Climate anomalies affect annual survival rates of swifts wintering in sub‐Saharan Africa

Abstract Several species of migratory swifts breed in the Western Palearctic, but they differ in reproductive traits and nonbreeding areas explored in Africa. We examined survival and recapture probabilities of two species of swifts by capture–mark–recapture data collected in northern Italy (Pallid Swift Apus pallidus in Carmagnola, Turin, and Common Swift Apus apus in Guiglia, Modena) in the breeding season (May–July). Apparent survival rates were relatively high (>71%), comparable to other studies of European swifts, but showed marked annual variations. We used geolocators to establish the exact wintering areas of birds breeding in our study colonies. Common Swifts explored the Sahel zone during migration and spent the winter in SE Africa, while the Pallid Swifts remained in the Sahel zone for a longer time, shifting locations southeast down to Cameroun and Nigeria later in winter. These movements followed the seasonal rains from north to south (October to December). In both species, we found large yearly differences in survival probabilities related to different climatic indices. In the Pallid Swift, wintering in Western Africa, the Sahel rainfall index best explained survival, with driest seasons associated with reduced survival. In the Common Swift, wintering in SE Africa, the El Niño–Southern Oscillation (ENSO) cycle performed significantly better than Sahel rainfall or North Atlantic Oscillation (NAO). Extreme events and precipitation anomalies in Eastern Africa during La Niña events resulted in reduced survival probabilities in Common Swifts. Our study shows that the two species of swifts have similar average annual survival, but their survival varies between years and is strongly affected by different climatic drivers associated with their respective wintering areas. This finding could suggest important ecological diversification that should be taken into account when comparing survival and area use of similar species that migrate between temperate breeding areas and tropical wintering areas.


| INTRODUC TI ON
In the Western Palearctic, several species of swifts breed in the same geographic area, where different species may form mixed colonies in natural breeding sites (Mazzotto, Cucco, & Malacarne, 1996).
They may also breed in buildings constructed by humans (Cucco & Malacarne, 1987). However, swift species that mix in the breeding area differ noticeably in their biological traits and circannual timing and may spend the winter in different parts of the African continent.
Traits that differ between two swifts breeding in the same area, the Common Apus apus and the Pallid Swift Apus pallidus (Pellegrino et al., 2017), include number of clutches produced per year (Cucco, Malacarne, Orecchia, & Boano, 1992), diet composition (Cucco, Bryant, & Malacarne, 1993), and moult strategy (Boano, Pellegrino, & Cucco, 2015). At the same time, they show similar movement adaptations during the nonbreeding period including continuous flight, but extending for different lengths of time, with Common Swifts staying airborne around 10 months (<1% landing ;Hedenström et al., 2016), and Pallid Swifts 5 months (<1% landing ;Hedenström et al., 2019). The large difference in breeding period between the two sympatric swift species could potentially originate from differences in preferred winter destinations including length of wintering periods. However, due to their strictly aerial habits , observations during winter are hard to collect, and until recently, little was known about both location of wintering areas explored (Åkesson, Klaassen, Holmgren, Fox, & Hedenström, 2012;Norevik et al., 2019) and the annual survival of these continuously flying insectivorous species.
Most survival estimates for swifts have been collected in a single or a limited number of locations. In particular, the survival of Common Swift, corresponding to one of the most widespread and abundant swift species in the Western Palearctic, has until now only been studied in Great Britain and France. Moreover, no study so far has investigated the influence of climate drivers in wintering or passage grounds on the survival of swifts. One of the few studies focusing on the effect of weather on survival solely considered the effect of local climate experienced on the breeding ground, leading to reduced adult survival associated with low temperatures in July (Thomson, Douglas-Hhome, Furness, & Monaghan, 1996). Thus, it is important to investigate what effects climate drivers may have on the swifts' survival in other periods of the year. In particular, this is central for conservation of the species, in order to understand under which period of the annual cycle the swifts face the largest mortality risk (Marra, Cohen, Loss, Rutter, & Tonra, 2015).
Our aim was to compare the climatic factors affecting survival of two species of swifts, the Common and the Pallid Swift, sampled among populations breeding in the same area (northern Italy), but wintering in different parts of the African continent. In the present study, we provide one of the few estimates of survival rates for adult Common Swift in Europe outside England and France, and the first study considering the effect of climate outside the breeding season.
To record the wintering areas of our study populations, and to assess the nonbreeding movements of breeding adults, we used miniature data-logger geolocators to track adult breeding Common Swifts and used information from a recent study reporting nonbreeding ranges for our Pallid Swift population .

| Pallid Swift
The study site is localized in two old buildings in the center of the town of Carmagnola (Turin, Italy) (Lat. 44.84°, Long.: 7.72° E, 239 m asl; Boano, 1974). During the study period, the colony has grown from about 30 to > 100 breeding pairs. Our main study activity was done in a subset of the colony with nests accessible from inside the private house. Here, we could inspect about 25 nest holes occupied annually by 7-20 pairs. These nests were inspected around every fortnight (from 1982 to 1986) or daily (from 1987 to 1990) for concurrent ecological and ethological studies (Boano & Cucco, 1989;Cucco & Malacarne, 1996a, 1996b, 1996c. From 1991 and thereafter, we reduced the effort to 2-3 visits per season (mainly in July) to ring all young and most adults. For survival estimation of this population, we used the data of ringed adults from 1984 to 1992, when the capture effort was higher (Boano et al., 1993), and those of a second period from 2002 to 2012, with a lower capture effort but in the same years of the Common Swift sample, so particularly suited for comparisons.

| Common Swift
The study was performed in the Regional Park of Sassi di Roccamalatina, Guiglia, Modena, in the medieval tower of Castellaro (44. 39°N, 10.95°E, 490 m asl). At this breeding colony, nests were checked at least twice per year for over 22 years (from 1991 to 2012), the first time in late May, to capture breeding adults and to record number of the eggs for each pair, and again in late June to ring the chicks and adults feeding young. In the first 10 years, the nests were checked more irregularly with some years lacking, but from 2001 the monitoring was regular up to the final year 2012. Thus, the data used here are from 2001 to 2012, including only the period with regular checking. During this period, the colony increased from 17 to 51 breeding pairs (averaging 35 per year), thanks to some management of the vegetation nearby the tower and of the nest facilities (Minelli et al., 2014).
In both study sites, all birds captured were ringed with metal rings and controlled when captured again. Only adult birds were used for the analyses, because the very low philopatry of young swifts (Lack, 1956;personal observations) prevents the possibility to recapture them after fledging.

| Tracking nonbreeding movements by geolocation
We utilized geolocation data collected in the same swift colonies where we assessed survival. We used archival light loggers (Model Intigeo W55B1 and W65B1) from Migrate Technology Ltd. (all without a stalk) to track the nonbreeding movements of adult breeding swifts. The adult breeding Common and Pallid Swifts were captured inside the nest boxes located in the building after sunset at night. We timed the catching and attachment of geolocators to the late stage of the breeding period, when the young were near to leaving the nest, but the parents were still actively feeding the young (second half of the breeding period; Åkesson, Bianco, & Hedenström, 2016;Åkesson et al., 2012). We used a full body harness (soft braided nylon string) to attach the geolocators to the bird by three loops around neck and each wing, respectively (total mass 0.7-1.3 g including harness, never exceeding 3% of the bird's body mass), as described by Åkesson et al. (2012). After capture and immediate logger attachment, the birds were released at the colony. We did not include individuals fitted with geolocators in the analysis of survival due to their documented lower return rate for logged birds to the breeding colony (Morganti et al., 2017). The wintering area utilized by the Pallid Swifts breeding at our study site was taken from the results reported in a recently published study based on geolocation . For further information on capture method of Pallid Swifts, see Norevik et al. (2019).
In 2010-2012, we attached 28 geolocators to adult Common Swifts in the breeding colony in Italy, and we were able to recover 4 logged individuals. At recapture, we did not find any negative effect on plumage or skin caused by the attachment of the geolocators on the retrapped swifts.
We used the program Intiproc v.1.03 provided by the manufacturer Migrate Technology Ltd, to perform the initial linear correction function for the clock drift and extracting times for sunrise and sunset using 2 as the light threshold. We used the critical sun angle corresponding to a light-level value of 2 on the arbitrary geolocator light scale (Migrate Technology Ltd.) minimizing the difference in latitude between pre-and postequinox, and at the same time minimizing the uncertainty in latitude close to the equinox for periods when the birds were stationary as defined by the estimations of longitude. We used 0.5 and 0.3 steps of critical sun angle extracted and evaluated across a range of sun angles (8-12 per bird) to define the one resulting in lowest difference in latitude between pre-and postequinox periods. We used the "Hill-Ekstrom" procedure (Ekstrom, 2004) to evaluate which sun angle we should use for respective track as outlined in Åkesson et al. (2012), Åkesson et al. (2016). The sun angles used varied between −3.7° and −6.7° depending on model. We recorded minimal clock drift (0-2 min) over one year, and no consistency in the drift for our loggers. We excluded light data on latitude and longitude corresponding to ca 14 days before and after autumn and vernal equinoxes, respectively, and 21 days after and before autumn and vernal equinoxes, respectively, from the evaluations. The errors recorded by archival lightlevel geolocators are influenced by geographic location, time of year, habitat type, and weather and correspond to values of 143 ± 62 km (mean ± 95% confidence interval) in terrestrial systems (Fudickar, Wikelski, & Partecke, 2011), and 186 ± 114 km (mean ± SD) in marine environments (Phillips, Silk, Croxall, Afanasyev, & Briggs, 2004) for latitude. The corresponding values for errors of longitude estimates are lower 50 ± 34 km (mean ± 95% confidence interval) for terrestrial environments (Fudickar et al., 2011) and 85 ± 47 km (mean ± SD) at sea (Phillips et al., 2004). Reports show that weather, topography, and vegetation have the strongest impact on accuracy in geolocator tracking data for terrestrial birds, leading to shading and variations in light intensity (Lisovski et al., 2012). Conversely, for swifts and other aerial birds who spend a large fraction of their time on the wing mainly weather influence the geolocator precision, resulting in typically very clean light measurements and no shading effects (Åkesson et al., 2016).

| Mark-recapture analysis
Survival estimates derived from analysis of recaptures or resightings of living marked birds are widely used in association with proper stochastic open-population models (Cormack, 1964;Seber, 1982;Williams, Nichols, & Conroy, 2002). Estimates derived from capture-recapture experiments should be considered as minimal (or "apparent") survival rates, although the bias is frequently negligible for adult philopatric bird species. In this study, capturerecapture data were used in association with open-population models (Cormack-Jolly-Seber, CJS hereafter, and related models), and associated model selection criteria (e.g., Lebreton, Burnham, Clobert, & Anderson, 1992;Nichols, 1994;Williams et al., 2002).
These models produce survival estimates that are not influenced by variations in recapture probability, and therefore are more reliable than those based on return rates alone (Martin, Clobert, & Anderson, 1995;Nichols & Pollock, 1983). Survival probability (Фi) was defined as the probability that an animal, living at period i, is still alive and available for recapture at period i + 1, and recapture probability (pi) is the capture probability of an animal alive and in the population at the sampling time i.
According to previous considerations, survival probability complement (1-Ф) includes both mortality and permanent emigration from the study site and is here referred as "apparent" survival according to Thomson et al., (2009) that we follow for all the relevant terminology.
Analysis started with program U-CARE (Choquet, Lebreton, Gimenez, Reboulet, & Pradel, 2009) to compute the goodness-offit test (GOF) of the most general model, where Фi and pi vary only with time, and other specific tests for transience and trap dependence. GOF procedures are used to identify a general statistical model "fitting" the data. Then, models making further restrictions were fitted with program MARK version 9.0 (Cooch & White, 2002;White & Burnham, 1999), including models where temporal variation in survival is modeled as a logit-link function of specific weather conditions. The selection of the most appropriate model was based on the Akaike's information criterion (AIC) (Anderson & Burnham, 1999;Burnham, Anderson, & White, 1995). In this study, we adopted AICc values (AIC approximated for small samples) or QAICc (quasi-AICc) when the data were overdispersed as evidenced by ĉ (the variance inflation factor) >1 (White & Burnham, 1999). If overdispersion is present, then model selection should be based on QAICc (QAICc = AICc/ĉ). We also applied analysis of deviance (ANODEV) to assess the statistical significance and the fraction of temporal variation explained by each covariate used in the model (Grosbois et al., 2008;Lebreton et al., 1992;Rolland, Barbraud, & Weimerskirch, 2008;Skalski, Hoffman, & Smith, 1993).
To investigate the relationships between survival rates and climate conditions potentially influencing swift survival, three main climatic variables were considered, that is, (1) Sahel rainfall, (2)  This index was found to potentially correlate with bird survival as, in a number of European trans-Saharan migrants, precise relationships have emerged between annual breeding numbers or survival and annual rainfall in African wintering areas (Newton, 2008;Zwatrts, Bijlsma, Van der Kamp, & Wymenga, 2009). Moreover, to investigate the effect of extreme low rainfall values, the years were categorized to reflect whether or not a given rainy season was "very dry" following the numerical limits explicitly defined by Landsea, Gray, Mielke, and Berry (1997), coding the dryer years as "1" and all other years as "0" in the categorical model.

NAO: The North Atlantic Oscillation Index is a large-scale
oscillation in atmospheric masses between the subtropical high and the polar low and is an index useful as a measure of the general climatic conditions in large parts of Europe (Hurrell, Kushnir, Ottersen, & Visbeck, 2003). Furthermore, Mediterranean precipitation is correlated with NAO, with negative anomalies for the positive phase of the oscillation and northward shifts of the storm track during the positive phases of the oscillation (Baldi, Cesarone, Carella, Crisci, & Dalu, 2004;Delitala, Cesari, Chessa, & Neil, 2000). More interestingly for our purposes, the NAO is now considered to show an association with rainfall variability and some influence on the productivity of diverse African regions, including eastern (Stige et al., 2006) and southeastern Africa (McHugh & Rogers, 2001). Values of NAO for the month from December to March are taken from https://clima tedat aguide.ucar.edu/sites /defau lt/files /nao_stati on_djfm.txt.

ENSO: As index of the El Nino-Southern Oscillation (ENSO)
cycle, we used the Oceanic Niño Index (ONI) for the months December, January, and February. ONI is based on sea surface temperature anomalies and is defined as the three-month (in our case Dec-Jan-Feb) running-mean SST (sea surface temperature) departures in the Niño 3.4 region, based on the NOAA ERSST data. El Niño is characterized by ONI ≥ +0.5°C, while La Niña is based on ONI ≤ −0.5°C. An El Niño or La Niña episode is defined when the above thresholds are exceeded for a period of at least 5 consecutive overlapping 3-month seasons. We used ONI data for the trimester Dec-Jan-Feb downloaded from http:// www.cpc.noaa.gov/produ cts/analy sis_monit oring /ensos tuff/ ensoy ears.shtml. Because of strong effect in an area visited by Common Swift in winter (see Figure 1), we also tested a categorical model obtained simply by coding the La Niña years as "1" and all other years as "0." The complex effects of these climatic drivers on different African regional climates are summarized in Figure 1.
According to the best model, survival probability ranged from 0.62 in the "very dry" season to 0.84 in the "other" seasons. Figure 2a In the second study period (2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012), the model evaluation procedure of the program MARK (results in Table 2) was found as better model the one with constant survival (Ф = 0.715; SE = 0.075) and variable capture probabilities in the time (Table S3), but the second classified model (ΔQAICc < 2) was still the one considering the "very dry" versus "other" seasons in Sahel, with survival estimates of 0.59 in very dry against 0.87 in"other" years (Table S4). These values were very similar with respect to the ones obtained in 1984-1992. In this case, however, ANODEV indicated that the variable did not add any significant explanation to the interannual variation in adult survival (F 1,3 = 0.416, p = .585).  Table   S1). The GOF test of the general CJS model Ф(t)p(t) (χ 2 = 43.444; df = 33; p = .106) suggested that the CJS model's basic assumptions were met; ĉ estimated by χ 2 /df was 1.316 and this value was used to correct for overdispersion (Cooch & White, 2006 (Table S5) is highlighted in Figure 2a

| Nonbreeding movements tracked by geolocation
We were able to track the nonbreeding movements of four adult breeding Common Swifts from our study colony in Modena (

| Survival probabilities for adult swifts
Swift are long-lived birds, and indeed, the survival of Apodiformes is very high when compared to passerine aerial feeders such as swallows (Ferro & Boano, 1998;Masoero, Tamietti, Boano, & Caprio, 2016). In this study, we found a mean annual survival similar to those reported in the previous studies available, both for the Common (0.78) and the Pallid Swift (ranging from 0.71 to 0.76).
A correct comparison of the survival values reported in different studies should, however, consider the methodology employed for data collection and analysis. In some studies, the method involves the use of ringing recoveries, that is, the observations of all birds that were found dead or alive in a vast area (UK scheme : Baillie & Green, 1987;Dobson, 1990). In other studies, the method involved the use of recaptures, that is, observations based on capture-mark-recapture of live individuals in the same locality. The values of survival obtained by only recapture data should be considered as apparent (or "local") survival. The method can slightly underestimate the global survival, because some individuals that were alive can instead be considered as dead if they move (emigrate) to a different and not examined site. In the particular case of the Pallid and Common Swift, the difference between the recovery and recapture methods is likely very small. The adult swifts show a high philopatry and fidelity to their breeding colony, and even to the single nest (Boano et al., 1993;Lack, 1956;Weitnauer, 1980). Hence, the difference in estimates related to individuals moving to other sites should be minimal.
The higher mean value reported by Baillie and Green (1987) for Common Swift is probably due to the fact that this study is one of the few taking into account recoveries over the entire species range and not only from live recaptures at a single colony. In effect, a similar recovery analysis by Dobson (1990) obtained a lower survival value (0.76); the same values obtained by Lebreton et al. (1992) and Thomson et al. (1996) with live recaptures and the survival estimates for Pallid Swift are strictly comparable or eventually a bit lower (but not statistically significant) (Table 4).

F I G U R E 3
Nonbreeding movements (black dots: 2-day averages) and location of extended stopover periods (filled circles) in autumn (a) and spring (b) for Common Swifts tracked by geolocation from northern Italy. Tracking data are color-coded for individual birds The uniformity of survival values found both within different latitude populations of the same species of swift, and between our two species examined here, is someway unexpected. The observation of latitudinal gradients in bird life-history traits has strongly affected the study of avian life-history patterns and evolution (Lack, 1947).
The well-documented latitudinal trend in clutch size or number of clutches per year (Jetz, Sekercioglu, & Böhning-Gaese, 2008) was also found in our study species, where the Common Swift has a northern distribution range and lay a single clutch, while the Pallid Swift with a southern range most often lay two clutches per season.
In a perspective of fecundity-survival trade-off (Saether, 1988), the latitudinal gradient in number of clutches should be matched by an inverse trend in survival probability. Indeed, some studies reported observations that temperate birds have lower survival probabilities than their tropical counterparts (Martin, 1996;Murray, 1985;Skutch, 1985

| Wintering areas and movement of common and pallid swifts
The geolocation tracking by Commons Swift in this study showed that the birds performed their migration earlier in autumn as compared to more northern populations (Åkesson et al., 2012Hedenström et al., 2016) and that their final winter destination were located further to the southeast in Africa (Figure 3). The locations of the nonbreeding areas were further to the east as compared to commons swifts tracked from breeding sites in Germany (Wellbrock, Bauch, Rozman, & Witte, 2017 Perrins (1971) (2) Great Britain 1954-1993 0.76 ± 0.02 a -Φt Thomson et al. (1996) ( 3)

| Relationships between survival and winter climate in Africa
In our study, we found a strong difference of climatic factors influ- and "other" condition. Probably, this result could be related to the recent rainfall recovery with lack of extreme dry events in this 11-year period (Munemoto & Tachibana, 2012;Sanogo et al., 2015). Similar effects of weather were previously found in other species wintering in Western Africa, for example, the Common Nightingale (Luscinia megarhynchos) wintering in Guinean coast influenced on the way of return to Europe by a negative effect of the very dry seasons (Boano, Bonardi, & Silvano, 2004).
The Common Swifts from our study colony seems to winter mainly in Mozambique and nearby countries in southeast Africa, being strongly affected by the ENSO cycle. The ENSO is a phenomenon that originates in the Pacific Ocean with years in which the current is warmer than usual (El Niño) and years when it is colder (La Niña). Various studies of climatologists have found worldwide effects, including some in Africa. For example, in the years of La Niña, in southeast Africa there are often exceptional rains and exaggerated flooding, while a little further north in East Africa there can be droughts instead (Figure 3). According to our study, mortality of our Common Swifts is higher during La Niña years. La Niña involve large shifts of rainfall patterns to the southwest into Australia, Indonesia, and southern Asia. This leaves less rain for eastern African countries including Uganda, Kenya, Ethiopia, and Somalia (Nicholson & Selato, 2000;Schrage et al., 2004), and more south in Africa the precipitation can at the same time be very strong, long-lasting, and widespread causing very severe and extensive flooding (e.g., in 2008; Lukamba, 2010). The areas affected by these rain anomalies correspond to the areas where we know that our marked swifts are spending their nonbreeding period in winter (see maps in Figure 1).
Summing up, the results from our study suggest that the variable climatic conditions, as found in the wintering area, show evident effect on swift's survival only when the adverse effects exceed a certain limit. In this scenario, the swifts' survival is mostly affected when particularly worse conditions, that is, extreme drought or heavy rainfall, occur in the two major regions explored in Africa by the two species, respectively.

| CON CLUS IONS
We have found that two species of swifts that differ in biological traits and wintering area used show similar yearly survival, but are influenced by different climatic drivers, resulting in annual difference in survival. We could confirm that the important climatic variables can be predicted, since precise winter locations are known for our study populations. It is interesting to note that adult average annual survival rate is similar between the two species despite differences in migratory strategy, wintering areas, and breeding biology, with a double brood in Pallid and single brood in Common Swift. The difference in breeding investment between the two species, involving a longer period spent in the breeding areas for Pallid Swifts as compared to the Common Swifts, is perhaps balanced by a shorter migration in Pallid Swifts. Future studies, however, need to investigate why two highly mobile aerial insectivores spending the complete nonbreeding period on the wing are not able to escape from extreme weather conditions by changing wintering area, even if they may be capable of some adjustment as showed by Åkesson et al. (2012) and Norevik et al. (2019).