Wolverines Gulo gulo are polyphagous, which enables them to switch between different food sources when one prey becomes scarce. In Scandinavia, the wolverine primarily scavenges on large ungulates (Haglund 1966). Wolverine is sympatric with both wild and domestic reindeer Rangifer tarandus, which constitutes its most important winter food source (Pulliainen 1968, 1988, Myhre & Myrberget 1975, Semenov-Tyan-Shanskii 1982, Landa, Strand, Swenson & Skogland 1997). Hares Lepus timidus, ptarmigan Lagopus spp. and small rodents are also significant winter food sources for wolverines (e.g. Pulliainen 1968, Myhre & Myrberget 1975, Landa et al. 1997) and may be the most important foods during summer (Myrberget & Sørumgård 1979, Magoun 1987, Landa et al. 1997). Larger animals in the diet of wolverines are probably obtained mainly as carrion (Magoun 1987, Banci 1987, 1994, Landa et al. 1997), but the wolverines may prey heavily on domestic sheep Ovis aries (Olstad 1945, Myrberget & Grotnes 1969, Kvam, Overskaug & Sørensen 1988, Mortensen 1995, Børset 1995) and domestic reindeer (Bjärvall, Franzén, Nordkvist & Åhman 1990).

In Norway, upland areas are grazed by large numbers of free-ranging sheep. In many areas, there are serious problems with wolverine predation on sheep, although sheep numbers seem to have little effect on wolverine numbers or reproduction (Landa & Tømmerås 1996, 1997, Landa et al. 1997). In the upland areas, sheep are released on mountain pasture in June and graze unattended until the beginning of September, when they are collected. If the future of the wolverine in these areas is to be secured, the conflict created by wolverine predation on untended sheep on summer grazing must be solved. This conflict has recently spread to new areas (Landa 1997). A study was therefore initiated to establish how captive wolverines react to the presence of oils and chemicals (Landa, Tømmeråis & Skogland 1993, Landa & T0mmerds 1997). Based on the results of that study Landa & Tømmerås (1996) tested the release rates of different chemicals in the laboratory and developed a dispenser which made it possible to use chemicals as long-lasting repellents. Volatile repellents were then attached to sheep in free-ranging flocks. The results showed that 50% fewer lambs were lost in treated groups than in untreated groups (Landa et al. 1993, Landa, Tømmerås & Bergersen 1994, Landa & Tømmerås 1996), which led to the conclusion that repellents reduced predation rates, but essential questions remained unanswered by the small-scale tests. What would happen if all sheep in a large area were treated with repellents and if wolverines no longer had a choice between treated and untreated sheep? Would they simply switch back to their natural prey, or would they become habituated to the volatile agents? Landa & Tømmerås (1996) suggested that a large-scale experiment should be performed before a conclusion could be reached on whether or not aversion agents would be an effective method to reduce predation on large numbers of untended sheep grazing on open range. Hence, the purpose of this study was to test the effect of the repellent in a large-scale experiment, in which wolverines no longer had lambs without repellents as alternative prey within a large area. The experiment was designed to determine whether wolverines would switch back to their natural prey or whether they would become habituated to sheep with volatile repellents.

Study area and material

The previous small-scale tests were carried out on two upland plateaus in south-central Norway, Snøhetta and Trollheimen. While Landa & Tømmerås (1996) found no correlation in the entire Snøhetta area between sheep losses and the recorded number of attempts by wolverines to reproduce, Gudvangen (1995) found that in some smaller parts of the area losses were higher in years when wolverines reproduced. To ensure that the large-scale experiment was carried out where there was potentially high wolverine predation, thereby covering a representative variation in predation pressure, four geographically distinct areas were selected, according to the following criteria: 1) wolverine breeding had been recorded recently in the area, 2) documentation of sheep losses was available for a period of more than 10 years, and 3) high losses of lambs due to wolverine predation had been documented in recent years. All lambs in the four areas were fitted with repellents.

The four areas were located in Ulvådalen, Rauma, in the county of Møre & Romsdal, Åmotsdalen, Oppdal, in the county of Sør-Trøndelag, Høvringen, Sel, in the county of Oppland, and Sagfjord, Hamarøy, in the county of Nordland (Fig. 1). A total of 5,876 lambs were treated in the four areas. The approximate sizes of the respective grazing areas were 75 km² (Ulvødalen), 207 km² (Åmotsdalen), 140 km² (Høvringen) and 84 km² (Sagfjord).

Olfactory and taste agents were placed in low-density polyethylene dispensers, as in previous experiments (Landa, Tømmerås & Krogstad 1995, Landa & Tømmerås 1996). The dispensers were attached to a conventional ear tag or to a specially designed wool clip on an elastic collar constructed to keep the dispenser in position in the upper neck region. At Sagfjord, the dispenser was attached to a collar made by Os Husdyrmerkefabrikk AS. The lambs were fitted with the dispenser when they were released onto summer grazing in late May to mid-June. The flocks were monitored by trained personnel throughout the grazing season. Post-mortem analyses were performed on sheep carcasses using the method described by Myrberget & Sørensen (1981) and Sørensen, Mysterud & Kvam (1984). A representative sample of repellents was collected from each flock during gathering in autumn to check whether the dispenser functioned technically correct.

Figure 1.

Location of the four test areas (A-D) in Norway.

Survival model

To remove variations ascribed to other variables than the dependent variable and to be able to estimate the separate effect of repellents more precisely, we used the following model of the general pattern in losses during the period from 1982 to 1996. We assumed that the dependent variable, the number of individuals lost out of the total number released, is binomially distributed, and that the probability, p, that a single individual is lost is related to a linear predictor (β^Tx = (β₁x₁ + β₂x₂ +... + (β_nx_n of parameters and covariates through a link function according to the relation f(p) = β^Tx. The model thus falls into the category of generalised linear models (McCullagh & Nelder 1993). Such models are extensions of more traditional regression models in that distributions other than the normal can be assumed for the response variable. In addition, for different choices of link functions f, different relationships, other than linear ones, between the response variable and the linear predictor can be assumed. The choice of f should preferably be based on the biological problem in question.

We assume that the probability that an individual dies in a short time interval t, t + dt is λdt provided it is alive at time t. It then follows that the survival time is exponentially distributed. The probability that an individual is lost (due to predation or other causes) when exposed for a time interval of length T (the length of the summer), is then

By letting λ depend on different covariates and parameters, the effects of different variables can be estimated and tested. When analysing survival data, it is common to assume that different mortality factors (or covariates) act multiplicatively on the mortality rate λ (e.g. in Cox & Oakes' (1984) proportional hazards model). For example, it is reasonable to assume that an animal exposed to some illness would experience a multiplicative, say two-fold, increase in the predation risk. With several such covariate factors acting together in this manner, the hazard rate can be expressed as A. = exp(β^Tx). However, some mortality factors may certainly act additively on the mortality rate. In the present study, for example, many lambs died from ‘natural’ causes, such as becoming stuck in impassable terrain. Although the risk of this taking place may increase with the predation pressure, it is also clear that it happens regardless of the predator density in the area, in an additive manner. Such a form of mortality can be incorporated into the model by letting

where parameter C represents this additive mortality. By substituting equation (1) into equation (2) and rearranging, we obtain the following relation between the linear predictor and the probability of dying:

where C′ = CT and β′ = In Tp are combined estimable parameters. Note that p tends to approach the limit p ≈ C′ (for small C′), when the predation pressure becomes small (i.e. when β′^Tx→ - ∞). The parameter C′ can thus also be interpreted as the background mortality. It should also be noted that equation (3) corresponds to the commonly used comple mentary log link function in the case of C′ = 0 (McCullagh & Nelder 1993).

We used the following variables as covariates: time, locality, interaction between time and locality, age of the individuals (lamb or sheep) within each population, and a dummy variable representing the occurrence of wolverine reproduction in the grazing areas. To test for the effect of the fitting of lambs with repellents, a dummy variable for treated/untreated lambs was included. In addition, the effect on ewe mortality of having treated lambs in the same population the same year was estimated by including a dummy variable, which was 1 for populations of ewes in the same area and year as treated lamb populations, and otherwise 0.

A χ²-test was included to test for differences in losses independent of factors between the year of the large-scale experiment (1996) and the previous year.

Results

Losses of lambs and sheep increased from 1983 to 1996 in all four test areas. A significant interaction between year and locality showed that the trends in losses differed from one area to another. In the model, the background mortality rate of lambs and ewes was estimated to be C = 1.6%. More lambs than ewes were lost and higher total losses were found in years when wolverines reproduced within the same areas. Finally, the model showed no reduction in the losses of lambs or ewes in response to the repellents (Table 1, Fig. 2).

Table 1.

Results of the generalised linear model to test for differences in mortality among test areas, sheep losses with time, loss difference between ewes and lambs, the effect of wolverine reproduction within grazing areas, and possible reductions in losses due to the use of repellents. For each locality, the losses predicted by the fitted model are given for 1996.

Figure 2.

Losses of lambs within the four test areas (A-D) during 1982 – 1996.

Table 2.

Total losses of ewes and lambs on summer grazing in the four test areas in 1995 when no lambs were treated, and in 1996 when all lambs were treated.

In the four areas, 1,034 lambs (17.6%) were lost out of a total of 5,876 lambs with repellents released onto summer pasture (Table 2). Of the 208 dead lambs that were found, 159 had been killed by wolverines. The losses in each area during the experiment were: 21.8% in Ulvådalen where wolverines had killed 45 of the 73 lambs found, 20% in Åmotsdalen where wolverines had killed all the 23 lambs that were found, 9.6% in Høvringen where wolverines had killed 23 of the 36 lambs found, and 19.7% in Sagfjord where wolverines had killed 68 of the 76 lambs that were found.

In Ulvådalen, the losses were higher in 1996 (294 of 1,346) than in 1995 (89 of 1,289), (χ² = 156.2, P < 0.001). In the other areas, no significant differences were found in the losses between the two years.

In all the areas combined, significantly fewer ewes were lost during the year of the experiment (73 of 3,185, χ² = 4.06, P= 0.044) than during 1995 (103 of 3,321). When the figures for the individual areas were examined separately, the difference was only significant in Høvringen (χ² = 3.95, P = 0.047). During the experiment, the carcasses of 16 out of 73 missing ewes were found, and seven had been killed by wolverines.

In two flocks where elastic collars had been used, 17.4% of the wool clips had loosened and the repellents had become displaced. The proportion of repellents lost when the dispenser had been attached to an ear tag was 30% in Åmotsdalen, 2.7% in Ulvådalen and 0.9% in Høvringen (Table 3). The ear tag caused wounds at the attachment spot on nine lambs (1.7%). No dispensers attached to elastic collars were lost, but 0.3% of the dispensers attached to ear tags were lost. The chemical leaked out of 4.1% of the dispensers (Table 4).

Tabel 3.

Functionality of the wool clip and the ear tag.

Table 4.

Repellent functionality and the quality of the ampoule.

Discussion

Based on the long-term data from the areas concerned, wolverines mainly killed lambs and losses increased in years when wolverine reproduction took place within the grazing areas. The wolverine denning period overlaps with the mating period (Rausch & Pearson 1972) and the relatively restricted home ranges used by females with cubs compared with non-breeding females (Homocker & Hash 1981, Banci 1994) are frequently visited by adult males and juveniles from earlier litters (Magoun 1985). Although wolverine females accompanied by cubs have been observed preying on lambs (A. Landa, unpubl. data), it is likely that the positive relationship between lamb losses and wolverine breeding within the grazing areas is a result of high wolverine densities within the breeding areas, resulting in higher predation pressure on lambs.

If the repellent reduced wolverine predation, as had been shown in small-scale experiments (Landa & Tømmerås 1996), the losses could be expected to be smaller than in previous years. However, the effect of the repellents was difficult to determine because we did not have control groups in this large-scale experiment as in the previous experiments (Landa & Tømmerås 1996). Thus, we do not know how large the losses would have been without repellents. In Ulvådalen, the proportion lost due to wolverine predation was higher than in previous years when the lambs did not have repellents. This may be explained by a recent establishment of wolverines within the area and surplus killing by a few wolverines (Landa, Krogstad & Tømmerås 1996). However, we found no demonstrable effect in reduced lamb losses in any of the other areas. Moreover, in all the areas, a relatively large number of lost lambs were recovered and were found to have been killed by wolverines. The lamb losses were generally high and no reduction could be demonstrated in any area when compared with the previous year or the trend in losses during the period from 1983 to 1996. Previous experiments had shown a reduction in the loss of lambs equipped with repellents if wolverines had ordinary lambs as alternative prey (Landa et al. 1993, 1995, Landa & Tømmerås 1996). The lack of effect in reduced lamb losses, when all the lambs were fitted with repellents, suggests that wolverines coped with the situation by killing treated lambs when they had no untreated lambs as alternative prey. Repellents of the kind used may therefore only be a potential tool to reduce losses in target groups of untended sheep on summer grazing.

Carnivore predation can be separated into several steps of behaviour, for example; search, identify, approach, attack, kill and consume (i.e. Kruuk 1972, Mysterud 1980, Seidensticker & McDougal 1993). Protecting livestock can be viewed as an attempt to interrupt the behavioural sequence at one or more of these steps. Ideally, the process should be interrupted as early in the sequence as possible (Linnell, Smith, Odden, Kaczensky & Swenson 1996), as was the main objective of our repellent. A pertinent question is if repeated use in a particular area may cause habituation. However, this would be unlikely as long as the wolverines have untreated lambs without repellents as an alternative. Moreover, the sheep graze for only a short period each year and the turnover in wolverine populations is probably relatively high (Banci 1994). We conclude that, on the basis of this large-scale experiment, the repellent cannot be recommended as a tool to reduce overall predation.

There were relatively few technical problems with the repellent. However, some variation was observed from one flock to another regarding how many dispensers attached to ear tags were lost during the grazing season. This is probably related neither to the attachment of the dispenser to the ear, nor to morphological differences between different breeds of sheeps, but rather to differences between grazing areas. The ear tag may be more vulnerable in dense vegetation than the conventional collar made by Os Husdyrmerkefabrikk AS.