Seed and berry crop data
We used yearly seed crop size from silver and downy birch (Betula pendula and Betula pubescens, combined as birch species Betula spp.) and Norway spruce (Picea abies, hereafter spruce) and fruit crop size of rowanberry (Sorbus aucuparia) in Finland (Gallego Zamorano et al. 2018). The crop estimates for spruce, silver and downy birch are based on the monitoring programme of the Natural Resources Institute Finland; the crop sizes of the trees have been estimated for tens of stands across the country since 1979 (detailed methodology in Gallego Zamorano et al. 2018). The crop sizes of rowanberry have been collected as a part of the Finnish winter bird monitoring since 1986, where observers estimate the initial rowanberry crop in early winter using the relative abundance scale of six categories from no berries to very abundant (Gallego Zamorano et al. 2018).
Bird migratory data
The spring and autumn migration counts of migratory birds have been collected at the Hanko Bird Observatory (59°48′ N, 22°53′ E), Southwest Finland since 1979. Here, we used 29 years (1986–2014) of autumn median departure days of eleven common frugivorous bird species in Finland (Table 1). These species are mainly short-distance migrants that winter usually in western, central or southern Europe, or are partially migratory, i.e. some individuals over-winter in Finland. Spring arrival median days included 27 years of data (1987–2014). Data from the spring 1990 are missing due to lack of observation effort. The phenology of migratory bird species was monitored at Hanko using standardised migration counts (including four hours standardised migration observation from the sunrise and counts of staging birds) and trapping data (including standardised mist-netting sites) from 25 July to 5 November (Vähätalo et al. 2004; Lehikoinen 2011; Lehikoinen and Jaatinen 2012). The combined data of all observation activities including the number of trapped birds resulted in a daily bird count. Observation activity covered over 95% of observation days annually and there was no trend in observation phenology (Lehikoinen 2011). Complete spring arrival data were available for eight bird species during 28 years; in three species, the number of years was 17–26. The migration date was recorded as the yearly cumulative number of days from 1st January [day 1 is 1st January; for more details see (Vähätalo et al. 2004)].
Table 1 The mean autumn and spring migration dates (day 1 is 1st January) with minimum (min) and maximum (max) dates for eleven short-distance migratory species feeding mainly on fruits (F) or seeds (G), n refers to the number of years Weather data
We calculated the autumn and early spring temperatures from southern Finland between 660 and 690 latitudes in the Finnish uniform coordinate system (59°40′–62°10′ N, 21°30′–32°00′ E), using weather data provided by the Finnish Meteorological Institute (Venäläinen et al. 2005). We used the mean September temperature for autumn migration, except for the late migrating species Bohemian Waxwing Bombycilla garrulus, Eurasian Blackbird Turdus merula, Fieldfare Turdus pilaris, Redpoll Carduelis flammea and Northern Bullfinch Pyrrhula pyrrhula (Lehikoinen and Vähätalo 2000), where we used the mean temperature of September and October. In spring, we used the mean March temperature in the analyses to describe the severity of the early spring.
Statistical analyses
At the scale of southern Finland, tree crop sizes show strong spatial autocorrelation (Gallego Zamorano et al. 2018). To generate an uncorrelated crop size of tree species, we extracted principal components from the yearly crop size of the tree species (birch, spruce and rowanberry; Table 2).
Table 2 Correlation of yearly variation (n = 29 years) in the crop size with the principal component score (PCA1) in three species of trees We used linear mixed models (LMM) to test which factors explain the median migration days of species during autumn and spring separately. In the full model, the fixed explanatory variables were crop size, temperature (September/October for autumn and March for spring) and year to account for potential linear trends in the migration dates. As random effects, models had for each study species separate intercepts and slopes for the crop size, the temperature and the year. The crop size and the temperature values were normalized (mean zero, standard deviation 1) before the analyses.
Furthermore, to test the connection between the autumn and spring migration dates, we used LMM, where the spring arrival dates of species were explained by the autumn migration dates of the previous year and year as a continuous variable. The autumn migration dates were centred species specifically before the analyses to account for species having different autumn migration periods. Species was a random factor in the analyses to account for that species have different spring arrival periods.
LMMs were calculated using the add-on package lmer (package of lme4) (Bates et al. 2015) and package visreg for drawing figure (Breheny and Burchett 2017) in R version 3.4.1 (RCore 2018). Correlations and principal component analyses were performed using the IBM SPSS statistical package, version 23.