A songbird compensates for wing molt during escape flights by reducing the molt gap and increasing angle of attack

Avian plumage not only functions as a protective barrier and insulation layer, but is also essential for locomotion, forming the aerodynamic shape of the animal's body, tail and wings during flight. As a result, avian feathers are under strong selective pressure for optimal flight performance (Jenni and Winkler, 1994). Environmental and biological factors such as sunlight, weather and parasites cause feathers to degrade over time, reducing their quality and compromising all activities of an individual (Barbosa et al., 2002; de la Hera et al., 2010; Swaddle et al., 1996; Weber et al., 2005). Thus, birds need to replace their feathers in order to retain quality in the so-called molt process (Jenni and Winkler, 1994; Pap et al., 2007). This is particularly important for the flight feathers, as a degraded or damaged feather deck is likely to impact flight performance (Swaddle et al., 1996).

The molting process is energetically costly because individuals need to grow new feathers and maintain tissues for feather production (Lindström et al., 1994; Murphy and King, 1991, 1992). It also has to be timed correctly in the season because if molt is delayed or hastened, or when it overlaps with other stages in the annual cycle, it may compromise plumage quality (Dawson, 2004; Jenni and Winkler, 1994; Nilsson and Svensson, 1996; Vágási et al., 2012). Therefore, allocation of this expensive stage in the annual cycle of a bird is an important life-history decision (Barta et al., 2008; Hemborg et al., 2001; Holmgren and Hedenström, 1995; Jenni and Winkler, 1994).

The energetic requirements of growing new feathers, however, is not the only reason why molt is costly. During molt, birds are also forced to fly with missing wing feathers, which forms gaps on their wings and causes additional energetic costs (Chai, 1997; McFarlane et al., 2016; Swaddle and Witter, 1997; Swaddle et al., 1999; Williams and Swaddle, 2003). Such molt gaps are detrimental to bird flight owing to a reduction in the wing area, altered wing shape and a consequent increase in wing loading, hindering the ability to generate aerodynamic lift or causing additional aerodynamic drag (Achache et al., 2018; Chai, 1997; Hedenström and Sunada, 1999; Kleinheerenbrink and Hedenström, 2017; McFarlane et al., 2016). Both the gap size and position are detrimental for flight, with a strong decline in performance when the gaps are situated inside the wing, which is the case of early molt stages (Achache et al., 2018; Hedenström and Sunada, 1999).

There are different strategies that individuals use to reduce the costs of molt, such as molting just one or few feathers at once and allocating molt to moments of the year when there are no other costly events such as breeding or migration (Barta et al., 2006, 2008; Jenni and Winkler, 1994). Still, it is not uncommon for some birds to start to molt while still breeding, even if this means that they will pay additional costs of overlapping molt and breeding (Echeverry-Galvis and Hau, 2013; Hemborg, 1999; Hemborg and Lundberg, 1998).

In a previous study, we investigated the costs of molting while breeding and showed that male pied flycatchers (Ficedula hypoleuca) with simulated molt gaps in their wings suffer from flight costs measured as distance gained per wingbeat, a parameter with potential energetic implications (Tomotani et al., 2018b). However, this reduction in performance did not reflect in a reduction in flight speed of birds with simulated molt gaps (Tomotani et al., 2018b). Similarly, a study with starlings showed that birds with simulated molt gaps had a low take-off speed immediately after manipulation, but that effect disappeared over time (Williams and Swaddle, 2003). These results suggest that individual birds may be able to behaviorally compensate for the detrimental effects of wing feather gaps on take-off performance (Tomotani et al., 2018b; Williams and Swaddle, 2003). Here, we investigated whether and how birds compensate for the detrimental effects of molt gaps on escape flight performance via adjustments in their flight kinematics. We studied the upward-directed escape take-off maneuvers of pied flycatchers with and without experimentally induced gaps in their wings simulating early molt stages (henceforth ‘control group’ and ‘molt group’, respectively). We used video recordings of 146 escape take-off flights in a vertical flight chamber of 19 birds with simulated molt gaps and 22 control birds to create a dataset of the three-dimensional body and wing movements throughout the escape flight. Based on these data, we described in detail how gaps of early molt stages affect wing morphology and escape flight performance, and how pied flycatchers adjust their flight kinematics in order to negate the negative aerodynamic effects of molt.

Flying animals flap their wings to produce aerodynamic forces required for flight. During steady horizontal flight, the animal needs to produce an upward-directed aerodynamic lift force that is equal in magnitude to the weight of the animal, and a forward-directed thrust force produced by the flapping wings that cancels aerodynamic drag mostly produced by the body (Alexander, 2004).

During vertical escape flights, in contrast, the animal should maximize the upward-directed aerodynamic force (F_aero) in order to accelerate upwards as fast as possible. The resulting high upward accelerations throughout an escape flight lead to a high escape speed as well as a short time duration of the escape maneuver. Both characteristics are associated with a high escape performance (Muijres et al., 2014; Swaddle et al., 1996), as they minimize the chance of being captured. Equally, these metrics could also be used to quantify capture performance in predators (Hedenstrom et al., 2001).

The advance ratio of the escape take-offs in the present study are relatively low, and translation velocities of the bird are relatively small compared with wingbeat- and tailbeat-induced velocities. Because aerodynamic forces scale with velocity squared (Anderson, 1985), we ignore aerodynamic forces that are primarily the result of the relatively low translational velocities. Therefore, we assume that body-induced aerodynamic forces are negligible in our aerodynamic model for escape take-off maneuvers in birds (F_body=0). Note that because wing molt most likely does not change body drag directly, even if body drag forces are not negligible, this simplification will most likely not affect our study of the effect of wing molt on flight kinematics and aerodynamics.

The aerodynamic model as described by Eqns 1–5 will be used to study how wing molt affects the flight kinematics, aerodynamics and performance of escape take-offs in pied flycatchers. Based on this model, we hypothesize that the primary detrimental effect of wing molt is that molt gaps cause a reduction in S₂ of the wings, which will have a negative effect on force production by the wings (Eqn 4). This could then lead to a reduction in escape flight performance as expressed by a reduction in F*_aero (Eqns 1, 2). However, our previous study suggests that instead of having a reduced escape performance, our molting pied flycatchers adjusted their flight kinematics in order to negate this negative effect (Tomotani et al., 2018b). Therefore, using our aerodynamic force production model for the wings and tail (Eqns 4 and 5, respectively), we will investigate how these pied flycatchers adjusted their wingbeat and tailbeat kinematics in order to compensate for wing molt.

According to Eqn 4, birds can increase aerodynamic forces produced by the molting wings by increasing S₂ (e.g. by spreading their remaining wing feathers), increasing the angle of attack of the beating wings, and increasing the (angular) speed of the wings. Likewise, birds can increase force production by the tail by spreading the tail (increasing b_tail), increasing the tail angle of attack, and increasing the speed of the tail (Eqn 5). Therefore, we measured these parameters in escaping flycatchers, and tested how they vary between birds with and without simulated molt gaps. Note that, especially at relatively low flight speeds, pied flycatchers have an inactive upstroke whereby the wing does not produce significant aerodynamic forces (Muijres et al., 2012; Norberg, 1975). Therefore, we will focus on the wingbeat kinematics particularly during the aerodynamically active downstroke.

The pied flycatcher, Ficedula hypoleuca (Pallas 1764), is a small long-distance migratory bird that reproduces in Europe and Western Asia and winters in West Africa (Lundberg and Alatalo, 1992; Ouwehand et al., 2016). The field part of the experiment was conducted from early April until late June 2015 in the forests of the Hoge Veluwe National Park (The Netherlands; 5°51′E, 52°02′N). We provided approximately 400 nest boxes year-round in an area of 171 ha; these are occupied in spring by cavity-nesting passerines, such as pied flycatchers. Every year, this pied flycatcher population is monitored and data on arrival dates of males, nest building of females, female egg-laying dates, chick hatching dates, brood success and adults and chick basic biometrics are collected. Voucher material of this population was deposited in the ornithology collection of the Naturalis Biodiversity Center (Leiden, The Netherlands) under the inventory numbers RMNH 592347, RMNH 592348 and RMNH 592349.

Birds used in the present study were part of a previous field–laboratory experiment designed to test the effects of simulated molt gap on fitness (Tomotani et al., 2018b). Adult males were captured when feeding their 7-day-old chicks and randomly assigned to a treatment. If a male was in the molt group, we simulated early molt stages by plucking primaries 2 and 3 of both wings, following the molt sequence (Jenni and Winkler, 1994). If a male was in the control group, it was handled as a molt group male, but no primary feathers were removed. Our treatment mimicked the natural molt process in pied flycatchers, with the exception that we removed primaries 2 and 3, instead of 1 and 2. We opted to not remove the first primary feather because this allowed us to assess the date of the natural molt onset as the moment when the first primary was dropped. Nevertheless, our treatment still created a similar-sized gap in a location very close to where the natural molt would start. After this experimental treatment, all birds were released. Later, when chicks were 12 days old, those males were captured a second time and taken to the Netherlands Institute of Ecology (Wageningen, The Netherlands), where we recorded their flight.

All procedures were carried out under licenses of the Animal Experimental Committee of the Royal Netherlands Academy of Sciences (KNAW) (protocol NIOO 14.13). The molt treatment consisted of pulling feathers from the wings of the males upon capture, a process that lasted a few seconds and was only performed by experienced researchers. Moreover, the return rates of molt and control males in the following year did not differ (see Tomotani et al., 2018b). More details regarding the design and results of this field experiment can be found in Tomotani et al. (2018b).

Escape flight experiments were performed in a vertical flight arena with a stereoscopic videography system (Fig. 1A), as described in Tomotani et al. (2018b). The flight tunnel consisted of a release chamber, a flight chamber and a collection chamber. The flight chamber had dimensions 50×50×150 cm (length×width×height), and the release and collection boxes were each 50×50×30 cm in size. The release and collection chambers were removable and identical in design, such that they were interchangeable and could be used as a transport cage. Each cage had a perch and a sliding door (50×50 cm) that could be quickly opened manually by the experimenter.

Before each experimental session, a single bird was transferred from its housing cage to the release chamber and transported to the experimental room. There, the release chamber was connected to the bottom of the flight arena and the sliding door was quickly opened. This would trigger the bird to fly upward and land on the perch of the collection chamber on the top. After this, the experimenter would close the sliding door of the collection chamber, switched the release and collection boxes, and performed a second flight experiment by again quickly opening the sliding door of the release chamber.

The upward flight maneuvers were filmed with a stereoscopic videography system, consisting of three synchronized Basler piA64-210gm cameras, each with a Nikkor f/2.8 lens and a 300 W halogen floodlight (GE lighting, PAR56) for illumination. Each camera had a spatial resolution of 648×488 pixels and a gray-scale bit depth of 8 bits, and operated at 150 or 200 frames s⁻¹ (Fig. 1B, Movies 1, 2). The stereoscopic camera system was calibrated at least once a week using the direct linear translation (DLT) method (Hatze, 1988), based on a calibration frame with 22 randomly placed calibration points, and using an open-source MATLAB (The MathWorks) DLT calibration software package (Woltring and Huiskes, 1990). The accuracy of each DLT calibration was estimated as the mean absolute calibration error, defined as the mean absolute distance between the location of each calibration point and its three-dimensional reconstruction; for our study, this mean absolute calibration error was 7.5±0.6 mm (mean±s.d., n=7 calibrations).

The stereoscopic camera system filmed a volume of approximately 40×40×40 cm on the bottom half of the flight chamber, and thus the mean absolute calibration error equals 1.1% of the diameter of this volume of interest. We chose to film this region in the bottom half of the flight chamber because we assumed that in this section the birds were producing maximum aerodynamic forces in order to accelerate upward. Closer to the take-off perch, the birds might still be transitioning from the push-off phase to the flight phase, and more towards the collection chamber they might start to slow down in order to prepare for landing.

During the experiments, the camera system was continuously recording to a buffer of 1000 video frames (5 or 6.7 s) for each camera. When the system was manually triggered after a bird performed a flight maneuver, recording was stopped and the final 1000 video frames recorded by each camera before triggering were saved and stored for later analysis (Movies 1, 2).

Throughout each recorded stereoscopic video, we manually tracked 14 morphologically distinct markers on the body, wings and tail of the upward flying bird (Fig. 1C), using a MATLAB tracking software package (Hedrick, 2008). The body and tail markers included the tip of the beak, the rump, and the left and right tail tip. On each wing, we tracked five markers: the shoulder, the wrist, the wing tip defined as the tip of the eighth primary feather (P8), and the tip of the first and fourth primary feather [P1 and P4, respectively; P1 and P4 were adjacent to the feathers that we removed in the molt-simulated group (P2 and P3)].

We used the open-source DLT calibration code (Woltring and Huiskes, 1990) to convert all video-tracked marker positions into their three-dimensional positions. For each of these three-dimensional reconstructions, we determined the mean absolute reconstruction error, defined as the mean distance between the measured location of a marker on each camera sensor and the re-projected location on the camera sensor of the estimated three-dimensional marker position. For all three-dimensional reconstructions, the mean absolute reconstruction error was 2.7±2.6 pixels (mean±s.d., n=39,503 reconstructions), which equals 0.3% of the diameter of each camera sensor.

The resulting three-dimensional tracks were filtered using a linear Kalman smoother (Muijres et al., 2015), which provided us with filtered estimates of position, velocity and acceleration of all data points. For the Kalman smoother, the measurement noise covariance matrix was set to identity, the process noise matrix was set to 10, and the cross-product of the error covariance matrices was set to zero. A comparison between the unfiltered and Kalman filtered data of a flight maneuver is shown in Fig. S1.

The filtered data were used to determine the various kinematics parameters throughout each measured wingbeat. We first separated each flight sequence into distinct wingbeats, by manually identifying the video frames at which the wingbeat transitioned from downstroke to upstroke, i.e. when the wingtip switched from a downward to upward movement. Based on this, we defined the temporal dynamics throughout the wingbeat as normalized time τ=t/Δt, where Δt is the time difference between two consecutive downstroke-to-upstroke transitions. And thus τ=0 at the start of each upstroke, and τ=1 at the end of the next downstroke. The flapping frequency of each wingbeat was calculated as f=1/Δt. We used the tip of the beak to determine the flight path, flight speed U_body(τ) and weight-normalized net aerodynamic force F*_aero(τ) (Eqn 2) throughout each wingbeat.

More detailed wingbeat kinematics analysis was performed by dividing the wing into four triangles, each spanned by three tracked markers (Fig. 1C): the inner wing triangle T_in, the mid wing triangle T_mid, the outer wing triangle T_out and the simulated molt gap triangle T_gap. Thus, for the molt group, the molt gap was defined as the triangle spanned by the shoulder joint and the wing tips of feathers P1 and P4, and simulated molt gap width (b_gap) as the distance between the tip of P1 and P4.

For each wing triangle, we calculated its area S, its second moment of area S₂ relative to the shoulder marker, its velocity vector U as the average velocity of its three markers, and the angle of attack α as the angle between the velocity vector U and the surface of the triangle (Fig. 1C). The average wing speed U_wing and angle of attack α_wing for the bird were estimated as the average speed and angle of attack of the inner, middle and outer wing triangles of both wings combined. For control birds without a simulated molt gap, total wing area and S₂ were estimated as the sum of S and S₂ for all the wing triangles of both wings combined, respectively. For birds with simulated molt gaps, the gap triangles were not included in the S and S₂ calculation.

We defined the tail as a triangle spanned by the rump marker and the two tail tips. From this tail triangle, we calculated tail velocity U_tail and tail angle of attack α_tail, using the same method as for the wing triangles (Fig. 1C). Tail width b_tail throughout each wingbeat was calculated as the distance between the two tail tip markers.

All statistics were performed using R version 3.4.3 (https://www.r-project.org/). We tested how wing molt affected upward-directed flight dynamics of pied flycatchers using linear mixed-effect models. Mixed-effect models were fitted to each flight performance, morphology and kinematics component (R package lme4, Bates et al., 2015) as a response variable, with treatment as fixed effect and bird ID as a random effect to take into account that each individual was tested multiple times. Treatment effects were tested using a Kenward–Roger approximation for F-tests, comparing models with and without treatment (R function KRmodcomp from the pbkrtest package, Halekoh and Højsgaard, 2014); data did not violate model assumptions and critical P-values were subsequently corrected for multiple testing using the Holm–Bonferroni method (Holm, 1979).

The tested flight performance, morphology and kinematics components included all variables identified as important for aerodynamic force production in upward-directed avian flight (Fig. 1C). The flight performance metrics were flight speed and weight-normalized net aerodynamic force; the wing morphology parameters were molt gap size and second moment of area of both wings combined; the wingbeat kinematics parameters were the average wing speed and angle of attack of both wings combined (Eqn 4); and the tail kinematics parameters were tail speed, tail angle of attack and tail spread (Eqn 5).

For the flight performance metrics flight speed and normalized force, we used the wingbeat average values. For all other parameters, we used the average values near the moment within the wingbeat when force production was maximum (F*_aero≈F*_aero,max). This was around mid-downstroke, within the wingbeat-normalized time window of 0.5<τ<0.6. Our rationale for analyzing the kinematics parameters near maximum force production is that in this time window, the effect of these parameters on flight performance are also most likely at a maximum (Eqns 4 and 5).

To test which flight kinematic components best explain the force production, we used a linear mixed-effect model with normalized force as the response variable and second moment of area, flight speed, wing speed, wing angle of attack, tail speed, tail spread and tail angle of attack as fixed effects, again using bird ID as a random effect. To define the minimal model, we used backwards model selection, dropping non-significant terms in each step. Once more, effects were compared with a Kenward–Roger approximation for F-tests.

In addition to the isolated comparisons, we also carried out a principal component analysis (PCA) to visualize all metrics together. The PCA reduces the number of dimensions of data by geometrically projecting the data into lower dimensions (principal components, PCs). It thus reduces the complexity of high-dimensional data but retains trends and patterns (Jolliffe, 2002; Lever et al., 2017). All analyzed metrics were included in order to detect whether data of the two treatments would cluster and which metric(s) would be related to the treatment effects. The PCA was based on the standardized measurement values (mean centered at 0, standard deviation at 1) of the variables.

Pairs of control and molt males (n=29 pairs, 58 males) with the same hatch date and same brood sizes were selected throughout the season, covering the full range of hatching dates. This ensured that the treatment groups did not significantly differ in terms of average chick hatching date or brood size (see Tomotani et al., 2018b). From these 58 nests, however, we only analyzed recordings of 41 males; the remaining birds were either not recorded (e.g. natural molt, desertion, see Tomotani et al., 2018b) or recordings were not precise enough to trace the whole wing movement. However, this subset of nests still did not differ in terms of brood size (F_1,40=−1.30, P=0.20) or hatching date (F_1,40=2.48, P=0.80).

Based on 10 years of molt data, male pied flycatchers in this population start to symmetrically molt on 13 June on average (Tomotani et al., 2018a). In the year of the experiment (2015), males started to molt on average on 15 June, while flight trials took place between 28 May and 18 June. Natural molt onset was monitored in all individuals and was not affected by treatment; individuals observed in natural molt prior to flight trials were excluded from all analyses (also see Tomotani et al., 2018b).

We recorded and analyzed 73 upward-directed escape flight sequences of 22 control birds, and 73 sequences of 19 birds with simulated molt gaps (see Movies 1 and 2 for respective example videos). By manually tracking the 14 body, wing and tail markers in 4147 frames of these 146 stereoscopic videos, we determined the wing, body and tail kinematics throughout a total of 410 complete wingbeats (a full dataset is available from the Dryad Digital Repository: doi:10.5061/dryad.g28t010).

The flight speed of the control birds and birds with simulated molt gaps varied throughout the wingbeat, with a consistent offset in flight speed between the control and molt groups (Fig. 2A). Despite this offset, the average flight speed throughout the wingbeat was not significantly different between the two groups [U_body,control=2.53±0.03 m s⁻¹ (mean±s.e.m.), n=73 flights; U_body,molt=2.47±0.03 m s⁻¹, n=73 flights; F_1,38.48=0.94, P=0.34; Fig. 2E], and thus both the control and molt groups flew upward with a flight speed of approximately 2.5 m s⁻¹.

The net weight-normalized aerodynamic forces also varied throughout the wingbeat, and these dynamics were strikingly similar between the control and the molt groups (Fig. 2B): for all birds, normalized forces increased on average from a g-force of 1.6 g at the start of the wingbeat (τ=0) to a maximum of 2.6 g near mid-downstroke (τ≈0.55). The resulting wingbeat-average normalized forces were not significantly different between the two groups (F*_aero,control=2.16±0.05, n=73 flights; F*_aero,molt=2.09±0.05, n=73 flights; F_1,37.89=0.69, P=0.41; Fig. 2F). Thus, throughout the escape maneuver, both the control and molt birds produced similar net aerodynamic forces of on average 2.1 g, and that peaked near mid-downstroke at a value of 2.6 g.

Based on the tracked wing markers, we measured the temporal dynamics of the second moment of area of both wings combined throughout the wingbeat (Fig. 2C). Because the wing markers are only clearly visible during its downstroke movement, we were only able to accurately estimate S₂ (and any other wing kinematics parameter) within the time window of 0.25<τ<0.8. Within this time window, the second moment of area first slowly increased to a maximum at roughly mid-downstroke (τ∼0.5), and then dropped off towards the end of the downstroke. Throughout the complete measured wingbeat section, the second moment of area was larger for the control group than for the molt group (Fig. 2C); also, the average second moment of area near maximum force production (0.5<τ<0.6) was significantly higher for the control birds (S_2,control=1.61±0.05 dm⁴, n=66 flights; S_2,molt=1.28±0.04 dm⁴, n=65 flights; F_1,35.72=20.61, P<0.01; Fig. 2G). These results show that the birds with simulated molt had wings with a 20% lower second moment of area compared with the control group.

We tested how this reduction in second moment of area relates to the introduction of the molt gap by comparing the distance between the tips of primary feathers P1 and P4, which for the birds in the molt group is representative of the simulated molt gap width (Fig. 1C). This P1–P4 distance was on average 31% larger for the control birds than for the birds with simulated molt gaps (b_gap,control=5.20±0.08 cm, n=66 flights; b_gap,molt=3.55±0.14 cm, n=65 flights; F_1,37.41=48.59, P<0.01, Fig. 2D,H), and thus the birds with molt gaps had a reduced size of this gap. This molt gap reduction partly negated the detrimental effect of molt on the second moment of area of the wing.

The molt-induced reduction in second moment of area causes that the birds in the molt group have less S₂ available to produce the same aerodynamic forces (Eqn 4, Fig. 2). To achieve this, birds with a simulated molt gap should adjust their wing and tail kinematics. We tested how the birds in the molt group do this by comparing wing and tail kinematics between the two groups.

Birds can increase aerodynamic forces produced by their tail, by adjusting the spread, speed and angle of attack of the tail (Eqn 5), and thus we tested those three parameters. None of these differed significantly between the molt and control groups (Table S1), suggesting that pied flycatchers do not use their tail to compensate for wing molt (U_tail,control=3.63±0.07 m s⁻¹, n=72 flights; U_tail,molt=3.41±0.05 m s⁻¹, n=73 flights; F_1,38.46=1.89, P=0.18; b_tail,control=5.25±0.23 cm, n=72 flights; b_tail,molt=5.08±0.26 cm, n=73 flights; F_1,37.77=0.15, P=0.70; α_tail,control=32.1±1.76 deg, n=72 flights; α_tail,molt=33.94±2.07 deg, n=73 flights; F_1,38.47=0.47, P=0.50).

Birds can increase the aerodynamic forces produced by their flapping wings primarily by increasing the wing speed and by adjusting the angle of attack (Eqn 4). The temporal dynamics of wing speed throughout the wingbeat is similar between the birds in the control and molt groups: the speed of both the inner and mid wing sections remain roughly constant throughout the downstroke (Fig. 3A,B); for the outer wing triangle, the wing speed first increases to a maximum of roughly 12 m s⁻¹ at τ=0.4, after which it decreases again (Fig. 3C). Although the temporal dynamics are similar between the control and molt groups, the speeds are on average higher for the birds with simulated molt gaps (Fig. 3), which is also the case for the average speed of the complete wing (Fig. 4A). Comparing the average wing speed at maximum force production between the control and molt groups shows that the average speed was not significantly different between the groups (U_wing,control=6.52±0.16 m s⁻¹, n=66 flights; U_wing,molt=7.00±0.16 m s⁻¹, n=65 flights; F_1,35.08=4.28, P=0.05; Fig. 4C).

The temporal dynamics of the angle of attack throughout the wingbeat also differed between the control and molt groups (Fig. 3D–F), particularly near mid-downstroke, when force production is maximal (0.5<τ<0.6). Around this phase, the angle of attack dips for all wing sections of both groups, but this dip is consistently less pronounced in the molt group (Fig. 3D–F). The same difference is observed for mean angle of attack of the complete wing (Fig. 4B), and as a result, the average wing angle of attack at maximum force production is significantly higher for birds with a simulated molt gap (α_wing,control=19.4±0.8 deg, n=66 flights; α_wing,molt=23.7±0.8 deg, n=65 flights; F_1,33.20=15.78, P<0.01; Fig. 4D).

Independent of molt treatment, the different birds produced various amounts of mean normalized forces throughout their maneuvers (Fig. 5). This allowed us to test how these birds adjusted their wingbeat kinematics to control their aerodynamic force production during upward-directed escape maneuvers. When testing for the variables that explain the variation in normalized force production, only flight speed, wing speed and tail spread related significantly to force (Fig. 5A–C, Table S2: normalized force per flight speed slope=0.53±0.14 m⁻¹ s, F_1,99.62=13.98, P<0.01; normalized force per wing speed slope=0.08±0.03 m⁻¹ s, F_1,126.86=6.12, P=0.01; normalized force per tail spread slope=0.07±0.02 cm⁻¹, F_1,120.81=16.70, P<0.01). Thus, birds that flew faster also produced higher normalized forces, suggesting that these birds work harder throughout the escape maneuver. The results also suggest that normalized forces are enhanced by increasing the wing flapping speed (a g-force increase of 0.08 g per 1 m s⁻¹ wing speed increase) and by increasing tail spread (a g-force increase of 0.07 g per 1 cm increase in tail spread). What is striking is that normalized force is not related to wing angle of attack (slope<0.01±0.01 cm⁻¹, F_1,121.94=0.44, P=0.51).

We retained the PCs with variance above l, leaving us with the first three PCs that, combined, explained 58% of the variation. All of these three PCs differed significantly between the control and molt groups (PC1: F_1,38.45=6.88, P=0.01; PC2: F_1,38.00=5.80, P=0.02; PC3: F_1,37.83=26.32, P<0.01; Tables S3 and S4), but only PC2 and PC3 explained the variation in S₂ (Table S2).

When the first three PCs are represented in the biplots PC1–PC2 and PC1–PC3 (Fig. 6), there is a clear clustering of birds in control and molt groups, albeit with some overlap. This clustering is mostly evident along the PC2 and PC3 axes (Fig. 6A,B). The vector (loadings) plots are consistent in showing that birds in the molt group are characterized by lower values of S₂, lower values of gap size and higher values of wing angle of attack (Fig. 6C,D). The wing angle of attack vector is oriented in the opposite direction of the second moment of area and gap size vectors, which supports the results of the separate tests: birds with a smaller S₂ operate at higher wing angle of attack.

In contrast, PC1 mostly explains the variation of the normalized force, flight speed and wing speed, with all vectors pointing in a similar direction (Fig. 6C,D). In these plots, the normalized force and wing speed vectors were both close to perpendicular to the control and molt group distributions. This is in support of the above analysis that the upward escaping birds increase wing speed to enhance normalized force production, and not to control for molt.

The study of the aerodynamic effects of molt has received relatively little attention, with few studies looking at the effects of natural molt on take-off (McFarlane et al., 2016; Swaddle and Witter, 1997; Williams and Swaddle, 2003), gliding (Kleinheerenbrink and Hedenström, 2017; Tucker, 1991) or hovering flight (Achache et al., 2018; Chai, 1997). Here, we tested how experimentally induced wing molt affects the upward-directed escape flight performance of a passerine bird after 1 week of habituation, and how these birds have adjusted their flight kinematics in response to molt.

Our results show that, after habituation, birds with simulated molt gaps are able to maintain their escape flight speed and aerodynamic force production via behavioral adjustments of their flight dynamics. These behavioral adjustments consist of two aspects: an adjustment of wing morphology and a change in wingbeat kinematics.

Wing molt gaps lead to a reduction in the second moment of area of the wing, which consequently reduces aerodynamic force production during flight at low advance ratios (McFarlane et al., 2016). For our experimental birds, this molt-induced decrease in S₂ was partly compensated for by a reduction in the size of the molt gap. This was similarly demonstrated for gliding flight in a jackdaw (Corvus monedula), which modified its wing posture across molt stages in order to reduce the molt gap size (Kleinheerenbrink and Hedenström, 2017). Because there was still a molt gap present between P1 and P4, the reduction in molt gap size is not likely to be the result of feather interlocking after preening. One possibility is that the gap reduction is achieved actively via muscle tension; another possibility is a passive closure owing to the lack of support from boundary feathers once they are dropped. In any case, the result is an adjustment in wing morphology, which allows molting birds to partly negate the detrimental reduction in second moment of area caused by molt.

Because wing morphing only partly negated this molt-induced reduction in S₂, the molting birds also needed to adjust their flight kinematics to fully compensate for the reduction in flight performance. This could be achieved by adjusting both the wingbeat and the tail kinematics. None of the tested tail kinematics parameters significantly differed between the molt and control groups, suggesting that the tail did not contribute to this kinematics compensation. These results are in line with several previous studies showing a relative small effect of tail dynamics on aerodynamic force production in passerines (Johansson and Hedenström, 2009; Muijres et al., 2012), but they contradict models that show an important contribution of the tail to lift (Norberg, 1994; Thomas, 1993, 1996).

Throughout the wingbeat, the wings of molting birds operated at both higher wing speeds and higher angles of attack (Fig. 4A,B, respectively), but the average wing speed at mid-downstroke was not significantly different between the control and molt groups (Fig. 4C). In contrast, the average wing angle of attack at mid-downstroke was significantly different between these groups (Fig. 4D). This suggests that molting birds primarily increase the angle of attack of the wing near mid-downstroke to compensate for the molt-induced reduction in second moment of area.

Among the different analyzed flights, we observed variations not only in the second moment of area, but also in aerodynamic force magnitudes (Fig. 5). The latter variation in our data allowed us to determine how our upward escaping birds adjust their flight kinematics to control their aerodynamic force production. The analysis showed that aerodynamic force production was positively correlated with wing speed and tail spread, suggesting that birds use these two metrics to control aerodynamic force production.

The above conclusions are supported by our PCA, which showed that wing speed and normalized force were both primarily associated with the first PC, whereas S₂ and wing angle of attack were both primarily associated with the second and third PCs (Fig. 6; Jolliffe, 2002). The PCA therefore gives some insights into the flight control mechanisms during upward escape maneuvers. These results point to a relatively simple and modular flight control system, whereby the kinematic adjustments for varying aerodynamic forces and for molt gap control are mostly independent: to compensate for a reduction in S₂, an upward escaping bird primarily adjusts the wing angle of attack at mid-downstroke, whereas to boost aerodynamic force production, the bird increases the wingbeat-induced velocities. This modularity might reduce the burden on the neuro-muscular flight control system (Dickinson and Muijres, 2016; Lentink et al., 2007; Tobalske and Dial, 1994), but testing this would require additional research.

Molt is a complex process that involves tissue regeneration that impacts both the energy balance and behavior. Therefore, it is also important to look experimentally at the effects of flying with molt gaps separately from the physiological costs of molt (Swaddle and Witter, 1997; Swaddle et al., 1999). A few previous experiments have looked at the effect of simulated molt on flight dynamics of starlings (Sturnus vulgaris) and sparrows (Passer montanus). They showed that birds with simulated molt gaps have a slower take-off speed and impaired predator evasion and maneuverability, as well as changes in body mass and behavior (Lind, 2001; Lind et al., 2004; Swaddle and Witter, 1997; Swaddle et al., 1996, 1999). Curiously, after this initial impact, Swaddle and Witter (1997) also report a slow recovery of flight performance, which hints at a compensatory behavior, such as changes in the pattern of wing movement. Our results support this observation. The pied flycatchers used in the present study were tested 1 week after being manipulated in order to also measure the impacts of our manipulations on fitness (Tomotani et al., 2018b). This may have given the birds the opportunity to adjust their behavior to retrieve the same flight speed as the controls, and for us to assess the compensation mechanism.

The study of flight performance of molting birds may help us to understand the variation of molt strategies, for example, the segregation of molt from other annual cycle stages (Bridge, 2011; Tomotani et al., 2018a,b). Molt may force birds to avoid costly and risky activities, as the combined aerodynamics and physiological costs of molt could be too damaging to allow molt to co-occur with other stages (Swaddle and Witter, 1997). Still, molt–breeding overlap is common in male but not female songbirds (Jenni and Winkler, 1994).

Our results suggest that, after a habituation period of 1 week, early stages of molt do not negatively affect escape speed and aerodynamic force production. This is achieved by the molting birds by primarily increasing the wing angle of attack with approximately 4 deg at mid-downstroke. Nevertheless, the wing molt gaps and associated wingbeat kinematic adjustments are expected to incur energetic costs: a molt gap locally reduces lift produced by that wing section, causing a dip in the spanwise lift force distribution. This decreases span efficiency and, consequently, increases induced drag (Hedenström and Sunada, 1999; Muijres et al., 2011); because an increase in angle of attack is associated with increased aerodynamic drag on the wing, the energetic power requirement for flight is expected to also increase as a result of molt-induced wingbeat kinematic adjustments (Usherwood, 2009). Thus, the detrimental effect of molt on flight performance in passerines may not be expressed as a reduction in escape speed, but instead as an increase in energetic cost of flight.

Flycatchers forage on the wing by catching insects using rapid flight maneuvers similar to the upward-directed maneuvers that we studied (Davies, 1977). Our results suggest that primarily the energetic costs of such maneuvers are increased, and less so their swiftness. The resulting increase in the energetic costs of foraging and predator escape would force the males with molt–breeding overlap to allocate more energy to self-maintenance, and consequently less to their offspring. This notion helps to explain the observed response of our male pied flycatchers with molt–breeding overlap (Tomotani et al., 2018b): the molt group did not have a reduced fitness in terms of breeding success and next-year return rate compared with the control males, but males with simulated molt gaps did reduce parental care by visiting their nest fewer times, which the females compensated for by working harder. Thus, the increased power requirement of flight with molt gaps, forced males with molt–breeding overlap to prioritize their own survival (future reproduction) over their current reproduction success, which may come at the expense of their female partner (Hemborg, 1998, 1999; Hemborg and Merilä, 1998; Tomotani et al., 2018b).

We thank Marcel Visser for his guidance and support during the field–lab experiment, Hennie Uittenhout for constructing the flight tunnel, and Remco Pieters for developing the experimental setup and for his technical support during the flight experiments and videography analysis. We are grateful to the board of the National Park ‘De Hoge Veluwe’ for the permission to conduct our research on their property. We thank Corry Teichmann, Emma Rietveld, Femke van Kampen, Cynthia Lange, Iván de la Hera, Jip Ramakers and Henri Bouwmeester for assistance with general fieldwork and experimental manipulations, and Marylou Aaldering, Ruben de Wit and Coretta Jongeling for taking good care of our birds. We also thank two anonymous referees for their constructive comments that helped us to improve the manuscript.