Global Change Biology (2012), doi: 10.1111/j.1365-2486.2012.02739.x

The influence of mean climate trends and climate variance on beaver survival and recruitment dynamics RUAIRIDH D. CAMPBELL*†‡1, PIERRE NOUVELLET§1, CHRIS NEWMAN†, D A V I D W . M A C D O N A L D † and F R A N K R O S E L L * * Department of Environmental and Health Studies, Faculty of Arts and Sciences, Telemark University College, N-3800 Bø, Telemark, Norway, †Wildlife Conservation Research Unit, Department of Zoology, University of Oxford, The Recanati-Kaplan Centre, Tubney House, Abingdon Road, Tubney, Abingdon, Oxfordshire, OX13 5QL, UK, ‡Animal, Conservation and Education Department, Highland Wildlife Park, The Royal Zoological Society of Scotland, Kincraig, Inverness-shire PH21 1NL, UK, §Centre for Study of Evolution, School of Biological Sciences, University of Sussex, Brighton, BN1 9QG, UK

Abstract Ecologists are increasingly aware of the importance of environmental variability in natural systems. Climate change is affecting both the mean and the variability in weather and, in particular, the effect of changes in variability is poorly understood. Organisms are subject to selection imposed by both the mean and the range of environmental variation experienced by their ancestors. Changes in the variability in a critical environmental factor may therefore have consequences for vital rates and population dynamics. Here, we examine
90-year trends in different components of climate (precipitation mean and coefficient of variation (CV); temperature mean, seasonal amplitude and residual variance) and consider the effects of these components on survival and recruitment in a population of Eurasian beavers (n = 242) over 13 recent years. Within climatic data, no trends in precipitation were detected, but trends in all components of temperature were observed, with mean and residual variance increasing and seasonal amplitude decreasing over time. A higher survival rate was linked (in order of influence based on Akaike weights) to lower precipitation CV (kits, juveniles and dominant adults), lower residual variance of temperature (dominant adults) and lower mean precipitation (kits and juveniles). No significant effects were found on the survival of nondominant adults, although the sample size for this category was low. Greater recruitment was linked (in order of influence) to higher seasonal amplitude of temperature, lower mean precipitation, lower residual variance in temperature and higher precipitation CV. Both climate means and variance, thus proved significant to population dynamics; although, overall, components describing variance were more influential than those describing mean values. That environmental variation proves significant to a generalist, wide-ranging species, at the slow end of the slow-fast continuum of life histories, has broad implications for population regulation and the evolution of life histories. Keywords: Castor fiber, Climate Change, Climate Variability, Demography, Environmental Stochasticity, Fecundity, Fluctuating Environments, Population Dynamics, Seasonality, Survival Received 06 December 2010; revised version received 22 March 2011 and accepted 26 March 2012

Introduction Indeterminacy in the development of future climatic conditions presents a major issue for both human societies and natural systems (Grosbois et al., 2008). Although warming trends are well established (e.g. Smith 2011), increasing variability in regional weather patterns (IPCC, 2007) must also be modelled into future prediction scenarios. Intensification in the frequency of extreme precipitation events has been predicted (Karl et al., 1995; Groisman et al., 1999; Easterling et al., 2000), whereas temperatures are predicted Correspondence: Prof. David W. Macdonald, tel. + 44 (0)1865 393100, fax: + 44 (0)1865 393101, e-mail: david.macdonald@zoo. ox.ac.uk 1

Equal first authors

© 2012 Blackwell Publishing Ltd

to show lower diurnal and intra-annual variation (i.e. seasonal amplitude), but more frequent heat waves (Karl et al., 1995; Easterling et al., 2000; Meehl & Tebaldi, 2004; Scha¨r et al., 2004). There is mounting evidence that climate change is matching these predictions (e.g. O’Gorman & Schneider, 2009; Karl et al., 1995; Brunetti et al., 2000; Luterbacher et al., 2004; Scha¨r et al., 2004; Min et al., 2011). Ecologists are presented with a considerable challenge to develop models able to capture the full breadth of these climatological impacts effectively (Tuljapurkar 2010; Grosbois et al., 2008). Due to the complexity of patterns of climate change, where mean trends are set against a background of changes in frequency, amplitude and variation, it is important to understand how these various components that comprise climate impact

1

2 R . D . C A M P B E L L et al. on species. For example, there is increasing evidence that both changes in climate averages (e.g. Parmesan & Yohe, 2003; Hughes, 2000; Walther et al., 2002) and changes in climate stability (e.g. McLaughlin et al., 2002; Boyce et al., 2006) are causing detrimental impacts on biodiversity (Gaillard & Yoccoz, 2003). However, at the population level, not all the impacts of increased environmental variability appear to be negative (e.g. Drake, 2005) and the effects of projected increases in weather variability remain poorly understood (Weltzin et al., 2003; Boyce et al., 2006). To survive and reproduce, individuals of a given species must adapt life history decisions to environmental changes (McNamara et al., 1995; Fero´ et al., 2009). Assessing future conditions accurately in environments that are more variable has a higher potential for error compared with environments that are relatively predictable (Gaillard & Yoccoz, 2003). As a consequence, the degree of stochasticity in the environment can have a major influence on the optimal phenotype (Tuljapurkar et al., 2009). Physiological adaptability and decision-making processes are honed by evolutionary selection pressures according to the probability distribution of the environmental conditions experienced by the organism’s ancestors (McNamara et al., 2001). Changes in the patterns and extent of fluctuation in the environment can therefore generate changes in the selective pressure on life histories (Ruzzante et al., 2008; Boyce et al., 2006). Here, we explore the effects of the different components of temperature and precipitation on the survival rate and recruitment of a temperate herbivore, the Eurasian beaver, Castor fiber. Beavers are generalist foragers with a broad geographical range (see Muller-Schwarze & Sun, 2003) and their life history strategy places them towards the slower end of the fast-slow continuum (sensu Promislow & Harvey, 1990 and sensu Bielby et al., 2007). These traits are thought to reduce the likelihood, and extent to which, populations will be influenced by environmental variation (Johnson, 1998; Boyles & Storm, 2007; Dalgleish et al., 2010; Tuljapurkar et al., 2009). Therefore, the beaver is a particularly informative model species with which to examine the effects of environmental variability vs. means. Furthermore, beavers can have a marked impact on their environment and on other wildlife due to their river damming and tree-felling activity (Rosell et al., 2005), giving the beaver the status of a keystone species (Paine, 1995). Understanding how beaver populations may be influenced by climate change is therefore important in the context of wider conservation because of the implications beaver activity has for the biodiversity and abundance of wetland in boreal ecosystems (Rosell et al., 2005).

To investigate climatic influences on beaver population dynamic parameters, we combine an examination of weather trends in our study area with capture-markrecapture (CMR) analyses, based on long-term population monitoring data collected at the individual scale (Sandercook, 2006; Clobert et al., 1987; Lebreton et al., 1992), to test how: 1 intra-annual variability around long-term temperature and rainfall means, and 2 seasonal variability around these means, affect: (i) beaver annual survival rates and (ii) annual recruitment rates, and consider how these impact (iii) the optimality of population vital rates (Fero´ et al., 2009).

Materials and methods

Study site The study site was centred on three large rivers (c. 20–150 m wide), the Straumen (59°297′ N, 09°153′ E), the Gvarv (59°386′ N, 09°179′ E) and the Sauar (59°444′ N, 09°307′ E), in Telemark, southern Norway. The bedrock is granite gneiss with a thin layer of fluvial (Straumen and Gvarv) or marine (Straumen and Sauar) deposits in valley bottoms. The climate is cool continental and lies on the boundary of the Dfb and Dfc (Hemiboreal and Boreal) classes in the Ko¨ppen–Geiger climate classification system (Kottek et al., 2006) with a mean annual temperature of 4.6 °C and a mean annual precipitation of 790 mm. Mean monthly 24 h temperature dips below 0 °C for 5 months a year between November and March. Due to the presence of both natural lakes (Gvarv and Sauar) and manmade impoundments (Straumen) along their length, which moderate fluctuations in water temperature (Webb & Walling, 1996), all rivers exhibit reduced ice cover in winter. We, nevertheless, observed that all beaver families in the study area build winter food stores (caches, branches plugged into the lake or river bed) whether ice cover occured or not. All three rivers form part of the catchment of Lake Nordsjø. None of the focal beaver families built dams during the 13-year study period, as rivers are wide enough to make damming unnecessary (Hartman & To¨rnlo¨v, 2006). Illegal culling is believed to be rare in the area (B Hovde & F Bergan pers. comm.) as few beaver families are in conflict with human land use. A close relationship was maintained with local hunters, which allowed us to keep track of the majority of direct anthropogenic mortality in our population. Hunting pressure was low with 23 of a total of 268 animals (including floaters: animals that have dispersed from their natal territory, but have not established a territory of their own) shot or kill-trapped during the course of the study. Predation pressure was also low, as wolves, Canis lupus, have been extirpated from the area for over 100 years, bears, Ursus arctos, only occasionally pass through and lynx, Lynx lynx, are present, but at low densities (Rosell & Sanda, 2006). Following extirpation from overhun-

© 2012 Blackwell Publishing Ltd, Global Change Biology, doi: 10.1111/j.1365-2486.2012.02739.x

EFFECTS OF CLIMATE VARIABILITY ON THE BEAVER 3 ting and subsequent protection, beavers have naturally recolonized part of their former range and have been in the study area since the 1920s (Olstad 1937).

Population monitoring Between March and November 1997 to 2009, inclusive, beavers were monitored through an extensive live-trapping programme, using hand-nets from a motorboat (Rosell & Hovde 2001). Captured individuals were immobilized in cloth sacks, sexed based on the colour of the anal gland secretion (Rosell & Sun 1999), tagged with a microchip (Avid or Trovan) and marked with unique colour-plastic (Dalton) and metal (National Band and Tag Co, Newport, KY, USA) ear-tag combinations. All animal handling procedures were approved by the Norwegian Directorate for Nature Management and by institutional ethical review. Animals were assigned to an ageclass (0 years = kit, 1 years = yearling, 2 years = subadult and
3 years = adult) based on body weight (Campbell et al., 2010). Recaptures resulted from either the trapping or sighting of marked animals (Clobert et al., 1987, 1994). Dominance status was determined by previous trapping and sighting history and incidences of lactation in females. For example, individuals dispersing into a territory were posited to have obtained a dominant breeding position, an assumption generally corroborated by data indicating the disappearance (including usurpation) of the previous dominant individual of the same sex and, for females, by evidence of lactation (signified by nipple length >0.5 cm). In the majority of cases, dominant status could be attributed to the adult territorial resident of each sex. On two occasions, where more than one animal was a potential dominant individual, we attributed dominance to the heaviest individual, following Smith et al. (1994). Mean annual trapping effort (the number of nights in each year spent on the rivers specifically trapping and sighting beavers) was 30 ± 15 SD (Straumen 15 ± 11; Gvarv 10 ± 10; Sauar 9 ± 8 nights). Additional sighting data were collected on an ad hoc basis during other research activities on the rivers. Trapping effort was concentrated over spring and summer, with 39.3% of trap nights in March–June, 44.6% in July–August and 16.1% in September–November, with no trapping in December–February due to ice on the rivers. Throughout the study we base our analyses on a demographic history file documenting the capture events of 242 individual beavers, each of which was caught between 1 and 11 times (appendix S1), some moving through the age-classes as the study progressed. Of these, 66 were recorded in the population as kits, 95 as yearlings and 141 as subadults. Of the 149 individuals contributing to the adult age-class, 106 attained dominant breeding positions while the remaining 43 did not.

26 m, 59°383′ N, 09°183′ E, Gvarv-Lindem 1989–1994, elevation 71 m, 59°387′ N, 09°202′ E, Gvarv-Nes 1994–2009, elevation 93 m, 59°383′ N, 09°201′ E) were available from 1919 to 2009; Monthly precipitation figures from Lifjell (height 354m, 59°455′ N, 09°037′ E) from 1896 to 2009. All beaver territories were within 18 km of these weather stations. These weather stations at Gvarv were all within 1.4 km of each other. Nevertheless, to make sure that historical relocation of weather stations in the study region had not influenced the continuity of temperature readings, we also obtained temperature records from 1954 to 2007 from Tveitsund in Telemark (elevation 455 m, 59°027′ N, 08°521′ E, approximately 60 km south-west of Gvarv). Temperature readings were highly correlated with those from all weather stations at Gvarv and there was no evidence that changes in the weather stations from which we utilize meteorological records affected temperature values (Appendix S2). For each year we derived indices that reflect the climatic conditions (i.e. temperature and rainfall). Particular regard was given to exploring the effects of weather conditions on a meaningful biological scale, with emphasis on decoupling the influence of variability and changing average trends (Grosbois et al., 2008). A data-year was defined to start on the 1st of March, to follow beaver life cycle patterns. The population dynamics of the closely related North American beaver (C. canadensis) are known to be influenced by climatic conditions predominantly during the vegetation growing season (e. g. spring – autumn) and far less so by climatic conditions over winter (Jarema et al., 2009). In northern latitudes, both beaver species cache food stores underwater in the autumn, thus beavers can remain entirely within their lodges or bank dens, and do not venture above ice cover to forage throughout the substantial majority of the winter (Muller-Schwarze & Sun, 2003). It is therefore reasonable to infer that the effects of winter weather on beavers, thus protected and insulated, will be much less influential than weather conditions during other times of year, with the quality of lodge construction mainly effecting their internal den environment (Dyck & MacArthur, 1993). Furthermore, as winter precipitation is cumulative (creating snow pack), precipitation variability is likely to be less important than the sum total precipitation over winter. Similarly, the temperature of the water (which the beaver will experience in winter as it swims between lodge and food store) will fluctuate much less than that of the air, again resulting in the mean air temperature being more influential than air temperature variability. As no biologically meaningful inferences can be made between winter weather variability and means, we therefore excluded winter weather and thus evaluate each of the yearly climate indices described below using the climate data between the 1st April and the 30th of September (days 32 to 214 of the data-year). We do, however, acknowledge that mean winter weather may, nonetheless, influence the vital rates of this population.

Climate data Daily temperature and monthly precipitation figures were obtained from the Norwegian Meteorological Institutes eKlima online database (http://eklima.met.no/): Daily temperatures from Gvarv (Gvarv weather station from 1919–1989, elevation

Temperature. Temperature, as expected from the changing angle of incidence of the sun at the latitude of the study area, demonstrated clear evidence for seasonal variation (Fig. 1). To model and extract informative components of

© 2012 Blackwell Publishing Ltd, Global Change Biology, doi: 10.1111/j.1365-2486.2012.02739.x

4 R . D . C A M P B E L L et al. 100

Log10 (rainfall)

Temperature

20 10 0

−10

100

200

10

1

300

Days

0

Fig. 1 Daily temperature in Gvarv over 1 year from 1st March 2007. A seasonal trend within a year is clear, in the form of an underlying sinusoidal curve with period equal to a year. The solid line represents the yearly mean temperature, lT, plus the amplitude of changes, aT in 2007. Some variation around these predictions (the solid line) remains and can be characterized by its variance, hence the definition of rT.

temperature – given the sinusoidal nature of the variation in temperature within a year, we apply a ‘sinusoidal model’ (Simmons 1990). We characterize yearly temperature with the terms: Mean (average) temperature for that year: lT. Amplitude of seasonal temperature changes: aT. Residual variance around the daily predicted temperature values provided by these two previous indices. This residual variance can be interpreted as high-frequency variation in temperature. To derive a coherent scale across these terms, rather than utilizing variance, we calculated the standard deviation in temperature: rT. In our model, the temperature T on day d of year y is thus characterized as: Td;y ¼ lT;y þ aT;y cosðdcÞ þ ey where: (i) ey is the residual variance associated with year y, each ey value is taken (or drawn) from a normal distribution with mean 0 and variance r2T;y and (ii) c is a constant (2p/365) such that the cycle is finished after 1 year. Using a regression procedure, based on least square (function ‘regress’ from Matlab; The MathWorks Inc., Natick, MA, USA), for each year we then estimate lT,y, and aT,y, for all temperature data available between 1919 and 2009. The sum of squared residuals over each year allows us to derive the yearly temperature variance, r2T;y and thus rT,y. We explored the variation in climatic indices over time using linear regressions with year as the predictor, and the correlation between indices using Pearson’s correlations.

Rainfall. From inspection of daily rainfall (to include all accumulated precipitation as millimetres of liquid water) records within a year, no significant within-year seasonal trends were evident (Fig. 2). This permitted rainfall, within any given year, to be characterized by mean rainfall, lR, and the coefficient of variation cvR. The two indices were calculated for a period of 115 years from 1895 to 2009.

100

200

300

Days Fig. 2 Daily rainfall over Lifjell over 1 year from 1st March 2007. No seasonal trends were observed within a year.

We used the coefficient of variation in rainfall rather than standard deviation, as the latter is highly correlated with the mean rainfall. This approach for measuring rainfall variability is consistent with numerous climatic studies (Le Houe´rou et al., 1988; Fay et al., 2003; Nippert et al., 2006). Again, linear regressions were used to explore variation in climatic indices over time, and Pearson’s correlation was used to assess correlations between indices.

Influence of climate indices on population dynamics Informed by previous studies on North American beaver (Payne 1984; McNew & Woolf, 2005), we separated beaver survival rates by age-class. Population losses attributed to mortality could in actuality be due to dispersal out of the study area. As dispersal is almost exclusively restricted to nondominant individuals, the problem of distinguishing mortality and dispersal was addressed by including dominance status in the model. If the variable influenced mortality and not dispersal, we would expect to see effects on both dominant and nondominant individuals. Conversely, any variable influencing dispersal and not mortality would be apparent through effects on recapture rates for nondominant but not dominant individuals. We thus distinguish between: kit survival φk (from first emergence from the natal lodge at 2 months to 1 year old), juvenile survival φk (older than one, but less than 3 years old, see Appendix S3) and adult (greater than 2 years old), separated into dominant φDa or nondominant φNa (philopatric offspring). This constituted our Minimum Adequate Model (MAM; sensu Gillespie et al., 2008) on which further analyses were conducted by including climatic covariate(s). In all model scenarios, capture probabilities were considered constant with age-class, but variable between years, and between the three main study sites (reflecting differential sampling effort). Further details and explanation of models presented here are included in Appendix S3.

Survival rates. Survival rates were estimated for each of the three age-classes (kit, juvenile and adult) along with the

© 2012 Blackwell Publishing Ltd, Global Change Biology, doi: 10.1111/j.1365-2486.2012.02739.x

EFFECTS OF CLIMATE VARIABILITY ON THE BEAVER 5 (a)

(b)

(c)

5

3

σT

T

16

α

μT

8

12

2 2 1920

1960

2000

8 1920

Years

1960

2000

1920

Years

1960

2000

Years

Fig. 3 Long-term trends in Mean (a), amplitude of seasonal changes (b) and residual standard deviation (c) in daily temperature (°C) between April and September for each year between 1919 and 2009.

(a)

(b)

2.5

cvR

μR

4 3

2

2 1900

1950

Years

2000

1900

1950

2000

Years

Fig. 4 Mean (a) and coefficient of variation (b) in daily rainfall (mm) between April and September of each year between 1895 and 2009.

additional consideration of dominance status within the adult age-class (creating four classes ‘c’). We used the ‘Cormack-Jolly-Seber’ (CJS) model (Jolly, 1965; Lebreton et al., 1992), implemented in the program Mark (White & Burnham 1999; version 6.0), with modification in the Parameter Index Matrix (PIM) to account for age and dominance status (see Appendix S3 for details). ‘Dominant status’ was attributed to any individual that had attained a dominant breeding position in the population by either the end of the study period or the last time it was recorded. Of 106 individuals that attained dominant positions, 26 were of known age and the remainder were adult (
3 years old) when first recorded. Of these 26 known-age individuals, the mean age at which an individual achieved dominance status was 4.3 ± 1.8 years. Fifteen individuals (14.2%) held nondominant status as adults for 1–5 year(s), prior to achieving dominant breeding positions. These individuals were considered as dominant during analyses. Survival rates were used to construct a primary model, dependent upon year (y), {φ(y, c)}. This model was checked for goodness of fit using bootstrap methods (White 2002), allowing approximation of a corrected variance inflation factor ð^cÞ. A logit transformation for the estimation of survival rates (being probabilities) was used in all models presented. Survival rates were then reestimated, allowing the logit of survival estimates to follow a linear relationship with climatic indices. Models varied in which climatic indices they included as covariates and were compared using a multimodel inference procedure (Burnham & Anderson, 2002, 2004).

In terms of biological influence, weather conditions could affect the survival rates of the age/dominance classes differently. We thus constructed models with and also without interaction terms (accounting for differential and similar relationships between survival and climatic indices for each age-class). Given the resolution of our data, relationships between the kit age-class and climatic indices could not be evaluated with interactions (i.e. due to problems of model convergence). In response, we grouped together nonreproductive age individuals (kits and juveniles) and assumed that their survival rates (although different, see above and Fig. 5) were in effect influenced by climatic indices similarly. As well as evaluating models without interaction terms, we were thus able to evaluate models including interactions with respect to three functional age-classes, kit-juvenile/adult nondominant/ adult dominant. Using a multimodel inference procedure (Whittingham et al., 2006; Burnham & Anderson, 2002, 2004), we compared CMR models where survival rates from the MAM were constrained by climate indices (in addition to the age/dominance classes, c, defined above). This permitted investigation of the linear relationship (on a logit scale) between the survival estimate and climatic indices, using the Capture-Mark-Recapture (CMR) framework, as described by Grosbois et al. (2008). In multimodel inference, it is important that all variables are represented equally in the analysis. Given that five climatic indices were available per year, we were able to construct 32 models (i.e. one model with no climatic covariate, five models with one climatic covariate and ten models with two, ten models with three, five models with four and one model with five climatic covariates). We thus created a balanced set of candidate models that represented all climatic covariates equally. In addition, 31 models (each model including climatic covariate(s)) were constructed comprising interaction term(s) with age/dominance class. We therefore evaluated 63 models, where in 62 instances we were able to examine for similarities or differences in age/dominance class survival rate responses to the climatic indices. For each model, we derived the QAICc (AIC corrected for small sample size ‘c’ and adjusted for overdispersion ‘Q’: a lower value indicates a better supported model) and used this to derive Akaike weights for each model (Buckland et al., 1997) to rank the support for each model. A different estimation of the parameter h, linking the climate component in the model with the logit of the survival rate, was estimated for each model.

© 2012 Blackwell Publishing Ltd, Global Change Biology, doi: 10.1111/j.1365-2486.2012.02739.x

6 R . D . C A M P B E L L et al.

(a)

1

1

1

1

1

0.5

0.5

0.5

0.5

0.5

0 4

Survival rates

(b)

0

10

15

0 2

2.5

3

0 2.5

3

3.5

4

0 1.5

1

1

1

0.5

0.5

0.5

0.5

0.5

5

6

7

0

10

15

0 2

2.5

3

0 2.5

3

3.5

4

0 1.5

1

1

1

1

1

0.5

0.5

0.5

0.5

0.5

5

6

7

0

10

15

0 2

2.5

3

0 2.5

3

3.5

4

0 1.5

1

1

1

1

1

0.5

0.5

0.5

0.5

0.5

0 4

Recruitment

7

1

0 4

(d)

6

1

0 4

(c)

5

5

6

7

0

10

15

0 2

2.5

3

0 2.5

3

3.5

4

0 1.5

1

1

1

1

1

0.5

0.5

0.5

0.5

0.5

0 4

5

μ

T

6

7

0

10

α

15

0 2

T

2.5

σ

T

3

0 2.5

3

3.5

μR

4

0 1.5

2

2.5

2

2.5

2

2.5

2

2.5

2

2.5

cvR

Fig. 5 Survival rate estimates for kits (less than a year old) (a); juveniles (between 1 and 3 years old) (b); adults (3 years old or more) either as dominant (c) or nondominant (d) and recruitment as a function of each climatic index. Solid lines represent significant relationships (i.e. where the confidence interval for the coefficient does not overlap 0, see Table 2–4). Estimates of survival rates, recruitment and their confidence intervals (error bars) are based on model averaging. Adult survival rate is divided into dominant and nondominant classes. Juvenile and nondominant adult survival rates are similar and both are lower than for kits and the dominant adult class. A negative relationship is evident between survival rate and the coefficient of variation in rainfall. Note that kit and juvenile age-classes were combined for interaction analyses; the apparent differences in the slopes for these two age-classes are artefacts of the logit back transformation.

This allowed us to estimate the relative influence of each climatic component as the sum of the Akaike weight of models containing each particular index (Burnham & Anderson, 2004). We were then able to employ model averaging to estimate the h‘s with their associated confidence intervals from all 63 models. Model averaging accounts for uncertainty in model selection by calculating the mean value for a parameter of interest through averaging over all models in the candidate model set containing this parameter of interest, weighted by normalized AIC weights (Buckland et al., 1997; Burnham & Anderson, 2002, 2004). To conform with theoretical developments (Burnham & Anderson, 2002), confidence intervals were based on estimated unconditional variance, accounting for two variance components: the conditional sampling variance, given a model, and the variation associated with model selection uncertainty.

Recruitment. We used the Pradel model (Pradel 1996) within the CMR framework to estimate yearly recruitment parameters

(measured as the number of kits per adult). Ostensibly, this recruitment parameter equates with apparent population fecundity, however, the Pradel approach does not allow the specification of age-classes, and thus we performed separate analyses for survival (i.e. incorporating variation in survival between ageclasses, as presented above) and recruitment. Recruitment estimates were inferred as the number of recruits per individual per year, directing us to a log-link function model. Again a set of 32 models was constructed, where recruitment, f, was dependent (or not) upon the climate indices from the preceding year. Survival rate values and capture probabilities in all recruitment models were fixed to predetermined values, estimated using a CJS model with survival rates constant across age-classes, but dependent upon year, and with capture rates dependent on year and location (see Appendix S3). Further model evaluation was performed as above, allowing us to determine the estimated influence of the coefficients of each climatic index and their confidence intervals.

© 2012 Blackwell Publishing Ltd, Global Change Biology, doi: 10.1111/j.1365-2486.2012.02739.x

EFFECTS OF CLIMATE VARIABILITY ON THE BEAVER 7 Results

Climate indices Temperature. A long-term significant trend of increasing yearly mean temperature, lT, was detected (b = 0.02 SE 5.103; F87 = 22.5, p < 0.001, r2 = 0.20), equivalent to a rise of 2 °C since 1919 (Fig. 3a). A significant decreasing long-term trend in yearly change in seasonal temperature amplitude, aT, was also detected (b = 0.02 SE 6.103; F87 = 6.6, p = 0.012, r2 = 0.07), equivalent to a 1.8 °C reduction in the difference between winter and summer since 1919 (Fig. 3b). In addition, there was a long-term trend of increasing yearly residual standard deviation in temperature, rT (b = 3 9 103 SE 1.103; F87 = 9.35, p = 0.003, r2 = 0.10) (Fig. 3c). While a negative correlation was evident between lT and aT (Pearson correlation, ρ = 0.79, p < 0.001), no such correlation was observed between lT and aT (Pearson correlation, ρ = 0.17, p = 0.11), or between aT and aT (Pearson correlation, ρ = 0.06, p = 0.58). Rainfall. No long-term trend in yearly mean rainfall or in the coefficient of variation in rainfall was detected (for lR: b = 1 9 103 SE 2.103; F113 = 0.196, p = 0.66,

r2 < 0.01 and for cvR: b = 7 9 104 SE 1.103; F113 = 0.911, p = 0.34, r2 < 0.01) (Fig. 4a, b). A negative correlation was, however, detected between the two indices (Pearson correlation, ρ = 0.58, p < 0.001). Importantly, no correlation between lR and cvR was in evidence between 1997 and 2009 (the period for which demographic data are available, Pearson correlation, ρ = 0.40, p = 0.17).

Influence of climate indices on population dynamics Modelling constant survival rates across years, irrespective of weather conditions, revealed that average annual survival rate was highest for the kit age-class (0.92; SE 0.05), followed closely by that of dominant adults (0.87; SE 0.02), juveniles (0.72 SE 0.04) and of nondominant adults (0.64; SE 0.06). Average recruitment rate across all years was 0.20 (SE 0.01). Survival rates. Of all 63 models ranked according to their relative statistical support (see Table 1 for the five most supported models, and Table S3 and Table 2 for full version), models that did not include interaction terms with age/dominance status ranked highest. The

Table 1 Statistical summary of the five most supported models linking survival rate to climate indices. Models were constructed by adding climatic variable(s) to a MAM and thus have the form ‘phi’ ({climatic variable(s)}, c), p(terr, year)’ where phi represents survival rate and p represents recapture probability. ‘No Int’ indicates models without interaction, whereas ‘Int’ specifies models with interaction. For the models without interactions, survival rates are linked with climate indices consistently, irrespective of age/dominance class. From Akaike weight (wi), the most supported model is cvR, No Int. This model and cvR, Int have ca. twice the support of the third ranked model (lR, cvR, No Int). h represents the coefficient linking survival rates to each climate index within the logistic equation, with standard errors in parenthesis. The full version of this table, including a summary of all 63 models, is presented in Appendix S3

Coefficient for kit and juvenile survival

Coefficient for adult dominant survival

Coefficient for adult nondominant survival

Model rank

1

2

3

4

5

Climatic variable(s) used in the MAM AICc ΔAICc wi hlT haT hrT hlR hcvR hlT haT hrT hlR hcvR hlT haT hrT hlR hcvR

cvR, No Int

cvR, Int

lR, cvR, No Int

lR, cvR, Int

rT, Int

914.7 0 0.14

914.9 0.2 0.12

915.8 1.1 0.08

916.0 1.3 0.07

916.3 1.6 0.06

3.25 (1.44) 3.84 (1.87)

6.51 (2.46)

0.40 (0.38) 4.04 (1.84)

1.14 (0.50) 7.67 (2.44)

0.39 (3.04)

0.40 (0.38) 4.04 (1.84)

0.29 (0.59) 0.13 (3.06)

0.97 (4.49)

0.40 (0.38) 4.04 (1.84)

0.06 (0.84) 1.74 (4.80)

0.90 (1.21) 3.84 (1.87)

1.96 (2.10) 3.84 (1.87)

© 2012 Blackwell Publishing Ltd, Global Change Biology, doi: 10.1111/j.1365-2486.2012.02739.x

8 R . D . C A M P B E L L et al. Table 2 Model averaging on the parameters linking survival rates to climate indices (within the logistic model). The relative influence of each parameter (based on Akaike weights) is presented, along with the model-averaged estimated values of these parameters with confidence intervals, based on the estimated unconditional variances. Estimates with confidence intervals that do not cross zero are marked with an asterisk

Influence h Lower 95 CI Upper 95 CI h Lower 95 CI Upper 95 CI h Lower 95 CI Upper 95 CI

Kit and juvenile survival

Adult dominant survival

Adult nondominant survival

lT

aT

rT

lR

cvR

0.19 0.00 0.04 0.04 0.01 0.05 0.03 0.00 0.04 0.04

0.20 0.00 0.02 0.02 0.00 0.02 0.02 0.00 0.02 0.02

0.34 0.41 0.85 0.03 0.27* 0.54 0.00 0.01 0.41 0.43

0.34 0.21* 0.38 0.03 0.05 0.19 0.09 0.09 0.24 0.07

0.75 3.72* 5.06 2.37 2.01* 3.41 0.61 1.74 3.61 0.13

relative influence (sum of Akaike weights) of models without interactions was 0.68 compared to 0.32 for the relative influence of models with interactions. It was thus evidenced that weather conditions affected beavers of different age-class and dominance status in a highly consistent manner. Based on the support for each model, we established that the coefficient of variation in rainfall (cvR) was overwhelmingly the most influential (relative influence, as the sum of Akaike’s weights, reached 0.75). The influence of the mean rainfall (lR) and the residual standard deviation in temperature (rT) followed, with the mean temperature (lT) and the amplitude of seasonal change (aT) having the least influence and least discernible significant effect on any age-class or dominance status (Table S3 and Table 2). Beaver survival proved to be associated negatively with both indices of rainfall (cvR, lR) and negatively with the residual standard deviation in temperature (rT) (confidence intervals, not including 0; see Table 2). This negative relationship with the coefficient of variation in rainfall (cvR) was significant for the survival rates of both the kit-juvenile class and for dominant adults, whereas bordering significance for the survival rate of nondominant adults (see Table 2; Fig. 5). The negative relationship with amplitude of seasonal tem-

perature change (aT) was significant for dominant adult survival rates only, but bordered significance for the kit-juvenile class (see Table 2; Fig. 5). The negative relationship with mean rainfall (lR) was significant for the kit-juvenile class only (see Table 2; Fig. 5). Recruitment rates. The recruitment of beavers into the adult population was associated positively and significantly with the amplitude of seasonal temperature change, negatively with both mean rainfall and standard deviation in temperature, and positively with the coefficient of variation in rainfall (Tables 3 and 4; Fig. 5). Comparing all 32 models (see Table 3 for the top five models, and Table S3 and Table 3 for all 32 models) revealed that the model including only the amplitude of seasonal temperature changes (aT) achieved similar support to the model including a much more extensive set of weather variables, in addition to aT. Amplitude of seasonal temperature changes (aT), mean rainfall (lR) and residual standard deviation in temperature (aT) were the most influential covariates predicting recruitment (Table 4) with amplitude exhibiting the most substantial influence (greatest Akaike weight). As with survival, based on 95% confidence intervals, the effects of these three covariates were statistically significant (i.e. confidence intervals did not include 0; see Table 4).

Table 3 Statistical summary of the five most supported models linking recruitment to climate indices. Based on Akaike weight (wi) the first model was slightly more supported than the next three models, which all rank similarly. h represents the coefficient linking recruitment rate to each climate index within the log transformation, with standard errors in parenthesis Model rank

Climatic variable(s) used in the model

AICc

ΔAICc

wi

1 2 3 4 5

aT, rT, lR, cvR lT, aT, aT, lR, cvR aT aT, lR lT, aT, rT, lR

2206.89 2207.70 2207.82 2207.93 2208.84

0 0.81 0.93 1.04 1.95

0.18 0.12 0.11 0.10 0.07

hlT 0.299 (0.268)

0.482 (0.289)

haT

hrT

hlR

hcvR

0.315 (0.091) 0.426 (0.134) 0.478 (0.068) 0.419 (0.078) 0.554 (0.114)

2.559 (1.222) 2.645 (1.166)

0.831 (0.384) 0.903 (0.394)

3.468 (1.625) 3.006 (1.649)

1.488 (0.980)

0.424 (0.304) 0.763 (0.385)

© 2012 Blackwell Publishing Ltd, Global Change Biology, doi: 10.1111/j.1365-2486.2012.02739.x

EFFECTS OF CLIMATE VARIABILITY ON THE BEAVER 9 Table 4 Model averaging for the parameters linking recruitment to climate indices (with log transformation), giving the relative influence of each parameter (based on Akaike weights), the model averaged estimated values of each parameter and associated confidence intervals, based on estimated unconditional variances. Estimates with confidence intervals that do not cross zero are marked with an asterisk

Recruitment

Relative influence h Lower 95 CI Upper 95 CI

lT

aT

rT

lR

cvR

0.37 0.105 0.006 0.215

0.996 0.434* 0.36 0.508

0.54 0.997* 1.849 0.146

0.64 0.433* 0.694 0.173

0.5 1.222* 0.09 2.355

The positive effect of the coefficient of variation in rainfall was also statistically significant.

Discussion There is a growing recognition among ecologists that a consideration of environmental variance is essential to understand fully many of the important patterns and processes in nature (e.g. Tuljapurkar, 2010; Rosell & Sun, 1999; Easterling et al., 2000; Clark et al., 2001; Schlesinger 2006; Hulme, 2005; King, 2005; IPCC, 2007; Grosbois et al., 2008). Our primary prediction, that climate variability would prove at least as influential on vital rates as changes in mean values, was confirmed: While mean rainfall values influenced both survival and recruitment, the effect of climate variability, in the form of variance in rainfall and temperature and seasonal amplitude in temperature, was also highly influential in a variety of predictive relationships for survival and recruitment rates. Notably, our continuous observations within the study area recorded no large-scale changes in habitat type, or other apparent causes for changing habitat productivity, which might otherwise explain variation in survival and recruitment.

Effects on survival rate Survival rates were affected negatively by the coefficient of rainfall variation, by the residual standard deviation in temperature, and to a lesser extent by mean rainfall. Whereas net primary productivity is known to respond positively to absolute rainfall, increases in temporal variability in water availability can exhibit a negative relationship with primary productivity (Fay et al., 2002). Thus, the counter-intuitive negative effect of mean rainfall that we observed on survival likely arises from water-logging in wet years with consequential reduction in riparian plant growth (Johansson & Nilsson, 2002; R. Campbell & F. Rosell, unpublished data); as a consequence beaver forage may be affected accordingly. The negative effect of both rainfall and temperature variation on survival rates underscores the

importance of considering variance in environmental factors when examining biological systems. Age specific survival rate responses to weather indices could provide further resolution of the mechanisms involved in this population regulation (see Coulson et al., 2001; Cowley & Sirwardena, 2005). Models without age/dominance status interactions were overall more supported than models with interactions. Contrasting the different climate index model-averaged parameter estimates in each age/dominance class reveals that all significant, or near-significant, effects on survival rate conformed to the same trend across the classes. This points to commonality in the mechanisms underscoring climate indices effects, emphasizing the importance of models without interactions. Nevertheless, differences were evident between age/dominance classes: A significant negative effect of mean rainfall was found only in the nonadult age-class. Indeed, the nonadult age-class showed steeper slope of the parameter estimates for cvR, lR and rT, although the latter fell below significance. This may arise either because younger, less-experienced, individuals suffer more from adverse weather than older animals, or because some of the 2-year-old animals in the juvenile age-class emigrated to breeding positions created outside the study area as a consequence of higher mortality in dominant adults (i.e. an additive effect of lower survival and increased emigration). It is not immediately apparent why survival in the nondominant adult class stood out by showing no response to the climate indices. If nondominant adults were affected similarly to dominant adults by the climate indices, the additive effect of survival and emigration described above should also lead to a more obvious response of survival in nondominant adults compared with dominant adults. Notably, only 43 nondominant adults persisted in the population, and therefore small sample size could have reduced analytical power.

Effects on recruitment As with survival, that recruitment also showed a negative association with residual temperature variance, but no clear response to mean temperature, further

© 2012 Blackwell Publishing Ltd, Global Change Biology, doi: 10.1111/j.1365-2486.2012.02739.x

10 R . D . C A M P B E L L et al. underscores the importance of considering environmental variability in biological systems. The negative effect of mean rainfall on recruitment complements the relationship between mean rainfall and survival and is likely due to the same water-logging mechanism. Seasonal temperature amplitude, aT, is a measure of the difference between the cooler part of the growing season (spring and autumn) and the warmer part (summer). Greater amplitude indicates either a cooler spring/autumn and/or a warmer summer. As we found that aT correlates negatively with lT (i.e. there was a greater amplitude in cooler years), we conclude that years with high amplitude had cooler spring and/ or autumn temperatures. Phenologically, cooler spring weather in years with higher amplitude may thus explain the positive association between aT and recruitment: cooler spring temperatures create conditions of slower forage plant development and therefore the period for which quality early growth is available to foragers will be extended (Saether, 1985; Nolet et al., 2005; R. Campbell & F. Rosell, unpublished data). Recruitment arises predominantly from within this study population (e.g. between 1997 and 2008, 143 beavers were recorded as born within the study area, whereas only 22 moved into the study area during the same period), and consequently the primary effects on recruitment are liable to reflect fecundity and reproductive success responses. The strong negative effect of cvR on survival rate does, however, imply that the positive effect of cvR on recruitment may be due to individuals immigrating, to replace dominant animals that may have died during years with high rainfall variability. Winter conditions were not expected to be especially influential on beaver survival and recruitment in our population, as beavers remain primarily within bank dens and lodges during this time (Dyck & MacArthur, 1993), and feed on food caches collected during autumn (Muller-Schwarze & Sun, 2003). December–January temperature does, however, have a significant effect on the weight of yearlings and subadults in the following year (R. Campbell & F. Rosell, unpublished data). Nevertheless, our preliminary data exploration here demonstrated that models without winter data were more effective in describing beaver survival and recruitment; that is winter temperature variability appears not to play a major role. Specific analyses of the effects of winter warming may warrant investigation if warming trends continue.

Potential future impacts of climate on the study population Trends in climate indices in our study area conformed, in part, with the changes predicted for

Northern Europe (e.g. Karl et al., 1995; Groisman et al., 1999; Easterling et al., 2000; Meehl & Tebaldi, 2004; Scha¨r et al., 2004). All temperature indices followed these projections, but neither rainfall index exhibited a trend. In particular, we observed changes in the components of annual temperature regime in accordance with the predictions of Karl et al. (1995) and Easterling et al. (2000); that is, increasing mean (lT) and residual variance (rT), but lower seasonal amplitude (aT). The negative correlation between lT and aT leads us to conclude that the trends in both components arise primarily from warmer springs and/or autumns and not from warmer summers (see discussion above). Groisman et al.’s (1999) study found both mean summer rainfall and the frequency of heavy precipitation events in summer increased in Norway over the past century and therefore the lack of trends in our study area may reflect local conditions or that the dataset has not been homogenized (Hanssen-Bauer & Førland, 1994). Increasing rT and decreasing aT trends over the past 90 years followed patterns that, over the 13 years of this study, impacted on recruitment and (for rT) dominant adult survival negatively. Conversely, a trend of increasing lT over the past 90 years was not matched with a negative effect on either vital rate. This may arise because an increase in lT is very likely to be associated with an increase in the length of the growing season (Jarema et al., 2009); where consequent improvements in forage availability would outweigh any negative impacts arising from dislocated adaptation to former, local and historic conditions. Kjellstro¨m et al. (2007) compared several predictive models of temperature extremes in Europe and found that for all models, the increase in winter minimum temperatures in southern Scandinavia was greater than the increase in summer maximum temperatures, suggesting that the trend of declining aT will continue. Under these scenarios, the vital rates of the beavers may be impaired. For example, were trends in aT to continue at current rates, recruitment rates would decline by approximately 50% over the next 90 years. Scha¨r et al., 2004 predict an increase in rT during June–August in southern Scandinavia of approximately 20% by 2100, whereas we found an increase in rT over April–September of approximately 12% over the past 90 years. Such a change in rT would have an important additive effect on reducing recruitment rates in the population: an increase in rT of 12% over the next 90 years would lead to a decline in recruitment rates of approximately 20%, whereas if both rT and aT continued their historic trends, the additive effect would reduce recruitment rates by 60% over 90 years. Due to the low slope of relationship between rT and adult dominant survival, a continuing decline in

© 2012 Blackwell Publishing Ltd, Global Change Biology, doi: 10.1111/j.1365-2486.2012.02739.x

E F F E C T S O F C L I M A T E V A R I A B I L I T Y O N T H E B E A V E R 11 rT over the next 90 years would, however, result in only a small decline in survival for this age-class of approximately 1%. Using an otherwise comprehensive species-climate envelope model, Jarema et al. (2009) predicted that the abundance of North American beaver would increase with projected changes in climate, whereas our results here show that climate variability effects oppose this prediction. Although we found no positive effects of increasing mean temperature on beaver vital rates, it is plausible that if trends towards an extended growing season and reduced winter ice cover persist, then these effects may, in the longer term, outweigh the predicted negative effects of climate variability.

General implications As a slow-living, generalist and wide-ranging species, the beaver would be predicted to exhibit resilience in vital rates to environmental variation (Johnson, 1998; Boyles & Storm, 2007; Dalgleish et al., 2010; Macdonald et al., 2010; Tuljapurkar et al., 2009). Nevertheless, it appears that the optimality of beaver annual routines is compromised by reduced capacity to prospect risk under conditions of uncertainty (Fero´ et al., 2009), leaving them more susceptible to climatic perturbations through effects on survival and fecundity (McDermott et al., 2008). The stronger responses in both recruitment and in survival in younger age-classes (with reference to slopes) suggests that increasing variability in temperature and rainfall and declining seasonality will reduce recruitment into the breeding population, potentially amplifying the risk of population extinction and exerting changes in selective pressures on individuals (McNamara et al., 2001; Berteaux et al., 2004). This comprehensive vulnerability to environmental variation therefore has implications for population growth, the evolution of life histories and for conservation (Tuljapurkar 2010).

Acknowledgements The authors would like to thank Frode Bergan for technical support, Bjørnar Hovde for trapping, Lasse Asmyhr, Claire Buchanan, Laura Daniells, Rita Goncalves, Orsi Haarberg, Jan Herr, Kristian Ingdal, Ba˚rd Andreas Lassen, Leigh Murray, Bruno Pinto, Christian Robstad, Jørn-Ingar Sanda, Matthias Scherger, Jo¨rg Schlichter, Fiona Sharpe, Kjartan Sjulstad, Liat Romme Thomsen, Helga Veronica Tinnesand, Jan Marc Tu¨rschmann and Lisa Wallis for field assistance, Akkerhaugen Marina for providing moorings, the Norwegian Meteorological Institute for providing weather data and finally the local residents of the three rivers for their relaxed and patient attitude to our nocturnal research activities. Christina Buesching provided useful comments on earlier drafts and Paul Johnson provided statistical insight. This study was supported financially by a Telemark

University College grant to RDC and FR and through a grant from the Peoples Trust for Endangered Species to DWM. We thank J.M. Gaillard and three anonymous reviewers for comments on this manuscript.

References Berteaux D, Re´ale D, McAdam AG, Boutin S (2004) Keeping pace with fast climate change: can arctic life count on evolution? Integrative and Comparative Biology, 44, 140–151. Bielby J, Mace GM, Bininda-Emonds OR et al. (2007) The fast-slow continuum in mammalian life history: an empirical reevaluation. The American Naturalist, 169, 748–757. Boyce MS, Haridas CV, Lee CT (2006) Demography in an increasingly variable world. Trends in Ecology & Evolution, 21, 141–148. Boyles JG, Storm JJ (2007) The perils of picky eating: dietary breadth is related to extinction risk in insectivorous bats. PLoS ONE, 2, e672. Brunetti M, Buffoni L, Maugeri M, Nanni T (2000) Precipitation intensity trends in northern Italy. International Journal of Climatology, 20, 1017–1031. Buckland ST, Burnham KP, Augustin NH (1997) Model selection: an integral part of inference. Biometrics, 53, 603–618. Burnham KP, Anderson D (2002) Model Selection and Multi-Model Inference. Springer, New York. Burnham KP, Anderson DR (2004) Multimodel inference: understanding AIC and BIC in model selection. Sociological Methods Research, 33, 261–304. Campbell RD, Feber R, Macdonald DW, Gaywood MJ, Batty D (2010) The Scottish beaver trial: ecological monitoring of the European beaver Castor fiber and other riparian mammals – initial methodological protocols 2009. Scottish Natural Heritage Commissioned Report No. 383. Clark JS, Carpenter SR, Barber M et al. (2001) Ecological forecasts: an emerging imperative. Science, 293, 657–660. Clobert J, Lebreton JD, Allaine D (1987) A general approach to survival rate estimation by recaptures or resighting of marked birds. Ardea, 75, 133–142. Clobert J, Lebreton JD, Allaine D, Gaillard JD (1994) The estimation of age-specific breeding probabilities from recaptures or resightings in vertebrate populations. 2. Longitudinal models. Biometrics, 50, 375–387. Coulson T, Catchpole EA, Albon SD et al. (2001) Age, sex, density, winter weather, and population crashes in Soay sheep. Science, 292, 1528–1531. Cowley E, Sirwardena GM (2005) Long-term variation in survival rates of Sand Martins Riparia riparia: dependence on breeding and wintering ground weather, age and sex, and their population consequences. Bird Study, 52, 237–251. Dalgleish HJ, Koons DN, Adler PB (2010) Can life-history traits predict the response of for populations to changes in climate variability? Journal of Ecology, 98, 209–217. Drake JM (2005) Population effects of increased climate variation. Proceedings of the Royal Society B: Biological Sciences, 272, 1823–1827. Dyck AP, MacArthur RA (1993) Seasonal variation in the microclimate and gas composition of beaver lodges in a boreal environment. Journal of Mammalogy, 74, 180– 188. Easterling DR, Meehl GA, Parmesan C, Changnon SA, Karl TR, Mearns LO (2000) Climate Extremes: observations, Modeling, and Impacts. Science, 289, 2068–2074. Fay PA, Carlisle JD, Danner BT et al. (2002) Altered rainfall patterns, gas exchange, and growth in grasses and forbs. International Journal of Plant Sciences, 163, 549– 557. Fay PA, Carlisle JD, Knapp AK, Blair JM, Collins SL (2003) Productivity responses to altered rainfall patterns in a C4-dominated grassland. Oecologia, 137, 245–251. Fero´ O, Stephens PA, Barta Z, McNamara JM, Houston AI (2009) Optimal annual routines: new tools for conservation biology. Ecological Applications, 18, 1563–1577. Gaillard J-M, Yoccoz NG (2003) Temporal variation in survival of mammals: a case of environmental canalization? Ecology, 84, 3294–3306. Gillespie DOS, Russell AF, Lummaa V (2008) When fecundity does not equal fitness: evidence of an offspring quantity versus quality trade-off in pre-industrial humans. Proceedings of the Royal Society B: Biological Sciences, 275, 713–722. Groisman PY, Karl TR, Easterling DR et al. (1999) Changes in the probability of heavy precipitation: important indicators of climatic change. Climatic Change, 42, 243– 283. Grosbois V, Gimenez O, Gaillard J-M et al. (2008) Assessing the impact of climate variation on survival in vertebrate populations. Biological Reviews, 83, 357–399. Hanssen-Bauer I, Førland EJ (1994) Homogenizing long Norwegian precipitation series. Journal of Climate, 7, 1001–1013.

© 2012 Blackwell Publishing Ltd, Global Change Biology, doi: 10.1111/j.1365-2486.2012.02739.x

12 R . D . C A M P B E L L et al. Hartman G, To¨rnlo¨v S (2006) Influence of watercourse depth and width on dambuilding behaviour by Eurasian beaver (Castor fiber). Journal of Zoology, 268, 127–

O’Gorman PA, Schneider T (2009) The physical basis for increases in precipitation extremes in simulations of 21st-century climate change. P Natl Acad Sci USA, 106,

131. Le Houe´rou HN, Bingham RL, Skerbek W (1988) Relationship between the variability of primary production and the variability of annual precipitation in world arid lands. Journal of Arid Environments, 15, 1–18. Hughes L (2000) Biological consequences of global warming: is the signal already apparent? Trends in Ecology & Evolution, 15, 56–61. Hulme PE (2005) Adapting to climate change: is there scope for ecological manage-

14773–14777. Olstad O (1937) Beverens (Castor fiber) utbredelse i Norge. Statens viltundersøkelser. Særtry. Nytt Mag. Naturvidensk, 77, 217–273. Paine RT (1995) A conversation on refining the concept of keystone species. Conservation Biology, 9, 962–964. Parmesan C, Yohe G (2003) A globally coherent fingerprint of climate change impacts across natural systems. Nature, 421, 37–42.

ment in the face of a global threat? Journal of Applied Ecology, 42, 784–794. IPCC (2007) Climate Change 2007: Synthesis Report. Intergovernmental Panel on Climate Change, pp. 52. Jarema SI, Samson J, McGill BJ, Humphries MM (2009) Variation in abundance across a species’ range predicts climate change responses in the range interior will exceed those at the edge: a case study with North American beaver. Global Change Biology,

Payne NF (1984) Reproductive rates of beaver in Newfoundland. The Journal of Wildlife Management, 48, 912–917. Pradel R (1996) Utilization of Capture-Mark-Recapture for the study of recruitment and population growth rate. Biometrics, 52, 703–709. Promislow DEL, Harvey PH (1990) Living fast and dying young: a comparative analysis of life-history variation among mammals. Journal of Zoology, 220, 417–437.

15, 508–522. ˆ E, Nilsson C (2002) Responses of riparian plants to flooding in freeJohansson MA flowing and regulated boreal rivers: an experimental study. Journal of Applied Ecology, 39, 971–986. Johnson CN (1998) Species extinction and the relationship between distribution and abundance. Nature, 394, 272–274. Jolly GM (1965) Explicit estimates from capture-recapture data with both death and

Rosell F, Hovde B (2001) Methods of aquatic and terrestrial netting to capture Eurasian beavers. Wildlife Society Bulletin, 29, 269–274. Rosell F, Sanda JI (2006) Potential risks of olfactory signaling: the effect of predators on scent marking by beavers. Behavioral Ecology, 17, 897–904. Rosell F, Sun L (1999) Use of anal gland secretion to distinguish the two beaver species Castor canadensis and C. fiber. Wildlife Biology, 5, 119–123. Rosell F, Bozse´r O, Collen P, Parker H (2005) Ecological impact of beavers Castor fiber

immigration-stochastic model. Biometrika, 52, 225–247. Karl TR, Knight RW, Plummer N (1995) Trends in high-frequency climate variability in the twentieth century. Nature, 377, 217–220. King D (2005) Climate change: the science and the policy. Journal of Applied Ecology, 42, 779–783. Kjellstro¨m E, Ba¨rring L, Jacob D, Jones R, Lenderink G, Scha¨r C (2007) Modelling

and Castor canadensis and their ability to modify ecosystems. Mammal Review, 35, 248–276. Ruzzante DE, Walde SJ, Gosse JC, Cussac VE, Evelyn Habi T, Zemlak TS, Adams EDM (2008) Climate control on ancestral population dynamics: insight from Patagonian fish phylogeography. Molecular Ecology, 17, 2234–2244. Saether BE (1985) Annual variation in carcass weight of Norwegian moose in relation

daily temperature extremes: recent climate and future changes over Europe. Climatic Change, 81, 249–265. Kottek M, Grieser J, Beck C, Rudolf B, Rubel F (2006) World Map of the Ko¨ppen-Geiger climate classification updated. Meteorologische Zeitschrift, 15, 259–263. Lebreton JD, Burnham KP, Clobert J, Anderson DR (1992) Modeling survival and testing biological hypotheses using marked animals: a unified approach with case studies. Ecological Monographs, 62, 67–118.

to climate along a latitudinal gradient. The Journal of Wildlife Management, 49, 977– 983. Sandercook BK (2006) Estimation of demographic parameters from live-encounter data: a summary review. Journal of Wildlife Management, 70, 1504–1520. Scha¨r C, Vidale PL, Lu¨thi D, Frei C, Ha¨berli C, Liniger MA, Appenzeller C (2004) The role of increasing temperature variability in European summer heatwaves. Nature, 427, 332–336.

Luterbacher J, Dietrich D, Xoplaki E, Grosjean M, Wanner H (2004) European seasonal and annual temperature variability, trends, and extremes since 1500. Science, 303, 1499–1503. Macdonald DW, Newman C, Buesching CD, Nouvellet P (2010) Are badgers ‘under the weather’? direct and indirect impacts of climate variation on European badger (Meles meles) population dynamics. Global Change Biology, 16, 2913–2922.

Schlesinger WH (2006) Global change ecology. Trends in Ecology and Evolution, 21, 348 –351. Simmons LF (1990) Times-series decomposition using the sinusoidal model. International Journal of Forecasting, 6, 485–495. Smith MD (2011) The ecological role of climate extremes: current understanding and future prospects. Journal of Ecology, 99, 651–655.

McDermott R, Fowler JH, Smirnov O (2008) On the evolutionary origin of prospect theory preferences. The Journal of Politics, 70, 335–350. McLaughlin JF, Hellmann JJ, Boggs CL, Ehrlich PR (2002) Climate change hastens population extinctions. PNAS, 99, 6070–6074. McNamara JM, Webb JN, Collins EJ (1995) Dynamic optimization in fluctuating environments. Proceedings of the Royal Society B: Biological Sciences, 261, 279–284. McNamara J, Houston A, Collins E (2001) Optimality models in behavioral biology.

Smith P, Berdoy M, Smith RH (1994) Body weight and social dominance in anticoagulant-resistant rats. Crop Protection, 13, 311–315. Tuljapurkar S (2010) Environmental variance, population growth and evolution. Journal of Animal Ecology, 79, 1–3. Tuljapurkar S, Gaillard J-M, Coulson T (2009) From stochastic environments to life histories and back. Philosophical Transactions of the Royal Society B: Biological Sciences, 364, 1499–1509.

SIAM Review, 43, 413–466. McNew LB Jr, Woolf A (2005) Dispersal and Survival of Juvenile Beavers (Castor canadensis) in Southern Illinois. The American Midland Naturalist, 154, 217–228. Meehl GA, Tebaldi C (2004) More intense, more frequent, and longer lasting heat waves in the 21st century. Science, 305, 994–997. Min SK, Zhang XB, Zwiers FW, Hegerl GC (2011) Human contribution to more-

Walther G-RR, Post E, Convey P et al. (2002) Ecological responses to recent climate change. Nature, 416, 389–395. Webb B, Walling DE (1996) Long-term variability in the thermal impact of river impoundment and regulation. Applied Geography, 16, 211–223. Weltzin JF, Loik ME, Schwinning S et al. (2003) Assessing the response of terrestrial ecosystems to potential changes in precipitation. BioScience, 53, 941–952.

intense precipitation extremes. Nature, 470, 376–379. Mu˝ller-Schwarze D, Sun L (2003) The beaver: natural history of a wetlands engineer. Cornell University Press, New York. Nippert JB, Knapp AK, Briggs JM (2006) Intra-annual rainfall variability and grassland productivity: can the past predict the future? Plant Ecology, 184, 65–74. Nolet BA, Broftova L, Heitkonig IMA, Vorel A, Kostkan V (2005) Slow growth of a translocated beaver population partly due to a climatic shift in food quality. Oikos,

White G (2002) Discussion comments on: the use of auxiliary variables in capturerecapture modelling. An overview. Journal of Applied Statistics, 29, 103–106. White GC, Burnham KP (1999) Program mark: survival estimation from populations of marked animals. Bird Study, 46, S120–S139. Whittingham M, Stephens P, Bradbury R, Freckleton R (2006) Why do we still use stepwise modelling in ecology and behaviour? Journal of Animal Ecology, 75, 1182– 1189.

111, 632–640.

© 2012 Blackwell Publishing Ltd, Global Change Biology, doi: 10.1111/j.1365-2486.2012.02739.x

E F F E C T S O F C L I M A T E V A R I A B I L I T Y O N T H E B E A V E R 13 Supporting Information Additional Supporting Information may be found in the online version of this article: Appendix S1. Histogram showing the number of times individual beavers were captured during the course of the study. Appendix S2. Effect of short distance changes in weather station location on temperature records. Appendix S3. Details on the structure of the basic demographic model used in these analyses to estimate demographic parameters (survival rates or recruitment) with climatic variables correlations.

Please note: Wiley-Blackwell are not responsible for the content or functionality of any supporting materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.

© 2012 Blackwell Publishing Ltd, Global Change Biology, doi: 10.1111/j.1365-2486.2012.02739.x