*Correspondence: Linda J. Beaumont, Laboratoire d’Ecologie Alpine, UMR CNRS 5553, Université J. Fourier, BP 53, 38041 Grenoble Cedex 9, France. E-mail: email@example.com
Aim We explore the impact of calibrating ecological niche models (ENMs) using (1) native range (NR) data versus (2) entire range (ER) data (native and invasive) on projections of current and future distributions of three Hieracium species.
Location H. aurantiacum, H. murorum and H. pilosella are native to Europe and invasive in Australia, New Zealand and North America.
Methods Differences among the native and invasive realized climatic niches of each species were quantified. Eight ENMs in BIOMOD were calibrated with (1) NR and (2) ER data. Current European, North American and Australian distributions were projected. Future Australian distributions were modelled using four climate change scenarios for 2030.
Results The invasive climatic niche of H. murorum is primarily a subset of that expressed in its native range. Invasive populations of H. aurantiacum and H. pilosella occupy different climatic niches to those realized in their native ranges. Furthermore, geographically separate invasive populations of these two species have distinct climatic niches. ENMs calibrated on the realized niche of native regions projected smaller distributions than models incorporating data from species’ entire ranges, and failed to correctly predict many known invasive populations. Under future climate scenarios, projected distributions decreased by similar percentages, regardless of the data used to calibrate ENMs; however, the overall sizes of projected distributions varied substantially.
Main conclusions This study provides quantitative evidence that invasive populations of Hieracium species can occur in areas with different climatic conditions than experienced in their native ranges. For these, and similar species, calibration of ENMs based on NR data only will misrepresent their potential invasive distribution. These errors will propagate when estimating climate change impacts. Thus, incorporating data from species’ entire distributions may result in a more thorough assessment of current and future ranges, and provides a closer approximation of the elusive fundamental niche.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
There are several reasons why the realized niche of a species may not be conserved spatially or temporally. Historic and geographical constraints may prevent a species from occupying its entire fundamental niche. An example is provided by the New Zealand mudsnail, which is invasive in Australia and North America. As the mudsnail originated from a small geographical region in New Zealand, models formulated on only the data from its native range failed to adequately predict known invasions in Australia and North America, i.e. the climatic range of the snail in invaded regions was wider than in its native New Zealand (Loo etal., 2007). In non-indigenous environments, release from enemies and competitors that may limit range margins in native regions may also result in a realized niche change (Broennimann etal., 2007; Fitzpatrick etal., 2007), and during the early stages of invasion a species may not be at equilibrium with its environment (Hulme, 2003; Welk, 2004). Thus, descriptions of the species–environment relationship will not be accurate because sampling across the entire gradient of environmental space in which the species could occur cannot take place (Heikkinen etal., 2006). While historical and geographical constraints may influence the realized niche, evolutionary changes occurring due to genetic shift or selection operating in the introduced range may also affect the fundamental niche of species (Broennimann etal., 2007; Lavergne & Molofsky, 2007). Given the potential for niche shifts, models based on data from both invaded and native ranges (i.e. closer to the species’ fundamental niche) may provide a more representative range of conditions under which the species may spread (but see Welk, 2004; Mau-Crimmins etal., 2006) (Fig. 1).
In this study we assessed shifts in the realized climatic niche occupied by native versus invasive populations of three species of perennial herbs in the genus Hieracium. We then investigated the impact of calibrating ecological niche models using data from the (1) native range (NR) only versus (2) entire range (ER) (native and invasive) on projections of current and future distributions.
The genus Hieracium (family Asteraceae) contains over 9000 species, most of which are native to temperate and montane areas of the Northern Hemisphere. Some Hieracium species are highly invasive, particularly in Australia, New Zealand and North America. The ability of Hieracium species to form dense monocultures and outcompete native species, agricultural crops and pastures in New Zealand and North America has resulted in these species being classified as emerging weeds of significance to the Australian grazing industry (Barker etal., 2006). Four Hieracium species native to Europe have naturalized in Australia: H. aurantiacum L., H. murorum L., H. pilosella L. and H. praealtum Vill. Ex Gochnat. These species may threaten tussock grasslands and tablelands in alpine and temperate regions of eastern Australia (Barker etal., 2006), and have been found in Eucalypt woodlands and alpine heathlands (Williams & Holland, 2007). Although the distributions of these four species in Australia are currently restricted, the potential threat posed by their presence has resulted in them being placed on the National Alert List of invasive species (Brinkley & Bomford, 2002).
Current climate data were obtained from WorldClim (version 1.4) (Hijmans etal., 2005) and included annual precipitation (AP), mean temperature of the coldest quarter (MTCQ), maximum temperature of the hottest month (MTHM), mean diurnal range (MDR), precipitation of the driest month (PDM) and precipitation seasonality (PS). These variables were selected as they provide an uncorrelated mix of means, extremes and seasonality. Data were downloaded at a spatial resolution of 10 min then re-projected to an equal area grid of 10 km2 using a bilinear interpolation in ArcGIS (version 9.1) (ESRI, 2006). Climate scenarios for Australia were extracted from OzClim (version 2.0, http://www.cmar.csiro.au/ozclim/index.html) for the reference year 2030. To account for some of the uncertainty in future climates we obtained data from two climate models (CSIRO Mk2.0 and NCAR CCSM 3) as well as two emissions scenarios (A1F with high climate sensitivity and B1 with low climate sensitivity). The climate models were selected because (1) objective analyses of CSIRO MK2.0 suggested that it provides good simulations of current climate across Australia (A. Pitman, personal communication.) and (2) simulations from NCAR are commonly used in impacts assessments and provide an independent source of data to CSIRO MK2.0. The B1 storyline describes a convergent world with a global population that peaks in mid-century and declines thereafter but with rapid changes in economic structures toward a service and information economy and the introduction of clean and resource-efficient technologies. Cumulative CO2 emissions from 1990 to 2030 total 369.9 GtC, and human population is set at 8.196 billion. The A1 storyline describes a future world of very rapid economic growth, global population structure the same as the B1 storyline and the rapid introduction of new and more efficient technologies. The A1F scenario focuses on maximum fossil fuel use. By 2030 human population is estimated to be 8.122 billion with cumulative CO2 emissions from 1990 totalling 420.9 GtC (Nakicenovic etal., 2000). CSIRO Mk2 and NCAR estimate an increase in Australia's annual mean temperature of 0.48 and 0.47 °C, respectively, under the B1 low climate sensitivity scenario by 2030, and 1.00 and 0.97 °C under the A1F high climate sensitivity scenario.
Distribution records of the native and invasive ranges of H. aurantiacum, H. murorum and H. pilosella were downloaded from a variety of online databases, including the Global Biodiversity Information Facility (GBIF, http://www.gbif.org/), Australia's Virtual Herbarium (AVH, http://www.flora.sa.gov.au/avh/), Arctos (http://arctos.database.museum/home.cfm), Eunis (http://eunis.eea.europa.eu/) and E-Flora BC (http://www.eflora.bc.ca/). Insufficient distribution records were available to model H. praealtum. The Alexandria Digital Library (http://www.alexandria.ucsb.edu/) was used to determine, where possible, the latitude and longitude of records that were missing coordinate data. Records from several countries known to contain the three Hieracium species were excluded as coordinates could not be determined (H. aurantiacum: Romania; H. murorum: Slovenia, Romania, Portugal; H. pilosella: Croatia). In total, the data bases yielded 2684, 11237 and 29404 records with coordinate data for H. aurantiacum, H. murorum and H. pilosella, respectively. To reduce the impact of spatial autocorrelation we re-projected location records to a resolution of 10 km2 and removed duplicate records. Similarly, as the majority of records came from Austria, France and Germany, we randomly reduced these by 50%, thereby decreasing the potential bias that may occur as a result of these areas being over-represented in the model calibration. A total of 1088, 1281 and 2721 records for H. aurantiacum, H. murorum and H. pilosella remained, representing the entire range (ER) of each species. Of these, 505, 33 and 285, respectively, represented invasive populations.
Locations where species are known not to occur are not included in the data bases examined. As absence data were required for the ecological niche models, pseudoabsences were generated. To avoid the selection of false absences (i.e. locations that lie within the climatic tolerance of the species but where the species has not been recorded because of insufficient surveying, dispersal limitations or biotic constraints) we selected pseudoabsences from sites that lay outside the species’ known climatic envelope (Austin & Meyers, 1996; LeMaitre etal., 2008), as defined by the six climate variables; AP, MTCQ, MTH, MDR, PDM and PS. The geographical distribution of each species’ climatic envelope was mapped in ArcGIS, and locations outside of this region were randomly selected as pseudoabsences. The number of pseudoabsences selected equalled the number of presence records for each species multiplied by 10. Two sets of pseudoabsences were created for each species based on location data from the (1) native ranges only and (2) species’ entire ranges. To ensure that results were not dependent on a single selection of pseudoabsences, we assessed the accuracy of the SDMs against 10 alternative sets of pseudoabsences.
Climatic niches of native and invasive ranges
Following Broennimann etal. (2007), we used principal component analysis (PCA) in the R CRAN library ‘ade4’ to quantify the positions of native and invasive climate niches. The magnitude and statistical significance of differences between the native and invasive occurrence clouds on the PCA graph were assessed using a between class analysis, which yielded a between class inertia ratio. The significance of this ratio was further tested by conducting 99 Monte Carlo randomizations.
Ecological niche models
Alternative ecological niche models have different levels of accuracy under different circumstances (Elith etal., 2006) and there is no single ‘best’ method. Using a single modelling technique does not provide any information as to whether that method gives the best predictive accuracy for the data set used (Araújo & New, 2007). Thus, multiple ENMs can be used to create an ensemble of simulations enabling a range of projections to be explored (Thuiller, 2004; Araújo & New, 2007). To model species’ distributions, we used eight statistical techniques available in the BIOMOD package (version 2006.01.26) developed by one of the authors (Thuiller, 2003; Thuiller, 2004; Araújo & New, 2007). These consisted of two regression methods (GAM: generalized additive model, GLM: generalized linear model), three machine learning methods (ANN: artificial neural networks, GBM: generalized boosted model, RF: random forests), two classification methods (CTA: classification tree analysis, MDA: mixture discriminant analysis) and a surface range envelope (SRE). Models were calibrated with data from (1) the realized niche of each species’ entire range (i.e. native and invaded – ER) and (2) the realized niche of native ranges only (NR), as defined by the six climatic variables. Presence/pseudoabsence data were randomly divided into two subsets with 70% of records used to calibrate the models and the remaining 30% of records used to evaluate the predictive power of each model using the area under the receiver-operating characteristic (ROC) curve and Cohen's kappa statistic. Probabilities of occurrence derived from all models (except SRE) were transformed into presence/absence data using the area under the ROC curve (AUC). The calculation of ROC curves traditionally requires both presence and absence data. If true absence data are unavailable and pseudoabsence or background points are used in their place, then the maximum achievable AUC is less than 1 (Wiley etal., 2003; Phillips etal., 2006). We applied a conservative classification where AUC < 0.8 is a poor or null model; 0.8 < AUC < 0.9 is fair; 0.9 < AUC < 0.95 is good and 0.95 < AUC < 1.0 is very good (Thuiller etal., 2006). Models with a kappa < 0.6 were classified as poor or null; 0.6 < kappa < 0.8 as fair; 0.8 < kappa < 0.9 as good and > 0.9 as very good.
Models calibrated with (1) NR data and (2) ER data were projected onto current European, North American and Australian climate surfaces, as well as future Australian climate surfaces.
A ‘frequency histogram’ approach, with models being given equal weights, was used to create ensembles of model projections (see Marmion etal., 2009) using ArcGIS. Thus, areas classified as suitable by the majority of the models can be identified.
Climatic niche shifts across native and invasive ranges
A principal components analysis was conducted to reduce the dimensionality of each species’ realized climatic niche. The first two axes of the PCA were primarily associated with temperature and moisture availability, respectively. Together these axes explained 65–69% of variation in the climatic niches. The PCA graphs for each species show that the niche centroids of native and invasive populations differ strongly (between class inertia ratio of 36%, P < 0.01, for H. aurantiacum; 5%, P < 0.01 for H. murorum; 19%, P < 0.01 for H. pilosella) (Fig. 2). For H. murorum, the climatic niche of invasive populations is primarily a subset of that of its native range. In contrast, the climatic niches of invasive populations of H. aurantiacum and H. pilosella differed substantially from their native populations. For each species, a number of invasive populations occurred in conditions for which there is no current climate analogue in Europe.
Sufficient location data were available to compare invasive populations of H. aurantiacum and H. pilosella located in western North America, eastern North America, and Australia and New Zealand. As shown in Fig. 2, the climatic space occupied by the two species across these invasive regions differs: populations in eastern North America realize a climatic niche that is mostly outside that of native populations. For H. pilosella, this represents a shift to wetter conditions while for H. aurantiacum, the shift is to warmer, wetter conditions. Populations of H. aurantiacum in western North America have the broadest climatic niche, and generally occur in warmer conditions than native populations. In contrast, western North America populations of H. pilosella occupy a subset of the native range climatic niche. Populations in Australia and New Zealand occupy a climatic niche more similar to their native ranges than to those of North American populations.
Current range projections
The accuracy of the eight ENMs was assessed using both threshold-dependent (kappa) and threshold-independent methods (AUC). Model accuracy was consistently high, with most AUC and kappa scores exceeding 0.95 and 0.9, respectively (exceptions were ANN and MDA which scored ‘good’ in several runs). These results were irrespective of the selection of pseudoabsence points. Across 10 alternative selections of pseudoabsence points, the standard deviation of the AUC scores was typically less than 0.01 (exceptions were ANN and RF where the standard deviation was < 0.03). More variation occurred across the kappa scores with standard deviations typically less than 0.05, although this increased up to 0.09 for ANN.
The inclusion of invasive range data had little impact on the overall size of the projected European distributions for the three Hieracium species: differences primarily lay in greater consensus achieved by the eight ER models (Fig. 3). This was particularly so for H. aurantiacum with few NR models predicting areas around Romania as suitable. Projected distributions for all three species included large regions of eastern Europe for which few records exist.
In contrast to native ranges, projected North American and Australian distributions for the three species varied substantially. NR models projected substantially narrower ranges that, for the most part, failed to capture the known distributions of these species (Fig. 3). Invasive populations were correctly predicted by a minimum of two ENMs calibrated with ER data, with 78% (H. aurantiacum), 16% (H. murorum) and 79% (H. pilosella) of these populations predicted by all eight ENMs (Fig. 4). Conversely, 64% and 23% of invasive populations of H. aurantiacum and H. pilosella failed to be correctly predicted by any of the NR models.
Projections of Australian ranges under future climates
The consensus across ENMs was for a decrease in the Australian distributions of the three Hieracium species in response to climate change. For ER models, averaged across the four climate scenarios and the eight ENMs, distributions for H. aurantiacum, H. murorum and H. pilosella were projected to decrease 15%, 9% and 9%, respectively. For NR models they decreased by 11%, 10% and 10% (Fig. 5). Thus, the use of NR or ER data to calibrate models had little impact on the proportion by which ranges were projected to decrease. As with current distributions, however, substantial differences occurred in the overall size of projected distributions, with NR models predicting very limited ranges (Fig. 6). For example, areas identified as suitable by at least five of the eight NR models were 93%, 50% and 47% smaller than when ER models were used, for H. aurantiacum, H. pilosella and H. murorum.
Niche change across native and invasive ranges
Three species of Hieracium, native to Europe and invasive in Australia, New Zealand and North America, occupy different climatic niches in their invasive versus native ranges. Climatic niches of invasive populations also vary across broad geographical regions. Furthermore, numerous invasive populations occur in areas for which there is no climate analogue in Europe (Fig. 2). Two general hypotheses can be put forward to explain these results: barriers to the native range may not be present in the invaded range, and/or the climatic tolerances of invasive populations may have changed as a result of local adaptation, genetic drift or phenotypic plasticity.
In their native range, biotic constraints (such as competitors, herbivores, parasites and pathogens), historical and land use legacies, dispersal limitations and geographical barriers may prevent species from occupying larger climatic ranges. The release that occurs due to the absence of these constraints in novel regions may enable species to occupy environments different to their native ranges. Release from enemies and disease are two of several reasons suggested for the differences in the climatic niche occupied by fire ants in North America compared to their native range in South America (Fitzpatrick etal., 2007). Several specialist insect taxa have been recorded on some Hawkweed species in Europe and may cause significant damage (Sárospataki, 1999). However, few specialists appear to feed on H. pilosella in its native range (Sárospataki, 1999). Similarly, although a variety of generalist herbivores are found on Hawkweeds in New Zealand, none of these cause serious damage, and specialist taxa do not occur on these plants (Syrett & Smith, 1998). In the Nordic countries, geographical barriers to the migration of numerous Hieracium have been identified (Tyler, 2000); however, there is generally little known about the factors limiting range margins of Hieracium in other regions.
Successful invaders may expand their range into varying habitats by undergoing local adaptations or as a result of genetic drift (Callaway & Ridenour, 2004; Lavergne & Molofsky, 2007). The genus Hieracium is extremely complex, resulting from recent speciation, hybridization, polyploidy and diverse reproductive strategies (Trewick etal., 2004). Populations of H. pilosella in New Zealand have a high degree of genetic and genome size variation as a result of interspecific hybridization. The polymorphic nature of this species has resulted in a wide range of morphotypes and its invasive success in New Zealand has been attributed to its genetic variability (Morgan-Richards etal., 2004). In contrast to H. pilosella, the genetic diversity of H. aurantiacum throughout its invasive range in North America and its native range is very small; this species has an unusually high proportion of asexual offspring and Loomis (2007) asserted that global populations of H. aurantiacum may be clones. The ability of this species to maintain populations under various conditions may be the result of phenotypic plasticity (Loomis, 2007) rather than from rapid adaptation. Low levels of genetic variability also limit local adaptation within populations of Verbascum thapsus, an invasive perennial of North America. The success of this species across a wide environmental gradient (from mediterranean to alpine climates) is attributed to phenotypic plasticity (Parker etal., 2003).
While various hypotheses can be invoked to explain niche change among the three Hieracium species, an alternative explanation is that these shifts are simply an artefact of the accuracy and availability of species location data, and the scale of our study. Our study assumed that the location data used are representative of the actual distribution of each species. Similarly, we assumed that all invasive populations have been recorded. Violation of either of these assumptions will clearly influence our results. Apparent niche shifts may also be scale dependent, and it is unknown to what extent the microclimate of invasive populations differs from that of the coarse (10 min) scale used in this study. However, we believe that a change in the realized niche is the most likely explanation, as there are several examples of invasive species confined to certain habitats due to land use or competitors in their native range (i.e. fire ants (Fitzpatrick etal., 2007), ice fig, W. Thuiller, personal observation). These species thus occupy a relatively small amount of their physiological fundamental niche, and have become highly aggressive invaders when introduced to new habitat types, released from native range pathogens or competitors, and/or further distributed by humans. The ice fig (Carpobrotus edulis) is, for instance, confined to natural areas in the Cape Floristic Region in South Africa but has spread dramatically around the globe principally due to its use by humans for horticulture.
In a more pragmatic sense, such realized niche change offers promising avenues to investigate the true potential eco-physiological limits of species without resorting to transplant experiments or experimental manipulations. In theory, the sum of all these expressed realized niches should provide the closest possible determination of the fundamental niche of the species.
Change in realized niche and ecological niche modelling
The impact of using a subset of the realized niche to model species distributions is demonstrated clearly by projections derived from Species Range Envelope (SRE) model. This method, commonly known as BIOCLIM, interpolates a species’ bioclimatic envelope, which is a summary of the climate at locations from where the species has been recorded. The SRE model describes the climatic envelope of a species as a rectilinear volume, that is, it suggests that a species can tolerate locations where values of all climatic parameters fit within the extreme values of its climate envelope. Thus, if the entire climatic envelope is not captured (e.g. if NR data only are used in model building rather than ER data), then the predicted distribution will not cover all locations that are climatically suitable for the species (Fig. 1): distributions for the three Hieracium species predicted by SRE models were substantially smaller when NR data only were used to calibrate models.
While we have advocated using ER data to model the global and Australian distributions of these species, H. pilosella provides an interesting example where the climatic niches of invasive populations in western and eastern North America do not overlap (Fig. 2). What data should be used to project invasive distributions in these two regions? Without experimentation or reciprocal transplants to determine fitness of individuals from the eastern populations grown under conditions analogous to western North America, or vice versa, it is unclear whether ER data or a subset of this would produce the most reasonable distribution models. However, given the genetic variability of this species in New Zealand, the use of ER data would provide a ‘worst case’ scenario.
By 2030, climatically suitable habitat for three species of Hieracium in Australia is estimated to decrease. Although similar decreases in the ranges of the Hieracium species were predicted regardless of the location data used, distributions obtained from NR models were 47–93% smaller than those from ER models: NR models suggest that climatically suitable habitat for H. aurantiacum on the Australian mainland will almost completely disappear within the next few decades. Thus, consistent with Broennimann & Guisan (2008), we found that underestimations or misrepresentations of species’ current ranges due to different expressions of realized niches will have cascading impacts on projections of future distributions.
For all three Hieracium species, range decreases resulted from a contraction of range margins to higher elevations or latitudes. Current potential distributions of these species already incorporates the highest elevation areas in Australia, and future increases in temperature and decreases in precipitation in the south-east would be expected to result in further contractions. However, these species are yet to realize their full geographical range in Australia and substantial climatically suitable habitat remains currently available.
Intermodel comparisons and ensembles of model projections
Modelling species distributions requires the selection of a technique(s) from an increasing number of alternatives (Elith etal., 2006). Although it is generally accepted that presence/absence models result in more accurate projections compared to presence-only models (Brotons etal., 2004), there is no single technique that is consistently superior to others (Elith etal., 2006; Heikkinen etal., 2006; Pearce & Boyce, 2006). To an extent, the accuracy of different models may vary with the characteristics of individual species (Thuiller, 2003; Segurado & Araújo, 2004; Luoto etal., 2005), and small differences between current distributions derived from alternative models can be magnified when projecting ranges under future climates (Thuiller, 2003). Several authors have suggested projecting species distributions using more than one ENM, allowing uncertainty among models to be addressed (Araújo etal., 2005; Thuiller etal., 2005; Lawler etal., 2006; Araújo & New, 2007). This is analogous to using multiple climate change scenarios to explore the uncertainty in future climates. In this study, although there was a general agreement in the projections derived from CTA, GAM and GLM, projections of other models diverged from each other, and the ability of models to extrapolate was also species-specific. Furthermore, there was a tendency for some models to predict larger distributions than others (e.g. ANN and RF). Thus, consistent with previous studies (Thuiller, 2004; Lawler etal., 2006), we found substantial differences among projections of current and future distributions made by the various ENMs we used, highlighting the importance of using more than one technique to model species’ distributions.
While the AUC and kappa values of the eight SDMs ranged from ‘good’ to ‘excellent’, this may be an artefact of our method of generating pseudoabsences where we deliberately included areas that were climatically distant from the presence locations. This approach would increase the rate of well-predicted absences and result in higher AUC scores: the more climatically distant the absences, the better they will be predicted with a bad model (Lobo etal., 2007). The appropriateness of using AUC scores to assess the accuracy of species distribution models has recently been questioned (Lobo etal., 2007) and improved methods of model evaluation have been called for (Araújo & Guisan, 2006). However, to the best of our knowledge, no alternative measures are available and a plethora of papers have conversely championed these accuracy measures (e.g. Fielding & Bell, 1997; Pearce & Ferrier, 2000).
There is a growing body of literature demonstrating that some species occupy different climatic niches when they move into novel regions. Failure to take this into account when projecting distributions can result in underestimation of potential invasive ranges. This in turn can affect forecasts of future range changes and has clear implications for invasive species management. We suggest that the realized climatic niche of both native and invasive ranges should be explored before modelling. Furthermore, care must be taken when exploring the potential for the spread of species in invasive regions, as these species may not be at equilibrium with climate and niches may continue to expand. If additional areas are invaded, ENMs should be refined to include this new information. Our results also suggest that modelling non-invasive species’ distributions based on a subset of location records may result in similar under-representations of current and future distributions.
Finally, we suggest that further investigations be conducted to measure and sum the expressed realized climatic niche of notorious invasive species, thus identifying, as closely as possible, their fundamental niche. We may discover that the invading capacity of species is simply a reflection of the width of their fundamental niche. This could help to explain the failure of traditional comparative analyses to give a functional prototype of the perfect invader.
The OzClim model was jointly developed by the International Global Change Institute (IGCI), University of Waikato and CSIRO Atmospheric Research. This research was supported by the Australian Research Council, the NSW Department of Environment and Climate Change and Macquarie University. WT received support from European Commission's FP6 MACIS (Minimization of and Adaptation to Climate change Impacts on Biodiversity, No. 044399) and ECOCHANGE (Challenges in Assessing and Forecasting Biodiversity and Ecosystem Changes in Europe, No. 066866 GOCE) projects. We thank three anonymous reviewers for their constructive comments.