Study area

The study area includes a shallow water coastal area, Nepisat (64°10′N, 51°53′W), at the mouth of the Nuuk fjord system (Fig. 1). Within the Greenland Open Water Area, the Nuuk fjord and coastal area constitutes key wintering areas for both King Eiders S. spectabilis and Common Eiders, with estimated numbers of c. 12 000 and 57 000 birds, respectively (Merkel et al. 2002). The municipality of Nuuk inhabits a human population around 14 000, of which 2486 persons were licensed as recreational hunters and 132 as commercial hunters in 2001. In the period 1993–2001 the hunters bagged on average 11 579 ± 1071 eiders (both species) during fall, winter and spring, or approximately 15% of the total annual eider kill toll in Greenland (The Greenland Home Rule Department of Hunting and Fishing). The Nepisat study area is primarily occupied by Common Eiders, feeding on a mixture of soft-bottom and hard-bottom prey species, primarily bivalves and polychaetes (Merkel et al. 2007). The Common Eiders in the Nepisat area are primarily diurnal feeders (Merkel et al. 2008). According to systematic counts in the study area in 2002 (8 and 26 February, 30 March) 2.4% of the eiders in Nepisat were King Eiders, however, in the following observation protocol we did not distinguish between the two species. Up to c. 8000 eiders was observed during the study period. The Nepisat area is located only 6 km from Nuuk city and as a consequence it is often exposed to high levels of boating activity (Merkel 2004b, Merkel et al. 2008). Tide levels change approximately 4.5 m from low tide to high tide.

Figure 1.

The capital Nuuk and the Nepisat study area, Southwest Greenland.

Behavioural observations

In the winter of 2002 we made behavioural observations from dawn to dusk (i.e. including the twilight periods before sunrise and after sunset) from a small island (camp island) in the Nepisat area at nine dates: 26–27 January, 2–6 February, 31 March, and 1 April. Observations were made from a hide at an elevated point (∼10 m) using a Leica Televid 77 telescope (32×). We recorded eider activity in 12 predefined transects, which were laid out in 10° intervals, starting at 0° (North). Transect widths were fixed according to the field of vision of the telescope when pointing at a given angle (0°, 10°, 20° … 110°). Transects had a mean length of 2249 m (range 1510–2500 m) and covered an area of 3.9 km². The size of the total area that could be observed from the observation point was c. 12 km². Eiders were primarily at the inner 1300–1500 m of the transects, at which we measured water depths (close to high tide) between one and 40 m (mean 19 m). Beyond 1500 m depths gradually increased and averaged 48 m (range 14–181 m). For each transect we counted the total number of birds and subsequently observed randomly selected groups of birds (usually 1–3 groups of 5–30 individuals/transect) continuously for 75 s to classify their behaviour into three categories: feeding (diving and inter-dive pauses), locomotion (swimming or flying), and other activity (resting, comfort and social behaviour). The number of birds engaged in feeding (including inter-dive surface time) was estimated by counting the number of birds disappearing from the surface during the scan period. The duration of the scan period (75 s) was based on sequences of focal-animal sampling (Altmann 1974), which showed that inter-dive pauses rarely (5%) exceeded 75 s (mean 48.4 ± 1.3, range 22–114 s). Feeding did not include the longer resting and diet processing phase (8–17 min, Guillemette et al. 1992) which complete a feeding bout. We surveyed all transects in sequence each hour throughout each day. Eider activity was observed at night or in twilight periods using a Swarovski NC2 night vision scope (3×). We walked the coastline of the camp island, on average three times per night, and included observations of a roosting area located just behind the camp island. The results of the night observations are reported elsewhere and only referred to in the discussion (Merkel et al. 2008). Between dawn and dusk the study area was kept under observation for disturbances, i.e. boats. We determined the track of each boat as it passed through the study area by measuring its angle and distance from the hide at regular intervals. Similarly, locations of flocks were determined from angle and distance measurements, using laser binoculars (Zeiss Harlem II 8×56 and Leica Geovid 7×42 BDA). Eider flocks and boat routes were subsequently mapped in GIS, allowing us to calculate the distance between boats and eiders. Time is referred to as local time (UCT-3 h). Information gathered after 30 March is corrected (minus one hour) for daylight saving time (summertime) to make comparison possible throughout the observation period.

Data analysis

To analyze the influence of different variables on the feeding activity of Common Eiders we used the PROC GENMOD procedure in SAS, applied the logit transformation and assumed a binomial error structure (SAS 2001). The genmod package allows groups of e.g. birds to be the observation unit and not the individuals, in contrast to traditional logistic regression. Our response variable for feeding was the number of birds engaged in feeding relative to the total number of birds observed in the flock. We corrected for overdispersion by scaling the error variance (quasibinomial error structure) and by applying the F-test of deviance when comparing different logistic regression models. Overdispersion occurs when the residual deviance is substantially larger than degrees of freedom, and was in our case maybe caused by social interactions among birds, i.e. the behaviour of one bird may be influenced by others. Prior to logistic regression analyses we explored the independence of observations using an autocorrelation procedure in the statistical package R. For each day, correlation coefficients r were calculated between consecutive observations and between observations with time lags up to c. 20 observations. For consecutive observations correlation coefficients varied between -0.24 and 0.35 and were significant in four of nine dates. In one of these four days the autocorrelation continued to be significant for observations with time lags of two and three observations (r = 0.31 and 0.19). Apart from this there was no meaningful detection of autocorrelation in the data, and we therefore chose not to model autocorrelation in the logistic regression analyses.

Initially, we considered 7 independent factors as explanatory variables for feeding activity. These were: the tide level (calculated in meters on an hourly basis); the nearest twilight period (time in min to the nearest twilight period, either dusk or dawn); the chill factor (from observations of temperature and wind speed); abundance (the number of birds present in a given transect) ; the passed undisturbed time since the last disturbance (in min); the disturbance distance to the last disturbance (to the nearest 10 m); and the number of repeated disturbances (number of boats within a given distance and time period). Continuous variables were organized into discrete variables to reflect expectations that changes in e.g. abundance, disturbance distance or repeated disturbances had larger impact on feeding activity at the low end of the sampling scale compared to the high end. The intervals used for discrete variables (see Table 2) were based on visual inspections of data and general considerations not to generate badly skewed datasets.

When constructing the logistic regression models we followed the procedure described in Hosmer and Lemeshow (1989). First, we analyzed each explanatory variable separately to identify candidates qualified for multiple logistic regressions (P < 0.2, F-test). Subsequently, multiple logistic regressions were applied to all qualified main effects plus all first order interactions. Second order interactions (or higher) were not considered, except in one case, where two first order interactions included the same variable. The model was subsequently reduced by removing non-significant factors one by one according to the Type 3 F-test of deviance. A significance level of 10% was applied in the model reduction procedure (Hosmer & Lemeshow 1989).

To make the maximum use of data, we constructed two logistic regression models (Table 2). Modell is based only on days that included human disturbances and included the variables undisturbed time, disturbance distance and repeated disturbances, whereas Model 2 also includes days with no human disturbances and treats disturbance as presence or absence. The level of feeding activity is presented as the observed proportion of individuals actively feeding or as the change in feeding odds (odds ratio) when one variable changes (e.g. from low tide to high tide) while the others are fixed.

The testing of differences between two proportions, representing the means of two sets of proportions, we compared by a two-sample t-test after arcsine transformation (Zar 1999). For correlation analyses we used the Pearson product moment correlation coefficient r. To test for differences in mean number of birds over time we used one-way ANOVAs. Assumptions of normality and homogeneity of variances were tested using the Andersen-Darling test and Bartlett's test, respectively. For tests other than logistic regression analyses, we used a significance level of 0.05. Average numbers are reported as means ± SE.

Overall activity

As counted by the hour the study transects had a mean sum of 844 ± 93 birds present between dawn and dusk (n = 82 counts, range 69–1696). There was a steady increase in the number of eiders around dawn and the birds gradually moved out of the area again around dusk. Despite fluctuations within days, the mean number of birds did not change significantly between 9:00 and 16:00 when averaged for all days (814 ± 46; ANOVA, F_7,62 = 0.87, P = 0.53). The only detected change was a seasonal change: significantly fewer eiders occupied the study area in late March/early April compared to January/February (386 ± 27 vs. 956 ± 42; n = 28 vs. 54; ANOVA, F_1,80 = 84.7, P < 0.001).

Figure 2.

The daily number of disturbances (boats) in the Nepisat study area and the daily mean (±SE) proportion of birds engaged in feeding or locomotion among Common Eiders wintering in Nuuk, Southwest Greenland.

As calculated on a daily basis there was a decrease in the feeding activity as the number of human disturbances (number of boats) increased (r = -0.77, P = 0.015). More and more time was instead allocated to locomotion activity as the number of disturbances increased (r = 0.79, P = 0.012). Approximately, the observed locomotion activity tripled when comparing days with few or no disturbances with days having the highest number of disturbances (Fig. 2).

Table 1.

Logistic regression statistics based on test of deviance on Common Eider feeding activity on days with human disturbances (Model 1) and on days with and without disturbances (Model 2). See Table 2 for details on parameters.

Birds exposed to human disturbances (Model 1)

The initial simple logistic regression step showed that there was a tendency for higher feeding activity at the low end of the chill factor scale (P = 0.02). However, this result was caused by observations on a single very cold day, and the number of observations were too few (n = 25) to adequately represent the other variables in the proceeding analyses. All other explanatory variables qualified for multiple logistic regression (P < 0.2). A stepwise reduction of model 1 (P > 0.1) resulted in the removal of the twilight factor and all first order interactions except for one, the undisturbed time × abundance interaction term (Table 1). Based on the F-test of deviance, the undisturbed time factor was the single most important factor in explaining the variation in the estimated feeding probability of the eiders, partly as a main effect partly through the interaction with abundance. Birds that were recently disturbed (<1 h vs. 1–5 h) were less likely to feed (Table 1), and this tendency was stronger at high abundances (<100 vs. >100 birds per transect). Thus, birds gathered in large flocks after a recent disturbance had very low feeding odds, while birds in smaller groups approached normal feeding activity much faster. The variables disturbance distance and tide both came out as significant main effects. The odds for feeding at low tide was a factor 2.1 times higher than at high tide and there was a gradual increase in feeding odds as the distance between the observed eiders and the boats increased (Table 2). In addition, there was an effect of repeated disturbances, with feeding odds being a factor 2.2 lower than for birds not disturbed or disturbed only a single time (Table 2). When repeated disturbances occurred within a relatively short period of time and distance (<1 h, <1 km) there seems to be a cumulative effect on the feeding activity, approaching zero when disturbed three times within one hour (Fig. 3, lower curve). In contrast to this, the feeding activity observed 1–5 hours after one or several disturbances appeared to be rather constant and independent of the number of boats (Fig. 3, upper curve). At disturbance levels of 1–3 boats/h significantly more time was allocated to locomotion activity (0.20 ± 0.04) compared to periods 1–5 h after the last disturbance (0.07 ± 0.02; t₁₃₂ = 2.72, P = 0.007).

Table 2.

Parameter estimates for logistic regression models on Common Eider feeding time allocation on days with human disturbances (Model 1) and on days with and without disturbances (Model 2). Variables that are part of an interaction term are not listed as main effects. The three columns to the right show the observed proportions of individuals feeding (mean ± SE) corresponding to the variables included in the logistic regression models.

Figure 3.

The observed feeding activity (mean proportion of individuals ± SE) for Common Eiders in the Nepisat study area in relation to the number of disturbances (boats; ±SE), within the specified distance and time interval since the last disturbance.

Disturbance vs. no disturbance (Model 2)

The mean feeding activity observed on days with no human disturbances was 0.46 ± 0.03, and this value was significantly smaller than the feeding activity observed in undisturbed periods on disturbed days (0.59 ± 0.03, t₂₅₈ = 2.39, P = 0.017). The mean proportion of eiders engaged in locomotion was 0.16 ± 0.02 on days with no disturbances.

Logistic regression on the full dataset, including days with no human disturbances, differed from Model 1 in several ways. The variable twilight remained significant throughout the analysis and a dichotomous variable for disturbance (Yes/No) interacted with both twilight and tide (Table 1). The impact of tide was highly significant during midday (twilight 2–6 h) when birds were undisturbed, whereas this tendency was levelled out when exposed to disturbances (Table 2). The odds for feeding were lowest when high tide occurred around midday on days with no human disturbances. At all other periods feeding odds were higher in the absence of human disturbances (Table 2).

Evidence of disturbance impacts

We observed a decrease in feeding activity in the study area when the daily number of boats in the study area increased (Fig. 2). This indicates that boating generated disturbances that induced direct costs to the eiders in the form of lost feeding opportunities, including the time needed to relocate the same or a similar ideal feeding spot. In addition, as feeding activity went down locomotion activity went up and induced increased energy expenditure due to additional movements. This was even more evident on a short temporal scale (Fig. 3). As the number of boats approached three per hour the feeding activity was close to zero and locomotion activity nearly tripled compared to undisturbed periods. Increased energetic cost due to risk avoidance is the most common consequence reported for waterbirds in studies of human disturbances (see review by Madsen & Fox 1995). A factor less commonly considered, and indeed quantified, is the extra cost of feeding when conditions are less favourable (but see Béchet et al. 2004). Our results indicate that eiders avoided feeding during high tide if undisturbed, and when undisturbed they intensified feeding in the morning and again prior to the dusk twilight period (Table 2). In contrast, it appears that when disturbed, the eiders attempted to compensate for lost feeding opportunities by feeding more at high tide and during midday. On days with human disturbances the feeding activity was significantly higher in undisturbed periods compared with days that had no human disturbances at all. The preference to feed more at the start and at the end of the day is probably related to the fact that eiders in the Nepisat area normally do not feed at night (Merkel et al. 2008). The cost for not doing so is not straightforward, since it will depend on feeding conditions the remaining period of the day. However, the cost of feeding more at high tide levels will clearly compromise the energy budget. Work against buoyancy and drag are the dominant components in the costs of feeding in diving birds (Stephenson et al. 1989, Loworn & Jones 1991). In our study area the tide level changed approximately 4.5 m from the low tide to high tide, corresponding to a mean water depth of c. 14.5 m at low tide and c. 19 m at high tide (the inner 1500 m of the transects).

Disturbance vulnerability

Bell & Owen (1990) referred to the Common Eider as particularly sensitive to hunting disturbance. The responses to human disturbances observed for the eiders in our study area may be used as an argument to support this. However, as argued by Gill et al. (2001) and Stillman et al. (2007), the response or the amplitude of the response to human disturbance may not be directly related to disturbance vulnerability. They argue that species showing a strong avoidance of human disturbances may not be those most seriously affected. Strong avoidance may be interpreted as having the capacity to respond, whereas no response may reflect that birds are already stressed and therefore need to accept the higher risk of ignoring the threat. In the same way Gill et al. (2001) argues that birds that do not leave a disturbed site may not do so simply because they have no alternative site to go to. They may therefore be more severe impacted than those leaving the disturbed area. The same argument can be used to explain the relationship between flock size (abundance) and the undisturbed time factor in our study (Table 2). We found that eiders in small flocks resumed feeding activity much faster than eiders in larger flocks. This does not necessarily mean that large eider flocks are more severely affected by disturbances than small flocks, rather that Common Eiders have a tendency to gather in large flocks when they are not feeding. The tendency for non-feeding birds to gather in large flocks is a well known behaviour for eider ducks in Greenland (Merkel et al. 2002) and elsewhere (Goudie et al. 2000).

An alternative indicator for disturbance impacts is the dynamic behavioural response, which in contrast to the direct response (e.g. avoidance distance) considers the response on a longer time scale (e.g. feeding time allocation on a daily basis). This tool has not been widely used in impact studies of human-induced disturbances (but see Béchet et al. 2004). In our case the analyses of feeding activity clearly show that eiders in the Nepisat area have the capacity to adapt to human disturbances, as indicated by the change in feeding time allocation when disturbed. What remains unanswered is whether the observed change in behaviour fully compensates for the unbalanced energy budget imposed by the lost feeding opportunities and the additional work effort. If nocturnal feeding is a durable strategy to balance the energy budget, then it would appear that they possess a large potential for compensating even more if needed. Among days where we know the history about daytime events, the eiders only had to rely on nocturnal feeding a single time (Merkel et al. 2008). If, on the other hand, nocturnal feeding as a supplement to diurnal feeding is durable only as a short-term solution, then the incidence of nocturnal feeding could indicate that the number of disturbances on this particular day were close to the limit of what they were capable of compensating for. We tend to believe that the last suggestion is the more likely one. If supplemental feeding at night was a competitive strategy we would expect this to occur more often. Although we realize that we have studied the eiders for only a limited period of time, we know from observations in remote fjord habitats east of Nuuk that Common Eiders are capable of practicing nocturnal feeding as an exclusive foraging strategy (Merkel et al. 2008). However, feeding conditions are different in these areas, in such a way that visual cues are probably less important (Merkel et al. 2008). Eiders in our study area spent between 43% and 66% of daylight hours feeding if not frequently disturbed. This is at the very high end of that reported on Eider feeding time during winter (Goudie et al. 2000) and gives support to the interpretation that the eiders were at the threshold limit for maintaining energetic needs at the day with the highest disturbance level (after which nocturnal feeding was observed). Although the net-consequences of the disturbances were not quantified in this study, the factors identified as important for eider feeding activity should be considered when dealing with eider management in Southwest Greenland or similar wintering areas highly exposed to human disturbances.