Timing of spring departure of long distance migrants correlates with previous year's conditions at their breeding site

Abstract

Precise timing of migration is crucial for animals targeting seasonal resources at locations encountered across their annual cycle. Upon departure, long-distance migrants need to anticipate unknown environmental conditions at their arrival site, and they do so with their internal annual clock. Here, we tested the hypothesis that long-distance migrants synchronize their circannual clock according to the phenology of their environment during the breeding season and therefore adjust their spring departure date according to the conditions encountered at their breeding site the year before. To this end, we used tracking data of Eurasian curlews from different locations and combined movement data with satellite-extracted green-up dates at their breeding site. The spring departure date was better explained by green-up date of the previous year, while arrival date at the breeding site was better explained by latitude and longitude of the breeding site, suggesting that other factors impacted migration timing en route. On a broader temporal scale, our results suggest that long-distance migrants may be able to adjust their migration timing to advancing spring dates in the context of climate change.

1. Introduction

Animal migration is an adaptation to seasonal variations in food resources and environmental conditions [1]. The phenology of migration and most importantly the timing of spring migration are crucial to match the timing of reproduction with the peak of food resources and thereby maximize breeding success [2,3]. When migrants leave their wintering grounds, they should ideally adjust their departure with spring phenology at their breeding site. While for short-distance migrants local cues at the wintering site can be good predictors of conditions at the breeding site [3], this is likely not the case for long-distance migrants [4]. Even large-scale climatic indices such as the North Atlantic Oscillation index seem to be poor predictors of bird migration timing [4,5]. Therefore, for long-distance migrants, environmental conditions at their breeding sites are likely to be unknown when they depart. In this case, it has been proposed that the timing of spring departure is controlled by an internal circannual clock, itself influenced by the photoperiod and other stimuli able to fine-tune events year-round [2,3,6,7]. The underlying mechanisms synchronizing the circannual clock and triggering the departure of spring migration are still poorly understood [2,6].

To link the timing of migration with environmental conditions, one needs to assess spring onset over a large geographical scale. At the scale of a continent, this can be done by monitoring the change in colour of the vegetation with satellite imagery, to identify when budburst starts in deciduous forests [8]. Even though green-up of the vegetation might not be the direct cue influencing the timing of migrants, it gives a good proxy of spring phenology and is a particularly relevant approach when studying the breeding habitat of populations with a large breeding range [9–11].

Here, we propose to test whether long-distance migrants schedule their spring departure according to the environmental conditions encountered the previous year at their breeding site, via the synchronization of their circannual internal clock during the breeding season, rather than through local cues at the wintering site. More specifically, we predict that the spring migration departure date is correlated to the breeding site green-up date experienced by the bird during the previous year, rather than to the breeding site green-up date of the same year, not known at the time of departure. We also predict that arrival dates at the breeding site are adjusted according to conditions encountered en route [2,12,13].

We used a dataset of GPS tracked Eurasian curlews (Numenius arquata arquata, hereafter curlew) nesting over a large area of Northeast Europe (including western Russia) and occupying contrasting western European wintering grounds. Spring phenology at the breeding sites was assessed by the green-up date of the vegetation measured by satellites.

2. Material and methods

(a) . Tracking data

A total of 35 complete spring migration tracks were obtained from 26 adult curlews in 2016 and 2017. Nine birds were tracked during two consecutive years. Deployments were made in three countries, either during the non-breeding period (mid-July to mid-April; Germany, Wadden Sea coast, n = 7 and France, Pertuis Charentais, n = 16) or during the breeding season (May; Estonia, n = 3). Details on the capture methods and deployment sites are given in the electronic supplementary material, S1 [20]. GPS tags were EOBS bird solar GPS-UHF tags (20 g, e-obs GmbH, Gruenwald, Germany) in Estonia, Ecotone Sterna GPS-UHF (35 × 16 × 10 mm, 7.5 g) in France and Ecotone Skua GPS-UHF-GSM (58 × 27 × 18 mm, 17 g) in Germany (Ecotone Telemetry, Gdynia, Poland). GPS tags recorded fixes every 5–60 min.

To extract nest positions from the GPS tracks, we isolated the breeding period and counted the number of positions within each cell of a grid (0.0001° × 0.0001°) covering the range of longitudes and latitudes visited by each individual during the breeding period. The position of the nest was defined as the centre of the cell with most counts.

Migration dates (day of year) were obtained by visual examination of the tracks on QGIS [14]. Departure and arrival were defined as the first and last locations on a migration bout (highly unidirectional flight, high speed). When birds showed a prospecting behaviour upon arrival at their breeding ground, the arrival date was determined as the first stop within a 10 km radius around the nest site.

(b) . Green-up date of the vegetation as an index of spring phenology

Satellite remote sensing has often been used to study the leaf phenology of various ecosystem types. We used the method by Delbart et al. [8], which determines the spring date when the ecosystem greens-up (hereafter green-up date). This method uses a combination of near- and short-wave infrared spectral bands, instead of the ‘classic’ vegetation spectral index based on the red and near-infrared bands; this avoids disturbance of the radiometric signal by snowmelt, which affects other algorithms [15,16]. Over boreal forests, the green-up date is closely related to the observed leaf budburst date of deciduous trees, with an 8-day root-mean-square difference (RMSD) with budburst observations [8,17]. This RMSD arises from the remote sensing data pre-processing that retains only the cloud-free observations and reduces the observation frequency, and from the phenological variability within the pixel that the method cannot catch. Nevertheless, this method allows the monitoring of interannual variations in phenology [17–19]. The remote sensing method is here applied to PROBA-V data from 2014 to 2017, and our study area is located where the method's reliability is the highest [8,16,18]. Details are presented in the electronic supplementary material, S2 [20].

For each nest site, green-up date was calculated for a 10 km buffer around the nest. Curlews are extremely faithful to their breeding site [21] and breed within less than 5 km of their previous nest site (P Bocher 2021, unpublished data). Thus, we assumed that they were breeding at the same location the previous year, which was the case for all curlews tracked for more than 1 year in our dataset. One bird captured in France bred in Germany in an area located in the temperate zone, whereas the remote sensing method has been designed for continental and polar bioclimatic zones [22]. Nevertheless, as it was not an outlier in our data, it was kept in the analyses. We tested whether green-up dates differed between 2 consecutive years, using a Wilcoxon signed-rank test (electronic supplementary material, figure S1 [20]).

(c) . Statistical analyses

Statistical analyses were performed under R v. 4.0.3 [23]. We used linear mixed-effect models (nlme package [24]) to take into account the fact that some birds were sampled twice and that we sampled at three different locations, following methods described in Zuur et al. [25]. We used a model selection approach based on Akaike information criterion (AIC) [26] and compared seven models to identify which variables best explained the migration dates of curlews (table 1). We modelled the departure or arrival date of the spring migration as a function of green-up date of the current year (model 1), the previous year (model 2) or two years prior (model 3). Green-up dates of successive years were correlated and were therefore not included in the same model. The green-up date two years prior was included as a negative control, i.e. we did not expect it to be a good predictor of timing of migration according to our hypothesis. A fixed effect ‘year’ was included because the timing of spring could differ between years over the whole study area. We also modelled departure and arrival dates as a function of the mean green-up date for the study period (2014–2017, model 4), and as a function of the nest latitude and longitude (models 5–7) to test whether curlews rather left according to the average conditions they experienced or based on the location of their nest. The random term contained ‘individual’ nested in the country of capture (France, Germany or Estonia) to take into account sampling groups and repeated measurements per individual. Models were compared using the AIC corrected for small sample sizes (AICc), and all models with a ΔAICc lower than 2 were considered equally good [27]. We graphically checked the selected models for homogeneity, independence and normality of the residuals [25]. A repeatability index (intraclass correlation coefficient) was calculated for individuals tracked for two successive years, using the R package ICC [28].

Table 1.

Summary of statistical models. GU_n = green-up date of the same year than the migration. GU_{n − 1} = green-up date of the previous year. GU_{n − 2} = green-up date two years prior (used as a control). AvgGU = average green-up date for the period 2014–2017. Longitude and latitude of the breeding site. Models with the smallest AIC corrected for small sample sizes (AICc) are represented in italics and their estimates are given below.

(a) departure date
model selection
model no.	model	d.f.	AICc	ΔAICc
1.0	intercept	4	254.12	30.54
1.1	GUn + year	6	226.66	3.08
1.2	GUn − 1 + year	6	223.58	0.00
1.3	GUn − 2 + year	6	235.81	12.23
1.4	AvgGU	5	233.26	9.68
1.5	latitude	5	230.38	6.80
1.6	longitude	5	245.34	21.76
1.7	latitude + longitude	6	230.21	6.63
parameter estimation
model no.	parameter	estimate	s.e.	t-value	p-value
1.2	intercept	34.01	10.71	3.18	0.0042
1.2	GUn − 1	0.56	0.08	6.75	0.0003
1.2	year 2017	−2.57	1.31	−1.96	0.0909
(b) arrival date
model selection
model no.	model	d.f.	AICc	ΔAICc
2.0	intercept	4	275.19	47.4309
2.1	GUn + year	6	243.61	15.8473
2.2	GUn − 1 + year	6	244.89	17.1270
2.3	GUn − 2 + year	6	256.30	28.5429
2.4	AvgGU	5	235.51	7.7500
2.5	latitude	5	233.96	6.2000
2.6	longitude	5	261.19	33.4300
2.7	latitude + longitude	6	227.76	0.0000
parameter estimation
model no.	parameter	estimate	s.e.	t-value	p-value
2.7	intercept	−86.50	19.15	−4.52	0.0002
2.7	longitude	0.42	0.13	3.15	0.0162
2.7	latitude	3.04	0.35	8.68	0.0001

3. Results

Tracked curlews bred over a large part of the subspecies breeding range [21]: from 9.35° E to 52.24° E and from 52.75° N to 64.54° N (figure 1). The mean migration distance was 2453 ± 754 (s.d.) km (electronic supplementary material, table S1 [20]). Median departure date was April 10th (range March 16th–May 4th). Median arrival date was April 23rd (range March 17th–May 14th). Departure and arrival dates were correlated to the green-up date of the vegetation at the breeding site (figure 2 and table 1). Individual spring departure date from the wintering site was best explained by the model containing green-up date of the previous summer and year as fixed effects (table 1a and figure 2). Arrival date at the breeding site was best explained by the model containing longitude and latitude of the nest (table 1b). The green-up dates at nest sites were significantly different between 2 successive years, with on average earlier green-up dates in 2016 than 2015 and 2017 (electronic supplementary material, figure S1 [20]; Wilcoxon signed-rank test: 2016: p-value = 0.04, effect size r = 0.50, n = 17; 2017: p < 0.001, effect size r = 0.87, n = 18). For nine birds tracked during two consecutive spring migrations, repeatability was 0.88 (95% CI: 0.60–0.97) for the departure date and 0.91 (95% CI: 0.68–0.98) for the arrival date.

Figure 1.

Nest sites of curlews and mean green-up dates for years 2015–2017. Dates are given as day of the year.

Figure 2.

Relationship between spring migration date and green-up date of the previous year. Orange: year 2016, blue: year 2017. Dots represent data. Lines represent the fitted values and 95% confidence interval of model 1.2 (table 1). All date units are day of the year.

4. Discussion

Using a unique dataset of long-distance migrant tracks gathered by three European research teams over two years with contrasting spring phenology, we demonstrate that the onset of spring migration in Eurasian curlews is correlated with environmental conditions at their breeding site (spring phenology) encountered the previous year. Our results thus suggest that curlews might fine-tune their annual endogenous clock during the breeding season for the coming year and use this clock to depart on time for the next spring migration. Arrival at breeding sites was, however, better explained by longitude and latitude of the nest, suggesting that other environmental factors impact migration timing en route [12].

From an evolutionary perspective, synchronizing the circannual clock and subsequent annual movements during the breeding season appears to be the most relevant strategy, as this period is critical for individual fitness, especially for migrants breeding in highly seasonal environments where the optimal window to reproduce is short [29]. By doing so, long-distance migrants use their own experience to schedule departure at a time when they cannot rely on local cues at their wintering site to anticipate conditions at their breeding ground. Since most bird migrant pairs neither migrate together nor overwinter at the same place [3], setting the circannual clock and the onset of spring migration according to the previous breeding season ensures synchrony among the pair, which is known to reduce the risk of divorces and improve the pair's reproductive success [30,31]. An alternative hypothesis could be that curlews may gain experience year after year and adjust their migratory behaviour based on more than one breeding event. This hypothesis was not supported by our results as the average green-up date did not better explain the departure dates. A longer dataset would be required to obtain across-lifetime green-up dates and further investigate this possibility.

Among all yearly events in a migratory bird's life cycle, the onset of spring migration appears to be the least variable, both in natural conditions [32,33] and for captive birds under constant conditions [2,34]. This most likely reflects the constrained time window to reproduce in seasonal environments [29]. Yet other parameters may also influence this timing, such as climatic conditions, winter habitat quality, social interactions or age [35–38], and as proposed here previous breeding conditions. Similarly, the timing of migration can be adjusted after the departure, along the way after the departure, according to local environmental cues that become increasingly informative the closer the bird gets to the breeding grounds [11–13,33]. The differences in models selected for the departure and arrival dates in our study suggest that some of the above-mentioned environmental factors influence migration timing en route. Yearly events are therefore dependent on the circannual clock, but constantly fine-tuned by environmental factors [2,3].

While we used green-up date of the vegetation as an indicator of spring phenology at the curlew's breeding site, it is probably a proxy only for other tightly linked cues [9] that directly influence their breeding success and that can be used by birds to synchronize their cycle. For curlews, it might be insect peaks and access to other invertebrate prey, essential to ensure chick provisioning and survival [39], or vegetation height, important for incubating parents and chicks to hide from predators in such ground-nesting meadow birds [40].

In the context of global warming, changes in migration dates have been observed at the population level [10,41,42]. By contrast, at the individual level, most studies tracking migratory birds across several years found a high individual consistency in the timing of migration (e.g. [5,32,43–47], but see [48]), including our study where the repeatability of migration dates was similar to previous findings from other migratory birds [32,44,47]. Previous studies conclude that there is a low phenotypic plasticity in the timing of migration and eventually a mismatch with food resources leading to negative consequences on fitness [49]. Individual consistency in the timing of migration in these studies could also be explained by the repeatability of spring timing of the environment from year to year and may not necessarily imply a low phenotypic plasticity. As shown by the correlation we found between spring departure date and green-up date of the previous year at the breeding site, we propose that there might be some plasticity in the timing of migration, which allows a synchrony of the breeding activity with phenological changes of the environment. Further investigations are needed to test this hypothesis on different species and with longer repeated measurements from the same individuals in order to better evaluate phenotypic plasticity and possible adjustments to ongoing global changes.

Ethics

Curlews in Germany were caught and treated with permission by the Ministerium für Energiewende, Landwirschaft, Umwelt, Natur und Digitalisierung of the federal state of Schleswig- Holstein (file nos. V 312-7224.121-37(42-3/13) and V 241-35852/2017(88-7/17)) as well as by the Niedersächsisches Landesamt für Verbraucherschutz und Lebensmittelsicherheit of the federal state of Lower Saxony (file no. 33-19-42502-04-17/2699). In Estonia, license was provided by Matsalu Ringing Center, Estonian Environmental Agency, for the program called ‘Program of marking Eurasian curlew’; ringing licenses were 3-2013 for R.M. and 4-2013 for J.E. In France, license was provided by the ringing and tagging program of the National Museum of Natural History (Paris) CRBPO, N°366 under the responsibility of Philippe Delaporte.

Data accessibility

The dataset supporting this article [50] is deposited at https://doi.org/10.5061/dryad.8sf7m0cp6.

Authors' contributions

P.S., R.M., J.F., S.G., J.E., P.D., P.R. and P.B. collected the data. F.A., N.D., F.D. and P.B. conceptualized the study. F.A. and N.D. analysed the data. F.A. wrote the original draft of the manuscript. All authors reviewed and edited the draft, gave final approval for publication and agree to be held accountable for the work performed therein.

Competing interests

We declare we have no competing interests.

Funding

This study was funded by Estonian Environmental Investment Centre, European Regional Development Fund (grant QUALIDRIS), Contrat de Plan Etat-Région, CNRS  (grant ECONAT), German Federal Agency for Nature Conservation (BfN) (grant nos. FKZ 3515822100 and FKZ 3519861400) and Ligue pour la Protection des Oiseaux.