1This study examined annual differences in redshank Tringa totanus winter mortality caused by predation by Eurasian sparrowhawks Accipiter nisus over an 11-year period at a rocky shore in south-east Scotland.
2Redshank numbers at the beginning of a winter showed no trend over the study period and winter mortality rates also showed no temporal trend. Mean winter mortality through predation by sparrowhawks was 30·6% for juveniles and 5·6% for adults.
3The study demonstrated density-dependent winter mortality in both juvenile and adult redshanks due to predation by sparrowhawks. Wind speed was also positively associated with juvenile winter mortality but had a weaker influence on adult mortality. This apparent influence of wind speed on juvenile mortality was not confirmed by correlations between monthly juvenile mortality and weather variables (minimum temperature, rainfall and wind speed).
4Density-dependent functions differed between adult and juvenile redshanks, consistent with differences in the competitiveness of the two age classes. It is suggested that individual differences in vulnerability to predation arose through differences in individual susceptibility to density-dependent competition during foraging. Companion studies indicated that as bird density increased more birds were forced to spend more time feeding in an area where risk of predation was high.
Much of the interest in predation is due to the potential effects of predators on the harvest of prey species by humans, which may often have economic implications. Hence, the effect of predators on populations has been the subject of considerable research for many years (e.g. Sih et al. 1985; Newton 1993, 1994, 1998). The impact of predation on prey populations is largely governed by changes in predator–prey interactions with changes in the density of prey, and whether predation compensates for other influences on prey abundance or has additive effects. Pioneering studies of predation often suggested that while predators could take much prey their effects on prey populations were negligible, but more recent studies often point to a major although variable effect of predators on their prey populations. For example, the pheasant Phasianus colchicus is an important quarry of hunters in many countries and Göransson (1975) concluded that most predation by northern goshawks Accipiter gentilis on pheasants in southern Sweden was compensatory and so had little effect on pheasant populations. On the other hand, later research in Sweden concluded that goshawks could reduce breeding populations of pheasants and predation was a largely additive source of mortality (Kenward 1977, 1986; Kenward, Marcström, & Karlbom 1981).
Theoretical considerations suggest that generalist raptors such as the goshawk which take many prey species can occur at a high stable abundance by switching between the different prey species according to the density of each prey (Andersson & Erlinge 1977; Hanski, Hansson & Henttonen 1991). This density-dependent predation regulates prey abundance and produces stability in prey populations. Density-dependent effects consequently are critical in affecting population dynamics and population size. For shorebirds (Charadrii), most attention has focused on density-independent effects, especially the influence of severe weather on winter survival, probably because severe weather mortality can be spectacular (e.g. Baillie 1980; Davidson & Evans 1982; Clark et al. 1993). Density-dependent processes have been less well studied and doubts have even been raised as to their importance in shorebird population dynamics (Evans & Pienkowski 1984).
Durell et al. (2000) highlighted three reasons why the effect of density dependence on mortality is difficult to study in shorebirds. First, shorebirds can be highly mobile making mortality difficult to measure. Second, density-independent mortality may mask density-dependent processes and makes their isolation difficult. Third, population size may not be sufficiently variable over a study period to detect density dependence. Despite these difficulties, Durell et al. (2000) showed that annual and winter mortality of oystercatchers Haematopus ostalegus on the Exe estuary, south-west England, was consistent with a density-dependent influence, probably acting through starvation and exacerbated by the severity of winter weather (Durell et al. 2001).
For large shorebirds such as the oystercatcher, winter mortality is probably mainly attributable to starvation, but in many smaller species, such as the redshank Tringa totanus, raptor predation can be an important source of winter mortality (Page & Whitacre 1975; Whitfield 1985a,b; Whitfield, Evans & Whitfield 1988; Cresswell & Whitfield 1994). Redshanks can experience catastrophic mortality due to severe weather on some wintering sites in Britain (e.g. Clark et al. 1993) that is probably density-independent, but on some sites they are also particularly vulnerable to raptor predation. On a rocky shore and a small estuary in south-east Scotland, wintering redshanks experienced heavy mortality due to predation by Eurasian sparrowhawks Accipiter nisus (Whitfield 1985b; Cresswell & Whitfield 1994). The predator–prey system involving sparrowhawks and redshanks on these sites is particularly amenable to an examination of density-dependent mortality because within a winter the redshank population is stable in membership, deaths can be recorded directly through recovery of kills, density-independent mortality through severe weather effects is rare, and numbers of redshanks can vary substantially from year to year. In this paper, I describe the results of an 11-year study of redshank mortality due to predation by sparrowhawks that examines whether redshank mortality was density-dependent on this rocky shore site.
The study site was an approximately 5 km stretch of rocky shore at Scoughall on the outer Firth of Forth, East Lothian, south-east Scotland (Whitfield 1985a,b, 1990). Observations were conducted in the ‘winters’ (August to April) of 1983–93, and each winter is denoted by the year in which it ended, e.g. the 1982–83 winter is termed the 1983 winter.
A complete census of redshanks over the whole study site was made around high tide when birds were forced into restricted feeding areas, at pre-roost sites and when arriving at or leaving roosts. Several complete counts were made over a series of high tides each month and all counts within a series always fell within 10% of the maximum count made during that series. Population size could also be estimated from the proportion of the population that was colour-ringed (Whitfield et al. 1999) and the number of colour-ringed birds alive. There was a very high correlation between these estimates and maximum monthly counts made within three days of each other (20 estimates calculated from arbitrary dates across the study period, Pearson's r = 0·99).
I assumed that the area of the study site remained constant over the study period and used the maximum count in early October for each winter as the relevant population density that could affect mortality. This count was used because in most winters no redshanks died before October and by this time of the year all birds that would use the site in the following months had settled. Very little emigration or immigration was recorded after October, until spring migration began in April and so effectively the overwintering population was a resident group of birds with a stable membership (Whitfield et al. 1999). The maximum count was used because the mean count for a series was less well correlated with population estimates derived from numbers of colour-ringed birds and counter acted small errors due to occasional counts which observations of colour-ringed birds suggested were incomplete. During the course of conducting counts, all birds were aged as juvenile (birds in their first winter) or adult (birds in at least their second winter) on plumage differences (Whitfield 1985b; Cresswell & Whitfield 1994; Whitfield et al. 1999). This allowed the numbers of juveniles and adults in October to be calculated for each winter.
Remains of dead redshanks were recovered by methodical searches of the shoreline and adjacent inland areas. Remains were classified as having been killed by sparrowhawks and aged as juvenile or adult using the criteria of Whitfield (1985b) and Cresswell & Whitfield (1994). Remains of sparrowhawk kills were very distinctive and not easily confused with remains of kills by other predators (see also Cresswell 1995). Observations of redshank kills by sparrowhawks showed that in 95% (n = 20), remains were left within the search area. It was estimated that 100% of redshank remains (n = 29) of all sparrowhawk kills within the search area were recovered by searches during tests of search efficiency (details in Cresswell & Whitfield 1994). If the death of a redshank had not been observed, the date of death of a redshank was estimated by reference to the date of the previous search of the area concerned and the appearance of the remains (Whitfield 1985b; Cresswell & Whitfield 1994). Records of the appearance of remains of known-date kills left in situ and tests of search efficiency suggested that estimation of the date of death was usually accurate to within 1 week (Cresswell & Whitfield 1994). Due to having many white feathers, remains of killed redshanks remained visible for several months (D.P. Whitfield unpublished data; see also Cresswell 1995), but searches were at most a month apart.
In the majority of cases, after capturing a redshank, sparrowhawks took their prey to the nearest cover above the shoreline (Whitfield 1985a; see also Cresswell 1995), but thereafter could take the corpse to several locations, and leave remains at each, if it was not all consumed immediately after capture. There was therefore a risk that a single corpse could be double-counted, or even more. Therefore, different sets of remains with a similar estimated date of death were only considered to be from different corpses if they had a body part in common (e.g. both sets had a right leg) (Whitfield 1985b; Cresswell & Whitfield 1994). Measurement of redshank outer primary feather length (primary 10: Whitfield et al. 1999), instigated in the winter of 1986, improved the estimation of the number of corpses that different sets of remains represented (since left and right primaries from the same bird are the same length ±1 mm: D.P. Whitfield unpublished data). From the 1986 winter onwards, if two sets of remains with a similar date of death did not share any body part but each had a primary 10 and its length differed by more than 1 mm they were assumed to represent two kills. As all remains from previous winters had been collected and retained (Whitfield 1985b), this method could also be applied to remains from previous winters retrospectively. Winter mortality rate of an age class was given by (n individuals of age class killed by sparrowhawks/n individuals of age class alive in October) × 100%.
Weather data were taken from the British Atmospheric Data Centre website (www.badc.rl.ac.uk) utilizing Meteorological Office daily surface records from the coastal station at Dunbar, approximately 5 km to the south of the study site. These data were used to examine the effect of winter weather on redshank mortality. Numerous studies have documented the effect of low temperature on redshank winter mortality (e.g. Baillie 1980; Davidson & Evans 1982). Several temperature variables were considered initially for analyses, including maximum daily temperature, minimum daily temperature, number of frost days, and number of 3-day periods with freezing minimum temperatures. All these variables covaried strongly, however, and so only minimum daily temperature was used in analyses. It was expected that mortality would be negatively associated with minimum temperature (MIN_T). Three other weather variables were used in analyses: mean daily wind speed (WIND), daily precipitation (RAINF) and number of days with precipitation (RAIND). These variables were chosen as both wind and rain have been implicated in increased risk of predation of redshanks by sparrowhawks (Hilton, Ruxton & Cresswell 1999): it was expected therefore that mortality would be positively associated with rainfall and negatively with wind speed.
Data from the Dunbar weather station were missing from parts of 1985 and 1986. For the missing periods, data derived from other weather stations were used as substitutes after having corrected the data for small differences in climate between the stations. This was done by regressing records from Dunbar on the equivalent records from Belliston, Fife (25 km from the study site: MIN_T, WIND) and East Linton (5 km from the study site: RAINF, RAIND) for the years 1983, 1984, 1987 and 1988.
No wind speed data were available from any station for the 1983 winter. Mean values of each weather variable were calculated for each winter for the period October to February, as most redshank mortality occurred in this period.
As noted earlier, the proportionate measure of winter mortality rate was the percentage of birds in an age class present in October that was found pre-dated during the following winter. The measure of density was the total number of birds (i.e. both age classes) in October There has been a considerable debate and literature over methods used to measure or document density dependence (reviewed by Den Boer & Reddingius 1996; McCallum 2000). A serious analytical problem identified by many authors occurs when mortality and density are measured using the same process (as in many time series census studies), so any methodological errors will be autocorrelated, predisposing the results towards a spurious link between mortality and density. While there are several analytical tools to overcome this problem (e.g. Rothery 1998), in the present study the methods used to assess the number of deaths and population size were independent and so any associated errors should also be independent. Mortality estimates were age specific and hence based only on an annually variable proportion of the total population that was used as a measure of density. Thus, as there should have been little autocorrelation between errors, relationships between mortality, weather and redshank density were explored with regression (Ito 1972; Durell et al. 2000; McCallum 2000). Measures of density (number of redshanks in October) and mortality were loge transformed to normalize variables and associated residuals, and to satisfy assumptions of variance constancy (checked as the most appropriate transformation through probability plots and residual plots from post hoc analyses of regression models). A stepwise regression technique was used, with the criterion of variable entry to a model being a 0·05 probability of F and a criterion of variable elimination from a model being a 0·10 probability of F. The two age classes were analysed separately due to age-dependent mortality (Whitfield 1985b; Whitfield et al. 1988).
Although only a proportion of the population (the number of birds in one of two age classes, varying annually) was used to produce a measure of mortality rate, and this should not have produced autocorrelation of errors (see also Durell et al. 2000), the analyses were re-run using loge (number of age class killed) as the dependent variable. This removed even the smallest risk of a spurious relationship between mortality and density through autocorrelation in measurement error, by removing any methodological link between measures of redshank deaths and population size. As the number of birds killed could rise if proportionate mortality rate was constant with respect to population size, I tested if the slope of the relationship was significantly greater than 1 (e.g. Rothery 1998). Significant deviation from a slope of 1 would indicate that deaths occurred at a higher rate than expected (i.e. that mortality was density-dependent). As the direction of the results was expected apriori from the regression analyses that used mortality rate as the dependent variable, these tests were one-tailed.
The redshank population varied between 90 and 170, with no temporal trend being evident over the study period (Spearman's rank correlation, rs = 0·05, P = 0·883) (Fig. 1). Mean winter mortality of juveniles was 30·6% (range 18%– 56·8%) and mean winter mortality of adults was 5·6% (range 2·8%– 17·7%) (Fig. 2). Winter mortality showed no relationship with time across the study period (juveniles: rs = −0·227, P = 0·502; adults: rs = −0·027, P = 0·937).
As a preliminary exploration of the possible influence of weather variables on mortality, correlation matrices between weather variables and monthly mortality rates (October to March, juvenile mortality only due to sample sizes) were calculated with corrections for multiple comparisons (Table 1). Bearing in mind the likelihood of a Type I error with such a process, very few strong relationships were evident and no weather variable was strongly associated with monthly mortality in more than one month. The only strong association of precipitation with mortality was in the opposite direction to that expected, and no weather variable was correlated with mortality in the three months when most mortality occurred, December to February.
Table 1. Pearson correlation coefficients, r, and associated Bonferroni probabilities for weather variables and monthly juvenile mortality rates. MIN_T = minimum daily temperature (°C), WIND = mean daily wind speed (knots), RAINF = mean daily precipitation (mm), RAIND = number of days with precipitation. n = 11 (winters) for each variable, except WIND (n = 10). Data for 29 February were ignored. Coefficients of P < 0·05 are emboldened and underlined, coefficients of P < 0·10 are emboldened
Individually, the relationship between loge mortality and loge population density was significant (P < 0·001) for both age classes (Fig. 3). In a multiple regression, loge winter mortality rate of juvenile redshanks due to sparrowhawk predation was strongly related to loge population density (P = 0·002) and the addition of loge population density explained 73% of the variation in loge winter mortality (as judged by the change in R2) (Table 2). The best regression model, however, included both loge population density and wind speed, as the addition of wind speed (P = 0·009) improved the fit of loge population density (P = 0·001) and explained an additional 17% of the variation in loge winter mortality. Other weather variables were eliminated from the model (minimum daily temperature, t = −0·606, P = 0·567; mean daily precipitation, t = −0·693, P = 0·510; mean days with precipitation, t = −1·598, P = 0·154). A model from a separate multiple regression, including loge population density and interactive weather variables (minimum daily temperature × windspeed, mean daily precipitation × windspeed) as independents, showed no improvement over the model that included only loge population density (F = 0·073, P = 0·794 and F = 0·436, P = 0·53 for weather variables, respectively).
Table 2. Results of regression models produced by a stepwise procedure for loge population density (log POPN) and weather variables on loge juvenile winter mortality due to sparrowhawk predation. Mean daily wind speed (October to February, WIND) was the only weather variable not eliminated from the model. B, estimated unstandardized coefficient; SE, standard error of B; t, t-value; P, regression significance
In a multiple regression, loge winter mortality rate of adult redshanks due to sparrowhawk predation was also strongly related to loge population density (P = 0·002) and the addition of loge population density explained 72% of the variation in loge winter mortality (Table 3). Addition of weather variables did not improve the regression model according to the criteria used, however, and so all weather variables were eliminated (mean wind speed, t = 2·054, P = 0·079; minimum daily temperature, t = 1·105, P = 0·306; mean daily precipitation, t = −1·194, P = 0·271; mean days with precipitation, t = −0·902, P = 0·397). It should be noted, however, that wind speed only marginally failed to pass the criterion set for inclusion in the regression model.
Table 3. Results of regression models produced by a stepwise procedure for loge population density (log POPN) and weather variables on loge adult winter mortality due to sparrowhawk predation. All weather variables were eliminated from the model. B, estimated unstandardized coefficient; SE, standard error of B; t, t-value; P, regression significance
Although the regression analyses did not identify the 1989 winter as an outlier or as being especially influential, mortality rates of both age classes were especially high in this winter (Fig. 2). Hence, the regression analyses were re-run excluding the 1989 winter. While the strength of the resulting regression models was diminished, there was little substantive change in the results. For loge juvenile mortality the best model included both loge population density and wind speed (log POPN, t = 3·657, P = 0·011; WIND, t = 2·801, P = 0·031) and for loge adult mortality the best model included only loge population density (log POPN, t = 2·653, P = 0·033).
In a regression of logen juveniles killed and loge population density the slope of the relationship was significantly greater than 1 (t = 4·962, d.f. = 10, P < 0·001). In a multivariate analysis with the four weather variables as additional independents, the best model included loge population density (85% variance explained), wind speed (12·5% variance explained) and mean days with precipitation (negative association, 1·5% variance explained). In this model the slope of the log–log relationship between the number of juvenile deaths and density was significantly greater than one (t = 5·548, d.f. = 8, P < 0·001). In a regression of logen adults killed and loge population density the slope of the relationship was significantly greater than one (t = 4·247, d.f. = 10, P < 0·001). In a multivariate analysis with the four weather variables as additional independents, the best model included only loge population density (82% of variance explained).
This study has demonstrated density-dependent winter mortality in both juvenile and adult redshanks due to predation by sparrowhawks. As pointed out by Durell et al. (2000), intuitively it might be expected that density dependence may start to operate above a threshold density. This was not obvious from the plot of juvenile mortality (Fig. 3a), but was apparent from the plot of adult mortality (Fig. 3b). The differences between age classes may be because adults are more buffered against the effects of competition than juveniles and so density dependence in adults operates above a higher density threshold than in juveniles. Density dependence in the study system appears to come about because as competition for food on intertidal areas (seaward from the strandline beaches) increases, more individuals are forced to feed on the strandline (around high tide marks) close to cover (D.P. Whitfield, unpublished data). Food intake on the strandline is greater than on intertidal areas but risk of predation by sparrowhawks is far greater (Whitfield, in press) hence as more birds feed on the strandline, more are killed by sparrowhawks (D.P. Whitfield, unpublished data). Competition on intertidal areas is asymmetrical between age classes, however, as adults are more likely to hold territories than juveniles (D.P. Whitfield, unpublished data). Hence, the effects of competition are likely to be felt at lower population densities in juveniles. These types of processes have been modelled by Goss-Custard et al. (1995a,b,c). The differences in the mortality functions between redshank age classes are as predicted on differences between the age classes in competitive ability (Goss-Custard et al. 1995a,b,c) and the present study provides broad empirical support for the approach of Goss-Custard et al. (1995a,b,c). The detailed observations of intake rates, food supplies and interference functions required to test fully the model predictions have not been made in the present study system, however, but it would be extremely valuable to do so. Simple measures such as the disposition to feed in predation-risky locations may provide simple surrogates of mortality in developing predictive population models from individual behaviour in species vulnerable to sparrowhawk predation.
The present study system bears a striking resemblance to that described by Cresswell (1994), whereby competition between juveniles and adults on a predation-safe, low-food area forces juveniles to feed in a predation-risky, high-food area. The present study would predict that redshank mortality through sparrowhawk predation on the study area of Cresswell (1994) is density-dependent and that the density-dependent mortality functions differ between age classes.
In recent decades the recovery of many European raptor populations and a restriction of their distribution into smaller areas due to changing agricultural practices or reduced numbers of prey shared by hunters and raptors (e.g. Newton & Haas 1984; Villafuerte, Viñuela & Blanco 1997) has heightened conflicts between game shooting and raptors. Hunters, conservation agencies, other stakeholders and the economic value of shooting gamebirds have driven an increasing interest and research into the effects of raptors on their prey. Most of this research has concentrated on identifying raptor responses to changes in prey density: the functional response describes how predation rates of individual predators respond to prey density, whereas the numerical response describes changes in predator density (Holling 1959; Sonerud 1992). This approach undoubtedly increases our understanding of predator–prey interactions and is very valuable in providing insights in to raptor–game conflicts (e.g. Redpath 1992; Redpath & Thirgood 1997, 1999; Nielsen 1999; Salamolard et al. 2000). However, it tends to ignore prey behaviour and any role this may play in influencing predatory responses and so may limit a full understanding of differences between predator–prey systems. The present study of sparrowhawks and redshanks suggests that density-dependent mortality of the prey came about principally through a density-mediated change in the behaviour of the prey. A lack of individually marked sparrowhawks made it difficult to know how hawks responded to this change, but there was little evidence for any substantial annual change in number of sparrowhawks on the site and hawk attack success was greater in winters with higher redshank density (D.P. Whitfield unpublished data). This suggests that a numerical response by hawks was probably relatively unimportant, and that hawks mainly responded functionally to the increased availability of vulnerable redshanks. It seems very unlikely that redshank mortality through hawk predation was compensatory. The area that redshanks preferred not to feed in enabled high food intake rates (D.P. Whitfield, unpublished data) but was where redshanks were most likely to be attacked and killed by hawks (Whitfield, in press). In the absence of sparrowhawks, redshanks would probably not have avoided an area that was clearly capable of providing food for many redshanks through a winter (see also Sih 1992).
Sparrowhawks have not been shown previously to be able to regulate the populations of their prey (Newton 1986). Previous studies of sparrowhawk predation involving songbird prey have concluded that, unlike the present study, prey was not regulated through predation by sparrowhawks and mortality was probably compensatory and not additive (Newton 1986; Newton, Dale & Rothery 1997; Thomson et al. 1998). Sinclair & Pech (1996) have argued that a mortality factor is probably seldom completely additive or exactly compensatory, but it is difficult to be sure as to why hawk predation should limit the number of wintering redshanks potentially entering the breeding population but not similarly limit small songbirds. The differences in life history traits of small songbirds and shorebirds may play a role.
It is most unlikely that sparrowhawks are agents of density-dependent mortality in all species of their wintering shorebird prey or at all sites where redshanks overwinter. The available data suggest that the influence of sparrowhawks on redshanks does vary between sites (Whitfield et al. 1988; D.P. Whitfield unpublished data) and this should be expected given the differences in mortality risk within sites (Cresswell 1994; Whitfield 1985b, in press). Conditions such as those found on sites where sparrowhawks are influential are probably widespread, however, and the effects of hawks on redshank numbers in late winter should be apparent over a large scale. Major historical changes in the presence of sparrowhawks in parts of Britain through the effects of pesticides (Newton & Haas 1984; Newton 1986) have been used to examine hawk impacts on songbird prey populations (Newton 1986). Similar comparisons of long-term late-winter counts of redshank numbers and changes in sparrowhawk presence could be illuminating. No trend in the early winter numbers of the redshank population was apparent from the present study. There may be no expectation of an adverse effect of hawk predation on the population at this stage of the annual cycle because redshanks ‘escaped’ hawk predation through migration to northerly breeding grounds. If redshank breeding success is negatively density-dependent (Goss-Custard et al. 1995b), hawk-induced losses in breeding numbers may be offset by gains in the subsequent autumn and early winter. Indeed, depending on the form of a negatively density-dependent productivity function, hawk predation may lead to higher redshank numbers in autumn.
The sparrowhawk is not the only predator of wintering redshanks in Britain (Whitfield 1985b; Whitfield et al. 1988; Cresswell & Whitfield 1994), and the peregrine Falco peregrinus can also take significant numbers during a winter. The conditions that favour sparrowhawk hunting success may be more restrictive than those that favour peregrine hunting success (Whitfield 1985b; Whitfield et al. 1988; Cresswell 1996), and so the influence of predation by peregrines on wintering shorebird dynamics is worth further investigation. Peregrines are less likely to take juvenile redshanks than sparrowhawks (Whitfield 1985b; Cresswell & Whitfield 1994), which may indicate that redshank mortality through peregrine predation is less likely to be the result of individual differences in redshank competitive ability. Peregrines show a greater propensity to chase prey (Cresswell 1996; D.P. Whitfield unpublished data) which may mean that ‘substandard’ individuals are more likely to be killed (Kenward 1978; Temple 1987; Whitfield et al. 1999). Taken together, these observations may infer that mortality through peregrine predation may be less likely to be regulatory. On the other hand, the influence of raptor predation on wintering shorebird mortality functions may depend on the strength of spatial gradients in shorebird vulnerability to predation. The scale over which peregrine impacts should be considered may need to be larger than that for sparrowhawks (Ray & Hastings 1996), reflecting differences in the two raptors’ winter ranging behaviour (Marquiss & Newton 1982; Ratcliffe 1993).
In light of the analyses that failed to find any consistent correlations between weather variables and juvenile monthly mortality across winters, it was surprising that wind speed contributed to the influence of density on juvenile mortality. Hence, the positive association between wind speed and winter mortality of juveniles may have been a Type I error. Due to the results of Hilton et al. (1999), it was expected that wind speed would be negatively related to mortality, in contrast to the finding that implied that mortality was greater in winters with higher mean daily wind speed. Durell et al. (2001) found that strong winds were positively associated with mortality in wintering oystercatchers. Higher wind speeds can lead to higher thermoregulatory energy requirement (Wiersma & Piersma 1994) and physical difficulty for foraging long-legged species like the redshank (Dugan et al. 1981). This could explain the finding of the present study as greater energetic requirements should lead to more risky foraging behaviour to meet the increased demand for energy, and thereby result in greater mortality from predation. Hilton et al. (1999), however, found that redshanks foraged further from cover in strong winds, implying a perception of enhanced predation risk that outweighed any energetic considerations. Hence, under these conditions wind speed should be negatively associated with predation mortality. It is possible that high wind speeds interfered with the foraging efficiency of redshanks on the present study site more than on the salt marsh study site of Hilton et al. (1999). Much of the foraging of redshanks on the present study site was visual and took place in large shallow rock pools, and high wind speeds may have caused greater rippling of the surface of pools, lowering rates of prey detection. The contrasting results from the two studies perhaps caution against assuming that weather will have the same influence across all sites.
The absence of any role of temperature in redshank mortality was unexpected. Low temperatures can increase the energetic requirements of both prey (Wiersma & Piersma 1994) and predator (Masman, Daan & Beldhuis 1988), sparrowhawks are more successful at catching redshanks at low temperatures (Hilton et al. 1999) and seasonal patterns of redshank mortality and redshank prey abundance are, respectively, negatively and positively associated with temperature (Whitfield 1985a; D.P. Whitfield unpublished data). The rocky shore habitat of the study site may be less prone to the large temperature-mediated differences in redshank prey availability that can occur on estuaries because most prey live in a substrate that rarely freezes. In the severe weather of the 1982 winter when estuarine redshanks on the Firth of Forth were struggling to meet energetic demands and several died, redshanks on the study site showed no signs of difficulty (Clark 1982). Thus, while temperature may affect redshank mortality on the study site within a winter (Whitfield 1985b; Whitfield et al. 1988), the rocky shore and redshank mortality may be comparatively inured against annual variations in temperature within the range observed.
This study was partially carried out while I was in receipt of awards from SERC and NERC at Edinburgh University. I am grateful to Mike Shewry for statistical advice, the British Atmospheric Data Centre and the Meteorological Office for permission to use weather data and to the Dale family for allowing access to the study site. Comments by three anonymous referees and Ken Norris improved an earlier version of the manuscript.