P. Byholm, University of Helsinki, Bird Ecology Unit, Department of Biological and Environmental Sciences, PO Box 65 (Viikinkaari 1), FI-00014 Helsinki, Finland. Tel.: +358 9191 57746. Fax: +358 9191 57694. E-mail: email@example.com
1Habitat heterogeneity has important repercussions for species abundance, demography and life-history patterns. While habitat effects have been more thoroughly studied in top-down situations (e.g. in association with predation), their role in bottom-up situations (e.g. in association with food abundance) has been less explored and the underlying mechanism(s) behind the ecological patterns have not commonly been identified.
2With material from 1993 to 2003, we test the hypothesis that the reproduction of Finnish northern goshawks Accipiter gentilis (L.) is bottom-up limited by habitat composition, especially in situations where the density of their main prey (grouse) is low. Special emphasis was placed on identifying the mechanism(s) behind potential habitat effects.
3While laying date and large-scale variation in the main prey density (but not habitat composition) were related to the number of eggs goshawks laid, small-scale differences in alternative prey density between different territories later influenced how many young were fledged via the mechanism of habitat-dependent partial-brood loss. As a result of this mechanism, a difference in nestling condition also arose between goshawk territories with differing habitat compositions.
4As the relative proportions of different landscape elements in a given landscape is a function of large-scale differences in geomorphology and land use, this means that the reproductive performance of goshawks as averaged over larger scales can be understood correctly only in respect to the fact that habitat gradients differ across landscapes.
5In addition to being one of the first papers identifying the mechanism of partial brood loss as being primarily responsible for the habitat-specific differences in the production of young, this study further illustrates the need to identify small-scale mechanisms to correctly understand the large-scale patterns of reproductive performance in territorial species. The repercussions of the observed habitat effect for local population development are discussed.
One problem that might complicate the interpretations of any potential habitat effect on demography, especially in bottom-up limited systems, is that the actual mechanisms behind the patterns are often poorly understood. First, possible bottom-up effects of habitat composition on reproductive performance have been investigated considerably less often than top-down effects (cf. Lahti 2001; Stephens et al. 2003; but see, e.g. Zanette, Doyle & Tremont 2000; Krüger 2002; Suorsa et al. 2003). For example, while considerable effort has been invested in studying habitat-dependent top-down effects of nest predation and brood parasitism on avian reproductive success (Lahti 2001; Stephens et al. 2003), much less effort has been put into exploring the bottom-up role of food availability (cf. Zanette et al. 2000). Furthermore, the mechanisms producing the observed patterns have usually remained unidentified. One reason for this might be that habitat effects seem to be of importance only in certain situations. For example, with work on raptors in northern Europe, it has been found that bottom-up habitat effects on reproductive success typically are present only in situations when prey are scarce, but not when prey are abundant (Hakkarainen et al. 2003; Lõhmus 2003; Lõhmus & Väli 2004). The reason for this variation is however, not yet well understood.
The northern goshawk Accipiter gentilis (from here onward goshawk) is a habitat-sensitive (Penteriani 2002) and long-lived forest-dwelling raptor with reversed size dimorphism that chooses breeding territories pre-emptively (Krüger & Lindström 2001). In northern Europe, spruce-dominated forests represent the optimal breeding habitat for goshawks (e.g. Widén 1997), and in Finland, grouse Tetraonidae are the most important prey (Tornberg 1997). However, largely due to increased timber production, the Finnish grouse populations have, on average, decreased by c. 50% during the recent four decades (Helle et al. 2003a; Fig. 1). During the same period, the proportion of grouse in the breeding-season diet of the Finnish goshawks has also declined by c. 50%, and grouse have largely been replaced with a mixture of alternative prey species such as corvids Corvidae, woodpigeon Columba palumbus L. and thrushes Turdus spp. (Sulkava, Linkola & Lokki 2006). As an apparent consequence of these dietary changes, the previous strong connection between grouse density and goshawk brood size (e.g. Tornberg et al. 2005) has been reduced in most parts of Finland (Byholm et al. 2002; Byholm 2004). However, the exact causes and mechanisms behind this breakdown are unknown. In particular there is no knowledge of how factors working at the scale of territories add to population-level patterns.
Here we test the prediction that goshawks breeding in forest-dominated, high-quality territories lay larger and earlier clutches and produce larger broods with nestlings in better condition than pairs breeding in territories having higher proportions of open habitat in the landscape around the nest. Special emphasis was placed on identifying the mechanism(s) behind the observed patterns. As different grouse species are quite evenly distributed across different habitats while the diversity and biomass of alternative prey are higher in forest habitats than on peat land (Siivonen 1972; Koskimies 1989; Väisänen, Koskimies & Lammi 1998), a particular aim was to test the hypothesis that the reproductive performance of goshawks differs between territories where the relative coverage of these habitats are distinct. Based on earlier knowledge (Lõhmus 2003), we expect that the effect of habitat on reproductive performance would be stronger in areas and/or years when grouse density is low compared with situations when grouse density is high. Acknowledging that the frequency of different habitat elements is a function of large-scale variation in geomorphology and land use, the importance of identifying small-scale patterns in habitat-effect studies is highlighted and the repercussions of habitat effects for population development is discussed.
Materials and methods
study site and goshawk data
Data on territory-specific goshawk reproductive performance (hatching date, clutch size, partial-brood loss, brood size, nestling condition) were collected in a 6300 km2 large study area in Suupohja (see Fig. 2), western Finland (62°00′−62°55′N, 21°05′−22°40′E) during 1993–2003. In 1993–98, all known goshawk territories were customarily visited two to four times during the nesting phase. The initial visit(s) occurred in May to judge whether the territory was occupied or not, and later visits occurred at the end of June to determine brood size and to ring chicks. As a result of increased research activity from 1999 onwards, most occupied nests were now also examined at least once in May to determine clutch size, as well as in June to ring the chicks and to determine the occurrence of partial-brood loss (‘some, but not all, members of a sibship dying from any and all causes’; Mock 1994) and brood size. In June, nestlings were wing measured to the closest millimetre and weighed with PESOLA spring balances (accuracy: 10–20 g). During the 11-year period, a total of 776 successful nesting attempts (at least one nestling to be ringed in late June) were recorded in 218 different territories that were occupied for 1–11 years. In analyses of clutch size, data from both successful and unsuccessful nesting attempts were included in the analyses, whereas in analyses of brood size only successful nesting attempts were considered. This procedure was followed as nesting failure is unrelated to main-prey density and landscape composition (nesting attempts fail equally often in all three territory types; Byholm & Nikula 2007) and the focus of this study is on how landscape composition and food supply influence realized goshawk reproductive success. Hatch date of chicks was estimated separately for each sex by regressing wing length on age for those chicks with a known hatch date (cf. Kenward, Marcström & Karlbom 1993; Byholm 2004). After controlling for size (wing length), residual weight (the deviance from the expected weight for a given age) was used as a measurement of nestling condition.
regional grouse density
Estimates of grouse density (individuals km−2) for capercaillie Tetrao urogallus L., black grouse Tetrao tetrix (L.), hazel grouse Bonasa bonasia (L.) and willow grouse Lagopus lagopus (L.) during 1993–2003 from the Finnish Game and Fisheries Research Institute (FGFRI) were based on used data collected under the Finnish wildlife triangle scheme (see Lindén et al. 1996). Basically, a wildlife triangle is a transect line in the form of a triangle, each side being 4 km long (i.e. 12 km in total). The wildlife triangles are censused by volunteers in August each year. Annually 23·5 ± 3·9 (mean ± SD) wildlife triangles were censused within the study area. Owing to the limited number of wildlife triangles annually censused within the study area, the triangles were more scattered geographically than the goshawk territories. As a result, we did not consider it to be meaningful to calculate separate grouse density estimates for specific goshawk territories from this material. However, as the synchrony in population dynamics of Finnish grouse decreases with increasing distance (e.g. Ranta et al. 2003), different grouse density estimates were calculated for four 50 × 50 km grid cells roughly laying on top of the study area (see Fig. 2) to account for potential geographical differences in grouse density and dynamics. As all grouse species are, to some degree, preyed upon by goshawks (Tornberg 1997; Sulkava et al. 2006), we analysed all species together. The subdivision of the study area was justified as grouse density differed among the four grid cells (General Linear Model, F3,214 = 7·30, P < 0·001), even if grouse populations fluctuated synchronously over time in all four grid cells (grid cell × year, F30,214 = 0·99, P = 0·48). Grouse density did not differ significantly between years (F10,214 = 1·84, P = 0·06), and especially if 2003, the worst grouse year during the study period (see Fig. 1), was omitted from the analyses, the effect of year was clearly far from significant (F9,197 = 0·51, P = 0·87). As grouse density differed significantly among the four grid cells, grid cell identity was used as a categorical factor when testing for the effects of variation in large-scale grouse density. Here it is important to note that grouse density varies much more when compared among the four grid cells (minimum average: 17·8 ± 1·7, maximum average: 35·6 ± 2·1; mean ± SE) than the annual averages calculated for the whole Suupohja study area during the study period (1993–2003) may at first indicate (minimum average: 13·7 ± 3·3, maximum average: 23·2 ± 3·9; cf. Figure 1). In other words, while grouse density varies clearly at the 50 × 50 km scale, this variation is averaged away when calculating grouse density estimates for larger areas.
territory-specific prey abundance
A modification of the point-count method (Koskimies & Väisänen 1991) was used to estimate territory-specific natural abundances of avian prey species in 33 goshawk territories in 2003. The counts were carried out by JK between 14 May and 13 June in mornings with calm and sunny weather between 03.45 and 11.00 h at 6–8 points (median: 7) within a belt of 500–1500 m from an occupied goshawk nest. All birds observed (species and numbers) were registered. Even though goshawks do not rely completely on forested areas for hunting in central Europe (Kenward 1982; Kenward & Widén 1989), birds in Fennoscandia almost exclusively hunt within forests (Widén 1989; Tornberg & Colpaert 2001). Thus, to obtain as biologically relevant measures as possible, all census points were located in forest environments. To minimize potential bias arising from possible differences in detection probabilities of species between locations, the points were evenly spaced in tree-dominated areas (including the tree-dominated border zone between bog and forest habitats). However, these points were always at least 200 m from the edges of permanently open habitat types (fields, tree-less mires, larger water bodies). As we wished to prevent unsuccessful nesting attempts from influencing our results on the relationship between realized reproductive performance and prey abundance/landscape composition, only bird count results from successful territories (n = 27) were used in analyses of these data. To make the analyses of the bird count material more straightforward, all observations were categorized into five groups reflecting the importance/suitability of the species as prey for Finnish goshawks (Tornberg 1997; Sulkava et al. 2006; Byholm et al. unpublished): (1) grouse; (2) ducks, black woodpecker Dryocopus martius (L.), doves, thrushes, corvids; (3) waders, gulls; (4) small passerines; and (5) other birds (see Appendix for exact species lists). Of these, grouse are the single most important prey, but species belonging to category 2 are also commonly preyed upon during the breeding season (Tornberg 1997; Sulkava et al. 2006). Species belonging to categories 3–5 are preyed upon more occasionally.
The proportions of different landscape components as determined from digitalized 1 : 250 000 maps (source: National Survey of Land) were calculated at different distances (500 m, 1000 m, 2000 m) from the centre of all goshawk territories over an area approximately representing the core-hunting range of breeding north European goshawks (Halley, Nygård & Wiseth 2000). When only one nest was present on a territory, the coordinate of this nest was considered to represent the centre of the territory. If the goshawks had two or more alternative nests (range: 2–7) within the range of a defined territory (cf. Hakkarainen et al. 2004), a mean coordinate of all nest-specific coordinates was used as a measurement of the territory centre. The landscape components calculated from digital maps are as follows:
1Forest includes all types of forests of all ages (including clear cuttings) predominated by either deciduous tree species (< 5% of all forests) or by a mixtures of the locally dominant conifers, the Norway spruce Picea abies (L.) and the Scots pine Pinus sylvestris (L.) (> 95% of all forests).
2The majority of the habitat classified as peat land are either tree-less mires or barren bogs with sparse stands of Scots pine. Spruce-dominated peat lands are also placed into this class (comprising < 5% of all habitats classified as peat land).
3The fields category primarily consist of cultivated fields and pastures, but also fallow fields, seminatural and natural grassland (comprising < 2% of all habitats classified as field) are included in this category.
4Water includes all types of open water bodies (> 0·3 ha in size and/or > 5 m in width) found in the landscape, such as lakes and streams.
5Built-up areas comprise areas with man-made buildings (e.g. villages, towns, industrial areas).
The proportion of landscape components was calculated with the software FRAGSTATS (McGarigal & Marks 1995). As we were interested in comparing whether territories with differing proportion of the focal goshawk habitat (forest) differed in respect to their reproductive performance, territories were divided into two main types on the basis of the proportion of land area that was covered with forest and peat land. These habitats comprise the two most common habitat categories in goshawk territories in the study area (Table 1). Although the proportions of the same landscape element correlated well between the different scales for every single habitat type (Spearman rank correlation, all P < 0·0001, all rs = 0·37), the radius of 1000 m was chosen as a rationale in the territory categorizing process instead of 500 m or 2000 m. This scale was preferred, as most hunting of breeding Finnish goshawks, according to radio telemetry data, occurs within a radius 500–1500 m from the occupied nest (R. Tornberg, personal communication) and because the minimum nearest neighbour distance between two simultaneously occupied goshawk nests in Suupohja was 1890 m. A territory was categorized as a ‘forest territory’ if the proportion of forest was ≥ 68% and the proportion of peat land was < 26%, while a territory was assigned to the category ‘bog territory’ if the proportion of forest was < 68% and the proportion of peat land was ≥ 26%. The percentages used as dividers (26% and 68%) were derived from the overall mean coverage of these habitat types in all territories at the 1000 m scale (Table 1). Of the 218 territories producing at least one fledgling in 1993–2003, 88% (n = 192) belonged to either one of these two groups. The remaining 12% of the territories were assigned to the group ‘field territory’, as the proportion of fields was higher among these territories in relation to other landscape components, compared with the two main territory types (Table 1). The proportions of water and built-up areas were low at all distances from the territory centre as well as in every territory category (Table 1) and were not used to classify territories. As a result of large-scale variation in landscape composition within Suupohja, the proportions of different territory types differed among the four 50 × 50 km grid cells approximately covering the study area (χ2 = 28·2, d.f. = 6, P = 0·0001). As peat land is less common close to the sea coast than it is further inland [due to historical land use (drainage) and differences in geomorphology], bog territories were sparser than forest territories in the grids close to the sea coast, whereas the relative proportions of forest and bog territories were the opposite in the inland grids (Fig. 2). The proportion of field territories did not differ among the grids (Fig. 2).
Table 1. The percentage (SD) of landscape components analysed at three different spatial scales (500 m, 1000 m, 2000 m) from the centre of 218 goshawk territories producing at least one nestling in 1993–2003. Territory-type specific (see methods) averages and grand average values are given
Distance from territory centre
Forest territory n = 109
Bog territory n = 83
Field territory n = 26
Grand average n = 218
As much of the goshawk breeding material was collected from the same territories for consecutive years, linear general mixed effects models (GLMMs) with each territory (territory-ID) defined as a random variable were used to analyse these data statistically. This can be considered the best option in situations when data are of a pseudoreplicative nature, as it estimates the differences in intercepts of the fixed factors as random effects arising from territory-specific patterns (e.g. Crawley 2002). The significance of fixed effects was assessed with F-statistics that take into account the repeated measures. In analyses of clutch size, brood size, hatching date and nestling condition, the data were modelled using an identity link with normal errors, while analyses exploring whether or not a clutch experienced partial-brood loss was modelled as a binomial response (0 = clutch intact, 1 = clutch reduced) and fitted with a logistic link function. Unless stated otherwise, model selection was performed following a manual step procedure by excluding nonsignificant explanatory variables (regional grouse density, territory-specific prey abundance, territory type, date, year) from the GLMMs, starting with second-order interactions, until the model contained only significant factors (Pinheiro & Bates 2000). At this stage, the significance of the random factor was assessed through variance comparisons. All GLMMs were implemented using S-Plus 6·1 software (Insightful corporation, Seattle, WA, USA), whereas all other analyses were conducted with SPSS ver. 12·0 (SPSS Inc., Chicago, IL, USA). For analyses of hatching date and clutch size, we did not specify territory-specific prey abundance as an explanatory variable because many of the species (see Appendix) are migratory and only arrive after goshawks have laid their eggs.
hatching date, clutch size and partial-brood loss
Hatching date was not related to grid-specific differences in grouse density, and differed neither between years nor between the three different territory types (GLMM, all P > 0·11). Moreover, only 14% of an appreciable between-territory variation in hatching date could be explained by territory-ID (σ2 = 4·45, 95% CI 2·40–8·35). Although material from 1999 to 2003 showed that clutch size did not differ between territory types (GLMM, F2,140 = 1·68, P = 0·19, territory type forced to stay in the model), clutch size was positively related to regional grouse density (F3,140 = 5·78, P = 0·0009), differed between years (F4,159 = 7·21, P < 0·0001), and decreased with advancing date (F1,159 = 27·49, P < 0·0001). There was no evidence of systematic differences in clutch size between different territories as territory-ID accounted for only 7% of the remaining variation (σ2 = 0·03, 95% CI 0·005–0·14). Even though clutch size did not differ among territory types, a subset of data revealed that partial-brood loss occurred more often in bog and field territories than in forest territories (F2,827 = 8·09, P = 0·0003; Fig. 3).
brood size and nestling condition
No annual variation in brood size was detected from 1993 to 2003, and brood size was not related to either hatching date or grid-specific grouse density. However, brood size differed significantly between territory types (GLMM, F2,215 = 4·61, P = 0·01). Brood size was significantly smaller in bog territories than in forest territories and field territories (Fig. 4). The inclusion of territory-ID did not significantly improve the fit of this model (σ2 = 0·06, 95% CI 0·02–0·14) and territory-ID explained only 7% of the total variance. In addition, the frequency distribution of brood sizes differed between territory types (χ2 = 20·1, d.f. = 8, P = 0·01), because broods with four nestlings were clearly more common in forest territories than in the other territory categories. When modelling territory-specific prey abundances, it was found that only the estimates describing the abundances of species belonging to prey category 2 added significantly to goshawk brood size (F1,25 = 11·48, P = 0·002). The abundance of species belonging to category 2 was significantly higher in forest (n = 17) territories than in bog (n = 9) territories (Mann–Whitney U-test, U = 31·5, P = 0·02), while the abundance of species belonging to other prey categories did not differ between the two territory types (all P > 0·14). Field territories were not analysed due to insufficient data (n = 1).
Data on nestling condition with measurements of 1239 nestlings from 1999 to 2003 revealed that nestlings were in better condition in bog territories and field territories than in forest territories (GLMM, F2,171 = 2·98, P = 0·05, Fig. 5a). As the proportion of different territory types varied regionally (cf. Fig. 2), nestling condition therefore also differed at the 50 × 50 km grid-scale (F3,171 = 3·29, P = 0·02). Average nestling condition was higher where the relative share of bog territories was high than where it was low (Fig. 5b). Furthermore, nestling condition differed between years (F4,1058 = 7·48, P < 0·0001). Even after territory type was accounted for, nestling condition to some degree still differed consistently between individual territories (σ2 = 1114·7, 95% CI 778·2–1596·6) and territory-ID accounted for 20% of the random variation. A subset of the same data set also revealed that nestlings were in better condition in nests where partial-brood loss had occurred than in nests where the number of nestlings at the time of ringing corresponded to the number of laid eggs (F1,651 = 4·34, P = 0·04; Fig. 6).
The predictions concerning the hypothesized landscape effects on goshawk breeding success were only partially confirmed. First, in common with some other species (Hakkarainen et al. 2003; Luck 2003; Lambrechts et al. 2004), including German goshawks (Krüger 2002), goshawks breeding in Suupohja produced larger broods in territories where the proportion of prime habitat was high. However, hatching date and clutch size were not related to territorial habitat composition. Second, contrary to expectations as well as to results reported in earlier studies (e.g. Lambrechts et al. 2004) goshawk nestlings were in worse condition in forest territories than in territories where the proportion of secondary hunting habitats (peat land or fields) was large. Third, as grouse density and territory type did not interact in any of the models designed to explain the observed variation in hatching date, clutch size and brood size, no support was found for the hypothesis that territorial habitat composition would be of importance in situations where main prey (grouse) are few but not in situations where grouse are more abundant (cf. Lõhmus 2003). This result differs from the general picture where habitat quality seems to have a positive influence on breeding success in bad/deteriorating food situations but not in good food situations (Hakkarainen et al. 2003; Lõhmus 2003; Laaksonen, Hakkarainen & Korpimäki 2004; Lõhmus & Väli 2004). Instead, goshawk clutch size was determined directly from laying date and large-scale variation in grouse density (at the 50 × 50 km scale; Fig. 7). Following the common pattern in birds, clutch size also decreased with advancing date and varied between years (cf. Byholm 2005).
As partial-brood loss hits harder in bog and field territories than in forest territories, this indicates that partial-brood loss is the mechanism that causes average brood size to differ between the different territory types (Fig. 7). Because only c.4% of all partial-brood losses are due to hatching failure (Byholm 2005), nestling number is in practice adjusted by post-hatching mortality. The identification of partial-brood loss as being responsible for the observed habitat effect is important, because in order to understand in detail how spatial pattern contributes to ecological processes, there is a need to identify the mechanisms producing the effect (McGarigal & Cushman 2002; Turner 2005). While food availability directly seems to be the key element in the few studies that have thus far identified such a mechanism (e.g. Suorsa et al. 2003; Lambrechts et al. 2004) the present study is to our knowledge the first to identify habitat-dependent partial-brood loss as being the mechanism producing the observed result irrespective of large-scale variation of main prey density (see Hakkarainen et al. 2003 for a result of an effect in poor food situations). Furthermore, the observation that partial-brood loss was habitat-dependent also highlights the importance of considering small-scale mechanism in order to correctly understand large-scale pattern: brood size as averaged over larger scales depends on the relative proportions of different territory types in the landscape. The proportion of different territory types in the landscape is in turn largely a function of historical land-use (e.g. drainage of mires and bogs to agricultural land) and even more, a function of large-scale differences in geomorphology. This underline the importance of considering that the strength of not only top-down (cf. Patten & Bolger 2003) but also bottom-up forces varies in response to habitat gradients across landscapes.
Considering that broods were larger in forest territories than in bog territories (even if clutches were initially of similar size), the fact that nestlings were in worse condition in forest territories than in bog territories is an interesting finding that calls for further discussion. If partial-brood loss is the main mechanism that adjusts brood size, nestling condition should be positively related to the occurrence of a brood-loss event. This was shown to be the case, and indicates that partial-brood loss ‘forces’ goshawks to trade-off the quality of their offspring with offspring quantity. As virtually all partial-brood loss occurs within the first 2–5 days after hatching (Byholm 2005), it is evident that the trade-off situation arises very soon after hatching. However, whether or not nestling condition has any long-time fitness consequences for the individuals of concern or for the population as a whole (cf. Lindström 1999; but see Lambrechts et al. 2004) is unknown at the moment. As with brood size, nestling condition also was formed by partial-brood loss (Fig. 7). Consequently, average nestling condition as measured over large spatial scales (the 50 × 50 km grid-scale) is best understood as a function of how abundant bog territories are in relation to forest territories regionally (as field territories are few in our study area, their role remains rudimentary). Because the most important alternative prey species (e.g. thrushes, doves, corvids), but not main prey species (grouse), are more abundant in forest territories than in bog territories, the key element behind this habitat effect seems to be alternative prey abundance (Fig. 7). Testing the validity of this hypothesis would be possible only through a controlled feeding experiment where extra food is provided to a subsample of territories belonging to both territory types while another subsample would function as a control. Even after this, it would be necessary to catch adult birds to distinguish whether habitat composition directly affects the fitness components or whether individual differences in phenotypic quality of breeders are more important.
Considering that habitat heterogeneity elsewhere has been found to relate to goshawk population size (Krüger & Lindström 2001), the main question here is whether the differences we have described in reproductive output between territories have consequences for population trend. Even if the difference in the average number of young produced in bog and forest territories (2·7 vs. 2·9) might seem to be too small to be of biological relevance, this conclusion might be more apparent than real. Subsequently, all other things being equal, assume that all goshawk pairs producing nestlings in bog territories during the last 5 years of study (1999–2003) would have bred successfully in forest territories instead. In this scenario, the population would have produced 100 (7·0%) more offspring than what was the case. This must be considered a significant difference for a species applying a K-strategy, even though the population would also be limited outside the breeding season by factors that are not habitat-dependent (e.g. by factors affecting adult mortality). Consequently, as Finnish goshawks recruit locally to a high degree (Byholm et al. 2003), the differences in reproductive performance between territory types suggest that the size of a local goshawk subpopulation (at the 50 × 50 km grid-cell scale) is linked to large-scale landscape composition. Before any rigid conclusions on this matter can be reached, it is, however, clear that more detailed studies must be performed. In particular the apparent trade-off between offspring number and condition requires more attention in this respect, as it is possible that offspring produced in bog territories gain a long-term fitness advantage over offspring produced in forest territories due to their on average better body condition (cf. Lindström 1999).
In conclusion, the present study sheds light on two important questions that have not been deeply explored in ecological work so far (Thompson et al. 2001; Turner 2005). First, in order to correctly understand the effect of the mechanism(s) behind a habitat effect, the conclusions drawn might easily fail unless it is taken into account that small-scale factors (habitat-dependent partial-brood loss in this case) and factors working across larger scales (grouse density and laying date) may interact to influence the results. Second, as goshawk reproductive performance (brood size, nestling condition), as averaged over larger areas, depends on the relative proportions of different territory types in the landscape, this means that the general strength and effect of bottom-up forces can be interpreted correctly only with respect to the fact that habitat gradients differ across landscapes. In the light of these results, we feel that there is a need to more thoroughly study the long-term population consequences of mechanisms behind habitat effects, theoretically and empirically, in the future.
We are grateful to Jon Brommer, Mike Fowler, Harri Hakkarainen, Robert Kenward, Jari Valkama and an anonymous referee for comments on previous drafts of the manuscript. Marcus Wikman is acknowledged for his effort in extracting the grouse material from the databases at the FGFRI. We express our gratitude to Kari Ketola, Ismo Nousiainen, Kari Palo, Jari Hernesniemi and Ville Yli-Teevahainen for help in the field. This work was financially supported by the Kone Foundation and the Academy of Finland (grant no: 208861 to PB).