In Finland and much of Fennoscandia, both long-term breeding success and size of black grouse Tetrao tetrix populations have faced severe declines during the past decades (Helle et al. 2002, Ludwig et al. 2006). Declines in grouse breeding success in Finland and elsewhere are commonly associated with increased rates of nest loss due to increased predation (Marcström et al. 1988, Baines 1991, Caizergues & Ellison 1997, Kurki et al. 1997, 1998, Kauhala & Helle 2002, Ludwig 2007), but the processes determining early brood survival are generally more difficult to study and are therefore less well known. Survival of juveniles from hatching until recruitment to the breeding population in the following year is a critical determinant of population growth in many species of grouse (Tetraonidae; Sandercock et al. 2005). Juvenile mortality rates in grouse are commonly the highest during approximately the first two weeks after hatching (Ludwig et al. 2006, Hannon & Martin 2006), and are therefore an important component of the variation in annual recruitment. The main causes of early brood mortality are generally considered to be predation, unfavourable weather conditions as well as chick and/or hen quality. For instance, high predation induced chick mortality during the first 10-14 days after hatching has been reported for capercaillie T. urogallus (Kastdalen & Wegge 1991), red grouse Lagopus lagopus scoticus (Park et al. 2002), ruffed grouse Bonasa umbellus (Larson et al. 2001), white-tailed ptarmigan L. leucura (Braun et al. 1993) and sharp-tailed grouse Tympanuchus phasianellus (Manzer 2004). On the other hand, exposure to cold and/or rainy conditions has been shown to be a major source of early mortality in willow ptarmigan L. lagopus (Erikstad & Andersen 1983, Erikstad 1985). Hen condition, which may be affected by e.g. weather conditions in the pre-nesting period, may further affect egg quality, and hence also chick quality and viability (e.g. Siivonen 1958, Moss & Watson 1984, Hannon & Martin 2006).

The role of weather conditions for breeding success in Finnish forest grouse such as black grouse, capercaillie or hazel grouse B. bonasia has been controversial (Lindström 1996). Recent results, however, have demonstrated that weather conditions, especially temperature, experienced by black grouse chicks during the first 10 days after hatching, strongly influence the annual breeding success (Ludwig et al. 2006). These results also underline the importance of accurately estimation and matching of the timing and the length of the weather period with the timing of hatching if one is to gain biologically meaningful inference. The effect of weather has further been shown to be exacerbated by climate change. Due to an advance of spring phenology, black grouse have advanced the timing of both the breeding and the hatching. Problematically, early summer has not advanced, i.e. has not warmed up, and hence grouse chicks are increasingly hatching under colder conditions (Ludwig et al. 2006). Little is known, however, about how individual properties (e.g. body mass of hen or chicks) and properties of the brood rearing environment (e.g. vegetation characteristics or weather conditions) interact with each other and/or with fluctuating abundance of predators to determine early brood survival.

Human alteration of brood habitat can influence early juvenile survival through increased predator densities, direct disturbance and changes in the spatial and temporal distribution of food or lack of cover (e.g. Miquet 1990, Kurki et al. 2000, Storch 2000). In central and southern Finland, the majority of black grouse breeding habitat has been heavily altered by modern forestry practices such as clear-cutting and large-scale drainage during the past decades (Kurki et al. 2000, Ludwig et al. 2008), and the consequences of it on black grouse nesting success have recently been demonstrated (Ludwig 2007). Until now, little knowledge exists on how interactions of humanly altered habitat properties, weather, predators and individual characteristics of hens and chicks affect early brood survival.

The main scope of our study was to evaluate both potential direct and indirect causes of early mortality in black grouse broods in a managed forest landscape in central Finland. Using an information-theoretic approach and a unique data set on the individual brood level, we aimed at identifying the most important explanatory variables and the extent to which variables interact with each other. We took temporal and spatial variation into account.

Material and methods

Statistical analyses

We used the nest survival model procedure and the logit link function of program MARK to estimate daily brood survival rates (DSR). The nest survival procedure is suitable for modelling brood survival within a given period of time, as long as the fate of the brood is treated as a categorical variable (brood alive = at least one chick alive). The model allows for variable brood exposure time, i.e. accounts for the problem of counting broods of different age (Mayfield 1961). For every model, MARK produces the Akaike Information Criterion corrected for small sample size (AIC_c), allowing models to be ranked according to the amount of information loss (Burnham & Anderson 2002). For a given set of candidate models, the model with the lowest AIC_c is considered the best model, i.e. the model with the least information loss. We also presented the difference in AIC_c (▵AIC_c), Akaike weight (w) and deviance (dev; in respect to the full model).

We checked all variables for multicollinearity, and whenever pairs of independent variables were strongly correlated (r ≥ 0.5) one would be excluded based on the importance of the variable as established through previous research. The variables we used in the candidate models were the following: female age (two categories: first breeder & older), body mass of the hen (in g), hatching date (1 = earliest nest of the study), microfilaria (n), trypanosoma (presence/absence), hematocrites (in % red blood cells), clutch size (n), mean chick mass (in g), bilberry cover (in %), heather cover (in %), grassy vegetation cover (in %), tree density (number/ha), shrub density (number/ha), canopy coverage (in %), visibility (in metres), drainage (two categories: drained/not drained), rainfall (in mm), temperature (in °C) and predator track density (tracks/24 hours/10 km). For the analysis, we arcsine transformed variables that express proportions.

We accounted for variability induced by study site and year by introducing a dummy control variable for each year and site in all of the models. In this way, site and year dependent variation was effectively standardised across all candidate models, and the differences between candidate models largely reflect differences in the predictive ability of the explanatory variables. For ease of interpretation we did not include control variables in the tables.

In addition to year and site, we included a maximum of two main effects. This we did because large numbers of predictor variables with respect to n response variables result in data too thin to accurately estimate parameters (Hosmer & Lemeshow 1989), thereby compromising comparison of different effects. Furthermore, it is well known that the variance of the model increases exponentially with the number of parameters to be estimated (Burnham & Anderson 2002). Using all possible combinations with up to two main effects, we developed a total of 190 candidate models to predict early brood survival. In addition to the main effect models, we introduced biologically meaningful interactions between main effects whenever considered necessary, to investigate whether the effect of a variable was dependent on the effect of some other variable.

In order to establish a subset of ‘best‘ models, we selected all models that performed better, in terms of AIC_c, than the model including only site and year variables. That is, from an information theoretic point of view, we are interested in all the models i for which the following condition is fulfilled:

where AIC_cthreshold represents the AIC_c of the site + year model. From that group of models, even the worst will be better, in terms of its AIC_c, than the threshold model (Burnham & Anderson 2002). Such a procedure commonly suffers from model selection uncertainty (Burnham & Anderson 2002). It needs to be understood, however, that we are mostly interested in the relative importance of explanatory variables, rather than a particular ‘best‘ model. We therefore made further inference based on the entire subset of selected models. We present Akaike weights w_i for each variable that is present in at least one selected model, and average the parameter estimates (and confidence intervals) across the models where a given variable occurs (Burnham & Anderson 2002). Variables with 95% confidence intervals that include zero can be considered as not significant.

Results

During 2003-2005, we monitored a total of 81 successfully hatched broods (out of 136 nests) and the number of locations per brood during the first 10 days varied within 3-9. Early brood loss varied between years (χ² = 6.19, P = 0.045); the frequencies of complete brood loss during the first 10 days were 0.36, 0.26 and 0.08 for 2003, 2004 and 2005, respectively. Mean values for all variables, both for successful and unsuccessful broods, as well as t-tests and P-values for comparisons between groups, are shown in Table 1.

Out of more than 200 candidate models (190 main effect models plus the interaction models), 14 models fulfilled the condition AIC_cthreshold - AIC_ci ≥ 2, where AIC_cthreshold represents the AIC_c of the model including site and year only (Table 2). The set of best models can be clearly structured into two groups; three out of the five best models were interaction models involving the variables temperature, hatching date and chick body mass. All other models included main effects only, and each of the remaining models included the variable bilberry cover. Inspection of the relative Akaike weights (w), revealed strong model selection uncertainty, the weight of the best model being only 0.173. However, considering the number of models in the candidate set this is hardly surprising. We therefore gained further inference by calculating relative weights, averaged β-estimates as well as averaged upper and lower confidence intervals for all variables being present in at least one of the selected models (Table 3).

Being present in 11 out of 14 models, the variable bilberry cover was clearly the best predictor (Fig. 1). The averaged β-estimate as well as an averaged confidence interval that excludes zero, suggest a strong positive effect of bilberry cover on DSR of broods during the first 10 days after hatching. We found no interaction of bilberry with other variables. The effects of temperature, hatching date and chick mass, on the other hand, were clearly dependent on interactions with each other, as none of the respective main effect models was included in the candidate set. As is often the case with interactions, the β-estimates may not be indicative of the effect's direction (positive/negative). In order to understand it better, we visualised DSR as a function of the two most important interactions, i.e. temperature*hatching date and temperature*chick mass (Figs. 2 and 3).

Predator track density was not included in any of the best DSR models, despite the fact that predator track density was significantly higher for unsuccessful nests (see Table 1). Furthermore, DSR of broods did not seem to depend on hen age, as age was not included in any of the best models either. Cross tabulation of brood fate against age showed no significant effect (c² = 0.33, P = 0.53, N = 81).

Discussion

The statistical approach we used allowed us to evaluate the relative importance of different covariates on early brood survival, both among and between different levels (i.e. individual brood and environment). When having a large number of closely related models, such as in our case, designation of a single best model may not be satisfactory, because that model is likely to be highly variable. In other words, the ‘best‘ model is likely to vary from data set to data set, even if replicate data sets would be collected under the same underlying process. In such a case, model and parameter averaging provides more stabilised inference (Burnham & Anderson 2002).

We defined early brood survival as whether or not at least one chick survived the first 10 days after hatching. Using brood size, as a continuous variable, rather than a binary outcome, may eventually provide more information. However, we checked broods without using pointer dogs, and often under difficult conditions, and a reliable estimate of brood size at the age of 10-14 days was difficult at the least. However, using two brood counts (at two and eight weeks of age), as well as features of individual hen behaviour, provided a solid basis for inferring whether or not at least one chick was still alive at this point of time. The proportion of hens without a brood during late summer is also commonly used in population studies based on line transects or wildlife triangle monitoring data (e.g. Kurki et al. 2000) to assess breeding success. This proportion will be made up by those that lose the nest, lose the brood during the early but critical stage, or lose the brood after the critical early stage. The use of the first 10 days after hatching as the critical period, on the other hand, is justified by a recent study (Ludwig et al. 2006). In our study, we demonstrated that the effects of weather conditions experienced after hatching are very critical for survival, but then rapidly fade after a period of 10 days. Similarly, intense mortality during the first 10-14 days has been documented in a number of grouse species (e.g. Lindén 1981a, Erikstad & Andersen 1983, Kastdalen & Wegge 1991, Braun et al. 1993, Caizergues & Ellison 2000, Larson et al. 2001). Such mortality has been experimentally shown to be due in part to chick viability, which ultimately is influenced by maternal condition and egg quality (Moss & Watson 1984). For instance, according to the ‘early vegetation’ hypothesis by Siivonen (1958), resource availability before and during egg-laying may affect hen condition and hence egg quality. In our study, chick body mass was strongly correlated with fresh egg mass r = 0.568, P < 0.000), and negatively correlated with egg quality, here defined as fresh egg mass - chick body mass (P = −0.365, P = 0.001). Hen body mass, on the other hand, was related to neither chick body mass nor egg quality (r = 0.178, P = 0.112 and r = 0.066, P = 0.558, respectively). Hence, in our study, the effect of chick body mass is not necessarily mediated by hen condition.

In our study, variation in early brood survival was mediated by the extent of bilberry cover, as well as through the interactions of temperature, chick body mass and hatching date. Being present in 11 out of 14 selected models, cover of bilberry was by far the single most important main effect. The lack of any interaction with other variables suggests the effect to be independent of other covariates. The proportion of bilberry has previously been shown to be of high importance to grouse chicks (e.g. Baines et al. 1994, Storch 1994, Wegge et al. 2005) and in capercaillie adults, summer home-range size was negatively correlated with proportion of bilberry cover (Storch 1993). The importance of bilberry is likely to be food mediated. Due to the high energy requirement in early life (Lindén 1981b), chicks rely on a protein rich invertebrate diet (Baines et al. 1994, Wegge et al. 2005). While chicks may prey upon a wide array of invertebrate taxa (e.g.Kastdalen & Wegge 1984, Picozzi & Hepburn 1984, Baines et al. 1994), lepidopteran larvae may be of particular importance (Spidsö & Stuen 1988), as they are particularly rich in proteins and easy to prey upon. Bilberry is commonly associated with occurrence of lepidopteran larvae (Wegge et al. 2005, Lakka & Kouki 2009). Furthermore, bilberry is the dominant type of ground vegetation in the so‐called Oxalis-Myrtillus forest-site type (Hyvämäki 2002), which is traditionally considered the most important brood habitat type in Finland (Rajala 1966, Lindén & Rajala 1981). During the second half of the 20th century, however, the total coverage of bilberry should have decreased by about 50% (Reinikainen et al. 2000). Our results therefore suggest that in addition to increased predation (effect on nest success; Ludwig 2007) and weather and climate change (effects on brood survival; Ludwig et al. 2006), declines in bilberry cover may have contributed to the long-term decline in reproductive success observed in Finnish black grouse populations (Ludwig 2007).

In addition to bilberry cover, the daily brood survival was affected by interactions of chick body mass, temperature and hatching date; in particular by the interactions temperature*hatching date and temperature*chick body mass. The respective models with main effects only were not included in the set of selected models, which suggests that the effects occur only by means of interactions. Averaged parameter estimates of main effects, as estimated from interaction models, are commonly difficult to interpret, especially the direction of the effect. The visualisations of the interactions (see Figs. 2 and 3) shed more light on how the covariates interact. Both interactions suggest that daily survival rates increase with temperature. However, the benefits of increased temperatures are more pronounced when hatching early (see Fig. 2). This is likely due to the development of temperature patterns in early summer: while the mean temperatures of the 10 day post-hatching periods tend to increase from early to late hatchers, the variance of experienced temperatures decreases rapidly (Ludwig et al. 2006). Our results therefore strengthen the view that early hatching is more risky than late hatching. This is especially true under the scenario of the concurrent climate change. Due to the advancement of spring phenology, black grouse have advanced reproduction and increasingly hatch earlier. Since early summer has not advanced, i.e. has not become warmer, chicks increasingly hatch into unfavourable and more variable conditions (Ludwig et al. 2006). Also, hens that lay too early may produce a poor quality of eggs, because they lay before their diet allows them to optimise the egg quality (Moss et al. 1981). Later hatchers, on the other hand, are likely to face warmer temperatures with less pronounced daily fluctuations, hence the interaction between temperature and hatching date. In a study of black grouse in the Swiss Alps, Zbinden & Salvioni (2004) found that if the mean temperature of the first five days after hatching falls below a threshold of about 11.5°C, the probability of brood loss increases significantly. In the Alps, this does not occur before the beginning of July (when chicks hatch in the Alps), but in central Finland it happens already by about mid June, or even earlier (when chicks hatch in central Finland). A similar threshold level could be functional in our study population (see Figs. 2 and 3).

Cold temperatures may act as a double jeopardy for newly hatched chicks. First, after having used the nutrients of the egg yolk sack the energy requirement of grouse chicks are particularly high at the age of 4-8 days (Lindén 1981b), and the chicks rely entirely on protein rich insect food (Baines et al. 1994, Wegge et al. 2005) before gradually shifting towards a more plant dominated diet from the age of 2-3 weeks onwards. Cold temperatures in early summer may well affect the availability of insect food. Secondly, the thermoregulatory capacity fully develops only at the age of approximately three weeks (Hissa et al. 1983). Before that, and especially before becoming fledglings, excessive heat loss is likely to be lethal. That is, when temperatures are cold and acquiring food is particularly important, chicks need to trade-off between getting warmed up by the mother and foraging.

The interaction between temperature and the chick's body mass is more complicated (see Fig. 3). Heavy chicks seem to be less successful at very low temperatures, but their survival prospects improve at a faster rate when temperature increases. Already at rather low temperatures (11-12 °C), large chicks gain advantage, and do consistently better than light chicks at more elevated temperatures. The mechanism of this interaction is less clear. Large chicks should be more resistant to hypothermia, but on the other hand, they might be more prone to food shortage, due to increased metabolic demands. It seems clear, however, that under normal conditions heavy chicks do better than the lighter chicks.

The remaining variables included in at least one selected model (see Table 2) are likely to be of lower importance. Their averaged Akaike weights are relatively low, and the confidence intervals of their main effects all include zero. Furthermore, all of them were included in a model along with bilberry cover, suggesting that their inclusion may be due to bilberry cover. However, among these less important variables, the effect of drainage may be worth further discussion. For the same study population, we have recently demonstrated a negative effect of drainage upon nesting success, especially at high predator densities (Ludwig 2007). Furthermore, using nationwide data, drainage has been shown to affect annual reproductive output in forest grouse in southern and central, but not in northern Finland (Ludwig et al. 2008). In the present study, however, the effect of drainage on early brood survival was surprisingly positive. That is, the mechanisms by which drainage affects nesting and early brood success differ from each other. While drained areas may suffer from increased nest predation when predators are abundant, this is not the case for chicks after they hatched. Another likely and often speculated direct negative effect of drainage is drowning. Before chicks fledge at the age of about one week, crossing ditches may be lethal unless they are dry. Although we do not have local data on water levels in the ditches during the post-hatching period, we may say from observations that 2003 (with a high mortality) was rather dry and the majority of ditches were dry. On the other hand, the two last years (2004 and 2005 with a lower mortality) were wetter and many ditches contained water. Drowning therefore did not seem to be a problem in our case. The observed slight, but positive effect, is likely to be microhabitat and/or food related. Fertilisation of drained areas, a common practice in Finland, may have positive effects on the growth of bilberry, at least at a local scale (Helle & Ludwig 2003). We found a negative, but non-significant, relationship (P = 0.13) between the proportion of brood locations occupying undrained sites and bilberry cover, i.e. drained areas had a somewhat higher cover of bilberry. This may, to some extent, explain the weak positive effect of drainage. Considering the rather low weight of the variable, as well as recent results on the population level (Ludwig et al. 2008) and on individual nesting success (Ludwig 2007), nevertheless, suggest that on the long term, drainage does have overall detrimental effects on breeding success.

Among the variables not included in any of our alternative models, predator track density is the most peculiar one. Predator track density did not seem to affect chick survival, be it as a main effect or as an interaction with other variables, despite the fact that successful and unsuccessful broods differed in terms of mean predator track density (see Table 1). However, it needs to be noted that the two years with moderate to high predator track densities (2003-2004) were also clearly colder than the year with a low predator track density (2005). For the same population of black grouse, predator track density was the single most important variable in determining spatial and temporal variation in nesting success (Ludwig et al., submitted). Mammal predation has been shown to be a major source of early chick mortality in capercaillie (Kastdalen & Wegge 1991) and williow ptarmigan (Steen et al. 1988) in Norway, and in sharp-tailed grouse in Alberta, Canada (Manzer 2004). Predation by raptors, on the other hand, was significant for early chick survival in red grouse (Thirgood et al. 2000) and ruffed grouse (Larson et al. 2001). For black grouse, however, there is little information about predation patterns during early life. A number of raptor species are known to prey upon black grouse chicks in Finland, but the predation occurs at later chick stages (Reif et al. 2004, R. Tornberg, pers. comm.). This and our results suggest that black grouse chicks are relatively safe from predation during the first 10 days after hatching. Black grouse chicks hatch more or less synchronously (within 12-24 hours), and after hatching is completed, the chicks may remain in the nest for a couple of hours before leaving. The risk of complete brood loss due to predation is therefore likely to be highest during this period, but that did not seem to be an issue in our study.

Targeting of management practices to relieve mortality at key stages is commonly given a high priority in declining populations of grouse. That is, management at a local scale should focus on retaining, restoring or even creating suitable brood habitat. As mentioned earlier, abundance of bilberry has significantly decreased during the past decades as a result of intensified timber production (Reinikainen et al. 2000). In terms of early survival, management targets favouring an increase in the local coverage of bilberry should be considered. Predator control to improve juvenile survival is commonly advocated (Hudson 1986, Larson et al. 2001), but in the scenario we present, there would probably be no significant benefit for early juvenile survival.

Little can be done about the effect of temperature. Our results on brood survival of individual hens are in accordance with our recent results on the population level mentioned above (Ludwig et al. 2006). No management practice can possibly improve weather conditions, but the weather conditions experienced by the chicks could be used as a management tool. We have shown (Ludwig et al. 2006) that spring temperatures can be used as proxy for the timing of hatching. By doing so, one could estimate local hatching times and hence evaluate early-life conditions, such as temperature and precipitation. Whenever post-hatch conditions are below the threshold, a bad breeding output is to be expected. In order to avoid exacerbation of juvenile mortality in such years, hunting should be reduced, or hunting should not occur at all.