Complex phenological changes and their consequences in the breeding success of a migratory bird, the white stork Ciconia ciconia

Transcription

1 Journal of Animal Ecology 2013, 82, doi: / Complex phenological changes and their consequences in the breeding success of a migratory bird, the white stork Ciconia ciconia Oscar Gordo 1 *, Piotr Tryjanowski 2, Jakub Z. Kosicki 3 and Miroslav Fulı n 4 1 Department of Zoology & Physical Anthropology, Complutense University of Madrid, Jose Antonio Novais 2, E-28040, Madrid, Spain; 2 Institute of Zoology, Poznan University of Life Science, Wojska Polskiego 71C, , Poznan, Poland; 3 Department of Avian Biology and Ecology, Faculty of Biology, Adam Mickiewicz University, Umultowska 89, , Poznan, Poland; and 4 East-Slovakian Museum Kosice, Hviezdoslavova 3, SK Kosice, Slovakia Summary 1. The timing of bird migration has shifted in response to climate change. However, few studies have linked the potential consequences of any phenological shift on individual fitness and even fewer have disentangled the role of plasticity and microevolution in the observed shifts. 2. The arrival date and breeding success of white storks (Ciconia ciconia) have been recorded since the 1880s in Slovakia. We used data for two periods ( and ), which were considered, respectively, as populations before and after the start of climate warming. About 4000 male and 2500 female arrival dates along with 3000 breeding attempts were studied. 3. Mean arrival dates did not differ between the two periods. During , males tended towards a slight delay for most fractions of arrival distribution. Protandry was reduced by 30% (144 days). 4. In both sexes, the early percentiles of the arrival distribution arrived later those years with warmer temperatures at the African wintering grounds, while late percentiles advanced their arrival when temperatures were higher in the European areas flown over during migration. 5. Mean breeding success of the Slovakian population has not changed since However, fecundity selection for arrival date reduced over the years: at the end of 1970s and 1980s, early breeders had more success than late breeders, but this seasonal trend disappeared towards the end of the study period. An early arrival and territory acquisition may have become less of an advantage due to the enhancement of feeding opportunities during the breeding season in recent decades. 6. A century ago, stork arrival varied spatially, with earlier arrivals at low altitudes, southern slopes and warmer and drier regions. This spatial variation mostly vanished, and at present, we found little correlations with topographical and climatic gradients. 7. We showed that long-term temporal changes in the timing of biological events may be complex because each fraction of a population and sex may show different temporal trends in their arrival dates. In addition, the effect of biotic and abiotic factors may change consistently in space and time, and thereby phenotypes value depends on the circumstances that are expressed due to its variable fitness consequences. Key-words: arrival date, climate change, long-term study, migration, plasticity, protandry, selection, sexual differences, Slovakia, temporal trends Introduction *Correspondence author. ogvilloslada@gmail.com Responses of organisms to climate change are becoming more and more apparent across the globe (Parmesan 2006; Rosenzweig et al. 2008). Among these, phenology has 2013 The Authors. Journal of Animal Ecology 2013 British Ecological Society

2 Migratory phenology and breeding success 1073 received much attention because shifts in the timing of biological events mirror climate fluctuations. In recent decades, warmer temperatures are hastening the spring activities of plants and animals (Root et al. 2005; Rosenzweig et al. 2008). The arrival of migratory birds has been a focus of special interest in phenological research due to negative fitness consequences found in those populations unable to keep pace with the advance of the trophic levels on which they rely during their reproductive period (Both et al. 2006, 2010; Jonzen, Hedenstr om & Lundberg 2007; Møller, Rubolini & Lehikoinen 2008; Jones & Cresswell 2010; Saino et al. 2011). Overall, spring arrivals of migrants are advancing by 4 days per decade (Lehikoinen & Sparks 2010). However, there are noteworthy differences among populations (Rubolini et al. 2007; Gordo & Doi 2012) as a result of the differential environmental pressures to which they are subjected in passage and wintering areas (Both & te Marvelde 2007; Gordo 2007), and some of their biological characteristics (Tryjanowski, Kuzniak & Sparks 2005; Spottiswoode, Tøttrup & Coppack 2006; Møller, Rubolini & Lehikoinen 2008; Vegvary et al. 2010). While it is essential to quantify accurately any shift in phenological traits (e.g. the number of days of advance or delay), it is necessary to scale this into phenological responses in trophic interactions to determine the adaptive meaning of phenological responses. Parallel phenological records for competitor species or lower/upper trophic levels would be necessary to put phenological responses into an ecological context (Visser & Both 2005; Both et al. 2009; Vatka, Orell & Rytk onen 2011). Similarly, a causal link needs to be established between phenology and individual fitness to put phenological responses into an evolutionary context (Both & Visser 2001; Both et al. 2006; Post & Forchhammer 2008; Gienapp & Bregnballe 2012; Lane et al. 2012). In contrast to research on the temporal trends of bird migratory phenology, topics such as the geographical variability of arrivals and the progression of the migratory wave across breeding territories remain poorly studied (Sparks & Braslavska 2001; Gordo, Sanz & Lobo 2008; Hulbert & Liang 2012). This imbalance is especially intriguing within a historical perspective because the aim of phenological studies of bird migration was originally the description of the spatial variability in arrival dates (Southern 1938). Unfortunately, at the time, there were no suitable tools to properly manage huge phenological data bases, and the introduction of new study techniques of bird migration, such as ringing, resulted in a premature abandonment of this topic (von Haartman & S oderholm- Tana 1983). Recent studies have demonstrated that spatial variability of bird arrivals closely follows climatic and geographical gradients, suggesting the existence of common environmental drivers for the spring progression of migratory birds until their arrival at the breeding grounds (Gordo, Sanz & Lobo 2007a,b, 2008). Moreover, in some species, high population density is related to an earlier arrival phenology (Gordo, Sanz & Lobo 2007b). These findings suggest complex interactions between the abiotic and biotic features of the environment affecting each population as well as the existence of spatial heterogeneity (i.e. among populations) in the strength of selection for an early arrival date at the breeding grounds (Gordo 2007; Gordo, Sanz & Lobo 2007a, 2008). Here, we have carried out a comprehensive study of the migratory phenology of a long-distance migratory bird species, the white stork (Ciconia ciconia, L.). The popularity of this migratory bird makes the white stork an excellent candidate for large-scale and long-term phenological studies. In Slovakia, a volunteer-based monitoring scheme has gathered information about the arrival and reproductive success of individuals since the 1880s. We used data for two periods ( and ) which are considered, respectively, as populations not subjected to, and subjected to, recent warming (Lapin 2004; Melo 2005; see Fig. S1, Supporting information). During the first period, only the arrival of males was recorded, while in recent decades, monitoring has been enhanced by including the arrival of the female and the number of fledged chicks in each nest. In our long-term study, we investigated whether or not any change in the timing of migration have been a result of plastic responses to climatic conditions during migration and wintering and how this affected breeding success of individuals depending on their arrival time. In addition, we also investigated the spatial variability of arrivals and the environmental factors affecting it. Materials and methods white stork data Data consist of 4005 male and 2549 female records of arrivals to their nesting sites in Slovakia (827 different sites). The arrival date was defined as the date when individuals were seen occupying the nest for the first time. Dates were extremely accurate because they were recorded by volunteers living near the nests or who visited them on a daily basis (Fulın et al. 2009). Our sample is representative of the Slovakian population owing to the fact that the overwhelming majority of storks build their nests on human-made structures, such as chimneys, roofs or electric pylons (Kalivodova, Valachovic & K urthy 1993; Fulın et al. 2009). Species misidentification is highly unlikely because of stork s popularity and characteristics, which are too distinct from any other species breeding in Slovakia. Studies with individually marked storks have demonstrated that males return and occupy nests before females (Cramp 1977; Tortosa & Redondo 1992; Barbraud, Barbraud & Barbraud 1999; Kosicki, Sparks & Tryjanowski 2004). Therefore, the first and the second individuals recorded for each nest were assumed to be male and female, respectively. Although there are records for the arrival of storks to Slovakia since 1882, we only used data for the periods and The volunteer-based monitoring scheme was interrupted due to the I World War, and it was not resumed until the mid-1970s. Records of female arrivals were available only during the later period. The number of fledglings (average 243, range 0 6) was available in 2916 breeding attempts during

3 1074 O. Gordo et al. the period and was used as a measure of individual breeding success. White storks do not refrain from breeding once the nest is occupied. Therefore, when the number of fledglings was zero, it indicated a true breeding failure between laying and fledgling. temporal changes in arrivals Temporal trends in stork arrivals during the period were analysed by quantile regression (Cade & Noon 2003) with year as explanatory variable using Blossom version W (Cade & Richards 2005). Temporal trends in the width of the distribution of arrival dates were explored by simple regression of the annual standard deviation of arrivals against the year to investigate whether arrival dates have become more synchronous. Furthermore, we calculated the difference between the arrival date of the male and the female of each nest and in each year (n = 2,538) and explored its temporal trend to determine changes in the degree of protandry (the earlier arrival of males relative to females). Temporal autocorrelation of data was checked by the autocorrelation function with a lag from 1 to 15 in all time series, and the significance of correlations was tested by the Ljung-Box Q statistic at each lag. We did not find evidence of temporal autocorrelation in any time series. Temporal changes in stork phenology over the past century were assessed by two independent comparisons of male arrival dates recorded in both study periods. First, a multivariate analysis of variance (MANOVA) was carried out to test for differences in average arrival dates of the 5th, 10th, 15th, etc. up to the 95th percentile (total of 19 percentiles as response variables) of each period (as categorical explanatory variable). Temporal autocorrelation was checked by the same procedure previously described, and we did not find evidence of temporal autocorrelation in any percentile time series (Fig. S2, Supporting information). In a second analysis, we performed a paired t-test for the median arrival date to 23 localities with records in both periods (Table S1, Supporting information). temperature and arrivals Based on accurate information of wintering areas, migration routes and migration periods of eastern European storks (Fiedler 2001; Berthold et al. 2002; Van den Bossche et al. 2002; Newton 2008), we selected 61 weather stations from southern, eastern and north-eastern Africa, the Middle East and south-eastern Europe available from the Global Historical Climatological Network version 2 data base (Peterson & Vose 1997; Peterson et al. 1998). Temperature time series for the period were gathered for the following three area/period combinations: January and February for stations in southern and eastern Africa; February and March for stations in Sudan and southern Egypt; March and April for stations located in the Middle East and south-eastern Europe (Fig. 1). We focused only on temperature because it is the most influential weather variable for soaring migratory birds available at this spatio-temporal scale. We calculated average temperatures in periods of two months to ensure the climate variables represented departure, passage and/or arrival of the entire population (i.e. from earliest to latest individuals). Southern and eastern Africa are the main wintering grounds of eastern European white storks, while the other areas represent the main migratory corridor for the eastern flyway. January and February are departure months from wintering areas, while migration lasts from February to April through the passage areas from Sudan to Slovakia (Reed & Lovejoy 1969; Leshem & Yom-Tov 1996; Newton 2008; Mestecaneanu & Mestecaneanu 2010). We used the same area/periods for both sexes because differences in their migratory schedule were too small (some days) for the monthly temporal resolution of our climatic variables. We adopted a dual approach to study temperature effects on stork migration. First, we made an accurate station-by-station assessment of temperature effects and the potential differences between sexes and study periods. For this aim, the median arrival date to Slovakia (only possible for females during ) was regressed against temperature time series, and the obtained slopes (referred to as sensitivity) were mapped and visually inspected. The difference in the sensitivity within each weather station was calculated between periods and sexes. When slopes were negative, values were multiplied by 1 to ensure comparability of differences obtained from negative and positive pairs of sensitivities to temperatures. Generalized least squares (GLS) models containing just an intercept were used to test whether or not the average of the calculated differences was >0. In the model for the differences between sexes, a positive and significant intercept would imply that females are significantly more sensitive to temperature than males. In the model for differences between periods, a positive intercept would imply that males are currently more sensitive than in the past. Only 11 out of the 58 weather stations with meteorological records during had also records for the period To account for the spatial autocorrelation of climate data, we used five types of spatial correlation structures (Gaussian, linear, exponential, spherical and rational quadratic) in the residuals of the GLS models with the latitude and longitude of weather stations as covariates (Zuur et al. 2009). We selected the exponential correlation structure without the nugget effect because they showed lowest value of the Akaike Information Criterion (AIC). Tests were one-tailed. GLS models were carried out with the nlme package version of R software (R Development Core Team 2011). In a second approach, we explored temperature effects for the 5th, 10th, 15th, etc. up to the 95th percentile both in males and females. For this purpose, we calculated a new set of more synthetic temperature variables. Weather stations were grouped following the previously reported spatio-temporal criteria. Therefore, a single average annual value was calculated for southern and eastern Africa, Sudan and southern Egypt, and the Middle East and south-eastern Europe (Fig. 1). The resulting single temperature time-series representative of each area/period were used as explanatory variables in multiple regression models with the arrival date of each percentile as the response variable (i.e. 19 independent models, one for each percentile). An information-theoretical approach was applied to determine the relative importance of the seven possible candidate models in each percentile. By a corrected AIC for low sample size (AIC c ), the Akaike weight (x i ) of each area/period was calculated and weighted model averaging was used for parameter estimation (Burnham & Anderson 2002). Due to the low number of available weather stations in Africa for the period , we did not run models for this period. breeding success The temporal trend during in the average number of fledged chicks per nest in the Slovakian population was

4 Migratory phenology and breeding success 1075 Fig. 1. Temporal trends of temperatures between 1977 and 2003 in wintering and passage areas of white storks breeding in Slovakia. In the left part of the figure, mean temperature graphs for each region. Temperatures were calculated as the average of all weather stations within each region during the months indicated in each case. Red lines show fitted linear regressions (b = slope, p = significance). In the right part, there is a map with spatially interpolated trends of temperatures for all weather stations. Black dots indicate weather stations location. In most cases, temperatures have increased in recent decades, but only 5 out of 46 significantly at a = 005. A few stations (12) recorded a cooling trend (none significant). See colour scale bar for correspondence between colour and the magnitude of slope. Slovakia is highlighted in green. determined by linear regression against year. In addition, we investigated the effect of migratory phenology and climate during the breeding period on the breeding success at the country-wide scale. In this analysis, the median arrival dates of males or females and monthly values of temperature and precipitation during May, June and July for Slovakia (Carrascal, Bautista & Lazaro 1993; Tortosa & Villafuerte 1999; Moritzi et al. 2001; Jovani & Tella 2004; Olsson 2007) were included as explanatory variables in multiple regression models where the annual average number of fledged chicks per nest was the dependent variable. The support of each competing possible model (2 7 = 128) was determined according to an information-theoretical approach (Burnham & Anderson 2002), as previously described. Those models with an AIC c three units greater than the best model were excluded from parameter estimation by weighted model averaging. To assess the interannual variability of arrival date effects on individual breeding success, we looked for heterogeneity in the slopes between arrivals and breeding success among years. A generalized linear mixed model (GLMM) with arrival date and year as fixed continuous variables and the number of chicks raised by each pair as a response variable was used. Fixed variables were standardized (l = 0, r = 1) prior to the analysis. Year was also included as a random factor to add a random intercept and slope to each year (Zuur et al. 2009). The interaction term (year x arrival) of fixed variables specifically tested for trends in the relationship between the arrival date and the number of chicks. A Poisson distribution with a log link function was used in the gmler function of the lme4 package version for R software (R Development Core Team 2011). The adaptive Gaussian Hermite approximation was used to improve the accuracy of model fitting (argument nagq = 5). Data showed some deviation from a canonical Poisson distribution because breeding success records showed underdispersion and had a small excess of zeros (i.e. breeding failures). This fact invalidates significance estimates (Zuur et al. 2009). We overcame this by simulating the distribution of the statistics by bootstrapping (Efron 1979). Random sampling with replacement was conducted within each year to keep the original analytical design. For each bootstrap sample, the model was run, and the obtained results were used to calculate the 25th and 975th percentiles of the bootstrap distribution. These percentiles form a good approximation of the 95% confidence interval of the estimated parameters of the fixed effects included in the model. We simulated bootstrap samples. Finally, the selection differential for arrival date was calculated both for males and females for all years using the following

5 1076 O. Gordo et al. formula: Selection differential ¼ P n i¼1 ðx i fr i Þ= P n i¼1 fr i x s where x i is the arrival date of an individual i (male or female) in a certain year, fr i is the relative number of chicks fledged by the individual i; x is the median arrival date of the population (of males or females) in the same year, and s is the standard deviation of the arrival dates for that year. The fr i was calculated as: fr i ¼ f i P n fi i¼1 n where fi is the number of chicks fledged by an individual i in a certain year, and n is the number of individuals breeding that year. We calculated the relative number of fledglings to avoid the potential scale effect of the different reproductive output attained each year in the population. In the same way, we scaled the selection differential to the standard deviation of the trait to get a comparable magnitude of the deviation of the weighted arrival date with regard to the median arrival date among years. The more negative the selection differential, the higher the relative breeding success for early vs. late arriving individuals (van Noordwijk, McCleery & Perrins 1995). The temporal trend of selection differentials was studied by a multiple regression model with year, sex and their interaction as explanatory variables. The interaction between year and sex tested specifically for the homogeneity of the slopes between males and females. spatial patterns and modelling of arrivals A grid layer for Slovakia with a resolution of 0042º (which corresponds to 150 s, or ~47 km) was created for the study of spatial migratory patterns of white storks. The median date of all arrival dates in each cell was calculated. Three different phenological maps were generated: (i) male arrivals during ; (ii) male arrivals during ; and (iii) female arrivals during Arrivals were modelled using a set of 11 explanatory variables, which were calculated for each cell using Idrisi 32 software (Clark Labs 2001). Variables were as follows: mean altitude, altitude range, terrain slope, aspect (mean direction of the slope), annual mean temperature, annual range of temperature, annual sum of precipitation, precipitation seasonality (coefficient of variation), white stork breeding population density, latitude and longitude. Topographical and climatic variables are related and represent important predictors of spatial patterns of arrival dates of migratory birds (Sparks & Braslavska 2001; Gordo, Sanz & Lobo 2007a,b, 2008). We expected these variables to be relevant because there are marked topographical gradients which impose highly contrasting climatic conditions across Slovakia in spite of its relatively small area (c km 2 ). White stork breeding population density was calculated as the number of nests in a radius of 15 km around the centre of each cell found in the last national census of Slovakia carried out in 2004 (M. Fulın, unpublished data). A previous study demonstrated that storks in Iberia breeding in higher density areas arrived earlier on average (Gordo, Sanz & Lobo 2007b). This variable was included only in models for the period because the storks have spread throughout Slovakia during the last century, and consequently, the distribution of 2004 is not representative of past times. The resampling radius of 15 km around each cell was used to match the km resolution of the 2004 national census data. Latitude and longitude together with their interaction were included to account for other potential spatial gradients (Legendre & Legendre 1998). We did not include polynomial terms in any of the predictors because a preliminary exploration of the relationships with arrival data did not show non-linear relationships. All explanatory variables were standardized (l = 0, r = 1) prior to analyses. We applied Partial Least Square Regression (PLSR; Carrascal, Galvan & Gordo 2009) for modelling the spatial variability of arrivals. PLSR combines original predictors into a number of orthogonal components designed ad hoc to maximize the variance explained in the response variable. Components account for successively smaller portions of variance in the response variables, and consequently, original multidimensionality can be reduced to one or a few relevant components. Components were computed by non-linear iterative partial least squares (NIPALS algorithm), and their significance was established by cross-validation. Components are interpreted by the weights for each of the original predictors. Weights provide the magnitude and sign of the effect of each original explanatory variable in the PLSR components. We used weights to make comparisons about the relative contribution of each predictor in the models obtained for each sex and period because the sum of their squares is equal to 1 (see details in Carrascal, Galvan & Gordo 2009). To improve the comparability between the male arrival model of and that of , the latter model was repeated without the nest density variable and only with those localities to the east of 196 E longitude. We selected those sites from the eastern half of the country because our data set for the period only has records eastwards from 196 E longitude. Differences in the spatial extent of data sets may influence the relative importance of explanatory variables in biological gradients (Chust et al. 2004; Rahbek 2005). Residuals were examined for possible spatial structure by calculating Moran s I autocorrelation coefficient with a Bonferroni-corrected significance level (Rangel, Diniz-Filho & Bini 2006) against twenty classes separated by a lag distance of 30 km (from 30 to 600 km). These analyses were conducted with GS+ version Most of the statistical analyses were conducted with Statistica version 7; otherwise, the specific software and version has been cited. Results temporal changes in phenology arrived on average the April 1 (940SD, range March 12 to May 2), while females arrived the April 5 (1032SD, range April 5 to May 15). During the period , all fractions of the Slovakian population of white storks tended to delay arrival both in males and females (Fig. 2). However, the confidence intervals showed that these tendencies were statistically different from zero at a = 005 only for those percentiles earlier than the 70th in males and only between the 5th and 15th percentiles in females. The annual standard deviation of the arrival dates showed a negative tendency both in males (slope = 0086, t 29 = 1760, P = 0089) and females (slope = 0058, t 29 = 1528, P = 0136). Therefore, the arrival date distribution tended to be narrower

6 Migratory phenology and breeding success 1077 Slope year (d yr 1 ) Slope year (d yr 1 ) Quantile Quantile Fig. 2. Temporal trends of male and female white stork arrivals to their nests in Slovakia during the period Values are estimated slopes from quantile regression with year as an explanatory variable. Negative values are advancements and positive values are delays. Stippled lines indicate 95% confidence intervals. because early fractions of the population delayed their arrival more than late fractions. During the period , males arrived on average 47 days earlier than their females to the nests, but this difference decreased by 144 days in the last three decades (slope = 0048 days/year t 29 = 2074, P = 0047). Arrival dates of males did not differ significantly between and (MANOVA: Wilks k = 0678, F 18,31 = 0817, P = 0668), although late fractions of the population during the latter period tended to arrive earlier than a century ago (Fig. S3). Furthermore, median arrival dates observed in those localities with records in both periods were not significantly different (paired t-test: t 22 = 053, P = 0599). temperature effects on arrival dates Regression coefficients (slopes) from simple regressions between the median arrival date to Slovakia during and temperatures in the weather stations of the wintering and passage areas of Africa and Europe showed similar spatial patterns in both sexes (Fig. 3; Table S2). Storks arrived earlier in years with warmer springs in south-eastern Europe and in the Middle East [range of the correlation coefficients (r) for males: 0268 (P = 0176) to 0582 (P = 0001); for females: 0220 (P = 0270) to 0651 (P < 0001)]. Interestingly, the strongest effect of and sensitivity to temperature was found in the three nearest stations to the Bosphorus Straits, a bottleneck in the migratory route of the eastern European populations of soaring birds (Fiedler 2001; Van den Bossche et al. 2002; Newton 2008) (average for males: r = 0533, slope = 112 days/ C, P = 0004; average for females: r = 0615, slope = 166 days/ C, P < 0001]. High temperatures from Kenya to South Africa were related to later arrivals at the breeding grounds, but only a few of the correlations from the southernmost sites were statistically significant [range of r for males: 0126 (P = 0531) to 0525 (P = 0005); for females: (P = 0472) to 0464 (P = 0015); see Table S2]. were on average 0308 days/ C more sensitive to climate than males (GLS model: t 57 = 2054, P = 0022). The effect of temperature on male arrivals was similar between periods: negative in Europe near to Slovakia and positive in the wintering areas of east and south Africa (Fig. 3a vs. b). Sensitivity was on average 0376 days/ C lower during the period compared with the later period, but differences were not statistically significant (GLS model: t 10 = 0636, P = 0269). The analysis of percentiles showed that the temperature effect was not homogeneous over all fractions of the population (Fig. 4). In agreement with the general pattern found for the median arrival date (Fig. 3), most percentiles advanced their arrivals in years with warm springs in the Middle East and south-eastern Europe and delayed their arrivals after warm winters in eastern and southern Africa (see regression coefficients in Fig. 4c,d). These regions were the most important, while the effect of temperature in the passage region of southern Egypt and Sudan was negligible both for males and females (note the low Akaike weights and the regression coefficients close to zero in the Fig. 4a,b). However, the relative importance of temperatures in eastern and southern Africa vs. the Middle East and Europe varied according to the percentile examined. Late percentiles were markedly influenced by temperatures in the last section of the route, while early percentiles were more influenced by temperatures in the African departure areas. Such differences in the relative influence of each part of the route for early and late individuals were even apparent in a comparison between males and females. are somewhat earlier than females, and interestingly, the relative importance of temperature in departure areas in the male models was higher in later percentiles than in the female models (Fig. 4c,d). However, temperatures in the last part of the route showed higher Akaike weights for females than for males. Finally, the explanatory capacity of multiple regression models had similar magnitudes and diminished in later percentiles both in males and females (Fig. 4e). The best fitted models (r 2 > 30%) were obtained between the 20th and 30th percentiles.

7 1078 O. Gordo et al. (a) (b) (c) Fig. 3. Maps of the regression coefficients between median arrival dates of storks in Slovakia and temperature time series in their wintering and passage areas. Map (a) is arrivals of males during , (b) arrivals of males during and (c) arrivals of females during Each dot represents a weather station. Regression coefficients have been interpolated among stations to create a continuous surface of spatial variation and improve visualization. Slovakia is highlighted in green. See colour scale bar for correspondence between colour and the magnitude of slope. temporal trends and climate effect in breeding success The annual average number of fledglings per nest did not change from 1977 to 2007 (r = 0141, t 29 = 0743, P = 0464) but was significantly correlated with both the annual median male (r = 0581, t 29 = 3709, P < 0001) and female (r = 0557, t 29 = 3306, P = 0003) arrival dates (Fig. 5). This single effect of arrivals was not mediated by climatic conditions during the breeding period, since the best multiple regression models for the breeding success included both arrival dates and temperatures in spring (Table 1). Warm temperatures in May also enhanced the productivity of storks. Whereas in the past early arrival positively affected reproductive success, in recent years, reproductive success is mostly unrelated to arrival time (GLMM interaction arrival x year was positive, Table 2; see Figs S4 and S5). This result can also be presented as a positive trend in selection differentials on arrival dates in males (r = 0508, t 29 = 3046, P = 00049) and females (r = 0522, t 29 = 3038, P = 00050) during the period (Fig. 6). Differences in the slope values between sexes were not statistically significant (GLM interaction year x sex: F 1,58 = 1194, P = 0279). patterns of spatial variability Storks in the period arrived consistently early at sites of low altitude, and with colder, moister and more marked seasonality in local climate (Table 3). The sum of the two components accounted for up to 47%, which suggest a marked spatial variability in arrival dates (Fig. 7a). In recent decades, the topographical and climatic gradients had similar effects on arrival dates both in males and females (Table 3), but the explained variance declined strongly (r 2 < 5%), resulting in a spatial variability without any evident pattern (Fig. 7b, c). Differences in the explained variability (r 2 ) became even greater when the model for male arrivals during was rerun by including only records from the eastern part of Slovakia and excluding the effect of population density (Table 3). Despite such manifest difference, weights for the first component were alike (Spearman rank correlation between variable weights: r S = 0645, t 9 = 2535, P = 0032), and the relative percentage of explained variability by each group of variables was consequently similar during both periods (Table 3). Spatial autocorrelation was not detected in the residuals from any of the models (Fig. S7). Discussion Many recent studies have devoted special attention to the impact of climate change on bird migratory phenology (Lehikoinen & Sparks 2010). Usually, they have focused on the study of temporal trends of arrival dates of first individuals and the effect of weather at the study site (Gordo 2007). Applying this habitual approach, we would reach few conclusions about the migratory behaviour of

8 Migratory phenology and breeding success 1079 (a) Variable weights (ω i ) Region 1 Region 2 Region 3 (b) Variable weights (ω i ) Fractions of the population (%) Fractions of the population (%) (c) (d) Fig. 4. Results for multiple regression models between white stork arrivals during and climatic variables in wintering and spring migration areas. The explanatory variables were temperatures in the Middle East and south-western Europe (Region 1), Sudan and southern Egypt (Region 2), and southern and eastern Africa (Region 3). All the possible models were made for each percentile and ranked according to the corrected Akaike Information Criterion (AIC c ). The Akaike weights (x i ) for each variable were calculated and are plotted in the upper graphs. The parameters of each explanatory variable (mid graphs) and the explanatory capacity of models (r 2 ; bottom graph) were estimated according to model averaging and are plotted for each percentile of the population. All results are shown both for males and females. Regression coefficient (days/ºc) (e) 0 40 r Fractions of the population (%) Fractions of the population (%) Regression coefficient (days/ºc) Fractions of the population (%) Average number of fledglings Mar 1-Apr 7-Apr 13-Apr Median arrival date Fig. 5. Relationship between the annual mean number of fledged chicks and the annual median arrival date of males and females of the Slovakian population of white storks. Lines comes from a simple linear regression model. the Slovakian storks during the last century: they have not shifted their arrivals because they have not responded to the increase in temperature in Slovakia. A comprehensive analysis of phenological data by studying the entire arrival distribution of the population, differences between sexes, links with reproductive success, and the spatial progression of spring arrival has discovered complex and to some extent cryptic changes in its phenology. In contrast to stork populations from Poland (Ptaszyk et al. 2003), Spain (Gordo & Sanz 2006) and Lithuania (Zalakevicius et al. 2006), the overall schedule of arrivals to Slovakia has not changed directionally in the long term. However, we did observe differential trends within different fractions of the population (e.g. early vs. late individuals, or males vs. females). In the studied species, the shape of the arrival date distribution has become narrower because early individuals are delaying their arrivals more than late ones. First arrival dates have been criticized because they have been shown not to fully represent the behaviour of the whole population (Tøttrup, Thorup & Rahbek 2006; Miller-Rushing et al. 2008). Our results suggest that this problem may extend to trends in measures of central tendency (mean or median), because they may not take into account changes in the width of distributions of phenological dates. This fact highlights the necessity of a thorough analysis of the entire distribution of arrival dates to understand how phenological date distributions move within the calendar and also how they vary in their shape (Gordo & Sanz 2009). Temperature effects on arrival date suggest that storks have some plasticity to adjust their migration to conditions encountered en route. Theoretically, this was

9 1080 O. Gordo et al. Table 1. Multiple regression models for the annual average number of fledged chicks in the Slovakian population of storks ( ). The explanatory variables were the median annual arrival date (of males or females), and monthly temperature and precipitation during May, June and July. Only those models with an increase of AIC c (DAIC c ) below 3 are shown. The AIC c and the weight (x i ) of each model are also shown. The weights (x), parameters (b) and standard errors (SE) for the explanatory variables are shown at the bottom of the tables. The explanatory capacity (R 2 ) of models is in the right-bottom corner of tables Temperature Precipitation Model Arrival date May June July May June July AICc AAICc x i, 1 X X X X X X X X X X X X X X X X X X x b R 2 = SE X X X X X X X X X X X X X X X X X X x b l R 2 = SE Table 2. Results for the fixed effects of the generalized linear mixed model for the breeding success of storks in Slovakia, The model also included a random intercept and slope effect of the arrival date within each year. P-values represent the probability of 0 in the bootstrap distribution of the estimates. N = 2844 for males and 2199 for females Fixed effects Estimate Bootstrap 95 % Confidence interval Arrival date ( , ) < Year ( , ) Arrival date 9 Year (0.0162, ) Arrival date ( , ) < Year ( , ) Arrival date 9 Year (0.0080, ) expected because soaring birds are less controlled by an endogenous programme to provide individuals with a more flexible environment-dependent behaviour during migration (Berthold et al. 2002; Shamoun-Baranes et al. 2003; Newton 2008; Mestecaneanu & Mestecaneanu 2011). Interestingly, such plasticity varied among fractions of the population and even slightly between sexes (Fig. 4). We may speculate that early individuals (i.e. more affected by climate) represent older and more experienced birds (Vergara, Aguirre & Fernandez-Cruz 2007), which are able to profit favourable weather conditions to P increase their migratory progression. On the other hand, the stragglers could be young or solitary individuals (Reed & Lovejoy 1969), which rely more on their internal cues or cannot take advantage of social interactions, respectively (Liechti, Ehrich & Bruderer 1996; Chernetsov, Berthold & Querner 2004). The opposite effect of temperatures at departure time and in the last part of the migratory route concurs with the ecology of most trans-saharan species. Heat in the wintering arid grounds of eastern Africa is associated with drier and more restrictive conditions for feeding (Tøttrup et al. 2012). However, no or no substantial, fattening or hyperphagia have been observed in the white stork during migration (Van den Bossche et al. 2002; Newton 2008; Zwarts et al. 2009). Thus, migration onset, while notably variable among individuals and among years within the same individual (Berthold et al. 2002; Van den Bossche et al. 2002), does not seem constrained by physiological preparation depending on food resources. In spite of this fact, ecological restrictive conditions induced by heat and drought at the end of the wintering phase in Africa may affect stork migration by two mechanisms linked to food availability. The winter distribution of eastern storks in Africa is related to insect availability (Verheyen 1950; Dallinga & Schoenmakers 1989; Van den Bossche et al. 2002; Zwarts et al. 2009), such locusts, which are strongly dependent on the amount and timing of rainfalls (Todd et al. 2002). Food is erratic, both spatially and

10 Migratory phenology and breeding success 1081 temporally, and storks exhibit a nomadic behaviour searching for rain-related outbreaks of insects and thus for the best foraging sites (Newton 2008; Zwarts et al. 2009). During warmer (and drier) years in East Africa, storks spend the winter in larger proportion in the southernmost wintering areas following the monsoon peak, which progresses southwards from the equator during the austral summer (Zwarts et al. 2009). Thus, late arrivals preceded by warm temperatures in eastern and southern Africa during the winter would represent longer migratory journeys as consequence of an overall more southern winter distribution. In addition, food scarcity may act directly on the physical condition of individuals by starvation. If ecological conditions during the winter have been restrictive, individuals may need to spend longer periods in the stopover sites en route to ensure their survival until their breeding areas (Gordo 2007; Tøttrup et al. 2012). Since journeys are longer and probably more exhausting for individuals and/or individuals may be already in poor body condition, these phenomena may have carry-over effects on breeding success (Berthold et al. 2002; Zwarts et al. 2009). This may explain why population productivity was lower in late years (Fig. 5) independently of weather condition in Slovakia during reproduction. In the Middle East and southeastern Europe, warm temperatures at the beginning of the spring are linked to improved migration conditions (Zalakevicius et al. 2006; Both & te Marvelde 2007; Gordo 2007; Gordo & Sanz 2008; Tøttrup et al. 2008). This means for a soaring bird more favourable conditions for flight thanks to increased thermal convection (Liechti, Ehrich & Bruderer 1996; Shamoun-Baranes et al. 2003; Mestecaneanu & Mestecaneanu 2011). Therefore, the last part of the migratory route could be covered in fewer days by increased distances flown every day and/or minimal delay in crossing barriers such as the Jubal or Bosphorus Straits. The small differences in temporal trends between the sexes are probably caused by a different influence of climate along the migratory route led to a decrease in the degree of protandry in the Slovakian stork population. This empirical evidence contrasts with the predicted increase in protandry in migratory birds under a global warming scenario (Møller 2004; Spottiswoode, Tøttrup & Coppack 2006; but see Rainio et al. 2007; Tøttrup & Thorup 2008; Baub ock et al. 2012). However, our observation is expected if directional selection towards earlier arrival dates is relaxed. Currently, storks have no pressure to arrive early because first arrivals do not provide benefits in terms of increased number of offspring. This relaxation of selective pressures on arrival date was stronger in males (e.g. compare the magnitude of the interaction of the arrival date with the year in Table 2) than in females. This concurs with the observed stronger delay of arrivals in males than in females. The observed relaxation in the effect of arrival on breeding success suggests that the timing of breeding in storks does not require an accurate synchronization with the seasonality of the environment at their breeding grounds, in contrast to some migrants (Both & Visser 2001; Both et al. 2006, 2009; Jonzen, Hedenstr om & Lundberg 2007; Saino et al. 2011; Gienapp & Bregnballe 2012). Breeding success in the white stork depends on the absolute abundance of food during spring (Moritzi et al. 2001; Tortosa, Caballero & Reyes-Lopez 2002; Tryjanowski & Kuzniak 2002; Denac 2006a) and weather-related mortality of chicks (Carrascal, Bautista & Lazaro 1993; Tortosa & Villafuerte 1999; Jovani & Tella 2004; Olsson 2007) instead of a temporal match with seasonal peaks of resources, as in other bird species (van Noordwijk, McCleery & Perrins 1995; Both & Visser 2001; Visser, Both & Lambrechts 2004; Nussey et al. 2005; Pearce- Higgins, Yalden & Whittingham 2005; Vatka, Orell & Rytk onen 2011). Nevertheless, a seasonal decline in the breeding success has been observed in some populations (Tortosa, Perez & Hillstr om 2003; Tryjanowski et al. 2004; Tryjanowski & Sparks 2008; but see Grishchenko 2006), and we indeed found this date-effect in some years. Such differences in the breeding success between early and late individuals are due to differences in age, experience, quality and status rather than to a within-season decline of food supplies (Vergara, Aguirre & Fernandez-Cruz 2007; Fulın et al. 2009). The relaxation of selection on arrival date may be a consequence of a homogenization of the quality of breeding territories. The organized map of arrivals shown by storks one century ago (Fig. 7a) could be reflecting a despotic distribution of individuals across Slovakia. The earliest, which are the best phenotypically individuals, would occupy the best territories and would have higher reproductive success (Vergara & Aguirre 2006). The best and most productive territories could be placed in those warmer and driest areas, and for this reason, the earliest individuals were recorded there one century ago (Table 3). This seems plausible since a cold and moist climate may be limiting the breeding success by a trade-off in the time budget between provisioning and chicks sheltering from weather (Moritzi et al. 2001). However, this trade-off could disappear or, at least, could become of little relevance for individuals, if balance between weather and food supplies has changed. Climate change in Slovakia during the last century (see Fig. S1; Lapin 2004; Melo 2005) may have improved weather conditions during the breeding season, especially in the formerly lower quality territories (colder and wetter), by reducing the adults time invested in sheltering and enhancing chick survival. In addition, food is probably more accessible at present due to exploitation of rubbish dumps (Tortosa, Caballero & Reyes-Lopez 2002; Tortosa, Perez & Hillstr om 2003; Kruszyk & Ciach 2010), changes in livestock farming practices (Tryjanowski, Jerzak & Radkiewicz 2005), changes in agricultural landscape (Dallinga & Schoenmakers 1989) or reduction in the use of pesticides (Newton 2008), providing feeding opportunities to all breeding pairs over their necessary threshold for rear their

11 1082 O. Gordo et al. Selection differential for arrival dates Year Fig. 6. Temporal trend of selection differentials for arrival dates in white storks. Data for males and females are shown separately. offspring. If food is no longer a constraint, weather impact on breeding may become minimal (Denac 2006a), and consequently, any variation in territory quality based on climate may disappear. Competition for territory occupation by early arrivals would no matter anymore, and some expected consequences would be a relaxation of selection for early arrivals, reduction in protandry, and finally a free distribution of individuals according to their arrival phenology, that is, what we find currently (Fig. 7b,c). The white stork spreading during the last decades (Sæther et al. 2006; Fulın et al. 2009) supports the idea of an increase in the carrying capacity of Slovakia by an enhancement of resources availability. However, it can be hypothesized that the observed loss of the spatial organization of arrival dates is linked to dispersal processes. During the early stages of range expansion and occupation of new territories, newly established individuals may express a migratory phenotype not fully suited to the local conditions. This process seems especially plausible in the white stork, since distances reported for natal dispersal in this species (Chernetsov et al. 2006; Kania 2006; Olsson 2007) would easily overcome the strong, but small-scale (i.e. hundreds of kilometres), environmental gradients of Slovakia. In fact, dispersal from core breeding areas has been postulated as a way for the notable recuperation of the European stork population since the 1980s (Zwarts et al. 2009). In addition, dispersal would be more frequent in a growing population since it is a denso-dependent process (Itonaga et al. 2011). Such decoupling between Table 3. Results of the four partial least square regression models of spatial variation in arrival dates per grid cell in Slovakian storks. For each variable, the weight in each component is shown (significant in bold). The explanatory capacity (r 2 ) and significance (P) of each component are also shown. At the bottom of the table, the percentage of the variance explained by each group of variables in each component is given. The subset model for males during the period included only data to the east of longitude 196 E and did not include nest density to improve comparison with the male model for (see methods for details) Between sexes Between periods Variable Male Male Female Male (subset) 1 st Component 1 st Component 1 st Component 2nd Component 1 st Component Topographical Altitude (mean) Altitude range Aspect Slope Climatic Temperature (annual mean) Temperature range Precipitation (annual sum) Precipitation seasonality Biological Number of nests (in 15 km) Spatial Longitude Latitude Longitude 9 Latitude r P < < < % of explained variability Topographical Climatic Biological Spatial

12 Migratory phenology and breeding success 1083 (a) (b) (c) Aguirre & Fernandez-Cruz 2007). Finally, although some classic experiments with storks demonstrated the genetic basis of bird migration (Sch uz et al. 1971), we still do not know anything about the genes controlling the migratory phenology of this species, how much heritable they are, and how the interaction genotype environment is for determining the expressed phenotype. In conclusion, long-lived animals, such as the white stork, are expected to rely more on their plasticity than microevolutionary responses on selection to adapt to rapid changes in their environment (Morris et al. 2008). We showed that different parts of the population vary in their response to environmental changes that differ both in space and time of the annual cycle. Fitness consequences of variation in arrival date diminished over time, which likely was the result of ecological changes at the breeding grounds (homogenizing spatial variation), rather than to changes in synchrony with other trophic levels. Acknowledgements Fig. 7. Maps of spatial variation in arrival dates of white storks to Slovakia. Data have been interpolated among the study sites to create a continuous surface and improve the visualization of spatial arrival variability. Because of limited data, male arrivals during the period has been restricted to the eastern sector of Slovakia (>196 E longitude). See scale colour bar at the bottom of the figure for correspondences between colours and dates. The total number of cells (n) available for the spatial analyses is shown. arrival dates of incomers and local conditions, as a result of the spread of individuals, would modify the spatial patterns of arrival dates, reflecting a transitory situation in the evolutionary history of a migratory species. One could expect that natural selection may create spatial gradients in phenology once again in the future by purging those individuals mistimed with their environment. However, this does not seem plausible because of the following reasons: (i) The notion of phenological mismatching is not relevant for this species, which has a long-lasting breeding period and does not rely in seasonal food peaks for breeding. (ii) We have demonstrated that selection on the arrival date is relaxing. Therefore, it is hard to imagine how spatial patterns would be re-established in future generations, if most of the variability in the arrival dates is not related to a differential reproductive success of individuals. (iii) Storks exhibit a notable behavioural plasticity for adapting to local environmental conditions, for instance by breeding in larger territories (i.e. diminishing population density; Barbraud, Barbraud & Barbraud 1999; Denac 2006b), seeking the most appropriate landscape structure in the areas surrounding the nest (Denac 2006a) or simply improving their experience with age (Vergara & Aguirre 2006; Vergara et al. 2006; Vergara, We thank all volunteer observers who participated in this study (www. bociany.sk). Christiaan Both and three anonymous referees carried out a thorough review that greatly improved the manuscript. Tim Sparks, Emma Nelson and Sarah Young help us with English editing. The study was partially supported by a grant from the Polish Ministry of Science N N to P.T. O.G. received a contract of the Juan de la Cierva program (ref. JCI ). References Barbraud, C., Barbraud, J.C. & Barbraud, M. (1999) Population dynamics of the White Stork Ciconia ciconia in western France. Ibis, 141, Baub ock, L., Miller-Rushing, A.J., Primack, R.B., Lloyd Evans, T.L. & Wasserman, F.E. (2012) Climate change does not affect protandry in seven passerines in North America. The Wilson Journal of Ornithology, 124, Berthold, P., Van den Bossche, W., Jakubiec, Z., Kaatz, M. & Querner, U. (2002) Long-term satellite tracking sheds light upon variable migration strategies of White Storks (Ciconia ciconia). Journal f ur Ornithologie, 143, Both, C. & te Marvelde, L. (2007) Spatial and temporal variation in climate change and their effects on timing of avian breeding and migration throughout Europe. Climate Research, 35, Both, C. & Visser, M.E. (2001) Adjustment to climate change is constrained by arrival date in a long-distance migrant bird. Nature, 411, Both, C., Bouwhuis, S., Lessells, C.M. & Visser, M.E. (2006) Climate change and population declines in a long-distance migratory bird. Nature, 441, Both, C., van Asch, M., Bijlsma, R.G., van den Burg, A.B. & Visser, M.E. (2009) Climate change and unequal phenological changes across four trophic levels: constraints or adaptations? Journal of Animal Ecology, 78, Both, C., Van Turnhout, C.A.M., Bijlsma, R.G., Siepel, H., Van Strien, A.J. & Foppen, R. (2010) Avian population consequences of climate change are most severe for long-distance migrants in seasonal habitats. Proceedings of the Royal Society of London - Series B, 277, Burnham, K.P. & Anderson, D.R. (2002) Model Selection and Multimodel Inference, 2nd edn. Springer-Verlag, New York. Cade, B.S. & Noon, B.R. (2003) A gentle introduction to quantitative regression for ecologists. Frontiers in Ecology and Environment, 1, Cade, B.S. & Richards, J.D. (2005) User Manual for Blossom Statistical Software. U. S. Geological Survey, Reston. Carrascal, L.M., Bautista, L.M. & Lazaro, E. (1993) Geographical variation in the density of the white stork Ciconia ciconia in Spain - influence of habitat structure and climate. Biological Conservation, 65,