Experimental evidence for density-dependent survival in mallard (Anas platyrhynchos) ducklings

Oecologia (2006) 149:203 213 DOI 10.1007/s00442-006-0446-8 POPULATION ECOLOGY Experimental evidence for density-dependent survival in mallard (Anas platyrhynchos) ducklings Gunnar Gunnarsson Johan Elmberg Kjell Sjöberg Hannu Pöysä Petri Nummi Received: 23 May 2005 / Accepted: 26 April 2006 / Published online: 31 May 2006 Springer-Verlag 2006 Abstract It is unresolved to what extent waterfowl populations are regulated by density-dependent processes. By doing a 2-year crossover perturbation experiment on ten oligotrophic boreal lakes we addressed the hypothesis that breeding output is density dependent. Wing-clipped mallard (Anas platyrhynchos) hens were introduced with their own brood and then monitored for 24 days. Predicted responses were that per capita duckling and hen survival would be lower in high-density than in low-density treatments. Survival was evaluated by model Wtting in program MARK. Density, year, and lake were used as main evects, while day after introduction, a weather harshness index, and presence of hens were covariates. Daily survival in ducklings was lower in the high-density treatment, but this evect was year dependent. The highest-ranking model for duckling survival also included a positive Communicated by Markku Orell G. Gunnarsson J. Elmberg Department of Mathematics and Natural Sciences, Kristianstad University, 291 88 Kristianstad, Sweden G. Gunnarsson (&) K. Sjöberg Department of Animal Ecology, Swedish University of Agricultural Sciences, 901 83 Umeå, Sweden e-mail: Gunnar.Gunnarsson@staV.hkr.se H. Pöysä Finnish Game and Fisheries Research Institute, Joensuu Game and Fisheries Research, Yliopistokatu 6, 80100 Joensuu, Finland P. Nummi Department of Applied Biology, University of Helsinki, P.O. Box 27, 00014 Helsinki, Finland evect of duckling age and presence of hens, and a negative evect of harsh weather. Density did not avect female survival although there was a prominent year evect. The highest-ranking model for female survival also included negative evects of day after introduction and harsh weather. This is the Wrst study to report density-dependent survival in experimentally introduced ducklings in a natural setting. Implications for population dynamics and management of harvested populations are far-reaching if such regulation occurs in some years, but not in others. Keywords Dabbling ducks Duckling Model Population regulation Waterfowl Introduction Density dependence is a fundamental ecological process explaining regulation and also, but not exclusively, limitation of populations. Examples of factors potentially avecting individuals in a density-dependent fashion are food abundance, nest site availability, predation, and pathogens (cf. Newton 1998, p. 374; Wilson 2000). The practical and theoretical signiwcance of density dependence has been acknowledged for a long time (Nicholson 1933), but opinions still diverge as to how frequently and how fundamentally natural populations are avected (e.g. Turchin 1990; Ray and Hastings 1996; Berryman et al. 2002; Grimm and Uchmanski 2002; Hixon et al. 2002; Both and Visser 2003). Inconsistent evidence and diverging interpretations have made the topic popular yet controversial for considerable time (e.g. Berryman 2004; White 2004). Observations that density-dependent evects may vary

204 Oecologia (2006) 149:203 213 in space and time and also avect individuals diverently have only added to this popularity (Caughley and Sinclair 1994; Caughley and Gunn 1996; Åström et al. 1996; Jonzén and Lundberg 1999). Understanding variations in density dependence is a basic aspect of population dynamics in general, being especially relevant to the sustainable management of harvested populations (Jonzén and Lundberg 1999). A common way to study density dependence is by retrospective analyses of long-term population data (e.g. Vickery and Nudds 1984; Pollard et al. 1987; Woiwod and Hanski 1992; Paradis et al. 2002; Nummi and Saari 2003; Pöysä and Pesonen 2003; Almaraz and Amat 2004). Such studies have produced widely diverging results with respect to breeding output; some report positive density dependence, some negative density dependence, and some no density dependence at all. There are even cases when patterns diverge for ecologically related bird species co-occurring in a limited geographical area (e.g. Paradis et al. 2002). However, Shenk et al. (1998) pointed out that most studies based on time series data do not account properly for sampling covariance, leading to inxated type I statistical errors, and that they may thus be invalid by overestimating the prevalence of density dependence. In addition, analyses of natural long-term data sets are frequently confounded by environmentally induced density-independent factors operating alongside putative density-dependent processes (Ruesink 2000; Coulson et al. 2004). Thus, Weld experiments addressing density dependence are much called for, but they remain rare in avian ecology (for nest predation see Sugden and Beyersbergen 1986; Reitsma 1992; Nams 1997; Larivière and Messier 1998). Arguably, density manipulations and other perturbations constitute a more successful way of unravelling the direct causality behind patterns of density dependence in vital rates (Kaminski and Gluesing 1987; Clark and Nudds 1991; Harrison and Cappuccino 1995; Berryman and Turchin 1997; Newton 1998). There is a number of studies in which long-term data from dabbling ducks have been explored for densityrelated patterns in regulation and/or limitation (e.g. Weller 1979; Hill 1983, 1984; Vickery and Nudds 1984; Kaminski and Gluesing 1987; Elmberg 2003; Elmberg et al. 2003; Pöysä and Pesonen 2003). Many concern the mallard (Anas platyrhynchos), which is a common breeder throughout the Holarctic and is perhaps the world s most well studied game species. The mallard has also served as a model organism for exploring whether hunting mortality is additive or compensatory to natural mortality; an issue where density-dependent processes constitute a necessary focus (Nichols 1991; Pöysä et al. 2004). Again, most work to date has been non-experimental, using either harvest juvenile/adult ratios or band recovery data as the dependent variable (e.g. Burnham and Anderson 1984; Kaminski and Gluesing 1987; Pöysä et al. 2004). Fundamental to population growth and to hunting bag recommendations alike is breeding output. Mallard ducklings are particularly sensitive the Wrst 2 weeks of life (Talent et al. 1983; Hill et al. 1987; Orthmeyer and Ball 1990; Rotella and Ratti 1992b; Sargeant and Raveling 1992), when ambient conditions may be crucial for survival (e.g. Chura 1961; Street 1977). Adult females, too, are potentially avected by conditions during this period, when a disproportionate part of the annual mortality occurs (Sargeant and Raveling 1992). As far as we know, there is only one previous experimental study of breeding dabbling ducks addressing density dependence in survival; Elmberg et al. (2005) demonstrated that an increased density of adult mallards lowered the number of hatched broods, but not the number of Xedged ducklings. Dzus and Clark (1997a) showed that per capita duckling survival in mallards decreased when broods of very young ducklings were enlarged, but they did not explore whether this avected vital rates in natural populations. Both studies concern small eutrophic lakes in landscapes with a strong impact from agriculture and other human activities. In other words, the generality of the results remains unclear because the experimental settings may not be very representative for the species, nor for dabbling ducks in general. We argue that the evect of population density on survival remains largely unexplored experimentally, in the mallard, in dabbling ducks, and in precocial birds in general. The present study addresses the hypothesis that breeding success is density dependent. We added mallard broods with their mother in two densities onto oligotrophic boreal lakes in a between-year crossover design, predicting that the high-density treatment would lead to lower per capita survival in ducklings as well as in females. Survival data were analysed by a modelling approach in which the possible evects of year, day after introduction, female presence, and harsh weather were also included. Materials and methods Study lakes This study was done in 2002 2003 in the province of Västerbotten, northern Sweden (64 N, 20 E). We used

Oecologia (2006) 149:203 213 205 ten typical lakes for the boreal forest biome; i.e. they were oligotrophic and surrounded by mixed coniferous forest and bogs. All the lakes freeze over in winter and waterfowl breeding on them are long-distance migrants (Fransson and Pettersson 2001). Lakes were selected according to the following criteria: (1) no regular human disturbance from cabins, Wshing, and other outdoor activities; (2) no major in- or outlets through which introduced mallards may escape; (3) isolation, i.e. there was no other lake within 300 m; (4) roundish shape and limited size (1 13.5 ha); (5) relatively easy to access, i.e. lakes were not too far from roads or tracks; (6) potential breeding sites for mallard, based on observations of broods and adults in previous years. Mallard introductions Wing-clipped mallard hens with their own brood were brought from a game farm 300 km southwest of the study area. Hens had either been wild-caught less than a year before, or they were daughters of wild females caught the previous season. At the game farm they lived under natural conditions on ponds surrounded by boreal forest and meadow-like shores. Captive birds fed on naturally available invertebrates and seeds, but they were also overed poultry pellets as a supplement. Hens mated with captive as well as with wild visiting males, and they placed their self-made nest in natural vegetation on the game farm. Clutches were complete in mid May, at about the same time as those of local wild mallards. At hatching, ducklings were collected with their mother, and brood size was standardised to ten by removal or addition. The hen and her ducklings accepted all ducklings added to an alien brood. Within 1 or 2 days of hatching, broods and hens were transported by car and introduced onto the experimental lakes. Because of natural variation in hatching date, a week lapsed from the Wrst to the last introduction each year. Ducks were shipped from the game farm to the study lakes at night to avoid high temperatures and to minimise stress. They arrived at their experimental lake early in the morning and were released promptly. Hens were always quick to swim away when released. For this reason we let the brood go before the hen so that mother and ducklings would not become separated. At lakes subjected to a high-density treatment (see below) we waited for each female and her brood to swim away and to hide before introducing the next. Females as well as ducklings were often seen foraging soon after release, from the water surface and in sheltering shoreline vegetation. By and large, hens and ducklings behaved naturally, at the game farm as well as on the experimental lakes. Based on our own experience and previous studies (cf. Pehrsson 1979; Gunnarsson et al. 2004) we argue that results and conclusions in this study are valid for wild mallards. In 2002 half of the ten study lakes were randomly assigned as high-density lakes, onto which four hens with one brood each were introduced. The other Wve were low-density lakes receiving only one hen with her brood. This treatment was reversed in 2003, producing a balanced crossover experimental design. Experimental introductions appreciably altered the density of mallards on each lake. Still, both treatment densities were within the range of what is natural in the area. This conclusion is based on the mean maximum number of wild waterfowl observed on the lakes (SD in parentheses): mallard, 0.8 (1.4); teal, 1.6 (3.3); broods of dabbling ducks, 0.8 (1.1); other waterfowl, 3.4 (2.3); broods of other waterfowl, 0.5 (0.8). A potentially biasing carry-over evect on survival may arise if birds introduced the Wrst year return to the same lake in the second. We are convinced this was not a problem in our study. Ducklings introduced in 2002 could not return in 2003 simply because they all died (see Results; Fig. 1). Further, female mallards are generally philopatric (review in Anderson et al. 1992). Since the introduced birds were not local, any surviving female would not consider the experimental lake as being her native one. Moreover, it is unlikely that a female would return to a lake where duckling mortality was very high. Waterfowl counts Following experimental introduction, each lake was visited daily for the Wrst 12 days and then every third day until day 24 (a total of 16 visits per lake). Remaining introduced birds and all wild waterfowl were counted on each visit, using the waterfowl point count method in Koskimies and Väisänen (1991). Most counts were made in the morning or in the evening, when duckling activity normally peaks (Ringelman and Flake 1980). No counts were made at night, and we also avoided very windy, rainy, hot, as well as cold weather. Considerable evort was made to approach the lakes without alarming the ducks. At each lake we did 20 min of still observation from a Wxed point. For practical reasons it was not feasible to totally randomise the visiting order of lakes. The latter comprised a northern and a southern cluster, between which the visit order was alternated. To minimise the distance driven, lakes were visited in a speciwed sequence within each cluster, the starting point of which was alternated between study days.

206 Oecologia (2006) 149:203 213 During the course of the study single or groups of wild mallards were seen on six of the lakes and single broods on two lakes in 2003. Teal (A. crecca) bred on four and common goldeneye (Bucephala clangula) on three lakes. Rare breeders were red-throated diver (Gavia stellata) (one lake), black-throated diver (Gavia arctica) (one lake), Slavonian grebe (Podiceps auritus) (two lakes), wigeon (A. penelope) (one lake), and tufted duck (Aythya fuligula) (two lakes). Potential breeders on single lakes were whooper swan (Cygnus cygnus), velvet scoter (Melanitta fusca), red-breasted merganser (Mergus serrator), and goosander (Mergus merganser), of which pairs or single individuals were observed. Models Introductions after which the female was never resighted were considered as unsuccessful. This occurred at least once at three lakes, reducing the sample size in a balanced modelling design to seven lakes for which there were complete data sets from both treatments (see below and Fig. 1). An encounter history for each duckling and each female was computed based on the 16 counts at each of the seven remaining lakes, i.e. day 0 12, 15, 18, and 21. In reality, day 24 was the last day of study, but data from that visit were entered as interval day 21 24. We obtained individual quantitative survival data for both categories of introduced mallards because: (1) wing-clipped females were easily distinguished from wild due to their lack of projecting primaries (the study was concluded before moult started), and (2) we made no observation indicating that any duckling permanently and successfully left its mother. We assumed missing birds to be dead because we had reduced the risk of ducks escaping from the lakes (see Study lakes). On a few occasions we observed more ducklings than on the preceding visit (on average, this happened 0 1.5 times per treatment group and year), in which case we adjusted the preceding count to correspond with the subsequent one. We argue that these events with missing birds were too rare and unimportant to consider capture recapture models including estimations of detection probabilities; accordingly we instead used the option known fate modelling to estimate survival in program MARK (version 4.1) (White and Burnham 1999). Intervals were set to either 1 (for the Wrst 12 days) or 3 (for day 15, 18, and 21) to get daily survival rates. Logit-link functions were used throughout because we added covariates in the design matrix. We Wrst ran a candidate set of a priori models, in which only the main evects (Density, Year, and Lake) and no covariates were included. Running models including Lake almost always led to over-parameterisation, and therefore this factor was left out in later modelling. However, because individual duckling survival cannot be assumed to be independent on a lake, nor within treatment category, the Lake factor was used along with Density in the separate model used only to explore and to account for over-dispersion (Burnham and Anderson 2002). Because of overparameterisation it was not possible to investigate over-dispersion on the brood level. To deal with the problem of potential interdependence, program MARK uses the variance inxation factor (Ç). By simulating the model with the most parameters, median Ç was estimated in a logistic regression option in program MARK. For each of 20 intermediate points, 50 iterations were run between the lower and upper boundaries (i.e. 1,100 iterations in all). The most complex model in the candidate set thus included Density, Year, and their interaction term. SimpliWcations of this model were run to Wnd the most parsimonious model. The latter was then used when constructing covariate models. Data for three covariates were entered in the design matrix in program MARK with one value for each interval: (1) Day after introduction (called Age in ducklings), (2) an index of Weather harshness, and (3) presence or absence of hens (Female). Day after introduction (Age) was relevant as a covariate because daily survival of mallard ducklings has been shown to increase non-linearly during the Wrst 3 weeks of life (e.g. Orthmeyer and Ball 1990; Mauser et al. 1994). To explore whether a nonlinear relationship of Day/Age was evident in our study too, such an evect was contrasted to an ordinary linear Day/Age evect. The latter was modelled by entering the study days for all lakes in the design matrix (i.e. 0 12, 15, 18, and 21). For the non-linear evect, we did not use a quadratic expression because such functions are humped with either a minimum or a maximum point and therefore not adequate for describing survival probability. To allow the daily survival function f(x) to increase gradually we instead dewned the arbitrarily chosen equation f(x) =a(1 b x ); where a is the start of the last interval after introduction (i.e. day 21), b is the slope of the graph (i.e. a value between zero and one, where the best Wtted value was used for ducklings and females, respectively), and x is the actual day after introduction. For the second covariate, data from two nearby oycial meteorological stations (Vindeln-Sunnansjönäs and Umeå airport) were used to create a weather harshness index. To do this we summed the z- standardised values of three variables: (1) temperature deviation, dewned as the diverence between the

Oecologia (2006) 149:203 213 207 Fig. 1 Mean proportion of introduced ducklings and females alive on experimental lakes. Filled circles denote low-density treatment lakes (n = 4 in 2002, n = 3 in 2003) and Wlled triangles high-density lakes (n = 3 in 2002, n =4 in 2003). Bars represent either the negative or the positive SE value alive (%) Females 100 80 60 40 20 100 80 60 40 20 0 0 5 10 15 20 25 0 0 5 10 15 20 25 100 100 80 80 alive (%) 60 60 Ducklings 40 20 40 20 0 0 5 10 15 20 25 Days after introduction 2002 0 0 5 10 15 20 25 Days after introduction 2003 mean temperature of a pre-visit interval and the highest mean diel temperature recorded during the entire study (i.e. 23.2 C on 16 July 2003), (2) mean wind speed in the pre-interval, and (3) precipitation in the pre-interval. High z-values thus denote harsh conditions (i.e. cold, windy, and rainy), for each variable as well as for the summed index. The last covariate (Female) was only used when modelling duckling survival to control for the fact that duckling survival is likely to depend upon their mother s. It was not necessary to add wild waterfowl as covariates, as there was no diverence between treatment groups in the cumulative use of a lake in any of the taxonomic groups considered: adult ducks (Anatinae, Aythyinae, and Merginae): T =9.00, P = 0.40; duck broods: T =4.50, P = 0.42; and other adult waterbirds (Gaviiformes, Podicipediformes, Cygninae): T =7.00, P = 0.46; Wilcoxon paired sample tests, n = 7 in all cases, two-sided probabilities. All covariates were added to the most parsimonious model in the candidate set, and their signiwcance was evaluated by comparison with a priori alternate models. The expression variable a variable b is used in this paper to denote an interaction term as well as its constituents in isolation (e.g. Density Year means Density + Year + the interaction between Density and Year). We used Akaike s information criterion (AICc) to rank models. In the case of over-dispersion we instead used the quasi-likelihood AIC (QAICc) (Akaike 1973; Burnham and Anderson 2002). Models were considered to diver in how well they describe data when the diverence in AICc (or QAICc) was 2.00 (Burnham and Anderson 2002). Model outputs for duckling and hen survival are given in Table 1 and include the null-model (with intercept only), the candidate models, and the covariate models. The slope (β-value) of individual factors included in the highestranking models is given in the main text. When the highest-ranking models for duckling and hen survival, respectively, had been identiwed, the relative importance of each of their variables was assessed. To do this we Wrst ran all possible models including the factors in the models with ΔAICc or ΔQAICc (diverence between AICc or QAICc of the current model and the minimum AICc or QAICc) < 2.00 (see Table 1). The weights of all models including a speciwc variable were then summed to obtain a value expressing the likelihood that this variable truly belonged in the highest-ranking models. Results Duckling survival Simulating the model including Density and Lake (duckling model 9 in Table 1; observed Ç =2.75) showed that duckling survival was over-dispersed. Median Ç was estimated to be 2.14 (SE = 0.05), with the lower boundary set to 1.00 and the upper to 3.00.

208 Oecologia (2006) 149:203 213 Table 1 Model output for survival in mallard ducklings and females Model a AICc or QAICc ΔAICc or ΔQAICc b w i K Deviance Ducklings 1. S Density Year + N.L.Age + Weather + Female 498.24 0.00 0.36 7 163.32 2. S Density Year + Weather + Female 498.71 0.47 0.28 6 165.82 3. S Density Year + L.age + Weather + Female 499.18 0.94 0.22 7 164.26 4. S Density Year + N.L.age + Female 501.44 3.20 0.07 6 168.55 5. S Density Year + N.L.age + Weather 501.93 3.68 0.06 6 169.04 6. S Density + Year 510.31 12.06 0.00 3 183.48 7. S Density Year 510.44 12.20 0.00 4 181.60 8. S Year + N.L.age + Weather + Female 511.52 13.28 0.00 5 180.66 9. S Density + Lake 511.60 13.35 0.00 8 174.64 10. S Density 513.52 15.28 0.00 2 188.72 11. S Year 526.14 27.90 0.00 2 201.33 12. S (.) 535.28 37.04 0.00 1 212.48 Females 1. S Year + L.Day + Weather 138.28 0.00 0.23 4 116.29 2. S Year + L.Day 138.38 0.10 0.22 3 118.44 3. S Year + N.L.Day + Weather 138.94 0.67 0.16 4 116.96 4. S Year 139.51 1.23 0.12 2 121.61 5. S Density + Year + L.Day + Weather 140.32 2.04 0.08 5 116.28 6. S Year + Weather 140.38 2.11 0.08 3 120.45 7. S Density + Year 141.47 3.19 0.05 3 121.54 8. S (.) 142.58 4.31 0.03 1 126.71 9. S Density Year 143.31 5.04 0.02 4 121.33 10. S Density 144.59 6.31 0.01 2 126.69 Due to over-dispersion the variance inxation factor (Ç) was used for ducklings, but this was not the case for female data. Accordingly, quasi-likelihood Akaike s information criterion (QAICc) values were used for the former, while AICc values for the latter when ranking models (higher rank with descending values) w i Model weight, K number of parameters, Deviance diverence in 2 log(likelihood) of the current model and 2 log(likelihood) of the saturated model a Models include intercept [(.)], main evects of Density, Year, and Lake, as well as covariates: (1) a linear (L.) evect of either age (ducklings: L.Age) or day after introduction (females: L.Day), (2) same as previous but non-linear (N.L.) (N.L.Age/Day), (3) Weather, and for duckling models (4) absence or presence of the hen (Female). The symbol denotes the inclusion of an interaction term as well as its constituents separately. Further dewnitions of covariates and details about the modelling procedure are given in Materials and methods b DiVerence between AICc or QAICc of the current model and the minimum AICc or QAICc value Of the models in the candidate set the one including Density and Year turned out to be the best (model weight, w i, = 0.47). The most general model also including the interaction between Density and Year had a high w i, too, (0.44) and we therefore used it when constructing the covariate models. All other models in the candidate set had w i <0.09. To see whether a linear or a non-linear Age evect approximated data the best we compared duckling models 1 and 3 in Table 1. The non-linear evect was optimised when b = 0.6, but because the diverence in QAICc between the two factors was < 2.00 (0.94) their inxuence on survival was about the same. By in turn considering the model including all covariates (Age, Weather, and Female), as well as simpliwed covariate models, we found that the highest-ranking model included all covariates along with the main evects; Weather and Female both improved the power of the model (compare model 1 with models 4 and 5 in Table 1), whereas Age evects gave ΔQAICc < 2.00 and thus neither improved nor lowered the descriptive power of the model (for non-linear Age, compare model 1 with model 2 in Table 1). Duckling model 8 was run to study how QAICc would change if the Density evect were excluded from the highest-ranking model. The observed diverence in QAICc (13.28, Table 1) shows that Density indeed was extremely important to the model s overall power to describe data. Examination of the β-values in the highest-ranking model (duckling model 1 in Table 1) showed that daily survival was higher in the low-density treatment than in the high-density one (β = 1.29; SE = 0.35), higher in 2002 than in 2003 (β = 0.46; SE = 0.46), higher if hens were present (β = 1.18; SE = 0.49), and lower in harsher weather (β = 0.15; SE = 0.07). Further examination of the interaction between Density and Year conwrmed that there was a negative evect of density in

Oecologia (2006) 149:203 213 209 2002 but not in 2003. This is visualised in Fig. 1, in which survival graphs for low and high densities of ducklings diverged more in 2002 than in 2003. As described above, the two Age covariates were almost equal in their Wt with data; the β-value was 0.03 (SE = 0.02) for non-linear Age and 0.04 (SE = 0.03) for linear Age (duckling models 1 and 3 in Table 1). The relative importance of each variable, i.e. the likelihood that it truly belonged in the highest-ranking models was (weight sum in brackets; see Models): Density (1.00), Female (0.92), Weather (0.86), non-linear Age (0.63), interaction between Density and Year (0.41), linear Age (0.37), Year (0.35). Hen survival A Lake evect could not be considered in the candidate models for female survival, as this would have resulted in unestimable parameters. Instead, the model including evects of Density, Year and their interaction (model 9 for female survival in Table 1) was investigated for overdispersion, with an observed Ç value of 0.67. Simulating for median Ç estimation gave a Ç value below 1.00, too. As Ç cannot be < 1.00 (there is no such thing as underdispersion ), we concluded that female survival data were not over-dispersed. Accordingly, we used AICc instead of QAICc values in the subsequent analysis. As for ducklings, the models in the candidate set were combinations of Density, Year, and their interaction term. The model including Year only had a higher w i (0.55) compared with the other models (w i <0.21), and therefore Year was the only main evect used when constructing covariate models. Linear and non-linear Day (Age) (b =0.1) evects gave very similar AICc values (cf. models 1 and 3 for females in Table 1). Both covariates were represented in the highest-ranking models (i.e. linear and non-linear Day as well as Weather; ΔAICc < 2.00, see Table 1). Model 5 for females divers from model 1 by including Density, which lowered the AICc by 2.04 units, thus implying that it did not avect survival signiwcantly. Studying the β-values of the parameters in the highest-ranking models revealed that the Year evect was similar to that in ducklings, with daily survival being higher in 2002 than in 2003 (β = 1.08; SE = 0.47; Fig. 1). Harsh weather had a negative inxuence on survival (β = 0.15; SE = 0.10) and the Day evect was negative, too (linear, 0.07, SE = 0.04; non-linear, 0.12, SE = 0.09). The relative importance of each variable, i.e. the likelihood that it truly belonged in the highest-ranking models was: Year (0.87), linear Day (0.60), Weather (0.55), non-linear Day (0.45). Discussion Duckling survival As shown by the model output and by the survival functions, per capita survival in mallard ducklings was negatively density dependent. The present study and that of Elmberg et al. (2005) are the Wrst experiments to conwrm previous suggestions that survival in breeding dabbling ducks may be truly regulated (Weller 1979; Hill 1984; Kaminski and Gluesing 1987; Elmberg 2003; Elmberg et al. 2003). Moreover, the present study indicates that the evect of density can be yeardependent (Table 1, Fig. 1). The latter is interesting and important considering the increasingly supported view that stochastic evects may be signiwcant for population growth and dynamics (Coulson et al. 2004; Saether and Engen 2004). As a consequence, density dependence and regulation of populations may be more easily overlooked if such processes are masked intermittently by stochastic events. For example, nonexperimental long-term data from Finland do not corroborate the hypothesis that breeding success in mallards is density dependent (Pöysä 2001; Elmberg et al. 2003). We think stochasticity and environmental noise may have been important to the outcome of the present study, too. This interpretation is supported by the occurrence of a year evect and the interaction between Density and Year. Adverse weather, representing environmental stochasticity, is known to decrease survival in mallard ducklings (Koskimies and Lahti 1964; Marcström 1966; Krapu et al. 2000). Conditions during the Wrst 1 or 2 weeks of duckling life are especially crucial to survival (Hill et al. 1987; Orthmeyer and Ball 1990; Rotella and Ratti 1992b; Sargeant and Raveling 1992; Pietz et al. 2003). Bad weather per se had an impact on duckling survival in our study, and virtually all mortality occurred during the Wrst 12 days (Fig. 1). It is important to note that the inconsistent evect of Density between years was not related to harsh weather. Daily survival of ducklings was higher in 2002, when weather conditions for ducks were worse, than in 2003 (mean of the weather index for the Wrst 12 days: 2002, 0.10, SD = 2.15; 2003, 0.72, SD = 1.92). Monitoring data from nearby Finland suggest that although both were relatively good mallard years in Fennoscandia, duckling productivity was actually higher in 2002 than in 2003 (Pöysä et al. 2003). We speculate that some other factor must have overridden the evect of density and thus contributed to the pattern of lower survival rate in 2003. Possibly, this result may be inxuenced by higher survival estimates for females in 2002 compared to

210 Oecologia (2006) 149:203 213 2003. Although the mallard is a precocial species, very young ducklings are generally thought to depend on their mother to protect them from cold, heat, rain as well as predators (Cramp and Simmons 1977). Our experimental data support this opinion, and that the evect of hen survival may override the density evect. Hen survival Although hen mortality in mallards is known to be disproportionately high during the breeding season (e.g. Sargeant and Raveling 1992), our study did not show any regulatory response at high densities. It is well known that hens do not rely as much as ducklings on emerging insects as food (Krapu and Reinecke 1992; Sedinger 1992; Nummi et al. 2000), and also that the availability of emerging insects is relatively weather dependent (Sjöberg and Danell 1982). This and the fact that hens are more robust than young ducklings may explain the lack of negative density dependence. Survival rate of hens decreased as a function of day after introduction, implying a lower mortality rate in the early part of the study period. We think that this result may be an artefact resulting from hens leaving the lake only at a point when all ducklings had died, which was more likely to occur some time into the experimental period. A biologically more relevant and alternative explanation might be that food shortage had a cumulative evect, resulting in direct or indirect mortality in hens only after some time. Sources of error One motive for using wing-clipped introduced birds is that they do not have any information about nearby lakes or the area in general. This reduces the risk that they guide their brood to another lake, like wild mallards sometimes do (Hill et al. 1987; Dzus and Clark 1997b; Sayler and Willms 1997). Wing-clipped birds may, however, experience an overall lower survival if they are less capable of escaping from predators. On four occasions, red fox (Vulpes vulpes) or pike (Esox lucius) were observed to attack broods, but the hens detected the threat early enough to escape. Nevertheless, successful predation events were observed thrice: (1) a buzzard (Buteo buteo) took one duckling from a brood, (2) a pike took one duckling that had left its hen and brood, and (3) a wing-clipped hen was found on the shore killed by an unknown predator. We use the term mortality for birds that were not re-sighted. With respect to ducklings this is probably appropriate because the decline in brood size was usually gradual. If entire broods had left their lake instead of being dead, then we would have noticed this as a sudden and total mortality. Despite the precautions we took to reduce escape probability, it is possible that some hens, though, left their lake when brood loss was complete. In summary, we are convinced that our conclusions about duckling mortality are valid, but we acknowledge that we may not have measured survival sensu stricta in hens. Regardless, the model for female survival remains valid in terms of lake utilisation and local survival. Conclusions We did not design the study to explore all possible processes that may induce a density-dependent survival response, but we argue that food resource limitation is one likely agent when it comes to ducklings (cf. White 2004). Food limitation has indeed been demonstrated in the study system (Gunnarsson et al. 2004, see also Sjöberg et al. 2000), and low food abundance in many oligotrophic boreal lakes may thus lead to densitydependent regulation of breeding success. We note that although the diverence in duckling survival between the two densities was obvious during the Wrst 2 weeks, all introduced ducklings were dead in both treatment groups and years by the end of the experiment (Fig. 1). We think food limitation eventually resulted in this total mortality, but our experiment succeeded in revealing the preceding density-dependent mortality pattern. In this respect our study design may not totally resemble a natural situation, in which a wild mallard female may have attempted taking her brood to better foraging lakes before all ducklings died (e.g. Ball et al. 1975; Rotella and Ratti 1992a; Dzus and Clark 1997b). This does not, however, invalidate the conclusion that density-dependent regulation of breeding success may prevail on boreal lakes; density-dependent mortality can occur anytime during the brood stage. Moreover, lake change and overland travel may alternatively increase duckling mortality (Ball et al. 1975; Rotella and Ratti 1992a, but see Dzus and Clark 1997b). Hence, if lake change is a response to brood density, as has been suggested for the buzehead (Bucephala albeola) (Gauthier 1987), it may strengthen instead of alleviating the evect of densitydependent mortality. Surprisingly, mortality rate in the present study was higher than in experimentally starved ducklings (Marcström 1966). As the latter study precluded predation, duckling mortality as we measured it was probably a combination of starvation and predation, a view further supported by the observed predation events. We

Oecologia (2006) 149:203 213 211 did not measure disease and other sources of mortality, but they, too, may have inxuenced survival. The present study supports the idea that density dependence occurs intermittently and partly as a result of exogenous feedback. For example, in some years weather may be severe enough to induce densitydependent survival through food limitation or increased predation. In good years, on the other hand, there may be no density-dependent evect on survival. Our work delivers two messages for sustainable management of harvested populations. Although what is a natural density can always be discussed, there is now experimental evidence that survival of breeding mallards may be density dependent. The hypothesis that there is a demographic response to hunting mortality (i.e. compensatory mortality and/or compensatory natality) rests on the assumption that density-dependent processes occur in natural populations (Anderson and Burnham 1976; Boyce et al. 1999). Such a view Wts with the present results if an increased harvest is compensated for by higher breeding success in the surviving fraction of the population. Secondly, if breeding success is regulated in some years but not in others, adaptive management is of great value, especially so a feedback process adjusting bag limits to recent breeding success (Nichols et al. 1995; Johnson et al. 2002). Acknowledgements We thank Per Wedholm, Thomas Trygg, Eric Andersson, and Åke Nordström for Weldwork assistance. All landowners are gratefully acknowledged for access, and so is Göran Karlsson at the Boda Game Farm for supplying mallards. Paul Flint helped us Wnding the best way to analyse our data by unselwshly devoting considerable time and evort to guiding us in program MARK. Jim Nichols and an anonymous referee provided useful comments. The study was supported by grants V- 124-01 and V-98-04 from the Swedish Environmental Protection Agency to Johan Elmberg. All handling of birds was done in accordance with permit A1-02 from the Research Animals Ethics Board at the Court of Appeal for North Sweden, Umeå. References Akaike H (1973) Information theory and an extension of the maximum likelihood principle. In: Petran BN, Csáki F (eds) Proceedings of the 2nd International Symposium on Information Theory. Akadémiai Kiadi, Budapest, pp 267 281 Almaraz P, Amat JA (2004) Multi-annual spatial and numeric dynamics of the white-headed duck Oxyura leucocephala in southern Europe: seasonality, density dependence and climatic variability. J Anim Ecol 73:1013 1023 Anderson DR, Burnham KP (1976) Population ecology of the mallard. VI. The evect of exploitation on survival. US Dep Int Fish Wildl Ser Res Publ 128 Anderson MG, Rhymer JM, Rohwer FC (1992) Philopatry, dispersal, and the genetic structure of waterfowl populations. In: Batt BDJ, Afton AD, Anderson MG, Ankney CD, Johnson DH, Kadlec JA, Krapu GL (eds) Ecology and management of breeding waterfowl. University of Minnesota Press, Minneapolis, Minn., pp 365 395 Åström M, Lundberg P, Lundberg S (1996) Population dynamics with sequential density-dependence. Oikos 75:174 181 Ball IJ, Gilmer DS, Cowardin LM, Riechmann JH (1975) Survival of wood duck and mallard broods in north-central Minnesota. J Wildl Manage 39:776 780 Berryman AA (2004) Limiting factors and population regulation. Oikos 105:667 670 Berryman A, Turchin P (1997) Detection of delayed density dependence: comment. Ecology 78:318 320 Berryman AA, Lima Arce M, Hawkins BA (2002) Population regulation, emergent properties, and a requiem for density dependence. Oikos 99:600 606 Both C, Visser ME (2003) Density dependence, territoriality, and divisibility of resources: from optimality models to population processes. Am Nat 161:326 336 Boyce MS, Sinclair ARE, White GV (1999) Seasonal compensation of predation and harvesting. Oikos 87:419 426 Burnham KP, Anderson DR (1984) Tests of compensatory vs. additive hypothesis of mortality in mallards. Ecology 65:105 112 Burnham KP, Anderson DR (2002) Model selection and multimodel inference: a practical information theoretic approach, 2nd edn. Springer, Berlin Heidelberg New York Caughley G, Gunn A (1996) Conservation biology in theory and practice. Blackwell Science, Massachusetts, Md. Caughley G, Sinclair ARE (1994) Wildlife harvesting. In: Caughley G, Sinclair ARE (eds) Wildlife ecology and management. Blackwell, Oxford, pp 279 290 Chura NJ (1961) Food availability and the preferences in juvenile mallards. Trans N Am Wildl Nat Res Conf 26:121 134 Clark RG, Nudds T (1991) Habitat patch size and duck nesting success: the crucial experiments have not been performed. Wildl Soc Bull 19:534 543 Coulson T, Rohani P, Pascual M (2004) Skeletons, noise and population growth: the end of an old debate? Trends Ecol Evol 19:359 364 Cramp S, Simmons KEL (eds) (1977) The birds of the western Palearctic, vol 1. Clarendon Press, Oxford Dzus EH, Clark RG (1997a) Brood size manipulation in mallard ducks: evects on duckling survival and brooding eyciency. Ecoscience 4:437 445 Dzus EH, Clark RG (1997b) Overland travel, food abundance, and wetland use by mallards: relationships with ovspring survival. Wilson Bull 109:504 515 Elmberg J (2003) Density-dependent breeding success in mallards Anas platyrhynchos on a eutrophic lake. Wildl Biol 9:67 73 Elmberg J, Nummi P, Pöysä H, Sjöberg K (2003) Breeding success of sympatric dabbling ducks in relation to population density and food resources. Oikos 100:333 341 Elmberg J, Gunnarsson G, Pöysä H, Sjöberg K, Nummi P (2005) Within-season sequential density dependence regulates breeding success in mallards Anas platyrhynchos. Oikos 108:582 590 Fransson T, Pettersson J (2001) Svensk ringmärkningsatlas (Swedish bird ringing atlas). Ljungföretagen Tryckeri, Örebro Gauthier G (1987) Brood territories in buzeheads: determinants and correlates of territory size. Can J Zool 65:1402 1410 Grimm V, Uchmanski J (2002) Individual variability and population regulation: a model of the signiwcance of within-generation density dependence. Oecologia 131:196 202 Gunnarsson G, Elmberg J, Sjöberg K, Pöysä H, Nummi P (2004) Why are there so many empty lakes? Food limits survival of mallard ducklings. Can J Zool 82:1698 1703

212 Oecologia (2006) 149:203 213 Harrison S, Cappuccino N (1995) Using density manipulation experiments to study population regulation. In: Cappuccino N, Price PW (eds) Populations dynamics. Academic Press, San Diego, Calif., pp 131 147 Hill DA (1983) Compensatory mortality in the mallard. Game Conserv Annu Rev 14:87 92 Hill DA (1984) Population regulation in the mallard (Anas platyrhynchos). J Anim Ecol 53:191 202 Hill D, Wright R, Street M (1987) Survival of mallard ducklings Anas platyrhynchos and competition with Wsh for invertebrates on a Xooded gravel quarry in England. Ibis 129:159 167 Hixon MA, Pacala SW, Sandin SA (2002) Population regulation: historical context and contemporary challenges of open vs. closed systems. Ecology 83:1490 1508 Johnson FA, Kendall WL, Dubovsky JA (2002) Conditions and limitations on learning in the adaptive management of mallard harvests. Wildl Soc Bull 30:176 185 Jonzén N, Lundberg P (1999) Temporally structured densitydependence and population management. Ann Zool Fenn 36:39 44 Kaminski RM, Gluesing EA (1987) Density- and habitat-related recruitment in mallards. J Wildl Manage 51:141 148 Koskimies J, Lahti L (1964) Cold-hardiness of the newly hatched young in relation to ecology and distribution in ten species of European ducks. Auk 81:281 307 Koskimies P, Väisänen R (1991) Waterfowl point count. In: Koskimies P, Väisänen RA (eds) Monitoring bird populations. Zoological Museum, Finnish Museum of Natural History, Helsinki, pp 41 44 Krapu GL, Reinecke KJ (1992) Foraging ecology and nutrition. In: Batt BDJ, Afton AD, Anderson MG, Ankney CD, Johnson DH, Kadlec JA, Krapu GL (eds) Ecology and management of breeding waterfowl. University of Minnesota Press, Minneapolis, Minn., pp 1 29 Krapu GL, Pietz PJ, Brandt DA, Cox RR (2000) Factors limiting Mallard brood survival in prairie pothole landscapes. J Wildl Manage 64:553 561 Larivière S, Messier F (1998) EVect of density and nearest neighbours on simulated waterfowl nests: can predators recognize high-density nesting patches? Oikos 83:12 20 Marcström V (1966) Mallard ducklings (Anas platyrhynchos L.) during the Wrst days after hatching. Swed Wildl 4:343 368 Mauser DM, Jarvis RL, Gilmer DS (1994) Survival of radiomarked mallard ducklings in northeastern California. J Wildl Manage 58:82 87 Nams VO (1997) Density-dependent predation by skunks using olfactory search images. Oecologia 110:440 448 Newton I (1998) Population limitation in birds. Academic Press, San Diego, Calif. Nichols JD (1991) Responses of North American duck populations to exploitation. In: Perrins CM, Lebreton J-D, Hirons GJM (eds) Bird population studies: relevance to conservation and management. Oxford University Press, Oxford, pp 498 525 Nichols JD, Johnson FA, Williams BK (1995) Managing North American waterfowl in the face of uncertainty. Annu Rev Ecol Syst 26:177 199 Nicholson AJ (1933) The balance of animal populations. J Anim Ecol 2:132 178 Nummi P, Saari L (2003) Density-dependent decline of breeding success in an introduced, increasing mute swan population. J Avian Biol 34:105 111 Nummi P, Sjöberg K, Pöysä H, Elmberg J (2000) Individual foraging behaviour indicates resource limitation: an experiment with mallard ducklings. Can J Zool 78:1891 1895 Orthmeyer DL, Ball IJ (1990) Survival of mallard broods on Benton Lake National Wildlife Refuge in northcentral Montana. J Wildl Manage 54:62 66 Paradis E, Baillie SR, Sutherland WJ, Gregory RD (2002) Exploring density-dependent relationships in demographic parameters in populations of birds at a large spatial scale. Oikos 97:293 307 Pehrsson O (1979) Feeding behaviour, feeding habitat utilization, and feeding eyciency of mallard ducklings (Anas platyrhynchos L.) as guided by a domestic duck. Swed Wildl 10:194 217 Pietz PJ, Krapu GL, Brandt DA, Cox RR Jr (2003) Factors avecting gadwall brood and duckling survival in prairie pothole landscapes. J Wildl Manage 67:564 575 Pollard E, Lakhani KH, Rothery P (1987) The detection of density-dependence from a series of annual censuses. Ecology 68:2046 2055 Pöysä H (2001) Dynamics of habitat distribution in breeding mallards: assessing the applicability of current habitat selection models. Oikos 94:365 373 Pöysä H, Pesonen M (2003) Density dependence, regulation and open-closed populations: insights from the wigeon, Anas penelope. Oikos 102:358 366 Pöysä H, Wikman M, Lammi E, Väisänen R (2003) Vesilintukannat ennallaan poikastuotossa vaihtelua. Riistantutkimuksen Tiedote 188:1 7 Pöysä H, Elmberg J, Gunnarsson G, Nummi P, Sjöberg K (2004) Ecological basis of sustainable harvesting: is the prevailing paradigm of compenstory mortality still valid? Oikos 104:612 615 Ray C, Hastings A (1996) Density dependence: are we searching at the wrong spatial scale? J Anim Ecol 65:556 566 Reitsma L (1992) Is nest predation density dependent? A test using artiwcial nests. Can J Zool 70:2498 2500 Ringelman JK, Flake LD (1980) Diurnal visibility and activity of blue-winged teal and mallard broods. J Wildl Manage 44:822 829 Rotella JJ, Ratti JT (1992a) Mallard brood movements and wetland selection in southwestern Manitoba. J Wildl Manage 56:508 515 Rotella JJ, Ratti JT (1992b) Mallard brood survival and wetland habitat conditions in southwestern Manitoba. J Wildl Manage 56:499 507 Ruesink JL (2000) Assesing density dependence using intra-annual counts. Ecology 81:3377 3390 Saether B-E, Engen S (2004) Stochastic population theory faces reality in the laboratory. Trends Ecol Evol 19:351 353 Sargeant AB, Raveling DG (1992) Mortality during the breeding season. In: Batt BDJ, Afton AD, Anderson MG, Ankney CD, Johnson DH, Kadlec JA, Krapu GL (eds) Ecology and management of breeding waterfowl. University of Minnesota Press, Minneapolis, Minn., pp 396 422 Sayler RD, Willms MA (1997) Brood ecology of mallards and gadwalls nesting on islands in large reservoirs. J Wildl Manage 61:808 815 Sedinger JS (1992) Ecology of prexedgling waterfowl. In: Batt BDJ, Afton AD, Anderson MG, Ankney CD, Johnson DH, Kadlec JA, Krapu GL (eds) Ecology and management of breeding waterfowl. University of Minnesota Press, Minneapolis, Minn., pp 109 127 Shenk TM, White GC, Burnham KP (1998) Sampling-variance evects on detecting density dependence from temporal trends in natural populations. Ecol Monogr 68:445 463 Sjöberg K, Danell K (1982) Feeding activity of ducks in relation to diel emergence of chironomids. Can J Zool 60:1383 1387

Oecologia (2006) 149:203 213 213 Sjöberg K, Pöysä H, Elmberg J, Nummi P (2000) Response of mallard ducklings to variation in habitat quality an experiment of food limitation. Ecology 81:329 335 Street M (1977) The food of mallard ducklings in a wet gravel quarry, and its relation to duckling survival. Wildfowl 28:113 125 Sugden LG, Beyersbergen GW (1986) EVect of density and concealment on American crow predation of simulated duck nests. J Wildl Manage 50:9 14 Talent LG, Jarvis RL, Krapu GL (1983) Survival of mallard broods in south-central North Dakota. Condor 85:74 78 Turchin P (1990) Rarity of density-dependence or population regulation with lags? Nature 344:660 663 Vickery WL, Nudds TD (1984) Detection of density-dependent evects in annual duck censuses. Ecology 65:96 104 Weller MW (1979) Density and habitat relationships of bluewinged teal nesting in northwestern Iowa. J Wildl Manage 43:367 374 White TCR (2004) Limitation of populations by weather-driven changes in food: a challenge to density-dependent regulation. Oikos 105:664 666 White GC, Burnham KP (1999) Program MARK: survival estimation from populations of marked animals. Bird Study Suppl 46:120 139 Wilson EO (2000) Density dependence. In: Wilson EO (ed) Sociobiology the new synthesis. Belknap Press of Harvard University Press, Cambridge, Mass., pp 82 89 Woiwod IP, Hanski I (1992) Patterns of density dependence in moths and aphids. J Anim Ecol 61:619 629