Ecology 1999, Demographic estimates from radio-tagging: models of age-speci c survival and breeding in the goshawk R. E. KENWARD{, V. MARCSTROÈ M{ and M. KARLBOM{ {Institute of Terrestrial Furzebrook Research Station, Wareham, Dorset BH20 5AS, UK; and {Institute of Zoophysiology, University of Uppsala, 751 22 Uppsala 1, Sweden Summary 1. Age-speci c survival and breeding (ASSAB) models were developed with data from 318 goshawks (Accipiter gentilis L.) tted during 1980±7 on the Baltic island of Gotland with tail-mounted radio-tags. 2. Comparisons with recaptures and recoveries of 238 ringed juveniles without radio-tags detected no marking bias among hawks radio-tagged at nests, nor survival bias from tag impacts or signal-loss e ects, but disproportionate adult deaths within a month of trapping. 3. Among 63 dead hawks found by radio-tagging, 35% were killed by humans and 65% died from natural causes, mainly starvation. Ring recoveries over-represented killing by humans. 4. Radio-tagging estimated rst-year male survival rates of 0 49±0 54 to the next spring, lower than estimates of 0 69±0 71 for females; combined survival was higher than in contemporary ringing records. Excluding rst-month data from trapped adults, second-year survival was 0 59 for males, 0 71 for females, and 0 79±0 83 for both sexes thereafter. 5. No hawks were observed to build nests or lay eggs in their rst year. Only 8% of radio-tagged females laid eggs in their second year, but 47% of older females. Second-year males were as often as older males in egg-laying pairs (70%), but were less likely to rear young. 6. Mark-recapture of 351 goshawks ringed in nests estimated that 262 edged on Gotland annually, from 151 nests with eggs. Emigration was 3% (SE 2 2%), with 1 ringed immigrant. 7. An ASSAB model with staggered-entry survival estimates predicted no population change when based on males, with breeding success of 75% from laying to edging, and a brood size of 2 27; independently recorded values for 165 nests were 75% (2 3%) and 2 24 (2 0 08). 8. The model required a 0 5% increase in female survival rate for equal numbers of hawks to pair, and an increase in edging success from 73% to 75% for a balanced population with a brood size matching the 2 27 of males. The model estimated a male : female ratio of 1 0 : 1 78 in the adult population, with 71% of males breeding annually but only 40% of females. Key-words: goshawk, population modelling, radio-tracking, sexual dimorphism, survival Ecology (1999) Correspondence: Dr R. E. Kenward, Institute of Terrestrial Furzebrook Research Station, Wareham BH20 5AS, UK. Tel: 01929 551518. Fax: 01929 551087. E-mail: reke@wpo.nevc.ac.uk Introduction Many bird and mammal species have larger males than females, with associated di erences in sizebased resource requirements, survival and sex ratios (e.g. Clutton-Brock, Guiness & Albon 1982). Raptors have unusual `reversed' size dimorphism (RSD), in which females tend to be larger than males. Although the evolutionary causes of RSD have been much discussed (Amadon 1959; Cade 1960; Storer 1966; Mosher & Matray 1974; Snyder & Wiley 1976; Newton 1979, 1986; Walter 1979;
1021 R.E. Kenward, V. MarcstroÈm & M. Karlbom Andersson & Norberg 1981), demographic implications have been little studied. It is hard to study the most size-dimorphic raptors, because their tendency to take agile, avian prey (Storer 1966; Newton 1979) is often associated with elusive behaviour, low abundance and destruction by humans to protect game or livestock. Thus, survival cannot readily be estimated by resighting visual markers, while data from ringing accumulate slowly and are vulnerable to recovery bias (Haukioja & Haukioja 1970; Lakhani & Newton 1983; Anderson, Burnham & White 1985). Nevertheless, Newton (1985) used 139 ring recoveries from the relatively common sparrowhawk (Accipiter nisus L.) in Southern Scotland to show that females survived better than males, and to infer that 40% of adult females might not be breeding. Only about 10% as many raptors need be radiotagged as ringed to estimate juvenile survival rates with similar accuracy, and without waiting for adult recoveries to accumulate (Kenward 1993). High proportions of nonbreeding adults have been inferred from survival rates of radio-tagged Haliaetus eagles (Bowman, Schemp & Bernatowicz 1995) and buzzards (Buteo buteo L.), with important implications for understanding evolutionary tness, predation and population ecology (Hunt 1998; R.E. Kenward & S.S. Walls, unpublished data). Breeding rates too can be found by radio-tracking, to test inferences from population models. However, estimates of survival by radio-tagging are vulnerable to bias from selective marking, tag impact and tag loss (White & Garrott 1990; Freeman, Morgan & Catchpole 1992). In this paper, we use data from radio-tagging to build age-speci c survival and breeding (ASSAB) models separately for male and female Swedish goshawks, which have mean body mass close to 850 g and 1250 g, respectively (MarcstroÈ m & Kenward 1981a). We complemented the radio-tracking with ring-recapture techniques to nd the immigration rate, emigration rate and total size of the island study population, and then waited for enough ring recoveries to test for bias in the radio-tag data. Finally, we tested the accuracy of the ASSAB models for predicting: (i) the productivity of the study population; (ii) equality in the numbers of each sex breeding; and (iii) breeding rates for males and females. The study was on Gotland, a 3100 km 2 Swedish island which lies in the Baltic approximately 90 km from the nearest mainland. During 1980±7, we sought all goshawk nests within an 846-km 2 study area across the centre of the island. The area was bounded by roads to the north and south, by the coast in the east and by a military training area in the west (Fig. 1). Most nest areas were found initially by systematic searching of mature woodland for plucking sites or hawk calls, although two of 48 sites were found by radio-tracking. Young hawks were marked at all these nests, and also at 47 nest areas outside the study belt, which were found mainly as a result of information from farmers and hunters or by radio-tracking. The intensive study area, which contained habitats representative of the whole island, was 54% wooded, mainly with Scots pine (Pinus sylvestris L.) and Norway spruce (Picea abies L.), but with some deciduous trees on the edges and in copses. The remaining 46% was mainly pasture and arable farmland, with 2% wetland and 2% houses, roads and gardens. Rabbits (Oryctolagus cuniculus L.) were eaten year-round by the hawks, with young corvids (Corvidae) and thrush (Turdus) species also important during breeding. Squirrels (Sciurus vulgaris L.), gamebirds and waterfowl were taken frequently during winter (MarcstroÈ m, Kenward & Karlbom 1990). Rabbits were especially abundant in a 136-km 2 eastern part of the study area, which had thin soil with frequent limestone outcrops; the density of goshawk Methods STUDY AREA Fig. 1. The Baltic island of Gotland, showing the boundaries of the intensive study area (Ð), of the boundary strip, and the locations of goshawk nests (.) and sites used for winter trapping (&).
1022 ASSAB models from radio-tagged goshawks nests there (1 per 10 5 km 2 ) was double that in the rabbit-poor central area (1 per 20 3 km 2 ; Kenward, MarcstroÈ m & Karlbom 1993a). TAGGING AND TRACKING HAWKS Nestlings were ringed (banded) when they were close to leaving the nest. At that time, 124 young goshawks were equipped with 10 g or 18 g legmounted transmitters (Biotrack, Wareham, BH20 5AX, UK), which represented 1 0±2 5% of their bodyweight at rst ight (Kenward et al. 1993a). When feather growth was complete, at 20±35 days after leaving the nest, 88 leg-tagged edglings and 97 other young hawks were caught in box-traps near the nest (Kenward, Karlbom & MarcstroÈ m 1983) and were equipped with more powerful 14 g tail-mounted radio-tags, sewn at the base of the two central rectrices (Kenward 1978). Traps near nests had dead rabbit or other meat accessible to the young hawks, with live pigeons in a separate compartment (Karlbom 1982). During autumn and winter, box-traps were set, when weather permitted, at 31 sites throughout the island (Fig. 1). The sites were selected for convenience of access and volunteer e ort, and were not near nests but where hawks were likely to be hunting. Fifteen sites were either in the area intensively searched for nests, or within a boundary strip of 2 km (approximating the radius of a circle with the mean area, 10 6 2 1 6 km 2, of MCP ranges recorded for 28 hawks in winter on Gotland). Trapped hawks were aged as juveniles if in brown plumage, as second-year if any brown feathers remained unmoulted (BruÈ ll 1964), and older if they showed only grey feathers. A total of 133 hawks in second-year and older plumage from these traps had tags attached to the second and third rectrices from the tail centre; these feathers usually did not moult until June or July. Trapping was most intensive for the 6 months starting in September, for the 4 years 1980±3, to obtain population estimates by recapture of those ringed in the nest. To capture breeding adults at the end of the study, in order to nd what proportion had been marked, spring-loaded nooses were set on 17 nests, over dummy eggs while the real clutch was kept warm below the nest, during 3 weeks from 29 April 1986. Full-grown hawks were sexed as female if their maximum wing-length ( exed carpal joint to tip of longest primary with feathers attened) exceeded 340 mm; this was 99 5% accurate in 643 Swedish goshawks sexed post mortem by examining their gonads (MarcstroÈ m & Kenward 1981a). Eighteen radio-tagged nestlings that were neither recaptured when full grown, nor found dead, were sexed by their tarsus width shortly before edging. This was 98% accurate among young that were recovered and sexed when full grown: only two of 87 males had a tarsal width greater than 6 5 mm and only one of 82 females had a tarsal width less than 6 5 mm. The radio-tags were monitored with RX-81 receivers (Televilt, 71050 StoraÊ, Sweden) and three-element handheld Yagi antennas. Survival and movements were recorded every 2±5 days immediately after marking, and at 2-week intervals after dispersal from nest areas until the rectrices moulted with the tag, normally in April or May. Survival checks were mainly at night, to ensure optimal signal detection from live hawks roosting in trees. Such signals could be detected at 5±10 km with a six-element Yagi antenna on a 2-m mast atop a Volkswagen minibus, so search-stops were at 5 km intervals, taking advantage of hills (up to 40 m above sea level) to increase detection distance. Posture-sensing mercury switches gave slow signal pulses from perched hawks; fast pulses occurred when a tag was horizontal, and were typically weak for a hawk dead on the ground, strong with constant volume for a hawk brooding eggs or young, and irregular for a hawk in ight (Kenward 1987). Hawks that roosted more than once within 200 m of nests in spring were classed as nesting, if any of the alternative nests at the breeding site contained new green vegetation, and breeding, if eggs were laid, in which case nests were checked again when any young were old enough to mark. If we failed to detect a signal by night, we searched by day more thoroughly in the area normally used by the hawk, because tags at ground level on dead hawks could sometimes not be detected beyond 500 m. Carcasses were sent to the State Veterinary Station in Stockholm for post mortem examination. A ight in a Cessna 152, with an outward facing three-element Yagi antenna attached to each wing-strut (Gilmer et al. 1981), was made to search for lost signals in several years. Signals could be detected at 10±20 km from the air, so the ight followed east±west tracks at 15 km separation, after an initial circuit round the edge of the island to detect any carcasses at the base of coastal cli s. SURVIVAL ANALYSIS Juvenile survival was estimated for the 9 months from 1 July to 31 March, after which tags were liable to be moulted. Adult survival was estimated for the year from 1 April. Survival rates were estimated: (i) with provision for seasonal change, allowing for staggered entry of tagged animals and exit when tracking ceased (Pollock et al. 1989), with error estimates from Cox & Oakes (1984); and (ii) on the basis of deaths per tag-day (for which errors could not be estimated robustly). If a tag moulted, or provided signals to indicate impending failure or was at the end of its expected battery life, the wearer was classed as live on exit from the study; those with
1023 R.E. Kenward, V. MarcstroÈm & M. Karlbom unexplained absence of signal were classed as lost. Live and lost individuals were censored from the analysis at the start of the month in which tracking ceased. Categorizing lost birds as live overestimates survival if the lost category includes traumatic deaths associated with destruction of tags. To correct for this bias, ring recoveries from autumn trapping and after the study were compared for the live and lost categories (Aebischer 1993; Kenward 1993). Ringing data were recorded at the National Museum of Natural History in Stockholm. AGE-SPECIFIC SURVIVAL AND PRODUCTIVITY MODELS In principle, if n j,k,0 is the number of young of sex k produced in year j, then the number of individuals of sex k present in age-class x is: n j;k;x ˆ n j x;k;0 Y x iˆ1 s k;i eqn 1 where s k,i is the survival rate for sex k in each of the preceding i years. The total population, summed to the maximum age o at which birds can breed, is: N j;k ˆ Xo xˆ1 j;k;x eqn 2 The number of breeders B j and the numbers of young produced Y j in year j are then: B j;k ˆ Xo n xˆ1 j;k;x b k;x eqn 3 and Y j:k ˆ Xo n xˆ1 j;k;x b k;x y k;x eqn 4 where b k,x is the age-speci c breeding rate and y k,x is productivity of each breeder of sex k in age-class x. By substitution of equation 1 in equation 4: Y j;k ˆ Xo b Y x xˆ1 k;x y k;x n j x;k;0 s iˆ1 k;i eqn 5 and, because Y j,k = n j,k,0 1 ˆ Xo b xˆ1 k;x y k;x l x k Y x s iˆ1 k;i eqn 6 which, with l as the annual population increase, resembles the population-growth equation of Lotka (1925). Equation 5 provided the basis for an ASSAB model that used age-speci c survival and breeding data from the radio-tagging. Equations 3, 4 and 6 were used to test the model's ability to make predictions that could be veri ed by independent observations of the study population. Target criteria were: (i) for model predictions to be within one standard error of independent observations; and (ii) for `divergence terms', which were applied to parameters so that the model matched observations, to be less than the coe cients of variation of the parameters. For detailed comparisons, breeding rates b k,x were estimated from proportions nesting h k,x and proportions of nesting birds with eggs e k,x (i.e. b k,x = h k,x e k,x ). Productivity y k,x was estimated from the rearing success for each clutch g k,x and the number of young in successful broods f k.x (i.e. y k,x = g k,x f k.x ). Maximum age (o) was set at 18 years, because goshawks have lived up to 19 years in the wild (Kramer 1973) and a hawk that survived 20 years in captivity did not breed after its 18th year (C. Saar, personal communication). The model was developed as a BASIC routine which is available from R.E. Kenward. A fundamental test was whether the model, based on survival and breeding rates of the female ( f) and male (m) radio-tagged hawks, predicted the mean productivity ( yâ) that was observed independently in the study area, i.e. from equations 3 and 4, whether: P o xˆ1 n f;x b f;x y f;x P o xˆ1 n f;x b f;x ˆ P o xˆ1 n m;x b m;x y m;x P o xˆ1 n m;x b m;x ˆ y: eqn 7 This test was repeated with age-speci c divergence terms b k,i applied to the survival rates of either sex to equalize numbers of males and females that had eggs, based on goshawks being monogamous, e.g. adjusting female numbers in equation 1 for equation 3: X o b xˆ1 f;x n f;0 Y x s iˆ1 f;i b f;i ˆ Xo b xˆ1 m;x n m;0 Y x s iˆ1 m;i eqn 8 Further divergence terms u k could be applied to the productivity in the numerators of equation 7 to equalize the number of young reared by each sex. In a balanced (no-growth) population, where l x k =1, an alternative to equation 7 was then, from equation 6: u f X o xˆ1 b f;x y f;x Y x iˆ1 s f;i ˆ X o u m b xˆ1 m;x y m;x Y x s iˆ1 m;i b m;j ˆ 1: eqn 9 Another alternative to equation 7, applicable at l x k = 1 without precise age-speci c breeding data (Hunt 1998), was to predict the mean breeding rate from the mean productivity: b 1 k ˆ y X o Y x xˆ1 iˆ1 s k;i: eqn 10 All tests of statistical signi cance were two-tailed unless stated otherwise. Results TRAPPING AND RING RECOVERIES Ring recoveries of goshawks marked during 1980±3, the period of systematic trapping in winter, were
1024 ASSAB models from radio-tagged goshawks Table 1. Numbers of goshawks that were marked with rings in nests on Gotland, in rows according to whether they were trapped after edging to attach tail-mounted radio-tags, remained untrapped at nests where there was trapping or were from nests where no trapping was attempted. Columns 4 and 5 show numbers (with percentages) of the marked birds that were recaptured alive during their rst winter, and those recorded alive or dead at a later date Area Treatment Marked Trapped during their 1st winter Recorded after their 1st winter Intensive Radio-tagged 113 29 (24%) 17 (15%) Untrappable 24 3 (13%) 3 (13%) No trapping 46 7 (15%) 5 (11%) Peripheral No trapping 168 39 (23%) 10 (6%) used to check the impact of radio-tagging and to make population estimates. To test for local variation in recovery rates, records for hawks marked in the intensive area and its 2 km boundary strip were initially analysed separately from those in the peripheral area. Within the intensive area, no attempt was made to trap hawks to t tail-mounted radios at 10% of the least accessible nests, nor at a further 15% in the boundary strip. At the remaining 75% of nests, 18% of the 137 edged hawks could not be caught to attach tail-mounted radios (Table 1), in six cases because they were recovered dead before they could be trapped. Within the intensive area, the percentage retrapped in their rst winter was highest for hawks with tail-mounted radio-tags (24%). The retrap rate was 23% among all hawks in broods used for radiotagging, compared with 15% for hawks from untagged broods in the intensive area. There was a probability of 0 132 that the proportion of rst-winter recaptures from broods with radio-tagging was biased below that of broods without radio-tagging (Fisher exact test, one-tailed). The overall retrap rate in the intensive area and boundary strip was 21%, similar to the 23% in the peripheral area. The recovery rate after their rst winter was 15% for hawks with tail-mounted radios and 13% for untagged hawks from the same broods, compared with 11% for hawks from untagged broods in the intensive area (Table 1). The probability that hawks from broods with radio-tagging were being recovered less after their rst winter than hawks from untagged broods was 0 267 (Fisher exact test, onetailed). The combined probability that the proportion of retraps and recoveries from radio-tagged broods was biased below that of broods without radio-tagging, treating recovery and retrapping data as independent samples, was 0 07. The overall recovery rate for hawks after their rst winter was 14% in the intensive area and boundary strip, higher than the 6% in the peripheral area (Fisher exact test, P = 0 02). Table 1 does not include 156 hawks trapped in their rst winter that had not been ringed in the nest: the total was 234 captures, of which 78 were part of the 351 ringed in nests. The 39 ringed hawks among 116 captured outside the intensive area was a very similar proportion to the 39 among 118 captured within the area. Data from both areas could therefore be combined to give a Lincoln±Seber estimate of 1046 (2 10 SE) young produced on the island during 1980±3, or 262 young per year, of which 34% had been ringed in the nest and another 15% were trapped as juveniles in their rst winter. Of the 18 adults trapped on 17 nests in 1986 (14 females, 4 males), six (33%) had been ringed as nestlings, another three (17%) as trapped juveniles and two as adults. The 29 hawks that were trapped in winter hunting areas after being radio-tagged at nests (Table 1) do not include the recapture of one hawk with a radio but without its ring. There were no further rings lost from 44 hawks found dead in their rst winter, so the single lost ring represents only 1 4% of the 74 hawks that had radios as separate markers in their rst winter. No rings had been lost from the 91 hawks trapped near nests that had been marked separately in the nest with leg-tag radios. Including 95 hawks ringed as nestlings in 1984, a total of 446 were ringed on Gotland during the 5 years from 1980. Excluding all recoveries made by radio-tracking, 49 of these hawks were reported after being recovered dead, of which 31 had died in their rst year (Table 2). The overall recovery rate of 0 10 is low compared with the rate of 0 14 for 250 hawks marked on Gotland during 1970±9, and 0 15±0 16 for 1504 goshawks ringed in nests on the Swedish mainland. However, none of the di erences were signi cant (P > 0 1), and a low public recovery rate on Gotland was expected during the study, because 63 hawks recovered dead by radio-tracking were not left in situ to be available for public recovery. The proportion recovered dead in their rst year on Gotland during 1980±4 (0 64) was close to the 0 66 recorded elsewhere in Sweden and the 0 61 recorded on Gotland in the preceding 10 years. A simple estimate of rst-year survival of goshawks ringed on Gotland in 1970±84 was 38%, with a 5% binomial standard error. Ringing also recorded three emigrants and one immigrant during the study. Two of the 446 hawks ringed as nestlings were recovered in Poland, and
1025 R.E. Kenward, V. MarcstroÈm & M. Karlbom Table 2. Numbers of goshawks ringed in Sweden north and south of 60 30 0 during 1970±9, and on Gotland during 1970±9 and 1980±4, with totals recovered dead and the proportion of rst year recoveries Area Period Marked n Total recovery proportion of marked First-year recovery n proportion of recovered Gotland 1980±4 446 49 0 10 31 0 64 1970±9 250 36 0 14 22 0 61 Southern Sweden 1970±9 813 127 0 16 84 0 66 Northern Sweden 1970±9 691 104 0 15 77 0 74 one of the 156 unringed birds trapped in winter was retrapped in southern Sweden. During the 1970s, one of the 250 ringed nestlings had emigrated to the island of OÈ land, close to the Swedish mainland, bringing the total emigration records to four of 696 hawks. One hawk ringed as a nestling in Finland was retrapped on Gotland in 1982 during its rst winter, two rst-winter hawks ringed in Finland were killed on Gotland during the 1970s. Immigrants and emigrants were of both sexes. BREEDING SUCCESS AND PRODUCTIVITY It is likely that some nests in the intensive study area that contained fresh lining material, indicating occupation by hawks, were not detected in 1980. Subsequently, the number of known occupied nests varied between 31 and 39 during the study (Table 3). Numbers of nests that reared young gave no evidence of an increasing or decreasing trend in the breeding population. Eggs were laid in 139 of 165 occupied nests in the intensive area (84%), and chicks were reared old enough to ring in 79% of the 136 that could be checked again. Combining these values, 67% of occupied nests reared young in the intensive area compared with 66% of 155 nests in the peripheral area (omitting four that were deserted following capture of males on the nest). Nine of the 108 nests with marked young were not checked at edging, but young ew from 94 of the remaining 99. Fledging success was therefore 75% of nests in which eggs were laid (binomial SE 2 3%) and 64% of occupied nests (2 3%). Sizes of successful broods did not di er signi cantly between years (F 4,81 = 0 74, P > 0 2), or between the intensive and peripheral areas (t 117 = 0 83, P > 0 2). Successful broods had 2 24 (SE 2 0 08) young in the intensive area and 2 29 (2 0 07) overall. Productivity was thus 1 45 young per occupied nest or 1 73 young per nest with eggs laid. FATE OF RADIO-TAGGED HAWKS From a total of 185 radio-tagged juvenile goshawks, 89 (48%) were tracked until they either died or reached the following April±June period in which they normally moulted their feather-mounted radiotags (Table 4). Tracking ceased for another 16% when radios were heard to fail or were found moulted before the following April, and signals were lost without explanation for another 36%. Combining the tags that failed or moulted early into Table 3. Numbers of nests and productivity data for goshawks in the intensive study area and peripheral areas on Gotland during 1980±4. Numbers of nests containing eggs, chicks and edglings are shown with the number checked as the denominator Intensive area Peripheral area nests containing clutch and brood size (n) nests containing eggs/lining chicks/eggs edglings/chicks eggs edglings chicks/lining 1980 23/26 19/23 19/19 3 14 (7) 2 06 (18) 8/27 1981 28/31 22/26 20/22 3 25 (16) 2 31 (16) 21/34 1982 26/33 19/25 17/19 2 78 (18) 2 38 (16) 20/36 1983 31/39 23/31 18/19 2 52 (27) 2 11 (18) 27/36 1984 31/36 25/31 20/20 3 06 (18) 2 39 (18) 17/26 Totals 139/165 108/136 94/99 2 81 2 24 103/159
1026 ASSAB models from radio-tagged goshawks Table 4. Outcome of tracking for juvenile and adult goshawks with tail-mounted radio-tags on Gotland during 1980±4 Number tagged Recovered dead Normal moult Premature moult Radio failure Loss of signal Juveniles 1980 31 5 0 2 3 21 1981 37 5 11 2 3 16 1982 38 16 9 3 0 10 1983 39 10 10 7 0 12 1984 40 8 15 11 0 6 Total 185 44 (24%) 45 (24%) 25 (13%) 6 (3%) 65 (36%) Adults Total 133 19 (14%) 87 (66%) 6 (5%) 6 (5%) 15 (11%) a category of known premature failures, the outcome of tracking di ered signi cantly between years (w 2 12 = 39 5, P < 0 001). The main change was a decrease during the study in unexplained signal losses (w 2 4 = 15 5, P < 0 01), with a corresponding increase in numbers tracked until April (w 2 4 = 10 7, P < 0 05). The known premature failures were caused mainly by faulty radios in the rst 2 years of the study and only by moulting in the last 2 years (Fisher exact test, P < 0 001). Adult hawks were radio-tagged in winter, with less time than juveniles before the moult. A higher proportion of adults than juveniles survived to moult (Table 4), and a relatively low proportion of their signals were lost (w 2 4 = 59 3, P < 0 001). There were too few deaths, premature failures and unexplained signal losses among adults to investigate variation between years. Among hawks that were presumed live when tracking ceased (because tags moulted in spring or were known to fail early), 14 of 76 juveniles (18%) and 32 of 99 adults (32%) were subsequently recaptured or recovered dead. For hawks whose radio signals were lost, 16 of 65 juveniles (23%) and ve of 15 adults (33%) were recorded again. The absence of any tendency for hawks in the lost category to be re-recorded less than hawks deemed live when tracking ceased indicates that loss of signal was not associated with disproportionate mortality. Most losses probably resulted from undetected radio failure: among the 21 hawks in the lost category that were later recaptured or recovered dead, seven were recovered still carrying radio-tags, and in every case the radio had failed. Only two deaths of radio-tagged hawks were attributed to disease (Table 5). The proportions that died from other causes of death did not di er signi cantly between juveniles and adults (w 2 3 = 0 6, P > 0 2). The majority (65%) of 63 deaths were from natural causes, the remaining 22 hawks being reported shot (which was legal in defence of poultry), found dead with shot damage, or recovered from concealment in gardens, dung heaps and refuse containers. The remaining 24% of hawks were too decayed for detailed post mortem examination, but they showed no sign of being killed by man or of traumatic injury. Starvation was the dominant natural cause of death. SURVIVAL RATES FROM RADIO-TAGGED HAWKS With no evidence of low ring recovery rates for hawks whose signals were lost unexpectedly, survival was estimated without a correction for tag-loss bias. Lost tags were censored at the time of loss, and the reliability of the data set was increased by adding 20 nestlings that were not tracked after loss from nest areas in 1985 and 1987 because they were tted only with leg-mounted tags. For modelling, survival of juveniles was estimated from 1 July to 31 March of the following year (Fig. 2), after which Table 5. Causes of death of juvenile and adult goshawks with tail-mounted radio-tags on Gotland during 1980±4 Killed by humans Other trauma Starvation Disease Natural death, cause unknown Juvenile (n = 44) 14 6* 11{ 2{ 11 Adult (n = 19) 8 2* 5{ 0 4 Totals (n = 63) 22 (35%) 8 (13%) 16 (25%) 2 (3%) 15 (24%) *Other goshawks had killed one juvenile and one adult; other traumas were impact injuries. {There were proventricular ulcers in three starved hawks, abundant intestinal nematodes and/or Coccidia in three. {Pasteurella infection caused one death and one was attributed to acute circulatory failure.
1027 R.E. Kenward, V. MarcstroÈm & M. Karlbom Fig. 2. The proportion of male and female goshawks, with 95% con dence limits, that survived from 1 July in the year of hatching to 31 March the next year. there was a rapid increase in the size of con dence limits as tags were moulted or exhausted their cells. Survival of older hawks was estimated for the year from 1 April to 31 March; egg laying was in early April. The survival rate of juvenile male and female hawks tended to diverge from September onwards. At the end of 9 months, the staggered-entry estimate of male survival was 0 49 (2 0 08), compared with 0 71 (2 0 07) for females (Table 6). Survival rates di ered signi cantly at 31 March (using Cox±Oates variances, z = 2 07, P < 0 05) and month-by-month during July±March (Wilcoxon matched pairs test, z = 2 31, P = 0 02). The many deaths of males during November±March, when sample sizes for the staggered-entry analysis had been reduced by censoring, drew this estimate of survival rate on 31 March below that based on tag days. However, there was much less di erence in the estimates for the whole rst year of life to 30 June. In this case, the staggered-entry survival rate for males was 0 49 (2 0 12) compared with 0 48 from 25 deaths in 12 360 tag days, and for females was 0 64 (2 0 12) compared with 0 61 from 19 deaths in 14 130 tag days. Of the 19 deaths recorded for radio-tagged adult hawks, ten occurred in the rst month after marking, including six killings by farmers. To minimize bias from hawks that were especially vulnerable, both to being trapped and to dying, all data from trapped adults were excluded for the rst month after capture. This reduced the number of deaths to seven for hawks more than 2 years old, and four for hawks in their second year (including two deaths after 31 March among hawks marked as juveniles). The estimated survival rates based on tag days were identical at 0 79 for both males and females more than 2 years old, so data from the 36 males and 42 females were pooled to obtain the staggered-entry estimate of 0 83 (2 0 09). For comparison with the rst-year survival rate estimates from ringing, which included many hawks Table 6. Age-speci c survival rates of goshawks radio-tagged on Gotland during 1980±4 Male Female 9 months 2nd year Older 9 months 2nd year Older Hawks 104 16 36 101 20 42 Tag-days 11136 1401 4734 12761 2137 6295 Deaths 25 2 3 17 2 4 Survival rate: From tag-days 0 54 0 59 0 79 0 69 0 71 0 79 Staggered entry 0 49 (2 0 08) ± 0 83 (2 0 09)* 0 71 (2 0 07) ± 0 83 (2 0 09)* *For hawks > 2 years old, survival data from both sexes were pooled in the staggered entry estimate.
1028 ASSAB models from radio-tagged goshawks Table 7. Age-speci c productivity of 116 radio-tagged goshawks on Gotland during 1980±4, together with data from 165 nests in the intensive study area Male Female Intensive study 1st year 2nd year Older 1st year 2nd year Older area (SE) Did not lay eggs (n) 16 3 5 30 11 16 ± Layed eggs, failed (n) 0 5 2 0 0 4 ± Fledged young (n) 0 3 10 0 1 10 ± Fledging success ± 0 36 0 83 ± 1 00 0 71 0 75 (0 03) Brood size ± 2 00 2 30 ± 2 00 2 20 2 24 (0 08) Young per clutch ± 0 75 1 92 ± 2 00 1 57 1 68 Young per pair 0 0 55 1 35 0 0 17 0 73 ± that lacked a sex category, data from radio-tagged juveniles of both sexes were pooled. The rst-year survival rate was then 0 58 (2 0 05), much higher (z = 2 83, P < 0 01) than the estimate of 0 38 (2 0 05) from 85 ring recoveries (Table 2). Combining age categories, and including deaths recorded within the rst month of tagging adults, the 35% of hawks killed by humans among 63 deaths recorded by radio-tagging did not di er signi cantly from the 48% reported killed among 50 independent ringing recoveries on Gotland during the 1980s (w 2 = 1 47, P > 0 2). However, restricting the analysis to adult hawks that died more than a month after being trapped, 12 had been killed among 18 ring recoveries compared with just two of nine recorded by radio-tracking (Fisher exact test, P = 0 04). THE POPULATION MODEL Table 7 contains age-speci c productivity data from radio-tagged hawks. These data were used with survival data in Table 6 to model the goshawk population on Gotland. No radio-tagged goshawks in their rst year were found near occupied nest areas in the breeding season, nor were feathers of other rst-year birds found at any nest. Only one of 12 second-year females was in a pair that laid eggs, compared with 14 of 30 older females (Fisher exact test, P = 0 03), so the 0 083 and 0 467, respectively, were used in the model for the proportions laying. The proportion of males paired to females that laid eggs was set at 0 714 for both adult age classes, because the eight of 11 paired second-year hawks was a similar proportion (0 727) to the 12 of 17 (0 706) older birds. The proportion of radio-tagged hawks in pairs that laid eggs was higher among males than females for both age classes, with the di erence signi cant in secondyear hawks (Fisher exact test, P = 0 004). There was also a tendency for second-year males in pairs that laid eggs to have less success rearing young than older males (Fisher exact test, P = 0 06), whereas rearing success of second-year and older females did not di er. The model therefore used the observed Table 8. Numbers and breeding parameters of male : female goshawks at 31 March each year on Gotland, modelled for 100 juveniles of each sex from survival and breeding rates of radio-tagged hawks, with minimal adjustment of survival (b) to equalize numbers of each sex with eggs (B m = B f ) and minimal adjustment of breeding success (u) for population stability (l =1) Staggered-entry survival Observed m:f B m =B f m:f l=1 m:f Tag-day survival Observed m:f B m =B f m:f l=1 m:f Intensive study area (SE) n 1 (juveniles) 49 : 71 49 : 71 49 : 71 54 : 69 54 : 69 54 : 69 ± n 2 (2nd year) 28 : 50 28 : 50 28 : 50 31 : 48 31 : 48 31 : 48 ± b (older hawks) 1 : 1 1 : 1 005 1 : 1 005 1 : 1 1 : 1 043 1 : 1 043 ± n 3+ (older hawks) 133 : 233 133 : 239 133 : 239 117 : 180 117 : 218 117 : 218 ± B (lay eggs) 116 : 113 116 : 116 116 : 116 106 : 88 106 : 106 106 : 106 ± b (prop. breeding) 0 55 : 0 32 0 55 : 0 33 0 55 : 0 33 0 52 : 0 30 0 52 : 0 32 0 52 : 0 32 ± u (all broods) 1 : 1 1 : 1 1 : 1 07 1 : 1 1 : 1 1 13 : 1 18 ± g ( edging success) 0 75 : 0 73 0 75 : 0 73 0 75 : 0 75 0 74 : 0 73 0 74 : 0 73 0 74 : 0 73 0 75 (0 03) f (brood size) 2 27 : 2 18 2 27 : 2 18 2 27 : 2 27 2 27 : 2 18 2 27 : 2 18 2 56 : 2 57 2 24 (0 08) y (young per clutch) 1 71 : 1 60 1 71 : 1 60 1 71 : 1 71 1 67 : 1 60 1 67 : 1 60 1 88 : 1 88 1 68 l (population change) 1 : 0 98 1 : 0 99 1 : 1 0 97 : 0 93 0 97 : 0 97 1 : 1 1 1
1029 R.E. Kenward, V. MarcstroÈm & M. Karlbom Fig. 3. An age-speci c survival and breeding model showing the percentage of goshawks remaining alive (open bars), and breeding (in black) in spring of each year of life; s1, s2 and s3 + are the survival rates in years 1, 2 and older; b1±b3 + and y1±y3 + are proportions breeding and young per clutch; o is maximum age. values of 0 75 young per breeding second-year male and 1 92 young per older male, with the average value of 1 6 for both female age classes. Two versions of the population model were tested (Table 8). The rst used staggered-entry estimates of survival rates for juveniles and for hawks more than 2 years old, and therefore took account of seasonal variation in survival. It used estimates based on tag days for second-year hawks, because too few deaths were recorded for reliable staggered-entry estimation; these tag-day survival rates were between the staggered-entry estimates for juveniles and for older hawks and, as among juveniles, they were higher for females than for males. The second version of the model used survival rates estimated from tag days for all age categories. Both versions of the model assumed that goshawks stopped breeding and died at age 18. They started with 100 young hawks of each sex and estimated the numbers present in spring each year. With the staggered-entry survival rates (Tables 6,7), the rst version of the model was balanced (l = 1) for males, but gave a 2% per annum population decline for females. An increase of 0 5% in the survival of adult females, by adjusting b from 1 to 1 005, balanced the numbers of each sex that laid eggs (i.e. B m = B f ) and was small compared to the 0 08 standard error of the survival rate estimate. Equal numbers of breeders could also be obtained by a 2 5% increase in the survival rate of second-year females (i.e. from 0 71 to 0 74), or by a 3% increase in survival of juvenile females. After adjusting the survival rate, the modelled female population declined by 1% per annum, but came into balance with a 7% increase in productivity. This value for u was most reasonably applied as a 3% increase in edging success, from the 0 73 recorded for radio-tagged females to the 0 75 (SE 2 0 03) recorded for radio-tagged males and for all nests in the intensive study area, and a 4% increase in the size of successful broods. These adjustments resulted in the same average brood size for males and females and were within the coe cients of variation for edging success (4%) and brood size (4%) in the study area. The second version of the model, based only on constant survival rates estimated from numbers of tag days, required a minimum increase of 4 3% in the survival of adult females to balance the number of each sex that bred, followed by productivity improvements of 13% for females and 18% for males to prevent a population decline (Table 8). These improvements could not be accommodated within the coe cients of variation for edging success (4%) and brood size (4%) in the study area. Alternatively, the number of males and females became the same as in the rst model, which balanced the male population and required a 7% productivity adjustment for females, if the survival rates for hawks > 2 years old were increased by 2 8% (to 0 81) for males and 6 3% (to 0 84) for females. These survival rate adjustments were again less than the standard error of the staggered entry survival rate estimates. Using only the staggered-entry estimates of survival rates, with the observed 1 68 young per clutch in the study area and no age-speci c breeding data, the simpli ed model (eqn 10) predicted that 56% of all males present in spring, and 33% of females, would breed in a balanced population (Fig. 3). This was not testable with independent data. However, without rst-year breeders, the prediction was that 73% of older males would breed, and 42% of females. Among the 28 adult males that were radio-tagged in winter hunting areas and survived to spring (Table 7), 71% bred and the 9% binomial standard error generously encompassed the predicted 73%. Di erences in the breeding rate between second-year and older females prevented the same test being carried out, but after taking account of the 8% breeding by second-year females the model predicted that 48% of older females would breed, very close to the 47% (2 9%) recorded for 30 radio-tracked older females. Discussion If animal performance, in terms of survival and productivity, is to be estimated accurately by radio-tag-
1030 ASSAB models from radio-tagged goshawks ging, it is essential to test for possible adverse e ects of tagging. If the goshawks with tail-mounted radiotags had survived worse than those without radios, their recapture rates should have been relatively low, unless radio-tagging also made them prone to recapture; for example, hawks that had been trapped to attach radios might have been `trap happy' through learning that traps contained food. However, recapture rates of hawks with radios were the same as for hawks outside the intensive areas that had no prior knowledge of traps (Table 1). Moreover, recording rates after the rst winter, that were not based on these traps, were 15% for radiotagged hawks and 12% for others in the intensive area. These rates were above the 6% rate outside the intensive area because of increased recording e ort in the intensive area, through the trapping of breeders with nooses on nests. Although a number of studies have shown adverse impacts of harnessmounted tags on survival (Johnson & Berner 1980; Small & Rusch 1985; MarcstroÈ m, Kenward & Karlbom 1989; Patton et al. 1991; Foster et al. 1992), no study seems to have shown that radios mounted on tail feathers reduced survival. Small sample sizes may sometimes preclude demonstration of adverse e ects (White & Garrott 1990). However, samples in this study were large enough to estimate a probability of only 0 07 that hawks from broods with radio-tags were being re-recorded less frequently, and therefore surviving less well, than those from untagged broods. If performance estimates are to lack bias, it is also important that the marked animals are a representative sample of the population as a whole. This was probably the case among tagged juveniles. Although 25% of nestlings in the least accessible broods were not targeted for tagging (Table 1), there was no reason for their survival to have di ered from the others. Although 18% of young in targeted broods were not caught to attach tail-mounted tags, known deaths among these were included in survival estimates and their long-term recovery rate was similar to that of captured hawks in the same broods (Table 1). Among adults, however, trapping bias was likely. Trapped hawks have lower weights than those sampled in other ways (Mueller & Berger 1970; MarcstroÈ m & Kenward 1981b). Excluding data from the rst month after capture would have omitted the most disadvantaged hawks (White & Garrott 1990). However, any remaining low-e ciency females may have been more likely to attempt breeding than low-e ciency males. This is because the breeding of accipiters depends on hunting by males until the young are several weeks old (Newton 1979, 1986). The importance of males being experienced could be seen in the poor breeding success of males in their second year, compared to older males (Table 7). We suggest that a few females with poor ability to contribute to broodraising may have managed to nest by choosing an e cient male, but then have had below-average edging success, as was observed for radio-tagged females more than 2 years old (Table 7). In contrast, the average productivity of radio-tagged adult males was 1 71 young per clutch, compared with 1 68 for the intensive area and 1 73 for the whole island (Table 8). Although the productivity of radio-tagged males was included in the overall estimates, their 20 nests were only 8% of the total 242 nests with eggs (Table 4). Data from the winter trapping showed that ringed hawks were equally likely to be recaptured (Table 1) and were a similar proportion of recaptures inside and outside the intensive area. This was probably because the mean rst-winter dispersal distances were more than 10 km in rabbit-rich areas and more than 20 km elsewhere (Kenward, MarcstroÈ m & Karlbom 1993b), thereby mixing birds from within and beyond the intensive area and o setting any e ects of di ering trapping and marking densities (Fig. 1). Data from both areas could therefore be combined to estimate the total population of hawks. Mark-recapture estimated an average juvenile population of 262 goshawks, of which 34% were ringed in the nest and 15% more at traps in winter. Further con dence in this estimate of the juvenile population came from the 18 breeders that were trapped later with nooses on nests: 33% had been ringed in nests, 17% at traps in winter. If all 262 young were produced on Gotland, with a mean productivity of 1 73 young per clutch (Table 3), there were an average 151 clutches laid on the island each year. Only one ringed Finnish hawk was caught during the 4 years of winter trapping, and the two others killed in the previous decade suggest that this immigration rate was typical. If one applies the 23% retrap rate for hawks ringed on Gotland to the Finnish birds, and the estimate from MarcstroÈ m& Kenward (1981b) that 4% of Finnish goshawks were ringed, there may have been approximately 25 hawks from Finland present each year. This would cause a 10% overestimate of the number of goshawk clutches laid on Gotland. However, rstwinter hawks trapped in southern Sweden tend to return later to northern areas (MarcstroÈ m & Kenward 1981b), so few immigrants may have remained to breed. Three emigrants were recorded among 696 hawks ringed as pulli on Gotland (0 43%), or four in 852 (0 47%) if one includes the trapped juvenile. Applying the average Swedish recovery rate of 0 15 (Table 2) to these birds, there was 3% (2 1 6%) emigration. Even if net immigration reduced l, such that survival was not exactly matched by breeding, this
1031 R.E. Kenward, V. MarcstroÈm & M. Karlbom would only cause a di erence in survival and breeding rates between the two sexes if immigration was sex-biased. There is male-bias in long-distance movements of goshawks (Sulkava 1964; MarcstroÈ m & Kenward 1981b). However, any disproportionate immigration of males to Gotland must have been slight, because staggered-entry survival estimates matched breeding data in males, and required little adjustment for females. Although survival rates that failed to include seasonal change gave a less adequate match with breeding data, all the survival estimates for juveniles were greater than contemporary estimates from ringing by more than 10%. Why? One possibility was that juvenile survival estimated from hawks ringed on Gotland in 1980±4 was biased downward by failure to complete recoveries of adults. However, the last hawk in the 1980±4 cohorts was ringed 14 years previously, and in 267 previous Swedish records the longest-lived was a hawk recovered at 14 years old. It is therefore unlikely that further records will increase the estimate of 0 38 (2 0 05) for the ringed hawks to within two standard errors of the 0 58 (2 0 05) estimated by combining radio-tracking data from both sexes. The estimate of 0 38 compares with other Scandinavian estimates of 0 38±0 36 from data recorded before 1980 (HoÈ glund 1964; Haukioja & Haukioja 1970; Saurola 1976). Other possible reasons for the survival of juvenile hawks to be underestimated by ringing were: (i) loss of rings, as recorded in this study and also in Germany (Ziesemer 1981); and (ii) ring recovery bias, through inexperienced raptors being more likely than adults to die in places where their rings would be reported, especially where many were killed by man (Haukioja & Haukioja 1970; Newton 1979). Thus, 48% of the ring recoveries on Gotland were from hawks killed by humans, compared with only 35% of hawk deaths recorded by radio-tracking in the same period (Table 5); after minimizing trapping bias by excluding data from the rst month after trapping in winter, only 22% of radio-tracked adults were killed by humans. The standard error of 0 09 was large relative to the survival rate of 0 83 estimated by radio-tracking hawks more than 2 years old, and changes in this parameter would have in uenced l strongly. Unfortunately, adult survival could not be estimated reliably from ringed goshawks on Gotland, as an independent check on the radio-tracking, because only one hawk more than 3 years old was recovered dead. However, the estimate of 0 83 from radiotracking was comparable with estimates of 0 81±0 85 from ringed Scandinavian goshawks in the same age classes (Haukioja & Haukioja 1970; Saurola 1976). Catchpole, Freeman & Morgan (1995) have shown theoretically that estimates of adult survival from ringing are much less vulnerable than juvenile survival to reporting bias. The similarity of adult survival rate estimates from ringing and radio-tagging among buzzards in Britain (R.E. Kenward & S.S. Walls., unpublished data), as well as for goshawks on Gotland, con rms this in the eld. Radio-tagging is most useful for estimating juvenile survival rates, not only to minimize nding bias but also for rare species, where detection of short-term changes in survival, correlations with environmental factors and causes of death may be important for conservation. Marking bias can now be avoided when estimating survival rates of young adults, by tagging nestlings with radios that transmit for several years (Kenward 1993), but ringing is an important check on these estimates. Ringing was also an important complement to radio-tagging on Gotland to show that unexplained losses from the radiotracked population did not represent deaths associated with destruction of tags. The poor reliability of tags in the early years of this study (Table 4), and moulting of tags throughout, gave, in due course, enough ring recoveries to show that hawks known to leave the tracked population alive were not more likely to be recovered later, and thus to have survived better, than those lost inexplicably. The high survival of juveniles shown by radio-tagging has consequences for our understanding of raptor ecology. By using the ring recoveries in Fennoscandia prior to 1970 (HoÈ glund 1964; Haukioja & Haukioja 1970), MarcstroÈ m & Kenward (1981b) estimated an autumn ratio of 1 9 adult goshawks per juvenile. In contrast, the ASSAB model (Fig. 3) predicts 2 9 adults per juvenile in autumn and 3 8 in spring. Another marked di erence from earlier studies is in the proportion of goshawks breeding in their rst year, where in depressed populations up to 35% of nests have had one or both sexes in juvenile plumage (Link 1986) compared with none on Gotland. In radio-tagged buzzards too, the incidence of early breeding increased towards areas being newly colonized and was least in an area with a stable breeding population (R.E. Kenward & S.S. Walls, unpublished data). With a minimal survival estimate of 0 66 in the rst year, modelling and survey showed that the stable buzzard population had only 25% of birds breeding in spring; for goshawks on Gotland the rst-year survival of 0 58 was associated with 41% of the spring population breeding (Table 8). High estimates of juvenile survival in healthy populations limit the scope for population regulation by densitydependence in this parameter, and thereby increase the potential importance of density-dependent delay in breeding. Prevalence of nonbreeders in healthy raptor populations may well not be limited to goshawks and buzzards. In 26 early analyses of raptor ring recoveries, the highest estimate of rst-year survival was