Cyclic dynamics in field vole populations and generalist predation

Authors


Xavier Lambin, Department of Zoology, University of Aberdeen, Tillydrone Avenue, Aberdeen, AB24 2TZ, Scotland. Tel.: 01224 273259. Fax: 01224 272396. E-mail: x.lambin@abdn.ac.uk

Summary

1. A geographical gradient in the relative impact of generalist and specialist predators on small rodent populations has been hypothesized to be responsible for the gradient in cyclicity found in Fennoscandia. Population oscillations resulting from weasel–vole interactions are said to be dampened by the increasing stabilizing impact of generalist predators in southern Fennoscandia resulting from: (i) a greater abundance and diversity of predators sustained by alternative prey; (ii) the absence of significant snow cover leading to constant exposure of voles to generalist predators; and (iii) a heterogeneous habitat that makes dispersing voles more vulnerable to predators.

2. Changes in the abundance of field voles (Microtus agrestis L.) in a man-made spruce forest in northern England were recorded during 1984–98 using sign indices at 14–18 sites calibrated with capture–recapture estimates of vole density.

3. Field vole populations exhibited cyclic dynamics which were in many ways similar to those reported from Fennoscandia, including population declines taking place during the breeding season and long periods with no recovery in numbers following population crashes.

4. The density dependence structure of the time series was explored by means of partial autocorrelation functions, which suggested second-order density dependence. Analyses based on two density estimates per year (spring and autumn) reveal significant negative values for lags of 1, 1·5 and 2 years, suggesting that the time-lag might be somewhat shorter than 2 years.

5. Estimates of predation on field voles by red foxes and tawny owls at high vole density were above the value predicted for this site and for the whole generalist predator community by a published model assuming that predation by generalist predators stabilizes vole populations. However, empirical estimates of the parameter used both for designing and testing the model are inherently imprecise.

6. A qualitative evaluation of the three variables (see 1) correlated to the Fennoscandian gradient and assumed to contribute to variations in generalist predation pressure did not support the hypothesis that low predation rates by generalist predators are necessary for vole dynamics to be dominated by the destabilizing impact of weasel–vole interactions. The specialist/generalist predation hypothesis must therefore be modified to account for the regular population cycles occurring in northern Britain.

Introduction

A clear biogeographical gradient has been demonstrated in the amplitude and the degree of cyclicity in small rodent population fluctuations from north to south in Fennoscandia (Hansson & Henttonen 1985; Henttonen, McGuire & Hansson 1985; Bjørnstad, Falck & Stenseth 1995; Turchin & Hanski 1997). Microtine populations in southern Sweden showed annual fluctuations of low amplitude and more northerly populations exhibited multiannual cycles of increasing amplitude and length. In Fennoscandia, the predation hypothesis is increasingly being used to explain this phenomenon. It states that the time-delayed numerical response of weasels to their rodent prey drives population cycles and that a geographical gradient in the relative impact of generalist and specialist predators on small rodent populations is responsible for the gradient in cyclicity (Hansson 1987; Hanski, Hansson & Henttonen 1991; Hanski & Korpimäki 1995; Norrdahl 1995; Turchin & Hanski 1997). Population cycles driven by weasel–vole interactions are said to be dampened by the stabilizing impact of generalist predators in southern Fennoscandia. A corresponding gradient in the strength of statistical direct density dependence in time series from across Fennoscandia is consistent with the hypothesis (Bjørnstad et al. 1995). The impact of generalists in the south is assumed to result from: (i) a greater abundance and diversity of predators sustained by alternative prey; (ii) the absence of a significant snow cover leading to constant exposure of voles to generalist predators; and (iii) a heterogeneous habitat that makes dispersing voles more accessible to predators. Modelling has shown the predation hypothesis to be a plausible explanation of microtine cycles, and there is strong correlative evidence from Fennoscandia consistent with this hypothesis (reviews in Korpimäki & Krebs 1996; Turchin & Hanski 1997). However, there is also a high degree of statistical confounding between the above environmental covariates, and empirical estimates of parameters used in models are subjected to a large uncertainty. In the absence of formal empirical testing, the predation hypothesis would be strengthened if the covariates invoked in the Fennoscandian context were found to be associated with microtine cycles outside Fennoscandia.

The reported lack of multiannual cycles in vole populations in Britain and continental Europe comes from analyses of restricted sets of time series (Alibhai & Gipps 1985; Hansson & Henttonen 1985; Henttonen, McGuire & Hansson 1985). The fact that biotic and abiotic conditions in Britain and continental Europe are generally more similar to those found in southern than in northern Fennoscandia, has led to the widespread belief that the Fennoscandian gradient extends southward to Europe, hence supporting the predation hypothesis. Small rodent populations in Britain and continental Europe are said to have: (i) no multiannual periodicity (Hansson 1987); (ii) mostly seasonal variation in numbers (Alibhai & Gipps 1985; Norrdahl 1995); (iii) no profound density differences between years (Hansson & Henttonen 1985); and (iv) no summer declines which occur in high amplitude cycles in northern Fennoscandia (Hansson & Henttonen 1985; Henttonen et al. 1987; Hanski et al. 1991; Hanski & Korpimäki 1995). Cyclic fluctuations of small rodents in mainland Europe are thought to be restricted to agricultural areas where habitat heterogeneity is low (Hansson & Henttonen 1985; Hanski 1987).

In this paper, we analyse data on field vole population fluctuations in a man-made spruce forest in northern England. Here, the duration of snow cover, the community structure and the abundance of predators appear similar to southern Fennoscandia and much of the landscape provides little cover for voles, yet multiannual cycles occur. We document the cyclic nature of field vole fluctuations, and consider whether the population dynamics are consistent with the predicted impact of generalist predation.

Methods

Study area

Kielder Forest is situated in Northern England (55°13′N, 2°33′W) (Fig. 1). It is one of the largest man-made forests in Europe (613 km2) and consists primarily of Sitka spruce (Picea sitchensis Bong.) and Norway spruce (Picea abies L.), with smaller amounts of larch (Larix spp. Miller) and pine (Pinus spp.) managed on a 40–60 year rotation. The study area covers eight valley systems which converge on Kielder Water (a reservoir of 1000 ha). During 1979–90, mean annual rainfall was 1420 mm; mean number of days per year with snow lying at 09·00 h was 40·6 days (max. 80 d), mean daily maximum temperature was 11 °C, and mean daily minimum temperature was 3·2 °C (Petty 1992; data from meteorological station at Kielder Castle).

Figure 1.

Location of Kielder Forest in northern England (circle), adjacent to the border with Scotland.

Harvesting of timber (mean 700 ha year−1 over the last 10 years) provides well-defined ‘islands’ of successional habitat that progress from clear-cuts, to grassland, to prethicket forest and finally reach thicket stage after 12–15 years. Clear-cuts range in size from 5 ha to more than 100 ha, with the smallest sizes in the valley bottoms. Approximately 16–17% of the forest is occupied by clear-cuts suitable for field voles. The field vole is common in these ephemeral habitats but completely absent from forested areas, which lack grass cover (Petty 1992).

The predator community in Kielder Forest comprises generalists, including (in decreasing presumed abundance order) the red fox (Vulpes vulpes L.; min. estimate 4–6 foxes 1000 ha−1 in spring, O’Mahony et al. 1999), tawny owl (Strix aluco L.; ≈ 2·4 pairs 1000 ha−1 in 1997), American mink (Mustela vison Schreber), badger (Meles meles L.) and stoat (Mustela erminea L.). Vole specialists such as weasels (Mustela nivalis L.) are probably abundant (I. M. Graham, unpublished data), but avian vole specialists such as common kestrels (Falco tunniculus L.; at least 0·41 pairs 1000 ha−1 in 1985) and short-eared owls (Asio flammeus Pontoppidan; at least 0·43 pairs 1000 ha−1 in 1985) are scarcer now than in the 1970s and 1980s, although long-eared owls (A. otus L.; at least 0·25 pairs 1000 ha−1 in 1985) may have slightly increased (Petty et al. 2000).

Field methods

From 1984, one of us (S.J. Petty) indexed the abundance of field voles three times a year, in spring (March), summer (May/June) and autumn (September) on 14–18 grass-dominated clear-cuts. Before new tree growth made the original sampling sites unsuitable for voles, replacement areas were selected as near to the original sites as possible (mean distance = 567 m, max. = 1655 m). Sampling areas were chosen to be on surface water gleys and were dominated by Deschampsia cespitosa Beauv., Agrostis tenuis Sibth., Juncus effusus L. and bryophytes. The largest distance between sampling areas was 15 km. At each sampling area, 25 randomly chosen quadrats (25 × 25 cm) were searched for the presence or absence of vole signs. These included runways, fresh droppings, old droppings, fresh (completely green) heaps of grass clippings, old (not completely green) heaps of grass clippings. Of these indices, signs based on the presence of fresh grass clippings explained more of the variability in vole trapping indices (VTI, see below, Petty 1992; see also Hansson 1979), and are used here. Grass clippings were also used because they do not remain fresh for longer than 1–2 weeks and so they reflect recent vole activity, whereas other vole signs (such as old clippings or runways) may reflect the cumulative vole activity over many months. The same area was used for each assessment and a similar route was followed each time. The position of each quadrat differed between assessments and was determined by walking 15–25 paces from the previous quadrat. A square metal frame was then thrown and the assessment undertaken where it landed, unless the quadrat landed on bare ground or a wood pile, in which case it was thrown again.

Calibration of vole abundance indices

Estimates of vole abundance based on sign indices (VSIs) were compared with actual densities of field voles and with the small quadrat method based on snap-trapping (Myllymäki et al. 1971; Hansson 1975). We calibrated the VTI with the VSI at one site, three times per year from 1984 to 1990, and subsequently measured vole density by live trapping and sign indices at eight permanent grids in 1996 and 1997.

To calibrate the VTI with the VSI, we randomly located 12 15 × 15 m quadrats within the sampling site. Three snap-traps, modified to catch without bait and adjusted to be triggered at 5 g, were set at the best location, usually across a vole run, within a 1-m radius of each corner of each quadrat. We checked 144 traps at 24-h intervals for 4 days (576 trap nights). We calculated the VTI as the ratio between the number of field voles caught and the number of trap nights minus the number of traps sprung without catching voles. We subtracted the number of sprung traps to remove any influence of precipitation on abundance indices as heavy rain and snow sometimes triggered the traps. Field voles were by far the most prevalent species (82·2%) among 1139 small mammals caught in 10 658 trap nights, followed by common shrew Sorex araneus L. 14·9%, and pygmy shrew Sorex minutus L. 2·2% (Petty 1992, 1999). Bank voles Clethrionomys glareolus (Schreber) and wood mice Apodemus sylvaticus (L). each amounted to 0·3% of the total catch. VTIs are expressed as the number of voles/100 snap-trap nights.

For live trapping estimates, we estimated abundance of field voles every month at eight permanent live-trapping grids from mid-April 1996 to October 1996 and March 1997 until October 1997. Each 0·3 ha grid consisted of 100 Ugglan Mouse Special traps set at 5 m intervals and baited with wheat and carrots. Traps were prebaited and vole sign indices were measured 3 days before each live-trapping session. In April and May 1996 traps were set at approximately 18·00 h and then checked at 12-h intervals for a total of five trapping sessions. In all subsequent months, traps were set at approximately 06·00 h and checked three times at 4-h intervals throughout the day. Traps were not set overnight. We then trapped for a further three sessions the following day to give a total of six trapping sessions per month. We assumed that no mortality or recruitment occurred during the five or six sessions and estimated population size using the program capture (Otis et al. 1978). In most cases, the model selected assumed heterogeneity in capture probabilities of individuals and differences in the trapping probability between sessions. We therefore used the estimator of Chao & Lee (1991) to calculate vole abundance at all grids. To check for seasonal differences in the relationship between density and VSI we pooled monthly estimates as follows: spring (March, April, May), summer (June, July, August) and autumn (September and October).

Data analysis

For comparison with other studies, we quantified the overall amplitude of fluctuations of field vole populations using the s-index [the standard deviation of log10(N + 1)]. We used autumn surveys only as advocated by Henttonen et al. (1985) and used data converted into snap-trapping indices (see Results) to avoid problems with zeros, and to make the indices produced directly comparable with those obtained by Fennoscandian studies based upon the small quadrat method.

To characterize multiannual patterns, we used estimates averaged over all VSI survey sites (n = 14 or 18) as this minimizes any sampling error in the estimation of vole density. We used data from the spring and autumn surveys in order to achieve a sufficient sample size to test for cyclicity (27 estimates of vole density at 6-month intervals). We did not include data from summer surveys, so meeting the requirement of equal time intervals between sampling periods. Any trend in a time series that might reflect change in habitat quality as a result of tree growth or other factors was removed by fitting second- and third-order polynomials, and removing insignificant (P > 0·05) terms by a step-down procedure. Detrending removed a weak downwards trend in vole abundance. We performed autocorrelations and spectral analysis on residuals after checking that they were normally distributed. Spectral analysis was performed using proc spectra (SAS) and was tested for significance by the Fisher Kappa test which compares the ratio of the largest periodogram ordinate to the average of all ordinates, and is designed to detect one sinusoidal component embedded in ‘white noise’ (Fuller 1978). We also used autocorrelation and partial autocorrelation functions calculated using log-transformed abundance estimates or differenced time series (log Abundancet– log Abundancet–1) to derive a preliminary estimate of the density-dependent structure of the time series assuming linear density dependence. We performed autocorrelation using both single and two-yearly abundance estimates. Autocorrelations were computed using proc arima (SAS 1989).

Much of the pioneering work by Middleton (1931), Elton, Davis & Findlay (1935) and Chitty (1996) on field vole populations took place between 1931 and 1939 in Newcastelton, 17 km west of the centre of our study area. The landscape was then dominated by fields, rough grasslands formerly grazed by sheep and young spruce plantations. They used both snap-trapping and the percentage of 100–175, 2 sq. yard quadrats with fresh vole faeces as indices of field vole abundance (see Chitty 1996 for details). The original data were kindly made available to us by D. Chitty.

Results

Calibration

Vole densities within each ‘season’ were linearly related to vole sign indices (Fig. 2, Table 1). The relationship did not vary significantly between grids (ancova, test for homogeneity of slopes F7,100 = 0·65, P = 0·71; test for homogeneity of Y-intercepts F7,108 = 0·71, P = 0·67) nor between seasons (ancova, test for homogeneity of slopes F2,110 = 1·15, P = 0·32; test for homogeneity of Y-intercepts F2,113 = 1·36, P = 0·26). However, as estimates for the intercepts differed by up to 14 voles ha−1 and because the primary concern was to minimize systematic bias in the resulting density estimates between seasons (see also Hanski, Henttonen & Hansson 1994), separate relationships were used for each season. The residuals of the regressions were normally distributed (all P-values > 0·3) but the measurement error associated with predicting density at any one site is large (Fig. 2). The measurement error of vole density at the landscape scale, which is based on 14–18 sites, should on average be minimal. Snap-trap indices were also closely related to VSIs and these relationships did not differ significantly between seasons (ancova, test for homogeneity of slopes F2,11 = 1·48, P = 0·23; test for homogeneity of Y-intercepts F2,15 = 1·44, P = 0·26). In this paper we use only snap-trapping indices for calculating the s-index based on the autumn relationship.

Figure 2.

Relationship between vole sign indices and vole density estimates derived from live trapping during the summer (grey). The line depicts the linear relationship from a least squares regression. Observations for spring (white), and autumn (black) are also shown

Table 1. . Linear regressions between vole sign indices (VSI), vole snap trap indices (VTI) and live trapping estimation of density. Values in brackets are standard errors of parameter estimates
SeasonRegressionnR2FP
Snap trap index
All seasonsVTI = 1·18*VSI + 1·55190·6734·98< 0·001
Vole density
Spring (March–May)N ha−1 = 13·10(1·76)*VSI + 9·96(11·16)360·6255·12< 0·001
Summer (June–August)N ha−1 = 14·11(1·33)*VSI + 14·60(9·34)480·71111·76< 0·001
Autumn (September–October)N ha−1 = 13·48(1·52)*VSI + 23·17(12·67)320·7277·96< 0·001

Multi-annual fluctuations

Field vole populations in Kielder Forest were unstable and reached peak density in 1984, 1987, 1991 and 1994 (Fig. 3). The interval between the first two peaks in mean density was 3 years, whereas that between the two most recent peaks was 4 years. The amplitude of the fluctuation averaged over all sampling sites was approximately 10-fold with minimum and maximum mean densities of 25 voles ha−1 and 215 voles ha−1, respectively, or 2·5–18·7 voles per 100 snap-trap nights. The minimum and maximum vole densities encountered at a single site were 14 and 333 voles ha−1, respectively (the minimum values correspond to the intercept of the regression for 0 vole per trap night or no positive quadrat out of 25). S-index values calculated with sign indices converted to number of individuals per 100 snap-trap nights range between 0·25 and 0·42 (median = 0·36). The index of amplitude of the fluctuation averaged over all sampling sites was lower (s-index = 0·26) reflecting a certain degree of asynchrony between sites.

Figure 3.

Fluctuations in the abundance of field voles in Kielder Forest (averaged over 14–18 sampling sites) for vole sign indices calibrated to densities. 95% confidence intervals are shown and estimates for spring (white), summer (grey) and autumn (black) are plotted for each year.

Fluctuations of field vole populations in Kielder Forest were periodic, as revealed by spectral analysis (Table 2). The time series differs significantly from white-noise (κ = 4·5, M − 1 = 5, P < 0·01) with periodograms characterized by a well-defined peak for a period of 3·3–3·5 years. The period length is similar if the spectral analysis is performed on log density or when the 18 time series are considered separately. However, Fisher's Kappa white-noise test fails to reach formal statistical significance at P = 0·05 for three 13-year time series out of 13 and one 11-year time series out of five (Table 2). Spectral analysis also revealed a rhythm in the first-differenced time series (growth rates).

Table 2. . Spectral analysis and autocorrelation functions (ACF) of time-series of field vole populations in Kielder Forest 1984–97. The dominant frequency is estimated by spectral analyses. ** denotes Kappa test P < 0·01, *P < 0·05. T is the period of the time series (period shown only if ACF(T) > 1 SE of ACF. ACF(T) is the estimated autocorrelation at lag T. Autocorrelations and partial autocorrelations are taken to be significant if ACF > 2 SE. All time series, except those based on differenced log densities were detrended using second- and third-order polynomes. the S-index is the standard deviation of the log-transformed vole numbers and is calculated for the autumn sample only
(a) Pooled sites
SiteNumber
of observations
KappaDominant
frequency (years)
T
(years)
ACF(T)Lag of other
significant (P < 0·05)
autocorrelations (year)
Lag (years) and estimates
of significant partial
autocorrelations
  
Autumn only
Abundance144·5**3·530·562(2)−0·79  
Log abundance144·7**3·530·462(2)−0·77  
Differenced134·2*3·230·472(2)−0·74  
Spring and Autumn
Abundance2710·1 **3·330·561·5, 2(1)−0·45(1·5)−0·48(2)−0·61
Log abundance279·9**3·330·521·5, 2(1)−0·43(1·5)−0·49(2)−0·62
Differenced265·7*3·230·432(2)−0·66  
(b) Individual sampling sites
Sites-index
autumn
samples
Number of
samples
KappaDominant
frequency
(years)
T
(years)
ACF(T)Lag of other
significant (P < 0·05)
autocorrelations
Value and lag (in brackets) of
significant partial
autocorrelations
  
Spring and Autumn
10·41273·43·3(5)−0·38   
20·46275·8**3·33·50·321·5, 2(1·5)−0·41(2)−0·54 
30·45278·3**3·33·50·521·5, 2(1·5)−0·43(2)−0·57 
40·45276·2**3·330·61*1·5(1·5)−0·62  
50·37277·3**3·330·61*1·5(1)−0·51(2)−0·60 
60·42277·3**3·330·481·5, 2(1·5)−0·39(2)−0·54 
70·46277·3**3·33·50·441·5, 2(1)−0·42(1·5)−0·42(2)−0·42
80·39263·53·32·50·361, 1·5(1)−0·44(1·5)−0·42(2)−0·39
90·43274·13·31·5(1·5)−0·41(2)−0·38 
100·32276·6**3·330·441·5(1·5)−0·42(2)−0·49(2·5)−0·47
110·38276·4**3·33·50·261·5, 2(1·5)−0·40(2)−0·48 
120·33217·4**3·53·50·531·5, 2(1·5)−0·49(2)−0·46 
130·35275·3**3·330·481·5, 2(1)−0·45(2)−0·55 
140·37277·7**3·33·50·441·5, 2(1·5)−0·58  
150·26225·8*3·63·50·352(2)−0·62  
160·40225·6*3·63·50·40(1)−0·44(2)−0·48 
170·40223·42·7(1)−0·56  
180·40225·8**3·640·411·5, 2(2)−0·57  

Common features of cycles in kielder forest

The calibration of VSI with vole density was used to describe the seasonal patterns of population growth common to each cycle (Fig. 3). In all three cycles, the increase phase covered a 2-year period with the highest growth rates occurring in early summer (Fig. 4), whereas the summer to autumn period (late summer) also contributed positively but often with slower growth. Overwinter growth rates were much less variable than summer rates (Bartlett's test B = 10·61, P < 0·005; Fig. 4) and they were mostly positive, with declines occurring only in the winter of 1984–85 and in 1988–89. In all cases, the decline phase began in the spring after the 2 years of increase. All four cyclic declines occurred during the summer months (the main breeding season) and the low-point of the cycle was typically the autumn in the year of the decline. During these declines there was no recovery during the breeding season, although in three out of four cases the decline rate appeared to be lower in the early summer than the late summer (most breeding typically occurs between April and June during the increase phases, J.L. Mackinnon unpublished data). Extended periods with no population recovery were observed in 1992 (18 months) and 1996 (12 months) following the declines of 1991 and 1995.

Figure 4.

The contribution of different seasons to the population growth rate during the course of the study. Growth rates have been calculated based on density estimates. We assumed that in each year there was a constant growth rate between autumn and spring.

Partial autocorrelation functions

The length of the time series available is insufficient for a thorough examination of the density-dependent structure of the time series. Therefore, for simplicity, we only explore linear density dependence. Partial autocorrelation functions based on a single value per year (n = 13 years) are characterized by strongly negative and significant values for a time-lag of 2 years (Table 2, Fig. 5c) suggesting second-order density dependence. The same structure is revealed when using abundance, log abundance or log growth rates. Partial autocorrelation functions based on two (spring and autumn) samples per year (27 estimates) reveal significant negative values for lags of 1, 1·5 and 2 years (Fig. 5d), suggesting that the time-lag might be somewhat shorter than 2 years. The shape of the autocorrelation function of each time series is broadly similar to that obtained from the pooled data (Table 2).

Figure 5.

(a) Autocorrelation (ACF), and (b) partial autocorrelation functions (PACF) of log-transformed autumn (a, b) and spring and autumn (c, d) vole abundance averaged from all sampling sites in Kielder Forest. Correlation coefficients shown in white are P < 0·05.

Populations of field voles reached peak density in 1930, 1933 and 1937 in Newcastleton during the afforestation phase (Fig. 6). The spectral density of the time series (calculated on log snap-trapping index after detrending) is characterized by a peak at 4 years which approaches statistical significance (k = 3·84, M − 1 = 7, P = 0·06) and significant negative autocorrelation was detected for a lag of 1·5 years.

Figure 6.

Vole fluctuations at Newcastleton, 17 km west of Kielder Water, between 1931 and 1939, redrawn from original data provided by D. Chitty. (a) Values are the percentage of 100–175, 2 sq. yard quadrats with fresh vole faeces (see Chitty 1996 for details). (b) The number of field voles caught per 250 trap nights (50 snap traps × 5 nights). White and black circles are from spring (April) and autumn (September) surveys, respectively.

Discussion

Field vole populations in northern England show cyclic dynamics, in many ways similar to those reported from central Fennoscandia. There is a clear multiannual periodicity with larger density differences between rather than within years. Populations occasionally grow in winter (October–March) and declines take place in summer, a feature associated with high amplitude cycles in northern Fennoscandia (Hansson & Henttonen 1985; Hansson 1987). Thus, the gradient in cycle length and amplitude present in Fennoscandia appears not to extend southwards to the British Isles, if that were the case, then populations of field voles should be noncyclic. There is further evidence that field vole populations experience regular 3- to 4-year cycles in western and northern Scotland (Petty 1992; Swann & Etheridge 1995; Chitty 1996). A field vole population studied by Tapper (1979) in southern England experienced wide amplitude fluctuations over 6 years but further data are needed to ascertain whether these were regular cycles and whether there is a north–south gradient of cyclicity of field vole populations within Britain. To our knowledge, there is no evidence of bank vole populations experiencing 3- to 4-year cycles in Kielder Forest (Petty 1999) or elsewhere in Britain. This lack of synchrony between rodent species may thus be similar in Britain to the transition zone centred around 60°N in Fennoscandia, where forest-dwelling bank voles tend to have more stable populations than field voles which inhabit clear-cuts and grasslands (Hansson & Henttonen 1985).

Most long-term studies of cyclic vole populations rely on indices of vole abundance rather than on measures of density derived from live-trapping data (see Henttonen et al. 1987 for a notable exception). In the present study, we used sign indices calibrated from direct estimates of vole density. The calibration against capture–recapture density estimates made it possible to quantify the measurement errors for individual estimates and these were relatively large. Nevertheless VSI averaged over 14–18 sites provide good estimates of vole abundance at the landscape scale.

The amplitude of the cyclic fluctuation of field voles in Kielder Forest appears lower than studies from northern Fennoscandia, but comparable to the regular cycles of grey-sided voles from Hokkaido in Japan (Saitoh 1987). The pattern of spatial asynchrony, which is akin to a travelling wave in our study area (Lambin et al. 1998), contributes to the low amplitude estimates. Indeed, even during periods of lowest vole abundance, some quadrats with fresh grass clippings were found in a substantial proportion of the habitat patches sampled (range 37–73%, mean 62%). In contrast, snap-trapping studies in Fennoscandia often yield no capture on the landscape scale. However, we only sampled good Microtus habitat in our study area. Around 70% of the landscape (closed-canopy, spruce forest) is unsuitable for field voles and remains unoccupied throughout all phases of a cycle (J.L. Mackinnon, unpublished data). Methodological considerations probably contribute to the difference in estimates of amplitude, but there is little doubt that voles decline to much lower densities in cyclic populations in Fennoscandia than in northern England. Otherwise, population cycles are remarkably similar in both regions. Therefore processes causing extremely low densities during the low phase in northern Fennoscandia, which probably contribute to spatial synchrony, are not necessary components of cycles. The high amplitude, prolonged periods with low density and large-scale geographical synchrony, which have sometimes been taken as defining features of population cycles (Hansson & Henttonen 1985; Hansson & Henttonen 1998) might be consequences of trophic interactions modifying lower amplitude, but nonetheless ‘real’ cycles in northern Britain, Hokkaido and the transition zone in Fennoscandia. Geographical gradients in amplitude could then reflect that population oscillations can occur within different communities of predators, which track cycles and modify their amplitude without necessarily being the primary cause of such cycles. The above interpretation is consistent with time series of cyclic microtine populations showing delayed density dependence (Hörnfeldt 1994; Saucy 1994; Bjørnstad et al. 1995; this study).

According to the predation hypothesis, variation in the impact of generalist predators on vole populations from north to south in Fennoscandia accounts for the gradient in cycle length and amplitude. In Turchin & Hanski's (1997) predator–prey model, the saturation rate of the whole generalist predator community (G) should reflect the combined effect of the abundance of generalists and variation in the susceptibility of voles to such predators caused by snow cover and landscape structure. Parameter G is a bifurcation parameter accounting for the type of dynamics (noncyclic, low amplitude 3- to 4-year or high amplitude 4- to 5-year cycles) of the model. Unfortunately, estimates of intake by predators are difficult to obtain and are notoriously unreliable for even a single species (Boutin 1995). Turchin & Hanski (1997) were therefore only able to estimate G for two sites along the entire Fennoscandia gradient. The first estimate was obtained from Korpimäki & Norrdahl's (1991a, b) work in central Finland (63°N), where vole populations are restricted to the edge of agricultural fields and experience 3- to 4-year cycles of amplitude similar to those observed in Kielder Forest. Turchin & Hanski's (1997) estimate of G assumes that three raptor species (kestrels, short-eared owls and long-eared owls) for which data were available account for the entire community of predators acting with no time delay in this area. A second estimate was derived from that of Erlinge et al. (1983) of abundance and intake rates of nine generalists predators that feed primarily upon rabbits (Oryctolagus cuniculus L.) and on field voles from noncyclic populations in southern Sweden, at latitudes similar to that of Kielder Forest (56°N). Turchin & Hanski (1997) then assumed that G decreases approximately linearly with latitude and they were able to reproduce the cline in amplitude and period observed in available time series from Fennoscandia. Turchin & Hanski (1997) also predicted that G should range between 70 and 120 voles ha−1 of vole habitat year−1 in Kielder Forest. O’Mahoney et al. (1999) estimated that foxes alone removed approximately 200–290 voles ha−1 vole habitat year−1 in Kielder Forest in a year of peak vole density. The approximate contribution of tawny owls, calculated using values from Petty et al. (2000) is 22–57 voles ha−1 vole habitat year−1. Estimates of the total number of voles removed by generalist predators would be further increased if the contributions of nomadic avian specialists and scarcer generalists such as mink, badger and stoat were included. However, information on density and diet of these species in Kielder Forest is far too imprecise to justify the calculation. Nevertheless, the number of field voles removed by the two most common generalist predators in Kielder Forest is well above the value predicted by Turchin & Hanski (1997) for the whole generalist predator community. We stress however, that the large discrepancy between the prediction of Turchin & Hanski (1997) and our data could result from the poor reliability of the parameter estimates used in designing and testing the prediction. Thus, even though G should reflect the impact of several variables on the variation in the impact of generalist predation pressure along the Fennoscandian gradient, such parameter cannot form the basis of a strong test of the specialist/generalist predation hypothesis.

An alternative approach is to consider variables correlated to the Fennoscandian gradient which are assumed to contribute to variations in generalist predation pressure. We attempt to determine whether these are causally linked to the cyclic nature of population oscillations or mere correlates of the gradient, by asking whether conditions are similar in regions with similar population dynamics.

Number of species, density and biomass of generalist predators (including raptors) decrease by approximately one order of magnitude with increasing latitude in Fennoscandia (Hanski et al. 1991). The functional response of generalists with relatively stable populations is thought to prevent population cycles in southern Fennoscandia. The assemblage of rodent-eating predators in Kielder Forest is similar to that present in southern Sweden, except for the scarcity of feral cats and the absence of common buzzards (Buteo buteo L.) until 1996 (S.J. Petty, unpublished data). Buzzards are however, present in similar conifer forests, open moorland and farmland to the west of Kielder Forest, where voles experience population cycles (Taylor 1994; Petty & Fawkes 1997; Leckie et al. 1999). Quantitative estimates of abundance are not available for all species but the abundance of four generalist predators has increased in the last decade with no noticeable changes in vole dynamics [buzzard colonizing in 1996, American mink invading in the early 1980s (S.J. Petty, personal observation), tawny owls in part of the forest increased from around 40 pairs in the late 1970s to just under 60 pairs by 1990/91 (Petty et al. 2000), and badger numbers have also increased in the 1990s (D. Anderson, personal communication)]. The availability of alternative prey decreases with increasing latitude in Fennoscandia and this should influence the effectiveness of generalist predators in regulating vole populations (Hansson & Henttonen 1985). In models of predator–prey interactions, alternative prey enables weasels to persist during periods of low vole abundance and lower the period and amplitude of population cycles (Hanski & Korpimäki 1995). When field vole density was less than 110 ha−1 on clear-cuts in Kielder Forest, red foxes increasingly relied upon carrion of roe deer (Capreolus capreolus L.) and rabbits (O’Mahony et al. 1999). Both these alternative prey species are scarce or absent from the Fennoscandia taiga. Tawny owls preyed mainly on common frogs (Rana temporaria L.), song birds and other small mammals when field voles were scarce (Petty 1999). Unlike most Fennoscandian owls, tawny owls in Kielder Forest are sedentary and in years of low density add to predation pressure on voles.

The lack of snow cover in southern Scandinavia is thought to increase the vulnerability of voles to predation by generalist predators, leading to more stable dynamics (Hansson & Henttonen 1985). Lindström & Hörnfeldt (1994) suggested that, in central Sweden, mild winters and reduced snow cover led to more stable vole dynamics via changes in the predator assemblage, in a region where vole cycles traditionally occurred at the southern edge of their geographical extent. The average annual duration of snow cover in Kielder (mean: 40 days, range: 5–85 days) is in the lower range for southern Sweden (20–120 days; Ångström 1974 in Lindström & Hörnfeldt 1994). Thus, sustained regular cycles do occur in the absence of significant snow cover. Demographic processes which occur during spring–autumn are the main determinants of multiannual differences in density in Kielder Forest (Fig. 4), emphasizing the point that snow cover is unlikely to be important.

The increasing heterogeneity of the landscape with decreasing latitude in Fennoscandia has been hypothesized to increase the diversity of alternative prey to generalist predators and the susceptibility of dispersing voles to predation, hence stabilizing their dynamics (Erlinge et al. 1983; Hansson & Henttonen 1985; Hanski 1987; Hanski et al. 1991). The decline in amplitude of field vole population oscillations in a landscape of newly afforested grass moorland in Wales as trees grew and shaded out the grass (Chitty & Chitty 1962), led Hansson & Hentonnen (1985) to assume that microtine population cycles in Europe were restricted to areas where habitat heterogeneity was low. Indeed, man-made forests in Britain differ profoundly from Fennoscandian taiga, as the former completely lack ground cover whereas ericaceous vegetation provides cover for all vole species, and the main habitat for Clethrionomys voles in Fennoscandia. However, we have shown that the population dynamics of field voles in our study area were similar in the 1930s to the present day. During the intervening years, the area has been profoundly modified by forestry. In the 1930s, the landscape was dominated by fields and grassland and heather moorland grazed by sheep and a few newly established spruce plantations. Vast areas were converted to spruce forest during the period 1930–60. Today, field voles are absent from mature forest, which covers around 70% of the area but they occur in ephemeral grassland patches created by clear-cutting and connected by linear grassy habitats alongside roads and watercourses. Thus, vole dynamics appear not to have changed in the last 60 years despite a massive increase in the proportion of the landscape without any ground cover, which would provide little protection to voles leaving grassland patches. Thus, the presence of poor quality habitats where dispersing voles experience high predation rates (R. O’Malley & X. Lambin, unpublished data) does not preclude cycles from taking place. Models exploring the effect of habitat heterogeneity on the stability of small rodent populations are nearly unanimous in concluding that it should be stabilizing (reviews in Ostfeld 1992; Stenseth & Ims 1993; Oksanen & Schneider 1995). In models assuming optimal patch use by predators, increasing the mortality and/or energetic cost associated with the use of non-prey habitats stabilizes prey dynamics (T. Oksanen, personal communication). Unfortunately, such costs cannot be quantified empirically, let alone be compared across sites or time periods.

To conclude, our study highlights the need to consider cyclic populations beyond Fennoscandia in attempts to disentangle the contributions of the many correlates of cyclicity (see also Stenseth & Saitoh 1998). Qualitative consideration of three variables correlated to the Fennoscandian gradient, and widely assumed to contribute to variations in generalist predation pressure, suggest they are only correlates of the gradient and may not have an important influence on vole dynamics. Our data challenge the hypothesis that low predation rates by generalist predators are necessary for vole dynamics to be dominated by the destabilizing impact of weasel/vole interactions or other processes. They do not, however, amount to a test of the role of weasel predation as a cause of population cycles. Weasels are present in Kielder Forest with densities in line with those suggested by Lockie (1966) for similar habitats, but well above those inferred by Korpimäki & Norrdahl (1998) from the small numbers they removed experimentally from large areas. We are presently conducting an experimental test of the hypothesis that their delayed numerical response to changes in vole abundance drives population cycles through sustained removal in replicated clear-fell patches separated by mature forest (I.M. Graham & X. Lambin, unpublished data).

Acknowledgements

James MacKinnon acknowledges the support of NERC studentship GT24/95/lspe/2. We thank Graham Gill, Forest Enterprise, for allowing us to work in Kielder Forest and for providing student accommodation. We dedicate this paper to Dennis Chitty and thank him for providing the data collected in the 1930s. We acknowledge R. Boonstra, D. Chitty, I. Graham, I. Hanski, H. Henttonen, T. Oksanen, P. Turchin and an anonymous referee for comments on previous versions of this paper. This project is supported by NERC (GST/02/1218).

Received 18 February 1998;revisionreceived 19 April 1999

Ancillary