Several invertebrate and vertebrate species have been shown to align their body relative to the geomagnetic field. Many hypotheses have been proposed to explain the adaptive significance of magnetic body alignment outside the context of navigation. However, experimental evidence to investigate alternative hypotheses is still limited. We present a new setup to track the preferential body alignment relative to the geomagnetic field in captive animals using computer vision. We tested our method on three species of migratory songbirds and provide evidence that they align their body with the geomagnetic field. We suggest that this behaviour is involved in the underlying mechanism for compass orientation and calibration, which may occur near to sunrise and sunset periods. Our method could easily be extended to other species and used to test a large set of hypotheses to explain the mechanisms behind the magnetic body alignment and the magnetic sense in general.
Many animals are known to have a magnetic sense and to use the geomagnetic field for orientation and navigation (Wiltschko and Wiltschko, 1995). In recent years, the phenomenon of magnetic body alignment of both invertebrates and vertebrates has received increasing attention (Begall et al., 2013; Wiltschko and Wiltschko, 1995). Based on data collected from satellite images, it has been shown that cattle on pastures align their body with the N–S axis of the geomagnetic field (Begall et al., 2008). Following this first finding, a growing number of studies have been published around this topic (Begall et al., 2013, and references therein) with the most recent studies largely focusing on vertebrates (e.g. Hart et al., 2013a, 2013b; Slaby et al., 2013; Malkemper et al., 2015; Obleser et al., 2016; Čapek et al., 2017; Nováková et al., 2017; Pleskač et al., 2017). The adaptive significance of magnetic body alignment remains elusive (Begall et al., 2013), and only in a few cases is there some support for plausible mechanisms. For example, the red fox (Vulpes vulpes) has been proposed to use magnetic body alignment while hunting to better estimate distances from their prey (Červený et al., 2011). In roe deer (Capreolus capreolus), magnetic body alignment has been suggested to facilitate a more effective escape strategy in response to approaching predators (Obleser et al., 2016). In newts (Notophthalmus viridescens), magnetic body alignment is linked to the sensing of magnetic map information (Phillips et al., 2002). Honeybees align with the magnetic field to resolve ambiguous visual landmarks standardizing the visual input to the compound eyes (Frier et al., 1996). Other hypotheses, i.e. group coordination, remnant trait from a migratory ancestor, spatial orientation, compass calibration, etc., remain to be formally tested (Begall et al., 2013).
The major constraint on the experimental testing of any hypothesis on magnetic body alignment is that it can be easily confounded by other external factors (e.g. wind, slopes or landmarks), and can furthermore be masked by orientation towards other cues (e.g. celestial bodies, olfactory signals) (Begall et al., 2013). Testing magnetic alignment under controlled conditions would allow us to selectively exclude external factors and/or to manipulate cues, and thus to eliminate all of the above-mentioned limitations. However, testing of magnetic body alignment under controlled laboratory conditions is mostly lacking for vertebrates except for nest-building rodents (e.g. Malkemper et al., 2015).
The magnetic sense was described for the first time in vertebrates using songbirds as a model organism (Wiltschko, 1968). Therefore, it is surprising that birds have only rarely been included in studies of magnetic body alignment. Moreover, when birds have been used in body alignment studies, they have only been observed performing specific activities in their natural habitat, such as landing in ducks and geese (Hart et al., 2013a), foraging in corvids (Pleskač et al., 2017) and drinking in domestic chickens (Čapek et al., 2017), and not studied in the lab. In contrast, magnetic orientation has been extensively studied in migratory songbirds in modified so-called Emlen funnels (Emlen and Emlen, 1966), in the past 50 years (e.g. Wiltschko and Wiltschko, 1972; Sandberg et al., 1988, 1998; Munro and Wiltschko, 1993; Åkesson, 1994; Åkesson et al., 2001; Muheim et al., 2006; Engels et al., 2014; Chernetsov et al., 2017). However, sometimes songbirds tested in Emlen funnels show a preferred direction that does not coincide with their migratory goal (e.g. Åkesson, 1993, 1994; Sandberg, 2003; Muheim et al., 2017). When orientation preferences different from those expected have been observed, and which do not coincide with the N–S geomagnetic field axis, this has been suggested to be a special case of magnetic body alignment (Begall et al., 2013). In this special case, birds do not align along the migratory direction or along the geomagnetic field axis, but they still rely on the magnetic field cues to select their body alignment. This suggestion is not supported by any hypothesis, but more importantly it has two methodological shortcomings. First, Emlen funnel experiments with songbirds are only performed during the migratory restlessness period of the target species (e.g. sunset for nocturnal migrants) and are often limited to just 1 h (Bianco et al., 2016; cf. Åkesson and Sandberg, 1994). Hence, it is impossible to disentangle whether the birds' body alignment during an Emlen funnel experiment is a result of a goal-oriented behaviour due to the bird's ‘migratory’ status or whether body alignment in songbirds also occurs during other periods of the day or while performing other activities. Second, the Emlen funnel has been shown to be an appropriate method for orientation (Emlen and Emlen, 1966), but not for body alignment studies (Bianco et al., 2016).
Here, we investigated the body alignment in three species of migratory songbirds: chiffchaff [Phylloscopus collybita (Vieillot 1817)], dunnock [Prunella modularis (Linnaeus 1758)] and European robin [Erithacus rubecula (Linnaeus 1758)]. We monitored their body alignment over 24 h in a controlled laboratory setting, removing all potential directional cues except the geomagnetic field. We demonstrate by using a computer vision tracking algorithm that we could gain detailed information on body alignment in migratory songbirds relative to the local geomagnetic field over the course of a full day. We propose that our experimental approach has the potential to improve our understanding of the magnetic body alignment behaviour in both migratory and sedentary animals without disturbance from human observers, and may in addition help to shed light on animals' perception of magnetic information.
MATERIALS AND METHODS
Permission was given by the Malmö/Lund Ethical Committee for Scientific work on animals (no. M33-13) for experiments, the Swedish Board of Agriculture for housing facilities (Dnr 5.8.18-12719/2017), and the Swedish Nature Protection Agency and the Swedish Ringing Centre (no. 440) for capture of birds.
Experimental birds and testing facility
First-year migratory songbirds (n=24) were captured with mist-nets near Stensoffa Ecological Field Station (55°41′N 13°26′E) in southwestern Sweden in October 2017. Birds were kept indoors in individual cages for a few days until they were moved to the testing facility. Birds of the same species (n=8 each) were put into groups of four individuals and placed in individual circular cages inside one of six identical experimental houses built of non-magnetic material (Ilieva et al., 2016). Each experimental house was equipped with a network camera that recorded the four cages from above and a constantly recording magnetometer (Ilieva et al., 2018). Birds were introduced into the cages in the early afternoon to familiarise themselves with the new environment, and there they were provided with food and water ad libitum. Video-recordings started at 00:00 h local time (02:00 h UTC) and lasted 24 h. The experimental houses exposed the birds to natural light conditions (through the semi-transparent roof) without providing visual information (i.e. environmental landmarks and the position of celestial bodies) or any wind-driven olfactory cue. Furthermore, to avoid the resting position of the bird biasing its body alignment, we equipped each cage with a 3D-printed circular perch (9 cm diameter; Fig. 1). The circular perch was positioned at equal distance from the bottom and the top of the cage, in order to avoid bias due to the vicinity of the water and food (provided on the floor of the cage) or exposing the bird to any visual landmark along the horizon (Ilieva et al., 2016).
Measurement of body alignment
The body alignment of the bird was determined from the recorded videos using an ad hoc script developed in Python version 3.4 (www.python.org) using the Open Source Computer Vision Library (OpenCV) version 3.4 (http://opencv.org). We relied on the background subtraction algorithm based on the mixture of Gaussian models MOG2 (Zivkovic, 2004; Zivkovic and Van Der Heijden, 2006) for segmenting each frame between foreground (i.e. the bird) and background pixels (Ilieva et al., 2018). However, although MOG2 is an efficient motion-tracking algorithm, it fails when there is little or no motion for longer periods in the scene (Bouwmans et al., 2008). As the aim of this study was to also measure the bird's body alignment during their resting periods, we implemented an additional local model for the background. Our model was localised where the bird stopped and became stationary (e.g. Fig. 1A). The last position of the bird before stopping was determined by measuring its speed and the number of foreground pixels of the MOG2 segmentation. When the speed was close to zero for several frames and the number of foreground pixels (i.e. the number of pixels representing the size of the bird) of the MOG2 segmentation become lower than a given threshold, i.e. when the stationary foreground was becoming part of the background model, we updated the background model by cutting the area around the bird and pasting the same area from a previous frame where the bird was in a different position (Fig. S1; Fig. 1B). Our algorithm implementation dynamically switched from the MOG2 to the local background model to find the best foreground model at each frame (see example in Movie 1). The body axis of the bird was then simply determined by the direction of the major axis of the ellipse fitting the foreground pixels (Fig. 1C; Bianco et al., 2016). The body axis was measured relative to the geomagnetic field between 0 deg (magnetic North) and 180 deg (magnetic South) using the information retrieved by digital magnetometers (Ilieva et al., 2018; Bianco et al., 2019). Inspection of the geomagnetic parameters recorded during our experiment showed that no correction to the measurement of the body axis was necessary (i.e. only minor variations of magnetic declination occurred during the day), and that it did not occur any large temporal fluctuations that could affect the behaviour of the birds (Bianco et al., 2019; see also Hart et al., 2013b).
We focused on body axis alignment and not directionality (i.e. heading); hence, we treated the data with the procedure of doubling the angles (bimodal or axial; Batschelet, 1981). Moreover, as animals tend to align either parallel or perpendicular to the magnetic axis (Begall et al., 2013), we also used the method of double-doubling the angles (quadrimodal or bi-axial) and provided the statistics for the case with the larger concentration parameters (Batschelet, 1981).
We determined the bird's body position and its body axis every second (Fig. 1D). For our analysis, we selected only the body axis measurements that complied with two criteria: (1) the bird was sitting on the perch (i.e. its position was within a fixed distance from the centre of perch) (Fig. 1E), and (2) the bird was in the same location for at least 1 min, that is it was in resting mode (Fig. 1F). Successively, we averaged the body axis angles in 20 min intervals and grouped the results in 6 h periods or for the full day (raw data for all birds are reported in Fig. S2). Averaging of the body axis over relatively long intervals was used to avoid autocorrelation in the data caused by periods when the birds kept the same body position for a prolonged time during extended resting or sleeping. The mean axis for single birds in the 6 h periods was only calculated when at least five measurements were available (Fig. S2).
Circular statistics analysis was performed with R software version 3.5.0 (http://www.R-project.org/) and circular package version 0.4-93 (https://CRAN.R-project.org/package=circular). Because our laboratory conditions did not allow a large sample size, we pooled all measurements of individual birds (Fig. 1, Table 1). Pooling data with comparable individual sample size is commonly used in body alignment studies (Hart et al., 2013b). However, first we inspected that the individuals mean showed the same pattern of directionality across time intervals and for all three species as the pooled data (Table 1, Fig. 2). Furthermore, we provide individual means and pooled data statistics in Table 1 for comparison purpose. An additional reason to pool the data was that in the period 00:00–06:00 h the two nocturnal migrant species (chiffchaff and European robin) were very active and some individuals spent little or no time resting on the perches (Fig. S2) with consequently lower individual replicates than the sample size (Table 1).
RESULTS AND DISCUSSION
The tested songbird species explored a wide range of body alignment directions over the course of 1 day (e.g. Fig. 1D). However, we focused our work on periods when the birds were sitting on the circular perches (Fig. 1E). The position of the bird on the perch clearly affected its preferred direction; that is, the bird's sitting location on the perch predicted its body alignment direction (e.g. Fig. 1E). However, when the bird was in resting mode, it showed a preferred resting position, and thus non-random body alignment, which was obvious in all three species (e.g. Fig. 1F). We studied birds during the resting period, and included three species enabling us to record alignment behaviour data throughout the day in both diurnal and nocturnal migrants. However, it would be easy to move the focus of the study to body alignment during other activities (e.g. foraging, drinking, etc.), as has already been done in previous field studies (e.g. Pleskač et al., 2017; Čapek et al., 2017).
All three songbird species aligned the body with the N–S geomagnetic field axis during the full experimental day (Fig. 2, Table 1). A significant body alignment with the N–S magnetic axis mainly occurred in the last part of the day (18:00–00:00 h; Fig. 2). The European robin, which migrates predominantly at night, also aligned its body axis with the magnetic field during the night and the first part of the day (00:00–06:00 h; Fig. 2). We note that the evening and morning alignments may suggest a possible involvement of magnetic body alignment in compass calibrations (Čapek et al., 2017). Birds possess the ability to use a multitude of compasses and it has been proposed that their coordination and calibration occur around sunrise/sunset (Åkesson et al., 1996; Cochran et al., 2004; Muheim et al., 2006; cf. Åkesson et al., 2014). During sunrise and sunset, the sun is clearly visible at the horizon and the polarization of the skylight is also more evident under overcast skies with the highest degree of polarization near the horizon (Hegedüs et al., 2007). At sunset and thereafter, geomagnetic parameters also become more stable (Skiles, 1985), and stars and celestial bodies again become visible (Åkesson et al., 1996). It is thus plausible that calibration of the different compasses should occur around this time (Sjöberg and Muheim, 2016).
Our results do not primarily support the hypothesis of magnetic body alignment as an escape strategy (Begall et al., 2013; Obleser et al., 2016). Our tested songbirds were mostly randomly aligned during the daytime (Fig. 2, Table 1), i.e. at a time when they are normally at higher risk of predation. However, this hypothesis could be further tested with our method in an ad hoc design in which birds can be exposed to artificial predator attacks (e.g. Kullberg et al., 1996). In a high-risk environment for predation, birds should align more consistently along a preferred axis without much inter-individual differences (e.g. Obleser et al., 2016). In this respect, species-specific behaviour could also be expected in which more social species may exhibit a stronger response. There is no reason, however, to expect any collective behaviour in songbirds that mainly migrate alone. We furthermore see the possibility of using the same laboratory approach in other taxa to investigate whether alignment behaviour is affected by, for example, group density (Slaby et al., 2013).
Interestingly, only the two nocturnal migrant species (chiffchaff and robin) exhibited a bi-axial response in the evening hours (18:00–00:00 h; Fig. 2) when these species initiate migration; whereas the diurnal species (dunnock) in the same hours showed an axial response (Fig. 2) pointing to possible different mechanisms involved in body alignment in nocturnal versus diurnal migrants. Furthermore, the axial response of pooled data is much weaker than the bi-axial response exhibited by chiffchaffs and robins (Fig. 2). A bi-axial orientation is not uncommon in body alignment studies across different taxa (Begall et al., 2013). In some organisms, a bi-axial orientation appears to be spontaneous and in others it appears to be one component of a goal-oriented response, e.g. honeybees (Martin and Lindauer, 1977), crayfish (Landler et al., 2019), flies and mice (Painter et al., 2013). A possible explanation of magnetic bi-axial orientation could be the presence of quadrimodal components of the complex 3D patterns generated by the light-based radical-pair magnetoreception mechanism (Hore and Mouritsen, 2016). In circumstances in which animals are unable to orient in a particular direction, they could spontaneously align themselves with the horizontal components of the magnetic input. However, this double-alignment response makes it more difficult to design experiments where the polarity of the geomagnetic field (i.e. magnetic North) is shifted with the use of a 3D coils array (e.g. Lohmann, 1991; Ilieva et al., 2016; Chernetsov et al., 2017). Such experiments usually shift the polarity by 90 deg (Sandberg et al., 1988; Åkesson, 1993, 1994; Sjöberg and Muheim, 2016, and references therein), and therefore it would be impossible to verify the effectiveness of a magnetic polarity shift in a bi-axial response. Hence, the magnetic field should be shifted by, for example, 45 deg instead and potentially provide weaker statistical power as a result of the scatter that is commonly found in magnetically shifted treatments.
For the diurnally migrating dunnock only, we found that during the expected period of migratory activity 06:00–12:00 h (Ilieva et al., 2018), the body was aligned along the expected SW migratory direction (Fig. 2). During other times of the day, dunnocks and the other two nocturnal migrant species (chiffchaffs, robins) did not align their body with their expected migratory direction. Hence, we suggest that magnetic body alignment studies should not be compared with results from Emlen funnel experiments, but they should be used as complementary tools to study the magnetic responses and the sense itself in birds and other animal taxa. Such studies could, for example, include investigations of the involvement of the radical-pair mechanism (see also above; Hore and Mouritsen, 2016), or effects of artificial human-generated oscillatory radio-frequencies (Engels et al., 2014).
The relatively short mean vector lengths in Table 1 confirm the findings of previous studies that the body alignment is a subtle phenomenon and can possibly be masked by other behavioural or external factors. Extreme care should be used in data handling and different statistical methods could be suitable for addressing specific questions. Finally, even in a laboratory setting it might be difficult to control for all external cues. Infra-sound and other possible sources of disturbance should be considered when interpreting the data.
In conclusion, our novel laboratory approach, with controlled environmental conditions and automatic body alignment tracking, can be easily extended to other vertebrate and invertebrate species without any user bias in data analysis. With a robust and objective method to record and evaluate body alignment in the laboratory, we see high potential to explore this method in different experimental situations investigating, for example, magnetic field perception. Ultimately, the possibility of combining computer vision tracking and cue conflict manipulations will allow a deeper understanding of the underlying mechanisms of magnetic body alignment and its ecological and evolutionary significance.
We are grateful to Christoffer Sjöholm for assistance during fieldwork. We thank John Phillips and one anonymous reviewer for their constructive comments on a previous version of this manuscript.
Conceptualization: G.B., M.I., S.Å.; Methodology: G.B., R.C.K., M.I., S.Å.; Software: G.B., R.C.K.; Validation: G.B., R.C.K., S.Å.; Formal analysis: G.B.; Investigation: G.B., M.I., S.Å.; Resources: S.Å.; Data curation: G.B.; Writing - original draft: G.B., R.C.K., M.I., S.Å.; Writing - review & editing: G.B., R.C.K., M.I., S.Å.; Visualization: G.B., R.C.K.; Supervision: S.Å.; Project administration: S.Å.; Funding acquisition: M.I., S.Å.
The project received financial support from the Swedish Research Council (Vetenskapsrådet) to S.Å. (621-2013-4361 and 2016-05342) and the Centre for Animal Movement Research (CAnMove) financed by a Linnaeus grant (349-2007-8690) from the Swedish Research Council (Vetenskapsrådet) and Lund University (Lunds Universitet). Financial support to infrastructure was received from the Royal Physiographic Society in Lund (Kungliga Fysiografiska Sällskapet i Lund) to M.I.
The authors declare no competing or financial interests.