., Prasad, A.M. & Schwartz, M.W. (1999) Modeling potential future individual tree-species distributions in the eastern United States under a climate change scenario: a case study with Pinus virginiana. Ecological Modelling, 115, 77–93. Journel, A.G. (1983) Nonparametric estimation of spatial distributions. Mathematical Geology, 15, 445–468. Lawesson, J.E. & Oksanen, J. (2002) Niche characteristics of Danish woody species as derived from coenoclines. Journal of Vegetation Science, 13, 279–290. Le Duc, M.G., Hill, M.O. & Sparks, T.H. (1992) A method for predicting the probability of species occurrence using data from systematic surveys. Watsonia, 19, 97–105. Leathwick, J.R. (1998) Are New-Zealand’s Nothofagus species in equilibrium with their environment? Journal of Vegetation Science, 9, 719–732. Leathwick, J.R. & Rogers, G.M. (1996) Modelling relationships between environment and canopy composition in secondary vegetation in Central North Island, New Zealand. New Zealand Journal of Ecology, 20, 147–161. Leathwick, J.R. & Whitehead, D. (2001) Soil and atmospheric water deficits and the distribution of New Zealand’s indigenous tree species. Functional Ecology, 15, 233–242. Legendre, P. (1993) Spatial autocorrelation: trouble or new paradigm? Ecology, 74, 1659–1673. Lehmann, A., Leathwick, J.R. & Overton, J.McC. (2002a) Assessing New Zealand fern diversity from spatial predictions of species assemblages. Biodiversity and Conservation, 11, 2217–2238. Lehmann, A., Overton, J.McC. & Austin, M.P. (2002b) Regression models for spatial prediction: their role for biodiversity and conservation. Biodiversity and Conservation, 11, 2085–2092. Lehmann, A., Overton, J.McC. & Leathwick, J.R. (2002c) GRASP: generalized regression analysis and spatial prediction. Ecological Modelling, 157, 189–207. Lenihan, J.M. (1993) Ecological response surfaces for North American boreal tree species and their use in forest classification. Journal of Vegetation Science, 4, 667–680. Lennon, J.J. (2000) Red-shifts and red herrings in geographical ecology. Ecography, 23, 101–113. Lipsett-Moore, G., McKenney, D.W. & Jones, S. (2003) Multiscale species modelling in Ontario: a workshop on needs and opportunities. The Forestry Chronicle, 79, 147–148.

Journal of Biogeography 33, 1750–1763 ª 2006 The Authors. Journal compilation ª 2006 Blackwell Publishing Ltd

1761

C. Coudun et al. Luoto, L., Heikkinen, R.K., Po¨yry, J. & Saarinen, K. (2006) Determinants of biogeographical distribution of butterflies in boreal regions. Journal of Biogeography, doi: 10.1111/ j.1365-2699.2005.01395.x. Maggini, R., Lehmann, A., Zimmermann, N.E. & Guisan, A. (2006) Improving generalized regression analysis for spatial prediction of forest communities. Journal of Biogeography, doi: 10.1111/j.1365-2699.2006.01465.x. Margules, C.R. & Stein, J.L. (1989) Patterns in the distributions of species and selection of nature reserves: an example from Eucalyptus forests in southeastern New South Wales. Biological Conservation, 50, 219–238. Marinoni, O. (2003) Improving geological models using a combined ordinary-indicator kriging approach. Engineering Geology, 69, 37–45. MathSoft, Inc. (1999) S-Plus 2000: programmer’s guide. MathSoft, Inc., Seattle, WA. McCullagh, P. & Nelder, J.A. (1997) Generalized linear models, 2nd edn. Chapman & Hall, London. McKenzie, D. & Halpern, C.B. (1999) Modeling the distributions of shrub species in Pacific northwest forests. Forest Ecology and Management, 114, 293–307. McKenzie, D., Peterson, D.W. & Peterson, D.L. (2003a) Modelling conifer species distributions in mountain forests of Washington State, USA. The Forestry Chronicle, 79, 253–258. McKenzie, D., Peterson, D.W., Peterson, D.L. & Thornton, P.E. (2003b) Climatic and biophysical controls on conifer species distributions in mountain forests of Washington State, USA. Journal of Biogeography, 30, 1093–1108. McPherson, J.M., Jetz, W. & Rogers, D.J. (2004) The effects of species’ range sizes on the accuracy of distribution models: ecological phenomenon or statistical artefact? Journal of Applied Ecology, 41, 811–823. Miller, J. & Franklin, J. (2002) Modeling the distribution of four vegetation alliances using generalized linear models and classification trees with spatial dependence. Ecological Modelling, 157, 227–247. Mills, E. (1996) An appreciation and natural history of the English field maple (Acer campestre L.). Arboricultural Journal, 20, 405–410. Motel, G. (1995) L’e´rable. Actes Sud, Paris. Mueller-Dombois, D. & Ellenberg, H. (1974) Aims and methods of vegetation ecology. Wiley, New York. Munoz, J. & Felicisimo, A.M. (2004) Comparison of statistical methods commonly used in predictive modelling. Journal of Vegetation Science, 15, 285–292. Odland, A., Birks, H.J.B. & Line, J.M. (1995) Ecological optima and tolerances of Thelypteris limbosperma, Athyrium distentifolium, and Matteuccia struthiopteris along environmental gradients in Western Norway. Vegetatio, 120, 115–129. Pearson, R.G., Dawson, T.P., Berry, P.M. & Harrison, P.A. (2002) SPECIES: a spatial evaluation of climate impact on the envelope of species. Ecological Modelling, 154, 289–300. Pearson, R.G., Dawson, T.P. & Liu, C. (2004) Modelling species distributions in Britain: a hierarchical integration of climate and land-cover data. Ecography, 27, 285–298. 1762

Peterson, A.T. (2003) Predicting the geography of species’ invasions via ecological niche modeling. The Quarterly Review of Biology, 78, 419–433. Peterson, A.T., Papes, M. & Kluza, D.A. (2003) Predicting the potential invasive distributions of four alien plant species in North America. Weed Science, 51, 863–868. Philip, G.M. & Watson, D.F. (1982) A precise method for determining contoured surfaces. Australian Petroleum Exploration Association Journal, 22, 205–212. Pinto, P. & Ge´gout, J.-C. (2005) Assessing the nutritional and climatic response of temperate forest tree species in the Vosges Mountains. Annals of Forest Science, 62, 1–10. Polasky, S., Camm, J.D., Solow, A.R., Csuti, B., White, D. & Ding, R. (2000) Choosing reserve networks with incomplete species information. Biological Conservation, 94, 1–10. Py, N. (2005) Cartographie pre´dictive de l’aire de re´partition et de la fertilite´ de quelques essences forestie`res. Application a` l’e´rable champeˆtre (Acer campestre), au heˆtre (Fagus sylvatica) et a` l’e´pice´a (Picea abies) dans les Alpes franc¸aises. French Institute of Forestry, Agricultural and Environmental Engineering, ENGREF, Nancy, France. Rameau, J.-C., Mansion, D., Dume´, G., Timbal, J., Lecointe, A., Dupont, P. & Keller, R. (1989) Flore forestie`re franc¸aise. Guide ´ecologique illustre´, 1, Plaines et Collines. Institut pour le De´veloppement Forestier, Paris. Randin, C.F., Dirnbo¨ck, T., Dullinger, S., Zimmermann, N.E., Zappa, M. & Guisan, A. (2006) Are niche-based species distribution models transferable in space? Journal of Biogeography, doi: 10.1111/j.1365-2699.2006.01466.x. Robertson, M.P., Villet, M.H. & Palmer, A.R. (2004) A fuzzy classification technique for predicting species’ distributions: applications using invasive alien plants and indigenous insects. Diversity and Distributions, 10, 461–474. Rushton, S.P., Ormerod, S.J. & Kerby, G. (2004) New paradigms for modelling species distributions? Journal of Applied Ecology, 41, 193–200. Schwartz, M.W., Iverson, L.R. & Prasad, A.M. (2001) Predicting the potential future distribution of four tree species in Ohio using current habitat availability and climatic forcing. Ecosystems, 4, 568–581. Schwarz, P.A., Fahey, T.J. & McCulloch, C.E. (2003) Factors controlling spatial variation of tree species abundance in a forested landscape. Ecology, 84, 1862–1878. Scott, J.M., Heglund, P.J., Morrison, M.L., Haufler, J.B., Raphael, M.G., Wall, W.A. & Samson, F.B. (2002) Predicting species occurrences: issues of accuracy and scale. Island Press, Covelo, CA. Segurado, P. & Arau´jo, M.B. (2004) An evaluation of methods for modelling species distributions. Journal of Biogeography, 31, 1–14. Shao, G. & Halpin, P.N. (1995) Climatic controls of eastern North American coastal tree and shrub distributions. Journal of Biogeography, 22, 1083–1089. Skov, F. & Svenning, J.-C. (2004) Potential impact of climatic change on the distribution of forest herbs in Europe. Ecography, 27, 366–380.

Journal of Biogeography 33, 1750–1763 ª 2006 The Authors. Journal compilation ª 2006 Blackwell Publishing Ltd

Distribution of Acer campestre in France Smart, J., Sutherland, W.J., Watkinson, A.R. & Gill, J.A. (2004) A new means of presenting the results of logistic regression. Bulletin of the Ecological Society of America, 85, 100–102. Swets, J.A. (1988) Measuring the accuracy of diagnostic systems. Science, 240, 1285–1293. Theurillat, J.-P. & Guisan, A. (2001) Potential impact of climate change on vegetation in the European Alps: a review. Climatic Change, 50, 77–109. Thornthwaite, C.W. & Mather, J.R. (1957) Instructions and tables for computing potential evapotranspiration and the water balance. Publications in Climatology, 10, 183–311. Thuiller, W. (2003) BIOMOD – optimizing predictions of species distributions and projecting potential future shifts under global change. Global Change Biology, 9, 1353–1362. Thuiller, W., Arau´jo, M.B. & Lavorel, S. (2003a) Generalized models vs. classification tree analysis: predicting spatial distributions of plant species at different scales. Journal of Vegetation Science, 14, 669–680. Thuiller, W., Vayreda, J., Pino, J., Sabate, S., Lavorel, S. & Gracia, C. (2003b) Large-scale environmental correlates of forest tree distributions in Catalonia (NE Spain). Global Ecology and Biogeography, 12, 313–325. Thuiller, W., Arau´jo, M.B. & Lavorel, S. (2004) Do we need land-cover data to model species distributions in Europe? Journal of Biogeography, 31, 353–361. Turc, L. (1961) Evaluation des besoins en eau d’irrigation et e´vaporation potentielle. Annales Agronomiques, 12, 13–49. Tutin, T.G., Heywood, V.H., Burges, N.A., Valentine, D.H., Walters, S.M. & Webb, D.A. (2001) Flora Europaea, Volumes 1–5. Cambridge University Press, Cambridge. Vayssie`res, M.P., Plant, R.E. & Allen-Diaz, B.H. (2000) Classification trees: an alternative non-parametric approach for predicting species distributions. Journal of Vegetation Science, 11, 679–694. Vetaas, O.R. (2002): Realized and potential climate niches: a comparison of four Rhododendron tree species. Journal of Biogeography, 29, 545–554. Williams, P.H., Hannah, L., Andelman, S., Midgley, G.F., Arau´jo, M.B., Hughes, G., Manne, L.L., Martinez-Meyer, E. & Pearson, R.G. (2005) Planning for climate change: identifying minimum-dispersal corridors for the Cape Proteaceae. Conservation Biology, 19, 1063–1074.

Yee, T.W. & Mitchell, N.D. (1991) Generalized additive models in plant ecology. Journal of Vegetation Science, 2, 587–602. Zimmermann, N.E. & Kienast, F. (1999) Predictive mapping of alpine grasslands in Switzerland: species versus community approach. Journal of Vegetation Science, 10, 469–482.

BIOSKETCHES Christophe Coudun, PhD and Environmental Engineer, is interested in modelling species–environment relationships. He contributes to investigating the quantitative response of forest plant species to edaphic and climatic factors in France. He is currently working on carbon-based models of forest growth for the Centre for Terrestrial Carbon Dynamics (CTCD) in the UK. Jean-Claude Ge´gout, PhD, is an ecologist whose research interests include forest plant ecology and sociology as well as geographical information systems (GIS). He is the main author of the French forest phytoecological data base EcoPlant. Christian Piedallu is a GIS engineer whose research interests include derivation of high-resolution ecological variables and indices and their testing in plant species predictive distribution models. Jean-Claude Rameau (1943–2005), Professor in Forest Ecology, was an international expert in botany, plant ecology, succession dynamics, habitat conservation and land management. He was the main author of many reports and books, in particular the two volumes of the Flore Forestie`re Franc¸aise.

Editor: Miguel Arau´jo This paper is part of the Special Issue, Species distribution modelling: methods, challenges and applications, which owes its origins to the workshop on Generalized Regression Analyses and Spatial Predictions: Grasping Ecological Patterns from Species to Landscape held at the Centre Pro Natura Aletsch in Riederalp, Switzerland, in August 2004.

Journal of Biogeography 33, 1750–1763 ª 2006 The Authors. Journal compilation ª 2006 Blackwell Publishing Ltd

1763