Are animals shrinking due to climate change?

Temperature-mediated selection on body mass in mountain

wagtails

January 30, 2019

Jorinde Prokosch1, Zephne Bernitz2, Herman Bernitz3, Birgit Erni4, Res Altwegg4,5

1. Department of Mathematical Sciences, Norwegian University of Science and Technology, 7034

Trondheim, Norway

2. Veterinary Consultant, Middelburg MPU, South Africa

3. Department of Oral Pathology and Oral Biology, School of Dentistry, University of Pretoria,

Pretoria, South Africa

4. Statistics in Ecology, Environment and Conservation, Department of Statistical Sciences, Uni- versity of Cape Town, Rondebosch 7701, South Africa

5. African Climate and Development Initiative, University of Cape Town, Rondebosch 7701, South

Africa

Author Contributions

ZB, HB and RA conceptualized this study based on ideas of the late Steven Piper. ZB and HB collected data. JP, RA and BE analysed the data. JP and RA wrote the manuscript. All authors contributed to revisions.

1 Abstract

Climate change appears to affect body size of animals whose optimal size in part depends on temperature. However, attribution of observed body size changes to climate change requires an understanding of the selective pressures acting on body size under different temperatures. We examined the link between temperature and body mass in a population of mountain wagtails

(Motacilla clara) in KwaZulu-Natal, South Africa, between 1976 and 1999, where temperature increased by 0.18◦C. The wagtails became lighter by 0.035g per year. Partitioning this trend, we found that only a quarter of the effect (0.009g / year) was due to individuals losing weight and three quarters (0.027g / year) was due to lighter individuals replacing heavier ones. Only the latter component was statistically significant. Apparently, the wagtails were reacting to selection for reduced weight. Examining survival, we found that selection was temperature- mediated, i.e. lighter individuals survived better under high temperatures whereas heavier individuals survived better under low temperatures. Our results thus support the hypothesis that temperature drove the decline in body mass in this wagtail population and provides one of the first demonstrations of the selective forces underlying such trends.

Key-words: body mass; climate change; survival; Motacilla clara; Bergmann’s rule

2 Introduction

Climate change is having fundamental impacts on ecosystems around the globe (Walther et al.,

2002; Parmesan & Yohe, 2003). One of the main components of climate change is temperature, which has increased globally by approximately 0.9◦C over the past 100 years (Blunden & Arndt,

2017). This change has affected many aspects of biological organisation (Parmesan, 2006), including range shifts (Thomas & Lennon, 1999), changes in phenology (Both & te Marvelde, 2007) and species interactions (Visser et al., 2004).

An intriguing idea is that rising temperature could lead to changes in animals’ body sizes

(Yom-Tov, 2001). The idea rests on the observation, known as Bergmann’s rule, that endothermic animals tend to be smaller in size in warmer environments than in colder regions (Bergmann, 1847).

Bergmann’s rule appears to hold broadly at least across mammals and birds (Ashton et al., 2000;

Ashton, 2002) but the mechanism responsible for this pattern is debated (reviewed in Blackburn et al., 1999; Watt et al., 2010).

Bergmann’s explanation for his observations was based on heat conservation (Bergmann, 1847).

He argued that homoeothermic animals living in relatively cold environments would profit from a relatively smaller surface area-to-volume ratio in order to radiate less of their body heat. How- ever, McNab (1971) pointed out that larger individuals have a larger surface and lose more heat in absolute terms. In a resource-limited environment, being large is therefore not necessarily ad- vantageous. More recent evidence shows that animals living in hot environments can be limited by the speed at which they can dissipate their body heat and higher area-to-volume ratios allow them to maintain higher activity levels without overheating (Speakman & Król, 2010). According to this theory, Bergmann’s rule could be the result of selection for small size in hot environments.

Alternatively, Bergmann’s rule may not be directly temperature driven. For example, larger indi- viduals could be more starvation tolerant during cold seasons with low food availability (Lindstedt

& Boyce, 1985; Goodman et al., 2012).

The role of temperature in driving patterns of body size is not clear (but see van Gils et al.,

2016). Given the importance of body size for the organisation of biological systems (Calder, 1983;

Werner & Gilliam, 1984) climate-change induced selection on body size could have important effects on community dynamics. If temperature is a direct driver of body size, we would expect

3 temperature-dependent selection on body size to be common.

In a recent comprehensive review, Teplitsky & Millien (2014) concluded that there is no direct evidence that decreases in body size in birds and mammals are an evolutionarily adaptive response to climate change. However, most studies they reviewed simply examined trends in body mass.

Once a trend is found, it can only either be consistent with warming (i.e. the animals became lighter) or not (the animals became heavier). Trends by themselves can therefore not provide a strong test of the hypothesis.

Studies that examined adaptive responses of body size to climate warming either found no evidence for selection on body size or no genetic basis for the observed size trends (Teplitsky et al.,

2008; Ozgul et al., 2009; Husby et al., 2011; Gardner et al., 2017). However, among migratory birds in North America, the observed trends in body size were apparently driven by selection on body size during the winter (Van Buskirk et al., 2010).

If climate warming causes evolutionary changes in body size, we predict that:

1. the observed trends should be due to smaller individuals replacing larger individuals, rather

than shrinking individuals, and that

2. there should be direct temperature mediated selection on an important fitness component,

i.e. large individuals have a higher fitness under cold temperatures and small individuals

have higher fitness under hot temperatures.

Using body mass as a measure of size, we tested these predictions in a population of mountain wagtails (previously long-tailed wagtail, Motacilla clara) living along a river in KwaZulu Natal,

South Africa. Between 1976 and 1999, average body mass in this population has decreased and temperature has increased. We show that the mass trend is due to lighter individuals replacing heavier ones. We further show that the effect of temperature on survival depends on body mass with lighter individuals surviving relatively better under hot conditions and heavier individuals surviving relatively better under cold conditions. Our results therefore confirm our predictions and are consistent with climate change being the driver of the observed mass changes in this population.

4 Methods

Data collection

The mountain wagtail (Motacilla clara) is a non-migratory passerine bird with a wide distribution across sub-Saharan Africa, including the east and south coast of South Africa. Mountain wagtails inhabit areas with small fast-flowing rivers in a largely arboreal environment. The birds hold life-long territories, which they rarely leave (Piper, 1990).

We studied mountain wagtails along a 7 km stretch of the Palmiet River, Westville, KwaZulu-

Natal (29◦490S30◦550E) in South Africa. From 1976 to 1999, we captured individuals with mist nets and ringed them with a numbered 3 mm steel ring and a unique combination of three or four color-rings (issued by SAFRING: South African Bird Ringing Unit, Animal Demography Unit,

University of Cape Town). Body mass was measured to the nearest 0.1 g on a spring balance. This study used data on territorial adults. The sexes are morphologically similar and we were not able to distinguish them in these analyses.

Each territory was systematically searched for surviving adults on a quarterly basis where the first quarter started in August (Q1: Aug-Oct (southern hemisphere spring), Q2: Nov-Jan

(summer), Q3: Feb-Apr (autumn), Q4: May-Jul (winter)). The onset of breeding falls in the second half of August and generally runs through to about mid December. In our study, we therefore consider the Wagtail year to extend from August until July, starting at the beginning of the southern hemisphere spring. Only territorial birds breed and they tend to use the same nest sites every year. They lay between one and four eggs, but on average 1.55 fledglings were produced per pair per annum. Individuals were either recaptured or identified by reading their color rings

(Piper, 2002).

We obtained mean quarterly temperature and total rainfall (another important environmental driver in South Africa) from a weather station in Palmiet (29◦49035.900S30◦55039.000E). We also ex- plored minimum and maximum temperatures but they showed similar trends as mean temperature and we did not pursue these further.

5 Statistical analysis

Climate

We decomposed mean quarterly temperature and rainfall time series into trend and seasonal effects using a state-space model (Durbin & Koopman, 2012) with the following observation equation:

2 yt = xt + γt + et, et ∼ N(0, σe )

and state transition equations:

2 xt = xt−1 + βt + νt, νt ∼ N(0, σν )

2 βt = βt − 1 + ωt, ωt ∼ N(0, σω)

t−1 X 2 γt = − γi + t, t ∼ N(0, σ ) i=t−4

Here yt is the observed temperature or rainfall value, xt is the underlying mean, which can change over time in this model, βt the slope or change in mean, and γt the seasonal effect. The last equation formulates an unstructured seasonal effect for four seasons. The et, νt, ωt and t are

2 2 error terms (with associated variances σx) that we assumed to be independent. We set σv = 0, to obtain smoother trend estimates. We used R package dlm (Petris, 2010; Petris & Petrone, 2011) to fit these state-space models. The advantage of a state-space modelling approach is mainly that slope and seasonal effects are allowed to change over time (Durbin & Koopman, 2012).

Birds

We estimated the linear trend in average body mass over time using linear models of mass against year, based on 574 measurements of 286 unique individuals. We also added season as a factor to account for possible seasonal variation in mass. These models were fitted using function ‘lm’ in program R 3.4.1 (R Core Team, 2017). We checked whether linear models provided a good description of the temporal trend by also fitting a generalized additive model using thin-plate

6 regression splines for the time trend. This model was fitted using function ‘gam’ of the R package

‘mgcv’ 1.8-17 (Wood, 2006).

We distinguished whether the mass trend was due to within-individual or between-individual changes using the following mixed effects model (van de Pol & Wright, 2009):

massij = β0 + βW (xij − x¯j) + βBx¯j + uj + ij

where massij is the ith mass measurement of individual j, xij is the time at which that measurement was taken, x¯j is the mean time of the measurements taken on individual j, β0 is the intercept, βW is the within-individual effect, βB is the between-individual effect, uj is an individual random effect and ij are the residuals. The random effects and residuals are assumed to follow a normal distribution with a mean of zero and variances to be estimated.

In this study, we use mass as a measure of body size. To see whether the patterns we found for mass are also reflected in a more structural measure of body size, we additionally examined trends in tarsus length using the methods described above.

We used capture-recapture data of 129 territorial adult wagtails recaptured 1295 times in total for analysing quarterly survival over the period 1976 to 1999. Because some marked individu- als may escape detection, ‘return rates’ (the proportion of released birds that is later recorded) underestimate survival probabilities (Lebreton et al., 1992). Therefore, the recapture probability

Pi, the probability that an individual is recaptured at time i given it is alive at this time, has to be taken into account when estimating the probability of surviving from occasion i to i + 1, Φ.

We used program MARK (version 6.1) (White & Burnham, 1999) to fit capture-mark-recapture models and to calculate survival and recapture probabilities on a quarterly basis (see Lebreton et al., 1992, for more details about the CMR methods). We cannot distinguish between mortality and permanent emigration from the study area and our estimates are thus of local survival rates.

However, territorial mountain wagtails are sedentary and we think that our estimates are close to true survival.

Our starting model allowed survival to vary among seasons and kept the recapture probability constant, i.e. model Φsp.. We chose this model as a starting model because our sample sizes did not allow us to estimate unconstrained time variation in either survival or recapture over the years

7 but we expected survival to vary among seasons. Since field effort was high and constant, we did not expect recapture probabilities to vary. We then examined alternative models that represent alternative hypotheses for how mass, temperature, rainfall and season affected survival, and ranked the models based on Akaike’s Information Criterion (Burnham & Anderson, 2002).

If temperature caused a change in wagtail body mass through selection on adults, we would expect light birds to survive relatively better in hot temperatures and heavy birds to survive relatively better under cooler temperature. Under this hypothesis, we thus expect interactive effects of body mass and the temperature measured during the corresponding quarter on survival of individual i during quarter t:

logit(Φit) = β0 + seasoni + β1 × massi + β2 × temperaturet + β3 × massi × temperaturet

In this model, the mass-specific effect (slope) of temperature on survival is: β2 + β3 × mass

with standard error:

p 2 var(β2) + var(β3) × mass + 2 × mass × cov(β2, β3)

We use the mass-specific effects – and their confidence intervals – of temperature on survival to examine at which body mass individuals are under temperature-dependent selection. Body mass was centred to a mean of zero prior to analysis to avoid numerical problems.

In addition to temperature, we examined the effect of total rainfall (r) during a specific quarter on the survival probability Φ during that quarter, since rainfall is predicted to change in this region

(Engelbrecht et al., 2009).

The CMR modelling approach used here assumes that there is no unmodelled heterogeneity in survival and recapture probabilities across the individuals in the sample (Lebreton et al., 1992).

We tested whether our data met these assumptions, using a goodness-of-fit (GOF) test for our starting model, Φsp.. We used the median − cˆ procedure implemented in program MARK (White

& Burnham, 1999), which suggested that our starting model fit the data adequately and showed no evidence for over-dispersion (cˆ = 1.003).

8 Results

Trends in climate

Mean quarterly temperature at our study site varied between around 18◦C in winter and 24◦C in summer (Fig. 1). The dynamic linear model suggested a slight increase in average temperature for most of the duration of the study, except for a clear dip from about 1994 to 1999. The mean slope of the average temperature level was positive (0.007◦C / year; median: 0.008◦C / per year), which is comparable to the long-term trend in global terrestrial temperature of about 0.06◦C per decade (Blunden & Arndt, 2017). Estimates for seasonal temperature effects were, relative to the mean: 2.4 ◦C for quarter 1, 2.6 ◦C for quarter 2, -3.2 ◦C for quarter 3, -1.8 ◦C for quarter 4 (SE

2 −10 2 = 0.07). Seasonality did not change over time. The variance σ was 2.5 × 10 compared to σw, the variance of the slope, which was 4.5 × 10−4.

Estimates for seasonal rain effects were, relative to the mean: 109.0 mm for quarter 1, 66.6 mm for quarter 2, -146.4 mm for quarter 3, -23.1 mm for quarter 4 (SE = 2.5 mm). According to AIC,

2 the model with σ = 0 was preferred to one where these seasonal effects were allowed to change over time. We found no evidence for long-term changes in the rainfall mean (Fig. 2).

Trend in body size

Body mass in our mountain wagtail population varied around a mean of 19.68 g, with standard deviation 1.25. A linear model showed that mass declined over time (coefficient for year: βyear =

−0.035, s.e = 0.007, t = -4.741, P<0.001, Fig. 3) but did not vary significantly among seasons

(F3,569 = 7.86, P=0.15). The generalised additive model showed that the linear model provides a good description of the mass trend (Fig. 3, dot-dash line is very close to the linear fit, i.e. the black line).

Partitioning the temporal trend in mass measurements into within-individual and between- individual components showed that the latter contributed more to the effect size than the former and only the latter was statistically significant (within-individual: βw = −0.009, s.e = 0.026, t

= -0.32, P=0.75; between-individual βb = −0.027, s.e = 0.010, t = -2.57, P=0.01, Fig. 3 dashed line).

Tarsus length also declined over time (coefficient for year: βyear = −0.20, s.e = 0.01, t =

9 -28.9, P<0.001). Both within-individual and between-individual changes contributed to this trend

(within-individual: βw = −0.13, s.e = 0.04, t = -3.4, P<0.001; between-individual βb = −0.20, s.e

= 0.01, t = -25.9, P<0.01).

Survival analysis

Model selection based on AICc showed that the model that best described the pattern in our data was the model that included the effects of season, mass, temperature and the interaction term between mass and temperature (Model 1, Table 1). The pattern recovered by this model is consistent with the hypothesis of temperature-mediated selection on body mass: heavy individuals survived relatively better in cold temperatures, and lighter individuals survived relatively better in warm temperatures (Fig. 4).

Since mass is a continuous trait, some individuals of intermediate mass experienced little temperature-dependent selection. For individuals in the mass range 19.3 g to 21.1 g, the 95% confidence interval for the estimated temperature-survival relationship included zero. Individuals below 19.3 g therefore experienced positive survival selection of increasing temperature and indi- viduals above 21.1 g experienced negative survival selection of increasing temperature. During the

first 10 years of our study, 35% of the individuals experienced positive selection and 12% experi- enced negative selection. During the final 10 years of the study, the proportion of the individuals experiencing positive selection rose to 47% and the proportion experiencing negative selection declined to 6%.

Even though temperature varied with seasons, model selection strongly favoured a model with both temperature and season effects. Seasonal variation in survival was thus not fully explained by variation in temperature. All models including an effect of rainfall on survival were poorly supported by the data, suggesting that rainfall had little effect on survival (Table 1).

On average, the quarterly survival estimates ranged from 0.96 (95% confidence interval 0.93 to 0.97) in the austral autumn to 0.87 (0.84 to 0.91) in the austral winter (Fig. 5). Based on these quarterly survival estimates, the annual survival rate was 0.72. The estimated recapture probability was 0.987 (0.979 to 0.992).

10 Discussion

We examined whether an observed change in body mass in a population of mountain wagtails may have been caused by climate change. The temperature trend in our study area is consistent with the trend in global terrestrial temperatures. If the change in body mass was an adaptive response to increasing temperature, we predicted that the mass trend was caused by lighter individuals replacing heavier ones and that the effect of temperature on survival should depend on mass.

We found evidence for both predictions and in particular found that lighter individuals survive relatively better under warm temperatures whereas heavy individuals survived relatively better under cold temperatures. Our results are thus consistent with the hypothesis that shrinking body size is a response to climate change (Gardner et al., 2011). One necessary condition for this hypothesis to be true is that selection on body size is driven by temperature, which we confirmed in our study. Whether this selection pressure leads to adaptive evolutionary change depends on the heritability of body size (Teplitsky & Millien, 2014), which we did not measure in our population.

We found that the trends in body mass were mirrored by a similar trends in tarsus length.

We did not have a clear a-priori expectation for tarsus. Some bird species can use their feet to loose excess heat (Martineau & Larochelle, 1988) but we are not sure whether this is the case for mountain wagtails. If it were, we would expect tarsus length to increase in a warming climate in accordance with Allen’s rule (Rensch, 1938). Alternatively, tarsus may just be a reflection of body size and decline along with mass. Our results are consistent with the latter expectation. It would be interesting to examine trends in bill size, which is increasingly being recognized as an important thermoregulatory organ in birds (Tattersall et al., 2017). Unfortunately, we did not have adequate data to conduct such an analysis.

Even though shrinking body size has been suggested as the third universal response to climate change, in addition to range shifts and phenological changes (Gardner et al., 2011), to date the ev- idence that the observed body size changes are indeed caused by climate change is mixed (Salewski et al., 2014). The few studies that did examine the adaptive nature of these trends either found no evidence for selection for smaller size or no genetic basis for the size change (see Teplitsky &

Millien, 2014, for a recent review). Given the importance of body size as a key ecological trait

(Woodward et al., 2005), more direct evidence for or against temperature-dependent selection on

11 body size is urgently needed.

While Bergmann’s rule explains body size clines in many taxa, the mechanisms that underpin these clines are much more unclear (Blackburn et al., 1999; Watt et al., 2010). Bergmann (1847) originally suggested that cold temperature selected against smaller individuals because of their greater surface to volume ratio. The more recently proposed heat dissipation limit theory in con- trast predicts that smaller individuals should be favored under warmer conditions because they can sustain higher metabolic activity without overheating (Speakman & Król, 2010). Our results are consistent with both mechanisms. The heaviest individuals survived relatively better than smaller individuals under cold temperatures but survived relatively less well under warm temperatures.

In contrast to the heat dissipation theory and our results, theories based on evaporative water loss predict that heavier individuals should be less affected by heat stress than lighter individuals

(McKechnie & Wolf, 2010) but evidence for this theory remains inconclusive (Whitfield et al., 2015;

McKechnie et al., 2016). In any case, mountain wagtails live near permanent water and should therefore not be limited by evaporative water loss.

Both Bergmann’s original hypothesis (Bergmann, 1847) and the heat dissipation theory (Speak- man & Król, 2010) build on thermodynamic considerations. Since metabolic rate scales with mass m as m0.75 (Kleiber, 1932), and mean body mass in our study population decreased by

0.035×10 19.7 = 1.8% per decade, the average observed decrease in body mass would have led to

0.9820.75 = 0.987 or a 1.3% decline in absolute metabolic rate per decade (McNab, 2010), which seems relatively modest.

Bergmann’s rule, and clines in body size in general, could also be due to variation in resources

(McNab, 2010). It is possible that resources vary seasonally in our study area, as a result of temperature seasonality. Cooler temperatures coincided with winter and relatively lower survival of small wagtails. Periods of food shortage could select against smaller individuals as they have shorter starvation times (Lindstedt & Boyce, 1985; Goodman et al., 2012).

Survival in our wagtail population varied among the seasons and this variation was only partly explained by variation in temperature. Few studies examined seasonal variation in survival of resident birds. Duckworth & Altwegg (2014) found no evidence for seasonal variation in survival of hadeda (Bostrychia hagedash). Among resident Dutch oystercatchers (Haematopus ostralegus),

12 winter survival was lower than during other seasons only in severe winters (Duriez et al., 2012).

Conclusion

Shrinking body size has been suggested as a universal response to climate change (Gardner et al.,

2011) but the evidence that climate change is indeed the selective force behind observed size changes is still scarce (Teplitsky & Millien, 2014). We found clear evidence for temperature-dependent selection on body mass of mountain wagtails and that the observed change in mass is due to turnover of individuals. Both lines of evidence are consistent with climate change being a driver of this observed body mass change. It remains to be seen, however, whether the observed temperature- dependent survival selection was due to direct effects of temperature through thermodynamic mechanisms or indirect effects via resource availability.

Acknowledgements

This paper is dedicated to the memory of Steven E. Piper who initiated this project and collected the data. This paper is largely based on a presentation Steven gave at the Pan-African Ornitho- logical Conference in 2008. We thank the National Research Foundation of South Africa (Grants

85802 and 114696) and the Alliance for Collaboration on Climate and Earth System Sciences for funding. The NRF accepts no liability for opinions, findings and conclusions or recommendations expressed in this publication.

Supplementary Information

The supplementary information contains the R code used to fit state-space models to the temper- ature and rainfall data, and the analysis of trends in the mass of mountain wagtails.

All applicable institutional and national guidelines for the care and use of animals were followed.

References

Ashton KG (2002). Patterns of within-species body size variation of birds: strong evidence for

Bergmann’ s rule. Global Ecology and Biogeography 11:505–523.

13 Ashton KG Tracy MC Queiroz AD (2000). Is Bergmann’ s rule valid for mammals? American

Naturalist 156:390–415.

Bergmann C (1847). Ueber die Verhältnisse der Wärmeökonomie der Thiere zu ihrer Grösse.

Göttinger Studien 3:595–708.

Blackburn TM Gaston KJ Loder N (1999). Geographic gradients in body size: a clarification of

Bergmann’s rule. Diversity and Distributions 5:165–174.

Blunden J Arndt DS (2017). State of the climate in 2016. Bulletin of the American Meteorological

Society 98:Si–S277.

Both C te Marvelde L (2007). Climate change and timing of avian breeding and migration through-

out Europe. Climate Research 35:93–105.

Burnham KP Anderson DR (2002). Model selection and multimodel inference: a practical

information-theoretic approach. 2nd edition, Springer, New York.

Calder WA (1983). Ecological scaling: mammals and birds. Annual Review of Ecology and

Systematics 14:213–230.

Duckworth GD Altwegg R (2014). Environmental drivers of an urban Hadeda Ibis population.

Ardea 102:21–29.

Durbin J Koopman SJ (2012). Time Series Analysis by State Space Methods. 2nd edition, Oxford

University Press, Oxford.

Duriez O Ens BJ Choquet R Pradel R Klaassen M (2012). Comparing the seasonal survival of res-

ident and migratory oystercatchers: carry-over effects of habitat quality and weather conditions.

Oikos 121:862–873.

Engelbrecht FA Mcgregor JL Engelbrecht CJ (2009). Dynamics of the conformal-cubic atmospheric

model projected climate-change signal over southern Africa. International Journal of Climatology

1033:1013–1033.

Gardner JL Peters A Kearney MR Joseph L Heinsohn R (2011). Declining body size: A third

universal response to warming? Trends in Ecology and Evolution 26:285–291.

14 Gardner JL Rowley E de Rebeira P de Rebeira A Brouwer L (2017). Effects of extreme weather on

two sympatric australian passerine bird species. Philosophical Transactions of the Royal Society

B: Biological Sciences 372.

Goodman RE Lebuhn G Seavy NE Gardali T Bluso-Demers JD (2012). Avian body size changes

and climate change: warming or increasing variability? Global Change Biology 18:63–73.

Husby A Hille SM Visser ME (2011). Testing mechanisms of Bergmann’s rule: phenotypic decline

but no genetic change in body size in three passerine bird populations. The American Naturalist

178:202–13.

Kleiber M (1932). Body size and metabolism. Hilgardia: A Journal of Agricultural Science 6:315–

353.

Lebreton JD Burnham KP Clobert J Anderson DR (1992). Modeling survival and testing biological

hypotheses using marked animals: a unified approach with case studies. Ecological Monographs

62:67–118.

Lindstedt SL Boyce MS (1985). Seasonality, fasting endurance, and body size in mammals. Amer-

ican Naturalist 125:873–878.

Martineau L Larochelle J (1988). The cooling power of pigeon legs. Journal of experimental biology

136:193–208.

McKechnie AE Whitfield MC Smit B Gerson AR Smith EK Talbot WA McWhorter TJ Wolf BO

(2016). Avian thermoregulation in the heat: efficient evaporative cooling allows for extreme

heat tolerance in four southern hemisphere columbids. The Journal of Experimental Biology

219:2145–2155.

McKechnie AE Wolf BO (2010). Climate change increases the likelihood of catastrophic avian

mortality events during extreme heat waves. Biology letters 6:253–6.

McNab BK (1971). On the ecological significance of Bergmann’ s rule. Ecology 52:845–854.

McNab BK (2010). Geographic and temporal correlations of mammalian size reconsidered: A

resource rule. Oecologia 164:13–23.

15 Ozgul A Tuljapurkar S Benton TG Pemberton JM Clutton-Brock TH Coulson T (2009). The

dynamics of phenotypic change and the shrinking sheep of St. Kilda. Science 325:464–467.

Parmesan C (2006). Ecological and evolutionary responses to recent climate change. Annual

Review Of Ecology Evolution And Systematics 37:637–669.

Parmesan C Yohe G (2003). A globally coherent fingerprint of climate change impacts across

natural systems. Nature 421:37–42.

Petris G (2010). An R package for dynamic linear models. Journal of Statistical Software 36:1–16.

Petris G Petrone S (2011). State space models in R. Journal of Statistical Software 41.

Piper SE (1990). Longtailed Wagtail Motacilla clara. In The complete book of southern African

birds pp. 376–377.

Piper SE (2002). Survival of adult, territorial Longtailed Wagtails Motacilla clara: the effects of

environmental factors and individual covariates. Journal of Applied Statistics 29:107–124.

R Core Team (2017). R: A language and environment for statistical computing.

Rensch B (1938). Some problems of geographical variation and species-formation. Proceedings of

the Linnean Society of London 150:275–285.

Salewski V Siebenrock KH Hochachka WM Woog F Fiedler W (2014). Morphological change to

birds over 120 Years is not explained by thermal adaptation to climate change. PLoS ONE

9:1–14.

Speakman JR Król E (2010). Maximal heat dissipation capacity and hyperthermia risk: neglected

key factors in the ecology of endotherms. Journal of Animal Ecology 79:726–46.

Tattersall GJ Arnaout B Symonds MR (2017). The evolution of the avian bill as a thermoregulatory

organ. Biological Reviews 92:1630–1656.

Teplitsky C Millien V (2014). Climate warming and Bergmann’s rule through time: is there any

evidence? Evolutionary Applications 7:156–168.

16 Teplitsky C Mills JA Alho JS Yarrall JW Merila J Merilä J (2008). Bergmann ’ s rule and climate

change revisited: Disentangling environmental and genetic responses in a wild bird population.

Proceedings of the National Academy of Sciences 105:13492–13496.

Thomas CD Lennon JJ (1999). Birds extend their ranges northwards. Nature 399:213.

Van Buskirk J Mulvihill RS Leberman RC (2010). Declining body sizes in North American birds

associated with climate change. Oikos 119:1047–1055. van de Pol M Wright J (2009). A simple method for distinguishing within- versus between-subject

effects using mixed models. Animal Behaviour 77:753–758. van Gils JA Lisovski S Lok T Meissner W Ożarowska A de Fouw J Rakhimberdiev E Soloviev MY

Piersma T Klaassen M (2016). Body shrinkage due to arctic warming reduces red knot fitness

in tropical wintering range. Science 352:819–821.

Visser ME Both C Lambrechts MM (2004). Global climate change leads to mistimed avian repro-

duction. In Birds and Climate Change: Advances in Ecological Research, Vol. 35 vol. 35, pp.

89–110.

Walther GR Post E Convey P Menzel A Parmesan C Beebee TJC Fromentin JM Hoegh-Guldberg

O Bairlein F (2002). Ecological responses to recent climate change. Nature 416:389–395.

Watt C Mitchell S Salewski V (2010). Bergmann’s rule; a concept cluster? Oikos 119:89–100.

Werner EE Gilliam JF (1984). The ontogenetic niche and species interactions in size-structured

populations. Annual Review of Ecology and Systematics 15:393–425.

White GC Burnham KP (1999). Program MARK: Survival estimation from populations of marked

animals. Bird Study 46:S120–139.

Whitfield MC Smit B McKechnie AE Wolf BO (2015). Avian thermoregulation in the heat: scaling

of heat tolerance and evaporative cooling capacity in three southern African arid-zone passerines.

Journal of Experimental Biology 218:1705–1714.

Wood SN (2006). Generalized additive models: an introduction with R. Chapman and Hall.

17 Woodward G Ebenman B Emmerson M Montoya JM Olesen JM Valido A Warren PH (2005).

Body size in ecological networks. Trends in Ecology and Evolution 20:402–409.

Yom-Tov Y (2001). Global warming and body mass decline in Israeli passerine birds. Proceedings

of the Royal Society of London B 268:947–52.

18 Model ∆AICc w K Deviance

1 Φs,m,T,T ×mP. 0.00 0.9868 8 893.624

2 Φs,m,T P. 9.78 0.0074 7 905.424

3 Φs,T P. 10.82 0.0044 6 908.488

4 Φs,mP. 15.42 0.0004 6 913.087

5 Φs,m,r,r×mP. 16.36 0.0003 8 909.987

6 Φs,m,m2 P. 16.68 0.0002 7 912.324

7 Φs,rP. 16.92 0.0002 6 914.587

8 ΦsP. 17.19 0.0002 5 916.872

9 Φm,T,m×T P. 22.04 0.0000 5 921.723

Table 1: Model selection of survival (Φ) and capture probability (P ) of mountain wagtails on

Palmiet river, South Africa, in relation to season (s), mass (m), temperature (T ), rainfall (r) and their interactions. Model selection was based on the sample-size adjusted Akaike’s Information

Criterion (AICc). Akaike weights (w) show the relative support of each model compared to the other models in the set. K is the number of estimated parameters.

19 Figure captions:

Fig 1 Observed quarterly mean temperature at Palmiet weather station in KwaZulu-Natal, South

Africa (open dots). We fitted a dynamic linear model to these data. The model included effects for seasonality. The smoothed trend for mean quarterly temperature is shown in red in the upper panel, with 95% confidence level (dashed lines). The gray horizontal line shows the average temperature across the time series. The lower panel shows the slope of the trend. The dotted lines are 95% confidence intervals.

Fig 2 Observed quarterly rainfall at Palmiet weather station in KwaZulu-Natal, South Africa

(dots) with smoothed mean level (red line) and 95% confidence levels (dashed lines), fitted using a dynamic linear model. Lower panel shows slope of trend with 95% confidence levels.

Fig 3 Change in mass of mountain wagtails (Motacilla clara) at Palmiet River, South Africa between 1976 and 1999. The black line is the best-fitting linear regression model. The red dot- dash line (largely coinciding with the black line) is a generalised additive model using thin-plate regression splines. The blue dashed line is the trend in mean mass due to individual turnover (as opposed to changes within individuals). This figure is in colour only in the online version.

Fig 4 The effect of temperature on quarterly survival of mountain wagtails. Even though mass was treated as a continuous variable, we are plotting the estimated relationship for a bird of mass 21.5 g (dotted line, representing the 95th percentile of mean body masses recorded in the population) and a bird of mass 17.5 g (black line, representing the 5th percentile of mean body masses recorded in the population). The gray lines show 95% confidence intervals, calcu- lated using the delta method. The estimated relationship between temperature and survival is logit(survival) = 1.61 + 2.73 × mass + 0.04 × temperature − 0.12 × mass × temperature (from

Model 1, Table 1).

Fig 5 Mean quarterly survival probabilities of mountain wagtails at Palmiet River in South

Africa. The vertical bars represent 95% confidence intervals (estimates from Model 8, Table 1).

20 ● ● ● ● ● ● ● ● ● ● ● ● 24 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● 22

● 20 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● Temperature [°C] Temperature ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 18 ● ● ● ● ● ● ● ● ●

0.20

0.15 Time 0.10 0.05

Slope 0.00 −0.05 −0.10 −0.15

1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000

Figure 1:

21 700 ● ●

600 ● ● ● 500 ● ● ● ● ● ●● ● ● 400 ●● ● ● ● ● ● ● ● ● ● ● ● ● ● 300 ● ● ● ● ● ● ● ● ● Rain [mm] ● ● ● ● ● ● ● ● ● ● ● ● ● ● 200 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 100 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0 ● ●

20

Time 10

0 Slope

−10

−20

1976 1978 1980 1982 1984 1986 1988 1990 1992 1994 1996 1998 2000

Figure 2:

22 ● ●● ● ●●● ● ● ● ● ● ● ●● ● ● ● ● ●● 22 ● ● ● ● ●●●●●●●●●● ●● ● ●●● ● ● ● ● ● ● ● ● ●●● ●● ● ●● ●● ● ●●●● ● ● ●●● ●●●●● ● ● ●●● ●●●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●●●●●●●●● ●● ●● ● ● ● ●●●●●●● ●●●●●●● ● ● ●● ● ●●●●●●●● ●● ● ● ●●●●●●●●●●● ● ●● ● ● ● ● ● ●●●●●● ●● ● 20 ● ●● ● ● ●● ●●●● ●●● ● ● ● ●● ● ● ●●●● ●●●●●● ●●● ● ● ●●●●●●●●●●● ●● ●● ● ● ● ● ●●●●●●●●●● ●● ● ●●● ● ● ●●● ●●● ●●●●●●● ● ● ●●●● ● ● ●● ● ●● ●● ●●●●●●●●●●● ●● ● ● ● ●● ●●● ●● ● ● ● ● ●● ●●●●●● ●●● ● ●●●●●●● ●●● ● ● ●●●●●● ● ● ● ● ● ●●●●● ●●● ● ● ●● ●● ● ●● ● ● ●● ●● ●●● ● 18 ● ● ● Mass [g] ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● 16 ● ● ● ●

● 14

1980 1985 1990 1995 2000

Year

Figure 3:

23 1.0

0.9

0.8

0.7

Heavy individual (21.5g) 0.6

Quarterly survival probability Light individual (17.5g)

0.5

16 18 20 22 24

Temperature [°C]

Figure 4:

24 1.0 ● ● ● 0.9 ●

0.8

0.7 Quarterly survival 0.6

Aug − Oct Nov − Jan Feb − Apr May − Jul Season

Figure 5:

25