Wing morphology, flight type and migration distance predict accumulated fuel load in birds

Millions of birds migrate each year, flying over vast distances of up to tens of thousands of kilometres to exploit seasonally variable resources (Berthold, 2001; Dingle, 2014; Somveille et al., 2015). The longest migratory flights recorded so far in birds include the approximately 11,000 km long, non-stop trans-Pacific commute of the bar-tailed godwit (Limosa lapponica baueri) (Gill et al., 2009) and the more than 90,000 km travelled annually by Arctic terns (Sterna paradisaea) from the Arctic to the Antarctic and back (Egevang et al., 2010; Fijn et al., 2013). During such strenuous flights, strong natural selection operates on the physiology, morphology and behaviour of migrants (Newton, 2007). The intensity of selection during migration is manifested by the high mortality rates experienced during this period, which are often higher than during any other part of the annual cycle (Guillemain et al., 2010; Owen and Black, 1989).

In order to deal with the exceptionally high energetic demands of sustained, high-intensity migratory flight, birds accumulate fuel prior to departure as well as at stopover sites en route to their wintering and/or breeding grounds (Bairlein, 2003; Lindström, 1991). This behaviour is called pre-migratory fattening, as birds consume food in excess (hyperphagia) and gain weight quickly. Daily gain in fat mass as a percentage of lean body mass averages 4–5% in different bird species (Lindström, 1991). Some individuals are capable of doubling their body mass within a short period of 1–3 weeks prior to departure (Biebach, 1996; Hedenström and Alerstam, 1992; Newton, 2007; Odum and Connell, 1956). Fuel reserves are especially important and reach maximum levels when birds embark on long spells of non-stop flight, usually over ecological barriers such as seas and deserts that are devoid of feeding opportunities (Bairlein, 2003; Berthold, 2001; Odum and Connell, 1956).

The migratory fuel of birds comprises mostly lipids (Jenni and Jenni-Eiermann, 1998). Lipids provide seven-to-nine times more energy per unit mass than alternative oxidative fuel sources, including proteins and carbohydrates; thus, lipids are both rich in energy and economic in terms of transport costs (McWilliams et al., 2004; Newton, 2007). In addition, as fat is deposited subcutaneously, it augments thermoregulation and buoyancy, and imparts mechanical protection (Lind et al., 1999; McWilliams et al., 2004; Witter and Cuthill, 1993). Birds were traditionally considered exceptional among vertebrates in their ability to fuel high-intensity, sustained muscle work with fatty acids that are transported from extra-muscular adipose tissues directly to working muscles by the circulatory system (Jenni-Eiermann et al., 2002; McWilliams et al., 2004). Migratory bats show similar convergent adaptations, as fat might provide the exclusive fuel to muscles during endurance flights (McGuire and Guglielmo, 2009). To achieve this, lipids in birds, as well as in volant mammals, need to be mobilised, transported and oxidised at much higher rates than the highest rates ever measured in non-volant mammals (Jenni-Eiermann et al., 2002). A switch to lipid-based metabolism was probably associated with the evolution of flight, which implies the need for reduced weight of stored fuel and the demand for continuous energy production to support muscles during endurance flights.

Besides these benefits, there is a wide variety of disadvantages to fuel accumulation. First, fuel accumulation is itself costly because of altered behavioural patterns that increase predation risk (Metcalfe and Furness, 1984). Second, an increase in body mass leads to higher metabolic rates, while an increase in wing loading results in elevated costs of lift, more energy consumed per unit distance travelled, decreased manoeuvrability, lower angle of ascent and higher risk of injury and predation (Hedenström, 1992, 2010; Pennycuick, 1975, 2008; Witter and Cuthill, 1993). Third, fuel deposited subcutaneously increases the projected frontal area, which has aerodynamic implications including increased body drag, decreased lift-to-drag ratio and elevated power required for flight (Hedenström and Alerstam, 1997, 1992; Hedenström, 2010; Pennycuick, 1975). As a consequence, increasing fuel load has diminishing returns and the size of the fuel load depends on a number of anatomical attributes. For instance, with increasing lean body mass for a given wing area, the power margin (i.e. the difference between the maximum power producible by flight muscles and the power required for flight) decreases (Hedenström and Alerstam, 1997). As a consequence, larger birds are more constrained by their narrower power margin and hence are expected to have smaller fuel loads relative to their lean body mass than smaller species (Hedenström and Alerstam, 1997, 1992). Similarly, aerodynamic attributes define costs and benefits of fuel transportation and might define the size of optimal fuel stores.

Migratory birds exhibit a range of morphological, physiological and behavioural adaptations to optimise their flight (e.g. Berthold, 2001; Dingle, 2014; Hedenström, 2008; Norberg, 1990; Pennycuick, 1998; Rayner, 1988, 1990; Vágási et al., 2016; Vincze, 2016). Two optimisation strategies are paramount, maximising time or energy efficiency (Lindström and Alerstam, 1992). High wing loading (i.e. weight per unit wing area) enables fast flight speed and helps minimise the time of migration, while high wing aspect ratio (i.e. squared wingspan divided by wing area, an index reflecting wing shape with larger values indicating a long and slender wing) and low wing loading ensure low energy expenditure by reducing flight power, but increasing the time of travel (Norberg, 1990). Wing morphology in particular is strongly related to flight type; high wing aspect ratio and low wing loading are often associated with soaring, while high wing loading is correlated with flapping flight (Ricklefs, 1996). Of all types, flapping flight is energetically the most demanding, while soaring, gliding and alternate flapping–gliding require lower power output (Ricklefs, 1996). It was shown that wing aspect ratio increases and wing loading decreases with increasing migration distance in birds (Vágási et al., 2016), indicating that long-distance migration co-evolves with wing traits to ensure energy-efficient flight. Although wing design and flight type determine the power required for flight, how these factors relate to energy storage prior to migration remains poorly understood.

Fuel store size must be carefully planned to safeguard premature energy depletion during migratory flight, as well as to reduce the costs associated with surplus weight. Empirical data on how morphological, behavioural and physiological factors explain interspecific differences in fuel accumulation are limited, however, probably because of the difficulties inherent in measuring fuel load, especially in a comparable manner across species with diverse fuel-accumulation strategies (Krementz and Pendleton, 1990). Filling this gap in our knowledge of the fuelling capacity of migratory birds has broad implications from both ecological and evolutionary perspectives (Bairlein and Hüppop, 2004).

In this study, we quantified fuel stores across species using data from the literature and checked the validity of these data using individual body condition indices and fat scores collected in the field during migration. Using species-specific fuel store data, we investigated how fuel load relates to phylogeny, migration distance, wing morphology and flight type. We predicted the use of larger relative fuel stores in species that: (1) travel longer distances during migration, irrespective of phylogeny; (2) have smaller body masses (i.e. wider power margin); (3) have wing architecture that is less economic in terms of energetic demand (i.e. low wing aspect ratio and high wing loading); and (4) utilise continuous flapping flight, in contrast to species with intermittent flight (flapping and soaring, flapping and gliding, and passerine-type flight; sensu Bruderer et al., 2010).

We photographed stretched wings of numerous European bird species that we captured in the wild or collected as fresh carcasses. We measured wing architecture from the photographs using ImageJ (http://rsbweb.nih.gov/ij/; see Vágási et al., 2016, for details). We described wing architecture as wing loading (kg m⁻²) and wing aspect ratio (dimensionless). Wing loading was calculated by dividing average body mass (kg, see source below) by wing area (m²), while wing aspect ratio was estimated as squared wingspan divided by wing area (both m²). Wing parameters were measured excluding body area between the wings, but these measurements are strongly positively correlated with wing morphology measurements that include body area (R²>0.99 for both wing aspect ratio and wing area in 152 species; Vágási et al., 2016). Species were assigned to one of the four flight type categories listed by Pennycuick (2008) and Bruderer et al. (2010), incorporating small modifications suggested by Pap et al. (2015): flapping and soaring, flapping and gliding, continuous flapping and passerine-type flight. Wing morphology and flight-type data are available in Table S1.

In order to quantify interspecific differences in accumulated fuel load, we extracted average and maximum body mass data from the literature (see Table S1 for species-specific data and references), separately for adult males and females, and separately for each species for which we had measured wing morphology. In cases when body mass was reported for multiple subspecies or populations, we chose the ones in which distribution overlapped with the site of collection of the wing morphology photographs. In all cases, average and maximum body mass as well as data on males and females were extracted from the same literature source. Species values of average and maximum body mass were then calculated as the arithmetic mean of the values extracted for the two sexes separately. Fuel factor was calculated as the ratio of species-specific maximum body mass to average body mass. This measure thus reflects how many times larger maximum compared with average body mass is for a given species (see also Hedenström and Alerstam, 1992). Our assumption is that fuel factor reflects accumulated fuel load as a function of lean body mass.

Body size can vary significantly among subspecies as well as across geographic regions inhabited by a single species (Cramp, 1998). Although we made sure to extract mean and maximum body mass from the same literature source, geographic variation in body size may bias our measure of fuel factor because of possible population mixing during migration. To counter this, we tested whether maximum body condition in a given species can be reliably estimated using fuel factor derived from the literature. We used data obtained at a migratory bird ringing station in the Danube-Delta Biosphere Reserve, on the coast of the Black Sea, Romania (44°32′N, 28°52′E). At this station, migratory birds were captured, measured and ringed throughout the peak migratory season, and it operates with a fixed number of mist nets that are usually open continuously. Activity at this ringing camp was concentrated around the peak of the autumn migratory season (from 4 August 2014 to 23 October 2014); individuals captured were almost exclusively migrants, as most do not breed in the area, and very few were re-captured weeks or months following the first encounter. We extracted body mass, wing length and fat score data from the ringing database. Fat score was recorded by assessing the visible subcutaneous fat deposit in each individual and was scored on a scale of 1 to 8 following Kaiser (1993). We built separate linear regression models for each species using log body mass (dependent variable) and log wing length (explanatory variable). To exclude outliers from these regressions, we calculated Mahalanobis distances (using function mahalanobis as implemented in R package stats) and excluded data points that had distances >9 to refit the model. The regression was only considered for species with at least 10 individual measurements available following the exclusion of outliers (a total of 12,214 captures, 45 species; see Table S1 for regression results). Individual body condition indices were extracted from the above regressions as residuals, reflecting the net difference between actual and expected body mass based on body size. In some poorly sampled species, the regression was weak, or in some cases a negative association between wing length and body mass was found, and it was therefore likely to reflect an erroneous fit. Consequently, the condition index was only considered further for species in which the association was at least close to significant (P<0.1, a total of 11,036 captures of 26 species).

Using individual body condition indices, we first tested whether the latter is correlated with individual fat scores using single-predictor linear regression models for each species separately. Second, to validate interspecific fuel factor data, we calculated species-specific maximum condition indices and analysed how fuel factor correlates with the latter across species. Maximum condition was calculated as the 95th percentile of individual condition data for each species, reflecting the high end of body condition or fuel accumulation of a given species during the autumn migratory period. We used the 95th percentile because extreme values, such as minima and maxima, are often unreliable and originate from measurement or data-handling errors. For this validation, we used a weighted linear regression between fuel factor (dependent variable) and maximum condition (predictor variable), with weights represented by the log number of captured individuals in the ringing database.

Distribution maps of the breeding and wintering ranges of western Palaearctic bird species (shape files) were retrieved from http://www.birdlife.org/datazone/info/spcdownload (BirdLife International and NatureServe, 2014). Geometric centroids of the spatial polygons for breeding (i.e. breeding and resident) and wintering (i.e. wintering and resident) ranges were then calculated from these maps using the gCentroid function of the R package rgeos (https://rdrr.io/cran/rgeos/), while migration distance was calculated as the geographic distance between the two centroids using a custom-written function in R (Vágási et al., 2016; Vincze, 2016; Vincze et al., 2015).

We conducted all analyses using phylogenetic generalised least squares (PGLS) models, implemented in the R packages nlme (https://CRAN.R-project.org/package=nlme) and ape (Paradis et al., 2004). We used fuel factor as a dependent variable with migration distance, mean body mass, wing aspect ratio, wing loading and flight type as explanatory variables. Body mass was log₁₀-transformed prior to the analyses. All models were weighted by the log sample size of individuals measured for each species. To control for similarity among taxa due to common phylogenetic descent, we downloaded 1000 phylogenetic trees from birdtree.org (Jetz et al., 2012), using the backbone tree of Hackett et al. (2008). The rooted consensus phylogenetic tree used in analyses was obtained using sumtrees software (Sukumaran and Holder, 2010), and models were fitted using maximum likelihood. We estimated phylogenetic dependence using Pagel's λ, set to take the most appropriate value in each model, as evaluated by likelihood ratio statistics. Model predictions and associated standard errors used for graphical presentation were extracted from PGLS models using the predictSE.gls function in the R package AICcmodavg (https://cran.r-project.org/package=AICcmodavg), while contrasts were extracted using the emmeans function from R package emmeans (https://CRAN.R-project.org/package=emmeans). We constructed models using all possible combinations of the five explanatory variables, and used second-order Akaike's information criterion corrected for small sample size (AICc) to compare them. We evaluated the importance of all candidate models using relative Akaike weights (ωi), and those of the predictors by the sums of their ωi across all models that contain the given predictor (Σωi). ωi and Σωi were computed for the whole model set; values close to 1 indicate high importance, while those close to 0 indicate low importance in explaining variance in the dependent variable. All analyses were conducted in R version 3.3.2 (http://www.R-project.org/) and all data are available in Table S1.

Given that wing aspect ratio and flight style are both key characteristics of bird flight, we ran a model to test how these two parameters are related. We built a PGLS model using wing aspect ratio as a dependent variable and flight style as the only independent variable. We controlled for phylogenetic inertia in a similar fashion to the models described above.

Individual condition was strongly positively correlated with individual fat scores in most species tested based on the ringing database (P<0.05 in 22 out of 26 species; see Fig. S1 for graphical presentations of the associations). This indicated that individual condition calculated as the residuals of a log–log linear regression between body mass and wing length reliably reflects body fat content within species. Fuel factor and maximum condition were strongly positively correlated across species (linear regression, n=26, β=2.29, s.e.=0.36, t=6.36, R²=0.64, P<0.0001; Fig. 1). Therefore, we conclude that fuel factor is a suitable cross-species measure for the maximum size of fuel stored during the migratory season.

In our sample of 213 European bird species, wing aspect ratio ranged between 3.51 (Poecile montanus) and 9.54 (Calonectris diomedea) (mean±s.e.m. 5.49±0.09), while wing loading ranged between 1.09 kg m⁻² (Phylloscopus collybita) and 23.44 kg m⁻² (Gavia arctica) (mean±s.e.m. 4.36±0.27 kg m⁻²). Fuel factor ranged from 1.06 (Picus viridis) to 1.80 (Sylvia borin) (mean±s.e.m. 1.26±0.01), while average body mass varied between 5.6 g (Regulus ignicapilla) and 10,735 g (Cygnus olor) (mean±s.e.m. 554.11±94.91 g). Migration distance varied from 0 km (residents) to 9856 km (Arenaria interpres) (mean±s.e.m. 3145.01±179.96 km).

The AICc-based comparison of all possible PGLS models explaining variation in fuel factor indicated that a single best-fitting model exists with an ωi value of 0.98 (see the entire model set and associated AICc values in Table S2). This model contained migration distance, wing aspect ratio and flight type as explanatory variables (Table 1). The second best-fitting model had ΔAICc=7.46, indicating a significantly lower fit. Summing Akaike weights for each parameter indicated that migration distance and flight style have the highest importance (Σωi=1), followed by wing aspect ratio (Σωi=0.98), while body mass (Σωi=0.02) and wing loading (Σωi=0.00) had little explanatory power.

As predicted, fuel factor increased with increasing length of the migratory journey (Table 1, Fig. 2). Nonetheless, significant variation remained unexplained across this regression slope (Fig. 2). Some of this residual variance was explained by wing aspect ratio, indicating that species with higher wing aspect ratio (i.e. slender wings) accumulated less fuel than species characterized by a low wing aspect ratio (i.e. rounder wings; Table 1, Fig. 2). Finally, the effect of flight type was also significant, indicating different fuel factors in species with different flight styles even after accounting for wing architecture and migration distance. Our results indicate that continuously flapping species accumulated the highest fuel stores, followed by flapping and soaring species, and flapping and gliding species, while the smallest fuel factors were observed in species with passerine-type flight (Table 1, Fig. 3). Note, however, that the difference was only significant between continuous flapping and passerine-type flight styles (t=5.70, P<0.0001), between flapping and soaring and passerine-type flight styles (t=6.33, P<0.0001), as well as between continuous flapping and flapping and gliding flight types (t=2.96, P=0.0180; Fig. 3). Although the association between wing aspect ratio and flight type is strong, the effects of these variables were consistent across models, and the results were not affected by collinearity (variance inflation factor was <4 in all models).

The results of the model testing the effect of flight style on wing aspect ratio indicated a significant effect (n=213, F=6.47, P=0.0003; Fig. 4). A post hoc test revealed that birds having a passerine-type flight have the lowest wing aspect ratio (i.e. broader and rounder wings), and this was significantly lower than that in birds having a flapping and soaring (t=3.33, P=0.0056) or flapping and gliding flight type (t=3.64, P=0.0020). No other flight style category pairs differed significantly in wing aspect ratio according to these results (Fig. 4).

Our study highlights a number of key traits that are related to the cross-species variation in migratory fuelling strategy of birds. First, we show that, as expected, accumulated fuel stores in birds significantly increase with the geographic distance covered during their migratory flight. This association holds even after controlling for phylogeny, but there is large variation among species in fuelling strategy once migration distance is controlled for. Second, species with high wing aspect ratio accumulate less fuel prior to migratory departure than species with low wing aspect ratio. Third, flight type further influences fuel stores irrespective of wing morphology and migration distance, implying that fuel stores are fine tuned in accordance with wing-beat frequency. Fourth, wing loading and body mass probably have little influence on the amount of fuel stored for migration.

Migration is one of the most energy-demanding behaviours in birds, with long-distance non-stop flights often assumed to be close to the edge of physiological endurance (Hedenström, 2010; Weber, 2009). Because of the energetic expense of avian flight and the frequent lack of refuelling sites en route, pre-migratory fuelling is an important determinant of flight range (Hedenström, 2010; Hedenström and Alerstam, 1992). The obvious positive association between fuel load and migration distance has previously been demonstrated at the species level. For instance, the size of the fuel reserves in three populations of redbilled queleas (Quelea quelea) differs in accordance with the distance each of these travels during their migration (Ward and Jones, 2010). A similar pattern of fuel storage was found across different barn swallow (Hirundo rustica) populations, where pre-migratory fat stores increased with the length of the ecological barrier to be crossed upon departure (Rubolini et al., 2002). Although here we quantified overall migration distance, and not the longest flight section, our results still reinforce the prediction that species optimise their fuel stores in proportion to the energy required for their migratory flight. Nonetheless, the considerable residual variance in fuel load highlights that additional factors influence fuelling strategy. Among these, the frequency and length of stopovers is certainly a key component that influences fuelling decisions (Rubolini et al., 2002).

Among the wing morphological characters tested here, wing aspect ratio is the most strongly correlated with flight energy efficiency in birds. Long and narrow wings (i.e. high wing aspect ratio), especially when combined with low wing loading, ensure energy efficient flight (Bowlin and Wikelski, 2008; Sapir et al., 2010; Vágási et al., 2016; Weber, 2009). Given that the induced drag is proportional to the inverse square of the wingspan, higher aspect ratio reduces the induced drag, at similar wing loading and flight speed. Accordingly, the cost of flight per unit distance is lower in species with a high wing aspect ratio, as maximal accumulated fuel load decreased with increasing wing aspect ratio across birds, indicating lower amounts of stored energy required when the expected cost of flight per unit distance is lower. The role of energy efficiency in determining fuel load can also be concluded from the significant effect of flight type on fuel factor, even after accounting for wing morphology.

The different energetic needs of various flight types appear to be supported by data (Norberg, 1990). Continuous flapping flight mostly occurs in birds with relatively high wing loading (Bruderer et al., 2010), which requires a large energy expenditure (Agostini et al., 2015; Pennycuick, 2008). It is therefore straightforward why continuously flapping species are forced to accumulate the largest amounts of fuel per unit length of migration distance in comparison to species belonging to the other three flight-type categories. In support of this, the longest non-stop flight ever recorded for birds was accomplished by a continuous flapping species, the bar-tailed godwit (Limosa lapponica) (Gill et al., 2009). Surprisingly though, species with flapping and soaring flight accumulated only slightly lower amounts of fuel than continuously flapping species and hence the two groups did not differ statistically. Soaring is often referred to as one the most energy-efficient ways of locomotion, requiring slightly more energy than sitting still, but this only applies if thermal currents are available (Hedenström, 1993). During migration, soaring birds often cross areas where thermal currents are weak, such as large water surfaces, and they need to switch to powered flight for prolonged periods (Agostini et al., 2015; Pennycuick, 2008). The generous fuel store observed in soaring species could therefore serve as a safety margin, offering a source of energy during prolonged powered flight under unfavourable conditions, such as storms or weak thermals (Pennycuick, 2008). An alternative strategy often observed in soaring migrants is to take large detours and avoid flying over large open water surfaces, with the aim of avoiding powered flight (Agostini et al., 2015). This strategy also depends on body mass, as a larger body mass entails larger energetic costs of powered flight, and consequently heavier soaring species are most reluctant to fly over open sea (Agostini et al., 2015; Panuccio et al., 2013).

The flapping and gliding flight style observed in birds such as swifts, falcons or bee-eaters is theoretically also energetically cheaper than continuous flapping flight (Sapir et al., 2010). Detailed data on migrating bee-eaters equipped with ratio transmitters indicates that they use flapping and gliding flight intermittently, with flapping flight depending on wind conditions (Sapir et al., 2010). Consequently, intermittent use of these two flight styles could help save energy, as gliding phases in these species are not dependent on thermal currents like in soaring birds, and could therefore explain the lower fuel factor of flapping and gliding species compared with continuous flapping or soaring species. Passerine-type flight is an extreme form of intermittent flight mostly confined to species with a relatively low wing aspect ratio (Bruderer et al., 2010; Sachs, 2015). This flight style involves short phases of high-frequency flapping interrupted by short glides (Bruderer et al., 2010). During the gliding phase the wings fold against the body and no mechanical power is required. Therefore, some models suggest there are energetic advantages to passerine-type flight when compared with continuous flapping (Sachs, 2015); however, this advantage is still disputed (Pennycuick, 2008). Our results corroborate the energetic advantage of passerine-type flight over continuous flapping, as we show that species in the former group accumulate the smallest fuel stores per unit distance of migration among the four flight types tested, after accounting for wing morphology. Although passerine-type fliers usually have small body sizes and consequently broad power margins (Norberg, 1996), they store the smallest amounts of fuel, which might indicate little energetic constraint in these species. Overall, our results on the effect of flight type on fuel factor are intriguing and the explanations we provided above are speculative; therefore, further research is needed to investigate the metabolic needs of various flight types during migration and how flight type relates to utilisation of stopover sites.

Our results on the effect of wing aspect ratio and flight style on fuel factor support the assumption that maximum fuel load in birds evolved to meet the energetic requirements of the flight apparatus and birds aim to avoid carrying extra load that is unlikely to be burnt. This suggestion is also supported by the fact that birds often maintain stable reserve levels that are well below the maximum that could possibly be attained (Biebach, 1996; Blem, 1990; Witter and Cuthill, 1993). This optimisation of fuel load is driven by the increased flight cost imposed by the weight of the reserves by means of elevated wing loading and associated elevated costs of locomotion (Hedenström, 2010; Witter and Cuthill, 1993). Large fuel reserves also impose elevated metabolic expenditure, increased risk of injury or predation, impaired foraging and take-off ability or impaired flight performance (Biebach, 1996; Witter and Cuthill, 1993). The fact that most organs suffer a significant reduction in size prior to or during migration to reduce the cost of transportation also highlights the costs of extra weight (Battley et al., 2000).

Species with large wing loading may require more energy to cover a given distance than species with lower body mass per unit wing area (Schmidt-Wellenburg et al., 2008), because wing loading is positively related to flight velocity (Norberg, 1990). Nonetheless, fuelling capabilities are also more limited in species with higher wing loading because of their lower power margin. Our results indicate that wing loading has very little predictive power for fuel factor and are in accordance with earlier studies demonstrating no association between wing loading and migration distance among species (Vágási et al., 2016; Winkler and Leisler, 2008).

In conclusion, our results indicate that fuel stores are optimised as a function of overall distance to be covered during the migratory flight as well of the energetics of species-specific locomotion. Species with slender wings and therefore more energy-efficient locomotion accumulate smaller fuel stores, pointing towards the importance of minimisation of the amount of stored fuel in accordance with the expected energy expenditure of the migratory flight. The optimal fuel store should safeguard against premature energy depletion, but also avoid costs associated with carrying surplus weight that is unlikely to be used and is associated with physiological and environmental threats. Our results are also important from a conservation point of view. Habitat fragmentation, degradation or disappearance of stop-over sites interferes with fuelling and re-fuelling during migration. Our results suggest that birds may possess little safety margin concerning their stored energy and the integrity of stopover sites is paramount for a successful migratory journey. Future studies should collect more detailed information about the longest flight spans, the number and position of stopover sites and body condition indices of numerous species at stopover sites for a better understanding of fuelling strategies of birds. Detailed data on fuelling strategy and the conditions of stopover sites together might also help to explain large-scale declines in migrant birds (Both et al., 2010).

We thank the members of the ‘Milvus Group’ Bird and Nature Protection Association for providing data from the 2014 Chituc Ringing Camp, Romania. We thank Gareth Dyke for his linguistic corrections on an earlier version of the manuscript. This study forms part of the PhD thesis (2017) of O.V. at the University of Debrecen, Hungary.