Long-term changes in autumn migration timing of Garden Warblers Sylvia borin at the southern Baltic coast in response to spring, summer and autumn temperatures

Abstract Many migrant European birds have been departing their breeding grounds earlier in recent decades in response to rising temperatures from climate change. We examined long-term changes in the autumn migration timing of immature Garden Warblers using data from ringing station Bukowo-Kopań on the Polish Baltic coast in 1967–2018. We calculated an Annual Anomaly (AA) of migration and the dates when 10%, 50% and 90% of migrating birds were caught at each station. We modelled AA and the dates of these percentiles of passage for each station against the year and bi-monthly spring, summer and autumn temperatures as explanatory variables in multiple linear regression models. The overall passage (AA) of immature Garden Warblers advanced by 7 days and the dates of the 50th and 90th passage percentiles advanced by 6–11 days. Overall passage, and dates for 50% of passage occurred later the higher summer temperatures rose in Finland. We suggest favourable foraging conditions in warm summers at these breeding grounds and at first stopover sites delayed the passage because these inexperienced birds took advantage of the conditions to accumulate fuel before migrating south. The overall passage, and the dates of 10%, 50% and 90% of autumn passage occurred the earlier the higher were spring temperatures on spring migration route in Italy, and the higher were mean summer temperatures at breeding grounds in southern Sweden and Norway. We suggest a trans-generation carry-over effect, as warm springs encourage early arrival and nesting of returning adults, thus earlier broods and hatching of juveniles, which also grow faster in warm summers. These effects enable immatures to migrate earlier in autumn after a warm breeding season. The long-term increase in Europe’s spring and summer temperatures thus help explain the shift in Garden Warblers’ migration timing.


Introduction
The timing of bird migration is driven by an endogenous rhythm controlled by seasonal changes in photoperiod (e.g. Gwinner 1996;Coppack & Pulido 2004;Dawson 2008, Knudsen et al. 2011), but external factors such as temperatures or feeding conditions often modify the timing of birds' life stages, including migration, breeding and wintering (e.g. Tryjanowski et al. 2002;Strode 2003;Newton 2008;Ockendon et al. 2013;Briedis et al. 2019;Lehikoinen et al. 2019). The timing of migrant passerines' passages and breeding has changed over recent decades in the northern hemisphere in response to climate change (e.g. Sokolov et al. 1999, Jenni & Kéry 2003Tøttrup et al. 2006a, 2006b, Miles et al. 2017Kluen et al. 2017;Lehikoinen et al. 2019).
Most studies show that progressively earlier arrivals in spring are more pronounced in short-distance migrants than in their long-distance counterparts (e.g. Hüppop & Hüppop 2003;Jenni & Kéry 2003;Tøttrup et al. 2006a). Yet many longdistance migrants are also arriving earlier in the northern European spring, for example the Willow Warbler Phylloscopus trochilus and the Blackcap Sylvia atricapilla (Remisiewicz & Underhill 2020, 2022a, 2022b. In many migrants the shift to an earlier spring passage has been attributed to higher winter and spring temperatures in the north, including Pied Flycatchers Ficedula hypoleuca in Finland (Ahola et al. 2004) and Song Thrushes Turdus philomelos at the Baltic coast in Poland (Redlisiak et al. 2021), among tens of other species across Europe and North America (Lehikoinen et al. 2019).
Researchers have focused on how climate change has affected the phenology of spring migration, but autumn migration has received less attention, so less is known about how these changes have affected the phenology of southward migration Rivrud et al. 2016;Haest et al. 2019). Passerines have responded to climate in varied ways, ranging from earlier to later autumn passages (Jenni & Kéry 2003;Gallinat et al. 2015;Miles et al. 2017). These shifts in autumn migration timing over recent decades have been attributed to individual species' breeding biology, including the number of clutches they lay a season , and to migration distance (Jenni & Kéry 2003). In single-brooded species, if adults begin and end breeding early in a warm spring the immatures often depart the breeding grounds early in autumn , as documented in Bluethroats Luscinia svecica in Sweden (Ellegren 1990) and Garden Warblers Sylvia borin in Denmark (Tøttrup et al. 2006b). In contrast, multiple-brooded species might use the breeding period extended by a warm spring to lay additional clutches, such as Barn Swallows Hirundo rustica in Denmark (Møller 2002). Immatures from additional clutches would be ready to depart the breeding grounds later than those from the first clutches and thus show a delayed autumn migration, as do Song Thrushes in Poland (Redlisiak et al. 2018) and Great Tits Parus major in Russia (Bojarinova et al. 2002). Short-distance migrants are under different pressures from long-distance migrants in autumn, with different effects on the timing of their migrations and other life stages (Jenni & Kéry 2003;Gordo 2007). For short-distance European migrants environmental conditions on their non-breeding grounds, often around the Mediterranean, improve as autumn progresses, bringing rainfall that ameliorates the effects of summer droughts (Gordo 2007). Yet short-distance migrants that arrive at their nonbreeding grounds too early might face unfavourable conditions, including phenological mismatches affecting foraging success, which would reduce their chances for survival: thus they might benefit from staying longer at the breeding grounds (Jonzén et al. 2007, Newton 2008). Long-distance migrants usually begin autumn migration right after breeding to take advantage of the peak abundance of food in early autumn at their stopovers and later at their non-breeding grounds (Morel 1973 cited by Gordo 2007, Redlisiak et al. 2018. Jenni and Kéry (2003) suggested that in response to climate change autumn migration has shifted later in short-distance migrants but earlier in long-distance migrants. Nevertheless, long-distance migrant passerines have adopted divergent strategies to cope with these changes: autumn migration of Blackcaps Sylvia atricapilla in Denmark shifted earlier over 1976-1997, but autumn migration of Wood Warblers Phylloscopus sibilatrix in Denmark and Willow Warblers and Whinchats Saxicola rubetra in the UK shifted later over 1955-2014, and no long-term change occurred among Common Whitethroats Sylvia communis in the UK in this period (Cotton 2003;Tøttrup et al. 2006b;Miles et al. 2017).
Few researchers have yet studied phenological shifts in the autumn migration of passerines, which vary widely . Responses to climate change might vary within a family, as among Sylvia warblers, even between the different migration phases of individual species (Fransson 1995, Tøttrup et al. 2006b). For example, in Garden Warblers the first half of autumn migration has advanced, but in Lesser Whitethroat Sylvia curruca the second half of passage shifted later at Fair Isle over 1955-2014, though these shifts have not been attributed to climate (Miles et al. 2017).
Garden Warblers are single-brooded, longdistance migrants that pass through Poland from the beginning of August to mid-September on their way south from breeding grounds in Fennoscandia (Norway, Sweden, Finland) to sub-Saharan Africa (Cramp 1992;Maciąg et al. 2017). We used monitoring data on migrant Garden Warblers collected during 1967-2018 at bird ringing station Bukowo-Kopań, on the Baltic coast of Poland. The ringing recoveries indicate that Garden Warblers migrating through this station come from the breeding grounds in Sweden, Norway and Finland ( Figure 1; Spina et al. 2022). These migratory populations breed mostly in June-July, migrate in autumn from August to mid-November, visit the non-breeding grounds from mid-November to mid-March, and migrate north in spring from mid-March to the end of May (Cramp 1992).
Most studies examining how climate change affects autumn migration focus on the influences of local weather that directly affect this passage, such as local summer temperatures on the breeding grounds (e.g. Cotton 2003;van Buskirk et al. 2009;Horev et al. 2010;Smith et al. 2011;Chambers et al. 2014;Ellwood et al. 2015;Bozó et al. 2018). Temperatures can serve as proxies for the myriad ecological 284 A. Pinszke and M. Remisiewicz conditions that govern a migrant's success, from the availability of suitable habitat to foraging conditions (Walther et al. 2002;Julliard et al. 2004;Thackeray et al. 2016). But few studies have examined the cumulative influence on migrants of factors such as temperatures along spring migration routes that might affect autumn migration indirectly through carry-over effects of phenological shifts in one life stage altering the timing of a later stage (e.g. Sokolov 2006; Barshep et al. 2011;Tomotani et al. 2018). We aimed to identify any multiyear and yearto-year changes in the timing of Garden Warblers' autumn migration. Given advances in autumn migration timing of Garden Warbler over 60 years at Fair Isle (Miles et al. 2017), we expected a shift to earlier autumn passage of this species at Poland's Baltic coast over 1967-2018. Considering the limited literature evidence of carry-over effects of temperatures on migrants, we aimed to determine any relationship between the timing of Garden Warblers' autumn migration and temperatures not only on the breeding grounds and along autumn migration routes, but also on previous spring migration routes. We expected the timing of immature Garden Warblers' autumn migration to be related to the temperatures they had experienced at the breeding grounds in Fennoscandia, and which the adults had encountered earlier during their spring migration to these areas. With this study we aim to fill in a gap in knowledge on a trans-generational carry-over effects of temperatures in migrant birds, which would extend our understanding of the ways the climate change might Figure 1. Location of ringing and weather stations used in the study, and recoveries of Garden Warblers ringed at Operation Baltic stations but recovered elsewhere. Blue star -Bukowo-Kopań ringing station (BK). Blue circles -recoveries of birds caught at Bukowo-Kopań, black circles-recoveries of birds caught at two other coastal stations (Hel, Mierzeja Wiślana), after Maciąg et al. (2017). Yellow trianglesweather stations near breeding grounds and on autumn migration routes: LEB -Łeba, JUU -Juupajoki, BRG -Bergen, KAL -Kalmar, GAR -Gardermoen, SKO -Skovde, GSK -group of three weather stations combined (Gardermoen, Skovde and Kalmar). Green squares -weather stations on spring migration routes: BRI -Brindisi, LUG -Lugano, EIN -Eindhoven. Arrows -general directions of migration of Garden Warblers towards the Polish coast from the breeding grounds in autumn (yellow arrows), and from the non-breeding grounds in spring (green arrows). affect migrants at subsequent stages of their life in different parts of their range.

Materials and methods
We used the autumn migration dates of immature Garden Warblers caught at Operation Baltic's ringing station Bukowo-Kopań (54°19ʹ38″-54°27ʹ54″N; 16° 13ʹ33″-16°25ʹ15″E) on the southern Baltic coast (Figure 1). This station is mostly a stop-over site for migrating Garden Warblers from the breeding grounds in Fennosciandia, but locally recaptured individuals (Nowakowski 2017) indicate that some birds come from local populations, which breed in nearby riparian broadleaved forest (Remisiewicz & Underhill 2022b). Data were collected for the period when these warblers migrate south through Poland, from 14 August to 15 October, in 1967-2018. Bukowo-Kopań station used the same standard datacollection protocol throughout this period (Busse & Meissner 2015). Migrating birds were caught daily in mist nets from dawn till dusk. The number of mist nets remained constant in each season, but ranged from 25 to 89 over the years. Birds were ringed, measured, and Garden Warblers caught in autumn were aged as immatures or adults by their plumage (Svensson 1992;Demongin 2016).
We used the daily numbers of immature Garden Warblers caught each autumn because too few adults were caught to analyse. We noted only the first capture of a bird in the season. The few gaps in data on days when catching was suspended for different reasons, mostly storms, were imputed with the mean number of birds caught at the station on those dates in the six previous years and the six following years, as in previous studies (Redlisiak et al. 2018;Remisiewicz & Underhill 2020, 2022a, 2022b. These gaps occurred in 12 years of the study and there was no more than three gap days in any season. We analysed data only for years in which more than 30 immature Garden Warblers were captured in autumn (Table S1), for a total of 4407 birds.
For each year we determined how many immature Garden Warblers were caught on each day of the season. Then we calculated this number as a percentage of the autumn total of immature Garden Warblers. These cumulative percentages were then converted to a scale of 0-1 for our statistical analyses. Based on these values we drew the daily cumulative curves, reflecting the pattern of autumn migration in each year of the study (Redlisiak et al. 2018;Remisiewicz & Underhill 2020). We determined the Annual Anomaly (AA) of passage in each autumn as the departure of the daily cumulative curve in that year from the manyyear (1967-2018) average daily cumulative curve, as in Remisiewicz and Underhill (2020). A negative Annual Anomaly showed that overall passage that autumn occurred earlier than the many-year average pattern, and a positive AA showed that season's passage had occurred later than the average (Remisiewicz & Underhill 2020, 2022a. To show changes in autumn migration timing in more detail, from the daily cumulative curves we derived the dates when 10%, 50% and 90% of the season's immature Garden Warblers were captured, as in other studies (e.g. Tøttrup et al. 2006b;Miles et al. 2017;Lehikoinen et al. 2019). These dates were used in further analyses as Julian days -the number of days from 1 January. We also calculated the duration of passage for 80% of the Garden Warblers each year as the difference (in days) between the dates when 10% and 90% of the birds had passed. We then used linear regression to check for any multiyear trends in the dates of 10% (beginning), 50% (middle) and 90% (end) of immatures' migration and the duration of migration for each year we studied.
To determine the relationship between temperatures and the timing of autumn migration we used mean temperatures from weather stations on the Garden Warblers' spring migration routes, at their breeding grounds and on their autumn migration routes (Table I, Figure 1). We selected these locations based on the distribution of ringing recoveries of Garden Warblers caught at three Operation Baltic's field stations: Bukowo-Kopań, and at Hel and Mierzeja Wiślana, located ca 150-200 km respectively, to the east ( Figure 1, Maciąg et al. 2017). Using mean daily temperatures from the selected weather stations from April to September 1967September -2018September (http://www.ecad.eu, 2018 we calculated bi-monthly mean temperatures of April-May, June-July and August-September, periods that corresponded with consecutive stages in the life cycle of the Garden Warbler (Table I). Strongly correlated temperature data from three weather stations in south Scandinavia for June-July and for August-September we combined as a single variable, GSK ( Figure 1, Table  S2). We checked if temperatures on the spring migration routes, at the breeding grounds and on the autumn migration routes showed any trends over 1967-2018 by using linear regression of the temperature variables against the year.
The variables were in different measurement units, so we standardised the AA, the Julian dates for the percentiles of passage and temperature variables so that each had a mean of 0 and a standard deviation of 1. We also checked for any correlations between the temperature variables using Pearson's correlation 286 A. Pinszke and M. Remisiewicz coefficient to check for multicollinearity (Dormann et al. 2013), which we considered in the interpretation of our results. We determined if spring, summer and autumn temperatures had influenced the timing of the immature Garden Warblers' autumn migration across the southern Baltic coast by building multiple linear regression models in which the AA and dates for 10%, 50% and 90% of passage (further in the text: "measures of migration timing") were response variables in the model against 13 explanatory variables of the bimonthly mean temperatures at the weather stations (Table I) and the year. Our calculations included many explanatory variables, so as our first step we analysed each response variable against the year and temperature variables from the weather stations for spring, summer and autumn separately. Most correlation coefficients between the temperature variables were within the acceptable level of |r| < 0.7 (Dormann et al. 2013), but eight variables were correlated within spring and summer season above this threshold (Table S2), which might cause a bias to the results of our models (Dormann et al. 2013), so we applied an iterative approach. Thus, in the second step we created two sets of partial models for each of these seasons, where we included only uncorrelated variables (Quinn & Keough 2008). We then assessed whether the explanatory variables selected in the best models derived from these partial models differed from the best models derived from the full models and we evaluated if the values and signs of the effects for each variable differed between the best models derived by these two approaches. When the sign and the strength of each effect were the same in the best models derived in these two ways, we assumed that any multicollinearity in the full model did not affect the selection of the model. Thus we used the best model derived from the full model in further considerations. When the best models derived from the partial models showed no significant effect of any variable, but these derived from the full model included the correlated variables, we further used this latter best model, which indicated that such variables might have a combined influence on bird migration timing. In our third step we analysed each response variable against the year and those temperature variables found to be significant in the best models for spring, summer and autumn. We plotted the residuals of the best models to verify if the models met the assumptions of multiple regression (Crawley 2013). To further evaluate any multicollinearity we calculated the variance inflation factor (VIF), which in all full and partial models and in the best multiple regression models for spring, summer and both seasons combined were all <10, indicating that any multicollinearity was low and did not harm our results (Quinn & Keough 2008;Dormann et al. 2013).
We selected the best models by Akaike Information Criteria corrected for small sample size (AICc) with "all subsets" of the variables using the dredge function in the package "MuMIn 1.43.17" (Bartoń 2020). Using the adjusted determination coefficient (AdjR 2 ) we evaluated the percentage of variation explained by all the explanatory variables of the best models. To evaluate the percentage of variation for each explanatory variable in the models we also determined partial regression coefficients (partial R 2 ) using the package "heplots 1.3-5" (Fox et al. 2007). We then plotted all the temperature variables that had any significant effect on the timing of the immature Garden Warblers' autumn migration against corresponding response variables to examine the influence of each temperature variable separately. The statistical analyses followed Remisiewicz and Underhill (2020) and Table I. Weather stations on Garden Warblers' spring migration routes and breeding grounds, from which we used the mean bi-monthly temperatures of months when the area was used by the birds. Abbreviationssymbols of the temperature variables used in the study, indicating weather station or grouping (as in Figure 1) and season.

Multiyear trends in timing of immature Garden Warblers' autumn migration across Polish Baltic coast
The overall autumn migration timing (AA) of immature Garden Warblers through Bukowo-Kopań shifted earlier by 7 days on average over 1967-2018 (Figure 2(a), Table S3). The timing of the 50th percentile of passage advanced by 6 days over these 52 years and of the 90th percentile by 11 days (Figure 2(b), Table S3). The duration of these immature Garden Warblers' autumn migration shortened by 9 days on average over 1967-2018, because the dates of the end (last 90%) of passage shifted earlier but those of the beginning (first 10%) of migration did not change (Fig. S1, Table S3).

Relationship between temperatures on spring migration north and at breeding grounds and timing of following autumn migration across Baltic coast
All the spring, summer and autumn temperatures we analysed increased significantly by 1.3 to 2.9°C over 1967-2018, except for summer temperatures at Juupajoki in Finland (Table S4). The first step of our analysis showed that autumn migration occurred earlier with higher spring temperatures, and higher summer temperatures were related to earlier or later dates for the median and end of passage at Bukowo-Kopań. Autumn temperatures showed no significant effects. The full models with four spring temperature variables and the year indicated a significant negative relationship between spring temperatures in Brindisi, Italy, and AA as well as the dates of 90% of passage (Tables S5-S10). The full models with four summer temperature variables and the year indicated that summer temperature at Juupajoki, Finland, had a significant positive relationship with the dates on which the median (50%) of immatures migrated through Bukowo-Kopań (Tables S11-S12). The dates for the end of migration (90%) at this station showed a significantly negative relationship with summer temperatures in Łeba, Poland (Tables S11-S12). The best models derived from the full models for spring (Table S6) and summer (Table S12) corresponded well with those derived from the partial models (Tables S8, S10, S14, S16). This showed that multicollinearity did not bias the model selection process, thus we used the best models derived from full models (Tables S5-S6, S11-S12). The models examining four autumn temperature variables and the year as explanatory variables showed no significant relationships with any of the measures of migration timing at Bukowo-Kopań (Tables S17-S18), so we excluded autumn temperatures from the next step of our analysis. In the third step, the best multiple regression models with those spring and summer temperatures found significant in the first step included from one to four temperature variables and explained 13%-35% of the variation in AA and in the dates of the percentiles of passage (Table II, Tables S19-S20). When combined in a single model, the higher were temperatures the preceding spring the earlier all phases of autumn passage occurred, but high summer temperatures were related to early or late timing in different phases of the autumn passage (Table II). The higher the spring temperatures on the migration route in Brindisi in Italy, the earlier Garden Warblers arrived the next autumn at Bukowo-Kopań, as indicated by the negative relationships between these temperature variables and all our measures of migration timing, especially the end (90%) of passage (Table II, Figure 3(a)). The higher the summer temperatures at GSK stations in south Scandinavia the earlier overall passage (AA), beginning (10%) and median (50%) of migration of Garden Warblers occurred at Bukowo-Kopań in autumn (Table II, Figure 3(b)). The higher the summer temperatures at Bergen, Norway the earlier the end (90%) of immature Garden Warblers' passage occurred at Bukowo-Kopań (Table II, Fig. S6(b)). But the higher the summer temperatures in Łeba, Poland, the later the first 10% of migrating Garden Warblers occurred (Table II, Fig. S6(a)). Also the higher the summer temperatures on the breeding grounds in Juupajoki, Finland, the later was the median (50%) and the overall passage of the immature Garden Warbler (Table II, Figure 3(c)).

Discussion
We found that immature Garden Warblers' autumn passage at Polish Baltic coast had ended progressively earlier over the five decades of our study and thus the duration of their passage had shortened. Our results showed that these changes likely occurred in response to summer temperatures at the species' breeding grounds in Fennocandia and Poland and to temperatures the previous spring on their migration routes in Italy. We discuss how the timing of Garden Warblers' autumn migration might respond to temperatures in spring, months before the immatures we examined had hatched, through a carry-over effect across the generations, and also through the direct effects on the immatures of conditions at the breeding grounds.

Shifts in timing of autumn migration on southern Baltic coast compared with other European sites
Our study showed that the timing of the end (90%) of Garden Warblers' autumn passage at the southern part of Baltic coast in Poland advanced over 1967-2018, as it had in 1976-1997 at the Danish island of Christiansø on the southern Baltic ca 130 km northwest of Bukowo-Kopań (Tøttrup et al. 2006b). Garden Warblers' earlier median date of autumn  (Table  S3). This might indicate similar effects of common environmental factors, including the temperatures we analysed, for the species' autumn migration through the Baltic and the North Sea region. As in our study, the last autumn departure dates advanced in 1971-2000 at Oxfordshire, England, (Cotton et al. 2003) though studies at other locations in Europe have shown different patterns. For example, the dates when 5%, 10%, 25% and 50% of Garden Warblers on passage were caught at Fair Island, Scotland, shifted significantly earlier over 1955-2014, but the dates for 75%-95% of passage did not (Miles et al. 2017), in contrast to our results. The median date of autumn passage (50%) of immature Garden Warblers caught at Ottenby Bird Observatory in Sweden showed no significant shift over 1950(Iwajomo et al. 2012, in contrast to our results, where the median passage shifted earlier (Table S3, Figure 2). At Rybachy station on the Courish Spit, Russia, about 315 km east of Bukowo-Kopań, the mean dates of this species' autumn migration shifted later over 1959-1976and 1985-1998, but no trend was found in 1976-1990(Sokolov et al. 1999, in contrast to the significant trend towards earlier median date and overall (AA) autumn passage over 1967-2018 at Bukowo-Kopań (Table S3, Figure 2). These different trends between neighbouring stations in the Baltic region likely resulted from the different periods being studied.
Changes in autumn migration timing might vary even for different phases of the same passage (e.g. Tøttrup et al. 2006b;Miles et al. 2017;Barton et al. 2018). Similarly, when the beginning and end of migration are advanced or delayed the duration of the passage might lengthen or shorten. The duration of Song Thrush migration in spring and autumn lengthened (Redlisiak et al. 2018(Redlisiak et al. 2021, as it did in Lesser Whitethroat (Miles et al. 2017) and in the spring migration of Willow Warbler (Remisiewicz & Underhill 2020). The duration of passage might also shorten, as for Redstart Phoenicurus phoenicurus (Miles et al. 2017). We suggest that the duration of immature Garden Warblers' autumn passage at Bukowo-Kopań field station shortened over the five decades we studied because of the earlier departures of the last cohorts of Garden Warblers from their breeding grounds and thus the last phases of their passage at stopover sites shifted earlier. This finding contrasted with those for most other short-and long-distance migrants in the temperate and boreal zones of the northern hemisphere, whose autumn migrations have extended because of an earlier beginning and a later end of passage compared with earlier decades (Lehikoinen et al. 2019). Metaanalysis of the duration of breeding has revealed that single-brooded species in the northern hemisphere have shortened their breeding season in the past few decades in response to increasing temperatures, in contrast to multi-brooded species, which have extended their breeding period . The condensed breeding season observed in single-brooded species likely results from them fine-tuning their egg-laying dates to match the maximum energetic demands of their offspring with the peak of food resources. Insects now often appear earlier in spring as temperatures trend upwards and foraging opportunities might be Table II. Relationships between overall Annual Anomaly (AA) and migration dates of 10th, 50th and 90th percentiles of autumn migration by immature Garden Warbler at Bukowo-Kopań over 1967-2018 with the temperature variables and year in the best models selected by Akaike Information Criteria (AICc). Statistics for the best model: AdjR 2 -adjusted coefficient of determination, F -result of GLM, p -significance of the best model, Estimate -coefficients of multiple regression, SE -standard error of the estimates, t -t-test, Psignificance of each estimate, VIF -variance inflation factor, R 2 -partial determination coefficient for each estimate. p < 0.05 marked in bold face and 0.05 < p < 0.1 in italics. Symbols of explanatory variables as in Table I  . Thus the shorter autumn migration period we found in Garden Warblers, a single-brooded, insectivorous, long-distance migrant that arrives on the breeding grounds in northeastern Europe in mid-May -late compared with other passerines -likely reflects a shortening of its breeding period.

Timing of autumn migration in relation to temperatures on spring migration
Few studies have examined how the effects of weather encountered by one generation of birds might be carried-over to the timing of autumn migration in the next generation (e.g. Sokolov 2006;Tomotani et al. 2018). Yet our results suggest that the timing of immature birds' autumn migration might be indirectly affected by temperatures their parents encountered at different stages in their life, including on spring migration routes. We found that the higher were spring temperatures in Italy on the Garden Warblers' migration northwards, the earlier the immatures migrating south arrived in autumn on the Polish Baltic coast. This influence could only be indirect, because the immatures we analysed would have hatched after spring, so these conditions could only have affected their parents. Warm springs might encourage the earlier arrival of adults at the breeding grounds and earlier  (Table II) with trends by linear regression. βregression coefficient. *** -p < 0.001, * -p < 0.05, + -p < 0.1. All variables are presented as standardised values. (a) AA and 10th, 50th and 90th percentiles against April-May mean temperature in Brindisi, (b) AA and 10th and 90th percentiles against June-July mean temperature in South Scandinavia, (c) AA and 50th percentile against June-July mean temperature in Juupajoki.
nesting than in years with a cold spring (Tomotani et al. 2018), thus the earlier hatching of the immatures followed by an earlier departure of both age groups from the breeding grounds (Ellegren 1990;Sokolov et al. 1999;Gordo 2007). The early spring passage of Garden Warblers at Hanko Bird Observatory in Finland corresponds with high temperatures in May on the migration routes north, confirming that a warm spring encourages the early arrival of adults at the breeding grounds (Halkka 2011).

Timing of autumn migration in relation to temperatures on breeding grounds
Several studies have shown that high summer temperatures at the breeding grounds affect the timing of autumn migration, but most of these studies have focused on temperatures at a single breeding area (e.g. Cotton 2003;Ellwood et al. 2015;Bozó et al. 2018). Few studies have investigated the effect of temperatures at widespread breeding grounds on subsequent autumn migration timing (e.g. Redlisiak et al. 2018;Haest et al. 2019), as we did. We found that temperatures at the different breeding areas in Fennoscandia directly influenced the migration timing of immature Garden Warblers at the southern Baltic coast. Warm summers enable immatures to grow faster, thus depart earlier from the breeding grounds, and thus then appear earlier at autumn stopover sites than after cold and rainy summers (Crick and Sparks 1999;Redlisiak et al. 2018). This effect likely explains why the overall passage (AA) and the first half (10%, 50%) of migrating immature Garden Warblers shifted earlier at Bukowo-Kopań in autumns after higher summer temperatures on the breeding grounds in south Scandinavia (GSK stations) in June-July (Table  II). An earlier end (90%) of Garden Warblers' passage through Bukowo-Kopań corresponded with higher summer temperatures in Norway (BRG) in June-July (Table II), as in Oxfordshire, UK, where the last 95% of Garden Warblers passed through earlier with higher local summer temperatures (Cotton 2003). The positive relationship between temperatures at local breeding grounds in Poland and the timing of the beginning of Garden Warblers' passage (10%) at Bukowo-Kopań (Table II), indicating later passage, probably accounted for the lack of a clear trend in the timing of this cohort of migrants. We found the later overall passage (AA) and median (50%) of migration at Bukowo-Kopań with higher summer temperatures in Finland (Table II), which suggests that warm summer at the breeding grounds and at early stopover sites might delay passage at the southern Baltic coast because the migrants take advantage of favourable feeding conditions to accumulate fuel for their long trip to sub-Saharan Africa. Garden Warblers gradually accumulate fuel reserves along their migration route and their median fat score in autumn at Bukowo-Kopań was a modest 2 on a scale from 0 to 7 (Ożarowska 2015). These inexperienced birds accumulated this fat at the breeding grounds or at their first stopover sites. Nevertheless, the timing of end of Garden Warblers' migration clearly shifted earlier, despite higher summer temperatures at some breeding grounds corresponded with a later passage of the first half of migrants. We suggest that temperatures that advance autumn migration -such as higher temperatures along the northward migration routes the previous spring, and at the breeding grounds in Norway and southern Sweden -have a stronger overall effect than the delay incurred at certain breeding grounds (Finland, Polish coast) on a cohort of immature migrants.
Different influences of summer temperatures at various breeding areas on subsequent stages of Garden Warblers' migration suggest that different cohorts arrive at the southern coast of the Baltic from different parts of the breeding grounds ( Figure 1) (Maciąg et al. 2017;Spina et al. 2022). Our results show that the higher the summer temperatures on the breeding grounds in southern Scandinavia, the earlier immature Garden Warblers arrive in autumn. If we assume that most Garden Warblers ringed at Bukowo-Kopań come mostly from southern Scandinavia as suggested by ringing recoveries (Figure 1) (Maciąg et al. 2017;Spina et al. 2022), this, combined with the earlier passage related to high spring temperatures, would explain the general trend towards an overall earlier autumn passage at this station.

Many-year trends and year-to-year variation in timing of immature Garden Warbler's autumn migration as a response to temperature
Spring temperatures on migration routes and summer temperatures at the Garden Warblers' breeding grounds rose over 1967-2018 (Table S4), likely as an effect of global climate change (Blunden & Boyer 2021). Considering the many different relationships between these temperatures and the timing of the species' migration that we found, the multiyear trend of rising temperatures partly explains the many-year trend towards an earlier end of the immatures' autumn passage and its shorter duration at the southern Baltic coast. The timing of bird migration 292 A. Pinszke and M. Remisiewicz is generally determined genetically (e.g. Coppack & Pulido 2004;Dawson 2008) but can be modified by temperature (e.g. Tryjanowski et al. 2002;Ockendon et al. 2013;Lehikoinen et al. 2019), therefore birds can fine-tune their migrations to adapt to climate change in the northern hemisphere (e.g. Jenni & Kéry 2003;Tøttrup et al. 2010). In several other long-distance migrants between Africa and northern Europe, the timing of spring migration at Bukowo-Kopań was shaped by the combined influence of certain climate variables showing longterm trends and others showing no trends (Remisiewicz & Underhill 2020, 2022a, 2022b. Similarly, the timing of immature Garden Warblers' autumn migration through the Baltic coast is likely a combination of their response to temperature variables showing long-term trends and to factors which showed no such trend, as summer temperatures in Finland (Table S4).

Conclusions
The relationship we found in Garden Warblers between an earlier end to the immature's autumn passage south at the Baltic coast and a warm spring during the adults' migration north to the breeding grounds suggests a carry-over effect with the timing of the life stages of one generation on the next. Tomotani et al. (2018) showed a similar carry-over effect in another migrant passerine, the Pied Flycatcher, where adults breeding early aided the survival of their offspring. To our knowledge no other study has shown a trans-generational carryover effect of temperatures influencing the migration timing of immature birds. In addition, the offspring respond directly to summer temperatures at the breeding grounds. Warm summers aid early maturity, thus early departure from the breeding grounds on autumn migration, but the birds might dally to take advantage of optimal foraging opportunities before heading south to sub-Saharan Africa. The many-year trend to an earlier end of autumn migration by immature Garden Warblers at Poland's Baltic coast along with the year-to-year variations we revealed are therefore likely shaped by the combined effects of the indirect influence of spring temperatures on adults en route to their breeding grounds and the direct influence of summer temperatures on their offspring.
to this paper by collecting the data at Operation Baltic's ringing stations over the decades. Data used on this study is available at the Global Biodiversity Information Facility database at: www. gbif.org, as "Ringing Data from the Bird Migration Research Station, University of Gdańsk" (Nowakowski 2017). We used the mean daily temperatures of the weather stations from the European Climate Assessment and Dataset (http://www.ecad. eu). Joel Avni commented on and edited earlier drafts of the manuscript.

Funding
Fieldwork, collection and digitalization of the data has been supported over the years by the Special Research Facility grants (SPUB)

Disclosure statement
No potential conflict of interest was reported by the author(s).