Measuring the food left in experimental trays when study organisms cease feeding on them [so-called giving-up densities (GUDs)] is an accepted technique for assessing predation risk and disturbance. However, in natural settings, accessibility and energetic harvest costs may vary spatially, and GUDs may be confounded. In this study, we assessed whether at a heterogeneous site, non-experimental GUDs could reveal the effect of disturbance. We measured initial and GUDs of tubers of Fennel Pondweed Potamogeton pectinatus, which form here the exclusive food source of Bewick's Swans Cygnus columbianus bewickii during their migratory stopover. We calculated giving-up net energy intake rates (GUNs) by correcting for biomass accessibility and foraging costs. The study area was a shallow lake consisting of nine creeks, three of which were open to the public (i.e. disturbed). GUDs in creeks open or closed to the public were not significantly different. In contrast, GUNs were generally higher in creeks open to the public, after correcting for initial net energy intake rate. The results suggest that natural GUDs may not reflect the effects of disturbance in heterogeneous habitats. When environmental differences are large within a site, GUNs may be a useful alternative as a behavioural indicator.
The growing popularity of outdoor recreational activities has brought about a conservation interest in determining the influence of these activities on the persistence and distribution of animal populations, and how recreation might be managed to permit biodiversity conservation (Boyle & Samson 1985, West & Caldow 2006, O'Connell et al. 2007). Many forms of recreation may indirectly impact on birds through disturbance, affecting the distribution, fitness, survival or reproductive success of animals. These effects are often not obvious, as they act through reduced access to resources, such as food supplies, and resting or breeding sites (Hockin et al. 1992, Gill et al. 2001, West et al. 2002, Weston & Elgar 2005, Rode et al. 2006). Nevertheless, it is becoming increasingly clear that disturbance may significantly influence population and community dynamics (Blanc et al. 2006, Cresswell 2008).
Birds are often easy to observe, and alert- and flight-distances are commonly used to quantify the response to human disturbance (Blumstein et al. 2005, Laursen et al. 2005, Rees et al. 2005, Møller 2008). However, such responses may tell little about the impact of the disturbance (Gill et al. 1996, Cardoni et al. 2008). Moreover, when individuals learn to avoid regularly disturbed areas, direct behavioural responses may be difficult to measure, and hence alert- and flight-distances are not helpful (Blanc et al. 2006).
Expressing the risk of predation from the animal's point of view is considered to be a major challenge in studying habitat selection (Sinclair et al. 2006). Animals often respond to humans in the same way as to potential predators, and hence human disturbance can be studied in a similar fashion to predation (Gill et al. 1996). At foraging sites, reduction in resource utilization is often used as an indicator of the perceived risk of predation. Animals trade off safety with energy intake: the amount of food left behind by a forager, the so-called giving-up density (GUD), is higher in food patches with a higher predation risk (e.g. Brown 1988, Thorson et al. 1998, Hochman & Kotler 2007). Therefore, quantifying leftovers in experimental food trays exposed to varying levels of safety risk has become an accepted method to investigate the effects of human disturbance (Bowers & Breland 1996). This approach has most often been applied in studies of animals that are difficult to observe, such as nocturnal mammals (see review in Brown & Kotler 2004).
Unfortunately, it is not always possible to use experimental food trays. Although by using GUD measurements in the field the actual habitat use of animals is assessed (Cresswell 2008), non-experimental GUDs are problematic because they are not solely influenced by the presence or absence of predators. The gross intake rate, accessibility of food and the energetic costs of foraging are also important in shaping GUDs (Brown 1988, Nolet et al. 2001). In a natural setting these may vary spatially, for example when part of the food of waders is buried too deep in the sediment to be reached (Zwarts & Wanink 1993) or when diving ducks spend more energy while foraging on food items in deeper water (Houston & Carbone 1992, Leeuw et al. 1999). When not accounting for such environmental conditions, natural GUDs cannot be linked directly to human disturbance. Alternatively, the use of net rate of energy intake incorporates all the differences in food density, intake rate, energetic costs, predation risk and missed opportunity costs of foraging (i.e. the value of alternative activities; Brown 1988). According to the marginal value theory, if the animal's goal is to maximize the net rate of energy intake, in the absence of predation the giving-up net energy intake rate (GUN; i.e. the net energy intake rate at the moment the animal leaves the patch) is expected to be the same over the entire habitat (Charnov 1976, Brown 1988, Nolet & Klaassen 2009). Therefore, GUN differences between patches with or without predation risk may be used to quantify the effect of predation (or disturbance) even in non-experimental studies. However, the ease of relying on simple measurements such as GUDs is attractive compared with the use of net energy intake, which involves the additional estimation of food accessibility and foraging costs. Therefore, it is unclear whether more complicated models are really needed.
In this study we look at GUDs and GUNs of Bewick's Swans Cygnus columbianus bewickii foraging on below-ground tubers of Fennel Pondweed Potamogeton pectinatus measured at the scale of a whole stopover site in the Netherlands. Earlier, Nolet and Klaassen (2009) found that Bewick's Swans within an undisturbed creek of the Lauwersmeer quitted foraging at the same GUN value throughout different sediment type and water depth patch type combinations, in accordance with the marginal value predictions. As Bewick's Swans are known to flee from boats (Mori et al. 2001), we compared measured GUDs and GUNs in creeks open or closed for boat traffic, as a measure of human disturbance. We expected Swans to prefer foraging in non-disturbed areas (i.e. closed to boat traffic), and thus to find higher GUD and GUN values in creeks open to boat traffic. In addition, we investigated whether decreasing distance from boat channels affected habitat use. Bewick's Swans forage both day and night, with peak intensity during the night (Nolet & Klaassen 2005). As the Swans are known to maximize their intake rate at the Lauwersmeer (Nolet et al. 2006b), we expected that they would try to compensate for feeding opportunities lost to human disturbance during daytime by using the disturbed sites more intensively at night (Blanc et al. 2006). In order to assess whether such compensation occurred, we compared the foraging duration of Swans inferred from daytime observations with the foraging duration predicted from measured food consumption (encompassing day- and night-time foraging).
The study was carried out in the Lauwersmeer (53°21′N, 6°12′E), a shallow lake in the northern part of the Netherlands, which is one of the last stopover sites of Bewick's Swans on their autumn migration. In the shallow parts of the lake the Swans refuel for a few weeks by feeding on below-ground pondweed tubers (Beekman et al. 1991), maximizing their net energy intake rate (Nolet et al. 2006b). The Lauwersmeer comprises nine creeks, all with different combinations of food density, water depth and sediment type, which affect the energetic profitability of a foraging site (Nolet et al. 2001, 2006a, Nolet & Klaassen 2009). Shooting is prohibited throughout the whole area. The lake is a popular recreational area, with some creeks open to the public (Fig. 1). In October–November, the period that Bewick's Swans are present in the Lauwersmeer, on average 4.0 (± 7.9 sd) boats per hour pass through these creeks during the daytime (weighted mean of observations in 2007 and 2008 on altogether 95 observational periods of 15 min to 1 h; S. de Klepper & M.S. Franken unpublished data). Besides boat traffic, disturbance by cars, pedestrians (occasionally with dogs) and cyclists also occurred in the three creeks of the Lauwersmeer open to the public. As Bewick's Swans are known to show a flight response to approaching boats (Mori et al. 2001), we categorized the creeks of the lake based on their accessibility to the public.
In October and November of four years (2005–2008), Bewick's Swans were counted every day on each creek, using a 20–60× telescope. As giving-up net energy intake rate GUN is affected by initial tuber density Di, giving-up tuber density Df, water depth d and sediment type s (Nolet et al. 2001, 2006a, Nolet & Klaassen 2009), each of these variables were measured at 90 locations each year (Fig. 2). Locations were chosen in a stratified random manner, at 10 randomly placed points per creek with a minimum distance of 100 m between locations in order to reduce spatial autocorrelation in tuber density (Nolet & Mooij 2002). Twelve to 16 cores (0.1 m in diameter) per sampling point were taken from the upper 0.35 m of the sediment, within (2007–2008) or around (2005–2006) 1-m2 plots (as above-ground biomass was removed in the summer for another purpose in these later years; Gyimesi et al. 2012). The extracted cores were washed through a metal sieve of 0.003 m mesh size to separate the tubers from the sediment. The tubers were stored in labelled plastic bags at 4 °C until they were dried in the laboratory for ≥ 48 h at 70 °C.
At each sampling point, water depth was measured and a sediment sample was taken for particle-size distribution analysis (Malvern analyser: Mastersizer 2000). Because water levels varied slightly during these measurements, water depths were expressed relative to the water level pursued by the lake's water manager (Waterschap Noorderzijlvest), yielding standardized water depth d. Sediment particle size was classified in six categories, which were submitted to a principal components analysis (PCA). Sediment type s was determined based on the extracted first component values: negative values were defined as sandy sediment, and positive values as clay (cf. Nolet et al. 2001).
GUNs were calculated by applying the patch-type-specific functional response of Bewick's Swans foraging on Fennel Pondweed tubers (Nolet & Klaassen 2009):
where GUN is the giving-up net energy intake rate (J/s or W) as a function of water depth and sediment type, ϕ(s) is the proportion of foraging time spent feeding (with head under water) as a function of sediment type, q is the assimilation efficiency of Fennel Pondweed tubers by Bewick's Swans and e is the energy density of the tubers (J/g). The functional response further includes the variables a(s), the attack rate (m2/s) as a function of sediment type; th, the handling time (s/g) (i.e. the time required for a forager to extract the food item from the substrate and consume it); Df(d), the giving-up tuber density, considering only the accessible part of the biomass, which is a function of water depth; and c(d,s), the energetic costs of foraging as a function of water depth (i.e. head-dipping in shallow water or up-ending in deep water) and sediment type. The values of the parameters and the calculation of the accessible tuber biomass have been described by Nolet and Klaassen (2009).
To reduce the effect of sampling points that Swans did not visit, we only considered the sampling points in our analysis where the final tuber density (GUD) was lower than the initial tuber density (n =192). Initial- and giving-up tuber densities were slightly skewed, and hence were square root-transformed to achieve normality. On the other hand, initial- and giving-up net energy intake rates were highly skewed and required ln transformation. First, however, all values needed to be positive (which was not always the case); this was achieved by adding 100. By increasing all values by a constant, differences remained the same, and therefore did not affect the analyses.
Initial values of sediment type, water depth, initial tuber densities and net energy intake rates were compared between creeks open and closed to the public by GLMs with ‘year’ as a random categorical predictor and ‘creek’ as a random categorical factor nested in the factor ‘open or closed to the public’. Subsequently, differences in GUDs and GUNs between creeks open and closed to the public were tested using GLMs (type III sum of squares), with ‘year’ as a random categorical predictor, ‘creek’ as a random categorical factor nested in the factor ‘open or closed to the public’, and ‘initial tuber density’ and ‘initial net energy intake rate’ as continuous predictors. In the final models, non-significant terms were removed, with the exception of the random effects year and creek.
Distance between a sampling point and the nearest boat channel was measured using arcmap 9.1 (1999–2005 ESRI Inc., Redlands, CA, USA) for points in creeks open to boat traffic and where the initial tuber density was higher than the final tuber density (n =60). In a subsequent GLM analysis (type III sum of squares), with ‘initial tuber density’ or ‘initial net energy intake rate’ as continuous predictors and ‘year’ and ‘creek’ as random categorical predictor, we tested the effect of distance to the boat channel on GUDs and GUNs, with all possible interactions where the factor ‘distance to boat channel’ was included. In the final models, non-significant terms were removed one by one, starting with the interaction term with the highest P-value and then non-significant main effects.
Due to the large number of tubers per unit time that Swans encounter in their foraging pits, Swans can probably estimate patch quality well, even at relatively low food densities, allowing them to approach prescient foraging behaviour (i.e. a forager that has an accurate knowledge of the patch quality immediately upon arrival; Valone & Brown 1989, Klaassen et al. 2006). Therefore, they might quickly abandon patches with energy intake rates below the threshold (k) that maximizes long-term average intake rate, whereas patches of better quality are harvested until this threshold is reached (Olsson & Brown 2006). We estimated the threshold giving-up energy intake rate k for creeks open and closed to the public by fitting least-squares solutions to the following set of equations:
where neii is the initial net energy intake rate.
Tuber consumption by Swans was calculated by subtracting the giving-up density from the initial tuber density, after correcting for the area (10.6%) that had been probed during our initial sampling. Based on the realized tuber consumption and the patch-type-specific functional response of Bewick's Swans on sandy and clay soils (fig. 4 in Nolet & Klaassen 2009), the expected foraging time (h/m2) was predicted (Tpred) per sampling point as:
where 56 J/s is the threshold giving-up net energy intake rate as measured in a previous study (Nolet & Klaassen 2009).
Finally, cumulative foraging times per creek (Tpred) were predicted for 2006 and 2007. Based on the initial tuber densities and observed tuber consumption, patch-type-specific (i.e. according to water depth and sediment type) functional responses were used to assess foraging times. Initial tuber densities were estimated based on satellite images of pondweed vegetation and sediment type and water depth based on interpolations of measured values (for detailed description of the whole procedure see Gyimesi et al. 2012). The cumulative number of Swans counted daily per creek provided the total observed foraging time in swan-days (Tobs). After ln-transformation for normality, the predicted total foraging time was compared with the total observed foraging time using paired t-tests, for creeks open and closed to the public separately.
None of the initial values showed a significant difference between creeks open and closed to the public (Table 1). When comparing GUDs between creeks open and closed to the public, the factor open/closed to the public was not significant, nor was the interaction term with initial tuber density. Consequently, both of these terms were removed from the final model (Table 2). GUDs differed between study years and among creeks, and were correlated with the initial tuber density (Table 2).
Table 1. Analysis of variance (df=1,181) comparing sediment type, water depth, initial tuber density and net energy intake rate in creeks closed and open to the public, with ‘year’ as random categorical predictor and ‘creek’ as random categorical factor nested in the factor ‘open and closed to the public’. Closed creeks, n =127; open creeks, n =61
Mean (± sd) closed
Mean (± sd) open
−0.1 (± 2.0)
0.1 (± 2.6)
Water depth (m)
0.44 (± 0.19)
0.49 (± 0.15)
Initial tuber density (g/m2)
20.7 (± 19.5)
23.0 (± 11.8)
Initial net energy intake rate (J/s)
89.0 (± 151.6)
100.4 (± 136.0)
Table 2. Results of the final GLM model comparing giving-up densities (GUDs) between creeks open and closed to the public (non-significant terms dropped from the final model), with year and creek as random categorical predictors (creek nested in the factor open/closed), and initial tuber density (Di) as continuous predictor
In the model comparing GUNs between creeks open and closed to the public, creek was a non-significant factor (Table 3). Although the main effect of open/closed to the public did not have a significant effect on GUNs either, the interaction term between open/closed to public and initial net energy intake rate was significant. In other words, GUNs increased with initial energy intake rates, but more steeply in creeks open to the public compared with closed creeks (Table 3); Swans gave up foraging at higher net energy intake rates in creeks open to the public than in creeks closed to the public. Estimations of the threshold GUN (i.e. k, the intake rate that maximizes long-term average intake rate) also resulted in marked differences between creeks open and closed to the public (148.2 and 84.3 J/s, respectively; Fig. 3).
Table 3. Results of the final GLM model comparing giving-up net energy intake rates (GUNs) between creeks open and closed to the public, with year and creek as random categorical predictors (creek nested in the factor open/closed) and initial net energy intake rate (neii) as continuous predictor
Sampling points within creeks open to the public were on average 214 (± 178 sd) m from the nearest boat channel. The measured distance to the boat channel had no effect on GUDs, either alone or in interaction with initial tuber density, creek and year, or in any of their combinations. After removing all non-significant terms one by one, in the final model GUDs were only correlated with initial tuber density (F1,58 = 166.4, P <0.0001). Similarly, GUNs were not affected by distance alone or in interaction with initial net energy intake rate, creek and year or in any of their combinations. As before, after removing all non-significant terms one by one, in the final model GUNs were only correlated with initial net energy intake rate (F1,58 = 619.0, P <0.0001).
We predicted that Swans would compensate for disturbance occurring during the daytime by foraging relatively more at night. Therefore, in creeks open to the public we expected foraging times calculated from food consumption to be more than foraging times calculated from daytime counts. Against our expectations, the predicted and observed values were similar (Table 4), both in the creeks open to the public (t5 = 1.3, P =0.2) and in the closed creeks (t12 = 1.9, P =0.08), providing no evidence for such compensation by the Swans.
Table 4. Observed (Tobs) and predicted (Tpred) foraging times (in bird-days) in creeks (shown with abbreviations of the creek names) closed (n = 6) and open (n = 3) to the public, as well as their totals (SUM) in the years 2006 and 2007
Potential differences in disturbance perceived by Bewick's Swans between creeks that were open and closed to the public were not revealed by GUDs, a measure that does not correct for other differences among creeks, such as sediment type and water depth. However, GUNs, which incorporate such environmental differences, reflected differences in disturbance between creeks open and closed to the public, with less foraging in creeks with human disturbance. However, within creeks open to the public, distance to the boat channel did not directly affect GUDs or GUNs. Based on the measured tuber consumption, the predicted foraging times were not significantly different from observed total foraging times, suggesting that Bewick's Swans did not compensate for lost foraging time in creeks open to the public due to disturbance during the daytime by feeding there more at night.
The difference in GUNs between creeks open and closed to the public was only prevalent at sampling points where initial tuber density and net energy intake rate were high enough. The influence of the initial tuber density on GUD and of the initial net energy intake rate on GUN can be explained by the presence of sampling points where the initial values are close to the GUD or the threshold GUN. In clumped food distributions, as is the case with pondweed tubers (Nolet & Mooij 2002, Klaassen & Nolet 2008), foragers can distinguish between patch qualities and concentrate their effort on the energetically profitable areas (Klaassen et al. 2006, 2007, Gyimesi et al. 2010). Consequently, grazing pressure is low or absent in patches at low initial food densities and below the threshold GUN (Olsson & Brown 2006). Nolet and Klaassen (2009) found the GUN to be constant (56 J/s) across patches, at a value that was close to our measured GUNs, but only in the undisturbed creeks (such as the one used in the study of Nolet & Klaassen 2009)).
Disturbance causing animals to flush leads to high flight costs (Goss-Custard et al. 2006). Alert behaviour brings about elevated heart rates, and thus extra energy consumption (Riddington et al. 1996, Ackerman et al. 2004). However, the fear of risk (i.e. predators or humans) may be even more important than these direct energetic costs in rendering disturbed sites less attractive (Gill et al. 2001, Brown & Kotler 2004). Animals that rely partly on nocturnal foraging, such as Bewick's Swans (Nolet & Klaassen 2005), might in theory be able to compensate for disturbance during the daytime by foraging more intensively at night (Hockin et al. 1992, Riddington et al. 1996). Night observations at creeks open to the public with a light intensifier confirmed the presence of foraging Swans (but actual counts were not possible due to the large distance or the poor light conditions). However, our comparison between foraging times predicted from measured food consumption and from daytime swan-counts did not reveal that Bewick's Swans adopt such a switching behaviour. This suggests that Swans may be reluctant to leave the currently employed foraging location for another one of unknown quality.
In a previous study, Bewick's Swans had higher quitting defecation rates, suggestive of higher GUDs, on beet fields close to roads than on beet fields further from roads (van Gils & Tijsen 2007). However, we did not find a negative relationship between GUDs or GUNs and distance to boat channels within creeks open to the public. At the Lauwersmeer, several types of disturbance sources occur, all fluctuating in frequency, both within and between open creeks and years. Therefore, animals foraging at different sampling points probably experienced varying levels of disturbance, which might overshadow the single effect of distance to, for instance, a boat channel (Taylor & Knight 2003, Rees et al. 2005, Preisler et al. 2006). In addition, the effect of disturbance also depends on the size, composition and condition of the group being disturbed (Weston & Elgar 2005, Rode et al. 2006, Sirot 2006, Stankowich 2008).
In addition to humans, animals perceive the presence of other predators at a site as risky, leading to different GUDs and GUNs (Brown et al. 1994). Bewick's Swans may be confronted with Red Foxes Vulpes vulpes and White-tailed Eagles Haliaeetus albicilla at the Lauwersmeer, but there are no local records of either of them predating on Swans. Still, in systems where predators do play a role and predation risk varies spatially, the GUD and GUN will comprise their effect as well, in addition to human disturbance.
Although in principle applicable to any predator–prey system, a main limitation of applying natural GUDs or GUNs as an indicator of human disturbance is that they can only be used to study the habitat use of foraging sites. The use of GUNs may require the collection of a large amount of information, and thus GUDs are preferred when the habitat is more or less homogeneous or areas with different characteristics are obviously separated by physical boundaries. However, when the environmental differences occur along a gradient on a large spatial scale without clear transition zones, GUNs may be a useful alternative. The advantages of using GUNs are especially distinctive in relatively simple systems like ours, in which a singular food source is being used by animals all at the same stage of their life cycle.
Bewick's Swans maximize their energy intake at the Lauwersmeer (Nolet et al. 2006b) and thus meet the assumptions of the marginal value theorem (Charnov 1976). The same species at another site or in another life-phase might have different goals in terms of energy gain (Brown et al. 1994, Brown & Kotler 2004, Heithaus et al. 2007, van Gils 2010). The tolerance to a certain level of disturbance might be different for animals that are satisfying their energy needs or that permanently stay at a site (Nonacs 2001, Stankowich 2008). Especially in the case of this latter group, habituation to disturbance must also be accounted for (Hockin et al. 1992, Taylor & Knight 2003). Migrating animals often visit a stopover site only for a few days or weeks, and hence barely have time to habituate to local disturbance factors (Hockin et al. 1992, Rees et al. 2005).
As the pressure on natural areas is ever increasing, it is important to measure the effect of disturbance on wildlife to inform reserve management. While experimental GUD measurements have successfully been used earlier to investigate the effects of human disturbance (see review in Brown & Kotler 2004), non-experimental GUNs may provide an alternative for expressing the cumulative long-term effect of disturbance on an animal population.
We thank all those who assisted during the fieldwork: Peter de Vries, Thijs de Boer, Koos Swart, Marcel Klaassen, Frederike Raassen, Liesbeth Bakker, Rommert Cazemier, Oane Tol, Boris Arevalo, Bert Hidding, Mariska Nieuwenhuijsen, Susan Zwerver, Sam Varghese, Casper van Leeuwen, Linda Franken and Stefan de Klepper. We thank Staatsbosbeheer, and Jan Willems in paticular, for permission to work at the Lauwersmeer. We thank Marcel Klaassen, Theunis Piersma, Matthieu Guillemain, Mike Weston, Ruedi Nager, Niall Burton and an anonymous reviewer for commenting on the manuscript. The project was financed by the NWO-ALW, grant 814.01.008. This is publication 5294 of the Netherlands Institute of Ecology.