A new approach to evaluate multimodal orientation behaviour of migratory passerine birds recorded in circular orientation cages Free

Orientation cage experiments have been commonly used in studies on the mechanisms of orientation and navigation behaviour of songbird migrants (e.g. Gwinner and Wiltschko, 1978; Helbig, 1992a; Helbig, 1992b; Helbig et al., 1989; Munro et al., 1993; Åkesson, 1993; Åkesson, 2003; Wiltschko and Wiltschko, 1996; Wiltschko et al., 2002; Ritz et al., 2004; Åkesson et al., 2005), while this method has more rarely been applied in studies on species-specific migratory directions or to explore potential differences in orientation direction within populations (e.g. Rabøl, 1985; Sandberg et al., 1988; Hilgerloh, 1989; Ehnbom et al., 1993; Sandberg and Gudmundsson, 1996; Ożarowska et al., 2004; Zehtindjiev et al., 2010; Ilieva et al., 2012). In most studies, unimodal orientation may be the expected outcome. However, in several situations where for instance species or population preferences have been studied, or when compass cues were set in experimental conflict, bimodal or multi-modal orientation may be expected. For example, in one of the most well-known studies investigating the genetics of bird migration, Helbig (Helbig, 1991) applied calculation procedures assuming unimodal orientation in his experimental birds. Nevertheless, Helbig did not give a clear explanation for the hereditary mechanism of the migratory direction under study and this was not the purpose of his study at the time. In fact, the gene expression encoding this feature has still not been well studied, so researchers need to be careful when applying current statistical procedures to actual biological data because the resulting orientation responses of birds may be expected to express more than one direction.

To date there is only one example of multimodal orientation behaviour commonly accepted, i.e. axial orientation behaviour (Batschelet, 1981). Despite this, a number of researchers have already noticed that the behaviour expressed by birds tested in orientation cages can be complex and generate multiple orientation preferences deviating less than 180 deg from each other (Holmquist and Sandberg, 1991; Rabøl, 1993; Busse, 1995; Busse and Trocińska, 1999). In his monograph on the ecology of bird migration, Newton (Newton, 2008) also points to this complexity. Moreover, there is growing evidence for more complex migratory routes (within the same population) revealed by tracking studies and of new evidence for individually distinct migratory strategies (Åkesson et al., 2012; Guilford et al., 2011; Vardanis et al., 2011).

The aim of this study was to analyse the behavioural patterns of songbird migrants tested in the most common type of orientation cage, the so-called Emlen funnel (Emlen and Emlen, 1966), and to propose a statistical method to analyse the data and to identify multimodal orientation responses.

The data set consisted of N=139 orientation cage tests with nocturnal songbird migrants of four different species; no tests repeated on the same individuals were included in the analysed data set. The birds were tested at the Biological Experimental Station Kalimok, NE Bulgaria (44°00′N, 26°26′E) in autumn 2001. Studied species included long-distance migrants wintering in Africa: the great reed warbler, Acrocephalus arundinaceus (Linnaeus 1758) (N=41), the sedge warbler, Acrocephalus schoenobaenus (Linnaeus 1758) (N=45), the willow warbler, Phylloscopus trochilus (Linnaeus 1758) (N=39) and the whitethroat, Sylvia communis Latham 1787 (N=14). The experiments were conducted in accordance with the national animal welfare legislation of Republic of Bulgaria.

Standard SE European Bird Migration Network working procedures were applied (Busse, 2000), such that the bird migrants were caught with mist-nets, ringed and measured (including fat score determination). Birds captured during the day were kept together in a large cage (2.5×1×1 m) with free access to mealworms and water. Orientation experiments were performed in modified Emlen funnels (Emlen and Emlen, 1966). The experiments were initiated shortly after sunset and lasted for 60 min.

Modality refers to the number of directions/headings preferred by an individual bird tested in the orientation cage, i.e. showing no (0=‘disoriented’), one or more preferred directions.

Unimodal behaviour refers to the situation when a tested individual preferred only one direction (Fig. 1). Multimodal behaviour refers to when a tested individual preferred more than one direction (i.e. bimodal, trimodal etc.; Fig. 1), and axial behaviour when the two directions were 180 deg (±30 deg) apart. Activity was recorded during a 60 min period as the number of scratches left on the correction paper covering the walls of the cage, and tested individuals were classified as inactive when the total activity was less than 40 scratches (e.g. Helbig, 1992a; Korner-Nievergelt et al., 2002).

We used ORIANA software (version 3; http://www.kovcomp.co.uk/oriana/) to analyse the data according to conventional methods applied in circular statistics, and ORIENT software (available from the Bird Migration Research Station, University of Gdansk, Poland, sbwp.ug.edu.pl; as this program requires specific data format, please contact K.M. before use) for the modelling based on Bayesian statistics. Both methods enable the analysis of individual tests in the first stage and analysis of the group of tests in the second stage. As we used this novel modelling procedure, as opposed to the conventional one, below we give a more detailed description of the method.

In the modelling procedure, the results of an individual test (scratches counted in each of eight sectors) were given as a histogram and each mode of cage data distribution was modelled as a wrapped (circular) Gaussian distribution (Fig. 2, Table 1) with the maximum likelihood method.

The modelling procedure at the group level is shown in Fig. 3. Calculations at this level were performed following the methodology given above, except for the input data, where in place of the number of scratches by the individual bird, the directions estimated in the individual tests were used (Fig. 3A). In polar plots, the continuous curve is the angular probability distribution of the group direction (Fig. 3B,C), while bars indicate the central values of each fitted mode with length proportional to the relative weight of each direction (Fig. 3C) in the sense outlined above.

To optimise calculation efficiency, model fitting was performed in two steps: (1) initial fit to the smoothed distribution of input data – the common standard deviation (s.d.) was used as a smoothing parameter, (2) final fit to the ‘raw’ input data. The weight of each group direction in the fitted model is proportional to the number of contributing individual directions. Consequently, as at the group level there are more data than in the individual test, the number of resulting directions for the analysed group is no longer limited to four. The number of these directions is in practice limited by the BIC criterion, in which with increasing BIC values the posterior probability of the model systematically decreases, although the goodness of fit of the model increases.

According to the fit quality of the final angular probability distribution to the initial distribution of all individual directions, the researcher is able to control the resulting number of directions. At the present moment it is recommended to use about a dozen different values of standard deviation and analyse corresponding BIC values and the number of directions of the resulting model. The comparison of the results of orientation tests and the recovery distributions of passerine migrants showed that a smoothing parameter standard deviation of about 10–15 deg is adequate (A.O. and K.M., unpublished data). Still, the BIC criterion based on the total number of directions is applied as a supporting criterion. The Matlab R20011b (MathWorks Inc., Natick, MA, USA) was used to create the models and to perform the calculations. ORIENT software is available as detailed above. All data used in the present study are available at either the Bird Migration Research Station (sbwp.ug.edu.pl) or the Biological Experimental Station Kalimok (Institute of Biodiversity and Ecosystem Research, Bulgarian Academy of Sciences, Bulgaria; www.iber.bas.bg).

In total, 96% of the orientation tests were included in further analyses when applying conventional circular statistical methods (Batschelet, 1981), while in our modelling procedure 85% were included (Table 3). Six individuals (4.3%) were classified as inactive. According to the modelling procedure, in 10.8% of the tests the model quality was too low (evidence<1). These tests were not included in further analyses. According to conventional circular statistical methods, in most (94%) of the tests the birds showed a unimodal pattern of orientation, while in 6% of the tests their directional behaviour was axial (Table 3). When applying the modelling procedure to the same available data set, the observed proportion of the different patterns of bird behaviour in the cages was far more complex (Table 3, Fig. 4).

Depending on the species, either unimodal or bimodal orientation dominated, while trimodal directional behaviour (Table 3) was observed in 12% of the tests. Moreover, when analysing bimodal orientation responses, i.e. when the two directions were within 150–180 deg of each other, they were axial in 36.8% of all tests (Table 4). This proportion varied across the four species from 20% to 45%.

The final results of both calculation methods applied to single tests can be compared in Figs 5, 6, 7 and 8.

When analysing unimodal distributions, the two methods were in good agreement (Fig. 5). All tests classified as unimodal based on the modelling procedure represented unimodal distributions according to the conventional procedures (Table 5). Most axial distributions according to the conventional procedures also represented this category when the modelling procedure was applied (Fig. 6, Table 5).

For the purposes of illustration, the calculations were applied to bimodal and trimodal distributions and the results of these calculations are given in Figs 7 and 8. The assumption of unimodality was not fulfilled; still, statistically significant results are obtained based on the conventional circular statistical methods. According to our modelling procedure, bimodal (excluding axial), trimodal and four-modal distributions comprise as much as 49.2% of all tests included in the analyses, so excluding them from the data set would not be advisable.

While at the individual level the two methods, when applied to certain types of distributions, show similar results, at the group level the observed pattern is different between the two methods because of the larger number of modes in the latter (Fig. 9, Table 6).

Conventional calculation procedures applied either to a single test or to the group of tests result in one, average direction, while the modelling procedure enables, at both levels, the analyses of far more complex patterns describing in more detail the directional preferences shown in the circular cages. With the latter method the only criterion for including the test in the sample set is whether the tested bird was active or not, while the researcher does not make any preliminary assumptions as to the distribution pattern, which is mandatory in the conventional statistical methods. Consequently, we avoid the risk of unintentionally breaking basic assumptions of some statistical tests and getting false results, and finally we are not ignoring the complexity of the behaviour of birds tested in the orientation cages.

Experiments for which migratory activity has been recorded for songbirds in circular cages or analysed by ringing recoveries often show unimodal expected migratory preferences, but also differential migratory orientation in individual experiments as well as at the group level (e.g. Sandberg et al., 1988; Busse, 1987; Åkesson et al., 1996; Mouritsen, 1998; Muheim et al., 2002; Remisiewicz, 2002). In most studies, standard procedures are deployed to treat the circular cage data from individuals for further statistical analyses (e.g. Sandberg et al., 1988; Helbig et al., 1989; Åkesson, 1994; Wiltschko and Wiltschko, 1996; Muheim et al., 2002; Wiltschko et al., 2002; Ilieva et al., 2012). After initial treatment of individual test data it is often valuable to compare mean orientation between groups. The conventional procedures of a single cage test data analysis include the Rayleigh test application in the first step, as outlined elsewhere (Batschelet, 1981), to find out whether the tested individual bird was disoriented or showed directional behaviour, i.e. one migratory direction. This is a suitable approach as long as the distribution does not have more than one mode, as the Rayleigh test requires unimodality (Batschelet, 1981). Currently, the only exception to this procedure is treatment of an axially bimodal distribution with the ‘doubling the angles’ procedure, after which the Rayleigh test can be employed. However, when circular data on the individual level are neither unimodal nor axially bimodal, Zar (Zar, 1999) recommends applying the ‘Rao spacing test’ (Batschelet, 1981; Zar, 1999). The problem is that there is no conventional circular statistical tool available for testing hypotheses on multimodal distributions. Therefore, it could happen that when following the conventional procedures, a researcher will not be able to identify that the orientation cage results show a multimodal pattern. Some multimodal distributions analysed with conventional methods, not appropriately applied, will thus give significant but false results. The problem could be overcome by excluding multimodal distributions from further analyses, but then in some cases, as in our study presented here, we would exclude nearly 50% of all tests and more importantly develop false results and interpretations.

Axial orientation behaviour needs special attention as this is the only example of a multimodal distribution when a conventional approach, the Rayleigh test, can be applied after the doubling angles procedure. However, when analysing the bimodal pattern of our experiments in detail we found that the experimental birds showed axial behaviour in only 36.8% of tests. This proportion differs according to the species. Still, there could be a high proportion of bimodal tests, which should be excluded from the analyses according to the conventional statistical methods.

The results of our study show that multimodal orientation behaviour of migratory passerines tested in orientation cages dominates and is characteristic of these types of data. Moreover, modelling results show that conventional calculation procedures must not be used for multimodal distributions, with the exception of true axial distributions. We therefore propose that care must be taken when evaluating orientation cage results such that analyses of multimodality should be included in the analyses of circular data as a standard method to investigate the pattern of distributions. Thus, we strongly recommend the modelling procedure applied in this study to orientation data analyses, which enables the researcher to distinguish among distributions differing in modality and subsequently to compare the probability of these distributions and thereafter to choose in an objective way the one that is the most probable. However, one disadvantage of such an approach is the problem of distributions in which none of the analysed models is favoured (evidence<1; in our study, 15 distributions). In future studies we plan to incorporate Bayesian model averaging (BMA), which overcomes this problem by conditioning not on a single ‘best’ model but on the entire ensemble of statistical models first considered (Kass and Raftery, 1995; Hoeting et al., 1999; Raftery and Zheng, 2003). The modelling procedure also offers alternative ways to analyse different aspects of birds' behaviour. For instance, the distributions can be analysed according to the number of individuals preferring a given direction, or the precision of their choice (based on standard deviation), or their activity (i.e. number of scratches) within the final direction. At the moment there is no software available that allow statistical comparison of two circular multimodal distributions. However, as the Orient program provides several parameters for each direction (e.g. angular value, standard error), contributing to the final pattern, we can compare two distributions by comparing pairs of corresponding directions.

There may be several reasons for the multimodal behaviour of migrants tested in orientation cages. Mechanisms controlling bird migration are necessarily conservative, but are under the influence of natural selection. Many features of migration, for example level and time of migratory restlessness and migratory direction, are encoded genetically (e.g. Berthold, 1983; Berthold, 1993; Berthold, 1996; Berthold et al., 1990a; Berthold et al., 1990b; Berthold et al., 1992; Berthold and Helbig, 1992; Berthold and Pulido, 1994; Helbig, 1991; Helbig, 1992b). Furthermore, reverse migration is regularly shown at ecological barriers (Åkesson et al., 1996); all these are regulated by the biological compasses (e.g. Gwinner and Wiltschko, 1978; Gwinner, 1996; Weindler et al., 1996; Wiltschko and Wiltschko, 1996; Wiltschko and Wiltschko, 2003; Åkesson and Hedenström, 2007). All these phenotypic characteristics of the migration programme enable immature passerines to reach species- or population-specific wintering grounds during their first autumn migration. Visual observations and ringing results have further proved that birds show breeding site fidelity, but also fidelity to intermediate stop-over sites and wintering sites (e.g. Moreau, 1972; Salewski et al., 2000; King and Hutchinson, 2001). Shifts in migratory direction with progress of the season have also been observed (Gwinner and Wiltschko, 1978; Helbig et al., 1989). At certain stages of the migration, birds may show both prior and new migration route preferences when recorded in cages after range expansion of the species or corrections of previous drift during migration flights, depending on the ecological situation and stage of individual migration.

The problem was also addressed by Phillips in his study on the vagrancy of Asiatic warblers observed regularly in Europe (Phillips, 2000). He pointed out that a factor which had been widely overlooked was that any experimentally tested group of migrants exhibited a wide range of orientation directions and one could not accept the mean direction as being valid while selectively ignoring individual data points from which it was derived. He also concluded that the ‘wrong’ directions had to be derived genetically, from the birds' parents, though details of the genetic inheritance of orientation directions had not yet been worked out.

When considering these facts it seems logical that migrants should exhibit not just one migratory direction but several directions and that these preferences need to be handled statistically.

Although several studies have focused on the genetics of bird migratory behaviour (e.g. Berthold, 1983; Berthold and Helbig, 1992; Berthold and Pulido, 1994), only two have focused on the inheritance of a migratory direction. Berthold and colleagues (Berthold et al., 1990b) found that cross-breeding of individuals from migratory and sedentary populations resulted in axial behaviour of the offsping along the NE–SW axis, i.e. the characteristic migratory direction of the parents. Helbig studied the basis of migratory direction inheritance (Helbig, 1991) and found that cross-breeding of individuals from two populations showing highly different migratory directions resulted in an intermediate migratory direction of the F1 generation. Consequently, a general assumption has been that individuals tested in orientation cages could only show one migratory direction (with one exception of bimodal behaviour, i.e. axial behaviour). However, it was not the result of Helbig's study (Helbig, 1991) that explained and documented the hereditary mechanism of migration direction in such a way that could justify the common assumption of a unimodal pattern of bird behaviour. Multimodal behaviour of migratory passerines tested in the orientation cages probably shows that the mechanisms of bird compass orientation and navigation are complex and need special attention, but also that the physiology of individual birds has a strong effect on directional choices (Sandberg, 2003). There are several advantages of differential orientation such as returning to the protective sites en route, i.e. sites already known, which provided the migrant with food and/or shelter (Åkesson et al., 1996). Other examples deal with the opportunity to expand breeding and wintering ranges (e.g. the blackcap, Asiatic warblers) (Berthold and Terrill, 1988; Busse, 1992; Berthold et al., 1992; Thorup, 1998; Phillips, 2000; Newton, 2008; Kopiec and Ożarowska, 2012). In this study we propose a way to objectively handle multimodal orientation responses in birds and to perform statistical evaluations of the data. Future evaluations of new results from light loggers (geolocators) currently applied in many studies and previously collected circular cage data will be important to further evaluate the modelling approach proposed.

This study is report no. 52 of the Biological Experimental Station Kalimok. We are grateful to Zübera and Lâtif Ismail for their assistance in the field.