We studied murres at the Coats Island, Nunavut, west colony (62°57′N, 82°00′W) during the chick-rearing period (15 July - 18 August) 1998–2011. As part of a long-term monitoring study 36 , 37 , 61 - 63 , we collected information on feeding rates, diet, attendance and chick growth rates annually. Starting in 2004, we also collected information on at-sea behavior from time-depth recorders. We found no impact of wind on time spent flying per day (see Results). Thus, in 2006 and 2009, we measured daily energy expenditure using doubly-labelled water to determine whether overall energy costs were impacted. As we found no effect of wind on daily energy expenditure (see Results), we then focused on measures directly associated with flight behavior by attaching GPS loggers and accelerometers in 2010 and 2011 to measure ground speed and wing beat frequency. Throughout, we included only birds with chicks 3–15 d old because feeding rates are constant for murres with chicks within that age range 63 . To reduce autocorrelation of weather over short time scales, we included data covering many years.

Chick-rearing birds at the Coats Island west colony forage almost exclusively within 100 km to the west of the colony 64 . Therefore, the core foraging area is bounded by the colony and the community of Coral Harbour, roughly 120 km from the colony. We recorded wind speed and wind direction daily (1800 h) at a fixed point near the Coats Island cabin, immediately adjacent to the murre colony, using a handheld anemometer (Davis Industries, Hayward, California). We also downloaded hourly wind speed and wind direction recorded at the Coral Harbour airport (www.weatheroffice.gc.ca). The daily average magnitude of wind in a particular direction at 1800 h was correlated between Coral Harbour and Coats Island camp (R = 0.58, P < 0.001). For those variables recorded at the finest time scales, correlations were strongest when weather variables were averaged over the scale of 24 hours, and we therefore used daily averages for the Coral Harbour data (e.g. ground speed: R² = 0.29 over 24 h, R² = 0.24 over 1 h). We used time of day (cos(π*Hours since solar midnight/12)) as a proxy for light levels 65 , 66 .

We augmented our more detailed information on murres with information from black-legged kittiwakes collected during chick-rearing (July 2010 and 2012) and incubation (30 May–16 June 2013) at the radar tower colony on Middleton Island, Alaska (59°27′N, 146°18′W; 67 - 69 ). The main benefit of the kittiwake dataset is that because kittiwakes do not dive, we did not need to encase the GPS units in resin and we could use lighter GPS units; we were therefore able to attach both GPS units and accelerometers simultaneously. We endeavoured to collect data on kittiwakes that complemented the data already collected on murres. In 2010, we collected data on time spent flying and feeding rates on chick-rearing birds. In 2010 and 2012 we collected data on chick growth rates. In 2013, we collected GPS-accelerometer data on ten incubating birds. We chose uniformly heavy birds (485 ± 20 g) to avoid the confounding effect of body mass on measurements of acceleration and wing beat frequency and to minimize any device effects (see next sections).

As there was a weather station associated with the Middleton Island Airport, and kittiwakes forage closer to the colony than murres, we used wind speed and direction recorded within 1 km of the colony at 20 minute intervals (http://cdo.ncdc.noaa.gov/qclcd/QCLCD). Finally, we also used circularly-transformed time of day (AKDT) as a proxy for light levels at Middleton Island.

We used chick growth rates as a proxy for fitness as chick growth rates are correlated with offspring recruitment rates (U. Steiner and A. J. Gaston, unpubl. data); virtually all chicks that hatch ultimately fledge, so fledging success is a poor indicator of fitness at our study site. Chick growth rates, however, link adult foraging behavior to the probability that a chick will recruit to the colony, which is a component of fitness. We recorded chick mass every two to three days for a subset of 25–50 chicks that were individually marked with metal bands 62 . We completed at least three continuous 24-h or 48-h feeding watches during each season (44 total watches). During the watches, we estimated visually the species and length of all fish delivered to offspring at ~30 breeding sites and used species-specific relationships between total energy and fish length to determine energy delivery rates 38 , 61 , 63 . We correlated average energy delivery rates per day (total and for each prey type) between 0600 h and 1000 h against wind speed. We were particularly interested in whether feeding rates of schooling fish, which require more above-water searching (flights and time spent flying) to locate, were impacted by wind. We used previously reported data for daily energy expenditure measured via doubly-labelled water 12 , 45 , 59 .

We attached time-depth temperature recorders (Lotek, St John’s, Canada)—5-g LTD1100 (sampling interval = 3 s; N = 140) in 2004–2007, 5-g LAT1400 (interval = 15 s; N = 20) in 2008 and 3-g LAT1500 (interval = 15 s; 2009–2011, N = 50) in 2009-2011—to the legs of parental murres and extracted time budgets (time spent flying, resting on water and resting at the colony) from the temperature log 64 , 65 . We also attached 17-g M190-D2GT biaxial recorders (sampling rate = 32 Hz; Little Leonardo, Tokyo, Japan; see 45 ) in 2010 (N = 42) and 2011 (N = 24), as well as 25-g GPS devices (interval = 15 s when ground speed above 2.8 m/s; CatTraQ™, Catnip Technologies, USA) in 2010 (N = 18) and 2011 (N = 20) to the back of parental murres using Tesa tape. Whereas the leg-mounted recorders weighing ~0.5% of body weight do not impact murre feeding rates 65 , the back-mounted recorders weighing ~1.8-3% of body weight reduced murre feeding rates 64 and imposed increased energy expenditure during flight 12 and reduced dive duration 70 . Because all birds were equipped similarly, we assume that all birds were similarly impacted by the devices.

As for murres, we used chick growth rates as a proxy for fitness in that chick growth rates are correlated with offspring recruitment rates at our study site 71 . We recorded chick mass every five days at roughly 40 nests on the radar tower colony 69 , 71 . Kittiwakes sometimes raise two chicks (we excluded any nests with three or more chicks), and we calculated chick growth rates separately for single, A-chicks and B-chicks 71 . As we only analyzed residuals (see Statistical Analyses section), we combined residuals from all three groups for analyses. We attached LTD1100 temperature recorders (same as previous section) to the legs of 30 kittiwakes for 48 h and videotaped each site from outside the tower. We used those data to calculate time spent flying, time spent on the water and time spent at the colony 72 . We also attached both 3-g Axy accelerometers (sampling rate = 50 Hz; Technosmart, Rome, Italy) and 14-g GPS devices (interval = 30 s; CatTraQ™, Catnip Technologies, USA) in 2013 (N = 20) to the back of parental kittiwakes using Tesa tape. The back-mounted GPS-accelerometers weighed ~3.5% of body weight, which is known to impact behavior in other bird species 73 - 75 . We assume that all birds were similarly impacted by the devices because all were equipped similarly.

In 2010, we also injected 0.5 mL of doubly-labelled water into the brood patch and obtained background, equilibrium (1 h of captivity) and final (48 h) blood samples from the brachial vein to measure daily energy expenditure in 37 birds 45 , 59 , 76 . Samples were timed as close to multiples of 24 h as possible to avoid circadian effects 77 . All samples were run blind to the identity of the bird and converted to values of daily energy expenditure using a single pool model with a fixed 25% evaporative water flux (equation 7 · 17: 78 ) and a respiratory quotient of 0 · 81 based on nutrient content of the diet (80% protein, 15% fat, 5% carbohydrate). Using this equation, estimates for energy expenditure based on doubly-labelled water in charadriiform birds are accurate within 2–18% relative to respirometry values from the same individual 79 .

To examine the accuracy of ground speed recorded at 15 s (murres) and 30 s (kittiwakes) intervals, we subsampled commuting (incoming/outgoing) flight tracks at longer time intervals. We recorded ground speed for longer intervals as a proportion of ground speed at 15 s intervals and fitted exponential models to the relationship between step interval and ground speed as a proportion of ground speed at 15 s. We also recorded the instantaneous ground speed as reported by the GPS logger.

Apparent (measured) ground speed decreased exponentially with step size (Figure 1). The average coefficient for the exponential function across all six individuals was 1.00087 ± 0.00014 (murres) and 1.00093 ± 0.003 (kittiwakes), revealing that at time = 0 the estimated instantaneous ground speed would be 0.087% higher than the ground speed recorded at 15 s intervals (murres) or 0.093% higher than the ground speed recorded at 30 s intervals (kittiwakes), even after accounting for random error associated with GPS measurements based on location error distribution of stationary GPS units. Thus, recording ground speed at 15 s to 30 s intervals accurately estimated ground speed with <0.1% average error. In contrast, the “instantaneous” onboard GPS speed readings averaged 0.86 ± 0.15 of the value measured at 15 s intervals (murres) and 0.96 ± 0.02 of the value measured at 30 s intervals (kittiwakes). Given that the step interval-ground speed graph showed a clear asymptote and that the instantaneous ground speed was lower than the ground speed estimated at 15 s increments, we concluded that the instantaneous “on-board” ground speed was less accurate than the 15 or 30 s estimates. The “on-board” flight direction also seemed more variable than the GPS tracks would suggest.

These results contradict those of Safi et al. 53 , who recommended the use of the instantaneous “on-board” GPS recordings, although they only compared their values to those obtained at 15 minute intervals and not necessarily for commuting flights; at 15 minutes, ground speed could be substantially underestimated within our dataset (Figure 1). Our step interval-ground speed showed a clear asymptote. In contrast, fractal movement, indicative of fine-scale searching, and which would imply that ground speed itself was dependent on scale, would be expected to have a linear relationship between step size and speed 80 . Thus, our relationship is applicable only to direct commuting flights and not to more convoluted searching behavior.

We completed all statistical analyses in R 2.14.2 (R Core Team 2013). We were interested in the response of behavioral metrics to wind speed after accounting for potential confounding variables. Therefore, for each metric used as a dependent variable we constructed a general linear mixed model with individual as a random factor using time of day (circularly-transformed), year (for parameters measured in multiple years) and wind speed in the direction of travel and wind speed perpendicular to the direction of travel as independent variables. To remove the potential for spurious correlations with wind speed due to potential correlations with date for variables at long time scales, we reported wind speed as the measured wind speed for that date – average wind speed for that date across all years for comparisons with parameters that are expected to vary with date: adult provisioning behavior, chick growth rate, daily energy expenditure and flight time. We included only wind speed and calendar date as independent variables for chick growth rate and daily energy expenditure, as those variables were calculated over >24 h periods.

We calculated the chick growth rate as the difference in body mass between subsequent measurements 81 . We calculated the average population-wide chick growth rate by fitting a linear equation to the average chick mass for a given age across all years. We then calculated the growth rate as the residual for any particular individual for any particular age 81 .

For flight speed analyses, we converted all GPS latitudes and longitudes to UTM coordinates (UN_t = UTM Northing at time t; UE_t = UTM Easting at time t) and calculated the average ground speed at time t: Speed = UN t + 1 – UN t 2 + UE t + 1 – UE t 2 0.5 / t t + 1 ‐ t t

For murres, the ground speed distribution was strongly bimodal with a minimum (<0.1%) of measurements at 5 m/s and thus we considered a flight to occur when there were six consecutive measurements > 5 m/s. For kittiwakes, we used the accelerometer profiles to determine the start and end of flights. We removed the first and last GPS measurement for each flight (which may be biased by the takeoff or landing) and calculated average ground speed for each flight.

To calculate wing beat frequency, we visually selected the ten first flights of each individual from the acceleration pattern and calculated the Fast Fourier Transform in the z component, excluding the first and last 50 s of flight (to avoid changes in wing beat frequency associated with take-off and landing; see Figures 2 and 3). We considered the frequency with the strongest maximum power in the frequency domain to reflect the wing beat frequency. We also calculated partial (murres, PDBA) and overall (kittiwakes, ODBA) dynamic body acceleration at 1 s intervals, excluding the first 50 s of flight, as an estimate of energy expenditure (L1-normalized; 44 , 45 ). During chick-rearing, murre outgoing direction averaged 101 ± 10° and incoming direction averaged 278 ± 11° (N = 38 birds; averages and standard errors generated from von Mises distribution). Consequently, we calculated wind speed in the direction of travel based on those average directions. As we focused on commuting flights, where turns were rare, we ignored energy costs associated with turning 5 , 19 . We were unable to apply a recent model for mechanical flight based on pigeon flight 82 because of transient effects associated with variation in wing beat frequency.

Ground speed decreased with the magnitude of the component of wind speed against the direction of flight (Table 1, Figure 4). Crosswind speed did not affect ground speed (Table 1). In the absence of a tail or headwind, the best-fit equation generated a prediction for average ground speed of 16.6 m/s (murres) or 10.6 m/s (kittiwakes; Figure 4). Air speed was lowest with a tailwind and highest with headwind, and higher with a crosswind than without a crosswind (Figure 5). Wing beat frequency, averaged across all inbound or outbound flights for each individual, decreased with the magnitude of the component of wind speed in the direction of flight (assuming outgoing direction of 101 ± 10° and an incoming direction of 278 ± 11° for murres). Wing beat frequency was higher for inbound murres, carrying fish, than for outbound murres (Figure 6).

¹Actual direction of flight was unknown, so we assumed 101 ± 10° for outbound direction (average from birds equipped with GPS).

²Average wind speed, rather than crosswind and tailwind components, used as independent variable. To account for the confounding effect of date on wind, average windspeed for each day was calculated as residual from the average value for a given date across all 10 years.

³Hatch date (murre: F_1,2876 = 10.9 P = 0.001; kittiwake: F_1,686 = 2.25, P = 0.13) was also included as a covariate.

Values in bold are statistically significant at the P < 0.05 level and values in italics are significant at P < 0.002.

Chick-provisioning rates (±SD) averaged 0.33 ± 0.12 feeds h⁻¹ for kittiwakes and 149 ± 25 kJ d⁻¹ for murres while daily energy expenditure during chick-rearing averaged 1.96 ± 0.28 kJ d⁻¹ g⁻¹ for kittiwakes and 2.04 ± 0.53 kJ d⁻¹ g⁻¹ for murres. Energy delivery rates decreased with wind speed (Table 1, Figure 7). Chick growth rate, daily energy expenditure and time spent flying were independent of wind speed (Table 1, Figure 7). For murres, energy delivered in the form of amphipods Parathemisto libellula increased with date (t₈₇₀ = 4.00, P < 0.0001) and headwind speed in the average direction of commuting (101 ± 10°; t₈₇₀ = −4.96, P < 0.0001; total R² = 0.26; Figure 7). In contrast, energy delivered in the form of two schooling fish, cod Boreogadus saida and sand lance Ammodytes hexapterus, decreased with date (cod: t₈₇₀ = −2.81, P < 0.0001; sand lance: t₈₇₀ = 2.70, P < 0.0001) and with headwind speed in the direction of commuting (cod: t₈₇₀ = −4.28, P < 0.0001; total R² = 0.22; sand lance: t₈₇₀ = −3.51, P < 0.0001; total R² = 0.18; Figure 7). Energy delivered for all other fish species was independent of wind speed.

Ground speed increased linearly with the component of wind speed in the direction of flight. There was no effect of crosswinds, demonstrating that birds completely compensated for crosswinds 47 - 54 . Wing beat frequency correlated with dynamic body acceleration (Figure 3a,b), as wing beat frequency is, by definition, a measure of how quickly the wing accelerates through the wing beat cycle. Variation in this relationship is likely due to variation in body mass/load, as birds carrying heavy loads will have higher wing beat frequency (work harder) but lower body acceleration (Newton’s second law). Dynamic body acceleration is correlated with energy costs in active animals, such as seabirds 44 , 45 , 83 , 84 , so it is not surprising that wing beat frequency is a proxy for instantaneous energy costs in flight 43 , 85 . For every increase in headwind of 1 m/s, ground speed decreased by only 0.5-0.6 m/s; birds increased wing beat frequency (power output) to compensate for decreasing ground speed. Thus, the increase in ground speed is not equal to the increase in wind speed and the ground speed is decreasing relative to what it would be if it were completely controlled by wind speed alone with no accommodation by the bird. The average ground speed at 0 m/s wind speed was below most past reports of murre/kittiwake ground speed (Table 2). The low flight speed may, in part, be an artifact of increased drag associated with the 25 g GPS loggers (17 g accelerometers increase murre flight costs by 21%; 12 ). Indeed, it is possible that the effects throughout this paper were exacerbated due to the increased load and drag associated with instrumentation.

Wind had a strong impact on the flight behavior and energy expenditure during flight of both species. The birds were able to compensate for crosswinds. In contrast, although they increased wing beat frequency in the face of increased headwinds, they were only partially able to compensate for the effect of increased headwinds.

Wind had a smaller impact on energy delivery rates than on flight costs and did not appear to influence offspring growth rates at all. Thus, parental birds buffered the effect of variation in wind so as not to impact their reproductive success. This is not simply because strong tailwinds on the inbound trip compensate for strong headwinds on the outbound trip; due to the nonlinear relationship between energy costs and wind speed, a return trip with 5 m/s outbound headwind and inbound tailwind imposes a 6% increase in overall costs for murres compared with the same trip with no wind. We suggest that the increased chick feeding rates during periods of high food accessibility (moderate or low winds) is due to parental murres buffering changes in food accessibility in such a way that they make up for periods of low food accessibility (high winds); parental birds may have also switched to alternative prey or sacrificed self-feeding during high winds to maintain provisioning rates.

In contrast to other studies of charadriiform seabirds 20 , 60 , we found no effect of wind speed on daily energy expenditure. Likewise, using a subset of the data presented here, and measuring wind speed twice a day at the colony, Elliott et al. 59 found no effect of wind on daily energy expenditure in murres. Seabirds may alter their behavior during windy days to minimize the effect of high winds; on windy days kittiwakes use formation flocks and fly at low altitudes, where wind speed is less due to the effect of the boundary layer 86 . Birds may also choose flight paths that minimize the negative effect of wind 19 .

We were surprised that there was no link between several days of windy weather and reduced chick growth rates as previous studies have found measurable effects of wind on the reproductive success, attendance, feeding rates and chick growth rates of piscivorous birds 13 - 19 , 23 - 29 , 96 , 97 , contra 98 . This implies that adults of both species were adjusting their behavior to compensate for poor weather. Temporal scale likely explained some of the variation as extreme wind speeds, measurable in data collected at the scale of seconds or minutes, was not maintained for days on end. Nonetheless, there was a noticeable effect on feeding rates and flight behavior even at the reduced wind speeds (~ ± 4 m/s) observed within the chick growth dataset. In support of a weather effect on above-water searching, weather in our study was unrelated to energy delivered in the form of prey items that require underwater searching but not above-water searching (e.g. benthic prey occurring well below the surface layer affected by weather). In contrast, energy delivered in the form of schooling prey that required above-water but not underwater searching (Arctic cod and sand lance) decreased on windy days. We suggest that during inclement weather (high winds), birds were unable to track schooling fish due to high wave height or difficulty flying, and switched to more abundant, but lower energy content 38 , amphipods and benthic fish. The effect of date was also apparent for several of these parameters as murres deplete food sources over the course of the season and switch to alternative, less profitable food sources 63 .

If parental birds have the capacity to increase feeding rates to offspring during poor weather, why then do they not do so during good weather? One possibility is that they use periods of good weather to replenish their own reserves, which are larger and less likely to be exhausted during poor weather. Furthermore, the gain to the parent in terms of increased probability of chick survival may be a decelerating function of energy delivery rates; a chick that starves represents zero reproductive success but a very heavy chick may actually have low survival if it has difficulty fledging 99 , 100 . Likewise, both the chick and adult are only able to assimilate a certain amount of food each day 12 , and so there is no reason to catch excess food. Finally, during periods of low food the parent may divert more energy towards the chick simply in response to increased begging, and that response may not be linearly tied to ultimate costs and benefits 100 - 102 .