Change Global Biology (2008) 14, 110, doi: 10.1111/j.1365-2486.2007.01461.x
Prediction of the distribution of Arctic-nesting pink-footed geese under a warmer climate scenario
R I K K E A . J E N S E N *, J E S P E R M A D S E N *, M A R K O C O N N E L L w 1, M A R Y S . W I S Z *, H A N S T M M E R V I K z and F R I D T J O F M E H L U M 2 *Department of Arctic Environment, National Environmental Research Institute, University of Aarhus, Frederiksborgvej 399, PO Box 358, DK-4000 Roskilde, Denmark, wDepartment of Biology and Biochemistry, University of Bath, Claverton Down, Bath, BA2 7AY, zDivision of Arctic Ecology, Norwegian Institute for Nature Research, The Polar Environmental Centre, N-9296 Troms, Norway, Natural History Museum, University of Oslo, PO Box 1172, Blindern, N-0318 Oslo, Norway
Abstract
Global climate change is expected to shift species ranges polewards, with a risk of range contractions and population declines of especially high-Arctic species. We built species distribution models for Svalbard-nesting pink-footed geese to relate their occurrence to environmental and climatic variables, and used the models to predict their distribution under a warmer climate scenario. The most parsimonious model included mean May temperature, the number of frost-free months and the proportion of moist and wet mossdominated vegetation in the area. The two climate variables are indicators for whether geese can physiologically full the breeding cycle or not and the moss vegetation is an indicator of suitable feeding conditions. Projections of the distribution to warmer climate scenarios propose a large north- and eastward expansion of the potential breeding range on Svalbard even at modest temperature increases (1 and 2 1C increase in summer temperature, respectively). Contrary to recent suggestions regarding future distributions of Arctic wildlife, we predict that warming may lead to a further growth in population size of, at least some, Arctic breeding geese.
Keywords: Anser brachyrhynchus, Arctic, biodiversity, climate change, climate envelope, species distribution models, Svalbard
Received 4 October 2006; revised version received 21 February 2007 and accepted 9 July 2007
Introduction
Global climate change is predicted to have strong effects on the distribution and abundance of Arctic animal and plant populations. Ranges of individual species may move polewards, expand or decline in extent, and in mountain areas, move towards higher elevations (Boyd & Madsen, 1997; Parmesan et al., 1999; Thomas & Lennon, 1999; Hickling et al., 2006). Particularly in the high-Arctic regions and islands where the range of species is limited by the Arctic ocean, it is predicted that effects will be mostly negative because the habitats
Correspondence: Rikke A. Jensen, tel. [phone], fax [phone], e-mail: [e-mail]
1 Present address: Hartpury College, University of West England, GL19 3BE Gloucestershire, UK.
Present address: Research Council of Norway, PO Box 2700, St Hanshaugen, N-0131 Oslo, Norway.
2
of tundra living species will be squeezed by higher vegetation (Zockler & Lysenko, 2000; ACIA, 2005). Migratory, Arctic-nesting birds have a narrow time window for breeding, moulting and preparation for return migration between the time of thaw and before the Arctic winter sets in. The northern limits of breeding range are largely determined by the minimum period physiologically required to complete the breeding cycle (e.g. Newton, 1977), provided that suitable habitat is present. In Svalbard, nesting goose species (i.e. the brent goose, Branta bernicla, barnacle goose, Branta leucopsis and pink-footed goose, Anser brachyrhynchus), arrive late May to early June and migrate south around mid September, a period of o4 months which coincides with the time of frost- free and snow-free conditions (Prop & De Vries, 1993; Madsen et al., 1998). Their timing and success of breeding are highly variable, depending on snow and ice conditions on arrival, summer and premigratory weather conditions (Owen & Black, 1989; Prop & De Vries, 1993; Madsen et al.,
1
r 2007 The Authors Journal compilation r 2007 Blackwell Publishing Ltd
2 R . A . J E N S E N et al.
2007). The size of all three Svalbard goose populations has increased in recent decades. In barnacle and pinkfooted geese, the proportions of successful breeding pairs and brood sizes have declined with increasing population sizes (Trinder et al., 2005; M. Trinder & J. Madsen, unpublished data), suggesting that densitydependent factors are now affecting their productivity. In barnacle geese, the available area for brood rearing appears to be a limiting factor (Drent et al., 1998). In pink-footed geese, it is more likely that availability of suitable nest sites is limiting (Madsen et al., 2007), possibly in combination with availability of spring staging feeding habitat in which the geese gain body reserves of critical importance for the subsequent breeding success. The growing goose populations give rise to management concerns in their wintering range due to conicts with agriculture (e.g. Van Eerden et al., 1996) and, increasingly, in the Arctic due to potential grazing effects on the fragile tundra ecosystems (e.g. Loonen & Solheim, 1998; Abraham et al., 2005; Van der Wal et al., 2007). Therefore, to inform management about the expected future directions of these conicts, it is important to assess the impacts of warming of the Arctic. We examine whether warming will have a pronounced effect on the potential breeding range of the geese in the Svalbard archipelago compared with their present distribution, focussing on the pink-footed goose which has the widest distribution of the three species. A landscape based nesting habitat suitability model (resolution 15 m) for a central part of its breeding range showed that pink-footed geese prefer to nest on south facing slopes in the lowland, in close connection to suitable feeding sites which are wet moss-dominated areas (Wisz et al., in press). Upscaling the model to cover entire Svalbard (resolution 1 km), we hypothesize that at the regional scale the present distribution of pink-footed geese is determined by (1) the length of the season with frost-free conditions, which sets the limit for whether geese can physiologically full the breeding cycle or not, (2) the temperature in May which indicates availability of areas providing geese with early feeding and nesting opportunities, (3) suitable feeding habitats and (4) elevation. Based on the present potential distribution, we predict future distributions of pink-footed geese under a warmer climate scenario. gium, with autumn and spring stopover sites in Norway. The population has increased from approximately 15 000 individuals in the 1960s to more than 50 000 in 2003 (Fox et al., 2005). The pink-footed goose breeds in loosely aggregated colonies, mainly along the west coast and in the interior lowlands of the western parts of Svalbard. It breeds on small islands, as well as on the open tundra, being capable of defending its nest against avian predators, as well as Arctic foxes, Alopex lagopus (Mehlum, 1998).
Spatial data layers
For modelling the potential distribution of nesting pinkfooted geese, we used known presence of nests, environmental and climatic predictors. Our sample units consist of grid cells of 926.6 m by 926.6 m. For simplicity we refer to these as grid cells of 1 km2. As large parts of Svalbard is covered with glacier (approximately 55 000 km2) only cells with no glacier were considered in the analysis. This area consists of 10 498 cells. Areas disturbed by human activity are included in the analysis, but their extent is negligible compared with the total area. Nest records. Data were derived from: (1) The database of the Norwegian Polar Institute with records from 1962 to 1996. Only data with geographical coordinates and records of nests were used, whereas records of broods were not used. The accuracy of the coordinates is generally within 1 km2. (2) Recent nest surveys in different parts of Nordenskio ldland: (a) Sassendalen in the interior of Isfjorden, 20032004 (Wisz et al., in press); (b) Nordenskio ldkysten, 2004 (J. Prop, unpublished data), (c) Reindalen, 2004 (J. U. Jepsen, unpublished data); (d) Varsolbukta, 2004 (C. Hubner, unpublished data) and (e) Adventdalen, 2004 (D. Kuijper, unpublished data) (Fig. 1). In the recent studies, nest positions were recorded with a GPS with an accuracy of c. 50 m. From these sources 692 nest records and accurate coordinates could be derived. After relating these to a 1 km2 grid and assigning one presence record for each cell containing one or more nests, 111 records of nest presences were available for analysis. Vegetation. A vegetation cover map with four major classes relevant for geese was developed for entire Svalbard, based on Landsat Thematic Mapper with a 28 m spatial resolution (H. Tmmervik, in preparation, see Wisz et al., in press). The relevant classes are (1) dry heath dominated by Dryas octopetala, Cassiope tetragona and Carex spp., (2) bare ground with sparsely vegetated patches dominated by Saxifraga oppositifolia, (3) moist
Materials and methods
Study population
The Svalbard-breeding population of the pink-footed goose winters in Denmark, the Netherlands and Bel-
r 2007 The Authors Journal compilation r 2007 Blackwell Publishing Ltd, Global Change Biology, 14, 110
C L I M AT E C H A N G E E F F E C T S O N A R C T I C G O O S E D I S T R I B U T I O N S
3
Surface temperature. We used MODIS satellite imageryderived land surface temperature and emissivity (Friedl et al., 2002; Petitcollin & Vermote, 2002) at a spatial resolution of 1 km2 grid cells. We chose a temporal resolution averaging values across the middle 8 days of each month. Monthly temperature values for the years 20012004 were obtained, and subsequently reprojected using bespoke MODIS data manipulation tools (USGS EROS Data Center, 2002). Pixel values were averaged across all available years (two to four) to create a single mean monthly value (to nearest 0.5 1C) for each 1 km2 grid cell over the entire area of Svalbard (Fig. 2). Thus, a frost-free month is dened as a month where the average temperature across the middle 8 days is above 0 1C. The cells with three frost-free months are concentrated along the west coast, in the valleys that debouch into the western ords and in the western lowlands of Edgeya. Nordenskioldland has some of the largest continuous areas with three or four frost-free months. Nest presences were observed in cells with a mean May temperature ranging from 8.87 to 1.59 1C, and in areas with two to four frost-free
Fig. 1 Elevation map of Svalbard with pink-footed goose nest presences shown by red dots. Localities where detailed nest surveys were carried out: (a) Sasssendalen, (b) Adventdalen, (c) Nordenskioldkysten, (d) Varsolbukta, (E) Reindalen.
moss dominated fen with mixed coverage of Bistorta vivipara, Salix polaris, Equisetum spp., Eriophorum spp. and Carex spp. and (4) wet moss carpet, dominated by mosses and Dupontia spp. For the purpose of the present analysis, we only used the area of moist- and wet moss-dominated habitat, expressed as proportion for each 1 km2 grid cell, which is the preferred feeding habitat during the prenesting and nesting period and a signicant predictor in the landscape nest site model (Wisz et al., in press). Nest presences were observed at proportions of moist/wet moss in a range from 0.0 to 0.85.
Elevation. A digital elevation model DEM with a 20 m spatial resolution was made available for the project from the Norwegian Polar Institute (Fig. 1). From this map mean elevation was calculated in the coarser resolution of 1 km2 grid cells. Elevation was a signicant predictor in the landscape nest site model (Wisz et al., in press). Nest presences were observed at elevations ranging from 0 to 632 m.
Fig. 2
Current surface map of frost-free months on Svalbard.
r 2007 The Authors Journal compilation r 2007 Blackwell Publishing Ltd, Global Change Biology, 14, 110
4 R . A . J E N S E N et al.
months. Very few nest records were recorded in areas with two frost-free months. Future climate scenarios. Future distributions of pinkfooted geese are predicted using an increase of the mean May temperature of 1 and 2 1C, respectively, and an increase of the frost-free period derived from a scenario where the mean temperatures of May, June, July, August and September, are elevated by 1 and 2 1C, respectively. These scenarios are supported by the temperature increase predicted by Frland et al. (2004), who downscaled a temperature scenario (based on ECHAM4/OPYC3 AOGCM with the transient GSDIO integration), predicting that in 2050 the mean summer temperatures at Svalbard airport in the central part of Isfjorden will have increased approximately 1 1C compared with the 19611990 normal period and that this will cause an increase of 2 months for the frost-free period. Our predictions for the warmer climate scenarios do not take into account that glaciers are likely to decrease in size and expose new potential nest sites. sample of sites with true absences but as a sample of all sites potentially available to nesting geese in Svalbard. Because of the random procedure, pseudo-absences may coincide with observed presences. Pseudo-absence records along with presence records have been widely used as a reliable surrogate for true presence/absence data (e.g. Ferrier et al., 2002; Elith et al., 2006). We generated 10 times more pseudo-absence records than the number of presences in order to ensure sufcient landscape variety in the pseudo-absences. To ensure that the presences and pseudo-absences contributed equally to the model and to avoid bias towards the larger sample (Sokal & Rohlf, 1995), we weighted each pseudo-absence as 1/10 in the model. Model selection was based on Akaikes Information Criterion (AIC), by backwards model simplication, dropping terms from the full model. Smoothing parameters were also selected using AIC. To account for spatial auto-correlation in the data, we also tted a model that included all the environmental predictors and an auto-covariate. This method is recommended by Augustin et al. (1996) to avoid ination of the signicance of the selected predictors. In our study the auto-covariate was the sum of the eight nearest neighbouring cells where nests have been recorded. The inclusion of the auto-covariate did not change the selected predictors, nor did it seem to inate their signicance. Model performance was slightly better than the model without the auto-covariate, and we concluded that spatial auto-correlation did not seriously inate the signicance of the selected predictors, and we do not address the subject further in this analysis.
Modelling framework
We used generalized additive models (GAM) (Hastie & Tibshirani, 1990) to model the probability of nding pink-footed goose nests within the 1 km2 grid cells as a function of four environmental predictors (all continuous variables): number of frost-free months, mean May temperature ( 1C), proportion of moist/wet moss vegetation cover and elevation (m). Unlike regression models, GAMs do not force a parametric relationship (e.g. linear, parabolic, etc.) between the response and the predictors. Instead, GAMs implement nonparametric smoothers in regression models. GAMs have been shown to be particularly useful in modelling species distributions (e.g. Austin, 2002; Elith et al., 2006) because the smoothing functions can describe the complex nonlinear relationships often seen in ecology (e.g. Guisan & Zimmermann, 2000). The statistical analysis was performed in R 2.2.0 statistical software (http:/ / www.R-project.org), using the GRASP v2.5 library (Lehmann et al., 2002) for R. We used a binomial error family with a logit link function. Predictions to the entire Svalbard archipelago were calculated in R and illustrated on maps using ArcGIS 9.2.
Model evaluation
To assess the amount of deviance the model could explain, we used D2, dened as D2 5 (null deviance residual deviance)/null deviance. This measure is equivalent to the R2 value known from ordinary regression (Guisan & Zimmermann, 2000). In order to evaluate the overall predictive performance of our model, we examined the models ability to discriminate between occupied and unoccupied sites and the reliability with which it predicts the probability of a site being occupied. To assess our models discriminatory ability, we calculated the threshold-independent Area Under the Receiver Operating Characteristic Curve (AUC) (e.g. Pearce & Ferrier, 2000). The Receiver Operating Characteristic Curve (ROC) is plotted on a unit square as sensitivity against (1- specicity) for a range of increasing threshold values. Sensitivity is dened as the proportion of sites correctly predicted to be occupied out of
Statistical methods
GAM procedures require absences, as well as presences for inference, so we computed so-called pseudo-absences randomly from the background area not covered by glacier. The pseudo-absences were not intended as a
r 2007 The Authors Journal compilation r 2007 Blackwell Publishing Ltd, Global Change Biology, 14, 110
C L I M AT E C H A N G E E F F E C T S O N A R C T I C G O O S E D I S T R I B U T I O N S
the total number of occupied sites in the sample. Specicity is dened as the proportion of sites correctly predicted to be unoccupied out of the total number of unoccupied sites in the sample. The AUC describes the discrimination capacity ranging from 0.5 for models with no discrimination ability, to 1 for a model with perfect discrimination (Pearce & Ferrier, 2000). An AUC score between 0.8 and 0.9 indicates good discrimination capacity, and above 0.9 an excellent discrimination capacity (Thuiller et al., 2005). We assessed the predictive reliability, also known as calibration (Pearce & Ferrier, 2000), using the correlation coefcient (COR) describing the simple correlation between the observed and predicted response, as recommended by Zheng & Agresti (2000). AUC and COR was calculated from a cross-validation based on ve randomly selected subsets of the entire dataset. Each subset contained 20% of the data points. The subsets were dropped from the model one at a time, and the model was retted to the remaining 80% of the data and nally predictions were made for the omitted data points. The threshold, discriminating between predicted occupied and unoccupied sites, was set by calculating the maximum (k) (Cohen, 1960). k is a measure of agreement, considering both omission and commission errors (Elith et al., 2006) and is useful for setting a nonarbitrary threshold for the predicted response.
5
Table 1 Table of the predictors for nest distribution of the pink-footed goose in Svalbard and specications of the selected model, based on 111 nest presences and 1110 pseudoabsences Smoothing df Predictors Mean May temperature Number of frost-free months Proportion of moist and wet moss vegetation Elevation Specications Null deviance Explained deviance D2 AUC COR P
1 1 1 Not selected
0.014 o0.001 0.039
308 104 0.34 0.86 0.41
Null deviance is the total amount of deviance in the sample and explained deviance is the amount which can be accounted for by the model. D2 is the proportion of deviance explained by the model. AUC is the Area Under the Receiver Operating curve and COR the simple correlation between observed and predicted values.
Results
The most parsimonious GAM model incorporated mean May temperature, number of frost-free months, and proportion of moist and wet moss vegetation (Table 1). Elevation not was selected by the model. The probability of nest occurrence increased with number of frost-free months (Fig. 3a) and with mean May temperature (Fig. 3b). Probability of nest occurrence also increased with increasing moist and wet moss vegetation coverage (Fig. 3c). The condence intervals around the additive contribution of each predictor suggest that the increase in the response was signicant for all the selected predictors (Fig. 3df). The model had a reasonable t to the data as D2 was 0.34, indicating that we can explain 34% of the total deviance. AUC was 0.86, indicating that the model predicts higher where the species is present than where it is absent in 86% of the locations used for predictions. The model was well calibrated as the COR for the cross-validation was 0.42 (Po0.001). We calculated the maximum (k) to be 0.75. This was used as the objective threshold, when discriminating between occupied and unoccupied areas from the predictions.
The predicted suitable nesting areas, under the current conditions, exist in the lowlands on the southwest coast of Svalbard, in some of the valleys debouching into the western fjords, as well as in the western part of Edgeya in the southeast of Svalbard (Fig. 4a). The predicted distribution under the 1 1C increase in mean summer temperature scenario shows that there is a marked increase in possible nest site areas compared to the current distribution (Fig. 4b). Using the k of 0.75 as the critical threshold, a large part of the west coast is pointed out as suitable nesting area, as are several of the valleys that debouch into the western fjords. Under the 2 1C increase scenario (Fig. 4c), most of the west coast and the majority of the western valleys are predicted to be suitable for nesting. Compared with the present potential distribution, which is approximately 1950 km2, the suitable nesting area is predicted to increase by 84% and 217% under the 1 and 2 1C scenario, respectively.
Discussion
Climate has a profound effect on species distributions, and our results show that this also applies for the distribution of pink-footed geese. Their current nesting distribution appears to be limited by climatic factors, (i.e.
r 2007 The Authors Journal compilation r 2007 Blackwell Publishing Ltd, Global Change Biology, 14, 110
6 R . A . J E N S E N et al.
(a)
Probability of occurrence 0.8 0.4 0.0 10 6 2 Mean May temperature
(b)
0.8 0.4 0.0 0 1 2 3 4 Number of frost - free months
(c)
0.8 0.4 0.0 0.0 0.4 0.8
Vegetation
(d)
s (maytemp, 1)
s (post temp, 1)
2 0 2 4
2 0 2 6 2 10 Mean May temperature
s (vegetation, 1 )
4
(e)
(f)
4 2 0 0.0 0.4 0.8 Vegetation
0 1 2 3 4 Number of frost - free months
Fig. 3 Response curves from the generalized additive models (GAM) based on the 111 nest presences and 1110 pseudo-absences. (ac): Probability of nest occurrence as a function of the selected predictor variables. The functions are generated by setting the contribution of all predictors in the model to 0, except for the predictor in question. (df): GAM response curves of the probability of nest occurrence as a function of the selected predictors where the y-axis can be interpreted as a transformation of the response. The GAM curves show the additive contribution of each variable to the response. Dashed lines are twice-standard-error curves. The black markings on the x-axis show observations of the predictor in question.
the number of frost-free months combined with early snow melt), which are indicators of the physiological requirements of successfully breeding geese, and availability of suitable feeding habitat. Under scenarios of even modest warming, the range of the species is likely to increase substantially due to a north- and eastward expansion. From this analysis we cannot conclude that that range of the species will increase altitudinal because elevation is not selected as an important predictor variable in our model. Wisz et al. (in press) documented the effect of elevation on a ner scale, and we nd it likely that suitable nesting areas may become available at higher altitudes as the temperature increase. Some possible limitations of nesting at higher altitudes may, however, exist due to wind exposure and longer travelling distances from nest to brood rearing areas along the coasts, lakes and rivers, increasing the risk of predation. Although our model shows good t and discriminatory ability, additional uncertainties arise when making predictions into the future (Araujo et al., 2005). First, in predictive habitat distribution models, species distributions are assumed to be in equilibrium with the conditions present during the sampling period (Guisan & Thuiller, 2005). This requires that the environment has not undergone dramatic changes in recent time, and
thus it is a postulate that the current distribution reects the current environmental conditions. In support of the equilibrium assumption, historical records of the breeding areas of pink-footed geese indicate that the range and key nesting areas have not changed from the 1960s to the 1990s (Mehlum, 1998), in spite of a threefold increase in population size. If the distribution of the pink-footed goose is not in equilibrium, our predictions may be an underestimate of the true potential range (Guisan & Thuiller, 2005). Second, effects of predictor variables can be confounded with interactions between species and in consequence bias predictions into the future (Loehle & Leblanc, 1996). An increase in numbers of pink-footed geese may potentially lead to increased competition for resources with barnacle geese. However, the two species partly occupy different niches and show different feeding habits and food plant selectivity when occurring sympatrically (Madsen & Mortensen, 1987; Fox & Bergersen, 2005; Fox et al., 2007). Hence, we expect that competitive interactions between the two species are not likely to affect substantially the overall distribution of the two species in Svalbard. Third, predicting species distributions should be restricted to be inside the ranges of the observed predictor
r 2007 The Authors Journal compilation r 2007 Blackwell Publishing Ltd, Global Change Biology, 14, 110
C L I M AT E C H A N G E E F F E C T S O N A R C T I C G O O S E D I S T R I B U T I O N S
7
Fig. 4 Predicted nesting distribution of pink-footed geese in Svalbard: (a) at present, (b) under the 1 1 1C, and (c) under the 1 2 1C mean-summer temperature climate scenario, respectively.
r 2007 The Authors Journal compilation r 2007 Blackwell Publishing Ltd, Global Change Biology, 14, 110
8 R . A . J E N S E N et al.
variables, because information lack about responses outside those ranges (Thuiller et al., 2004). However, in this study we make predictions to areas that go beyond the observed ranges from the data. By increasing mean May temperatures by 1 1C a large part of the coastal area on Svalbard will have a mean May temperature above 1.59 1C which is the highest observed temperature where nests have been reported. We have no doubt that the warming will generally have a positive effect on habitat suitability for the geese and we believe that the exclusion of the areas outside the temperature range would be more confounding to the analysis than retaining them in the predictions. The area with frost-free months, proportion of moist/wet moss vegetation or elevation outside observed ranges is negligible (at 12 1C nest presences were predicted in 159 cells with 5 frost-free months, 15 cells outside the elevation range, and 219 cells outside the vegetation range) and we do not nd it necessary to exclude these from the analysis. Fourth, the absence of important predictor variables in a model may lead to unrealistic predictions. Snow cover at the time of egg laying is a controlling factor of the distribution of geese (Prop & De Vries, 1993; Lepage et al., 1996; Madsen et al., 2007). Data on snow coverage for the entire Svalbard are not available and our closest approximation is the mean May temperature. Fifth, in this analysis we have not considered potential climate change effects on vegetation types and primary production. It is expected that increased temperatures will lead to changes in plant community composition (polar deserts will be displaced to some extent by northward and altitudinal movement of the tundra), increase in biomass and primary production as well as drying out of wet habitats (Callaghan, 2005). It is reasonable to expect that such changes will be benecial for the geese. In East Greenland and Iceland, pinkfooted geese nest in sub and low Arctic environments (Mitchell et al., 1999). Furthermore, during recent decades, barnacle geese have successfully spread from Arctic north Russia into the temperate Baltic Sea area (Ganter et al., 1999) and pink-footed geese have spread from the highlands into the lowlands in Iceland (Mitchell et al., 1999). These examples show that goose species adapted to Arctic conditions may possess the phenotypic exibility to exploit very successfully areas with different plant phenologies and food plant qualities (Van der Graaf et al., 2006). have a positive effect on the suitability of Svalbard for nesting geese in terms of range expansion into the northern and eastern parts of Svalbard which are currently unsuitable. Hence, it is possible that increased temperatures could release the population from the suggested present density-dependent regulation during the nesting period. Furthermore, an elongation of the frost-free season in Svalbard may relax their dependence on the acquisition of body stores before arrival (so-called capital breeding, sensu Drent & Daan, 1980), so that geese will have more time to acquire the necessary resources upon arrival and still breed successfully. Both factors are likely to have a positive effect on the population growth. Future assessments of effects of climate impacts on Arctic species should combine analyses of ecologically founded spatial predictions and population dynamics.
Acknowledgements
This project was carried out under the EU 5th Framework project FRAGILE (EVK2-2001-00235). We thank D. Kuijper, C. Hubner, J. Prop and J. U. Jepsen for providing unpublished information on goose distributions. The Norwegian Polar Institute granted us access to use the digital elevation model. Rene van der Wal kindly commented on the manuscript.
References
Abraham KF, Jefferies RL, Alisauskas RT (2005) The dynamics of landscape change and snow geese in mid-continent North America. Global Change Biology, 11, 841855. ACIA (2005) Arctic Climate Impact Assessment. Cambridge University Press, New York. Araujo MB, Pearson RG, Thuiller W, Erhard M (2005) Validation of species-climate impact models under climate change. Global Change Biology, 11, 15041513. Augustin NH, Mugglestone MA, Buckland ST (1996) An autologistic model for the spatial distribution of wildlife. Journal of Applied Ecology, 33, 339347. Austin MP (2002) Spatial prediction of species distribution: an interface between ecological theory and statistical modelling. Ecological Modelling, 157, 101118. Boyd H, Madsen J (1997) Impacts of global change on arcticbreeding bird populations and migration. In: Global Change and Arctic Terrestrial Ecosystems (eds Oechel WC, Callaghan T, Gilmanov T, Holten JI, Maxwell B, Molau U, Sveinbjornsson B), pp. 201217. Springer-Verlag, New York. Callaghan TV (2005) Arctic Climate Impact Assessment. Arctic Tundra and Polar Desert Ecosystems. Cambridge University Press, New York, pp. 243352. Cohen J (1960) A coefcient of agreement on for nominal scales. Educational and Psychological Measurement, 20, 3746. Drent RH, Black JM, Loonen MJJE, Prop J (1998) Barnacle geese Branta leucopsis on Nordenskioldkysten, western Spitsbergen in thirty years from colonisation to saturatio...