A long-term study of size variation in Northern Goshawk Accipiter gentilis across Scandinavia, with a focus on Norway

Modern comparative material To analyse changes in both modern and past populations of A. gentilis, metric data of the nominate Accipiter gentilis gentilis were collected from across the Nordic countries (Norway, Sweden, Denmark and Finland). The Northern Goshawk is a sedentary species, generally choosing to breed and winter in the same area. There are some exceptions to this in North America, Fennoscandia and Russia (Squires et al., 2020). However, individuals from Fennoscandia rarely migrate further than 300 Km (Squires et al., 2020). The nominate A. g. gentilis is distributed across Europe and east to the Urals, Caucasus, and Asia Minor, and southwards to NW Africa (Squires et al., 2020). The slightly larger subspecies Accipiter gentilis buteoides breeds in northern Fennoscandia and Siberia, wintering in south and central Eurasia (Ferguson-Lees & Christie, 2005). Goshawks found in northern Finland and northern Sweden likely represent the subspecies A. g. buteoides (Gladkov, 1941; Vaurie, 1965). Therefore, we excluded any material from these areas in our study. In addition, young Northern Goshawk from interior parts of Scandinavia are known to winter along the Norwegian coast in the north (Fransson & Petterson, 2001; Bakken et al., 2003). To avoid mixing of the larger subspecies A. g. buteoides within the modern comparative sample, we decided to use material from the south of Norway, Sweden and Finland. Modern comparative skeletons (collected over the past c. 150 years from 1861 to 2015) have been measured from the University Museum of Bergen, the Natural History Museum of Denmark and the Finnish Museum of Natural History (see Appendix 1). The specimens were originally collected from Norway (n = 65), Sweden (n = 30), Denmark (n = 93) and Finland (n = 52) (see Fig. 2). We measured all skeletal elements, except for the vertebrae and ribs of 240 modern partial and complete Accipiter gentilis gentilis skeletons of which 103 were female and 137 were male. It is not clear how sex of the specimens was initially determined, however, we presume that for the majority it was through internal inspection. Accipiter gentilis are highly sexually dimorphic, with the female being larger than the male (i.e., reversed sexual size dimorphism). There is very little to no overlap between the sexes of Northern Goshawk (Kenward, 2006). As a result, it was possible to identify 5 modern specimens that were likely to have been mis-sexed at the time of collection. Wrongly identified sex can be common in Accipiter subspecies and especially in Accipiter gentilis, as the paired ovaries can sometimes be mistaken for testes (Storer, 1966). The specimens which were wrongly sexed have been reclassified as the correct sex based on our osteological analysis and included within this study. The museum numbers of mis-sexed specimens are as follows; B 4462 and KL 31303 (originally recorded as males but fall within the female size range and have been reclassified as females), B 9016, NHMD 306590 and NHMD 306670 (originally recorded as females but fall within the male size range and have therefore been reclassified as males). All specimens were measured by SJW using digital callipers. Measurements followed the conventions set out in Von den Driesch (1976). Three additional measurements were recorded: the smallest depth of the distal shaft of the humerus (KB) found in Kraft (1972), depth of the ulna proximal end (Tp) and height of the symphysis of the carpometacarpus (HS) taken from Otto (1981). Data suggest that Finnish populations of the nominate A. g. gentilis are slightly larger in their wing length and body mass than other Scandinavian populations (Tornberg et al., 2006), although the reasons for this remain unclear. It is possible that they represent a slightly more northern clinal population with bigger proportions. Alternatively, it may reflect a slightly more continental climate than Norway and Sweden (Tornberg pers. comm.). In addition, we noticed that the Danish specimens included here were on average smaller than the other Scandinavian specimens, again possibly due to clinal variation. ANOVA tests were used to detect statistical differences between the modern populations for each country. The results show that Norway and Sweden did not differ; Denmark was often statistically different to Norway, Sweden and Finland; Norway and Sweden rarely differed from Finland (full ANOVA results in Supporting Information File 1 (SIF1)). For these reasons, we have grouped Norway and Sweden together but kept Denmark and Finland separate when drawing comparisons. Archaeological material In general, A. gentilis is not frequent within the archaeological record for Norway (Walker et al., 2019). Morphologically, the osteology of A. gentilis is not easily confused with any other species, and we are confident of specimen identification. Despite this, all specimens were confirmed using the extensive modern comparative collections held at the University Museum of Bergen. In total 89 Medieval specimens were included within this study (Table 1). It is worth noting here that the 89 bones do not represent 89 individuals, although there is a possibility that some of these bones would have come from the same individual. The Medieval bones date to 1030–1537 Common Era (CE) and come from only 8 sites, all from the urban contexts of Oslo, Bergen and Trondheim (Table 1). Most Medieval bones come from female individuals and are likely to be linked to the practice of falconry (Walker et al., 2019). Most of the archaeological bones were limb elements, this may be due to taphonomic bias as they are more robust than for example cranial remains. However, this limited the skeletal elements that could be used for comparison. Within this paper, we focus on the humerus, ulna, carpometacarpus, femur, tibiotarsus and tarsometatarsus. Data analysis We first explored differences in size between groups using descriptive statistics in PAST 4.03 (Hammer et al., 2001). All data were tested for normality by looking at the variances and the Shapiro-Wilk test for normality (see SIF1). Principal Components Analysis (PCA) was used to establish which measurements were most responsible for the observed differences. Two separate PCAs were performed, one on the modern specimens only (including Norwegian, Swedish, Danish and Finnish modern specimens; Supporting Information File 2 (SIF2)) and a second with both modern and archaeological specimens (including Norwegian, Swedish, Danish and Finnish modern and Norwegian Medieval specimens; Supporting Information File 4 (SIF4)). To test for main and interaction effects of time, sex and country on greatest lengths (GL) of the humerus and femur of modern Northern Goshawks from Norway, Sweden and Denmark, we performed an Analysis of Covariance (ANCOVA) in R Statistical Software (v4.1.1; R Core Team, 2021), with the factor Time as the covariate and the factors Sex (2 levels), and Country (2 levels: Denmark and (Norway + Sweden) grouped together). For the humerus, n = 132, of which n = 57 for Norway & Sweden combined (43 males, 14 females), and n = 75 from Denmark (46 males, 29 females). For the femur, n = 172, with n = 83 for Norway & Sweden combined (55 males and 28 females) and n = 89 from Denmark (49 males, 40 females). Finnish modern specimens were excluded from the linear regression (Fig. 3) and the ANCOVA because, of the 52 Finnish specimens available, only 7 predate the year 2000, preventing a detailed look into the past century. To test for statistical differences between the mean greatest lengths of Norwegian Medieval specimens and modern Norwegian specimens of A. g. gentilis, we used a 10.000-iteration Fisher’s permutation test in R. The permutation test is somewhat similar to the bootstrap but differs from it in that a permutation test resamples without replacement. First, the sample means for each group and the difference between these means is computed. The data are then pooled and randomly permuted. The means and difference in mean for the permutated samples are computed. This process is then repeated n times for all possible permutations of the data, resulting in a frequency distribution of the mean difference. The 95% confidence interval and p-values can then be calculated. We considered p-values ≤ 0.05 statistically significant.