Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Postglacial dispersal limitation of widespread forest plant species in nemoral Europe

Postglacial dispersal limitation of widespread forest plant species in nemoral Europe Understanding the factors controlling species ranges is a central issue in evolution, ecology, and conservation biology. Climate is often singled out as the primary factor limiting the geographical range of species, in particular at large geographical scales ( Pearson and Dawson 2003 ). Climate may limit species distributions by direct physiological effects such as death of individuals due to frost damage or reproductive failure, indirect effects on biotic interactions ( Case et al. 2005 ), or even more subtly by affecting the balance between population extinction and recolonization ( Holt et al. 2005 ). Other environmental factors are generally thought to affect species distributions primarily at landscape and local scales ( Pearson and Dawson 2003 ). For plants, soil conditions are generally seen as being of predominant importance at these smaller scales ( Ellenberg 1988 ). More controversially it has been suggested that dispersal may also provide an important range constraint. In the absence of insuperable geographic barriers such as oceans and mountain chains, it has been argued that even plants are relatively mobile and via migration have rapidly tracked climate changes, with negligible dispersal constraints ( Pitelka et al. 1997 , Clark et al. 1998 ). However, even in the absence of strong barriers, dispersal may limit species ranges (for a theoretical perspective, see Holt et al. 2005 ). Notably, there is increasing evidence that many tree species are still expanding from their ice age refuges and have strongly dispersal‐limited ranges (the postglacial migrational lag hypothesis). Recent estimates of tree migration rates <100 m yr −1 imply that many species have probably not reached equilibrium with the modern environment (notably climate and soil) in Europe ( Svenning and Skov 2007b ). Equilibrium refers to the situation where a species’ range is completely determined by the present environmental conditions; the species occurs in all suitable areas and is absent from all unsuitable areas ( Araújo and Pearson 2005 ). Furthermore, bioclimatic modeling, and observations of naturalization or population dynamics suggest that many temperate tree species are still expanding from their ice age refuges and have strongly dispersal‐limited ranges in Europe ( Svenning and Skov 2004, 2007b ) and elsewhere ( Johnstone and Chapin 2003 ). Similar evidence is available for European forest herbs ( Skov and Svenning 2004 ), with a recently reported 45‐yr transplant experiment clearly showing that the forest herb Hyacinthoides non‐scripta is dispersal‐limited at its northern range edge in western Europe ( Van der Veken et al. 2007 ). Richness of tree species with small ranges in Europe is concentrated in areas that had a relatively warm climate during the Last Glacial Maximum (LGM; Svenning and Skov 2007a ). In contrast, modern climate appear to be a much stronger determinant of the richness of all or just the widespread tree species in Europe than LGM climate ( Svenning and Skov 2007a ), although there are also strong non‐environmental broad‐scale spatial trends in the tree species richness pattern across Europe ( Svenning and Skov 2005 ). Such spatial patterns provide at best indirect evidence for postglacial migrational lag, but a simple measure of geographical accessibility to postglacial recolonization from glacial refuges explains 78% of the geographical variation in the region's tree diversity and was a much stronger diversity predictor than three key bioclimatic variables ( Svenning and Skov 2007a ). The importance of dispersal as a limiting factor for plant species distributions at regional and smaller scales, in the absence of strong geographic barriers, is also increasingly noted ( Tuomisto et al. 2003 , Normand et al. 2006 ). In Europe, climate is a weaker predictor of species composition of amphibians and reptiles than of plants, suggesting that these groups are even less in equilibrium with climate ( Araújo and Pearson 2005 , Araújo et al. 2008 ). Although there is increasing evidence for the importance of postglacial migrational lag as a large‐scale range constraint for certain groups of organisms in certain regions, notably for trees, amphibians, and reptiles in Europe ( Svenning and Skov 2004, 2007a,b , Araújo and Pearson 2005 ), many studies still conclude that large‐scale species ranges are largely in equilibrium with climate ( Pearson and Dawson 2003 , Tinner and Lotter 2006 ), especially as regards the more widespread species (cf. Svenning and Skov 2007a ). The postglacial migrational lag hypothesis is of strong applied importance. Predictive species distribution modeling is of crucial importance to conservation management and planning as well as for predicting the impact of climatic and other environmental changes ( Guisan et al. 2006 ), but is often based on the assumption that species are at least close to equilibrium with climate or, more generally, the present environment (the equilibrium postulate; Guisan and Zimmerman 2000 , Guisan and Thuiller 2005 ). Furthermore, modeling of the potential impact of near‐future climate change on species diversity have consistently predicted much greater losses of species when assuming negligible migration than if unlimited migration was assumed ( Skov and Svenning 2004 , Thuiller et al. 2005 ). In the present paper we use logistic regression modeling to test the importance of postglacial migrational lag and the two main competing hypothesized range constraints (climate and soil) of widespread northern‐nemoral forest plant species across a large geographical region, namely north‐central Europe. We used an information‐theoretic model selection and multi‐model inference approach ( Burnham and Anderson 2002 , Johnson and Omland 2004 ) as well as variation partitioning ( Legendre and Legendre 1998 ) to assess the relative support for each of them. The information‐theoretic model selection approach is specifically designed to assess the relative levels of support for a set of competing hypotheses and to allow inferences to be drawn from the whole set of competing models, e.g. model averaging can be used to make robust parameter estimates when several models have similar support ( Burnham and Anderson 2002 , Johnson and Omland 2004 ). We chose to study widespread forest species in order to investigate the range‐limiting factors for species that have experienced strong postglacial range expansion and are associated with the habitat (forest) that was most widely available in the region during the early and middle part of the Holocene, i.e. the species most likely to be in equilibrium with climate and soil (as discussed above). In our analyses, postglacial migrational lag was modeled using a measure of accessibility to postglacial recolonization from ice age refugia. We investigated two specific predictions of the postglacial migrational lag hypothesis: 1) accessibility to postglacial recolonization has strong relative support as a predictor of species ranges, and 2) species prevalence are higher in areas closest to the greatest number of ice age refugia, i.e. the species prevalence‐accessibility relationship is positive. We note that not all species are likely to be equally affected by postglacial migrational lags: one might expect postglacial migrational lag to differ among forest plant growth forms, being less important for ferns than for trees or herbs, as many fern species have effective long distance dispersal due to their lightweight spores with the potential to produce hermaphroditic gametophytes ( Barrington 1993 , Pausas and Sáez 2000 ). Furthermore, cold‐hardy boreal species probably had relatively northern refuge locations in central and eastern Europe and recolonized northern Europe from here ( Stewart and Lister 2001 , Palmé et al. 2003 , Willis and van Andel 2004 , Cheddadi et al. 2006 , Maliouchenko et al. 2007 ) and are therefore less likely to still experience postglacial migrational lag than truly nemoral species. Materials and methods Study area and species The study area was western and north central Europe between 47.0–60.0°N and west of 24.0°E ( Fig. 1 ). This study region roughly coincides with the nemoral biome, excluding its easternmost part, where species occurrence data are less complete. We refer to this area as nemoral Europe. 1 The study region (nemoral Europe; 881 grid cells of ca 50×50 km) and the number of northern nemoral forest plant study species (n=47) in each grid cell. The study species were northern nemoral forest plant species, operationally defined as tree and forest‐associated herb and fern (including fern allies) species native to Denmark (positioned in the middle of the northern margin of the nemoral zone). Hereby we ensured that the selected species had all experienced strong postglacial range expansion in the region. Nomenclature follows Jalas and Suominen (1970–1994). Trees were defined as self‐supporting woody species reaching at least 20 m in height, in accordance with Svenning and Skov (2004 , 2005). Forest‐associated herbs and ferns were those species whose habitat description for Denmark primarily includes forest and scrub habitats in at least one of two standard floras ( Hansen 1984 , Mossberg et al. 1994 ). Of the 49 qualifying species, the fern Dryopteris assimilis and the herb Stellaria neglecta were excluded due to inadequate recording (Jalas and Suominen 1970–1994), leaving 47 study species ( Fig. 1 ). The three growth forms, trees, ferns, and herbs, were represented by 13, 13, and 21 species, respectively. Furthermore, 12 species are widespread in the boreal zone of Eurasia, while the remaining species have distributions that are more restricted to the nemoral zone. Species distribution data were obtained from the Atlas Florae Europaeae (AFE), which maps distributions using an equal‐area grid with cells of ca 50×50 km based on the Universal Transverse Mercator projection and the Military Grid Reference System (Jalas and Suominen 1970–1994). Only the present native ranges were used in the analyses. After excluding two AFE cells due to missing soil data, the study area included 881 cells. Explanatory variables Climate and soil maps were available in, or resampled to, a resolution of 10′. Climate data for the period 1961–1990 were provided by the Hadley Centre for Climate Prediction and Research (< www.meto.gov.uk/research/hadleycentre >). Three climatic variables of well established importance for temperate plant distributions ( Sykes et al. 1996 ) were derived from mean monthly temperature and precipitation values: 1) growing degree days (for a 5°C base temperature, GDD5), 2) absolute minimum temperature (TMIN), 3) and water balance (WATB), computed as the annual sum of the monthly differences between precipitation and potential evapotranspiration ( Skov and Svenning 2004 ). To represent the main edaphic factors affecting plant distributions in the study area ( Ellenberg 1988 ), four topsoil variables were extracted from “The digital soil map of the world” ( FAO 2003 ): pH, base saturation, sand content (%), and CaCO 3 concentration (%). The climate and soil variables were related to the species data by computing the mean of each variable within each AFE cell. However, CaCO 3 concentration and base saturation were excluded from the analyses due to their high correlation with pH (Spearman r=0.812 and 0.929, respectively). The five remaining environmental variables were transformed to have skewness<∣1.0∣, if necessary. Hence, water balance was transformed by subtracting its original minimum value to have a minimum of 0 and cubic‐root transformed. To allow for nonlinear responses, quadratic terms were generated for all environmental variables. We assessed multicollinearity of the final 10 environmental parameters by computing their mutual tolerance values ( Quinn and Keough 2002 ). Tolerance was 0.30 for water balance, 0.33 for growing‐degree‐days, and 0.60–0.85 for the remaining eight variables. As tolerances were>0.10, multicollinearity among the environmental parameters was unlikely to be problematic ( Quinn and Keough 2002 ). Interactions between the environmental variables were not modeled. Postglacial migrational lag was represented by a measure of accessibility to postglacial recolonization from ice age refugia. If the range of a species is constrained by postglacial migrational lag, we expect the species to be more prevalent in areas closest to the greatest number of ice age refugia. Following Svenning and Skov (2007b) , accessibility (ACC) was computed as: the inverse of the summed distances (in km) to all grid cells in the source area (subsequently, multiplied by 10 6 to avoid very small numbers). Hence, the more distant a receiving grid cell on average is located from any source cell the lower its accessibility ( Fig. 2 ). The source area was set to be southern Europe at 43–46°N. Most nemoral species were probably restricted to areas south of 46°N during the LGM ( Petit et al. 2002 ; also cf. Magri et al. 2006 ). Furthermore, more southern refugia have not played a strong role as source areas for postglacial recolonization of central and northern Europe ( Petit et al. 2002 , Magri et al. 2006 ). The measure of accessibility used here is a strong predictor of overall tree species richness in the study area ( Svenning and Skov 2007b ). We note that it is a general measure that is not tailored to species‐specific refugia locations; notably, species that are widespread in the boreal zone of Eurasia may have recolonized northern Europe from refugia located further north and east ( Stewart and Lister 2001 , Palmé et al. 2003 , Willis and van Andel 2004 , Cheddadi et al. 2006 , Maliouchenko et al. 2007 ; Table 2 ). 2 Accessibility to postglacial recolonization from ice age refugia (ACC) across nemoral Europe. Accessibility was computed as the inverse of the summed distances to all grid cells in the source area (range 1.51–4.08 10 −6 ×km −1 ; the warmer the color, the higher ACC; the scale gives 10 equal intervals). The source area was set to be southern Europe at 43–46°N. 2 Results of logistic regression model selection for 35 forest plant species in nemoral Europe. The number of AFE cell occurrences is shown (n, total number of cells=881). The most strongly supported regression model M i (M best : only i is given; see Table 1 ), its Akaike weight (w best in %), and the summed Akaike weights for models including accessibility (w A ) or soil (w S ) are shown. The summed Akaike weights for models including climate (w C ) was always 100.0%. Odds ratios are given for accessibility (OR A ) and the two strongest climate/soil variables (OR C+S ) based on their model‐averaged regression coefficients; the superscript indicates their rank in terms of the absolute size of the coefficients, while bold face indicates species where OR A ≥4.0. B Widespread in the boreal zone of Eurasia. Growth form Species n M best w best w A w S OR A OR C+S Trees Pinus sylvestris B 450 ACSnl 94.1 94.1 100.0 1.84 8 0.03 1 (TMIN), 0.22 2 (GDD5) Carpinus betulus 548 ACSnl 100.0 100.0 100.0 30.03 1 0.27 2 (WATB 2 ), 3.41 3 (GDD5) Fagus sylvatica 575 ACSnl 100.0 100.0 100.0 24.52 1 0.05 2 (WATB 2 ), 5.84 3 (WATB) Quercus petraea 703 ACSnl 100.0 100.0 100.0 2.58 2 2.62 1 (TMIN), 0.43 3 (pH) Taxus baccata 327 ACSnl 99.5 99.5 100.0 1.61 3 0.60 1 (GDD5), 0.61 2 (TMIN) Ulmus glabra 735 ACSnl 95.4 100.0 95.4 3.44 2 0.16 1 (GDD5), 2.43 3 (WATB 2 ) Ulmus laevis 335 ACSnl 93.0 97.9 94.8 1.82 5 0.21 1 (TMIN), 4.11 2 (GDD5) Ulmus minor 472 ACS 56.8 100.0 85.3 3.04 2 13.02 1 (GDD5), 0.35 3 (TMIN) Ferns Asplenium adiantum‐nigrum 380 ACSnl 98.3 98.3 100.0 1.66 4 10.96 1 (TMIN), 2.73 2 (WATB) Blechnum spicant 675 ACSnl 90.5 99.3 100.0 2.11 4 9.12 1 (TMIN), 0.37 2 (GDD5) Dryopteris dilatata 657 ACSnl 99.3 100.0 99.3 2.02 4 3.91 1 (TMIN), 0.30 2 (GDD5) Equisetum sylvaticum B 713 CSnl 59.8 36.8 94.9 1.10 9 0.47 1 (GDD5 2 ), 0.47 2 (GDD5) Gymnocarpium dryopteris B 639 ACSnl 98.6 98.6 100.0 2.54 3 0.25 1 (GDD5), 0.37 2 (TMIN) Lycopodium annotinum B 493 ACSnl 73.3 74.2 97.7 1.43 7 0.07 1 (TMIN), 0.11 2 (GDD5) Matteuccia struthiopteris B 238 ACSnl 92.9 92.9 100.0 0.69 6 0.05 1 (TMIN), 0.22 2 (TMIN 2 ) Polystichum aculeatum 481 ACSnl 100.0 100.0 100.0 7.15 1 3.38 2 (TMIN), 0.31 3 (GDD5) Thelypteris limbosperma 478 ACSnl 99.5 100.0 100.0 4.97 1 0.26 2 (GDD5), 3.62 3 (TMIN) Thelypteris phegopteris B 596 ACSnl 66.2 66.3 99.7 1.36 7 0.32 1 (TMIN), 3.09 2 (WATB) Herbs Actaea spicata 558 ACSnl 100.0 100.0 100.0 6.71 2 0.13 1 (TMIN), 0.24 3 (GDD5) Anemone ranunculoides 512 ACSnl 100.0 100.0 100.0 7.81 1 0.13 2 (TMIN), 0.24 3 (TMIN 2 ) Cardamine bulbifera 396 ACSnl 54.9 59.6 95.1 1.16 9 0.13 2 (TMIN), 0.28 3 (TMIN 2 ) Cardamine flexuosa 638 ACSnl 99.6 100.0 99.6 3.71 3 5.09 1 (TMIN), 0.21 2 (GDD5) Cardamine impatiens 484 ACSnl 100.0 100.0 100.0 3.87 1 0.52 2 (pH 2 ), 0.52 3 (WATB 2 ) Corydalis cava 402 ACSnl 100.0 100.0 100.0 32.79 1 0.09 2 (TMIN), 0.20 3 (TMIN 2 ) Corydalis intermedia 349 ACSnl 67.8 73.1 94.4 1.25 8 0.11 1 (TMIN 2 ), 0.13 2 (TMIN) Corydalis pumila 108 ACSnl 85.0 85.0 100.0 0.60 8 0.04 1 (TMIN 2 ), 0.06 2 (TMIN) Fallopia dumetorum 578 ACSnl 82.2 82.7 99.4 6.44 1 0.17 2 (TMIN), 0.27 3 (WATB 2 ) Hepatica nobilis 498 ACSnl 78.3 99.8 78.4 2.30 5 0.06 1 (TMIN), 0.14 2 (GDD5) Humulus lupulus 723 ACSnl 63.6 99.6 72.2 3.55 3 0.16 1 (TMIN), 3.68 2 (GDD5) Lunaria rediviva 230 ACSnl 87.0 100.0 87.0 8.55 1 0.16 2 (TMIN), 0.24 3 (WATB 2 ) Ranunculus lanuginosus 370 ACSnl 100.0 100.0 100.0 39.55 1 0.06 2 (TMIN), 0.12 3 (GDD5 2 ) Rumex sanguineus 629 ACnl 67.6 100.0 32.4 2.93 3 5.47 1 (TMIN), 0.33 2 (GDD5 2 ) Silene dioica 775 ACSnl 96.5 97.0 99.5 2.38 2 0.24 1 (GDD5), 2.16 3 (TMIN) Stellaria holostea 771 CSnl 56.6 43.4 100.0 1.11 9 0.35 1 (GDD5 2 ), 2.41 2 (GDD5) Stellaria nemorum 594 ACSnl 100.0 100.0 100.0 9.82 1 0.11 2 (GDD5), 0.25 3 (TMIN) Data analysis We used logistic regression ( Harrell 2001 ) to assess if and how the probability of species occurrence was influenced by climate, soil, and/or accessibility to postglacial recolonization. As these factors are not mutually exclusive as range controls, we investigated a primary set of 13 logistic regression models for each species, as shown in Table 1 . To minimize overfitting, the minimum allowable number of events per parameter was 10 ( Peduzzi et al. 1996 , see also Harrell 2001 ). Given that the maximum number of explanatory parameters in the primary set of models was 11, only species with≥105 presences and ≥105 absences were used for the above analyses. All explanatory parameters were standardized. The inclusion of an autocovariate has been a standard approach to handle spatial autocorrelation in species distribution modeling ( Dormann 2007b ). However, this approach has recently been shown to consistently underestimate the effect of the independent predictor variables, resulting in biased estimates compared to non‐spatial logistic regression ( Dormann 2007a , Dormann et al. 2007 ). Furthermore, there is a concern that regression methods that explicitly model spatial autocorrelation may cause scale shifts in the coefficients estimates and generate complex patterns of collinearity between space and the independent predictor variables ( Hawkins et al. 2007 ). As a consequence, we used non‐spatial logistic regression in the present study. 1 The non‐intercept parameters in the logistic regression models (M i ; only i is given). GDD5, growing degree days; TMIN, absolute minimum temperature; WATB, water balance; pH, soil pH; SAND, sand content; ACC, accessibility. Quadratic terms are indicated by 2 . The ACSnl model was also refit using four trend‐surface variables (latitude, longitude, latitide 2 , and longitude 2 ) instead of ACC (M XYnlCSnl ). I GDD5 TMIN WATB pH SAND GDD5 2 TMIN 2 WATB 2 pH 2 SAND 2 ACC C × S × CS × × Cnl × × Snl × × CSnl × × × × A × AC × × AS × × ACS × × × ACnl × × × ASnl × × × ACSnl × × × × × The degree of support for each model and the three hypothesized range controls was assessed using the information‐theoretic approach ( Burnham and Anderson 2002 ). For each species, the relative support for each model was assessed using Akaike's information criterion (AIC=−2ln(L)+2K), which estimates the Kullbach‐Leibler information lost by approximating full reality with a given model. It simultaneously accounts for model fit (L, model likelihood) and complexity (K, number of model parameters, including the intercept; Burnham and Anderson 2002 ). Given that the maximum value for K was 12 and thus smaller than n/40=22 (n=881 cells), it was not necessary to use the small‐sample bias‐corrected version of AIC ( Burnham and Anderson 2002 ). To assess the relative support for each model, we computed AIC differences, ▵AIC(i)=AIC(i)‐AIC(min), where AIC(i) is the observed value of AIC for model i and AIC(min) is the minimum AIC value in the set of candidate models. Thus, ▵AIC=0 for the best model. Models with ▵AIC≤2, 4–7, and >10 are considered to have substantial, considerably less, and essentially no support, respectively ( Burnham and Anderson 2002 ). We also computed the Akaike weight (w) for each model ( Burnham and Anderson 2002 ). This value can be interpreted as the probability that a given model is the best model in the candidate set ( Burnham and Anderson 2002 ). To summarize the relative support for the three hypothesized range controls, we computed the probability that the best model for a given species includes the climate variables, soil variables, or accessibility as the sum of the Akaike weights for all those models in which they occurred ( Burnham and Anderson 2002 ). The proportion of variation in species occurrences explained by a given model was represented by the likelihood ratio R 2 (R L 2 =−2[ln(L 0 )–ln(L M )]/−2[ln(L 0 )], where L 0 is the likelihood function for the model containing only an intercept and L M is the likelihood function for the model in question), which was found to be the superior measure in a comparison of coefficients of determination for multiple logistic regression ( Menard 2000 ). To separately estimate the strength of three main explanatory factors (climate, soil, and accessibility), we used variation partitioning to assess the amount of variation uniquely explained by the set of parameters associated with a factor, after controlling for the other explanatory factors under consideration ( Legendre and Legendre 1998 ). For example, the amount of variation uniquely explained by accessibility after controlling for climate and soil (including nonlinear relationships) was computed as R L 2 (ACC unique )=R L 2 (M ACSnl )−R L 2 (M CSnl ). We note that the postglacial migrational lag hypothesis specifically predicts that the species prevalence‐accessibility relationship should be positive. We assessed whether or not the results conform to this prediction using model averaging to provide a robust estimate for the ACC regression coefficient based on the full suite of models described above ( Burnham and Anderson 2002 , Johnson and Omland 2004 ). The model‐averaged parameter estimate for ACC was computed as the weighted average (β A (ma)=Σw(i)β i ) across the 13 models for a given species, with β i =0 for models without the predictor and w i =the Akaike weight for model i ( Burnham and Anderson 2002 , Johnson and Omland 2004 ). This parameter estimate was then converted into an odds ratio (OR A , using the formula exp(−β A (ma)), which indicates the multi‐model predicted change in odds of presence to absence for a unit change in the standardized ACC. For comparison, we report the similarly estimated odds ratios for the two strongest climate and soil variables for each species. To allow for a broader range of historically controlled spatial patterns than represented by ACC, we replaced it in the most complex and most preferred model (M ACSnl , see Results) with a set of trend surface parameters (cf. Svenning and Skov 2005 ), namely standardized longitude (Lon) and latitude (Lat) and their quadratic terms (Lon 2 , Lat 2 ). These trend surface parameters can represent a wider suite of broad‐scale spatial patterns ( Legendre and Legendre 1998 ). For each species, the w, and R L 2 of the resulting model (M XYnlCSnl ) were compared to those obtained for M ACSnl . With respect to the trend surface parameters, the postglacial migrational lag hypothesis has a specific prediction: given that ice age refugia were located across southern parts of Europe, the species prevalence should decrease towards the north and, less markedly, towards the western and eastern margins, i.e. the presence/absence odds ratio (estimated, as explained above) for Lat and Lon 2 should both be <1.0. We used Kruskal‐Wallis tests to test for differences between the growth forms and boreal versus non‐boreal species in the importance of accessibility, as represented the summed Akaike weights for models including accessibility (w A ), amount of variation uniquely explained by accessibility (R L 2 (ACC unique )), and the multi‐model odds ratio for accessibility (OR A ). Since the goals of the presented analyses were effect estimation and hypothesis testing (sensu lato) and not prediction, model validation was not necessary ( Harrell 2001 : 82–83). The statistical analyses were computed using JMP 6.0.0 (SAS Inst. Cary, NC) and Microsoft Office Excel 2003 (Microsoft, Redmond, WA). GIS operations were performed in ArcGIS 9.1 (ESRI, Redlands, CA). Results Twelve species were too ubiquitous to be analyzed, with <105 absences (see Methods): the trees Alnus glutinosa (17 absences), Betula pendula (32 absences), B. pubescens (76 absences), Populus tremula (23 absences), and Quercus robur (24 absences); the ferns Athyrium filix‐femina (34 absences), Dryopteris filix‐mas (26 absences), and Pteridium aquilinum (37 absences); and the herbs Anemone nemorosa (41 absences), Moehringia trinervia (75 absences), Ranunculus auricomus (93 absences), and Ranunculus ficaria (25 absences). Hence, 25.5% of the study species were largely ubiquitous and clearly not strongly limited by postglacial migrational lag in north‐central Europe. Species that are widespread in the boreal zone of Eurasia were clearly overrepresented, with six species ( Betula pendula , B. pubescens , Populus tremula , Athyrium filix‐femina , Pteridium aquilinum , and Ranunculus auricomus ) among the 12 nearly ubiquitous species, when compared to the non‐ubiquitous species, of which the boreal species constitute just 17% ( Table 2 ; Fisher's exact test, one‐tailed p<0.05). No species were too restricted (i.e. <105 presences) to be analyzed. All subsequent results refer to the remaining 35 moderately widespread species ( Table 2 ). Relative importance of accessibility versus climate and soil Accessibility to postglacial recolonization was consistently selected as an important factor ( Table 2 ). There was on average 91.4% support for including accessibility in the best model. The support was <50% for just two species ( Equisetum sylvaticum and Stellaria holostea ), while it was 100.0% for 17 species. Climate and soil were also consistently selected as important factors ( Table 2 ). There was always 100.0% support for including climate in the best model, while the average support for including soil was 95.0%. In terms of specific models, the model including accessibility as well as non‐linear climate and soil relationships (M ACSnl ) was strongly preferred, being the best for 31 species and having substantial support for the remaining four species ( Table 3 ). 3 Details of the 13 regression models (M i , only i is shown). Shown are their total number of estimated parameters (K) including the intercept, their mean (±SD) ▵AIC across the 35 study species, the number of times a model was selected as best (n best ; ▵AIC=0), the number of times a model was determined not to be the best, but to have substantial support (n subst ; 0<▵AIC≤2), the mean (± SD) Akaike weight (w), and the mean proportion (± SD) of variation in species occurrences explained by a given model (R L 2 ). i K ▵AIC n best n subst w (%) R L 2 C 4 153.6±117.1 0 0 0.0±0.0 0.299±0.138 S 3 430.7±188.6 0 0 0.0±0.0 0.056±0.043 CS 6 142.1±116.3 0 0 0.0±0.0 0.312±0.133 Cnl 7 69.4±71.1 0 0 0.2±0.6 0.366±0.136 Snl 5 410.3±183.7 0 0 0.0±0.0 0.082±0.043 CSnl 11 45.8±64.0 2 3 8.5±16.2 0.389±0.127 A 1 352.0±165.0 0 0 0.0±0.0 0.120±0.146 AC 5 83.1±59.7 0 0 0.3±1.8 0.353±0.138 AS 4 315.0±147.6 0 0 0.0±0.0 0.162±0.141 ACS 7 72.8±59.5 1 0 2.1±9.7 0.363±0.137 Acnl 8 20.8±18.4 1 1 4.5±12.6 0.399±0.139 Asnl 6 299.3±147.8 0 0 0.0±0.0 0.180±0.141 ACSnl 12 0.1±0.4 31 4 84.4±21.9 0.418±0.134 Although consistently important, accessibility to postglacial recolonization only accounted for a small to moderate amount of variation (R L 2 ): accessibility (M A ) accounted for about a third of the variation accounted for by the 10 climate and soil variables (M CSnl ; Table 3 ). After controlling for climate and soil, accessibility on average uniquely accounted for just 2.9±3.6% SD of the variation in species occurrences (R L 2 (ACC unique )=R L 2 (M ACSnl )−R L 2 (M CSnl ); Table 3 ). The unique accessibility effect was largest for the ferns Polystichum aculeatum (10.6%) and Thelypteris limbosperma (7.4%), the trees Fagus sylvatica (9.5%) and Carpinus betulus (7.6%), and the herbs Ranunculus lanuginosus (12.4%), Corydalis cava (10.9%), and Lunaria redidiva (8.6%). The importance of accessibility (w A and R L 2 (ACC unique )) did not differ among the growth forms (Kruskal‐Wallis tests, p≥0.27). However, species that are widespread in the boreal zone exhibited particularly low importance of accessibility ( Table 2 ; w A : median 84% (boreal) and 100% (non‐boreal), p<0.005; R L 2 (ACC unique ): median 0.5% (boreal) and 2.4% (non‐boreal), p<0.005; Kruskal‐Wallis tests). Nature of the species prevalence‐accessibility relationship Considering the model‐averaged regression coefficients provided strong evidence in favor of a postglacial accessibility interpretation ( Table 2 ): the resulting odds ratios were greater than one for 33 out of 35 species (one‐tailed sign test, p<<0.0005). Hence, adjusted for the effects of climate and soil, the prevalence of the far majority of the 35 non‐ubiquitous species increased with increasing accessibility to postglacial recolonization ( Fig. 3 ). Two species ( Matteuccia struthiopteris and Corydalis pumila ) had odds ratios less than one, but in both cases these were weak (being on a 6th or 8th place in strength compared to the climate and soil predictors). In contrast, the coefficient for accessibility was stronger than any of the coefficients for climate or soil for 11 species and the second or third strongest coefficient for an additional 10 species ( Table 2 ). The general strength of the relationship is also clear from Fig. 3 . Furthermore, considering the four species with the largest OR A , our results suggest that they would all be much more widespread in the western and northern parts of the study region if migrational lag was minimized ( Fig. 4 ). 3 The predicted response (probability of occurrence) to accessibility to postglacial recolonization from ice age refugia (ACC) for the 35 non‐ubiquitous northern nemoral forest plant study species. The probability of occurrence was computed as (1+exp(β 0 (ma)+ β A (ma)×ACC)) −1 (following the coding in JMP), where β 0 (ma) and β A (ma) are the model‐averaged parameter estimates for the intercept and ACC (see Materials and methods); β A (ma), therefore, represents the ACC relationship controlled for the species’ response to climate and soil. Further note that the probability of occurrence shown represents the case where all climate and soil variables are at zero (their mean). ACC varies between its empirical minimum and maximum in the study area (unit 10 −6 ×km −1 ). The thick line shows the average predicted occurrence across the 35 species. 4 The observed native distribution of the four species with the strongest accessibility relationships (largest OR A ), their predicted probability of occurrence according to M ACSnl (this model had 100% support for all four species), and their predicted probability of occurrence according to M ACSnl , when accessibility (ACC) was set to its observed maximum for all grid cells (right column). The darker the color, the higher the probability of occurrence (range 0–100%; the scale gives 10 equal intervals). The accessibility relationship (OR A ) did not differ among the growth forms (Kruskal‐Wallis test, p=0.15). In contrast, the species that are widespread in the boreal zone also exhibited particularly low strength of the accessibility relationship ( Table 2 ; OR A : median 1.4 (boreal) and 3.4 (non‐boreal), p<0.01; Kruskal‐Wallis test). Trend surface variables as an alternative measure of postglacial recolonization accessibility When the most preferred (and most complex) model M ACSnl was compared to its trend surface variant (M XYnlCSnl ), where the direct accessibility measure (ACC) was replaced by four trend surface variables, the latter received strong support (average w=93.8±14.6% SD). However, on average it explained only a small amount of additional variation (mean R L 2 difference=4.4±3.4% SD). As predicted by the postglacial migrational lag hypothesis, most species were more prevalent to the south and center than could be explained by the environment. The odds ratios for Lat and Lon 2 were <1 for 31 and 25 species, respectively (median=0.34 (Lat) and 0.57 (Lon 2 ); n=35). It is noteworthy that the Lat and Lon 2 relationships were among the strongest in this model, being in the top five in terms of the mean and median absolute size of the regression coefficients (results not shown). Discussion This study was focused on a set of widespread forest species that represent the kind of species most likely to be in equilibrium with climate and soil, as they all have experienced strong postglacial range expansion and their habitat (forest) has been widely available in the study area since the early Holocene. Nevertheless, our findings are largely consistent with the postglacial migrational lag hypothesis: notably, for the 35 non‐ubiquitous species, accessibility to postglacial recolonization was consistently selected as an important factor, and prevalence for the far majority of these species (33 species) increased with increasing accessibility, and often strongly so ( Fig. 3 ). The alternative analysis using trend‐surface parameters also indicated that most of the non‐ubiquitous species were more prevalent to the south and center of nemoral Europe than could be explained by climate or soil. When considering the role of postglacial migrational lag as a range constraint for widespread northern nemoral species, it is important to note several additional aspects of our results. Firstly, of the 47 study species 12 species were near‐ubiquitous. Hence, a quarter of the species had effectively overcome any postglacial migrational constraints in the study area. Secondly, although the evidence for migrational lag as important range constraint is strong, there was also strong evidence for the importance of climate and, to a lesser extent, soil. The latter findings are consistent with traditional expectations ( Ellenberg 1988 , Pearson and Dawson 2003 ), and show that migrational lag and climate and soil simultaneously act to limit the ranges of widespread northern nemoral plant species. A similar conclusion has been reached in studies of plant species distributions in nemoral forests at smaller spatial scales ( Svenning and Skov 2002 ) as well as in other regions and biomes, e.g. New World tropical forests ( Duivenvoorden et al. 2002 , Tuomisto et al. 2003 , Vormisto et al. 2004 , Normand et al. 2006 ). Based on general considerations of the dispersal and colonization capacity of the growth forms, it could be expected that the importance of accessibility would have been strongest for herbs, intermediate for trees, and weakest for ferns. These potential differences among the three growth forms were not supported by our analyses. It is possible that fern dispersal capabilities have been exaggerated. Notably, a number of forest fern species, including our study species Athyrium filix‐femina , Dryopteris carthusiana , D. filix‐mas (and/or D. pseudomas ), Gymnocarpium dryopteris , Pteridium aquilinum , and Thelypteris phegopteris , are reported to be poor colonizers of secondary woodland in Europe ( Hermy et al. 1999 ). Elsewhere regional fern species distributions appear to be constrained by dispersal to nearly the same extent as for flowering plants ( Tuomisto et al. 2003 ). Furthermore, the mechanisms generating the migrational lags may be more complex than just insufficient propagule dispersal. Other factors such as biotic interactions and Allee effects could also contribute to postglacial migrational lags ( Case et al. 2005 , Holt et al. 2005 ), prohibiting simple growth form generalizations. However, little or no empirical evidence exists for the importance of these factors for range limits in plants, and biotic interactions are generally thought to primarily affect species distributions at small scales ( Pearson and Dawson 2003 ), except when modulated by climate or other large‐scale gradients in the environment. In contrast, several additional aspects of our findings provide clear support for the interpretation that the accessibility relationships reflect postglacial migrational lags: the southern European refugia region assumed in this study is primarily relevant for truly nemoral species, while cold‐hardy boreal species probably had more northern refuge locations in central and eastern Europe and recolonized northern europe from here ( Stewart and Lister 2001 , Palmé et al. 2003 , Willis and van Andel 2004 , Cheddadi et al. 2006 , Maliouchenko et al. 2007 ). These boreal species are less likely to still experience postglacial migrational lags, and even if they do, the accessibility measure used here is unlikely to provide an adequate measure of accessibility to postglacial recolonization for species with important higher‐latitude refugia. Fossil evidence and vegetation modeling suggest this was the case for Pinus sylvestris ( Cheddadi et al. 2006 ). As could be expected, this species also had a relatively weak response to accessibility (see OR A in Table 2 ; R L 2 (ACC unique ) is also low for this species, just 0.6%). Our results were also more broadly in agreement with this expectation; species that are widespread in the boreal zone of Eurasia were strongly overrepresented among the nearly ubiquitous species, and among the 35 non‐ubiquitous species the six species that are widespread in the boreal zone of Eurasia exhibited particularly low importance and strength of the accessibility relationship. Comparing the accessibility relationships for the trees to the known postglacial recolonization record for the different species ( Huntley and Birks 1983 , Lang 1994 , Petit et al. 2002 , Tinner and Lotter 2006 , Cheddadi et al. 2006 ) provide additional support for impotence of postglacial migrational lag in determining species present distribution. Unfortunately, the latter information is not available for most herb and fern species. If the accessibility relationships reflect postglacial migrational lags, we expect them to be strongest for species that expanded across nemoral Europe relatively late in the Holocene. This is exactly the result of our analyses; the two tree species with very strong, positive accessibility relationships are Carpinus betulus and Fagus sylvatica ( Table 2 ). These two species expanded across nemoral Europe much later in the Holocene than the other study tree species ( Huntley and Birks 1983 , Lang 1994 ). The role of migrational lag contra climatic control of the postglacial expansion of Fagus sylvatica has been much discussed, with several recent studies concluding that its range has been ( Giesecke et al. 2007 ) and still is much limited by migrational lag ( Svenning and Skov 2004, 2007b , Fang and Lechowicz 2006 ; but see Tinner and Lotter 2006 ). Our results suggest that they would be much more widespread in the western and northern parts of the study region if not for migrational lag ( Fig. 4 ). Providing direct evidence for this interpretation, Fagus sylvatica is widely naturalized in forests beyond its native range throughout the British Isles and the Norwegian lowland ( Lid and Lid 1994 , Peterken 1996 ). Carpinus betulus is likewise reported as naturalized throughout the British Isles, in the Norwegian lowland, and north of its native range in the Baltics ( Jalas and Suominen 1972 –1994, Lid and Lid 1994 ). The other two species with strong accessibility relationships are the forest herbs Corydalis cava and Ranunculus lanuginosus ( Table 2 ). Although our results suggest that most of the study region would be suitable for these species in terms of climate and soil ( Fig. 4 ), they have been much less planted and, therefore, have not had many opportunities to naturalize beyond their native range. However, Corydalis cava is reported to naturalize in woods and hedges in England and Wales ( Stace 1997 ) as well as in the Netherlands, west of its native range ( Jalas and Suominen 1972 –1994). Traditionally, climate is emphasized as the main control over large‐scale species ranges ( Pearson and Dawson 2003 ). While the present study confirms that climate is an important range constraint, it also provides evidence that the ranges of many widespread northern‐nemoral forest plants probably still are moderately to strongly limited by postglacial migrational lag, as suggested by the postglacial migrational lag hypothesis. This conclusion is consistent with several prior analyses of species distribution and diversity patterns, bioclimatic and dispersal modeling, and observations of extensive naturalization beyond native ranges, which have suggested that many temperate tree and forest herb species are still expanding from their ice age refuges and have strongly dispersal‐limited ranges in Europe ( Skov and Svenning 2004 , Svenning and Skov 2004, 2005, 2007a,b ). A recently reported transplant study also provides strong evidence that dispersal may limit broad‐scale ranges of forest plant species in north central Europe ( Van der Veken et al. 2007 ). Postglacial migrational lag appear to affect other organism groups as well, notably amphibians and reptiles ( Araújo and Pearson 2005 , Araújo et al. 2008 ). The support for the postglacial migrational lag hypothesis provided by the present study is especially noteworthy given its focus on widespread, northern species, i.e. the species least likely to be limited by dispersal ( Svenning and Skov 2007a ). The previous support for the postglacial migrational lag hypothesis have primarily concerned southern species with small ranges, i.e. species still associated with their ice age refugia ( Svenning and Skov 2004, 2007a , Araújo et al. 2008 ). Our results underscore the need to reconsider the equilibrium postulate as a basis for calibrating predictive species distribution models ( Guisan and Thuiller 2005 , Guisan et al. 2006 , Giesecke et al. 2007 ). The combined importance of climate and migrational lag as range controls suggests that although species ranges are influenced by climate, we cannot expect most forest plant species to closely track the expected climatic changes. Bioclimatic envelope studies indicate that strong dispersal limitation will exacerbate plant species range losses in Europe ( Skov and Svenning 2004 , Thuiller et al. 2005 ). This has even been shown specifically for northern nemoral forest plants and particularly so with respect to northern Europe, where, given efficient dispersal, climate change could allow some degree of compensatory range expansion ( Svenning and Skov 2006 ). As the reduced and fragmented status of European forests will strongly reduce the migration rates of forest plants ( Honnay et al. 2002 ), the limited ability of many plant species to track the changing climate will pose one of the greatest challenges to European nature conservation in the 21st century. Acknowledgements We thank the Atlas Florae Europaeae project for access to the distribution data. We also thank David Viner from the Climatic Research Unit, Univ. of East Anglia for allowing use of their climate data sets, and the Danish Natural Science Research Council for economic support (grant #21–04–0346 to JCS). http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Ecography Wiley

Postglacial dispersal limitation of widespread forest plant species in nemoral Europe

Download PDF
11 pages

Loading next page...
 
/lp/wiley/postglacial-dispersal-limitation-of-widespread-forest-plant-species-in-CXtYjydoWF

References (56)

Publisher
Wiley
Copyright
© 2008 The Authors
ISSN
0906-7590
eISSN
1600-0587
DOI
10.1111/j.0906-7590.2008.05206.x
Publisher site
See Article on Publisher Site

Abstract

Understanding the factors controlling species ranges is a central issue in evolution, ecology, and conservation biology. Climate is often singled out as the primary factor limiting the geographical range of species, in particular at large geographical scales ( Pearson and Dawson 2003 ). Climate may limit species distributions by direct physiological effects such as death of individuals due to frost damage or reproductive failure, indirect effects on biotic interactions ( Case et al. 2005 ), or even more subtly by affecting the balance between population extinction and recolonization ( Holt et al. 2005 ). Other environmental factors are generally thought to affect species distributions primarily at landscape and local scales ( Pearson and Dawson 2003 ). For plants, soil conditions are generally seen as being of predominant importance at these smaller scales ( Ellenberg 1988 ). More controversially it has been suggested that dispersal may also provide an important range constraint. In the absence of insuperable geographic barriers such as oceans and mountain chains, it has been argued that even plants are relatively mobile and via migration have rapidly tracked climate changes, with negligible dispersal constraints ( Pitelka et al. 1997 , Clark et al. 1998 ). However, even in the absence of strong barriers, dispersal may limit species ranges (for a theoretical perspective, see Holt et al. 2005 ). Notably, there is increasing evidence that many tree species are still expanding from their ice age refuges and have strongly dispersal‐limited ranges (the postglacial migrational lag hypothesis). Recent estimates of tree migration rates <100 m yr −1 imply that many species have probably not reached equilibrium with the modern environment (notably climate and soil) in Europe ( Svenning and Skov 2007b ). Equilibrium refers to the situation where a species’ range is completely determined by the present environmental conditions; the species occurs in all suitable areas and is absent from all unsuitable areas ( Araújo and Pearson 2005 ). Furthermore, bioclimatic modeling, and observations of naturalization or population dynamics suggest that many temperate tree species are still expanding from their ice age refuges and have strongly dispersal‐limited ranges in Europe ( Svenning and Skov 2004, 2007b ) and elsewhere ( Johnstone and Chapin 2003 ). Similar evidence is available for European forest herbs ( Skov and Svenning 2004 ), with a recently reported 45‐yr transplant experiment clearly showing that the forest herb Hyacinthoides non‐scripta is dispersal‐limited at its northern range edge in western Europe ( Van der Veken et al. 2007 ). Richness of tree species with small ranges in Europe is concentrated in areas that had a relatively warm climate during the Last Glacial Maximum (LGM; Svenning and Skov 2007a ). In contrast, modern climate appear to be a much stronger determinant of the richness of all or just the widespread tree species in Europe than LGM climate ( Svenning and Skov 2007a ), although there are also strong non‐environmental broad‐scale spatial trends in the tree species richness pattern across Europe ( Svenning and Skov 2005 ). Such spatial patterns provide at best indirect evidence for postglacial migrational lag, but a simple measure of geographical accessibility to postglacial recolonization from glacial refuges explains 78% of the geographical variation in the region's tree diversity and was a much stronger diversity predictor than three key bioclimatic variables ( Svenning and Skov 2007a ). The importance of dispersal as a limiting factor for plant species distributions at regional and smaller scales, in the absence of strong geographic barriers, is also increasingly noted ( Tuomisto et al. 2003 , Normand et al. 2006 ). In Europe, climate is a weaker predictor of species composition of amphibians and reptiles than of plants, suggesting that these groups are even less in equilibrium with climate ( Araújo and Pearson 2005 , Araújo et al. 2008 ). Although there is increasing evidence for the importance of postglacial migrational lag as a large‐scale range constraint for certain groups of organisms in certain regions, notably for trees, amphibians, and reptiles in Europe ( Svenning and Skov 2004, 2007a,b , Araújo and Pearson 2005 ), many studies still conclude that large‐scale species ranges are largely in equilibrium with climate ( Pearson and Dawson 2003 , Tinner and Lotter 2006 ), especially as regards the more widespread species (cf. Svenning and Skov 2007a ). The postglacial migrational lag hypothesis is of strong applied importance. Predictive species distribution modeling is of crucial importance to conservation management and planning as well as for predicting the impact of climatic and other environmental changes ( Guisan et al. 2006 ), but is often based on the assumption that species are at least close to equilibrium with climate or, more generally, the present environment (the equilibrium postulate; Guisan and Zimmerman 2000 , Guisan and Thuiller 2005 ). Furthermore, modeling of the potential impact of near‐future climate change on species diversity have consistently predicted much greater losses of species when assuming negligible migration than if unlimited migration was assumed ( Skov and Svenning 2004 , Thuiller et al. 2005 ). In the present paper we use logistic regression modeling to test the importance of postglacial migrational lag and the two main competing hypothesized range constraints (climate and soil) of widespread northern‐nemoral forest plant species across a large geographical region, namely north‐central Europe. We used an information‐theoretic model selection and multi‐model inference approach ( Burnham and Anderson 2002 , Johnson and Omland 2004 ) as well as variation partitioning ( Legendre and Legendre 1998 ) to assess the relative support for each of them. The information‐theoretic model selection approach is specifically designed to assess the relative levels of support for a set of competing hypotheses and to allow inferences to be drawn from the whole set of competing models, e.g. model averaging can be used to make robust parameter estimates when several models have similar support ( Burnham and Anderson 2002 , Johnson and Omland 2004 ). We chose to study widespread forest species in order to investigate the range‐limiting factors for species that have experienced strong postglacial range expansion and are associated with the habitat (forest) that was most widely available in the region during the early and middle part of the Holocene, i.e. the species most likely to be in equilibrium with climate and soil (as discussed above). In our analyses, postglacial migrational lag was modeled using a measure of accessibility to postglacial recolonization from ice age refugia. We investigated two specific predictions of the postglacial migrational lag hypothesis: 1) accessibility to postglacial recolonization has strong relative support as a predictor of species ranges, and 2) species prevalence are higher in areas closest to the greatest number of ice age refugia, i.e. the species prevalence‐accessibility relationship is positive. We note that not all species are likely to be equally affected by postglacial migrational lags: one might expect postglacial migrational lag to differ among forest plant growth forms, being less important for ferns than for trees or herbs, as many fern species have effective long distance dispersal due to their lightweight spores with the potential to produce hermaphroditic gametophytes ( Barrington 1993 , Pausas and Sáez 2000 ). Furthermore, cold‐hardy boreal species probably had relatively northern refuge locations in central and eastern Europe and recolonized northern Europe from here ( Stewart and Lister 2001 , Palmé et al. 2003 , Willis and van Andel 2004 , Cheddadi et al. 2006 , Maliouchenko et al. 2007 ) and are therefore less likely to still experience postglacial migrational lag than truly nemoral species. Materials and methods Study area and species The study area was western and north central Europe between 47.0–60.0°N and west of 24.0°E ( Fig. 1 ). This study region roughly coincides with the nemoral biome, excluding its easternmost part, where species occurrence data are less complete. We refer to this area as nemoral Europe. 1 The study region (nemoral Europe; 881 grid cells of ca 50×50 km) and the number of northern nemoral forest plant study species (n=47) in each grid cell. The study species were northern nemoral forest plant species, operationally defined as tree and forest‐associated herb and fern (including fern allies) species native to Denmark (positioned in the middle of the northern margin of the nemoral zone). Hereby we ensured that the selected species had all experienced strong postglacial range expansion in the region. Nomenclature follows Jalas and Suominen (1970–1994). Trees were defined as self‐supporting woody species reaching at least 20 m in height, in accordance with Svenning and Skov (2004 , 2005). Forest‐associated herbs and ferns were those species whose habitat description for Denmark primarily includes forest and scrub habitats in at least one of two standard floras ( Hansen 1984 , Mossberg et al. 1994 ). Of the 49 qualifying species, the fern Dryopteris assimilis and the herb Stellaria neglecta were excluded due to inadequate recording (Jalas and Suominen 1970–1994), leaving 47 study species ( Fig. 1 ). The three growth forms, trees, ferns, and herbs, were represented by 13, 13, and 21 species, respectively. Furthermore, 12 species are widespread in the boreal zone of Eurasia, while the remaining species have distributions that are more restricted to the nemoral zone. Species distribution data were obtained from the Atlas Florae Europaeae (AFE), which maps distributions using an equal‐area grid with cells of ca 50×50 km based on the Universal Transverse Mercator projection and the Military Grid Reference System (Jalas and Suominen 1970–1994). Only the present native ranges were used in the analyses. After excluding two AFE cells due to missing soil data, the study area included 881 cells. Explanatory variables Climate and soil maps were available in, or resampled to, a resolution of 10′. Climate data for the period 1961–1990 were provided by the Hadley Centre for Climate Prediction and Research (< www.meto.gov.uk/research/hadleycentre >). Three climatic variables of well established importance for temperate plant distributions ( Sykes et al. 1996 ) were derived from mean monthly temperature and precipitation values: 1) growing degree days (for a 5°C base temperature, GDD5), 2) absolute minimum temperature (TMIN), 3) and water balance (WATB), computed as the annual sum of the monthly differences between precipitation and potential evapotranspiration ( Skov and Svenning 2004 ). To represent the main edaphic factors affecting plant distributions in the study area ( Ellenberg 1988 ), four topsoil variables were extracted from “The digital soil map of the world” ( FAO 2003 ): pH, base saturation, sand content (%), and CaCO 3 concentration (%). The climate and soil variables were related to the species data by computing the mean of each variable within each AFE cell. However, CaCO 3 concentration and base saturation were excluded from the analyses due to their high correlation with pH (Spearman r=0.812 and 0.929, respectively). The five remaining environmental variables were transformed to have skewness<∣1.0∣, if necessary. Hence, water balance was transformed by subtracting its original minimum value to have a minimum of 0 and cubic‐root transformed. To allow for nonlinear responses, quadratic terms were generated for all environmental variables. We assessed multicollinearity of the final 10 environmental parameters by computing their mutual tolerance values ( Quinn and Keough 2002 ). Tolerance was 0.30 for water balance, 0.33 for growing‐degree‐days, and 0.60–0.85 for the remaining eight variables. As tolerances were>0.10, multicollinearity among the environmental parameters was unlikely to be problematic ( Quinn and Keough 2002 ). Interactions between the environmental variables were not modeled. Postglacial migrational lag was represented by a measure of accessibility to postglacial recolonization from ice age refugia. If the range of a species is constrained by postglacial migrational lag, we expect the species to be more prevalent in areas closest to the greatest number of ice age refugia. Following Svenning and Skov (2007b) , accessibility (ACC) was computed as: the inverse of the summed distances (in km) to all grid cells in the source area (subsequently, multiplied by 10 6 to avoid very small numbers). Hence, the more distant a receiving grid cell on average is located from any source cell the lower its accessibility ( Fig. 2 ). The source area was set to be southern Europe at 43–46°N. Most nemoral species were probably restricted to areas south of 46°N during the LGM ( Petit et al. 2002 ; also cf. Magri et al. 2006 ). Furthermore, more southern refugia have not played a strong role as source areas for postglacial recolonization of central and northern Europe ( Petit et al. 2002 , Magri et al. 2006 ). The measure of accessibility used here is a strong predictor of overall tree species richness in the study area ( Svenning and Skov 2007b ). We note that it is a general measure that is not tailored to species‐specific refugia locations; notably, species that are widespread in the boreal zone of Eurasia may have recolonized northern Europe from refugia located further north and east ( Stewart and Lister 2001 , Palmé et al. 2003 , Willis and van Andel 2004 , Cheddadi et al. 2006 , Maliouchenko et al. 2007 ; Table 2 ). 2 Accessibility to postglacial recolonization from ice age refugia (ACC) across nemoral Europe. Accessibility was computed as the inverse of the summed distances to all grid cells in the source area (range 1.51–4.08 10 −6 ×km −1 ; the warmer the color, the higher ACC; the scale gives 10 equal intervals). The source area was set to be southern Europe at 43–46°N. 2 Results of logistic regression model selection for 35 forest plant species in nemoral Europe. The number of AFE cell occurrences is shown (n, total number of cells=881). The most strongly supported regression model M i (M best : only i is given; see Table 1 ), its Akaike weight (w best in %), and the summed Akaike weights for models including accessibility (w A ) or soil (w S ) are shown. The summed Akaike weights for models including climate (w C ) was always 100.0%. Odds ratios are given for accessibility (OR A ) and the two strongest climate/soil variables (OR C+S ) based on their model‐averaged regression coefficients; the superscript indicates their rank in terms of the absolute size of the coefficients, while bold face indicates species where OR A ≥4.0. B Widespread in the boreal zone of Eurasia. Growth form Species n M best w best w A w S OR A OR C+S Trees Pinus sylvestris B 450 ACSnl 94.1 94.1 100.0 1.84 8 0.03 1 (TMIN), 0.22 2 (GDD5) Carpinus betulus 548 ACSnl 100.0 100.0 100.0 30.03 1 0.27 2 (WATB 2 ), 3.41 3 (GDD5) Fagus sylvatica 575 ACSnl 100.0 100.0 100.0 24.52 1 0.05 2 (WATB 2 ), 5.84 3 (WATB) Quercus petraea 703 ACSnl 100.0 100.0 100.0 2.58 2 2.62 1 (TMIN), 0.43 3 (pH) Taxus baccata 327 ACSnl 99.5 99.5 100.0 1.61 3 0.60 1 (GDD5), 0.61 2 (TMIN) Ulmus glabra 735 ACSnl 95.4 100.0 95.4 3.44 2 0.16 1 (GDD5), 2.43 3 (WATB 2 ) Ulmus laevis 335 ACSnl 93.0 97.9 94.8 1.82 5 0.21 1 (TMIN), 4.11 2 (GDD5) Ulmus minor 472 ACS 56.8 100.0 85.3 3.04 2 13.02 1 (GDD5), 0.35 3 (TMIN) Ferns Asplenium adiantum‐nigrum 380 ACSnl 98.3 98.3 100.0 1.66 4 10.96 1 (TMIN), 2.73 2 (WATB) Blechnum spicant 675 ACSnl 90.5 99.3 100.0 2.11 4 9.12 1 (TMIN), 0.37 2 (GDD5) Dryopteris dilatata 657 ACSnl 99.3 100.0 99.3 2.02 4 3.91 1 (TMIN), 0.30 2 (GDD5) Equisetum sylvaticum B 713 CSnl 59.8 36.8 94.9 1.10 9 0.47 1 (GDD5 2 ), 0.47 2 (GDD5) Gymnocarpium dryopteris B 639 ACSnl 98.6 98.6 100.0 2.54 3 0.25 1 (GDD5), 0.37 2 (TMIN) Lycopodium annotinum B 493 ACSnl 73.3 74.2 97.7 1.43 7 0.07 1 (TMIN), 0.11 2 (GDD5) Matteuccia struthiopteris B 238 ACSnl 92.9 92.9 100.0 0.69 6 0.05 1 (TMIN), 0.22 2 (TMIN 2 ) Polystichum aculeatum 481 ACSnl 100.0 100.0 100.0 7.15 1 3.38 2 (TMIN), 0.31 3 (GDD5) Thelypteris limbosperma 478 ACSnl 99.5 100.0 100.0 4.97 1 0.26 2 (GDD5), 3.62 3 (TMIN) Thelypteris phegopteris B 596 ACSnl 66.2 66.3 99.7 1.36 7 0.32 1 (TMIN), 3.09 2 (WATB) Herbs Actaea spicata 558 ACSnl 100.0 100.0 100.0 6.71 2 0.13 1 (TMIN), 0.24 3 (GDD5) Anemone ranunculoides 512 ACSnl 100.0 100.0 100.0 7.81 1 0.13 2 (TMIN), 0.24 3 (TMIN 2 ) Cardamine bulbifera 396 ACSnl 54.9 59.6 95.1 1.16 9 0.13 2 (TMIN), 0.28 3 (TMIN 2 ) Cardamine flexuosa 638 ACSnl 99.6 100.0 99.6 3.71 3 5.09 1 (TMIN), 0.21 2 (GDD5) Cardamine impatiens 484 ACSnl 100.0 100.0 100.0 3.87 1 0.52 2 (pH 2 ), 0.52 3 (WATB 2 ) Corydalis cava 402 ACSnl 100.0 100.0 100.0 32.79 1 0.09 2 (TMIN), 0.20 3 (TMIN 2 ) Corydalis intermedia 349 ACSnl 67.8 73.1 94.4 1.25 8 0.11 1 (TMIN 2 ), 0.13 2 (TMIN) Corydalis pumila 108 ACSnl 85.0 85.0 100.0 0.60 8 0.04 1 (TMIN 2 ), 0.06 2 (TMIN) Fallopia dumetorum 578 ACSnl 82.2 82.7 99.4 6.44 1 0.17 2 (TMIN), 0.27 3 (WATB 2 ) Hepatica nobilis 498 ACSnl 78.3 99.8 78.4 2.30 5 0.06 1 (TMIN), 0.14 2 (GDD5) Humulus lupulus 723 ACSnl 63.6 99.6 72.2 3.55 3 0.16 1 (TMIN), 3.68 2 (GDD5) Lunaria rediviva 230 ACSnl 87.0 100.0 87.0 8.55 1 0.16 2 (TMIN), 0.24 3 (WATB 2 ) Ranunculus lanuginosus 370 ACSnl 100.0 100.0 100.0 39.55 1 0.06 2 (TMIN), 0.12 3 (GDD5 2 ) Rumex sanguineus 629 ACnl 67.6 100.0 32.4 2.93 3 5.47 1 (TMIN), 0.33 2 (GDD5 2 ) Silene dioica 775 ACSnl 96.5 97.0 99.5 2.38 2 0.24 1 (GDD5), 2.16 3 (TMIN) Stellaria holostea 771 CSnl 56.6 43.4 100.0 1.11 9 0.35 1 (GDD5 2 ), 2.41 2 (GDD5) Stellaria nemorum 594 ACSnl 100.0 100.0 100.0 9.82 1 0.11 2 (GDD5), 0.25 3 (TMIN) Data analysis We used logistic regression ( Harrell 2001 ) to assess if and how the probability of species occurrence was influenced by climate, soil, and/or accessibility to postglacial recolonization. As these factors are not mutually exclusive as range controls, we investigated a primary set of 13 logistic regression models for each species, as shown in Table 1 . To minimize overfitting, the minimum allowable number of events per parameter was 10 ( Peduzzi et al. 1996 , see also Harrell 2001 ). Given that the maximum number of explanatory parameters in the primary set of models was 11, only species with≥105 presences and ≥105 absences were used for the above analyses. All explanatory parameters were standardized. The inclusion of an autocovariate has been a standard approach to handle spatial autocorrelation in species distribution modeling ( Dormann 2007b ). However, this approach has recently been shown to consistently underestimate the effect of the independent predictor variables, resulting in biased estimates compared to non‐spatial logistic regression ( Dormann 2007a , Dormann et al. 2007 ). Furthermore, there is a concern that regression methods that explicitly model spatial autocorrelation may cause scale shifts in the coefficients estimates and generate complex patterns of collinearity between space and the independent predictor variables ( Hawkins et al. 2007 ). As a consequence, we used non‐spatial logistic regression in the present study. 1 The non‐intercept parameters in the logistic regression models (M i ; only i is given). GDD5, growing degree days; TMIN, absolute minimum temperature; WATB, water balance; pH, soil pH; SAND, sand content; ACC, accessibility. Quadratic terms are indicated by 2 . The ACSnl model was also refit using four trend‐surface variables (latitude, longitude, latitide 2 , and longitude 2 ) instead of ACC (M XYnlCSnl ). I GDD5 TMIN WATB pH SAND GDD5 2 TMIN 2 WATB 2 pH 2 SAND 2 ACC C × S × CS × × Cnl × × Snl × × CSnl × × × × A × AC × × AS × × ACS × × × ACnl × × × ASnl × × × ACSnl × × × × × The degree of support for each model and the three hypothesized range controls was assessed using the information‐theoretic approach ( Burnham and Anderson 2002 ). For each species, the relative support for each model was assessed using Akaike's information criterion (AIC=−2ln(L)+2K), which estimates the Kullbach‐Leibler information lost by approximating full reality with a given model. It simultaneously accounts for model fit (L, model likelihood) and complexity (K, number of model parameters, including the intercept; Burnham and Anderson 2002 ). Given that the maximum value for K was 12 and thus smaller than n/40=22 (n=881 cells), it was not necessary to use the small‐sample bias‐corrected version of AIC ( Burnham and Anderson 2002 ). To assess the relative support for each model, we computed AIC differences, ▵AIC(i)=AIC(i)‐AIC(min), where AIC(i) is the observed value of AIC for model i and AIC(min) is the minimum AIC value in the set of candidate models. Thus, ▵AIC=0 for the best model. Models with ▵AIC≤2, 4–7, and >10 are considered to have substantial, considerably less, and essentially no support, respectively ( Burnham and Anderson 2002 ). We also computed the Akaike weight (w) for each model ( Burnham and Anderson 2002 ). This value can be interpreted as the probability that a given model is the best model in the candidate set ( Burnham and Anderson 2002 ). To summarize the relative support for the three hypothesized range controls, we computed the probability that the best model for a given species includes the climate variables, soil variables, or accessibility as the sum of the Akaike weights for all those models in which they occurred ( Burnham and Anderson 2002 ). The proportion of variation in species occurrences explained by a given model was represented by the likelihood ratio R 2 (R L 2 =−2[ln(L 0 )–ln(L M )]/−2[ln(L 0 )], where L 0 is the likelihood function for the model containing only an intercept and L M is the likelihood function for the model in question), which was found to be the superior measure in a comparison of coefficients of determination for multiple logistic regression ( Menard 2000 ). To separately estimate the strength of three main explanatory factors (climate, soil, and accessibility), we used variation partitioning to assess the amount of variation uniquely explained by the set of parameters associated with a factor, after controlling for the other explanatory factors under consideration ( Legendre and Legendre 1998 ). For example, the amount of variation uniquely explained by accessibility after controlling for climate and soil (including nonlinear relationships) was computed as R L 2 (ACC unique )=R L 2 (M ACSnl )−R L 2 (M CSnl ). We note that the postglacial migrational lag hypothesis specifically predicts that the species prevalence‐accessibility relationship should be positive. We assessed whether or not the results conform to this prediction using model averaging to provide a robust estimate for the ACC regression coefficient based on the full suite of models described above ( Burnham and Anderson 2002 , Johnson and Omland 2004 ). The model‐averaged parameter estimate for ACC was computed as the weighted average (β A (ma)=Σw(i)β i ) across the 13 models for a given species, with β i =0 for models without the predictor and w i =the Akaike weight for model i ( Burnham and Anderson 2002 , Johnson and Omland 2004 ). This parameter estimate was then converted into an odds ratio (OR A , using the formula exp(−β A (ma)), which indicates the multi‐model predicted change in odds of presence to absence for a unit change in the standardized ACC. For comparison, we report the similarly estimated odds ratios for the two strongest climate and soil variables for each species. To allow for a broader range of historically controlled spatial patterns than represented by ACC, we replaced it in the most complex and most preferred model (M ACSnl , see Results) with a set of trend surface parameters (cf. Svenning and Skov 2005 ), namely standardized longitude (Lon) and latitude (Lat) and their quadratic terms (Lon 2 , Lat 2 ). These trend surface parameters can represent a wider suite of broad‐scale spatial patterns ( Legendre and Legendre 1998 ). For each species, the w, and R L 2 of the resulting model (M XYnlCSnl ) were compared to those obtained for M ACSnl . With respect to the trend surface parameters, the postglacial migrational lag hypothesis has a specific prediction: given that ice age refugia were located across southern parts of Europe, the species prevalence should decrease towards the north and, less markedly, towards the western and eastern margins, i.e. the presence/absence odds ratio (estimated, as explained above) for Lat and Lon 2 should both be <1.0. We used Kruskal‐Wallis tests to test for differences between the growth forms and boreal versus non‐boreal species in the importance of accessibility, as represented the summed Akaike weights for models including accessibility (w A ), amount of variation uniquely explained by accessibility (R L 2 (ACC unique )), and the multi‐model odds ratio for accessibility (OR A ). Since the goals of the presented analyses were effect estimation and hypothesis testing (sensu lato) and not prediction, model validation was not necessary ( Harrell 2001 : 82–83). The statistical analyses were computed using JMP 6.0.0 (SAS Inst. Cary, NC) and Microsoft Office Excel 2003 (Microsoft, Redmond, WA). GIS operations were performed in ArcGIS 9.1 (ESRI, Redlands, CA). Results Twelve species were too ubiquitous to be analyzed, with <105 absences (see Methods): the trees Alnus glutinosa (17 absences), Betula pendula (32 absences), B. pubescens (76 absences), Populus tremula (23 absences), and Quercus robur (24 absences); the ferns Athyrium filix‐femina (34 absences), Dryopteris filix‐mas (26 absences), and Pteridium aquilinum (37 absences); and the herbs Anemone nemorosa (41 absences), Moehringia trinervia (75 absences), Ranunculus auricomus (93 absences), and Ranunculus ficaria (25 absences). Hence, 25.5% of the study species were largely ubiquitous and clearly not strongly limited by postglacial migrational lag in north‐central Europe. Species that are widespread in the boreal zone of Eurasia were clearly overrepresented, with six species ( Betula pendula , B. pubescens , Populus tremula , Athyrium filix‐femina , Pteridium aquilinum , and Ranunculus auricomus ) among the 12 nearly ubiquitous species, when compared to the non‐ubiquitous species, of which the boreal species constitute just 17% ( Table 2 ; Fisher's exact test, one‐tailed p<0.05). No species were too restricted (i.e. <105 presences) to be analyzed. All subsequent results refer to the remaining 35 moderately widespread species ( Table 2 ). Relative importance of accessibility versus climate and soil Accessibility to postglacial recolonization was consistently selected as an important factor ( Table 2 ). There was on average 91.4% support for including accessibility in the best model. The support was <50% for just two species ( Equisetum sylvaticum and Stellaria holostea ), while it was 100.0% for 17 species. Climate and soil were also consistently selected as important factors ( Table 2 ). There was always 100.0% support for including climate in the best model, while the average support for including soil was 95.0%. In terms of specific models, the model including accessibility as well as non‐linear climate and soil relationships (M ACSnl ) was strongly preferred, being the best for 31 species and having substantial support for the remaining four species ( Table 3 ). 3 Details of the 13 regression models (M i , only i is shown). Shown are their total number of estimated parameters (K) including the intercept, their mean (±SD) ▵AIC across the 35 study species, the number of times a model was selected as best (n best ; ▵AIC=0), the number of times a model was determined not to be the best, but to have substantial support (n subst ; 0<▵AIC≤2), the mean (± SD) Akaike weight (w), and the mean proportion (± SD) of variation in species occurrences explained by a given model (R L 2 ). i K ▵AIC n best n subst w (%) R L 2 C 4 153.6±117.1 0 0 0.0±0.0 0.299±0.138 S 3 430.7±188.6 0 0 0.0±0.0 0.056±0.043 CS 6 142.1±116.3 0 0 0.0±0.0 0.312±0.133 Cnl 7 69.4±71.1 0 0 0.2±0.6 0.366±0.136 Snl 5 410.3±183.7 0 0 0.0±0.0 0.082±0.043 CSnl 11 45.8±64.0 2 3 8.5±16.2 0.389±0.127 A 1 352.0±165.0 0 0 0.0±0.0 0.120±0.146 AC 5 83.1±59.7 0 0 0.3±1.8 0.353±0.138 AS 4 315.0±147.6 0 0 0.0±0.0 0.162±0.141 ACS 7 72.8±59.5 1 0 2.1±9.7 0.363±0.137 Acnl 8 20.8±18.4 1 1 4.5±12.6 0.399±0.139 Asnl 6 299.3±147.8 0 0 0.0±0.0 0.180±0.141 ACSnl 12 0.1±0.4 31 4 84.4±21.9 0.418±0.134 Although consistently important, accessibility to postglacial recolonization only accounted for a small to moderate amount of variation (R L 2 ): accessibility (M A ) accounted for about a third of the variation accounted for by the 10 climate and soil variables (M CSnl ; Table 3 ). After controlling for climate and soil, accessibility on average uniquely accounted for just 2.9±3.6% SD of the variation in species occurrences (R L 2 (ACC unique )=R L 2 (M ACSnl )−R L 2 (M CSnl ); Table 3 ). The unique accessibility effect was largest for the ferns Polystichum aculeatum (10.6%) and Thelypteris limbosperma (7.4%), the trees Fagus sylvatica (9.5%) and Carpinus betulus (7.6%), and the herbs Ranunculus lanuginosus (12.4%), Corydalis cava (10.9%), and Lunaria redidiva (8.6%). The importance of accessibility (w A and R L 2 (ACC unique )) did not differ among the growth forms (Kruskal‐Wallis tests, p≥0.27). However, species that are widespread in the boreal zone exhibited particularly low importance of accessibility ( Table 2 ; w A : median 84% (boreal) and 100% (non‐boreal), p<0.005; R L 2 (ACC unique ): median 0.5% (boreal) and 2.4% (non‐boreal), p<0.005; Kruskal‐Wallis tests). Nature of the species prevalence‐accessibility relationship Considering the model‐averaged regression coefficients provided strong evidence in favor of a postglacial accessibility interpretation ( Table 2 ): the resulting odds ratios were greater than one for 33 out of 35 species (one‐tailed sign test, p<<0.0005). Hence, adjusted for the effects of climate and soil, the prevalence of the far majority of the 35 non‐ubiquitous species increased with increasing accessibility to postglacial recolonization ( Fig. 3 ). Two species ( Matteuccia struthiopteris and Corydalis pumila ) had odds ratios less than one, but in both cases these were weak (being on a 6th or 8th place in strength compared to the climate and soil predictors). In contrast, the coefficient for accessibility was stronger than any of the coefficients for climate or soil for 11 species and the second or third strongest coefficient for an additional 10 species ( Table 2 ). The general strength of the relationship is also clear from Fig. 3 . Furthermore, considering the four species with the largest OR A , our results suggest that they would all be much more widespread in the western and northern parts of the study region if migrational lag was minimized ( Fig. 4 ). 3 The predicted response (probability of occurrence) to accessibility to postglacial recolonization from ice age refugia (ACC) for the 35 non‐ubiquitous northern nemoral forest plant study species. The probability of occurrence was computed as (1+exp(β 0 (ma)+ β A (ma)×ACC)) −1 (following the coding in JMP), where β 0 (ma) and β A (ma) are the model‐averaged parameter estimates for the intercept and ACC (see Materials and methods); β A (ma), therefore, represents the ACC relationship controlled for the species’ response to climate and soil. Further note that the probability of occurrence shown represents the case where all climate and soil variables are at zero (their mean). ACC varies between its empirical minimum and maximum in the study area (unit 10 −6 ×km −1 ). The thick line shows the average predicted occurrence across the 35 species. 4 The observed native distribution of the four species with the strongest accessibility relationships (largest OR A ), their predicted probability of occurrence according to M ACSnl (this model had 100% support for all four species), and their predicted probability of occurrence according to M ACSnl , when accessibility (ACC) was set to its observed maximum for all grid cells (right column). The darker the color, the higher the probability of occurrence (range 0–100%; the scale gives 10 equal intervals). The accessibility relationship (OR A ) did not differ among the growth forms (Kruskal‐Wallis test, p=0.15). In contrast, the species that are widespread in the boreal zone also exhibited particularly low strength of the accessibility relationship ( Table 2 ; OR A : median 1.4 (boreal) and 3.4 (non‐boreal), p<0.01; Kruskal‐Wallis test). Trend surface variables as an alternative measure of postglacial recolonization accessibility When the most preferred (and most complex) model M ACSnl was compared to its trend surface variant (M XYnlCSnl ), where the direct accessibility measure (ACC) was replaced by four trend surface variables, the latter received strong support (average w=93.8±14.6% SD). However, on average it explained only a small amount of additional variation (mean R L 2 difference=4.4±3.4% SD). As predicted by the postglacial migrational lag hypothesis, most species were more prevalent to the south and center than could be explained by the environment. The odds ratios for Lat and Lon 2 were <1 for 31 and 25 species, respectively (median=0.34 (Lat) and 0.57 (Lon 2 ); n=35). It is noteworthy that the Lat and Lon 2 relationships were among the strongest in this model, being in the top five in terms of the mean and median absolute size of the regression coefficients (results not shown). Discussion This study was focused on a set of widespread forest species that represent the kind of species most likely to be in equilibrium with climate and soil, as they all have experienced strong postglacial range expansion and their habitat (forest) has been widely available in the study area since the early Holocene. Nevertheless, our findings are largely consistent with the postglacial migrational lag hypothesis: notably, for the 35 non‐ubiquitous species, accessibility to postglacial recolonization was consistently selected as an important factor, and prevalence for the far majority of these species (33 species) increased with increasing accessibility, and often strongly so ( Fig. 3 ). The alternative analysis using trend‐surface parameters also indicated that most of the non‐ubiquitous species were more prevalent to the south and center of nemoral Europe than could be explained by climate or soil. When considering the role of postglacial migrational lag as a range constraint for widespread northern nemoral species, it is important to note several additional aspects of our results. Firstly, of the 47 study species 12 species were near‐ubiquitous. Hence, a quarter of the species had effectively overcome any postglacial migrational constraints in the study area. Secondly, although the evidence for migrational lag as important range constraint is strong, there was also strong evidence for the importance of climate and, to a lesser extent, soil. The latter findings are consistent with traditional expectations ( Ellenberg 1988 , Pearson and Dawson 2003 ), and show that migrational lag and climate and soil simultaneously act to limit the ranges of widespread northern nemoral plant species. A similar conclusion has been reached in studies of plant species distributions in nemoral forests at smaller spatial scales ( Svenning and Skov 2002 ) as well as in other regions and biomes, e.g. New World tropical forests ( Duivenvoorden et al. 2002 , Tuomisto et al. 2003 , Vormisto et al. 2004 , Normand et al. 2006 ). Based on general considerations of the dispersal and colonization capacity of the growth forms, it could be expected that the importance of accessibility would have been strongest for herbs, intermediate for trees, and weakest for ferns. These potential differences among the three growth forms were not supported by our analyses. It is possible that fern dispersal capabilities have been exaggerated. Notably, a number of forest fern species, including our study species Athyrium filix‐femina , Dryopteris carthusiana , D. filix‐mas (and/or D. pseudomas ), Gymnocarpium dryopteris , Pteridium aquilinum , and Thelypteris phegopteris , are reported to be poor colonizers of secondary woodland in Europe ( Hermy et al. 1999 ). Elsewhere regional fern species distributions appear to be constrained by dispersal to nearly the same extent as for flowering plants ( Tuomisto et al. 2003 ). Furthermore, the mechanisms generating the migrational lags may be more complex than just insufficient propagule dispersal. Other factors such as biotic interactions and Allee effects could also contribute to postglacial migrational lags ( Case et al. 2005 , Holt et al. 2005 ), prohibiting simple growth form generalizations. However, little or no empirical evidence exists for the importance of these factors for range limits in plants, and biotic interactions are generally thought to primarily affect species distributions at small scales ( Pearson and Dawson 2003 ), except when modulated by climate or other large‐scale gradients in the environment. In contrast, several additional aspects of our findings provide clear support for the interpretation that the accessibility relationships reflect postglacial migrational lags: the southern European refugia region assumed in this study is primarily relevant for truly nemoral species, while cold‐hardy boreal species probably had more northern refuge locations in central and eastern Europe and recolonized northern europe from here ( Stewart and Lister 2001 , Palmé et al. 2003 , Willis and van Andel 2004 , Cheddadi et al. 2006 , Maliouchenko et al. 2007 ). These boreal species are less likely to still experience postglacial migrational lags, and even if they do, the accessibility measure used here is unlikely to provide an adequate measure of accessibility to postglacial recolonization for species with important higher‐latitude refugia. Fossil evidence and vegetation modeling suggest this was the case for Pinus sylvestris ( Cheddadi et al. 2006 ). As could be expected, this species also had a relatively weak response to accessibility (see OR A in Table 2 ; R L 2 (ACC unique ) is also low for this species, just 0.6%). Our results were also more broadly in agreement with this expectation; species that are widespread in the boreal zone of Eurasia were strongly overrepresented among the nearly ubiquitous species, and among the 35 non‐ubiquitous species the six species that are widespread in the boreal zone of Eurasia exhibited particularly low importance and strength of the accessibility relationship. Comparing the accessibility relationships for the trees to the known postglacial recolonization record for the different species ( Huntley and Birks 1983 , Lang 1994 , Petit et al. 2002 , Tinner and Lotter 2006 , Cheddadi et al. 2006 ) provide additional support for impotence of postglacial migrational lag in determining species present distribution. Unfortunately, the latter information is not available for most herb and fern species. If the accessibility relationships reflect postglacial migrational lags, we expect them to be strongest for species that expanded across nemoral Europe relatively late in the Holocene. This is exactly the result of our analyses; the two tree species with very strong, positive accessibility relationships are Carpinus betulus and Fagus sylvatica ( Table 2 ). These two species expanded across nemoral Europe much later in the Holocene than the other study tree species ( Huntley and Birks 1983 , Lang 1994 ). The role of migrational lag contra climatic control of the postglacial expansion of Fagus sylvatica has been much discussed, with several recent studies concluding that its range has been ( Giesecke et al. 2007 ) and still is much limited by migrational lag ( Svenning and Skov 2004, 2007b , Fang and Lechowicz 2006 ; but see Tinner and Lotter 2006 ). Our results suggest that they would be much more widespread in the western and northern parts of the study region if not for migrational lag ( Fig. 4 ). Providing direct evidence for this interpretation, Fagus sylvatica is widely naturalized in forests beyond its native range throughout the British Isles and the Norwegian lowland ( Lid and Lid 1994 , Peterken 1996 ). Carpinus betulus is likewise reported as naturalized throughout the British Isles, in the Norwegian lowland, and north of its native range in the Baltics ( Jalas and Suominen 1972 –1994, Lid and Lid 1994 ). The other two species with strong accessibility relationships are the forest herbs Corydalis cava and Ranunculus lanuginosus ( Table 2 ). Although our results suggest that most of the study region would be suitable for these species in terms of climate and soil ( Fig. 4 ), they have been much less planted and, therefore, have not had many opportunities to naturalize beyond their native range. However, Corydalis cava is reported to naturalize in woods and hedges in England and Wales ( Stace 1997 ) as well as in the Netherlands, west of its native range ( Jalas and Suominen 1972 –1994). Traditionally, climate is emphasized as the main control over large‐scale species ranges ( Pearson and Dawson 2003 ). While the present study confirms that climate is an important range constraint, it also provides evidence that the ranges of many widespread northern‐nemoral forest plants probably still are moderately to strongly limited by postglacial migrational lag, as suggested by the postglacial migrational lag hypothesis. This conclusion is consistent with several prior analyses of species distribution and diversity patterns, bioclimatic and dispersal modeling, and observations of extensive naturalization beyond native ranges, which have suggested that many temperate tree and forest herb species are still expanding from their ice age refuges and have strongly dispersal‐limited ranges in Europe ( Skov and Svenning 2004 , Svenning and Skov 2004, 2005, 2007a,b ). A recently reported transplant study also provides strong evidence that dispersal may limit broad‐scale ranges of forest plant species in north central Europe ( Van der Veken et al. 2007 ). Postglacial migrational lag appear to affect other organism groups as well, notably amphibians and reptiles ( Araújo and Pearson 2005 , Araújo et al. 2008 ). The support for the postglacial migrational lag hypothesis provided by the present study is especially noteworthy given its focus on widespread, northern species, i.e. the species least likely to be limited by dispersal ( Svenning and Skov 2007a ). The previous support for the postglacial migrational lag hypothesis have primarily concerned southern species with small ranges, i.e. species still associated with their ice age refugia ( Svenning and Skov 2004, 2007a , Araújo et al. 2008 ). Our results underscore the need to reconsider the equilibrium postulate as a basis for calibrating predictive species distribution models ( Guisan and Thuiller 2005 , Guisan et al. 2006 , Giesecke et al. 2007 ). The combined importance of climate and migrational lag as range controls suggests that although species ranges are influenced by climate, we cannot expect most forest plant species to closely track the expected climatic changes. Bioclimatic envelope studies indicate that strong dispersal limitation will exacerbate plant species range losses in Europe ( Skov and Svenning 2004 , Thuiller et al. 2005 ). This has even been shown specifically for northern nemoral forest plants and particularly so with respect to northern Europe, where, given efficient dispersal, climate change could allow some degree of compensatory range expansion ( Svenning and Skov 2006 ). As the reduced and fragmented status of European forests will strongly reduce the migration rates of forest plants ( Honnay et al. 2002 ), the limited ability of many plant species to track the changing climate will pose one of the greatest challenges to European nature conservation in the 21st century. Acknowledgements We thank the Atlas Florae Europaeae project for access to the distribution data. We also thank David Viner from the Climatic Research Unit, Univ. of East Anglia for allowing use of their climate data sets, and the Danish Natural Science Research Council for economic support (grant #21–04–0346 to JCS).

Journal

Ecography – Wiley

Published: Jun 1, 2008

There are no references for this article.