- Top of page
The potential impact of predators on prey populations is an important and often controversial issue of interest not just to ecologists. For example the impact of alien predators on native fauna worldwide is currently a main conservation concern (e.g. O’Donnell 1996). Predators can influence their prey populations directly through mortality but indirect effects may also occur through the suppression of reproduction in prey species (for review see Lima & Dill 1990). Another recurrent theme in predator–prey dynamics is the ‘Erringtonian’ view that predation merely falls on a doomed surplus of individuals. Whether or not predators select individuals of a specific age-class or sex or health might affect the magnitude of their impact on the prey population (Caro & FitzGibbon 1992).
Cyclic fluctuations in populations of small mammals have stimulated a great deal of research and debate in ecology (e.g. Stenseth 1999). The specialist predator hypothesis, that mammalian specialists, in particular weasels, are responsible for driving population cycles of microtine rodents (Andersson & Erlinge 1977), has been perceived as the probable explanation for the cause of vole population cycles (Korpimäki & Krebs 1996; Hanski et al. 2001). Yet few of the studies on population cycles of voles have demographic data on the impact of predation on voles. We tested the specialist predator hypothesis experimentally by conducting a long-term, large-scale, replicated weasel removal experiment during the increase, peak and decline phases of a vole population cycle in northern England (I. M. Graham & X. Lambin unpublished data). Elsewhere, we presented the ‘numerical’ results of our weasel removal experiment, specifically the changes in the size and growth rate of vole and weasel populations. We found that reducing weasel predation pressure had no impact on the size and growth rates of experimentally manipulated field vole populations in Kielder Forest. However, as fluctuations in population size and growth rate result directly from changes in demographic variables, it is equally important to consider the mechanisms underlying the numerical patterns observed, particularly following an experimental manipulation (Krebs 1996; Yoccoz et al. 1998; Oli & Dobson 1999). This is all the more important given the necessarily limited replication and imprecise nature of large-scale experiments such as ours (May 1999; Lambin et al. 2002). Quantifying the demographic changes can (a) validate and help to explain the observed results, (b) be useful for making predictions, e.g. in predicting the response to a greater treatment effect, and (c) be used to generate estimates of parameters for simulation models.
There have been very few large-scale, predator removal experiments conducted to test the role of predation in regulating population cycles of microtine rodents and almost all have been carried out using fenced enclosures (e.g. Klemola et al. 2000). A notable exception is the predator removal experiment carried out in six unfenced areas (2–3 km2) in two years (1992 and 1995) during the crash phase of the vole cycle in western Finland by Korpimäki, Norrdahl, Klemola and colleagues (Klemola et al. 1997; Korpimäki & Norrdahl 1998). However, Korpimäki and colleagues used snap-trapping methods to monitor the small mammal populations; consequently they could say little about the demographic changes which caused an increase in vole numbers in manipulated areas only in 1995.
Most previous studies have tested the role of predation in regulating the population dynamics of small mammals in enclosures (Korpimäki & Krebs 1996). Both Erlinge (1987) and Reid, Krebs & Kenney (1995) observed increased survival rates in populations, of voles and lemmings, respectively, protected from predation, but attributed the failure of protected populations to grow to dispersal of individuals outside the fenced area where they suffered high mortality. Desy & Batzli (1989) also found increased survival in enclosed vole populations protected from predation, and found in addition that protected populations reached higher densities; however, the enclosures used were impermeable to voles such that any compensatory effect of dispersal could not be examined. Klemola et al. (2000) also used enclosures that prevented vole dispersal and similarly reported rapid population growth to high density in the enclosed vole populations, although they do not report any of the demographic variables for the enclosed populations. The difficulty with such enclosure experiments lies in knowing to what extent fences disrupt the demographic processes of enclosed populations, and to what degree the results are applicable to natural populations.
Several studies, including our own in Kielder Forest, have shown that weasels can be a major cause of mortality in voles (Jedrzejewski, Jedrzejewska & Szymura 1995; Norrdahl & Korpimäki 1995; X. Lambin et al. unpublished data). We therefore investigated the influence of removing weasels on field-vole survival in the increase, peak and decline phases of a vole population cycle in northern England. We tested the specific predictions that experimental suppression of weasel numerical response would (1) increase survival of adult female voles, and (2) increase survival of juvenile voles (and therefore enhance juvenile recruitment). To do this, we investigated the factors, both individual (sex and age) and external (treatment and weasel abundance), influencing survival in all six vole populations during the increase, peak and early decline phases of the cycle.
- Top of page
The experiment was carried out in three paired treatment and control clear-cut sites (5–12 ha) in Kielder Forest, northern England (55°13′N, 2°33′W), a large man-made spruce forest. The three treatment sites are referred to throughout as removal sites. Field-vole populations found in the grass-dominated clear-cuts exhibit three to four-year cyclic dynamics similar in many respects to those in central Fennoscandia. A detailed description of the study area is found in Lambin, Petty & MacKinnon (2000). Removal and control sites were separated by between 2 and 4 km as, in Kielder, at distances greater than 4 km vole populations tend to have asynchronous dynamics (Lambin et al. 1998); however, the pairs themselves were greater than 4 km apart.
From April 1998 onwards, weasels were live-trapped and removed from the three removal sites to suppress their numerical response. Weasels were removed from the entire clear-cut in each case. Twenty-five wooden box traps, built to the design specifications of King (1973), were set for an average of 6·5 ± 0·5 nights per month at each site. Traps were baited with previously frozen fish and checked at 24-h intervals. All weasels were ear-tagged, translocated from Kielder and released in similar habitat a minimum of 10 km from the initial capture site. No weasels were ever recaptured following translocation. A total of 80 weasels were removed from all three removal sites from May 1998 to May 2000 (Graham 2001).
In order to establish the efficiency of the weasel removal on removal sites and quantify weasel numerical response on control sites, weasel abundance was assessed at all sites using footprint tunnel tracking. Footprint tracking was carried out at five of the six sites from April 1998 to April 2000, and at the sixth from July 1998 to April 2000 excluding November 1998, according to the method described by King & Edgar (1977). For a detailed description of the methods see Graham (2002). The weasel Trap Index (TI) was derived from the number of tunnels in which weasel footprints were recorded in any given week using a calibration equation to correct for the effect of vole density on weasel density and activity (Graham 2002). In 1999, weasel abundance was reduced by an average of 57 and 60% in mid-summer and autumn, respectively, in each of the three removal sites (Graham 2001). The TI was the measure of weasel abundance used as a covariate in vole survival analyses.
Field-vole populations were trapped every month, March–October 1998, 1999 and March–May 2000, at all six sites by capture–recapture methods. Each site had a permanent live-trapping grid (0·3 ha) consisting of 100 Ugglan Special Mousetraps (Grahnab, Marieholm, Switzerland) set at 5-m intervals, which were baited with wheat and carrots. Traps were prebaited 2–3 days before each live-trapping session, then set at approximately 18:00 h on the first day. From March 1998 to March 2000 inclusive, traps were checked five times at roughly 12-h intervals, dawn and dusk. From April 2000, and at two sites in April 1998 when trapping was aborted owing to adverse weather conditions, traps were checked only three times. Voles were marked with uniquely numbered Hauptner ear-tags (Hauptner & Herberholz, Solingen, Germany) on the occasion of their first capture, either as adults or juveniles, and their mass and reproductive status noted at the time of first capture each month (primary session). During the study, vole densities ranged from 33 to 456 voles ha−1 (Graham 2001).
Field voles were categorized as ‘juvenile’ or ‘adult’ based on their fur: individuals with juvenile fur or in their first moult were classified as juveniles. Therefore ‘juveniles’ according to our classification would be less than 4 weeks old. Recruitment rates were calculated as the number of juveniles trapped in a given month per adult female in the previous month. Individuals recaptured having lost their tags were identified according to sex, mass, reproductive status, capture location or any other distinguishing feature. In most cases, this produced an entirely satisfactory match and should serve to avoid the negative bias in survival resulting from tag loss (Diefenbach & Alt 1998).
We estimated apparent survival and recapture probabilities in standard open population Cormack-Jolly Seber models (Lebreton et al. 1992) implemented in MARK (Cooch & White 1999; White & Burnham 1999). The full data set comprised the capture histories of 2576 field voles (683 females and 1020 males marked as adults and 430 females and 443 males marked as juveniles). Survival and recapture probabilities quoted in this paper refer to the probability of survival or recapture for a 28-day period: the longer over-winter intervals between trapping sessions were taken into account in the survival analysis. Our model selection strategy, as recommended by Lebreton et al. (1992), was as follows. (1) An appropriate, biologically meaningful global model was selected and its fit verified. (2) Variation in recapture rates was modelled before constraining variation in survival to increase the power of detecting variation in survival rates. (3) Parsimonious models capturing the time dependency in survival were selected on the basis of AICc (Akaike's Information Criterion) (Anderson, Burnham & Thompson 2000). (4) A priori predictions for the experiment were tested using likelihood ratio tests on selected models. The modified criterion, AICc, was used for model selection, as the number of parameters estimated by models was large relative to the sample size (Anderson et al. 2000). AICc was calculated as AIC = (−2 × log-likelihood + 2 × no. parameters) corrected for the effective sample size. Model notation is explained in Tables 2 and 3.
Table 2. Modelling the impact of weasel predation on field-vole survival for all sites in Kielder Forest. The recapture model used in the model selection procedure was pa2+pr+t (except for the global model for which pa2*s*g*t was used). Variance–covariance matrices were estimated using the 2nd Part procedure in MARK (apart from the global model). Estimates of the age class * weasel abundance interaction parameter are on a logit scale. Selected models are highlighted in bold
|Model no.||Model†||No. of parameters||Deviance||AICc||Parameter estimate for a2.mw ± SE|
| ||φa2*s*g*t||∼863||1711·2||∼8909·2|| |
|1.||φs+a2*g+a2*t+g*t|| 147||2286·8||7684·8|| |
|2.||φs*g+a2*g+a2*t+g*t|| 152||2282·3||7691·0|| |
|3.||φs*t+a2*g+a2*t+g*t|| 164||2257·1||7691·4|| |
|4.||φa2*g+a2*t+g*t|| 146||2307·0||7702·8|| |
|5.||φs+a2*mw+g*t|| 131||2340·4||7704·4||−0·211 ± 0·074|
|6.||φs+a2+g*t|| 130||2352·6||7714·5|| |
|7.||φs+a2*g+a2*(m+y)+g*(m+y)|| 94||2437·4||7723·8|| |
|8.||φs+a2*g+a2*t+pr*t|| 96||2448·9||7739·5|| |
|9.||φs+a2*g+a2*t|| 62||2524·9||7745·2|| |
|10.||φs+a2+pr*t|| 76||2517·9||7767·1|| |
|11.||φs+a2+t|| 40||2595·1||7770·4|| |
|12.||φs+a2*g|| 34||2964·0||8127·2|| |
|13.||φs*g+a2*g*t|| 185||2208·8||7688·3|| |
|Analysis of treatment (categorical)|
|14.||φs+a2*t+trt*t+pr*t|| 107||2418·8||7732·3|| |
|15.||φs+a2*t+a2*trt+trt*t+pr*t|| 108||2417·3||7732·9|| |
|16.||φs+a2*t+trt+pr*t|| 89||2475·0||7751·1|| |
|17.||φs+a2*t+pr*t|| 88||2476·8||7750·8|| |
|Analysis of treatment (difference in weasel abundance)|
|18.||φs+a2*dw+pr*t|| 78||2507·7||7760·9|| |
|19.||φs*dw+a2*dw+pr*t|| 79||2506·0||7761·3|| |
|20.||φs*a2*dw+pr*t|| 81||2504·5||7763·9|| |
|21.||φs+a2+dw+pr*t|| 77||2517·4||7768·6|| |
|Effect of mean weasel abundance|
|22.||φs+a2*mw+pr*t|| 78||2503·3||7756·6||−0·222 ± 0·072|
|23.||φs+a2*mw+m|| 32||2602·0||7761·1||−0·191 ± 0·060|
|24.||φs+a2*mw+t|| 42||2583·1||7762·5||−0·180 ± 0·062|
|25.||φs*a2*mw+m|| 35||2596·7||7761·8|| |
|26.||φs*mw+a2*mw+m|| 33||2601·6||7762·7|| |
|27.||φs+a2+mw+m|| 31||2614·9||7772·0|| |
Table 3. The impact of changes in adult and juvenile field-vole survival rates on monthly population-growth rates (λ). Population-growth rates were estimated using a standard two age-class Leslie matrix model
The CMR models used assume that: (1) every marked animal in the population immediately after time (i) has the same probability of surviving to time (i + 1); and (2) every marked animal present in the population at time (i) has the same probability of recapture (pi). We therefore carried out an initial goodness-of-fit (GOF) test on the global (most fully parameterized) model, φa2*s*g*tpa2*s*g*t. (with both survival and recapture probabilities dependent on age, sex, site and time), using RELEASE (Burnham et al. 1987), in MARK. The GOF test for the whole data set, split by age at first capture, sex and site, was not significant suggesting that the fit of the model was acceptable (Test 2 and 3, RELEASE: χ2 = 248·80, d.f. = 401, P = 1·00). However, Test 2.C in particular had very few degrees of freedom because of the sparseness of the data. We therefore examined the contingency tables (pooled within site) for Tests 3.SR and 2.CT produced using RELEASE (Burnham et al. 1987), as modified by Pradel (1993) for a systematic bias, i.e. an excess of tables in which the cells on one or other diagonal exceed the expected frequencies. We performed sign tests (Sokal & Rohlf 1995) on the frequency of contingency tables with a particular bias. Using this method, Tests 2.CT and 3.SR were marginally significant for one site only, site 6 (Test 2.CT: t = 3·00, d.f. = 8, P = 0·01; Test 3.SR: t = 2·10, d.f. = 50, P = 0·05). However, given the number of sequential tests, this could be due to experimentwise error rate; moreover for this and three other sites the sample size for Test 2.CT was unsatisfactory (Sokal & Rohlf 1995). Where the number of individual contingency tables examined exceeded 12, there was no evidence of lack of fit. We therefore started in the analysis with φa2*s*g*tpa2*s*g*t as our global model. In order to simplify the model-selection procedure and identify a reduced set of realistic models, we initially modelled variation in recapture and survival with age, sex and time using the data sets for individual sites. We fitted all additive combinations of p with age, sex and time (fully time-dependent and constrained time-dependent: season and year) and all combinations of φ with age, sex and time to the data sets for individual sites. We then identified a suitable, reduced starting model for the data set for all sites. To test our predictions on the effect of the experimental treatment (removal of weasels) on survival, we fitted models including treatment coded either as: (1) a categorical variable (removal or control) or (2) an external, continuous covariate (coded as the difference between mean weasel abundance during each period in paired control and removal sites). To incorporate our experimental design when fitting these models, we included a pair * time interaction term. Finally, to investigate the influence of weasel abundance on vole survival in all sites, we used the site-specific mean weasel abundance during each period as an external, continuous covariate. The amount of variation explained by using weasel abundance as a covariate was calculated by analysis of deviance (Skalski, Hoffman & Smith 1993):
- Amount of variation explained =[deviance (constant model) – deviance (covariate model)]/[deviance (constant model) – deviance (time & group-dependent)]
- Top of page
Our experiment demonstrates that adult field-vole survival can be increased by experimentally reducing weasel-predation pressure. Weasel numerical response was suppressed in removal sites, particularly in summer and autumn 1999, yet the vole populations in all sites declined to low density and there was no consistent divergence in population trajectories between control and removal sites, in the increase, peak or decline phases of the cycle. Examining the impact of weasel predation on vole survival provides a partial explanation for this.
The models we used to investigate the impact of the treatment (weasel removal) on field-vole survival incorporated the experimental design through the inclusion of a pair * time interaction term. Although this term did not fully explain all the independent temporal variation between sites, i.e. the models had greater AIC values and were therefore less parsimonious than the best model, they were nevertheless the only appropriate models for testing a priori hypotheses on the impact of the experiment on field-vole survival. Our analysis of the impact of actual weasel abundance on field-vole survival shows that the poorer fit of models incorporating the experimental design did not influence the interpretation of the results. However, the percentage of variation explained by doing so was large. Moreover the age class * weasel abundance interaction parameter was robust to how the independent temporal variation between sites was modelled. It was therefore appropriate to estimate survival probabilities using the simpler models as there were insufficient data to estimate the large number of parameters required by the models incorporating full independent temporal variation among all six sites.
While weasel predation accounted for up to 20% of the variation in vole survival at high weasel abundance, weasel densities were rarely high enough to cause such an impact. Indeed, the seasonal dynamics of the weasel population and absence of year-on-year changes in their abundance (I. M. Graham & X. Lambin unpublished data) in Kielder Forest, mean that, apart from a few months in summer, weasel predation explained only 5% or less of the variation in field-vole survival. This was exemplified by the extensive spatio-temporal variation in survival rates in our six sites that could not be explained by weasel abundance. A number of previous studies have estimated weasel predation rates on rodents to be 1–35% (MacLean et al. 1974; King 1980; Delattre 1984; Jedrzejewski et al. 1995). Although weasel predation rates cannot simply be translated into vole survival rates, it is striking that our estimate of the impact of weasel predation on field-vole survival is of a very similar order of magnitude.
The seasonal dynamics of weasels also constrained the scope for the experimentally induced contrast between treatments. However only where treatment was coded as a categorical variable was there any assumption concerning the success of the experiment. Where actual weasel abundance was used, whether the difference between paired sites or site-specific, there was no dependence on a permanent or strong difference in weasel abundance between treatment and control sites. Moreover, our conclusion on the impact of weasel predation on vole survival held regardless of the way in which variation in weasel abundance was modelled: removal vs. control, difference in weasel abundance between paired sites or site-specific weasel abundance. Given that the results of all three different approaches concurred our conclusion is strengthened.
Although population growth was strongly correlated with adult survival in the non-breeding season over winter, the relationship between growth and survival was very much weaker during the summer. Consequently, when the effect of suppressing weasel numerical response on vole survival is greatest, increasing survival might be expected to have little influence on the population growth rate other than indirectly, through the effect of increased female survival on recruitment rate. This seems to have been the case as population growth rates showed no consistent divergence between control and removal vole populations even in summer (I. M. Graham & X. Lambin unpublished data). Our results concur with the conclusions of Oli & Dobson's (1999) modelling study. Using a demographically based model, they tested the potential influence of a number of parameters, including juvenile and adult survival, on population cycles and concluded that changes in adult survival were neither necessary nor sufficient to generate population cycles.
Increased survival rates have been observed in populations of voles and lemmings protected from predation in enclosures in several previous studies (Erlinge 1987; Desy & Batzli 1989; Reid et al. 1995; Wilson, Krebs, & Sinclair 1999); however, our experimental results are unique in two aspects. We manipulated only weasel densities, and our experimental populations were unfenced. Where the enclosures used were permeable, the failure of such protected populations to grow has often been attributed to the relatively small size of the enclosures, often too small to encompass dispersal. Consequently, as individuals, which disperse or live outside the fenced areas, are exposed to high levels of predation, emigration may increase as a result of the disparity in densities inside and outside the enclosures. Thus Reid et al. (1995) postulated that the number of potential immigrants was reduced, resulting in a bias in dispersal of individuals out of the enclosures and Erlinge (1987) reported that individuals whose home ranges extended outside the enclosure suffered high mortality rates. The surprising result of our study, that juvenile survival decreased in response to the reduction in weasel-predation pressure, suggests, however, that the lack of population growth might not have been an artefact of using enclosures. In our experiment, weasels were removed from entire clearcuts therefore vole populations in removal sites were not in direct contact with vole populations at lower densities. Nevertheless, juvenile survival decreased when weasel-predation pressure was reduced. This decrease in juvenile survival may have partially compensated for the increase in adult survival, contributing to the lack of divergence between vole population trajectories in control and removal sites, even when control sites experienced much higher levels of weasel predation. We used a simple standard age-based Leslie matrix model to explore the idea that the decrease in juvenile survival in response to reduced weasel-predation pressure did compensate for the increased adult survival. We used appropriate estimates of juvenile and adult survival suggested by model 22 and used a two age-class model. In spite of all the simplifying assumptions of this approach, it illustrates that the similar, although opposite, effects of weasel abundance on juvenile and adult survival were probably not fully compensatory (Table 3).
Suppression of the weasel numerical response increased adult vole survival in removal sites but it effectively reduced juvenile survival. As apparent survival probabilities confound dispersal and mortality, this decrease in juvenile survival could have been simply due to increased mortality or alternatively to increased emigration. In removal sites, it is possible that the higher number of adults in the population caused more juvenile mortality through infanticide. Instances of infanticide have been reported in some vole species (e.g. Boonstra 1978; Lidicker 1979; Caley & Boutin 1985) and infanticide has been shown to contribute significantly to juvenile mortality in Townsend's voles (Microtus townsendii) (Lambin & Yoccoz 1998). This same study demonstrated quantitatively that nestling survival was significantly lower in areas where the relatedness among females was lower (their low-kinship treatment), which they attributed to increased infanticide.
Higher numbers of adults may, however, have caused increased juvenile dispersal by forcing juveniles to disperse greater distances and making it difficult for juveniles to settle and recruit into the population. Julliard et al. (1999) similarly found a negative effect of multimammate rat (Mastomys natalensis) density on survival of newly marked subadults, whereas the effect of rat density on adults and previously marked subadults was positive. A possible solution to the paradox of lower juvenile survival in spite of equal densities may lie with the change in population age-structure that would result from increased adult survival. Pusenius & Viitala (1993) found that field-vole spacing behaviour in wild populations showed pronounced variation, which they attributed to differences in behaviour related to age and social dominance. They found that the old, over-wintered females became increasingly territorial as the breeding season progressed, whereas the younger females had much smaller, more extensively overlapping home ranges even when reproducing. Similarly, Agrell et al. (1995) found that female field voles were more territorial in experimentally manipulated populations previously at high density and that smaller, less competitive females suffered as a result of the increased competition. The pressure to disperse may therefore be related to the age and social structure of the population. This would explain why populations that experienced high weasel-predation pressure, although similar in density to those in which weasels were absent, appeared to have higher rates of juvenile survival. The effectively ‘younger’ population, resulting from predation, may have been more permeable to juveniles allowing compensation for the high mortality rate through increased recruitment.
We found no evidence to suggest that weasel predation was sex selective, contrary to the claims of Klemola et al. (1997). Klemola et al. (1997) found that the mean proportion of pregnant females was lower in control areas than in predator-reduction areas following experimental manipulation, based on an examination of only three or fewer snap-trapped voles, for all but one of their six control sites after the manipulation. As the sex ratio was less male-biased in predator reduction areas, they postulated that small mustelids were selectively predating pregnant female voles, hence magnifying their impact on population growth. However, they lacked the detailed capture–mark–recapture data to substantiate this claim. Our data show that the increase in survival due to reduction in weasel predation pressure was the same for males and females, not greater for females as predicted if weasels selectively predate female voles. In addition, the sex ratio of our vole populations, based on numbers live-trapped, was rarely 1 : 1 during the course of our study. It seems likely that this did not always reflect the actual sex ratio of the population and may simply have been an artefact of method of sampling. For example, if different individuals have different home-range sizes, trappability or activity, their probability of being trapped will vary. Clearly, if the probability is differential between sexes then it will appear that there is a bias in the sex ratio.
Adult field-vole survival increased in response to suppression of the weasel numerical response in removal sites. It is clear therefore that the effect of weasel removal was not simply negated through compensatory predation by other members of the predator guild in Kielder Forest. Juvenile survival, however, decreased in response to reduction in weasel predation pressure but this reduction in juvenile survival may not have fully compensated for the increase in adult survival in terms of vole population growth. Nevertheless, in spite of the increase in adult survival due to the experimental treatment there was no divergence between control and removal vole populations even during the breeding season when weasel numbers were greatest. In conclusion, the experimental reduction of weasel numbers affected field-vole survival and, although weasel predation can account for a sizeable proportion of the variation in field-vole survival, most of the variation was explained by other factors, probably as a result of the limited seasonal weasel numerical response (Graham 2001). Given that reducing weasel predation pressure had no impact on the size and growth rates of experimentally manipulated field vole populations (Graham 2001), we conclude that changes in weasel predation rate were not responsible for driving the population cycles of field voles observed in Kielder Forest.