Climate effects on population fluctuations of the white-throated dipper Cinclus cinclus

Authors


Correspondence author. E-mail: anna.nilsson@bio.uio.no

Summary

1. Climate change may have profound consequences for many organisms. We have studied fluctuations in a population of the white-throated dipper Cinclus cinclus during 31 years (1978–2008) in a river system in southern Norway in relation to both large-scale and local weather conditions occurring during the non-breeding season.

2. Multiple regression and partial least squares regression were used to model the growth rate of the population, accounting for population size in the previous year.

3. Population growth was influenced by North Atlantic Oscillation (NAO), mean winter temperature, precipitation and timing of ice formation on the main lake in the river system in autumn. These variables explained 84% of the variation in population growth over the 31 -year study period.

4. Local winter conditions played a prominent role in explaining the population fluctuations, which is plausible because the dipper depends on open water for foraging. In the study area, winters can be harsh and rivers and lakes may freeze and severely affect the subsequent population size of the dipper in spring.

5. The breeding population of the dipper does not seem yet to have reached a level where all possible territories in the area have been occupied, even after mild winters, and the estimated carrying capacity is also decidedly lower (66 breeding pairs) than the number of available territories. If the trend of milder winters continues, the population might increase in the future. However, strong climate variation is expected to continue in the future, and hence periods of rapid growth of the dipper population will probably be followed by severe declines.

Introduction

Research indicates that climate change might be one factor (among others such as habitat change) involved in the recent population declines in a number of bird species (Jetz, Wilcove & Dobson 2007; Pimm 2008). This may well be because of changing conditions prior to breeding, affecting survival or mediating effects to the breeding phase via food resources or the timing of important life-history events. For instance, migratory birds not responding sufficiently to climate change by advancing their spring migration to the breeding grounds have been subject to population declines (Møller, Rubolini & Lehikoinen 2008; Both et al. 2010). In seasonal environments, reproductive success can be dependent on the timing relative to the peak abundance of available food resources, and a mismatch in timing of the two events can have serious effects on local populations, as shown for the pied flycatcher Ficedula hypoleuca (Both & Visser 2001; Both et al. 2006). Warmer winters can at intermediate and northern latitudes generally be expected to result in increased survival of resident birds as well as earlier spring conditions and improved primary productivity, but can also lead to deteriorating reproductive success with population consequences (Waite & Strickland 2006). In conclusion, how populations vary over time in response to stochastic factors in the non-breeding season is fundamental to understand how populations will respond to climate change.

Climate is a collection of weather variables and can be measured on different temporal and spatial scales ranging from local weather observations to global measures of for example air surface temperatures across hemispheres (Easterling et al. 1997). The North Atlantic Oscillation (NAO) index captures fluctuations in the atmospheric pressure differences between the subtropic (centred at the Azores) and subpolar North Atlantic (centred at Iceland; Hurrell 1995). NAO is often used as a large-scale descriptor of winter conditions in the Northern Hemisphere, as it is considered a good proxy for a suite of covarying weather variables within biologically relevant time windows (Stenseth & Mysterud 2005). For instance, positive index values in winter are in north-western Europe generally associated with warm and wet weather and negative with dry and cold weather. However, on a geographically smaller scale, climate is most likely to act on population dynamics through local weather (Stenseth et al. 2002; Stenseth & Mysterud 2005).

For altricial birds (birds with offspring that are incapable of moving around on their own soon after hatching, essentially all passerines, woodpeckers, birds of prey, owls, etc.) in northern temperate regions, climate effects on population fluctuations are predicted to be strongest during the non-breeding season, (Sæther, Sutherland & Engen 2004). These would be direct (i.e. not lagged at the scale of years) effects of climatic variation and mediated by for instance survival. Weather conditions during the winter months in the Northern Hemisphere may therefore have a large impact on the long-term trends in the population dynamics of these species (Rodenhouse et al. 2008). To investigate which climatic variables are most influential in determining the population dynamics, we have studied a white-throated dipper Cinclus cinclus population in southernmost Norway, during the period 1978–2008. The growth rate of this population has been shown to depend on both stochastic and density-dependent factors during 1978–1997, when about half of the environmental variance in population size was explained by mean winter temperature (Sæther et al. 2000). As our study adds 11 more years to the picture and we are particularly interested in the climatic variables, our aim is to conduct a more detailed analysis to further elucidate the influence of climate on the population fluctuations. In the light of the current climate change (and the longer study period), local climate might have changed dipper conditions by for instance reducing ice cover or altering river flow rates or, indirectly, by affecting water quality, prey detectability and/or food abundance. Our objective is thus to explain the fluctuations in population size by modelling the impact of climatic variables (large-scale as well as local weather) during the non-breeding season on the growth rate of the population.

Materials and methods

Study species

The white-throated dipper occupies a special niche among birds by catching invertebrate prey items in running freshwater. It is a small passerine (50–70 g) distributed in mountainous regions across the Palaearctic. Breeding is restricted to the immediate proximity of fast-flowing rapids. In autumn, part of the Scandinavian population undertakes short-distance migrations, while other birds remain resident on or close by the breeding grounds over winter (Cramp 1988). The exact proportions of migrants and residents are currently not known.

Study population and data assimilation

The field work was conducted in the river system of Lyngdalselva, in the county of Vest-Agder in southern Norway (58° 08′–58° 40′ N, 6° 56′– 7° 20′ E; Fig. S1). The study was initialised in 1973, but the fieldwork was not regarded as standardised until 1978, when virtually all possible breeding territories in the river system were known and regularly visited (for a complementary analyses excluding a small number of sites that was discovered after 1979, see Appendix S1). From then, observed population sizes reflect population fluctuations and not increased field effort. The study area reaches approximately 60 km inland from the river mouth and the river system of Lyngdalselva drains about 680 km2. Locations of breeding territories span from sea level to an altitude of about 600 m. a. s. l. In the middle of the river system, the river is interrupted by the Lake Lygne (surface area 7·56 km2). Breeding territories are concentrated in areas of rapids, although more gently flowing parts of the river and even lake shores are commonly used as feeding grounds. Besides not containing any artificial nest boxes for the species, the river system of Lyngdalselva is not utilized for water power and is protected against further development, thus providing a natural study system for dipper ecology.

Because of the complexity of the breeding system, including polygyny, re-lays and second clutches, the population size is defined as the annual number of breeding females. The total number of available breeding territories that have been occupied or where breeding attempts have been recorded in the river system of Lyngdalselva is 159. All known breeding territories along the river and tributaries are visited in the morning hours during the nest building season and scanned for dipper activity. Breeding is defined as positive when the nest building is completed. Dippers build outer nests, which can be used year after year, but the inner nests complete with inner lining are new for each breeding attempt. Within the river system, the breeding outcome of all occupied nests is recorded and nearly all young are ringed.

The North Atlantic Oscillation (NAO) index was provided by the Climate and Global Dynamics Division, National Center for Atmospheric Research, Boulder, CO, USA (Hurrell 1995; http://www.cgd.ucar.edu/cas/jhurrell/indices.html). Mean and minimum temperature and precipitation data were provided by the Norwegian Meteorological Institute (http://www.yr.no/verdata/; mean Dec–Feb) for the observation stations Konsmo-Eikeland (58° 15′ N, 7° 19′ E; 1978–1989), Konsmo-Hægeland (58° 16′ N, 7° 18′ E; 1990) and Konsmo-Høyland (58° 16′ N, 7° 22′ E; 1993–2008) located in the immediate proximity of the study area (Fig. S1). Unfortunately, the weather data contains missing values, as observation stations were temporarily completely closed down (1991–1992). The Norwegian Water Resources and Energy Directorate provided us with information on timing of ice formation in autumn and ice break-up in spring and duration of ice cover on Lake Lygne. Water flow (m3 s−1; mean March–May) at two observation sites (Møska and Tingvatn) in the river system was provided by the same source. Although water flow is measured during early stages of the breeding season, it is a variable that is strongly influenced by winter conditions, i.e. the snow depth but naturally also the melting progress in spring. Møska is the largest tributary that joins the main river close to the river mouth in Lyngdalsfjorden, while Tingvatn is just downstream from Lake Lygne in the main river, almost in the centre of the river system.

Statistics

The duration of ice cover at Lake Lygne was measured as the proportion of winter days (181 days, Nov–Apr) with ice, which then was log-transformed to reach normality. Exceptionally mild winters with no ice cover were excluded (1989–1990, 1992), as this condition was unsatisfyingly represented by the transformation, and ice variables per se obviously become irrelevant when there is no ice. The timing of ice cover and ice break-up on the lake Lygne are measures of onset and termination of harsh conditions for the species feeding underwater, 1st of November was set as day 1 and dates were sequentially numbered and used as a measure of first observed date with connective ice cover on the lake, resulting in the variable timing of initiation of ice cover. Ice cover break-up was defined as when there no longer was a connective ice layer across the southern end of the lake, where the outlet is. For the timing of ice cover break-up in spring, 1st of March was set as day 1. Timing of ice cover break-up on lake Lygne was logit-transformed. Years with no ice cover were treated as years with missing data for these variables in the same manner as for the duration of ice cover on the lake. Other investigated variables were water flow at Møska and Tingvatn, as these could reflect winter snow conditions and melting, the extended winter NAO index (Dec–Mar) as a measure of large-scale climate, and mean and minimum winter temperature and precipitation, as these variables are local measures of critical factors for the survival in terms of metabolic costs, as well as access to underwater food supplies. Water flow was log-transformed to reach normality, and mean winter temperature was log-transformed to yield a linear relationship with the population growth rate (GAMM).

We fitted a range of likely linear models (using maximum likelihood; function ‘glm’) with the breeding population size lnXt as the response variable and the population size the year before, lnXt-1, included as an offset. This is equivalent to modelling annual breeding population growth ln (Xt/Xt-1). As breeding population growth is dependent on the population size the year before, Xt-1 (Fig. 1; GAMM: R2 = 0·17, e.d.f. = 1, F = 3·98, P = 0·041), it was also included in the list of explanatory variables, along with environmental variables (yielding a stochastic version of the discrete Ricker model with covariates; see e.g. Turchin 2003):

image

where inline image are environmental covariates. To reduce the risk of confounding effects because of trends over time, we removed the linear trend and used the yearly residuals of all included predictor variables. The models were then evaluated with Akaike Information Criteria corrected for small sample size, AICc (Burnham & Anderson 1998). Carrying capacity was defined as the population size when the population is at equilibrium, ln (Xt/Xt-1) = 0, and was estimated from the equation above by substituting coefficients and climatic variables with the parameter estimates and mean covariate values of the best-fitting model.

Figure 1.

 Annual population growth, ln(Xt/Xt-1), in relation to the population size (defined as the number of breeding females) the year before, lnXt-1, controlled for the autocorrelation in annual population growth. Gray scale indicates mean winter temperature (symbol X denotes years with missing data on mean winter temperature).

Generalized additive mixed models (GAMM) of the relationship between annual growth and annual residuals of each investigated variables, controlled for the autocorrelation in the error structure, were fitted to investigate bivariate relationships and their functional form. We preferred the GAMM framework rather than ordinary GAMs, as it yields more conservative smooths and allows modelling of autocorrelated errors.

Because climatic variables strongly covary at local scales, we also took a multivariate approach to identify statistically independent environmental components. A partial least squares (PLS) regression (Martens & Næs 1989; Carrascal, Galván & Gordo 2009) was performed, using yearly residuals from a linear regression of annual breeding population growth on population size the year before as the response variable and the yearly residuals of the environmental variables as predictors. This technique decomposes the predictor–response covariance matrix into orthogonal components of decreasing covariance explained, analogously to principal component analysis. PLS components were interpreted in terms of component loadings and Pearson’s correlation coefficient between environmental variables and component scores. Also, we performed a decomposition of the total sum-of-squares for each PLS components into sums-of-squares associated with each environmental variable. This sheds light on how the individual variables affect the response variable by ways of the independent environmental components identified by PLS regression.

For investigating the predictive power of our models, we used the first 20 years of data (1978–1997) to predict the fluctuations in population growth the last 11 years (1998–2008). We fitted a range of likely linear models (‘glm’ approach) based on the first 20 years and evaluated them with AICc. The predictions of the most suitable model were compared with the observed fluctuations in population growth and also the predictions from the model performing best on the complete time series, but here fitted to the first 20 years of data.

All analyses were performed using the statistical programming environment ‘R’, version 2.8.1 (R Development Core Team 2008), with add-on package ‘pls’ (Wehrens & Mevik 2007) for PLS regression.

Results

The breeding population size of dippers in Lyngdalselva has increased since the study was initiated in 1978 (Fig. 2; d.f. = 29, t = 2·2, b = 1·19, R2 = 0·14, P = 0·04), although the increase has levelled off from 1985 onwards (1985–2008: d.f. = 22, t = 0·2, R= 0·001, P = 0·9). The population size has fluctuated in the range of 21–117 breeding pairs, with a mean of 69 pairs. Since 1978, mean winter temperature (Dec–Feb) has increased on average 0·1°C per year (Fig. 2; b = 0·11, d.f. = 27, t = 3·0, P = 0·006). Also, minimum temperature has increased, with 0·16°C per year (b = 0·16, d.f. = 28, t = 2·3, P = 0·03), as has total precipitation, with 11 mm per year (b = 11·0, t = 2·9, d.f. = 27, P = 0·008). Annual population growth rate was correlated with annual residuals of NAO, mean and minimum winter temperature, winter precipitation and duration of ice cover and tended to be correlated with timing of ice-laying (Fig. 3; NAO: r = 0·378, P = 0·04; mean temp: r = 0·651, P < 0·001; minimum temp: r = 0·559, P = 0·002; precipitation: r = 0·437, P = 0·02; duration of ice cover: r = 0·448, P = 0·02; ice-laying: r = 0·339, P = 0·084). Other investigated climate variables were not significantly correlated with annual population growth rate (Fig. 3).

Figure 2.

 Fluctuations in annual breeding population size (defined as the number of breeding females), Xt, in Lyngdalselva (dots and solid line) and fluctuations in mean winter temperature (broken line) since 1978. The annual breeding population size increased in the beginning of the study period and thereafter levelled off (broken line).

Figure 3.

 Exploratory scatterplots of the investigated weather variables and annual population growth (defined as the population of breeding females), ln(Xt/Xt-1). Generalized additive mixed models (GAMM), controlled for the autocorrelation in the error structure, were fitted to investigate trends in relationships and their functional form. Residuals of all weather variables are used. Significant relationships are marked with trend line and shaded confidence intervals. The nearly significant relationship with timing of ice-laying has significance level written out.

When evaluating the different models explaining annual population growth, we found the model consisting of population size the year before (Xt-1), mean winter temperature, winter precipitation, NAO and timing of ice laying on the main lake in the system to be the best model (lowest AICc; Table 1). Population size the year before had the largest impact, but mean winter temperature and precipitation also had high influence in terms of deviance reduction and partial R2 (Table 2). Using mean values of the included predictor variables (±1 standard deviation), carrying capacity was estimated to 66 breeding pairs for this model (±27). This model explained 84% of the variation in annual population growth (Fig. 4a).

Table 1.   Alternative models for explaining the fluctuations in annual population growth ln(Xt/Xt-1) in Lyngdalselva 1978–2008 with population size the year before, Xt-1, and climatic factors in a linear model. Number of estimated parameters (K), Akaike Information Criteria corrected for small sample size (AICc), AICc differences (Δi) and explanatory power of models (R2) are given
ModelKAICcΔiR2
  1. NAO, North Atlantic Oscillation.

Xt-1 +  mean temp + precipitation + NAO + timing of ice610·8800·84
Xt-1 +  mean temp + precipitation411·610·730·77
Xt-1 + mean temp312·791·910·73
Xt-1 + mean temp + precipitation + NAO + timing of ice + duration of ice + termination of ice813·722·840·85
Xt-1 +  mean temp + timing of ice + precipitation514·183·300·79
Xt-1 +  mean temp + duration of ice416·015·120·74
Xt-1 +  mean temp + min temp + termination of ice + precipitation616·755·860·79
Xt-1 +  mean temp + timing of ice417·006·120·73
Xt-1 + mean temp + timing of ice + precipitation + duration of ice + termination of ice718·687·800·81
Xt-1 +  mean temp + water flow Tingvatn + Møska520·379·490·69
Xt-1 +  mean temp + timing of ice + duration of ice + termination of ice621·7710·890·75
Mean temp230·8219·940·44
Xt-1 +  mean temp + NAO + timing of ice + precipitation + min temp + duration of ice + termination of ice + water flow Tingvatn + Møska1035·1824·300·81
Min temp236·8025·920·31
Xt-1239·0528·170·26
NAO243·4532·570·14
Table 2.   Annual population growth ln(Xt/Xt-1) explained by the best model (with the lowest AICc) consisting of population size the year before, Xt-1, and climatic predictor variables. Partial R2 and estimated slopes are included in the analysis of deviance table. R2 of the final model is 0·84. Observe that estimated slopes are not standardised and reflect the effect of the annual residuals of (transformed) predictor variables
 DeviancePartial R2Estimated slope
  1. NAO, North Atlantic Oscillation.

Intercept6·93  
Xt-12·510·66−0·013
Mean winter temp2·560·390·558
Precipitation0·360·380·002
NAO0·250·23−0·106
Timing of ice laying0·110·23−0·003
Figure 4.

 The observed breeding population size (defined as the number of breeding females), Xt (filled squares and thick solid line) and (a) fitted values from the linear model with the lowest AICc (grey dots and broken line) and from the model including the first five partial least squares regression components, PLS C1–C5 (open dots and dotted line) over time, and (b) fitted values from the model including first PLS regression component, PLS C1 (grey dots and broken line) and the second PLS regression component, PLS C2 (open dots and dotted line).

The second best model also contained population size the year before, mean temperature and precipitation (Table 1), but not NAO and timing of ice laying. It did not differ more than 0·73 units from the model with the lowest AICc, but explained less of the observed variation in population growth (77%, Table 1). Population size the year before, mean temperature and precipitation were also the variables with the largest impact on the best model, and we therefore conclude that both models are equally suitable. Also the third best model, with only population size the year before and mean temperature, did not differ with more than 1·91 AICc units, although it explained still less of the observed variation in population growth (Table 1).

We found the PLS regression components reflecting climatic variation during the study period to account for 75% of the variation in population growth after the population size in the previous year was accounted for (Table 3). Factor loadings and correlations between component scores and environmental variables (Table 3) suggest that the first PLS component, C1, can be interpreted as a regional measure (winter weather) of winter harshness with a larger-scale element (NAO), capturing a correlation pattern between NAO, winter temperatures and precipitation (Fig. 4b; Table 3). Mean winter temperature especially had high Pearson correlation coefficients with C1 (Table 3). Contributions of the explanatory variables to the sum-of-squares for each PLS components illustrated major contributions of weather variables to C1, namely winter temperatures and precipitation (Table S1). C2 reflected primarily the ice conditions (Fig. 4b), where timing of ice laying and ice break-up in spring explained a large proportion of the explained variance in C2 (Table S1). C3 reflected the duration of ice cover and water flow in the centre of the study system in conjunction with NAO, where both variables were major explanatory variables in C3 (Table S1). C4 reflected primarily the hydrological conditions, where water flow was the main contributor to the explained variance in C4 (Table S1). C5 was more difficult to interpret, but seems to be an indicator of general winter harshness (mean and minimum temperature, NAO, water flow at the tributary Møska and duration of ice cover). C5, however, explains only a small proportion of the covariance in the predictor variables (Table 3).

Table 3.   Factor loadings of the predictor variables for each partial least squares regression (PLS) component, C1–C9, explaining population growth after population size the previous year in Lyngdalsvassdraget (1978–2008) was accounted for. Numbers in brackets are Pearson’s correlation coefficients between each weather variable and PLS component scores. The total percentage of variation (response explanation) explained by the PLS regression model is 75·08 %, whereof the first five components listed below explained 74·85 %. Predictor explanation is the cumulative percentage of the total covariance in the predictor variables explained by each of the predictor variables C1–C9
 C1C2C3C4C5C6C7C8C9
  1. NAO, North Atlantic Oscillation.

NAO0·430 (0·85)0·148 (0·12)−0·550 (−0·32)0·162 (0·17)−0·363 (−0·21)0·017 (−0·01)−0·279 (−0·26)−0·337 (−0·16)−0·025 (−0·01)
Mean winter temp0·453 (0·89)0·293 (0·24)−0·106 (−0·06)−0·121 (−0·13)0·374 (0·21)0·492 (0·25)−0·054 (−0·05)−0·276 (−0·13)−0·178 (−0·07)
Min winter temp0·402 (0·79)0·086 (0·07)0·309 (0·18)−0·075 (−0·08)−0·415 (−0·24)0·135 (0·07)0·491 (0·47)−0·037 (−0·02)0·548 (0·23)
Precipitation0·429 (0·84)0·291 (0·24)−0·105 (−0·06)0·245 (0·26)0·211 (0·12)−0·730 (−0·36)−0·034 (−0·03)0·244 (0·11)−0·088 (−0·04)
Water flow at Tingvatn0·088 (0·17)0·074 (0·06)−0·767 (−0·45)0·754 (0·80)−0·341 (−0·19)0·287 (0·14)0·141 (0·13)0·463 (0·21)−0·177 (−0·07)
Water flow at Møska−0·188 (−0·37)−0·151 (−0·12)−0·086 (−0·05)0·778 (0·82)−0·333 (−0·19)−0·068 (−0·03)0·234 (0·22)−0·619 (−0·29)0·074 (0·03)
Timing ice break-up−0·300 (−0·59)0·749 (0·61)−0·119 (−0·07)−0·145 (−0·15)−0·204 (−0·12)−0·077 (−0·04)0·446 (0·42)−0·207 (−0·10)−0·527 (−0·22)
Duration of ice cover0·322 (0·63)−0·654 (−0·53)0·732 (0·43)−0·020 (−0·02)−0·406 (−0·23)0·170 (0·08)−0·120 (−0·11)0·108 (0·05)−0·565 (−0·24)
Timing of ice laying0·235 (0·46)−0·516 (−0·42)−0·064 (−0·04)−0·202 (−0·21)0·424 (0·24)−0·371 (−0·19)0·702 (0·67)−0·312 (−0·15)−0·162 (−0·07)
Predictor expl44·1455·0361·0477·7181·6584·5695·6198·04100·00
Response expl54·6763·5171·7073·1474·8575·0775·0775·0875·08

Most of the covariation in the predictor variables was explained by C1 (45%; Table 3). PLS regression components 1–5 explained 74·9% of the variation in population growth given population size the previous year (Table 3) and predicted population fluctuations from 1978–2008 fairly accurately (Fig. 4a). The PLS regression components C1 and C2 reflects regional weather versus local winter conditions, such as ice variables, and together explain 63·5% out of the total variation in the response variable. The individual performance of PLS regression components C1 and C2 (fitted values) could therefore be contrasted and compared with the observed fluctuations in population size (Fig. 4b).

With such a long time series, it is possible to use the first 20 years (1978–1997) for predicting the last 11 years (1998–2008). The model that best predicted population fluctuations during the first 20 years consisted of the population size the year before and mean winter temperature (AICc = −5·33, R2 = 0·90) (c.f. Sæther et al. 2000). The second best model differed with more than two units (Δi = 2·58, R= 0·90) from the best model. The best model for the whole time period (1978–2008), including population size the year before, mean winter temperature, precipitation, NAO and timing of initiation of ice cover, had a lower AICc when fitted during the first 20 years (Δi = 5·20, R= 0·91). However, its ability to predict the population fluctuations during the last 11 years was better than the most suitable model during the first 20 years (AICc = 12·86, R= 0·66; Δi = 3·78, R= 0·52; Fig. 5). It is evident that mean winter temperature is important in explaining population fluctuations in this system.

Figure 5.

 Annual population growth (defined as the population of breeding females), ln(Xt/Xt-1) (filled squares and solid line) over time and fitted values from the most suitable model for the first 20 years (open dots and dotted line) and from the most suitable model for the whole time period (grey dots and broken line), fitted to the first 20 years. The first 20 years (1978–1997) are then used to predict the fluctuations in population growth during the 11 last years (1998–2008). The break point in time is illustrated by the dotted vertical line.

Discussion

We found a large effect of large-scale climate and local weather factors on the population growth rate of the white-throated dipper in Lyngdalselva. The population fluctuated mainly in response to changes in population size the year before, mean winter temperature and precipitation, but also to NAO and timing of initiation of ice cover on the main lake in the system (Table 1). The contribution of large-scale predictors such as NAO in explaining population fluctuations was relatively small compared to the influence of regional weather such as winter temperatures and precipitation and indicators of the local winter conditions such as parameters describing the timing and extent of ice cover (Tables 3 and S1). An earlier study in the same system emphasized the potential for population increase and the influence of mean winter temperature (Sæther et al. 2000). With our study, we confirm a slight population increase, but the population has fluctuated widely, from 21 to 117 pairs, since the mid-1980s, and our main conclusion is on the importance of environmental variation contributing to the population fluctuations. For instance, in response to very cold winters, the population decreases dramatically. Thus, the effect of winter temperature will primarily be in explaining population decline, while other variables, such as population size the year before and precipitation, will be better in explaining periods of population growth.

In many animal species, climate is often found to explain fluctuations in natural population size well (Mysterud et al. 2001; Stenseth et al. 2003; Hallett et al. 2004; Berryman & Lima 2006; Lima, Previtali & Meserve 2006; Coulson et al. 2008). In our study population in Lyngdalselva, the breeding population size of white-throated dippers readily fluctuated in direct response to climatic conditions during the winter. In the framework provided by Royama (1992), climatic factors with a direct effect on survival or reproduction are said to have a “vertical perturbation effect” on the relationship between population growth rate and population size, i.e. climatic effects can be evaluated independently of population size. When climatic effects are mediated by a limiting resource, they will have a “lateral perturbation effect”, and climate must be evaluated in relation to population size. An evaluation of different models for the Soay sheep Ovis aries demonstrate the importance of population size followed by direct effects of climate in understanding population fluctuations, where interactions between population size and climate proved to be the least important factor (Coulson et al. 2008). In the dipper population, there is a strong direct effect of the studied climatic factors, causing a vertical perturbation effect, which together with the population size are the major drivers of the system. We do not analyse lateral perturbation effects, because these effects are likely to be of less importance in this system. Also, there were no interactions between climatic variables and population size indicating lateral effects – possibly with the exception of precipitation, but this could also be because of nonlinearity. In addition, we lack data on variables that are likely to reflect such effects in our study system.

During the study period, local climate in the study area has changed. Mean winter temperature has increased, as well as minimum temperature and total precipitation recorded during winter. Although local variability can be large at short time-scales, the geographical extent of spatial autocorrelation in these variables is large when they are aggregated at a scale of months (Stenseth & Mysterud 2005). Hence, we expect our local trends to reflect general trends for the region as a whole, such as the recent trends in NAO (Hurrell 1995). The population increased through the early 1990s, a period also characterized by an increasing trend in the NAO and has since then fluctuated without a trend. Over the whole period, there is a small overall population increase since the start of the study, although this increase is relying on the particularly low population sizes in the first years of study when winter climate was relatively hard. When winters have become warmer and wetter, the main river is less likely to freeze in winter, and as the food source is under water, it therefore remains accessible during the coldest and most critical periods. Mostly likely, the mild winters have directly facilitated winter survival, an important factor influencing population fluctuations (Royama 1992; Sæther, Sutherland & Engen 2004).

Climatic conditions reflected by high NAO values were associated with large breeding population sizes in the white-throated dipper population (Table 2). High NAO values are also known to have a positive effect on other birds such as the breeding densities of collared flycatchers Ficedula albicollis, a long-distance migratory bird, in central Europe (Sætre, Post & Král 1999). Similarly, in the Southern Hemisphere, high values of the Southern Oscillation Index have a positive effect on the survival of the marine bird species the blue petrel Halobaena caerulea occurring in waters off Antarctica, as cold sea-surface temperatures are associated with higher food abundance than in warm years (Barbraud & Weimerskirch 2003). In mammalian species such as red deer Cervus elaphus on the west coast of Norway, high NAO values also have a direct positive effect through increased survival (less snow and warmer weather), but the main influence is the positive effect on deer abundance through the indirect effects of improved spring and early summer conditions (increased plant biomass and quality; Forchhammer et al. 1998; Mysterud et al. 2003; Mysterud et al. 2008).

Large-scale climatic predictors such as NAO have often been surprisingly good predictors of processes on a population level compared to local weather factors (Hallett et al. 2004; but see also Sillett, Holmes & Sherry 2000 for a study with the El Niño Southern Oscillation, ENSO). The coarse measurement of local weather averaged over months often used by ecologists has been speculated as one reason for NAO to perform better in explaining population fluctuations (Stenseth et al. 2002, 2003; Hallett et al. 2004; Stenseth & Mysterud 2005), especially when the determining weather factor might fluctuate between years (Hallett et al. 2004). Another reason could be that long-time averages of individual variables might not encompass ecologically relevant time periods. However, in a number of studies including our own, local weather is apparently also an accurate predictor of population processes (Keller & Van Noordwijk 1994; Gaillard et al. 1997). In our study, the regional as well as the local climatic conditions are captured by separate and statistically independent PLS component and their combination seem to fine-tune the performance of the model explaining the observed population fluctuations (Fig. 4b).

In the light of the prediction of future increased frequency of extreme weather conditions (Beniston et al. 2007), we would expect the population to fluctuate widely in response; periods of rapid growth will be followed by population crashes, particularly so in response to very cold winters. Climatic variability has a large impact on population fluctuations, as illustrated by the good fit between observed population fluctuations and model predictions (Fig. 4a, b). The two analytical methods used reflect the importance of certain key climatic components (GLM approach), but also the multitude of covarying climatic components that are involved in determining winter conditions. The population thus constitutes a good study system for studying climate impacts, despite containing an unknown fraction of migrants (see also Sæther et al. 2000). The breeding population seems to be particularly sensitive to severe winter conditions, where birds might starve and freeze to death. The prey community might also change as a direct effect of a response to changes in temperature, precipitation, river flow, etc or as an indirect consequence of changes in for instance vegetation or decomposition rates, in response to changed environmental conditions (Giller & Malmqvist 1998; Petersen, Gíslason & Vought 2006; Durance & Ormerod 2007). This could have implications for the population size of white-throated dippers. The overall long-term trend in response to shifting mean values in climate is likely to be positive in this population, because of the rapid response to climatic conditions. However, the number of known potential breeding territories in this system is higher than has ever been occupied in a single breeding season (159 vs. 117) and considerably higher than the estimated carrying capacity of 66 breeding pairs, but there is a large variation in how often a territory is occupied, suggesting an influence of site-dependent population regulation (McPeek et al. 2001). The reason for this discrepancy may be the climatic fluctuations, causing repeated, severe population declines after harsh winters, where some milder winters seem to be needed for the population to completely recover (Fig. 2). We therefore conclude that climate change will have a strong direct impact on the white-throated dipper population dynamics, because of the immediate response to climatic fluctuations by the study population.

Acknowledgements

We thank all the people involved in the field work, and Kyrre Kausrud, Luis Cadahia and Morten Helberg for helpful discussions, and Leif Christian Stige for help with graphical representation. The long-term study of the biology of breeding dippers was conducted with support from the Directorate for Nature Management. The Swedish Research Council (VR) provided financial support (to ALKN). We are also grateful to Mauricio Lima and an anonymous reviewer for constructive comments on a previous version of this paper. The field work was conducted with respect for the animals’ well-being and complies with the laws and regulations for animals used in research in Norway.

Ancillary