Nest protection in mallards Anas platyrhynchos: untangling the role of crypsis and parental behaviour

Summary

• 1
  The covering of clutches with nest material is generally considered to improve the thermal environment of developing embryos. Here we tested an alternative hypothesis: that this behaviour reduces the risk of clutch detection by predators and hence, fulfils a cryptic anti-predation function in a ground-nesting non-passerine bird, the mallard. In addition, we assess the anti-predation function of the direct presence of an incubating parent on the nest for the first time in a ground-nesting non-passerine bird.
• 2
  We compared predation rates of real mallard nests with two types of artificial clutches: (i) covered with nest material and (ii) uncovered. In addition, the cryptic effectiveness of nest material, female body presence, and uncovered clutch were assessed using a simulated search for nests on photographs by human volunteers. This allowed us to evaluate separately the impact of overall crypsis (covering of the clutch by nest material and colouration of the female feather) and the direct protective capacity of the incubating female.
• 3
  Our data demonstrate that in mallards, concealment of the clutch with nest material reduces the risk of nest predation. Although the incubating female seems to provide less effective crypsis to the nest than nest material alone, the presence of the female on the clutch enhanced nest survival, suggesting a significant anti-predation capacity of the incubating parent in this species.
• 4
  Contrary to some previous studies, the relative effects of crypsis and parental anti-predation behaviour on nest survival did not differ with respect to nest concealment by surrounding vegetation.

Introduction

Investments in parental care are expected to maximize the fitness of parents by optimizing the trade-off between the survivorship and quality of their offspring on one hand and parental reproductive output during future breeding attempts on the other (Williams 1966). Variable (and sometimes opposing) selective pressures drive the evolution of parental care, such as maintaining conditions for development of embryos (Haftorn 1988), diminishing the risk of nest predation (Montgomerie & Weatherhead 1988) or provisioning for young. Particular behavioural adaptations associated with parental care may enhance parental fitness by various mechanisms, but direct fitness consequences often remain poorly understood.

The covering of clutches with nest material during parental absence occurs in a wide variety of animals, including both invertebrates (Shimoda, Shinkaji & Amano 1994) and vertebrates, such as fish (Fleming et al. 1996), amphibians (Orizaola & Brana 2003), reptiles (Burger 1976) and birds (Summers & Hockey 1981; Götmark & Ahlund 1984; Salonen & Penttinen 1988). However, direct evidence for fitness benefits resulting from this behaviour is limited. Traditionally, clutch covering is believed to increase hatching success by improving the thermal isolation of the clutch (Caldwell & Cornwell 1975). Nevertheless, the few studies that have attempted to test this hypothesis directly have produced mixed results (e.g. see Burger 1976 and Shimoda et al. 1994). Various alternative hypotheses have been suggested to explain the adaptive function of clutch covering; for instance, parents may bury eggs under the nest material as an anti-parasitic strategy (Clark & Robertson 1981). In addition, clutch covering is also believed to be involved in sexual conflict: a female may use clutch covering to hide her eggs thereby camouflaging the stage of her fertile period from a social partner or extra pair intruders (Valera, Hoi & Schleicher 1997; Low 2004). Moreover, clutch covering by various animal taxa is thought to reduce the risk of clutch detection by predators (Götmark & Ahlund 1984; Shimoda et al. 1994; Orizaola & Brana 2003) in a similar way to the cryptic colouration of eggshells (Haskell 1996) or of adults incubating the clutch (Martin & Badayaev 1996).

Nest crypsis (cryptic colouration of parents or eggshells, or a specific type of nest construction) may affect nest survival in combination with other factors. Parents often place nests in sites concealed by the surrounding vegetation, which may diminish the risk of nest detection by predators (Clark & Shutler 1999). In addition, parental incubation behaviour (such as parental presence at the nest) may reduce the risk of nest predation by directly defending the clutch (sensu Montgomerie & Weatherhead 1988); alternatively, parental presence itself may deter predators from approaching the nest (i.e. Opermanis 2004). At the same time parental incubation behaviour (such as moving on or around the nest during incubation breaks) may enhance the risk of the nest disclosure to predators (Skutch 1949). These mechanisms may act simultaneously to affect the probability of nest predation, and their relative contribution to reproductive success may combine in a complex, non-additive way. For example, the effect of parental incubation behaviour may be crucial for survival of nests that are poorly concealed by vegetation, but not for nests that are exposed to low predator detection risk due to sufficient vegetation concealment (Cresswell 1997; Weidinger 2002). Similarly, the anti-predation effect of crypsis provided by parents or nest structures (eggs, feathers) is believed to be lower when vegetation concealment is high (Cott 1940; Stuart-Fox & Ord 2004).

The aim of this study was to test whether incubation behaviours and clutch covering by females reduces nest predation risk in the mallard, a ground-nesting non-passerine bird for which the incubating female has cryptic body colouration. Since nest concealment by surrounding vegetation may affect nest success (Albrecht & Klvaňa 2004; Albrecht et al. 2006), we tested for an interaction between vegetation concealment and the influence of parental presence (called the ‘parental compensation hypothesis’; Cresswell 1997; Remeš 2005). In addition, our experimental approach allowed us to test whether the protective effect of vegetation concealment nullifies the anti-predation effect of nest crypsis (covering of the clutch by nest material and/or the cryptic incubating female; e.g. Cott 1940; Stuart-Fox & Ord 2004).

Parental body colouration may affect the risk of nest predation (Martin & Badayaev 1996); however, previous experiments examining the effect of parental behaviour on nest predation have often implicitly assumed that active nests incubated by parents and artificial nests, which are not affected by parental behaviour, are equally cryptic (e.g. Cresswell 1997; Weidinger 2002; Remeš 2005; Fontaine et al. 2007). Our study evaluates separately, to the best of our knowledge for the first time, the contribution of crypsis and parental incubation behaviour to nest success, by comparing the survival of naturally incubated mallard clutches (affected by parental activity, sensu Cresswell 1997) with that of two types of artificial clutch whose fate is not affected by parental incubation behaviour; that is, clutches either (i) covered or (ii) uncovered by nest material. Second, the cryptic effectiveness of nests covered by either nest material or the cryptic female, and uncovered nests, are compared using simulated nest searches of visually oriented predators (humans) as a proxy of crypsis (i.e. Cuadrado, Martin & Lopez 2001).

If clutch covering performs an anti-predation function, then covered clutches should have lower detection by human predators than uncovered clutches (i.e. they should suffer a lower predation risk by visually oriented predators). The evaluation of crypsis, using human volunteers, may help to roughly separate the effect of protective parental behaviour and crypsis. For instance, parental incubation behaviour could be said to be predominantly protective if survival of both types of artificial clutches is lower than survival of incubated nests and if these nest types do not differ in crypsis (or incubated nests are even less cryptic than artificial nests). The opposite pattern, that is, a higher survival rate for artificial clutches compared to natural nests, would indicate that nest disclosure due to parental movements around the incubated nest prevails over the protective effect of parental presence (i.e. Skutch 1949). On the other hand, our approach does not allow us to directly separate the effect of crypsis and parental incubation behaviour in the situation where one clutch type would be both more cryptic and more successful against real predators than the other. To test the hypothesis that the anti-predation effect of the vegetation concealment compensates the effect of parental activity (i.e. Cresswell 1997) and cypsis (i.e. Cott 1940), in both experiments, we further examined the significance of the interaction between nest type (clutches covered by nest material, incubating female or uncovered clutches) and vegetation cover.

Material and methods

study site

Field work was conducted from mid-April through mid-July of 2003 and 2004 on seven-selected fishponds (total study area; 40 km²) in the Třeboň Biosphere Reservoir, Doudlebia, Czech Republic (49°9′N, 14°43′E). We searched for nests on small artificial islands (5–30-m wide, 50–300-m long), where visually oriented avian predators (marsh harriers Circus aeruginosus and crows Corvus corone) cause the majority of nest losses (Albrecht et al. 2006). The vegetation around mallard nests included the common reed (Phragmites communis, 53% of all nests included in the experiment), stinging nettle (Urtica dioica, 16%) and sedge grass (Calamagrostis epigeos, 31%); see Albrecht & Klvaňa (2004) for details.

expriment 1: the effect of clutch covering and parental presence on nest fate

Artificial nest experiment

Experiments using artificial clutches enable the separation of the contributions of crypsis (Haskell 1996) and parental incubation behaviour (Cresswell 1997) to nest success. For each real nest (hereafter called ‘incubated nest’), we assigned a pair of artificial nests (not incubated, and therefore not affected by parental activity; Cresswell 1997) 7·5 ± 2·7 m (mean ± SE, range 5·5–15 m) from the incubated nest. For each pair of artificial clutches, one was covered by nest material taken from the spatially associated incubated nests, and the second was left uncovered (although we added some nest material around this clutch to mimic the scent and appearance of a real mallard nest (Guyn & Clark 1997). This experimental design enabled us to test the anti-predation function of clutch covering (survival of covered vs. uncovered artificial nests), as well as the contribution of parental incubation behaviour to the nest success (survival of incubated vs. covered artificial clutches; we assumed that an equivalent or lower quality of clutch crypsis is provided by the female feather colouration compared to nest material [see below]).

Artificial nests were constructed from dead vegetation shaped into a cup closely resembling natural mallard nests. Real inactive nests baited with an artificial clutch have previously been used to assess the effect of parental activity on nest survival in open cup shrub-nesting birds (Cresswell 1997; Komdeur & Kats 1999; Weidinger 2002). In the case of ground-nesting birds, clutch colouration rather than nest design is expected to be a crucial clue for predator nest disclosure (i.e. Weidinger 2001). Thus, we assume that the use of human-made artificial nests does not seriously affect the results of this experiment.

Clutches of artificial nests consisted of four brown chicken eggs. In the additional experiment performed in year 2006 on the same study plots (data not included in the results), artificial clutches baited with chicken eggs had a survival rate comparable to those baited with mallard eggs (28 artificial nest couples χ² = 0·65, d.f. = 1, P > 0·40; in fact mallard eggs were slightly more predated than chicken eggs, 19 vs. 16 nests predated, Kreisinger & Albrecht unpublished data). Hence, the difference in eggshell colouration between chicken and greenish mallard eggs is unlikely to bias our results.

Nest sites for artificial clutches were chosen to approximate the nest site of the associated incubated clutch in features that could affect the probability of predation (e.g. vegetation structure, height and density, proportion of dead vegetation, distance to the closest tree, shrub and water). We took special care not to manipulate or damage the vegetation around all nests. To control for possible bias in this aspect, vegetation concealment was measured for each clutch at the beginning of each experiment. A 20 × 20-cm grid composed of eight white 5 × 5-cm squares was placed directly on the nest bowl, and the percentage of white squares obscured by vegetation when viewed from 1 m directly overhead was scored. The mean vegetation concealment of all nests was 50·7 ± 24·0% (mean ± SE) with no differences in vegetation concealment between incubated and artificial clutches in each triplet (6·3 ± 8·7%[mean ± SE], Repeated Measures anova, F_2,173 = 0·658, P > 0·5).

The relative proportion of time that a female spends on the nest differs noticeably between the pre-incubation and incubation stages (Afton & Paulus 1992). Therefore, in this experiment, we only included real nests that survived to day 3–5 of incubation (based on the floating test of Westerkov 1950). Clutches found during pre-incubation or very early incubation (< 3 days) are often abandoned (e.g. Johnson, Nichols & Schwartz 1992), and therefore, were not included in our experiment since the influence of parental behaviour cannot be assessed for abandoned nests. Incubated nests included in this experiment were a random sub-sample of all mallard clutches found in our localities which fulfilled the criteria mentioned above.

Nests triplets were checked twice, the first time after 5·7 ± 0·8 (mean ± SE) and a second time after 11·4 ± 1·1 (mean ± SE) days from the start of the experiment. When we approached the nest, the female duck usually left the nest suddenly without covering eggs with nest material. When this happened, we covered clutches with nest material after the nest check to mimic the behaviour of an undisturbed female leaving the nest during incubation recesses (Caldwell & Cornwell 1975). During the second check, successful artificial clutches were removed.

Nests were defined as predated if the clutch was damaged or at least one egg was missing. This criterion makes our results rather conservative, because incubated nests with partial clutch loss, where the female continued the incubation were classified as predated (two incubated nests, one and three eggs missing, respectively). Nevertheless, in these cases we could not distinguish between partial clutch predation (i.e. Ackerman et al. 2003) and other causes of eggs loss (such as the ejection of an egg with a broken shell by the female). We excluded two nest triplets for which the incubated nest was abandoned by the female during the experiment and the clutch was found intact during a subsequent visit. The final analysis included 60 nest triplets.

Experiments using artificial nests increase natural nest densities, which may draw predators and lead to an artificial increase of predation rates. However, mallards already breed in high densities in our locality. As only a small proportion of incubated nests were included in the experiment (15–18% mallard nests found), nest densities were only increased by the experimental setup by 30–35%. Similarly, the mean distance between neighbouring natural nests (10–20 m) was only increased slightly by our experiment. We assumed that this modest increase in nest densities did not alter natural predator densities or predation rates, since experiments based on much larger differences in duck nest densities show no consistent effect of nest densities on predation rates (see Ackerman, Blackmer & Eadie 2004 for review).

experiment 2: cryptic qualities of eggshells, nest material and the female body

To compare the cryptic quality of nest material, the female feather colouration and uncovered eggs, 15 groups of photographs of mallard nests were taken during the breeding season. Each group contained three pictures: (i) a clutch covered by a stuffed female mallard; (ii) a clutch covered by nest material, and (iii) an uncovered clutch. In addition, 150 photographs containing no nest were included in the experiment (i.e. Cuadrado et al. 2001). Fifteen of these control pictures featured the vegetation microhabitat close to nests included into this experiment (1–2 m away; hereafter ‘paired control photos’). The remaining 135 photos represent random sites of mallard breeding habitat within our study localities (hereafter termed ‘unpaired control photos’). Human volunteers (8 males, 12 females, aged 18–62 years) were asked to search for nests located randomly in the 195 photographs (150 control photos + 45 photos containing nests) that were presented in a random sequence for 10 s on a computer monitor. The pilot experiment showed that a successful search rarely exceeded 10 s (Kreisinger, unpublished data). Volunteers were asked to use a mouse to click on the spot where they believed there was a nest. The search was considered successful when volunteers clicked on the target area where a nest was present, with the ‘target area’ occupying c. 2% of the photo (a circle with 4·5 cm diameter on 15″ monitor for all photos containing a nest). The vegetation concealment of each nest was measured following the procedure used in the previous experiment. All photographs were taken 4 m away from the nest, from the direction that provided the highest visibility of the clutch, using a Practica MTL5B with a Helios 50 mm lens and Kodak 200 ASA print film. The focal length of the 50 mm lens corresponds to humans visual system and the 4 m distance provide, based on our experience, an appropriate level of uncertainty about the nest location on photographs (a random nest detected with c. 50% probability). The height of the camera was 1·6 m above-ground level.

Humans are frequently used in similar experiments to mimic behaviour of wild visually oriented predators since their prey search is primarily based on visual perception (e.g. Cuadrado et al. 2001). However, several potential biases may arise when using humans as a model predator in experiments similar to ours. For instance wild predators usually search for an entire spectrum of potential prey, not only nests; however, our field experiments were performed on a small artificial islands where alternative prey is rather limited and we expect that these sites are visited by predators primary searching for duck nests.

In addition, avian predators, unlike humans, also perceive light in the UV spectrum (300–400 nm). To address this problem, we examined the reflectance of structures whose cryptic efficiency was studied in experiment 1 and 2 (down feathers from the nest material, brown chicken eggshells, mallard eggshells and feathers from the female dorsal surface) and samples of fresh and dry vegetation surrounding nesting microhabitats in our locality (Sedge grass, Common Reed). We recorded the reflectance of five independently collected samples of these materials using an Avaspec 2048® (Avantes, Netherlands) spectrometer equipped with micron fiber-optic probe at a 45° angle to the measured surface. Reflectance data were generated relative to a white standard. The reflectance profile was very similar for dry Sedge grass and dry Common Reed and for fresh green samples of these species. UV reflectance for both dorsal surface of a female mallard and down feathers were negligible. Brown chicken eggs exhibited lower reflectance in the UV spectrum (i.e. more similar to vegetation samples), compared to green mallard eggs, except the small reflectance peak around 320 nm. Since perception in UV spectrum by avian predators (raptors and corvids) is based on violet sensitive cones with maximal sensitivity around 405–420 nm (i.e. Hastad, Victorsson & Odeen 2005), this peak is unlikely do be detected by the most common predators of our experimental nests. Reflectance curves for particular structures are presented in Supplementary Appendix S1.

In conclusion, the use of humans is probably adequate for the estimation of crypsis in our experiment. If UV reflectance of eggshells attracts nest predators, the use of chicken eggs will only cause the estimate of the effect of crypsis and clutch covering in our study to be conservative due to their lower UV reflectance compared to mallard eggs.

statistics

Experiment 1

Nest survival was evaluated using the Mayfield method (Mayfield 1975); that is, the proportion of days survived was calculated for each nest and modelled as a response variable with a binomial error distribution (Aebischer 1999). Due to the short intervals between nest checks, the time of nest predation and hence the number of days of nest exposure were estimated as one-half of the interval between the last positive and first negative check of the nest (Mayfield 1975). Days of nest exposure differed among (but not within) nest triplets, because we did not check nests during unfavourable weather conditions in order to prevent clutch abandonment. Weighting for different lengths of nest observation (days of nest exposure) was included in the analysis (Aebischer 1999; Crawley 2002).

Because nests in the first experiment were spatially grouped in triplets, their outcome cannot be treated as independent in the statistical analysis (i.e. the probability of predation of a single nest may depend on the probability of predation of any of the other nests in the triplet). To control for this issue and to avoid pseudoreplication, the data were analyzed within the framework of the GLMM, (GLMM binomial error distribution, R 2·6·0 software) with the nest triplet identity included as a random effect (e.g. Faraway 2006).

The most complex model contained triplet identity as the random effect and a set of categorical explanatory variables (nest type: incubated, covered artificial or uncovered artificial nests, plus year), continuous variables (vegetation concealment, date of particular experiment initiation) and all two-way interactions between variables as fixed effects. We first assessed the importance of random effect by comparing explained variance of the model including random effect with the model where the random effect was not considered using Akaike information criterion (AIC, Faraway 2006). Backward elimination of non-significant terms was further used to select the best Minimal Adequate Models (MAM), that is, the most parsimonious ones with all fixed effects significant (Crawley 2002). The significance of particular terms on the explanatory power of a model was tested using the deletion test. Data were checked for normality and, in the case of vegetation concealment, data were root square arcsine transformed before calculation.

We also compared the survival of incubated mallard nests in experimental localities but that were not included into the experiment (n = 252) with the survival of experimental nests using Aebischer's (1999) method. Since experimental nests were grouped in triplets, we performed three separate models (i.e. non-experimental incubated nests vs. each one of three experimental nest types).

Experiment 2

Statistical procedures are based either on GLM or GLMM (triplet of photos of the same nest included as a random factor) with a binomial response (proportions of cases when different observers correctly identified a particular nest, or proportions of cases when the position of the nest was misidentified respectively [e.g. Cuadrado et al. 2001]). Vegetation concealment, clutch cover type and interaction were incorporated as explanatory variables. The procedure for the MAM model selection was identical to that in experiment 1.

Results

the role of clutch covering and parental presence

The backward reduction of the complex survival model revealed a significant effect of nest type and vegetation concealment on nest survival, whereas the effect of other variables (between year differences, date of experiment initiation and two-order interactions) were statistically negligible, including the focal interaction between nest type and vegetation concealment (Table 1). The relationship between nest survival (i.e. proportion of days survived) and vegetation concealment was positive, with comparable regression slopes for all experimental nest types; incubated nests (slope = 1·852 ± 0·934 SE), artificial covered clutches (slope = 1·717 ± 0·788 SE) and artificial uncovered clutches (slope = 2·033 ± 0·670 SE). When the random effect (nest triplet identity) was deleted from the model, the explained variability decreased significantly (Δ AIC = 3·99), suggesting covariance of nest survival within a particular triplet. The a posteriori reduction of nest type categories further revealed that uncovered artificial clutches had significantly lower survival rates than covered artificial clutches (incubated + [covered vs. uncovered], Δ d.f. = 1, χ² = 11·41, P < 0·001, Fig. 1). The survival rates of incubated clutches was significantly higher than survival of covered artificial clutches ([incubated vs. covered] + uncovered, Δ d.f. = 1, χ² = 5·06, P = 0·024, Fig. 1). The survival of incubated nests included in the experiment did not differ from the survival of incubated mallard nests found in the same localities not included in the experiment; Mayfield's daily survival rates (DSR) for incubated experimental nests (n = 60, DSR = 0·969 ± 0·0072 SE) and incubated nests not included in the experiment (n = 252, DSR = 0·972 ± 0·0027 SE) did not differ (GLM, ΔD.f. = 1, χ² = 0·66, P < 0·4, Fig. 1). On the other hand, and consistent with previous results, incubated nests that were not included in the experiment had a higher survival than artificial covered clutches (GLM, Δ d.f. = 1, χ² = 10·80, P = 0·001) and artificial uncovered clutches (GLM, Δ d.f. = 1, χ² = 51·79, P < 0·001, Fig. 1).

Table 1.  Survival model based on GLMM with binomial error structure (Aebischer 1999). Backward eliminations of non-significant terms to select the best minimum adequate model (MAM) with all their effects significant were used. Models were compared using the deletion test. Significant factors included in the MAM are in boldface

Factor	d.f.	χ^2	P
Nest type (active + covered + uncovered)	2	31·11	< 0·001
Vegetation	1	12·36	< 0·001
Date	1	2·53	0·112
Year	1	0·67	0·415
Year × Date	1	2·60	0·107
Date × Vegetation	1	0·34	0·562
Year × Vegetation	1	2·22	0·136
Nest type × Year	2	0·98	0·614
Nest type × Vegetation	2	1·54	0·462
Nest type × Date	2	1·16	0·560

Figure 1. Daily survival rates (± SE, Mayfield 1975) of active mallard nests that were not included in the experiment (Incubated Non-exp) and of reminding nest groups that were included in the experimental setup: experimental incubated mallard nests (Incubated exp), artificial clutches covered with nest material (Covered) and artificial clutches that were not covered with nest material (Uncovered). Different letters above columns represent significant differences between nest types (P < 0·05, based on the Generalized Liner Mixed Models).

cryptic quality of eggshells, nest material and the female body

On average, in 6·2 ± 1·2 (SE) % of cases, volunteers identified an incorrect object as a mallard nest on control photos. The frequency of these mistakes did not differ between paired control photos and unpaired control photos (GLM, Δ d.f. = 1, χ² = 0·63, P = 0·427). Similarly, the proportion of cases in which no nest was found and any object was incorrectly identified as a nests did not differ between pictures containing a nest and control pictures (GLM, Δ d.f. = 1, χ² = 0·56, P = 0·453), indicating that human observers were no more likely to record false positives on control photographs than on photographs with nests.

If the analysis to pictures is restricted to those containing a nest, the probability of clutch detection by humans decreased with increasing vegetation concealment (GLMM, Δ d.f. = 1, χ² = 13·66, P < 0·001) and was affected by the type of clutch cover (Δ d.f. = 2, χ² = 114·94, P < 0·001), but the interaction between these variables was not significant (Δ d.f. = 2, χ² = 3·00, P = 0·223). A negative relationship between the probability of nest detection and vegetation concealment was detected for all nest groups (regression slope for uncovered clutches: –0·031 ± 0·006 SE, for clutches covered by the nest material: –0·025 ± 0·006 SE and for clutches covered of the female body: –0·029 ± 0·005 SE). Uncovered clutches tended to be detected more often than clutches covered by a stuffed mallard female (Fig. 2); nevertheless when merging these two nest types in the model (i.e. [uncovered clutches vs. clutches covered by female] + clutches covered nest material], this difference was not significant (Δ d.f. = 1, χ² = 2·81, P = 0·093). On the other hand, clutches covered by nest material were significantly less likely to be detected than clutches covered by a stuffed mallard female (uncovered clutches + [clutches covered by female vs. clutches covered nest material]; Δ d.f. = 1, χ² = 70·83, P < 0·001).

Figure 2. Bars indicate proportions of nests (mean ± SE) with uncovered clutches (Uncovered), clutches covered by nest material (Covered) and clutches camouflaged by stuffed mallard female (Female) detected on photographs by human volunteers. Different letters above columns represent significant differences between nest types (P < 0·05, based on Generalized Liner Mixed Models).

Discussion

The use of feathers as a non-structural nest bowl lining in various bird taxa is assumed to primarily improve the thermal environment of the clutch and consequently the hatchability of young (Hilton et al. 2004; McGowan, Sharo & Hatchwell 2004), although this effect on fitness has rarely been demonstrated experimentally (but see Lombardo et al. 1995). Our data clearly show that clutch covering by nest material consisting of down feathers and dead grass may also fulfil a cryptic function. Consistent with previous studies performed on artificial nests (Götmark & Ahlund 1984; Salonen & Penttinen 1988; Opermanis 2004), in our study, covered nests suffered a lower predation risk than uncovered artificial nests (Fig. 1). A cryptic function of the clutch covering was simultaneously supported by the experiment with human volunteers, where covered clutches were less likely than uncovered clutches to be detected (Fig. 2). Our experiment does not allow the accurate measurement of anti-predation benefits arising from clutch covering, since clutch covering may fulfil its function only when the female is off the nest, which in the case of the mallard involves 10–15% of the incubation period and a large proportion of the pre-incubation stage (Caldwell & Corwell 1975; Afton & Paulus 1992). However, our data demonstrate that, apart from other mechanisms, predation can be an important factor favouring the evolution of clutch hiding behaviour, especially in species with poorly developed eggshell crypsis. This could be particularly true in ground-nesting species because it is primarily the clutch itself, rather than the nest bowl, that attracts the attention of nest predators (e.g. Weidinger 2002).

The anti-predation function of cryptic colouration is often assumed to increase with decreasing concealment by vegetation (Cott 1940; Stuart-Fox & Ord 2004), but previous studies have generally focused on species that nest in open habitats covered by sparse vegetation (Götmark & Ahlund 1984; Opermanis 2004, but see Salonen & Penttinen 1988), limiting the possibility of testing whether clutch covering actually compensates for poor vegetation concealment (e.g. Cott 1940). The results from our survival model (Table 1) and our human predator experiment, in fact fail to support any interaction between nest covering and vegetation concealment, contrary to previous studies that suggest that clutch covering can compensate for poor nest site concealment.

Although we clearly demonstrate that covering the nest with nest material may strongly improve nest survival in mallards, the fitness consequences of this behaviour may simultaneously involve both better thermoregulation and enhanced development of embryos. Future manipulative experiments should be focused on assessing the relative importance of potential pathways through which clutch covering enhances female fitness.

In this study, covered artificial clutches exhibited lower survival than incubated nests guarded by a mallard female (Fig. 1). This pattern is unlikely to be explained by a better cryptic quality of the female body, as nests covered by nest material did not differ in UV reflectance from nests covered by females and were detected less frequently by human observers. Similarly data reported by Opermanis (2004) indicate clutches covered by nest material to be more cryptic to non-human predators than clutches covered by a duck female. Since an incubating female does not necessarily cover its clutch with nest material during every break in incubation (Caldwell & Cornwell 1975), our data provide conservative estimates of nest protection due to parental incubation behaviour, because uncovered nests suffer higher predation risk. Consequently, low predation rates of incubated nests indicate that parental incubation behaviour via direct parental presence improves nest success in mallards. On the other hand, the effect of nest disclosure due to female activity (e.g. Skutch 1949; Martin, Scott & Menge 2000) is unlikely to markedly reduce nest success in our model system.

It is also worth noting that incubated nests included in our study are in fact a sub-sample of nests that survived until the onset of incubation (see also Cresswell, 1997; Weidinger 2002 for similar drawbacks in the methodology). Hence, our experiment might slightly overemphasize the effect of parental behaviour on nest survival by estimating the parental behaviour effect for survivors but not unbiased estimates for the studied population, particularly in the case of high inter-individual variability of the protective component of parental behaviour. Nevertheless, we believe this potential bias is relatively low, since incubated nests were not monitored for only a relatively small proportion of the incubation period (3–5 days, c. 15–20% of the incubation period) and losses of unincubated nests and nests during the early incubation stage (0–5 days) due to predation were relatively low (c. 5–10%) compared with losses due to clutch abandonment (c. 20%).

In line with observed anti-predation effect of nest guarding, experiments based solely on non-incubated artificial nests performed by Opermanis (2004) indicate that the mere passive presence of an inactive dummy female duck on an artificial ground clutch may deter some avian predators. The contribution of parental behaviour to nest success has been experimentally revealed in several open-cup nesting passerines (Cresswell 1997; Komdeur & Kats 1999; Remeš 2005). In waterfowl, nest guarding is known to decrease predation risk in species with large body size such as geese, swans or eiders (Andersson & Waldeck 2006; Samelius & Aulisauskas 2006). We show, however, that parental presence on the nest during incubation may provide protection of the clutch against predators even in ground-nesting ducks with a mallard-like body size.

Under the parental compensation hypothesis, the protective effect of parental incubation behaviour often, but not always (e.g. Weidinger 2002), contributes more to nest success in unsafe (i.e. poorly covered by vegetation) nest sites (Cresswell 1997; Remeš 2005). Several mechanisms have been suggested to explain parental compensation; for example, a lower intensity of nest guarding and higher parental activity for individuals breeding in nest sites exposed to low predation risk (i.e. dense vegetation, McLean, Smith & Stewart 1986; Montgomerie & Weatherhead 1988; Martin et al. 2000) can be expected if nest guarding is associated with energetic costs (Montgomerie & Weatherhead 1988; Komdeur & Kats 1999). On the other hand, breeding in dense vegetation may limit the prompt detection of a predator, and preclude the parent from modulating its activity to avoid nest disclosure (Götmark et al. 1995).

In mallards in this study, parental incubation behaviour improved nest success regardless of vegetation concealment, despite vegetation concealment itself predicting nest fate (Table 1), providing no support for the parental compensation hypothesis. Consequently our results indicate that none of the suggested mechanisms of parental compensation play a significant role in the mallard. Nevertheless, a detailed analysis of female behaviour, and predation rates nests with experimentally manipulated vegetation concealment, should be performed (i.e. Remeš 2005).

Acknowledgements

Authors thank David Hardekopf, Heidi C. Hauffe, Pavel Stopka Juan Soler and anonymous refereers for valuable comments on earlier drafts of the manuscript. The study was supported by Grant Agency of Charles University (projects 181/2005/B-Bio and 192/2007/B-Bio). Authors are also grateful to the Ministry of Education, Youth and Sport of the Czech Republic and the Academy of Sciences of the Czech Republic, whose grants MSMT 0021620828 and AV0Z60930519 formed a framework for a part of this study. T.A. is partially supported by the Research Centrum project No. LC06073.