Introduction

Habitat exploration by birds helps ensure their food supply and safety (e.g., Mettke-Hofmann 2017). Exploratory movements allow individuals to gather information on local habitat quality before deciding where to settle during subsequent days, seasons, or even years. Previous studies reported exploring for molting sites (Fondell et al. 2013; Lewis et al. 2010), breeding sites (Arlt and Pärt 2008; Fijn et al. 2014; Ponchon et al. 2015), and sites for feeding (McEvoy et al. 2015; Vergara et al. 2016). Exploratory movements may also enable birds to adapt to habitat changes, e.g., due to new predators or climate change (McEvoy et al. 2015; Mettke-Hofmann 2017). Exploratory trips may thus have important impacts on individual fitness in terms of survival and reproductive success (Huang et al. 2016; Reed et al. 1999).

Individual frequency and patterns of exploratory movements are poorly understood (Koenig et al. 1996; Ponchon et al. 2013). There are two main reasons for this: First, before the development of small, powerful GPS devices, it was difficult to track the movements of individual birds over a broad geographic and temporal scale. Second, software that could be used to objectively identify the exploration movements has only recently been developed (Gurarie et al. 2016; Kays et al. 2015). Since the improvement in these two methods, studies of exploratory movements have been conducted that revealed some patterns associated with exploratory movements in various bird species: Exploratory movements were shown to be induced by radically changing habitat conditions (McEvoy et al. 2015), failed breeding (Fijn et al. 2014; Ponchon et al. 2017), and were shown to be performed intensively just before periods when individuals were constrained to small ranges, for example before molting of flight feathers including flightlessness (Fondell et al. 2013; Lewis et al. 2010) and breeding (Reed et al. 1999).

Birds living in dynamic habitats in particular would be expected to perform exploration movements (Campioni et al. 2017; Fijn et al. 2014; McEvoy et al. 2015).

One spatially and temporally dynamic habitat for birds is the Wadden Sea. Lunar cycles, wind speed, and direction influence the duration for which a place is flooded, the magnitude and the direction of the current, and thus the composition of the substrate. In addition, winter storms and ice abrasion can affect the distribution of the sediment (Bartholomä et al. 2009). These hydrodynamic factors create a variety of habitats for small macrobenthic animals that live in and shape the mud (Brouwer et al. 2000), and which provide food for millions of birds (Beukema et al. 1993; Horn et al. 2020). Seasonal and annual changes in abiotic environmental conditions, as well as flocks of feeding birds, affect the numbers and body masses of macrobenthic animals, their species composition, and the geographic distribution of larval and adult stages (Beukema et al. 1993; Horn et al. 2020).

In the German–Danish Wadden Sea, we studied the exploratory movements of one bird species: the common shelduck (hereafter Shelducks, Tadorna tadorna). Shelducks in the Wadden Sea feed mostly on small macrobenthic invertebrates by sifting the top layer of the sediment (Anders et al. 2009; Oelke 1979). Their annual cycle includes a complete molt of wing and tail feathers, resulting in a flightless time of 25–31 days between June and September (Hoogerheide and Hoogerheide 1958). During this period, shelducks from most of northwestern Europe have gathered in the wider Elbe estuary (Hoogerheide and Kraak 1942) for at least 200 years (Oelke 1981; Todt 1985). However, the core of the traditional molting area in the Elbe estuary shifted about 20 km southwest about 120 years ago (Glutz von Blotzheim 1988; Goethe 1957), and has since shifted back toward the present core area in recent decades (Kempf and Eskildsen 2000; Todt 1985). Moreover, numbers of molting shelducks in the traditional molting area have recently decreased from about 200,000 (1996–2002) to about 155.000 (2003–2009), corresponding with increased numbers 240 km to the west in the Dutch Wadden Sea from several thousand during the 1990s up to about 60,000 (2010–2012), suggesting that at least some individuals have changed their molting site (Kempf and Kleefstra 2013; Kleefstra et al. 2011). This indicates that shelducks might select the molting site based on ongoing explorations.

This study aimed to identify how, when, and to what geographic extent individual shelducks explored the Wadden Sea during several months in which they were tracked with GPS devices. We identified exploration movements using change point analyses. Specifically, we aimed to: (1) describe how common exploration movements occurred; (2) determine if shelducks explored by roaming movements, visiting many places along the way, or via short-lasting long-distance movements; (3) examine patterns to see if individuals explored places in advance and returned to those places later, or if they decided to stay at the explored places immediately; and (4) investigate if individual molting places were explored with more effort than other areas.

The flightless time during molting is a critical period with potentially strong effects on the fitness. We hypothesized that individual molting sites would be explored with more effort than other areas.

Materials and methods

Study area

Our study area was the eastern part of the German Wadden Sea (Fig. 1). The area is characterized by tidal flats including tidal channels, sandy shoals, seagrass meadows, sandbars, and estuaries. Shelducks were caught and tagged in nearby breeding grounds (Fig. 1, Table 1).

Fig. 1
figure 1

Study area in the Wadden Sea. BHK Beltringharder Koog, KW Katinger Watt

Full size image
Table 1 Overview of solar-powered GPS-tagged common shelducks
Full size table

GPS tagging

We equipped 11 shelducks, aged ≥ 2 years, with 30-g solar-powered GPS devices (Microwave Telemetry, MD, USA) during the breeding seasons 2011 to 2013 (Table 1). Age was determined by the absence of white tips on the secondaries and inner primaries and by pure white wing coverts (Hori 1965), and sex was determined by visual differences in bill and plumage cues (Patterson 1982). We used two types of devices (Table 1): devices using ARGOS satellites to transmit the data (n = 9) were programmed to locate the birds twice a day during the winter (November to February), four times a day during the breeding season (May to June), and hourly otherwise (Fig. 2), and devices using GSM satellites (n = 2) located the birds according to the battery charge, resulting in more frequent location events, sometimes every minute (Table 1). Females (n = 3) were caught at their breeding burrows using a small mist net (mesh size 3 cm) tacked at the entrance. Males (n = 8) were caught using a whoosh net (10 × 5 m, mesh size 4 cm; Pete Reid, UK) combined with a mobile phone-based remote-control release (Kristian Lilje, Germany), placed near burrows where they gathered in groups in the mornings.

Fig. 2
figure 2

Tracking periods for three female and six male shelducks. Three panels are displayed for each individual: the upper panel shows the movement behavior according to BCPA results, the central panel when the individual was located in a possible breeding habitat (usually the catching place) and in the molting area (compare Fig. 1), and the lower panel refers to periods when the individual was incubating or undergoing molting including flightless

Full size image

GPS devices were attached to the birds as a backpack harness with a 3.8 mm wide Teflon cord (Bally Ribbon Mills, USA) according to Roshier and Asmus (2009). The devices accounted for a mean of 3.1% of body weight for females and 2.5% for males. Birds were measured, weighed, and ringed on their leg with a metal ring (Helgoland bird ringing center) and a coded color ring. One female stopped breeding after catching, the other two bred successfully. Two of the female shelducks were found dead, probably killed by a bird of prey and a fox, respectively, during an unusual cold spell in February 2012 (Schwemmer et al. 2014). Three of the tagged males were observed foraging several weeks after tagging and their behavior and body condition appeared normal. Four of the tagged shelducks were observed after the end of the tracking period and loss of the devices was confirmed (Table 1). We caught more males than females, partly because of the better ratio between the weight of the device and the larger body mass of the males, and partly to avoid females abandon their nest.

Individuals for which we obtained tracking data for more than three months (Table 1) were included in the further analyses. This resulted in the exclusion of two individuals (Table 1).

Classification of behavioral stages

We identified significant changes in the movements of the shelducks using behavioral change point analysis (BCPA; Gurarie et al. 2016). We used the package bcpa version 1.1, implemented in R (R Core Team 2019). BCPA identifies the most likely points in an animal’s track where the movement patterns changed (‘change points’, CPs). The analysis was carried out in two steps: first, persistence velocity (Vp) was calculated for each time step by multiplying the speed of the animal with the cosine of its heading, such that Vp described the displacement of the animal and the tortuosity of its path; second, a user-selected time span (‘window’) moved over estimated values of Vp, its variance, and an autocorrelation parameter, and a CP was set at significant changes of these values. The output file included information on which of the three parameters changed significantly. Daily locations were averaged prior to the analyses because we were interested in broad-scale changes in movement behavior, rather than diurnal patterns (Gurarie et al. 2016). We chose a window size of 40 days to be sufficiently larger than potential changes in the animal’s annual cycle (e.g., incubation time, flightless time during molt; Gurarie 2013; Hoogerheide and Hoogerheide 1958; Patterson 1982). We used the flat output of the analysis and a threshold of 4, i.e., CPs detected by at least four windows were considered (Garstang et al. 2014; Gurarie et al. 2016).

Based on the BCPA results, we assigned movement behaviors to each tracking day: sedentary (≥ 10 days in an area with standardized Vp persistently close to zero), and exploration (standardized Vp moderate, standardized variance high; McEvoy et al. 2015). Exploration thus described a phase during which an individual moved through the landscape, mostly off the mainland, and visited several areas for only a few hours or days, resulting in a high variance of Vp. An example of the assignments can be found in the appendix (Fig. A1 in supplementary material).

The molting period was assigned following Cimiotti et al. (2022): The molting shelducks stayed near the tidal channels during the day at low tide for 24–35 days. During this time, they also did not cross any barriers such as dikes or travel long distances in a short time that would have exceeded the duck’s swimming speed and tidal flow.

Statistical analyses

We tested if different habitat types were explored with more effort. We first classified habitat types (offshore waters, tidal flats, sandbars, foreshore areas, inland wetlands, and inland) using aerial photographs in QGIS (QGIS Development Team 2020). Second, we added the habitat information to all available individual locations. Third, we calculated generalized linear models (GLM) from the package lme4 (Bates et al. 2015). The number of counts per individual in exploratory and the sedentary behavior were included as a dependent variable, and the habitat type was used as exploratory variable. We chose a quasi-binomial error distribution to account for over-dispersion (Korner-Nievergelt et al. 2015; Zuur et al. 2010).

Generalized additive mixed models (GAMMs) were implemented using the package gamm4 (Wood and Scheipl 2016). Circannual patterns were investigated by applying a binomial GAMM with behavior classification per day as a dependent variable and time as an exploratory variable. To display the exploratory effort depending on the flightless period during molt, we centered the time for each individual to the first day of molt. We included time as a smoother using cubic regression splines to account for the nonlinear effect. Individuals were included as the random factor and only time spans for which we had obtained data from at least five individuals were included (Zuur et al. 2014). The number of knots was set to 50 (Ruppert et al. 2003, in Zuur et al. 2014).

We defined the individual molting site by drawing a minimum convex polygon around locations during molt with a buffer of 200-m buffer. This value was chosen because it was about the median distance of successive data points in all GPS-data (172.9 m). Each explored and sedentary area was defined in the same manner.

We determined if the molting area (Fig. 1) was particularly explored in the period before (30 days) compared with the period after molting (30 days) using a GLM. The number of days with exploratory behavior in the molting area was included as a dependent variable, with time as the binary explanatory variable (30 days before vs. after molt). We chose a quasi-binomial error distribution to account for over-dispersion (Korner-Nievergelt et al. 2015; Zuur et al. 2009).

We studied the exploration effort in relation to the distance from sedentary-used areas. For this, we modeled the frequency of exploration in relation to the distance from the edges of sedentary-used areas using a GLMM from the package mgcv (Wood 2011) with a quasi-binomial error distribution to account for overdispersion (Korner-Nievergelt et al. 2015; Zuur et al. 2009). The number of times that an individual was located exploring within a 1-km interval from a sedentary area was used as the dependent variable, and the distance to the closest sedentary area was the exploratory variable. Individuals were included as random factors.

For all the model analyses, we tested violations of the model assumptions visually based on appropriate residual plots (Korner-Nievergelt et al. 2015; Zuur et al. 2009). We tested for over-dispersion by comparing the residual deviance with the residual degrees of freedom (Korner-Nievergelt et al. 2015).

We used non-parametric Kruskal–Wallis tests to test for differences (1) between the distance traveled between sedentary areas, (2) in time spent exploring the molting area compared with other sedentary areas, and (3) in the number of times the molting area was visited compared with other sedentary areas.

Exploratory movements were classified as long-distance exploration movements when they were performed > 45 km from a sedentary area. The value was chosen because it was half of the maximal distance detected between successively used sedentary areas.

Results

Frequency of exploration movements

We detected exploratory behavior on 22% of all days throughout the tracking periods for all nine shelducks (median). The values varied between individuals (range = 6–39%, n = 9; compare Fig. 2). Two shelducks did not molt during the tracking period, presumably due to a lack of time between rising chicks and winter (Cimiotti et al. 2022). Those individuals explored for more days (33% and 39%, respectively) than individuals that did molt (median = 20%, range = 6–30%, n = 7, only data outside of molt included). Differences among individuals were large even when only months for which we had obtained data from all individuals were considered (compare Table 1 and Fig. 2: Jul–Oct; only individuals that molted during the annual cycle were considered; median = 16%, range 6–39%).

Explored habitats

The most frequently visited habitat type was tidal flats (Fig. 3). The sites explored were of the same habitat type as the sedentary sites (Fig. 3). The only difference was found for inland wetlands, which were visited more frequently during sedentary phases (GLM, p = 0.03; Fig. 3).

Fig. 3
figure 3

Habitat types visited by individuals during sedentary and exploratory phases

Full size image

Spatial scale of exploration movements

Exploration movements were mostly near areas that were also used for sedentary purposes during the study period (Fig. 4; GLMM: R2 = 0.59; deviance explained = 61%, p < 0.001). Almost 25% of locations visited during exploration phases were < 300 m from sedentary locations and about 50% were within 3000 m. Long-distance exploration movements, i.e., ≥ 45 km from sedentary-used areas, accounted for 0.4% of the locations (Fig. 4), and only two shelducks performed such long-distance exploration trips. During one such trip, the individual flew to a salt marsh 63 km north of all sedentary-used areas for 2 days, before returning to the sedentary-used area where it had been throughout the breeding season. The other long-distance exploration movement was performed between molting and winter, when the individual traveled 165 km to the north, left the Wadden Sea, visited the Nissum Fjord, Denmark (56.3°N, 9.2° E), and returned to its previous sedentary area 8 days later. The furthest distance from the sedentary areas explored by most individuals was much closer (median = 23 km, interquartile range (IQR) = 20–31 km). The inter-individual variation was high (goodness of fit, χ2 = 489, df = 8, p < 0.001), with a range of 4–165 km. Map examples of some individual tracks can be found in the online appendix (Fig. A2 in supplementay material).

Fig. 4
figure 4

Exploration effort in relation to distance from sedentary-used areas. Gray dots represent individual proportion explored offset not explored within each kilometer (n = 9 birds). Black line represents the fit from a GLMM with quasi-binomial error distribution and 95% confidence interval (dashed line)

Full size image

Exploration patterns of sedentary areas

Shelducks explored about half 55% (n = 44) of the sedentary areas. Some of them were visited during exploration periods long after they were used sedentarily, but many of the sedentary areas (48%) were explored before they were occupied. Shelducks usually visited sedentary areas once (median 1, IQR = 1–3 times). On average, the last visit was performed 10 days prior to the sedentary usage (median, IQR = 1–55 days). Each exploratory visit to a subsequently used sedentary area lasted a median of 2 days (IQR = 1–6 days).

When shelducks switched between sedentary areas, they usually did so in an exploratory manner: rather than flying directly to the subsequently used sedentary area, they visited many places along the way, resulting in a tortuous path (examples see appendix, Fig. A2 in supplementay material). It usually took about 7 days to travel from one sedentary area to the next (median, IQR = 3–13 days). The distance was 4.5 times greater than necessary for a direct flight (median, IQR = 2.1–9.6, median distance = 109 km, IQR = 58–240 km vs. a possible direct distance of 28 km, median, IQR = 10–46 km). These patterns did not differ significantly between individuals (Kruskal–Wallis test, temporal: H = 12.6, df = 8, p = 0.13; spatial: H = 12.2, df = 8, p = 0.14).

Exploration of molting sites

The exploratory effort varied throughout the tracking period (GAMM; edf = 10.8; χ2 = 44.95, p < 0.001, n = 7; Fig. 5, only individuals that underwent molt including flightlessness). During the period of 30 days before the start of molting, the probability of exploring was twice as high as during the rest of the tracking period (odds ratio = 2.04, confidence interval = 1.13–3.69). During that time, most locations (median proportion = 1, IQR = 0.74–1) were detected in the molting area (Fig. 2). The molting area was explored more intensively during the 30 days before the onset of molt compared with the 30 days after molt was finished (GLM, p = 0.02; Fig. 6). Combining all locations, shelducks spent a median of 59 days (IQR = 55–67 days) per year in the molting area.

Fig. 5
figure 5

Seasonal exploration patterns of shelducks that underwent molt including flightlessness (n = 7). Black line represents generalized additive mixed model fit, gray area represents 95% confidence interval, red line indicates the 24–32 days lasting flightless period during molt

Full size image
Fig. 6
figure 6

Difference in exploration effort in the molting area 30 days prior to molting compared with 30 days after finishing molting (n = 7). Gray dots: fitted values, black vertical bars: 95% confidence intervals from a GLM with quasi-binomial error distribution (p = 0.02), boxplots: distribution of raw data (Tukey style), EX exploration behavior

Full size image

Individual molting places were not explored for significantly longer before use (median = 3 days, IQR = 0–10 days) compared with other sedentary areas in the molting area (median = 1 day, IQR = 0–8 days; n = 7; Kruskal–Wallis test, H = 9.37, df = 12, p = 0.67), and were not visited more often (for both: median = 1 time, IQR = 0–3 times; Kruskal–Wallis test, H = 10.6, df = 6, p = 0.10).

Two birds (M3 and M4) were tracked during consecutive molting seasons (Table 1), and both molted at places, 15 and 50 km respectively from the ones used in the previous year. One of the two birds had visited the second molting place five times (9 days) during the previous year. The other bird had visited the second molting place for only about 2 h at the day before the onset of molting, but not at all during the preceding tracking period of > 1 year.

Discussion

Our results showed that shelducks explored extensively, with individual birds using about a quarter of their tracked periods to explore. These findings suggest that birds in dynamic habitats may spend large amounts of time exploring their surroundings.

Our results allow us to interpret the purpose of exploration movements to some extent. Most of the exploration trips in our study were performed by roaming relatively close to intensively used areas, and more-distant exploration movements (> 45 km) were rare. By exploring in small scales, individuals may obtain ongoing information about local environmental changes and habitat cues, such as food, predators, and competing birds (Piper 2011). This information can allow individuals to switch quickly to suitable places if the environmental conditions deteriorate or, in turn, to move to locations with better habitat conditions. We found that sites during exploration phases were mostly located in habitat types that were also used during the sedentary phases. Consequently, tidal flats were most frequently explored compared to other habitat types. This suggests that finding good places to forage may be of high priority.

However, we detected that the probability of exploring was twice as high before molt compared with after molting. This could be explained by several, not mutually exclusive hypotheses. (1) Birds may select a place for the flightless period based on their high vulnerability to predators and downgrading habitat conditions (Dircksen 1968). This hypothesis was supported by our finding that the molting area was explored thoroughly before molt. (2) Alternatively, shelducks may explore intensively before molting to find suitable feeding places for the pre- and post-molting periods. Indeed, we found that shelducks that underwent molting spent a median of 59 days in the molting area, even though the duration of flightlessness was only 25–32 days (Hoogerheide and Hoogerheide 1958). (3) The high levels of exploration coincided with a period when shelducks were expected to feed more, given that hyperphagia and gain of mass before the onset of molt have been detected in numerous geese and duck species (Fondell et al. 2013; Fox et al. 1998; Portugal et al. 2007). Shelducks loose body mass during the molting period (Kempf 1993; Oelke 1969; Patterson 1982), suggesting that they aim to accumulate body reserves before molting.

Long-distance exploration trips involve higher costs compared with small-scale exploration, and the purposes may be different. Long-distance exploration trips may allow individuals to adapt to broad changes in environmental conditions, or to discover better quality areas. Birds can remember specific what/where/when events precisely, sometimes for years (Mayer et al. 2013; Mettke-Hofmann 2017; Roth et al. 2012). For example, adults of some bird species explored for possible breeding sites during or after the breeding period (Arlt and Pärt 2008; Fijn et al. 2014; Ponchon et al. 2015), suggesting that they could remember the explored habitat conditions for longer periods and return later when needed. Our finding that distant areas are explored is consistent with these findings. We suggest that individuals explore distant areas intentionally to enable switching to them. However, we recorded relatively few long-distance exploration movements, suggesting that individuals have limited information to resort to or that they have enough information but their current location is good enough not to leave it.

The exploration effort indicated in our study might have been over- or underestimated due to methodological limitations. Past studies showed that tortuous paths could be associated with exploration movements and foraging movements (e.g., Zhang et al. 2015). To exclude the effect of greater feeding efforts, we assigned the movement phases (exploratory vs. sedentary) based on mean daily locations (Gurarie et al. 2016; see above). However, individuals might leave feeding areas more frequently during periods of high food intake, e.g., because prey density decreases (Anders et al. 2009; Rosa et al. 2008), leading us to overemphasize the amount of exploration. In contrast, the exploration effort might have been underestimated: we obtained geographical data for some of the tracked shelducks at low temporal resolution, e.g., once in 2 h (compare Table 1), possibly leading to missing some short-lasting exploration trips. Individuals briefly visited sedentary-used places in advance, by either leaving their previously used sedentary area and returning to it or switching to a more sedentary behavior in the newly explored area. Other studies found similar results, showing that birds usually used explored habitat conditions shortly before using the areas, i.e., when the information was most valuable and reliable (Lewis et al. 2010; McEvoy et al. 2015; Ponchon et al. 2015). We speculate that shelducks, migrating from the western part of their European range toward the molting area in the Elbe estuary, exhibit similar similar exploratory behavior, and consequently visit many different sites west of the molting area in the Elbe estuary. This hypothesis is supported by Kleefstra et al. (2011), who observed that more shelducks molted in the Dutch Wadden Sea in years with more Sandhoppers (Corophium volutator), a major prey species of shelducks (Buxton and Young 1981).

The selection of sedentary-used areas is likely to be influenced by social contexts, i.e., individuals relying on conspecific social presence (Danchin et al. 2004; Helm et al. 2006). Shelducks are particularly social, e.g., in gregarious courtship groups, feeding flocks during the winter and during the molting period (Patterson 1982). Areas might thus be selected based on a combination of experiences from previous years, contemporary explorations of habitat features, and social information from flocks. The high levels of exploratory behavior found in the current study might thus evolve into collective knowledge of flocks, which might in turn provide reliable information for conspecifics.

We showed that individual shelducks frequently explored the Wadden Sea mostly on small spatial scales and sometimes over long distances. Habitat conditions may change frequently in the Wadden Sea. Exploratory movements of shelducks are likely aimed at finding sites with good habitat conditions for daily needs during different phases of the Shelduck’s annual cycle. Climate change is predicted to have a strong impact on the Wadden Sea and to change habitat conditions for shelducks. Therefore, shelducks may be forced to adapt to these changes, likely involving exploratory movements. Shelducks feed opportunistically on small macrobenthic species that live in the upper layer of the mud. They mainly predate small polychaetes and shellfish < 0.8 cm, but also eat algae, eelgrass, insects, seeds, and diatoms (Anders et al. 2009; Buxton and Young 1981; Evans et al. 1979; Oelke 1979; Thompson 1982; Viain et al. 2011). It is difficult to predict the effects of climate change on the numbers, biomass, and distribution of small macrobenthic species because of different abiotic scenarios (Dissanayake et al. 2012; Hofstede and Stock 2018), as well as the complex ecological interplay among species (Beukema et al. 2009; Buschbaum et al. 2012). The spatial distribution and biomass of small benthic prey species of shelducks is predicted to change in the coming decades. (Beukema et al. 2009; Singer et al. 2016; van Roomen et al. 2012). We speculate that the normal exploratory movements of shelducks will compensate for small-scale geographic shifts and changes in the species composition of small microbenthic animals.

However, predicted sea-level rises suggest that the tidal flats in the Wadden Sea will diminish until the end of the century or will become submerged (Dissanayake et al. 2012; Hofstede and Stock 2018). The rarity of long-distance exploratory movements may not allow shelducks to adapt to such changes. Changing environmental conditions have been shown to induce exploratory movements in other species (McEvoy et al. 2015). It would thus be interesting to see if declining habitat conditions lead to increased exploration by shelducks. We suggest that such information is crucial for estimating the ability of a species to cope with future changes in environmental conditions.