Predator hunting behaviour and prey vulnerability

Authors

  • J. L. Quinn,

    Corresponding author
    1. Edward Grey Institute, Department of Zoology, University of Oxford, South Parks Road, Oxford, OX1 3PS, UK
    Search for more papers by this author
  • Will Cresswell

    1. Edward Grey Institute, Department of Zoology, University of Oxford, South Parks Road, Oxford, OX1 3PS, UK
    Search for more papers by this author
    • Present address: School of Biology, University of St Andrews, Bute Building, St Andrews, KY16 9TS, UK


J.L. Quinn. E-mail: john.quinn@zoo.ox.ac.uk.

Summary

  • 1Game theoretic models of how animals manage predation risk have begun to describe predator responses to prey behaviour relatively recently. This is partly because our understanding of how terrestrial predators select vertebrate prey is often limited to numerical and functional responses to measures of prey abundance. Prey vulnerability, however, may improve our understanding of predation because predators could maximize foraging success by selecting prey on this basis.
  • 2We tested the hypothesis that sparrowhawks (Accipiter nisus L.), a typical generalist predator, hunt redshanks (Tringa totanus L.), a favoured prey species on coastal shores, primarily on the basis of their vulnerability rather than their abundance.
  • 3Five direct measures or indicators of redshank behaviour predicted sparrowhawk attack success in a multipredictor statistical model and therefore serve as measures of redshank vulnerability.
  • 4These and other vulnerability measures influenced whether sparrowhawks decided to hunt redshanks on saltmarsh habitat. A model that included most of these measures predicted correctly whether sparrowhawks hunted redshanks (attack decision) 90% of the time and accounted for up to 75% of variation. Prey abundance accounted for no additional variation.
  • 5Thus the hunting behaviour of some predators can only be predicted well by several highly dynamic and interacting factors related to prey vulnerability. These results mean that, theoretically at least, the management of prey populations may sometimes be achieved best by manipulating prey vulnerability, rather than by culling their predators.

Introduction

Predation is one of the key forces driving animal evolution (Darwin 1871; Ricklefs 1969; Dawkins & Krebs 1979). How animals manage predation risk continues to be one of the most studied topics in behavioural ecology. From an applied perspective, predation often gives rise to controversy when it impinges on human economic interests (Newton 1998; Ormerod 2002). In both theory and application, the focus has been largely limited to the prey and predation events, rather than on the behaviour of the predators or on their response to that of their prey. For example, most predator–prey models assume that predation rates are constant over time within habitats and that predators do not move between patches (Lima 2002). Because this assumption implies lack of adaptation to changing foraging conditions, it suggests that predators do not conform to one of the strongest tenets in behavioural ecology − that animals forage optimally (Stephens & Krebs 1986; Houston & McNamara 1999).

There are few empirical data to confirm that vertebrate predators of other vertebrates forage optimally or that they select the most vulnerable rather than the most available prey. Many data that are available come from studies of invertebrates or fish (Ebert 1991; Christensen & Persson 1993; Kohda 1994; Krause & Godin 1995; O’Keefe, Brewer & Dodson 1998; Snyder & Peterson 1999), many of which are laboratory-based and may reflect behaviour in wild populations only poorly (Lima 1998). For terrestrial predators there are fewer studies (Scheel 1993), although many studies make inferences about prey choice rules from the ratio of prey killed to its abundance (Götmark & Post 1996; Tornberg 1997) or to factors that affect hunting success (Fitzgibbon 1990; Funston, Mills & Biggs 2001). This paucity of data is due largely to predators being difficult to study and predation events difficult to witness. Many studies also infer how terrestrial predators select vertebrate prey based on simple models of how predation rates of individual predators change (a functional response) or how numbers of predators change (a numerical response) in response to prey abundance or density (Solomon 1949; Redpath & Thirgood 1999; Whitfield 2003), but can say little of the underlying mechanisms, making it difficult to predict predation mortality accurately. Adaptive variation in foraging behaviour should theoretically provide a better mechanism for understanding population regulation than do simple numerical responses (McNamara & Houston 1987; Anholt & Werner 1995b). This study therefore assesses the extent to which prey vulnerability, rather than abundance, determines prey selection and predation mortality by a terrestrial vertebrate predator.

A predator can theoretically assess potential prey vulnerability on the basis of many measures, thereby minimizing variability in hunting success. Prey vulnerability should be highest when animals are under energetic stress (Lima 1998) because they are forced to allocate fewer resources to antipredation behaviour (Caraco, Martindale & Pulliam 1980; Fitzgibbon 1989; Bachman 1993). Energetic stress in temperate regions increases during mid-winter when the number of daylight hours in which to find food is limited (Goss-Custard 1969), when temperatures are low (Elgar 1986; Pravosudov & Grubb 1998; Boysen, Lima & Bakken 2001), and during strong winds because of higher thermodynamic costs (Wiersma & Piersma 1994). Prey vulnerability may increase under energetic stress for reasons other than reduced vigilance. Animals may be forced to feed in areas good for foraging but with an inherently high predation risk (Lima & Dill 1990; Lima 1998; Hilton, Ruxton & Cresswell 1999b), or to feed in smaller groups in order to limit interference competition (Caraco 1979), but may then be at greater risk of being killed through slower predator detection (Elgar 1989; Cresswell 1994b; Beauchamp & Livoreil 1997). To further avoid interference competition, animals can also change their spacing within flocks (Charnov, Orians & Hyatt 1976; Goss-Custard & le V. dit Durell 1987; Cresswell 1997), but again this may increase vulnerability through reduced detectability of alarm signals and slower response times (Hilton, Cresswell & Ruxton 1999a). Variation in vulnerability can also be caused by social interactions between species. For example, subdominant species within mixed flocks may be forced to forage in risky areas (Ekman 1986, 1987).

In this study, we examine how the vulnerability of a potential prey species influences predator attack behaviour using a well-studied system of sparrowhawks attacking redshanks on a Scottish estuary. Previous studies with this system have shown that the normally difficult-to-witness behaviours of attacking and killing can be seen frequently (Cresswell & Whitfield 1994; Whitfield et al. 1999). Redshanks are also particularly vulnerable to starving in cold weather compared to all other wintering waders on European estuaries (Davidson & Evans 1982). In our study system redshanks respond to energetic stress by taking risks − they feed in areas where sparrowhawks are more likely to attack, in smaller flocks and closer to predator-concealing cover, all of which increases intake rate or shelter but only at the expense of increased vulnerability to predation (Cresswell 1994a; Cresswell 1994b). They also reduce vigilance rates and feed further apart, which probably also increases their vulnerability to predation (Cresswell 1994b; Hilton et al. 1999a).

Here we test the general hypothesis that sparrowhawks decide to hunt prey (their attack decision) according to the prey's vulnerability, rather than their abundance. First, the relationship between redshank behaviour and attack success in sparrowhawks is explored, hence identifying behaviours that indicate vulnerability. Second, temporal changes in attack decisions are described and the assumption that this correlates directly with redshank predation risk and hence mortality is tested. Third, temporal changes in redshank vulnerability are described. Finally whether sparrowhawk attack decision was predicted best by these measures of prey vulnerability or to other measures such as prey abundance was tested. If hunting decisions are dependent on the vulnerability of their prey, then we predicted that attack probability should increase when prey vulnerability increases.

Methods

general approach

The study area consisted of saltmarsh habitat backed by woodland or dunes at Tyninghame Estuary, East Lothian, Scotland (see Whitfield 1985 for study site details). Data were collected during three consecutive winter seasons, from October to early March (1989–90, 1990–91 and 1991–2) and during a fourth season 10 years later (2001–2). In all seasons, data were recorded during observation periods once or twice a day throughout each season on the saltmarsh from fixed locations.

attack success and vulnerability

All sparrowhawk attacks and kills were recorded during these observation periods. An attack was defined as a rapid flight directed towards a flock or a single bird. A ‘kill’ was defined as when the hawk captured a redshank, irrespective of whether the redshank was eaten by the hawk. Although rare (< 5% cases), redshanks sometimes escaped after being caught through the intervention of mobbing carrion crows Corvus corone (Cresswell & Whitfield 1994).

To determine whether sparrowhawk attack success was influenced by redshank behaviour, when possible flock size (FS) and distance from predator-concealing cover (DFC) were estimated for all flocks that were attacked during the four seasons. A flock was defined as a cluster of birds in which the maximum nearest neighbour distance (NND) was less than 25 m and less than one-tenth of the distance between clusters, with intercluster distance always being greater than 25 m, and varied in size from one to 200 birds. Markers were placed at regular intervals around the saltmarsh, 20 m and 50 m from the edge of the saltmarsh to facilitate estimating DFC. Additional indirect measures of redshank behaviour − mean daily temperature (T, °C), wind speed (W, m s−1) and day length (DL, h) − were also recorded. In 1989–92, weather data were obtained from the local Dunbar weather station (see below for 2001–2). In the case of temperature and day length, the starvation/predation risk trade-off predicts that redshanks should be less vulnerable when both increase because they can devote more time to being vigilant (see Cresswell 1994a). The same is true for decreasing wind speed as thermodynamic costs become lower. These indirect measures of probable vigilance rate are justified because of extreme difficulty in determining vigilance using direct behavioural measures (Lima & Bednekoff 1999).

attack decision, predation risk and redshank vulnerability

Whether sparrowhawks decided to hunt redshanks on saltmarsh (‘attack decision’) during a given observation period was recorded, from which ‘attack probabilities’ could be estimated. This binary variable was chosen as the dependent variable instead of total number of attacks during observation periods for two reasons. First, the data consisted predominantly of 0 s and 1 s, were highly skewed and could not be transformed to normality. Secondly, when multiple attacks occurred, these often occurred consecutively in a short space of time and clearly represented repeated non-independent attacks from the same hawk. Attack decision was used as the dependent variable in three separate analyses: (1) to describe seasonal trends in sparrowhawk attack decisions using data from all four seasons, (2) to relate the proportion of observation periods with sparrowhawk attacks to the number of kills found in any one month, recorded for the first three seasons (see below) and (3) to use data from the final season to test for how vulnerability measures influence sparrowhawk attack decisions.

Relatively few kills were witnessed during observation periods compared to the total number that were killed. To examine whether the binary variable, attack decision, was likely to reflect actual predation levels, the proportion of watches in any month on which attacks were seen was compared to the total number of kills found in the woods in that month. Redshanks that had been killed by sparrowhawks were recovered from the woodland from monthly searches in the first three winters of the study; almost all redshanks killed by sparrowhawks were recovered by the regular searches (Cresswell & Whitfield 1994). A high correlation between the two would confirm that attack decision was not just a direct measure of sparrowhawk behaviour, but also an indirect measure of redshank predation risk.

Attack decision was modelled on estimates of prey abundance and prey vulnerability during raptor watches in the fourth season alone. The total number of redshanks (NR) and the number of flocks (NF) were recorded on the saltmarsh every 30 min during observation periods and mean values calculated for each observation period. Flocks were defined as above and varied in size from 1 to 200 birds (mean ± SE = 28·8 ± 1·2). At the same time the following was recorded for every flock on the saltmarsh: FS, DFC and the mean NND within flocks (sampled from individuals chosen randomly throughout the flock, using bird length units (blu) that were subsequently converted to cm; 1 blu = 28 cm). For each observation period, we then derived single measures of each of these by calculating the mean of the following values: the minimum DFC, the minimum FS and the maximum NND of any flock on the saltmarsh in each 30-min sample count. To avoid cumbersome variable names and to discriminate between estimates of NND, FS and DFC from the single flocks attacked in the attack success analysis described above, these means were called FSM, DFCM and NNDM, respectively, the subscript denoting their derivation from minimum or maximum values (variable names summarized in Table 1). Wind speed (W, m s−1) and air temperature (T, °C) were logged every 15 min at a weather station in an open field 1 km south-west of the study site during the fourth season. Values were averaged over the hours of day length (DL), which itself was estimated directly using a solarimeter. This measured the amount of time for which any solar (short wave) radiation was detected. Thus our measure of day length was strongly correlated to, but slightly longer than, the time elapsed between sunrise and sunset. Time was expressed in terms of days since 1 September (DY). Duration of observation periods (D, h) was also examined. Most observation periods (80%, total N= 102) were 2–3 h in duration (mean ± SE = 2·48 ± 00·05 h). Observation periods were classified into one of three different periods of the tidal cycle: high (when mid-point of the observation period fell ± 1 h either side of high tide), falling or rising (± 3 h either side of high tide period) and low tides (all other times).

Table 1.  Eleven single predictor models of whether sparrowhawks attacked redshanks on saltmarsh (logistic regression). All variables are continuous (d.f. = 1), apart from Tide (d.f. = 2) and based on mean values during observation periods. DFCM and FSM were loge transformed; NR and NF were loge+ 1 transformed. No of observation periods varied between 84 and 103; this is mainly because some variables were redundant for given values of others, e.g. most behavioural variables when NF or NR = 0
ParameterWaldPR2Percentage predicted correctly1
Name (acronym)2EstimateNo attackAttackTotal
  • 1

    How well models predicted whether attacks occurred in each observation period.

  • 2

    The subscript M in acronyms denotes derivation from minimal or maximal values of all flocks during the observation periods (see Methods).

Number of flocks (NF)0·77  6·080·0140·1047·668·858·9
Number of redshanks (NR)0·53  8·690·0030·0046·972·959·8
Flock size (FSM) −0·26  1·590·2070·0233·377·156·7
Distance from cover, m (DFCM) −0·62  4·310·0380·0754·864·660
Nearest neighbour distance, m (NNDM)0·49  4·650·0310·0836·871·756·0
Day length, h (DL) −31·0523·90< 0·0010·3874·563·369·2
Wind speed, m s−1 (W) −0·4014·01< 0·0010·247073·571·7
Temperature, °C (T) −0·04  0·380·5390·014646·946·5
Days since 1 September (DY) −1·32  4·510·0340·066244·953·5
Tide (TD)  1·720·4200·0258·055·156·6
Duration observation period, h (D)0·45  1·290·2570·0279·234·757·8

statistics

Most statistical models were derived using logistic regression in SPSS (Norusis 1990) and the Nagelkerke statistic was used to describe the amount of variation accounted for (R2). How effective models were at correctly classifying (CC) whether attacks were made was also determined. Probabilities (P) associated with the attack decision model were obtained by the equation P= 1 ÷ [1 + (1 ÷ eY)] where Y is the predicted logit estimated from the linear predictor. All predictors involving bird counts and distances were loge-transformed and helped to increase the amount of variance accounted for by reducing the effect of extreme values.

Redshank vulnerability and abundance measures were selected for a multipredictor model using the method of variable reduction (Green 1979), i.e. if they had a significant effect in models of attack decision on their own and if they were correlated with other predictors to a level not greater than r= 0·5. All second-order interactions were included in the maximal model and the minimal model was derived by backwards elimination; predictors were retained in the final model only when P < 0·05. When displaying interactions graphically, data from continuous variables were classified into categories (for example early/late or low/high) in equal proportions based on cumulative frequency distributions.

Results

attack success and vulnerability

Five measures or indicators of redshank behaviour had an effect on whether sparrowhawks made a successful attack, either as independent effects or as interactions (Table 2). Attack success was highest when flock size was small and distance from cover short. It was also especially high when both distance from cover was short and temperatures were low to moderate (Fig. 1a). Temperature also interacted with day length showing that attacks on warm, long days when vigilance levels in redshanks should be relatively high had a very low success rate (Fig. 1b). Attack success decreased with wind speed (Fig. 1c). In summary, all variables tested affected the attack success of sparrowhawks. Temperature, flock size, distance from cover, day length and wind speed could all be taken therefore to be indices of redshank vulnerability.

Table 2.  Factors, and hence measures of redshank vulnerability, that determined attack success of sparrowhawks on redshanks during an observation period. With the exception of days since 1 September (DY), all variables entered were retained. R2 = 0·30; attacks during observation periods were predicted correctly as follows: 87·6% (all attacks), 98·3% (unsuccessful attacks, n= 300), 29·1% (successful attacks, n= 55). d.f. = 1 for all parameters except d.f. = 3 for season. DFC and FS were loge-transformed
Name (acronymn)Parameter
EstimateWaldP
  1. Non-significant terms: DY, DFC × S, FS × S, DFC × FS, S × T, S × W, FS × W, DY × DL, DY × T, S × DL, FS × DL, DFC × W, DFC × DL, DY × W, DY × FS, W × T, FS × T, DY × DFC, DY × S, W × DL.

Distance from cover (DFC) −2·2020·00< 0·001
Flock size (FS) −0·6215·72< 0·001
Wind speed, m s−1 (W) −0·04  4·330·038
Day length, proportion daylight hours (DL)0·60  2·030·154
Temperature, °C (T)0·36  0·570·449
DL × T −0·13  4·730·030
DFC × T0·22  6·930·008
Season (S) 11·53  4·560·033
                 21·82  8·450·004
               31·42  6·880·009
               4  9·140·027
Constant1·184  0·140·705
Figure 1.

Factors predicting sparrowhawk attack success when hunting redshanks over the four seasons. Attack success was particularly high: (a) when temperature and distance from cover were both low (DFC × T), (b) on short, cold days (DL × T) and (c) when wind speeds (W) were low to moderate. For the purpose of illustration, observations are approximately equally distributed between subclasses of each individual variable. Number of attacks indicated in each bar. Statistics and parameter details given in Table 2.

attack decision and predation risk

Even though the previous analysis showed that there was no temporal variation in attack success (see non-significant terms in Table 2), there was considerable temporal variation in sparrowhawk hunting behaviour (Fig. 2). Sparrowhawks were most likely to attack redshanks on saltmarsh in late December during the last three seasons, 3 weeks later during the first season, and the annual peak attack probability varied from c. 0·6 to c. 0·9 (Fig. 2). Attack probability was relatively high very early (mid-October) in the second and third seasons, thereafter increasing gradually. In contrast, early in the remaining two seasons, attack probability was relatively low but thereafter increased sharply. From late January onwards in all four seasons, attack probability fell sharply and was close to zero by late February.

Figure 2.

Temporal trends in whether sparrowhawks decided to attack redshanks on saltmarsh habitats. Only fitted attack probabilities are shown for clarity. Logistic regression model: DY, Wald (W) = 3·01, P= 0·083; DY2, W = 0·08, P= 0·78; DY3, W = 8·09, P= 0·004; S, W = 8·07, P= 0·045; S × DY, W = 8·14, P= 0·043; S × DY2, W = 8·82, P= 0·032); 1 d.f. for all, except 3 d.f. for S (season) and both interaction terms; (R2 = 0·30). The model predicted correctly whether attacks occurred in each observation period as follows: 71% (all periods), 78% (periods with attacks), 64% (no attacks). All acronyms as in Table 1. Numbers of observation periods were 60, 86, 127 and 115 for seasons 1–4, respectively.

The number of redshank kills found during the first three seasons was also predicted strongly by the proportion of days in a month on which attacks occurred (Fig. 3). Thus probabilities derived from attack decision models provided a good measure of predation risk and mortality from predation throughout the winter generally.

Figure 3.

The proportion of watches in which attacks were seen per month strongly predicted redshank kill rate (F1,20 = 21·1, P < 0·001; adjusted R2 = 0·49; season dropped out of the model at F2,18 = 0·5, P= 0·64). Overall regression line is redshanks killed per month = 4·2 + 30·2 × proportion watches with attacks.

changes in redshank vulnerability

Trends in several vulnerability measures suggested that redshanks were likely to be under energetic stress during mid-winter. Apart from the predetermined trend in day length (DL, Fig. 4a), in mid-winter redshank usage of the high predation-risk saltmarsh area peaked (NF and NR, Fig. 4b,c) and redshanks fed closer to cover (DFCM, Fig. 4d), although there was no evidence that flock size changed (FSM, Fig. 4e). In late winter, individuals became spaced further apart within flocks (NNDMFig. 4f). Wind speeds were also lowest in mid-winter (W, Fig. 4g), while temperature fluctuated widely through the season (Fig. 4h). The vulnerability of redshanks was described, therefore, by several underlying seasonal trends around which there was considerable day-to-day variation.

Figure 4.

Temporal trends in potential vulnerability measures during season 4. Underlying trends indicated by polynomial equations together with adjusted R2 values for (a) day length, (b) number of birds, (c) number of flocks, (d) distance from cover, (e) flock size, (f) nearest neighbour distance, (g) wind speed and (h) temperature.

modelling attack decision

Seven of the 11 single predictor models tested had significant effects on attack decisions made by sparrowhawks (Table 1). Sparrowhawks were more likely to attack when redshanks were more abundant (number of flocks and total numbers), fed closer to cover, were spaced further apart in flocks, on short days and earlier in the season. Wind speed had a negative effect on attack probability. Flock size, temperature, observation period duration and tide had no significant effects.

Temperature, tidal state and duration of watches were excluded from all analysis because of their high individual P-values (P = 0·54, 0·42 and 0·29, respectively: Table 1). Exclusion of observation period duration could be justified further because it did not vary systematically over time (quadratic model of days since 1 September, DY, F2,99 = 0·27, P= 0·76, R2 = −0·02; 1st-order model of DY, F1,100 = 0·50, P= 0·48, R2 = −0·01). Number of flocks (NF) was correlated highly with three variables and was excluded from all further analyses (Table 3). Spacing within flocks (NNDM) and flock size (FSM) were also correlated. Of the two, only the former had a significant individual effect. Flock size was nevertheless retained because it was strongly related to hunting success (Table 2) and was therefore likely to be a direct measure of redshank vulnerability. Other variables were at most intercorrelated only weakly and were included in the multipredictor attack decision model.

Table 3.  Correlations (r) between explanatory variables considered for modelling attack decision in season 4 (see Table 1 for acronyms). N varied between 84 and 103 observation periods. Values of r > 0·50 are in bold
DLDYFSMDFCMNNDMNRWTNF  
DLr      1·000        
P        
DYr      0·252      1·000       
P      0·011       
FSMr      0·063 −0·001      1·000      
P      0·555      0·989      
DFCMr      0·344      0·239      0·301      1·000     
P      0·001      0·023      0·004     
NNDMr      0·036      0·020 −0·518 −0·241      1·000    
P      0·745      0·855      0·000      0·027    
NRr0·269 −0·259      0·361 −0·399      0·147      1·000   
P      0·008      0·010      0·000      0·000      0·183   
Wr      0·522      0·206      0·201      0·328      0·015 −0·173      1·000  
P      0·000      0·041      0·058      0·002      0·890      0·091  
Tr      0·192 −0·219      0·028      0·254      0·111 −0·145      0·405      1·000 
P      0·054      0·030      0·795      0·016      0·317      0·157      0·000 
NFr −0·313 −0·219 −0·249 −0·584      0·503      0·733 −0·224 −0·0141·000
P      0·003      0·039      0·018      0·000      0·000      0·000      0·034      0·894

The minimum model classified correctly 89% of all watches and accounted for 70% of all variation in attack probability (Table 4). Of the seven variables entered originally, only number of redshanks was not retained in the final model, either as an independent effect or in an interaction with another predictor. Attack probability was particularly high when both distance to cover and flock size were small (DFCM × FSM, Table 4, Fig. 5a), and when both wind speed was low and individuals were widely spaced within flocks (NNDM × W, Table 4, Fig. 5b). Attack probability was higher on short than on long days and the effect was greatest when spacing within flocks was relatively high (NNDM × DL, Table 4, Fig. 5c). Finally, attack probability was highest on relatively long days early compared to late in the season (T × DL, Fig. 5d). Sparrowhawks were therefore most likely to hunt on the saltmarsh when, on average, flocks were small and close to cover, when individuals in flocks were widely spaced, when there was little wind and on short days.

Table 4.  Predicting whether sparrowhawks attacked redshanks during observation periods in season 4. R2 = 0·70; attacks during observation periods were predicted correctly as follows: 85% (all periods), 89% (periods with attacks), 79% (no attacks). d.f. = 1 for all parameters. Main results displayed graphically in Fig. 5. Data for all variables available for 84 observation periods
ParameterWaldP
NameEstimate
  1. Non-significant terms: NR, NNDM × DY, W × DFCM, FSM × DL, DY × W, DFCM × DY, FSM × W, DL × W, NNDM × FSM, FSM × DY, DFCM × DL, NNDM × DFCM, NR × DL, NR × W, NR × DFCM, NR × DY, NR × NNDM, NR × FSM.

Days since 1 Sept (DY)52·197·050·008
Day length (DL)138·894·870·027
Wind speed, m s−1 (W) −0·312·200·138
Nearest neighbour distance, m (NNDM)15·925·570·018
Flock size (FSM) −8·677·940·005
Distance from cover, m (DFCM) −4·244·760·029
DY × DL −167·877·120·008
NNDM × DL −62·375·910·015
NNDM × W1·106·930·008
FSM × DFCM2·648·080·004
Constant −28·512·220·137
Figure 5.

Sparrowhawks were especially likely to attack when redshank(s) (a) flock sizes and distance from predator concealing cover were small (FSM × DFCM), (b) were spaced widely within flocks on wind-free days (NNDM × W), (c) within flocks were spaced widely during short days (NNDM × DLM) and (d) on long days early compared to late in the season (DLM × T). Statistics and parameter details given in Table 4.

Discussion

prey vulnerability

Empirical field evidence in support of the prediction that reduced vigilance leads directly to a greater risk of being killed is scarce (Fitzgibbon 1989; Fitzgibbon 1990; Krause & Godin 1996). Although the measures of vigilance in this paper are indirect, their links to vigilance are well established. Furthermore, because vigilance estimates derived from body posture alone do not necessarily reflect accurately the ability of an individual to detect an approaching predator (Lima & Bednekoff 1999), indirect estimates may actually indicate vigilance rates just as, if not more, effectively. We are confident, therefore, that the detected effects of these variables on attack success reflect variation in vigilance.

Previous work with this system identified that temperature, flock size and distance from cover were all related to vulnerability, either through a link with vigilance or the amount of time available in which to detect predators (Cresswell 1994a; Hilton et al. 1999b). In addition to confirming the independence of these effects, two further factors were identified in this paper. Sparrowhawks were more successful when attacking redshanks that were constrained by the number of hours of daylight available for foraging. They were also more likely to be successful when there was little wind. Aerodynamic theory provides a compelling explanation for this: a result in the opposite direction to that predicted by a starvation/predation risk trade-off. We argue that sparrowhawks are less likely to catch prey in open spaces on windy days because turbulence makes it difficult to control flight, especially when landing (Stinton 2002). Paradoxically, wind is also most likely to affect flight control adversely in inherently stable species with long tails, as in the sparrowhawk, than it is in unstable species with short tails, for example the redshank (Thomas & Taylor 2001). Precise flight control is essential for predators when targeting prey but is probably less so for escaping prey that normally have multiple escape paths from which to choose.

predator hunting behaviour

When faced with a choice of hunting their more typical, generally smaller prey in other terrestrial habitats, or of hunting relatively large redshanks on saltmarsh, our results suggest that sparrowhawks first assessed the vulnerability of redshanks. The effect of prey availability was weak and over-ridden by that of prey vulnerability. The increase in attack probability, therefore, was not due to a simple functional or numerical response by sparrowhawks to changes in redshank abundance (Andersson & Erlinge 1977) but to changes in prey vulnerability. While changes in the abundance or availability of prey may well be related to changes in their vulnerability, they are not necessarily equivalent.

Optimal diet theory is poor at predicting behaviour of predators that hunt mobile prey. One of the suggested explanations is that variations in vulnerability are often more important than variations in prey choice (Sih & Christensen 2001) and our results show this could be true. That sparrowhawks made an optimal decision when they decided to hunt redshanks is supported strongly by the finding that, even though attack probability and intensity varied greatly with time and in relation to vulnerability or resource depression (Charnov et al. 1976), hunting success did not. This supports current theory, which predicts that a predator should only decide to hunt prey when the probability of capture is above a given threshold (Brown et al. 1999).

Against predictions from (1) a trade-off between vigilance and foraging and (2) the preceding attack success analysis, temperature apparently had no effect on attack decisions. During another winter season at Tyninghame, attacks were more likely on cold days (Hilton et al. 1999b) and it is assumed therefore that the effect varies annually. Supporting this, an independent study in a similar system found conflicting evidence for an effect of weather on redshank mortality (Whitfield 2003). Two additional, unexpected effects were identified. The first was that attack probability decreased with wind speed for reasons already discussed above, an effect that has been observed for another avian predator (Willem 2001). The second was that there was a general decline in attack probability over the season − sparrowhawks were more likely to attack redshanks on long days early, than late, in the season. Furthermore, the sudden and rapid drop-off in attack probability from late January in all four seasons is consistent with redshanks suddenly becoming unprofitable prey for sparrowhawks. Because the model controlled for all other variables, some other vulnerability trait must have decreased over time. The most likely such traits are the age structure of flocks on the saltmarsh (Cresswell & Whitfield 1994), some morphological factor that makes an individual more vulnerable to being caught (Janzen, Tucker & Paukstis 2000) or experience with predators (Mech et al. 2001). If sparrowhawks select on the basis of an intrinsic trait or traits, the observed decline in attack rate could be explained by a decrease in the frequencies of these traits within the population, a pattern that has been described previously at Tyninghame in relation to the proportion of young birds (Cresswell 1994a).

other effects

Most of the statistical models in our analysis were better at predicting when sparrowhawks attacked redshanks than when they did not, suggesting that additional factors were also important. One probable such factor is the availability of alternative prey in adjacent habitats. If the availability of the most common prey is high and sparrowhawks are successful hunting them, then redshanks are unlikely to be attacked, irrespective of their own vulnerability.

The effect of day length on attack probability could arguably have reflected the energetic requirements of sparrowhawks, rather than those of redshanks. Redshanks are substantially larger than the prey most frequently taken by sparrowhawks (e.g. chaffinch (Fringilla coelebs L.) and blackbird (Turdsus merula L.), Newton 1986) and it may be more profitable to select them when their own energetic needs are high in mid-winter. However, ringing recoveries suggest that sparrowhawks are most vulnerable to starvation in March–April when prey availability is lowest (Newton 1986). Similarly, attack success was also higher on short days, thus providing a more direct explanation for a greater propensity to attack on such days. Thus we believe that the effect of day length on attack probability was due primarily to the energetics of redshanks themselves.

implications for management

Far from hunting redshanks simply according to their availability, sparrowhawks were clearly constrained by redshank behaviour, probably additionally by their own and also by the nature of the habitat in which they hunted. Because many vertebrate predators face these constraints generally, as suggested by optimal foraging theory (Stephens & Krebs 1986) and a growing body of empirical evidence, this supports the idea of an ‘ecological’ approach to the control of predation (Jimenez & Conover 2001). Theoretically, predation can be limited by manipulating, in particular, (1) energetic constraints of the predators or prey themselves, (2) habitat features that facilitate predator hunting success and (3) the intrinsic vulnerability of prey.

Regarding the first of these, attempts so far to provide diversionary feeding to the predators themselves have had mixed results (Redpath, Thirgood & Leckie 2001). Even though diversionary feeding caused hen harriers (Circus cyaneus L.) to take fewer red grouse (Lagopus lagopus L.) there was no net effect on the grouse chick survival, possibly because other predators made up the difference. Our results suggest, however, that in some predator–prey conflicts, supplementary feeding might be best directed towards the prey themselves because this should allow them invest more in resources that help to limit predation from all, not just a single, predator species. We found reference to only one study that suggested that vertebrate prey could be provided with supplementary food to limit the effects of predation in harsh winters (Lamoureux, Crete & Belanger 2001), despite there being experimental evidence to support the theoretical claim that the approach should be feasible (McNamara & Houston 1987; Anholt & Werner 1995a).

Secondly, manipulating habitat structure is a well-known means of making it more difficult (Jimenez & Conover 2001; Kenward et al. 2001; Redpath et al. 2002) or easier (Palomares 2001) for predators to kill prey. In our study system, removing woodland adjacent to the saltmarsh would effectively increase distance from cover and make it harder for the hawks to kill redshanks. Finally, our data suggest that sparrowhawks may assess the intrinsic vulnerability of their prey before attacking. This principle is already used to manage predation by some pigeon keepers, who maintain lofts housing pigeons of low monetary value close to those housing pigeons of higher value in order to deflect peregrine falcon (Falco peregrinus T.) predation from the latter (Shawyer, Clarke & Dixon 1999). In conclusion, when predator hunting behaviour and success are heavily constrained, our results support the view that, rather than embarking on a culling programme, a subtler approach could be more effective in controlling predation linked mortality.

Acknowledgements

This research was carried out with funding from a Royal Society University Research Fellowship to W. Cresswell and a Leverhulme Trust Research Fellowship to W. Cresswell and J. Quinn. We thank Bobby Anderson, East Lothian District Council, the Tyninghame Estate, Mai Yasue and Sue Holt for logistical help. We would like to thank Crop-chemicals Ltd for access to weather data at Tyninghame estuary and to S. Redpath and S. Griffith for commenting on the manuscript. Comments by one anonymous referee and J.S. Brown were gratefully received.

Ancillary