Methods of estimating population density of animals and plants pdf

An effective system for monitoring wildlife populations enables management and decision-making by providing reliable data on the number of animals, distribution, individual growth rate, reproduction and sex/age composition. Over the years several different survey methods have been used to monitor ungulate populations, including aerial surveys, pellet group counts, direct observations (e.g. hunter observations and drive counts), hunter harvest statistics and snow-track counts (Timmerman 1974, Mayle et al. 1999, Solberg & Sæther 1999). Three of these methods are commonly used within Fennoscandian moose Alces alces management, i.e. aerial surveys, pellet group counts and hunter observations (Haagenrud et al. 1987, Lavsund et al. 2003, Wennberg DiGasper 2006).

However, the methods vary in terms of reliability (accuracy and precision), costs, information obtained and time period surveyed (Fuller 1991, Mayle et al. 1999, Barnes 2001, Campbell et al. 2004, Smart et al. 2004, Månsson et al. 2007; Table 1). Such differences (see Table 1) make it difficult to select the most suitable method for management, indicating that further evaluation of the different methods is needed. For moose, aerial surveys have probably been the most accepted and frequently used method for monitoring population densities and trends (Timmerman 1974, Jachmann 2002, Wennberg DiGasper 2006, Pople et al. 2007). Two other methods commonly used to estimate ungulate populations are pellet group counts (Neff 1968, Barnes 2001, Wennberg DiGasper 2006) and direct observations such as hunter observations (Haagenrud et al. 1987, Ericsson & Wallin 1999, Solberg & Sæther 1999, Wennberg DiGasper 2006). These two methods differ from aerial surveys in that they result in an indirect measure, i.e. an index of the number of animals (Neff 1968, Ericsson & Wallin 1999, Solberg & Sæther 1999, Andersen et al. 1992). Hunter observations are provided by moose hunters and include the number of moose observed, the number of active hunters, and the time (hours) spent hunting per day during the first week of the hunting season.

Table 1.

Comparison of different methods used to survey moose.

The objective of our study was to combine information from several survey methods in a long-term study to: 1) improve our understanding of the accuracy, concordance and usefulness of these methods, and 2) improve our understanding of the population development within a moose management area in terms of animal numbers, age-sex structure and migration.

We did this by comparing estimates of moose density provided by the four census methods (cohort analysis, aerial counts, pellet group counts and number of moose observations per hunter day) used within the Grimsö Wildlife Research Area. Data from a 30-year period was used to reconstruct the population size using cohort analysis (Fryxell et al. 1988, Ferguson 1993, Solberg et al. 1999). Age-specific natural mortality rates were estimated using radio-collared moose in the area in order to provide input parameters for the reconstruction of the moose population. By comparing methods we also estimated a conversion factor for transforming indices obtained from pellet group counts into aerial counts, and evaluated the importance of migration as a contributory factor to population development.

The 135 km2 Grimsö Wildlife Research Area, located in south-central Sweden (59°5′N, 15°5′E), is a rugged plateau with elevations ranging within 100–150 m a.s.l., and is composed of low flat ridges with till and boulders interspersed with bogs and swamps. The area comprises 72% forest, 18% bogs, 7% lakes and rivers, and 3% meadows and farmlands. Mature forest stands are dominated by Scots pine Pinus silvestris, Norway spruce Picea abies, and birch Betula pubescens and B. pendula. Rowan Sorbus aucuparia, aspen Populus tremula, and willows Salix spp. are preferred moose browsing species but occur rarely within the mature stands. Forest management is intensive, with clear-cutting of 3–10 ha patches and old forest replaced by planting. The period of rotation in the forest stands is 80–100 years. Early succession after logging consists of birch, aspen and willow with an under-storey of common hair grass Deschampsia flexuosa, bilberry Vaccinium myrtillus, cowberry V. vitis-idea and heather Calluna vulgaris. Climate is typical for inland, central Sweden, with winter temperatures down to −20°C and summer temperatures up to 25°C. Mean daily temperature is 16.3°C and −4.4°C in July and January, respectively. Snow cover is normally present from December to March with a mean snow depth of 25–30 cm in February. Annual precipitation averages 670 mm with a maximum in July (average=86 mm; Swedish Meteorological and Hydrological Institute). Other ungulates within the research area include roe deer Capreolus capreolus, whose population densities ranged between approximately 1–5/km2 during the study period (Å. Pehrson, unpubl. data). Potential moose-predators, such as wolf Canis lupus and brown bear Ursus arctos, were not present within the study area during the study period (Wabakken et al. 2001, Swenson et al. 1998).

A cohort analysis is mainly based on age and sex information from harvested animals, but natural mortality has to be accounted for in order to obtain reliable estimates of population size. Consequently, we estimated mortality rates attributable to causes other than hunting for radio-collared moose (both adults and calves). During 1980–2000, 63 adult females and 44 adult males were captured and equipped with radio-collars. The age of moose captured and collared was estimated by tooth wear of the incisors (Skuncke 1949). Radio-tracking was normally performed once a week (Cederlund & Lemnell 1979). For further details on adult moose tagging, see Cederlund et al. (1989).

Over the entire study period, radio-collared males and females were followed for an average of 2.4 (SD=1.8) and 5.6 (SD=3.6) years, respectively. Mean age of all radio-collared males and females was 4.8 (SD=2.9) and 9.9 (SD=5.3) years, respectively, whereas the maximum age for males and females was 12 and 21 years, respectively. In the spring of 2005, none of the males and only nine females were still alive. In total, 68% of the males were shot during the hunting season, 18% died of natural causes, whereas contact was lost with 14% of the collared moose due to malfunctioning radio-collars. Among females, 48% were shot during the hunting season, 30% died of natural causes, 8% were lost from the study for unknown reasons (possibly due to malfunctioning radio-collars) and 14% were still alive in the spring of 2006.

Survival rates for calves through their first winter were estimated by observing radio-collared moose cows and counting their calves on different occasions during the year, and also by using radio-collared calves. Newborn calves (1–10 days old) were captured within Grimsö Wildlife Research Area during 1993–2001. See Ericsson et al. (2001) for further details on moose calf tagging.

Since 1973, the number of moose shot each year within the Grimsö Wildlife Research Area has been recorded. The data used in the present study included 2,065 observations of shot animals of known sex and age. Another 53 moose (2.5%) were shot in the research area but were excluded from further analyses due to missing information on age or sex. For all adult moose shot, mandibles were collected and used for age determination in the laboratory by counting the number of layers in the cementum annuli of the first molar (Markgren 1969).

During 1977–2006, the size of the moose population in the Grimsö Wildlife Research Area was estimated by aerial counts from helicopter on 12 occasions. The method used during 1977–2002 was total counts of moose (N=11) based on line-transect surveys, with 300 m between transect lines (LeResche & Rausch 1974, Tärnhuvud 1988). The surveys were performed in winter after the moose hunt, 1–2 days after snowfall and only when at least 20–40 cm of snow covered the ground and temperatures were below 0°C. In our comparison between methods, we used aerial counts not corrected for sightability, whereas the aerial counts were corrected for sightability in the defecation rate estimation (with a sightability factor estimated from the aerial survey in 2006).

In 2006, an aerial count was performed as a sample of the research area, comprising 15 plots of 2×2 km (equal to 44% of the research area). A correction factor for sightability in 2006 was achieved by applying a mark-recapture procedure (Krebs 1999) of plots that were surveyed twice by using two different helicopters and observers. The first helicopter to survey a plot worked along north-south transects and positioned all moose observations by GPS. These observations were treated as marked moose. The second helicopter repeated the survey along east-west transects as soon as the first helicopter left the plot (recapture). Hence, the two helicopters flew on transects perpendicular to each other to ensure that both previously observed and non-observed moose could be observed from the second helicopter. The estimated sightability factor should therefore be close to unbiased. Furthermore, the sightability factor was estimated from an expanded survey covering more than the Grimsö Wildlife Research Area (42 surveyed plots and 244 moose observations) to increase the estimation accuracy.

During 1984–2005, hunters recorded the number of moose observed during the first week of the regular hunting season within the research area. This method has been widely used in both Sweden and Norway since the mid-1980s as an index of moose population size (Jaren 1992, Ericsson & Wallin 1999, Solberg & Sæther 1999) and is based on the assumption that a change in observation rate per time unit reflects a true change in the population. Each observation of an individual moose is recorded either as adult male, adult female without calves, adult female with one calf, adult female with two calves, lone calf or moose of unknown status. In this study the data is presented as the number of moose observations per day per hunter (Table 2).

Table 2.

Descriptive statistics for data used in estimating moose population size by: aerial counts, pellet group counts, hunter observations and cohort analysis. The data were collected within the Grimsö Wildlife Research Area, Sweden, during 1973–2006. Numbers of observed moose per 1,000 ha from the aerial counts are given without correction for sightability. The ratios of marked moose observed during the aerial surveys are given as Number of marked moose observed:Total number of marked moose in the area.

Moose pellet group surveys have been carried out annually in the research area since 1977. During 1977–1998 only the southeastern part of the research area (approximately 20% of the total research area) was surveyed. Permanent squared sample plots (5×10 m) were distributed 100 m apart along transects 200 or 400 m apart. The number of sample plots included in the survey was approximately 400, except for the first two years when only 175 and 209 squares were surveyed (see Table 2).

Starting in 1997, a different sampling method was used (‘new pellet group survey’), based on 32 permanent 1×1 km squares systematically distributed over the total research area. Each square had 20 (five along each side) permanent circular sample plots of 100 m2 (Fig. 1), resulting in a total of 640 sample plots. The old and the new pellet group surveys were both performed in 1997 and 1998 to allow validation that the two methods gave comparable results.

Description of the data from the new pellet group survey. Average number of pellet groups per 100 m2 within the 32 subareas of the Grimsö Wildlife Research Area. The four subregions (NE, SE, SW, NW), used in the subsequent statistical analysis are also shown.

In both surveys, all sample plots were checked annually and cleaned for moose pellet groups in autumn (early October), while the number of new pellet groups was counted in spring (late April to early May). Consequently, we collected data on the number of pellet groups produced by moose during a specific time period (i.e. number of days) during the winter season.

The number of adult moose, and population structure prior to the hunting season, was reconstructed for the period 1973–2005 using data collected from animals shot within the research area during that period. The maximum age recorded for shot females and males was 21 and 13 years, respectively. Therefore, all males born before 1990 and all females born before 1984 were assumed to have died before the end of the study period. These are the ‘complete cohorts’, whereas male cohorts born after 1990 and female cohorts born after 1984 are referred to below as ‘incomplete cohorts’. For animals shot between 1973 and 1990, 99% of the females were shot before the age of 15 and 99% of the males were shot before the age of seven.

Following the method developed by Fryxell et al. (1988), and extended by Solberg et al. (1999), we reconstructed the complete cohorts for each sex separately from the following equation (Equation 2 in Fryxell et al. (1988)):

The number of animals in each sex-age class was calculated using Equation 1 recursively with Ni,t=Ki,t for males in age class i=13 and for females in age class i=21. The adequacy of this method assumes that there is a maximum age beyond which no animals survive, that age-specific survival is known and is constant over years, and that the population is closed with no migration (Fryxell et al. 1988). We also performed a cohort analysis with a lower maximum female age of 15 years (following Solberg et al. 1999), which had little effect on the population development (correlation between the two time series was 0.987) but resulted in a 20% population size decrease. We therefore used 21 years as maximum female age in our final analyses.

The expected number of animals born before 2005, and expected to be shot after 2005, was calculated following Solberg et al. (1999), where the expected number of animals to be shot in the future is predicted from sex-specific curves of cumulative proportions shot in relation to age in the past (Fig. 2). The number of animals in each cohort born after 1990 was calculated from the observed numbers shot within the cohort and the number expected to be shot in the future using Equation 1.

Cumulative proportion of shot moose within the Grimsö Wildlife Research Area estimated from data covering 1973–1990.

The estimates of the expected number of animals to be shot in the future assume constant hunting effort each year and also that there has been no change in hunting strategy. During the years from 1984 to 1998 (when complete data on total number of hunter days was available) the number of hunter days was proportional to the reconstructed population size (ln(hunter days)=0.97ln(population size), SE=0.008). It was therefore concluded that hunting was constant. The hunting strategy, however, changed around 1982 from a male-biased hunting strategy to a strategy for an even sex ratio. Consequently, the sex-specific curves of the cumulative proportion shot differed before and after 1982. However, in additional cohort analyses (results not shown), these differences did not generate any considerable changes in the reconstructed population. The sex-specific curves of cumulative proportion of moose shot were therefore based on all years from 1973 to 1990.

The applied cohort analysis is a deterministic method based on several estimated parameters and uncertain assumptions, and it is therefore important to perform sensitivity analyses (e.g. Eberhardt 2002) to check how uncertainty in these parameters affects the output. We calculated the effect of small changes in natural mortality, cumulative proportion shot and number of moose shot on estimated population size. We also checked the assumption of no migration by comparing estimated adult sex ratio obtained from the cohort analysis and aerial surveys.

To estimate the natural mortality needed for the cohort analysis, we estimated both survival of calves through their first winter and adult survival. The probability of calf survival was modelled with logistic regression (Proc GENMOD in SAS). Calves from the same mother were treated as repeated measurements. The explanatory variables included were: age=the age of the mother in years; age2=age×age; year=the birth year of the calf; year2=year×year; twin=binary dummy variable equal to 1 if the calf was born as a twin and 0 otherwise. Sex was not included as a possible explanatory variable because this characteristic was not always determined on capture.

Adult survival was analysed with Cox regression and possible differences between sexes and cohorts were tested. The survival library in the statistical package R (R Development Core Team 2004) was used for these analyses.

The new pellet group counts of moose were analysed in order to gain an understanding of the geographical distribution of moose density within the research area. The number of pellet groups per sample plot was analysed as a dependent variable in a generalised linear model with log link function (i.e. a log-linear model). The explanatory variables were year (class effect), four subregions (class effect) and the interaction of year and subregion (class effect). In a preliminary analysis the distance from the mid-point of the research area was also included as a covariate. The four subregions correspond to the four large quadrates (NE, SE, SW, NW) in Figure 1. The GLM library in the statistical package R (R Development Core Team 2004) was used for these analyses and P-values were based on likelihood ratio tests.

The daily defecation rate per moose in the study area was estimated based on the population size obtained from the aerial counts in 2002 and 2006 and the number of pellet groups counted in these two years. These were the only two years with available estimates based on both aerial counts and pellet group surveys from the entire research area. The estimates were obtained as x/(yz) where x is the estimated number of pellet groups in the area, y is the estimated number of moose, and z is the number of days over which the pellets have accumulated between early October and late April. The variances (Vx, Vy and Vz) of the three estimates (x, y and z, respectively) used in the formula were combined using the delta method (e.g. Casella & Berger 2001) to obtain a standard error of the daily defecation rate:

A correlation analysis was performed to compare the similarities between the estimates of moose density. The pairwise comparisons were based on a variable number of years because the different methods started at different times during the total study period.

In each pairwise comparison, the time series y1(t) and y2(t) were analysed as two random walks around the means μ1 and μ2. The modelled random walks u1(t) and u2(t) included an auto-correlation with a one year time lag such that:

The maximum likelihood estimates of μ1, μ2, σ21, σ22, r1, r2 and ρ were obtained with Fisher scoring (e.g. Pawitan 2001). The main parameter of interest is ρ, and so we tested the null hypothesis ρ=0 with a likelihood ratio test. This analysis was not performed on the aerial counts since there were several years between each count and the effect of auto-correlation could therefore be ignored. Furthermore, the number of years in which both the old and the new pellet group surveys were performed were too few for a time series analysis.

Moose harvest in the research area increased during the 1970s and peaked in 1982 with harvest rates > 4 times larger than in 1973 (Fig. 3). In the late 1970s, a local hunting strategy within the research area was applied with the objective of reducing the number of adult males per female in the living population. Consequently, harvest of adult males was 5–6 times larger than the harvest of adult females during 1976–1983. Since 1984, the sex ratio of harvested adult moose has been relatively stable, averaging 51.5% males and 48.5% females over the years, while the proportion of calves of all harvested moose has been around 54%.

Number of calves and adults shot per year within the Grimsö Wildlife Research Area.

During 1993–2001, 85 marked calves (27 females, 37 males and 21 calves of unknown sex) from 32 different females were checked for survival from August to April. Of the 85 calves, 19 (22.3%) were born as twins, and nine (10.6%) died of natural causes during their first winter, resulting in an average survival rate of 0.89 for all calves. Survival was not significantly related to birth year, birth year squared, twin or single, age of mother, or age of mother squared (P>0.2).

Survival analyses, excluding harvest-related mortality of adult moose, showed no significant effects of birth year (P=0.25), and no significant differences between sexes (P=0.22). Consequently, survival was analysed for males and females combined (Fig. 4). The survival rate for both sexes combined corresponds approximately to an age-specific survival of 0.95 for 1–13 year-olds, 0.9 for 14-year-olds, 0.85 for 15-year-olds and 0.6 for older animals (see Fig. 4). These were the natural survival rates used in the cohort analysis.

Estimated survival curve of adult moose. Dashed curves show 95% confidence limits and smoothed curve corresponds to the approximation used for the reconstruction.

The cohort analysis showed that the population size grew from an all-time low in 1973 with 5–6 moose per 1,000 ha to a peak in 1981 with 19 moose per 1,000 ha (Fig. 5). After 1981 the moose population decreased continuously during the 1980s and 1990s, reaching a level of approximately 7–8 moose per 1,000 ha in the late 1990s and early 2000s. The pattern of a sharp increase during the 1970s and a somewhat slower decline in the late 1980s follows the general pattern of the Swedish moose population (Cederlund & Markgren 1987, Lavsund & Sandegren 1989).

Population density (per 1,000 ha) estimated from the reconstruction. The population density was estimated as the total number of adults prior to the hunting season divided by the size of the research area (135 km2). The reconstruction results are compared to aerial counts (moose per 1,000 ha), pellet groups (per 100 m2) and hunter observations (observed moose per hunter per day). In the figure, the aerial counts have not been corrected for sightability.

Aerial counts did not show the same major increase during the 1970s as compared to the reconstructed population. However, aerial counts did not start until 1977 and the first year produced a relatively high estimate compared to aerial counts in the two following years. Despite the lack of a consistent pattern in the late 1970s, a significant correlation between the aerial counts and the reconstructed population size was found for the period 1977–2002 (r=0.69, N=11, P=0.02; Table 3). The estimated correlation between hunter observations and the aerial counts was also high (r=0.76), but not significant (N=6, P<0.10). However, estimates from these two methods were comparable for only six years. All other correlations between time-series estimates were low and non-significant.

Table 3.

Auto-correlations within time-series (lower diagonala) and the correlation between the time-series (upper diagonalb). Number of years in each analysis is shown in parentheses.

The time-series estimated by cohort analysis showed high auto-correlation (see Table 3), which was expected since the calculation of year-specific population size in the cohort analysis depends on the previous year's population size. The hunter observation data also showed high auto-correlation (see Table 3), which indicates that hunter observation can be used to follow population trends.

The estimated proportion of animals observed during the aerial count in 2006 was 0.73 (SD=0.03). This corresponds relatively well to the average proportion of animals with radio-collars observed in previous aerial counts (0.67, SD=0.22; see Table 2). It is therefore likely that the aerial counts prior to 2006 underestimated the true population size, and that the sightability of 84–95% suggested by Tärnhuvud (1988) generally produces underestimates of the true population size.

The daily defecation rate was estimated from the aerial survey and pellet group counts in 2002 and 2006. The number of moose within the research area, estimated from aerial counts, was 225 (in 2002) and 163 (in 2006). The estimated total number of pellet groups within the research area was 638,500 (SD=1,441) and 440,100 (SD=1,218), and the number of days over which the pellet groups had accumulated was 197 (SD=5.0) and 202 (SD=7.1) in 2002 and 2006, respectively. The estimates of daily defecation rates were 14.41 (SE=0.71) in 2002 and 13.36 (SE=0.83) in 2006.

Data from the older pellet group surveys were obtained from the southeastern quadrate of the research area whereas the new pellet group survey includes sample plots throughout the research area (see Fig. 1). Therefore it was possible to test whether the southeastern subregion was a representative part of the research area. This proved not to be the case. The moose density varied significantly (P<0.001) between different sub-regions (see Fig. 1), which is a probable explanation of the poor correspondence between the old pellet group survey and the other three measures of moose density (see Fig. 5), and also the poor correspondence between the old and the new pellet group survey in 1997 and 1998 (see Fig. 5). In the same analysis, we found no evidence (P=0.2) that the moose density increases (or decreases) in the peripheries of the research area, which means that the border effect is not greater (nor less) than could be expected from the size of the research area. The deviance divided by the degrees of freedom in the log-linear model was close to one (1.3), indicating that the number of pellet groups was randomly distributed throughout the research area (after adjusting for the class effects of year, subregion and the interaction of year and subregion).

During1976-1983, an excess of male adults were shot (see Fig. 3). Contrary to what we expected, however, the proportion of adult males was unrealistically high during 1973–1980 in the reconstructed population (Fig. 6). This suggests net immigration of males during these years.

Sex ratio of adults (males:female). Estimates from the reconstruction and aerial counts given as line and squares, respectively.

The reconstructed population was sensitive to the estimated age-specific survival of females >15 years old, previously estimated to be 0.6. This parameter was estimated from the survival analysis of radio-collared moose and had relatively large standard errors (see Fig. 4). A decrease in the age-specific survival of old females from 0.6 to 0.5 resulted in a change in the maximum density in 1981 from 19.0 to 22.8 moose per 1,000 ha, whereas an increase in old female survival from 0.6 to 0.7 resulted in a lowered density in 1981 of 17.6 moose per 1,000 ha.

The main cause of the sensitivity in 1981 was that one female was born in 1979 and shot at the age of 21 in 2000. In the cohort analysis, this single female corresponds to 54 calves born in 1979 given that the age-specific survival of old females is 0.6, and to 163 calves born in 1979 if the same age-specific survival is 0.5. Calibrating this parameter by the population size estimated from the aerial counts indicates that the age-specific survival of old females would be 0.52.

After 1990, the reconstructed moose population was based on incomplete cohorts and so the precision of the estimated moose density declines towards the end of the study. For the complete cohorts an additional female calf shot increases the total number of females by one, whereas if an additional female calf was shot in the cohorts born after 2002 the total number of females increases by 1.5. Consequently, the estimates were rather insensitive to the number of calves shot.

We also studied how a small change in the cumulative proportion shot (see Fig. 2) affected the reconstructed population. If the cumulative proportion shot was reduced by one percentage unit for all ages up to the age of 10, then the estimated number of females increased by 1% in 1997, 4% in 2002 and 5% in 2005. Consequently, the density estimates for 2002 and later were especially sensitive to the cumulative proportion shot, and the reconstructed population therefore may have been less accurate after 2002.

We have compared and evaluated the results from four different methods of estimating moose density in a long-term study: aerial counts, cohort analysis, pellet group surveys and hunter observations. All methods gave similar general results in terms of the size and development of the moose population during the 30-year period. However, different methods gave somewhat different absolute estimates of the population size and of the population development during the last 10 years of the study.

Aerial counts provide absolute estimates of population size at a certain point in time during the annual population cycle. This method may yield estimates of high accuracy given that relevant correction factors for the proportion of non-observed moose are available. A number of studies in North America have provided a wide range of estimates of sightability (mean: 0.71, range: 0.38–0.97; Timmermann 1993), or a sightability correction factor (SCF; 1.03–2.60) between studies, where the estimates depend on a number of factors such as type of aircraft (helicopter or fixed wing), type of forest, moose density, experience of crew, and snow and weather conditions (Timmermann & Buss 1998).

Using a number of aerial counts from different areas with radio-collared moose in Sweden during the 1980s, Tärnhuvud (1988) suggested correction factors that mainly were based on the type of weather conditions during aerial counts. However, the average proportion of radio-collared moose observed (67%) of the total number present (Ntotal=82) within our study area during aerial counts was clearly lower than the estimates presented by Tärnhuvud (1988) for good (95%) and intermediate (84%) snow conditions using a helicopter with experienced personnel. This is also corroborated by the fact that the sampling-resampling technique used at the end of the study period (2006) provided estimates of sightability (73%) that corresponded well with the average proportion of radio-collared moose observed over the whole period. We conclude that general application of sightability estimates should be avoided and that area and survey specific estimates should be developed.

A pellet group survey may constitute a good method for analysing population trends but merely gives an index of population density if not combined with an estimate of animal defecation rate.

Pellet group counting is a widely used method in the management of deer populations (Neff 1968, Timmermann 1974, Mayle et al. 1999). The accuracy of this method has been questioned (Fuller 1991, 1992, White 1992), but several studies have also shown realistic estimates and consistency in population trends and size between different independent methods (Neff 1968, Mandujano & Gallina 1995, McIntosh et al. 1995, Barnes 2001). Pellet group counts provide indices that can be calculated into absolute numbers of animals (Neff 1968, Timmerman 1974, Mayle et al. 1999). However, reliable interpretation of absolute numbers requires estimates of the rate of defecation (i.e. number of pellet groups produced per individual per day) and knowledge about the length of the period of pellet accumulation. However, the defecation rate of moose varies between studies and seems to depend on the type of habitat, forage quality and forage composition (Neff 1968, Andersen et al. 1992). A variation in defecation rate has a large impact on the calculated absolute number of individuals in an area.

Our results show that the defecation rate derived by comparing aerial counts and pellet group counts in the field for specific years (14 pellet groups per moose per day) is similar to, or in the lower range of, those reported for moose in captivity (Franzmann et al. 1976a,b, Oldemeyer & Franzmann 1981) and for some free-ranging populations (Jordan & Wolfe 1980, Joyal & Ricard 1986, Andersen et al. 1992, Jordan et al. 1993). Several authors (Andersen et al. 1992, Neff 1968, Person 2003) have shown that the defecation rate may depend on the herbivore population structure and the amount and quality of available forage. Therefore, pellet group surveys should be used with caution in areas where the population structure or the availability of different browsing species changes considerably between years or over time. In this study, moose population size did not change dramatically between the two aerial estimates (2002 and 2006) when the defecation rate was estimated. Neither did forestry practices nor age distribution nor composition of forest stands change dramatically during this 5-year period. This fact may be an important cause of the high correspondence between the two estimates.

Moreover, estimation from the old pellet group survey showed low correlation to other estimates of population size but the overall trend, estimated as 3-year averages, was similar to the trend in estimates from other methods. Our analysis of the data based on the new pellet group survey indicates the importance of distributing sampling units over the entire study area.

Solberg & Sæther (1999) and Sylvén (2000) have shown that hunter observations can be a useful tool for moose population management, since several important population measures can be derived from these: population size index, sex ratio and recruitment rate. Sylvén (2000) found that an area of 500 km2 is needed if the hunter observations are to be used as an accurate population index. However, our results indicate that hunter observations can also be a useful tool for estimating long-term population trends even in smaller areas (130 km2). However, Mysterud et al. (2007) found that hunter observations of red deer Cervus elaphus was also influenced by the harvest techniques. Thus, changes in harvest techniques within a hunting district as well as differences in harvest techniques among hunting districts have to be taken into account when evaluating hunter observations.

Sylvén (2000) did not include spatial auto-correlation between hunter observations in smaller areas. However, because hunting districts close to each other are more likely to have similar observation statistics, combining data from several neighbouring small districts may improve the precision of moose observation indexes for management purposes. Hunting districts close to each other are more likely to have similar observation statistics even though the area of each district is small. Our results indicate that hunter observations can be useful over a long time period in moderately sized areas, and if support was available from surrounding hunting districts, then this should be improved even further.

Reconstruction of historical population size can be a good complement for analysing trends in population development. However, this method depends on a number of assumptions and requires independent estimates of natural mortality in the study population. Unlike most other moose populations, natural mortality could be estimated from radio-collared animals for our study population, but was biased towards young and middle-aged individuals. Unfortunately, reconstructed population size proved very sensitive to survival estimates among old females (>15 years), resulting in relatively large confidence intervals for this age class, thereby contributing to uncertain estimates of population size.

Our cohort analysis rests on the assumption of no net immigration or emigration, an assumption that was shown to have been violated in the current study during the initial phase of the study period due to seasonal immigration of males during the mating and harvest season. However, this was probably a result of an extreme harvest strategy aimed at drastically reducing the male segment of the adult moose population. From the cohort analysis it was estimated that >80 adult males must have immigrated during the second half of the 1970s in order to have a proportion of males equal to 0.5 in the population, as estimated from aerial counts. However, in the years after 1980, we found no evidence for changes in net migration since the proportion of males was constant (see Fig. 6). Nevertheless, if no aerial counts had been performed during this period, erroneous conclusions about population size (see Fig. 5) and the adult sex ratio (see Fig. 6) may have been drawn. Furthermore, a cohort analysis is difficult to use as a management tool since the estimates of population size in the most recent years are the ones most uncertain.

Estimates of the reconstructed population size yielded consistently lower estimates than aerial counts. A number of factors may be invoked to explain these differences. For example, seasonal migration into the study area during the winter season may have resulted in higher winter estimates compared to the reconstructed population. However, earlier studies of radio-collared moose do not support this explanation (Cederlund & Sand 1991, 1994). The only logical explanation that we could find for the discrepancy between aerial counts and reconstructed population size is that the natural mortality of radio-collared moose has been overestimated and that reconstruction estimates were sensitive to the existence of a few old shot females in the data set. Calibrating the reconstruction with the aerial counts corrected for 70% sightability gives a natural mortality for old females of slightly less than 50%, which is lower than the average estimate for this age class, but well within the confidence interval of the survival analysis (see Fig. 4).

Our results emphasise the importance of developing area-specific estimates of sightability correction factors for aerial counts of moose. This could be done either by using radio-collared moose or by using the sampling-resampling technique of sample plots during field counts.

If pellet group counts are to be used as a measure of population size and trends, this method should in some years be combined with aerial counts to estimate site-specific rate of defecation.

Our results also indicate that hunter observations can be used to follow long-term population size development, even in relatively small management areas.

Lars Rönnegård gratefully recognises the grant from Sparbankstiftelsen Dalarna that made this study possible. The study was supported by the Swedish Environmental Protection Agency within the programme ‘Adaptive Management of Fish and Wildlife Populations’, the private foundations ‘Olle och Signhild Engkvists Stiftelser’ and Sveaskog (Stockholm, Sweden).

1.

2.

3.

4.

5.

6.

7.

8.

9.

10.

11.

12.

13.

14.

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

25.

26.

27.

28.

29.

30.

31.

32.

33.

34.

35.

36.

37.

38.

39.

40.

41.

42.

43.

44.

45.

46.

47.

48.

49.

50.

51.

52.

Page 2

Rumination behaviour is classically considered in behavioural ecology studies investigating activity budgets of free-ranging ruminants: a focal individual is ‘ruminating’, as opposed, for instance, to ‘feeding’ or ‘being vigilant’. ‘Feeding’ or ‘being vigilant’, however, are often further described at a finer scale. Information on bite rate or step rate while feeding (Ruckstuhl 1998, Ruckstuhl et al. 2003) or on scan rate or scan frequency while being vigilant (Hunter & Skinner 1998) is thus widespread in the literature. Contrary to this, very few field studies have investigated rumination at a finer scale, i.e. at the scale of a bolus (e.g. Ginnett & Demment 1997), and in particular in relation to life history strategies of free-ranging animals (e.g. Blanchard 2005). However, by decreasing particle size of ingested forages, and thereby exposing more surface area to microbial degradation (Pond et al. 1984, Pan et al. 2003), chewing plays a key role in digestion efficiency, and in particular during rumination (Trudell-Moore & White 1983, Chai et al. 1984). Accordingly, rumination parameters such as the number of chews per bolus or the bolus duration, have been extensively investigated in agricultural sciences. Numerous studies on livestock reported effects of forage nutritional characteristics on rumination behaviour (Gibb et al. 1999, Tafaj et al. 2005a), and others reported effects of rumination behaviour on digestion efficiency (Domingue et al. 1991). Hence, despite clear evidence for their direct importance in animal feeding biology, the investigation of rumination parameters surprisingly seems limited to applied agricultural research (but see Gross et al. 1995, 1996, Ginnett & Demment 1997, Blanchard 2005).

The goal of our study was to investigate the seasonal effects on rumination parameters in free-ranging impalas Aepyceros melampus, a dimorphic African ruminant experiencing a strong seasonality in climate and food quality. We tested two predictions. Forage quality affects rumination parameters (Pérez-Barbería & Gordon 1998). In particular, more fibrous food requires more chewing. Using experimentally controlled diets, several studies on cattle reported a positive influence of fibre content on the chew number and/or on bolus duration (Moon et al. 2004, Tafaj et al. 2005a). Thus, because herbivore diet quality, including that of impalas (Skinner et al. 1983, Meissner et al. 1996), is lower during the dry than during the rainy season, our first prediction was an increase in bolus duration and in the chew number during rumination in the dry season as compared to the rainy season. Impalas are mixed feeders (Hofmann 1989) known to exhibit great dietary flexibility (Meissner et al. 1996, Sponheimer et al. 2003). In the rainy season, impalas mostly graze, whereas their food intake is more balanced between grazing and browsing in the dry season (Skinner et al. 1983, Klein & Fairall 1986, Meissner et al. 1996, Wronski 2002). Because plant characteristics directly impact on rumination parameters, our second prediction was a decrease in the variability in chew number and bolus duration when animals mostly fed on a single type of food, i.e. when grazing during the rainy season. Overall, we thus expected lower average values and lower variability for chew number and bolus duration in the rainy season than during the dry season. If confirmed, these results could promote future applied studies on rumination at a fine scale. A seasonal variation in rumination parameters, suggesting an effect of food quality, is the first step before investigating to which extent inter-year or inter-population fluctuations of food quality, and thus, ultimately, of population performance, may be assessed by variation in chew number or bolus duration. We also recorded the effect of sex on rumination parameters as sexual dimorphism in body size is likely to lead to differences in digestive efficiencies, and thus potentially to compensatory behaviour for the smaller sex (Ruckstuhl & Neuhaus 2002), such as increasing mastication investment (even at an intra-specific scale; Gross et al. 1995, 1996, Ginnett & Demment 1997).

Hwange National Park is located on the northwestern border of Zimbabwe (19°00′S, 26°30′E) and covers an area of ca 15,000 km2. Vegetation is typical of southern African, i.e. dystrophic wooded savannas with patches of grasslands (Rogers 1993). Altitude varies from 800 m to 1,100 m a.s.l. The long-term annual rainfall average is 606 mm with most rain falling between November-April. In the Hwange system, young impala are generally born around the end of November or early December. Lactation generally lasts until early April, when adult males start to exhibit rutting behaviours that last until early June, with a peak in May. Our study was carried out in the Main Camp region of the park, where impala density is ca 1 individual/km2 (S. Chamaillé-Jammes, M. Valeix, H. Fritz, M. Bourgarel & S. Le Bel, unpubl. data for the Zimbabwe National Parks and Wildlife Management Authority).

The data were collected in 2005 during two 10-day periods, one during 12–22 February in the rainy season and one during 3–13 September in the dry season. Using 10×42 binoculars, a single observer (PB) performed all the observations from an open-top car. We only focused on adults. Impalas were habituated to cars, and easy to observe. Most of the observations took place from 20–50 m. Focal individuals were chosen according to head orientation, since the face had to be clearly visible in order to record jaw movements. Each observation began with the regurgitation of a bolus chosen randomly, and lasted until the fifth bolus was swallowed. We recorded the amount of time required to process five boli using a stopwatch, and the total number of chews performed during the focal (Blanchard 2005). Observations were discarded if the focal individual stopped chewing for at least five seconds. We also recorded the sex (males have horns whereas females do not).

Individuals were not captured or marked as part of this study. Therefore, as impalas were not individually recognisable, we may have observed the same animal more than once (although not on the same day). We performed a total of 102 observations: 40 and 32 females observed in the rainy and dry seasons, respectively (out of respectively 67 and 63 adult females in the studied population), and 16 and 14 males observed in the rainy and dry seasons, respectively (out of respectively 25 and 24 adult males). Therefore, by observing about the same proportion of individuals for each sex, we avoided increasing the pseudoreplication problem for one sex in respect to the other.

We used linear mixed models (Pinheiro & Bates 2000) to investigate the effects of sex and season on both the number of chews and the time required to process five boli. When finding a group of individuals ruminating, we often performed several observations within the same group. Therefore, we included ‘group identity’ as a random factor in the analysis in order to control for the non-independence between these observations. To investigate the sources of variation in the number of chews and the total duration of five boli, we first tested the effect of the two-way interaction (sex by season) by testing the difference in log-likelihood between the models with and without the interaction. We then removed the non-significant interaction and successively withdrew each of the two main factors, testing for their significance by comparing the difference in log-likelihood between the models with and without each of the factors.

We compared the variation in rumination parameters using the coefficient of variation (CV), expressed for small samples as CV=(1+1/(4×N))×(standard deviation/mean)×100 (Sokal & Rohlf 1995), with N for the sample size. All statistical analyses were done using R software (R Development Core Team 2005).

Season clearly affected fine-scale rumination patterns. From the rainy to the dry season, impalas increased both the number of chews performed per bolus (240.3 vs 268.6 chews for five boli in the rainy and dry seasons, respectively; likelihood ratio=11.5, difference in df=1, P<0.001; Fig. 1) and the duration of a bolus (196.1 and 247.0 seconds for five boli during the rainy and dry seasons, respectively; likelihood ratio=35.0, difference in df=1, P<0.001; Fig. 2), irrespective of their sex (number of chews for five boli:likelihood ratio=1.9, difference in df=1, P=0.17 and interaction sex*season: likelihood ratio=1.6, difference in df=1, P=0.21; duration of five boli:likelihood ratio=2.1, difference in df=1, P=0.14 and interaction sex*season:likelihood ratio=0.72, difference in df=1, P=0.40).

Number of chews performed to process five boli in the rainy (N=56) and dry (N=46) seasons, respectively, for impalas observed in Hwange National Park. The line across the box indicates the median. The box represents the interquartile range that contains the 50% of values. The whiskers are lines that extend from the box to the highest and lowest values, excluding outliers. Outliers (values 1.5–3 box lengths from the upper or lower edge of the box) are represented by open circles.

Duration (in seconds) of five boli in the rainy (N=56) and dry (N=46) seasons, respectively, for impalas observed in Hwange National Park. The line across the box indicates the median. The box represents the interquartile range that contains the 50% of values. The whiskers are lines that extend from the box to the highest and lowest values, excluding outliers. Outliers (values 1.5–3 box lengths from the upper or lower edge of the box) are represented by open circles.

The CV in the recorded parameters were affected by both sex and season (Table 1). The CV in the number of chews increased in dry season, as predicted by our second prediction, for males (8.2 and 12.1% in the rainy and dry seasons, respectively) as for females (9.1 and 14.5% in the rainy and dry seasons, respectively). However, sex impacted on the effect of season on the CV of boli duration, with an increase in dry season for males (11.5% as compared to 7.1% in the rainy season), but not for females (12.4% as compared to 12.2% in the rainy season).

Table 1.

Coefficients of variation (CV) in the number of chews and the time required (in seconds) to process five boli according to sex and season in impalas, Hwange National Park.

Whereas behaviours such as foraging or vigilance are extensively investigated in free-ranging ruminants both at the scale of the time budget and at a finer scale (e.g. records of bite rate or scan rate), rumination at fine scale (i.e. at the bolus scale) remains largely overlooked in literature, despite its particular importance for ruminant feeding ecology. Here, we focused on seasonal variation of rumination patterns in free-ranging impalas. Our data suggest that sex and season impacted on chew number and bolus duration. Male and female impala increased both chew number and bolus duration while ruminating in the dry season as compared to the rainy season, which is consistent with our first prediction based on previous studies reporting a negative effect of food quality on the average values of these rumination parameters (Moon et al. 2004, Tafaj et al. 2005a), and a diet of better quality in the rainy season compared to the dry season for impalas (Skinner et al. 1983, Meissner et al. 1996). Sex had no influence on the average values of chew number or bolus duration. Sexual differences in body size are likely to lead to differences in feeding behaviour, including rumination parameters (Gross et al. 1995, 1996, Ginnett & Demment 1997). However, the sexual dimorphism displayed by our studied animals (about 20%; M. Bourgarel & H. Fritz, unpubl. data) was probably too small compared to those reported by previous studies (about 135% for Nubian Ibex Capra ibex nubiana; Gross et al. 1995) to easily detect potential sexual differences in rumination patterns. The CV in the number of chews increased for both sexes between the rainy season and the dry season, whereas only males increased CV in bolus duration. Once again, it is broadly consistent with our prediction that the greater variability in the food items consumed in the dry season (Skinner et al. 1983, Meissner et al. 1996, Wronski 2002) should lead to an increase in the variability in the chew number and in bolus duration from the rainy season to the dry season.

Rumination parameters reflect the physical and chemical characteristics of previously ingested forages (Pérez-Barbería & Gordon 1998). In particular, more fibrous forages require higher chewing effort. In an indoor trial conducted with red deer Cervus elaphus fed either fresh perennial ryegrass Lolium perenne or chicory Chicorium intybus, Hoskin et al. (1995) reported higher chewing effort, including chew number, for deer fed with the more fibrous ryegrass. Moon et al. (2004) reported longer bolus duration in dairy cows fed with experimental diets increasing in fibre concentration. Here, we report that both bolus duration and chew number increased in the dry season as compared to the rainy season, irrespective of sex. Impalas are mixed feeders (Hofmann 1989), moving to more woody browse in the dry season, when grass quality becomes too low (Klein & Fairall 1986, Meissner et al. 1996, Wronski 2002). Therefore, the decrease in the quality of the forages ingested by impala during the dry season (Skinner et al. 1983, Meissner et al. 1996) probably results in the longer bolus duration and in the increase in the chew number we report as compared to the rainy season.

Food quality, and in particular fibre content, has also been reported to affect the total time devoted to rumination (Moon et al. 2004), and intake rate may impact on rumination parameters (Bae et al. 1979, Tafaj et al. 2005b). Further studies performed at both fine and large scales for a single population, could improve the understanding of the relationships between food quality, food quantity, time budget and rumination parameters. Further, because many of the indices used by managers to assess nutritional status of free-ranging populations require the capture or the killing of the animal and/or are expensive or irrelevant (Blanchard et al. 2003), future studies should investigate the reliability of rumination parameters in assessing the variability of food quality for a given species according to season, year for a given season, locality or population.

We report that season broadly impacted on the variability in rumination parameters, probably explained by a broader range of food items selected in the dry season as compared to the rainy season, as suggested by previous studies (Skinner et al. 1983, Meissner et al. 1996, Wronski 2002). This explanation is also consistent with personal observations in the study area where impalas were mostly seen grazing during the wet season, whereas they relied on more various sources of food in the dry season, including browse and Acacia pods but also dried grass, which may form the bulk of their rumen fill. The increase in the variability in the food items consumed during the dry season therefore probably explains the increase in the variability in the chew number reported for both sexes, with more chews being performed when ingesting lower quality items (i.e. lower than the average forage quality, already lower than mean forage quality in rainy season).

The CV in bolus duration was lower in the rainy season than in the dry season for males, as predicted by our second prediction, but the CV in female bolus duration was not affected by season. Females showed about the same variability in both seasons, with more variability than males in the rainy season. Differences in the reproductive status of females may explain this result. Some of the females observed during the rainy season (i.e. the rearing period) were probably lactating while others probably were not (48 juveniles were present in the study area together with a total of 67 adult females in the rainy season). In the dry season, however, all young were weaned so that all adult females had the same status. The heterogeneity in reproductive status in the rainy season may have been translated into a heterogeneity in bolus duration. Blanchard (2005), focusing on the rearing period, investigated variation in rumination parameters as a function of presence/absence of lamb in bighorn sheep ewes Ovis canadensis, and suggested that lactating females increased chewing effort, in response to an increased energetic demand and risk of predation, as compared to yeld females. In order to avoid foraging longer than yeld females to meet the energetic costs of lactation, and thus to enjoy the benefit of group foraging (Kie 1999, Sevi et al. 1999) through synchronisation of activities (Ruckstuhl 1998), lactating females may compensate for an increasing intake rate by increasing chewing effort during rumination (Blanchard 2005). In the present study, lactating females may decrease bolus duration as compared to yeld females, in order to process more boli during the same amount of time spent ruminating. This would mean that lactating females increase rumination speed, as reported for bighorn sheep (Blanchard 2005).

Future studies should investigate the impact of reproductive status on foraging behaviour on marked female impalas, as this interpretation remains speculative. Also, if lactating females ruminate faster than yeld females (Blanchard 2005), the CV in rumination speed among females during the rearing period should crudely scale with the ratio of lactating female to yeld female, i.e. the young:female ratio, an index often used by managers to infer ungulate population dynamics (Bonenfant et al. 2005).

Rumination, particularly at a fine scale, seems to have often been overlooked in field studies. More studies are thus required to improve our understanding of the relationship between rumination behaviour at both large and fine scales (Ginnett & Demment 1997), food characteristics and population dynamics. Because ungulate population dynamics is strongly influenced by changes in density and climatic conditions (Sæther 1997, Gaillard et al. 1998), mostly through their effect on food availability and quality, a proxy of nutritional status would be useful for managers interested in wildlife demography (Blanchard et al. 2003). Future studies, based on long-term data sets of marked free-ranging individuals should investigate to which extent measures as easy to record as chew number or bolus duration could be used to assess factors as important as resource properties and thus, ultimately, population performance in ruminants.

this project was developed within the HERD Project (CIRAD/CNRS). We thank Simon Chamaillé-Jammes, Jean-Michel Gaillard and two anonymous referees for comments on the manuscript. We are grateful to Zimbabwe Parks and Wildlife Management Authority for their support, and to the CNRS-NRF PICS programme ‘Plant-herbivore dynamics in changing environments - developing appropriate models for adaptive management’ for funding. Many thanks also to Sébastien Le Bel (CIRAD-Zimbabwe) for facilitating the operations.

1.

2.

3.

4.

5.

6.

7.

8.

9.

10.

11.

12.

13.

14.

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

25.

26.

27.

28.

29.

30.

31.

32.

33.

34.

35.

36.

Page 3

Wildfowl Anatidae are sensitive to human activities during the hunting season (e.g. Bell & Owen 1990, Tamisier et al. 2003, Triplet et al. 2003, Blanc et al. 2006) and often gather in large numbers in disturbance-free protected areas (Cox & Afton 1997, Fox & Madsen 1997, Madsen et al. 1998). Refuges can also benefit hunters when they are used by birds as roosts. Ducks departing from refuges for nocturnal feeding or other activities are more exposed to hunting activities (Bellrose 1954, Griffith 1957, Anonymous 1961, Bell & Owen 1990, Mathevet & Tamisier 2002). Some duck species commute two times within a 24-hour period, moving from day-roosts to nocturnal foraging areas. These two habitats are considered their functional unit (Tamisier 1978). Regular commuting patterns during dusk and dawn increase the ducks' vulnerability to hunting. Different studies have shown that the average distance ducks fly between roosting and foraging areas varies among species and environmental conditions (e.g. adverse weather) and can range from 0.8-50 km (Fog 1968, Frazer et al. 1990). However, it seems intuitive that ducks would minimise travel distance given the energetic cost of flight. Distance between day-roosts and nocturnal foraging habitats is therefore generally limited to a few kilometres (Tamisier & Tamisier 1981, Jorde et al. 1983, Guillemain et al. 2002, Legagneux et al. 2008). It is hypothesised that hunting areas adjacent to day-roosts, whether protected or not, experience greater hunting harvests. One study in Camargue in the south of France demonstrated that habitats protected from hunting and comprising a major roost influenced surrounding land management practices, which became more hunting-orientated. As a result, the price of hunting leases in the vicinity of the reserve increased to more than 1800 euros per hunter per year (Mathevet & Tamisier 2002). More information concerning duck movements is necessary to assess the distance travelled in regards to land management procedures. Knowledge of duck movements could influence decisions to protect or manage nocturnal foraging habitats in addition to day-roosts, or to establish buffer areas around reserves (Fox & Madsen 1997, Rodgers & Smith 1997). This information could also be important for understanding the local dispersal of propagules (seeds or invertebrates (Green et al. 2002)) or the spread of diseases by ducks, such as avian influenza viruses (Gauthier-Clerc et al. 2007, Saad et al. 2007 for teal Anas crecca).

Our research evaluated hunters' harvests from various private hunting estates in the Camargue, southern France, to test the relationship between hunters' harvests and distance from hunting estates and nearest day-roosts.

The Camargue encompasses 145,000 ha of the Rhône Delta (see Tamisier & Dehorter 1999 for a general description). The Camargue is a site of international importance to waterbirds and exceeds the criterion of international importance for several duck species (Scott & Rose 1996). More than 85,000 ha of the delta are wetlands, and 78% of the wetlands are divided into approximately 180 private estates that integrate hunting activities (Dehorter & Tamisier 1996). As part of a research programme on hunting practices and wildfowl population dynamics in the Camargue (Mondain-Monval et al. 2006), hunting bags are collected annually for as many estates as possible. In the present study, we used data from 45 estates which provided duck bags for at least three hunting seasons, together with an indication of hunting pressure (number of hunter-days). Data ranged from the 1926–1927 to the 2004–2005 hunting seasons. However, most of the estates provided data sets beginning in the 1980s or 1990s (Appendix 1), which allows for comparison by site and also relates to the period when the list of duck day-roosts was established by aerial counts (see below). The number of annual data ranged from 3-68 years between the 45 estates (median=9 years). The inclusion of the 13 estates with <5 years of data led to a 22-33% increase in the variance of the daily bag per hunter, depending on duck species, compared to the reduced data set made of the 32 estates with ≥5 years of data. However, the results concerning the relationship between daily bag and distance from the nearest roost were exactly the same in terms of significant and non-significant patterns. The complete 45 estate data set was thus used to cover the largest geographical area possible. Data sets were interrupted in some cases, due to changes in landlords and/or hunting lease holders. Three data sets also abruptly ended when the hunting estates became natural reserves. In each estate, the average daily bag per duck species per hunter was calculated for each year, and then averaged for all available years (see Appendix 1).

Although owners of hunting estates sometimes manage water levels, salinity and frequency of disturbance so as to promote duck roosting within their property, most ducks spent daylight hours in the >20,000 ha of protected areas in the Camargue. Based on long-term (1964–65 to 1994–95) aerial monitoring of wintering wildfowl, Tamisier & Dehorter (1999: 352–353) provided a list of the main day-roosts for the different duck species based on the average number of birds of each species they hosted over the monitoring period. We determined the central point of each of these roosts for each species through GIS software (Arcview 3.1 GIS, ESRI 1998). The central point of each of the 45 hunting estates was also computed using the same methodology, allowing calculation of the distance in kilometres between each estate and the nearest day-roost for each duck species (NB: the nearest day-roost was possibly different for different duck species in a given hunting estate). This method therefore established a distance equal to zero in the case of day-roosts within hunting estates.

The distribution of hunting bag values was non-normal (Kolmogorov-Smirnov tests: all d values >0.20, all P<0.01), and usual transformations (such as logarithm, square-root, etc) did not solve this problem. Given the distribution of the data, we used maximum-likelihood ratio statistics based on a quasi-Poisson distribution of the variables. The relationship between hunting bag and distance from the nearest roost in each duck species was analysed with a non-linear regression model, using a negative exponential relationship. Pintail Anas acuta, red-crested pochard Netta rufina and tufted duck Aythya fuligula were not included in the analyses, because they were either absent or represented only a small part of the total bag of the 45 studied hunting estates (0.005-0.063 individuals per hunter per day on average on 20 estates, and 25 estates with no single individual ever harvested in tufted duck). The data set therefore included mallard Anas platyrhynchos, teal, gadwall Anas strepera, shoveler Anas clypeata, wigeon Anas penelope and pochard Aythya ferina hunting bags.

All six duck species except mallard showed the same pattern of decreasing daily bag per hunter with increasing distance from the nearest day-roost, although this was non-significant in shoveler and marginally so in wigeon (Fig. 1, Table 1). All species' hunting bags were highly variable among estates close to day-roosts. The range was as wide as 0.1-4.5 teal per hunter per day for estates <2 km from a day-roost.

Relationship between daily bag per hunter and distance between hunting estate and nearest day-roost in the six duck species. See Table 1 for statistics. Regression curves are shown where significant.

Table 1.

Non-linear regressions of average daily hunting bags per hunter and the distance between hunting estates and the nearest day-roost, per duck species. The equation is given where model fit was significant.

In all cases, the decrease in hunting bag with increasing distance was very rapid, since estates >5 km from a roost harvested very limited numbers of individuals of a given species (see Fig. 1). Using the significant trends for the teal, gadwall and po-chard models (see Table 1), we calculated the average expected daily bag per kilometre of distance from the nearest roost, and determined the threshold distance above which this value no longer exceeded 10% of the expected bag within the first kilometre. For teal, gadwall and pochard, the threshold values were 9, 5 and 3 kilometres, respectively (i.e. an estate located more than 3 km from any pochard day-roost could not expect a daily bag per hunter higher than 10% of the daily bag of an estate within 1 km from such a roost).

All daily hunting bags decreased (though a significant threshold was not reached in wigeon and shoveler) in an exponential manner with increasing distance from the nearest day-roost with the exception of mallard. This is consistent with central place foraging and refuge theories, in which animals try to limit their travelling distance between a central protected area and their feeding areas (e.g. review in Stephens & Krebs 1986, Cox & Afton 1996). The fact that daily bags differed among estates <1 kilometre from a roost probably reflects management options as well as the differences in hunting estate sizes. Estates are managed and decisions are made depending on the priority species that the managers aim to attract (e.g. by selecting different water regimes, water levels or salinity) and the habitat constraints (e.g. water salinity). For example, if an estate is close to both a teal and a pochard day-roost, the management options and habitat will favour one of the species. Furthermore, some hunting estates adjacent to or encompassing day-roosts allow shooting during a few daylight hours during the morning (as opposed to dusk and dawn shooting during duck commuting flights between day-roost and nocturnal feeding areas), which is obviously a more efficient hunting technique. Depending on species, ducks virtually disappeared from hunting bags in estates located >3–10 km from the day-roosts and in all cases were mostly shot within 2–3 km range. It is interesting to note that these distances between roosts and selected foraging areas are within the same ranges determined from earlier studies using radio-tracking methodologies. Radio-tracking studies suggest that ducks generally travel from a few hundred metres up to 10 km to feed, this range being species-specific (Tamisier & Tamisier 1981, Jorde et al. 1983, Frazer et al. 1990, Cox & Afton 1996, Guillemain et al. 2002, Legagneux et al. 2008). The present study suggests that larger home ranges (including up to 50 km travels) (e.g. Bellrose 1954, Fog 1968, references in Jorde et al. 1983) probably represented extreme travel distances or areas where no suitable habitat was available to the birds within their preferred travel distance. Some of these previous studies considered ring recoveries, indicating that the distance travelled may have been over several days.

The insignificant relationship between daily bag and distance from roost in mallard was unexpected, but could be explained by hunting management practices for this species. Most mallards in the Camargue are harvested in the beginning of the season and many of which are young individuals hatched on the estate during the previous spring or raised and released in spring and summer, a common practice in France. The exact number of such released mallards is not known, but has been estimated to be around 20–30,000 individuals in the Camargue (Tamisier 2004). The uptake of locally-released or locally-hatched individuals within the hunting estates at the beginning of the season may therefore hide a potential relationship between hunting bag and distance from a roost that may occur in natural wild conditions. This practice is, however, unlikely to have affected the uptake of other ducks, since the average daily bag in estates releasing (N=10 properties) or not releasing mallard (N=35 properties) did not differ significantly in any of the species (t-tests, all t absolute values <1.33, all P>0.1917). The lack of significance in shoveler and wigeon was essentially due to the high variance in hunting bags for estates close to day-roosts (0-0.8 wigeon per hunter per day and 0-1.3 shoveler per hunter per day for estates <1 km from nearest roost). However, the wigeon and shoveler average daily hunting bag also tended to decrease exponentially with increasing distance from the roost (see Table 1).

From a hunting management point of view, this study suggests that management practices leading to roost establishments within the hunting estate may indeed lead to larger daily bags (especially if this allows diurnal shooting). Similarly, the present results show that the higher prices of hunting leases around protected duck day-roosts observed by Mathevet & Tamisier (2002) can be explained by higher expected shooting opportunities in adjacent estates. However, private properties are not affected by roosts located >2–3 km away. On the other hand, we suggest that where the protection of duck species is a priority over hunting considerations, the establishment of hunting-free buffer areas around core wildfowl refuges (e.g. Fox & Madsen 1997) could be an effective strategy since the radius of such buffer zones may be as little as a few kilometres (as opposed to 20–50 kilometres in some earlier studies) for most species.

If they do disperse seeds, invertebrate eggs or disease locally during their feeding travels, the management implications of our results are that dabbling and diving ducks are most likely to do so over short distances during their daily movements. This does not take into consideration the long distances they may cover during migration which can be associated with long-distance dispersal of seeds or invertebrates (Green & Figuerola 2005). This conclusion is only based on indirect information provided by birds shot by hunters, therefore future research using electronic devices to determine actual nocturnal movements of ducks should be carried out. This may also allow for future testing to determine if wintering wildfowl actually have longer daily travel distances in order to avoid hunted estates and move between protected areas as observed in the past in western France (Guillemain et al. 2002).

we are grateful to the many Camargue landowners and holders of hunting leases who provided information on their annual hunting bags. We would also like to thank Michel Gauthier-Clerc, Andy Green, Jesper Madsen, Jean-Baptiste Mouronval and Lisa Ernoul for their useful comments on an earlier version of this manuscript as well as Hervé Fritz for valuable statistical advice. Anne-Laure Brochet is funded by a Doctoral grant from Office National de la Chasse et de la Faune Sauvage, and receives additional funding from a research agreement between ONCFS, the Tour du Valat, and the Laboratoire de Biométrie et de Biologie Evolutive UMR 5558 CNRS Université Lyon 1 in France, together with the Doñana Biological Station (CSIC) in Spain. This work also received funding from the Agence Nationale de la Recherche through the Santé Environnement - Santé Travail scheme (contract number 2006-SEST-22).

1.

2.

3.

4.

5.

6.

7.

8.

9.

10.

11.

12.

13.

14.

15.

16.

17.

18.

19.

20.

21.

22.

23.

24.

25.

26.

27.

28.

29.

30.

31.

Hunting pressure (average annual number of hunter-days) and hunting bag (average number of pieces of game per hunter per day) for each of the 45 hunting estates studied in the Camargue, in the south of France. Values are means±SE. The names of the estates are not provided because of confidentiality agreements with the landowners and holders of hunting leases.

Page 4

Ungulates are characterised by a U-shaped mortality curve with low newborn survival (Caughley 1966). Variations in juvenile survival are much higher than variations in adult survival, so that “high yearly variability in juvenile survival may play a predominant role in population dynamics” (Gaillard et al. 1998). In several species of temperate ungulates, fawn mortality is concentrated in the perinatal period and is related to the fawn's body weight at birth (e.g. Coulson et al. 2003).

In order to get unbiased estimates of important phenotypic traits, such as birth date and birth weight, we need to know the age of captured newborns. This datum allows us to evaluate the birth date and to obtain an estimate of birth weight, whether a linear growth function may be estimated using multiple recaptures (e.g. Pelliccioni et al. 2004).

Fallow deer Dama dama adopt a hiding strategy and fawns lie hidden in vegetation, isolated from conspecifics, for about 4–5 days at which time they become able to follow their mothers. During the short hiding period, the mother visits the fawn several times a day (more often at twilight) to nurse and move it to a new bed site (Chapman & Chapman 1975).

Since 1988, the Istituto Nazionale per la Fauna Selvatica has developed a long-term monitoring programme for the fallow deer population in the Preserve of Castelporziano (Roma, Italy; Focardi et al. 2001). As part of this programme, we have captured and marked neonates since 2000 in order to investigate juvenile survival in this species.

For several species of ungulates it is possible to estimate the age of newborn fawns (white-tailed deer Odocoileus virginianus: Haugen & Speake 1958; roe deer Capreolus capreolus: Jullien et al. 1992; fallow deer: Pélabon 1995). Pélabon (1995) concluded that hoof abrasion, status of the umbilical cord and coat appearance were unlike to yield accurate age estimates for fallow deer fawns >2 days old. The aim of our study was to evaluate if there is a set of traits which can be used to reliably estimate the age of fawns during the hiding phase and which can be collected easily during captures in natural conditions.

The simplicity of the method is very important in order to reduce the handling time of fawns. By minimising stress at capture, we should get unbiased data on survival and growth. Because of this requirement, we excluded the use of blood samples and complex biometrical measurements. In the light of experience made on roe deer using Jullien's et al. (1992) ageing method (Pelliccioni et al. 2004), we adopted a similar approach to calibrate neonate age of fallow deer.

Observations were made in the deer research facility ‘Antonio Servadei’ (University of Udine, Pagnacco, Italy) during two experimental periods in 2003 and 2007. In both years, 18 individually-marked adult females (different individuals in the two study years), were captured and weighed to determine their pregnancy status. The animals were divided into two groups. Each group was allocated to a 1.0-ha paddock and was allowed to graze a pasture dominated by tall fescue Festuca arundinacea, with the presence of foxtail bristlegrass Setaria italica, hairy crabgrass Digitaria sanguinalis, meadow buttercup Ranunculus acer and nettles Urtica spp. Observations were made from vantage points during 28 May-13 August, 2003, and during 7 June-24 July, 2007.

In some cases, parturition was observed or the doe showed signs of recent delivery. Females were always checked morning and evening from a distance, so that we could establish if parturition had occurred.

Most first captures were performed after observation of the fawn, but for several recaptures, fawns were detected upon systematic search.

Fawns were captured and inspected each day after birth when it was possible to capture them using a long-handled (150 cm) circular net (diameter 80 cm). To avoid desertion by the mother, we always waited some hours after delivery before capturing the newborn and latex gloves were always used when handling animals. During the first capture, fawns were marked with a numbered plastic ear-tag.

At capture the following parameters were recorded: status of the umbilical cord, behavioural reactions during capture and release, reaction to ear-tagging and/or handling and fawn's resting position at detection, before capture.

The umbilical cord was described by its status (protruding or in regression), appearance (wet, withered), solidity (soft, hard) and colour (pinkish, dark brown, black). Blood was defined as fresh, dry or absent. According to Jullien et al. (1992), we distinguished between three resting positions: 1) curled-up with head tucked against the flank, either with the belly down or lying on the flank (position lovée), 2) lying as an adult with the rear legs behind the belly, fore legs stretched along the ground and the head erect (position couchée) and 3) moving/standing up.

Like Pélabon (1995), we also used correspondence analysis (Andersen 1994) to evaluate whether or not recorded traits can provide a reliable estimate of fawn age. Using correspondence analysis (PROC CORRESP; SAS Institute, Inc. 2000) makes it possible to explore the structure of a multi-way contingency table (for such a table the usual χ2-test is not appropriate). The derived structure is illustrated in a point diagram, which represents the categories of the studied variables on a plane. A profile is a vector obtained by dividing a cell frequency for the totals of its row or column. It is possible to compute both observed and expected profiles under the null hypothesis of no association between variables. This method allows us to reduce the number of variables, since only the two largest eigenvalues of the residual matrix (Dimension 1 and 2, respectively) usually explain a large part of sample variance. In the interpretation of the plot, three main rules have to be followed (cf. Focardi et al. 2000):

• 1)  the categories near the origin have a frequency distribution close to the expected one; the larger the distance from the origin, the larger the residuals are;

• 2)  to compare categories of the same variable it is possible to use the between-point distance; the larger the distance, the greater the difference;

• 3)  the association between categories of different variables cannot be compared with distances, but the orientation of the line connecting the point to the origin can be used; categories with similar orientation are positively associated and categories with angular differences around π are negatively associated. Angles around π/2 between two categories denote that the observed frequencies are those expected under independence.

The first births were recorded on the 18 June in 2003 and on 11 June in 2007, while the last births were recorded on 8 August and 18 July in 2003 and 2007, respectively. During the fawning period, 15 (six males and nine females) and 13 fawns (eight males and five females) were born in 2003 and 2007, respectively. In 2003, we excluded two fawns from our analysis because they died prematurely and three were excluded because they were born after the observation period ended. In 2007, we excluded five fawns; one because it died during the experiment and four because they were born before the observation period started.

Thus, a total of 18 newborn fawns (eight females and 10 males) were monitored in the two years and data were recorded systematically during the first three days of life; but on Days 4 and 5, we were able to capture 13 and nine fawns only.

Only in one case (a very light newborn fawn weighing 3 kg) were we able to recapture a fawn after Day 5, but data for later days were omitted from the present analysis.

Inspection of the data showed systematic variations in the recorded traits as a function of age (Table 1). During the first and sometimes even the second day of life, we were able to capture the fawns by hand; on later days, it was necessary for us to use the net because flushing distance increased. Observations made during the study showed that during the first two days flushing distance was lacking or very short, and the fawn froze when somebody got close. At the age of 3–4 days the flushing distance was ≥2 m, but it was still possible to capture the fawns using the net. For older fawns capture probability decreased very rapidly.

Table 1.

Distribution of the different categories of the studied variables in the first five days of life of newborn fallow deer, reared in the research facility at the University of Udine, Italy.

Sometimes the fawns emitted distress calls when marked or handled inducing some specific reaction in the mothers, which usually were observing from some distance away.

Dimension 1 and Dimension 2 accounted for 42.0 and 17.0% of total sample variance, respectively. The association of the categories of the different variables showed a clear difference among the five one-day classes of age (Fig. 1). Day 1 was very different from the other days, which were arranged in a series showing continuous variations in the associated traits. Day 1 was also characterised by a non-random distribution of traits and it was associated with a pinkish, fresh, wet and protruding umbilical cord. Day 2 was associated with ‘position lovée’, withered blood and weaving movements. Day 3 was characterised by a dark brown cord and the presence of distress calls. Interpretation of traits associated with Days 4 and 5 followed straightforwardly.

Categories of the studied variables plotted into the space defined by the two largest eigenvalues corrected by Greenacre adjustment. The numbers (1–5) represent the age (in days) of neonates. See text and Table 1 for definitions of the categories. To improve readability when two words overlapped, their position was shifted the minimal amount necessary to avoid overlapping text.

Between-individual variability is shown in Figure 2, and so are the coordinates of the five age classes (note that some symbols overlap). Comparing the angle of each individual point with the angle of the appropriate category of variable Day, one can see that even at the individual level the different ages are clearly differentiated, particularly during the first three days of life. The overlapping between Days 3 and 4 is, however, small. More overlapping is present between Days 4 and 5. There are no significant between-sex differences (Wilcoxon test: Z=−0.40, P=0.69 for Dimension 1, and Z=0.64, P=0.52 for Dimension 2). The between-day differences of cosine of the angles connecting each individual point to the origin is highly significant (Kruskal-Wallis ANOVA: χ42=63.8, P<0.0001).

Coordinates of the individual animals and the age of neonates (in days; 1–5) plotted into the space defined by the two largest eigenvalues, to show the differences among individuals of the same age. Different symbols refer to different ages (i.e.•= Day 1, *=Day 2;⋆=Day 3, ◊=Day 4 and □=Day 5). A line connecting categories of the variable Day to the origin allows an easier understanding of the discrepancies between each animals and the expected profile for each day. Note that some individual points overlap completely (because animals presented exactly the same features). True sample sizes for Days 1–5: 18, 18, 15, 13 and 9, respectively.

Our simple study shows that it is possible to reliably age fallow deer neonates during their first five days of life using a set of easy-to-collect variables. Our sample size was too small to extend our analysis further (for instance to compute probabilities of correct classification), but we believe that the bias introduced by systematic ageing errors, in future analysis of fallow deer demography under field conditions, will be very limited. Interestingly, this classification is valid for both sexes. As it happens for many qualitative traits, there is some degree of subjectivity in their definition, and it is clearly necessary that the members of the capture team standardise procedures of data collection.

On Day 1 fawn age is clearly identifiable (Pélabon 1995). In field studies performed in enclosures or in very open landscape (e.g. Birgersson et al. 1998, Mc Elligott et al. 2002), it is often possible to detect and capture fawns during their first day of life, when they can be aged ‘by eye’. The importance of having a suitable ageing methodology for fallow deer fawns was clearly evidenced by San José et al. (1999) who could not capture all fawns on their first day of life as they were working in a wild area. At Castelporziano, which is characterised by dense woods, only 9.2% (N=76) of fawns were captured on their first day of life, so a reliable age calibration is necessary (A. Galli & S. Focardi, pers. obs.).

According to Lent's (1974) classification, fallow deer is a species (e.g. like red deer Cervus elaphus) with a quite reduced hiding period. From an evolutionary point of view (Fisher et al. 2002) this trait is well explained by the preference of this cervid for relatively open habitats and by the tendency to form large groups (Focardi & Pecchioli 2005). Because does exhibit intermediate body size (45–50 kg) and give birth to one fawn only, neonates are born relatively large and in few days they can attain a size which allows them to follow their mothers. When fawns join a group they may attain a certain degree of protection against predators.

we thank N. Chapman for the final correction of the paper.

1.

2.

3.

4.

5.

6.

7.

8.

9.

10.

11.

12.

13.

14.

15.

16.

17.

Page 5

Wolves Canis lupus are coming back to Sweden. In 1998, at least 50 animals were estimated to be in Sweden, and the latest estimate is 128 wolves (Aronsson & Svensson 2007). The attitudes of the public have life and death consequences for wolves and may be part of the reason why the wolf popu-lation is not growing as fast as expected (Wabakken et al. 2001). Our goal in this article is to outline some factors that point to potential negative swings of opinion in Sweden and to discuss some proposed remedies based on our surveys. In 2005, a wolf immigrating into Sweden acted as if it had not read our research reports. The ‘Ringvattnet’ wolf attacked livestock and repeatedly visited a village in spite of attempts to scare it off. Our surveys show that the general public opinion supports hunting and killing wolves if 1) wolves do damage to live-stock or private animals, and if 2) a wolf loses its fear of humans (Ericsson et al. 2004). The ‘Ringvattnet’ wolf had done both and was predictably shot by the authorities.

Our meta-analysis of 38 attitude studies from around the world published between 1972 and 2000 shows majority support for wolves and restoration across studies (Williams et al. 2002). This support has been constant across wolf attitude studies for nearly 30 years. We believe that the major shift from negative to positive attitudes came as American and western European societies continued to urbanise and industrialise between the 1930s and the 1960s. Our recent surveys in Sweden show that support for the existence of wolves in Sweden is strong and widespread among the general public (Ericsson et al. 2006). Even a majority of hunters who live in the wolf areas support the right of wolves to exist (Ericsson & Heberlein 2003). Attitudes of the general public in Sweden towards wolves have been stable or have become even more positive between 1976 and 2001 (Ericsson & Heberlein 2003).

Though attitudes are usually stable, they are not immovable. Attitudes towards wolves may become more negative now that the wolves are back. It is easy to support hypothetical wolves, but real wolves, such as the ‘Ringvattnet wolf’ which may kill hunting dogs and livestock and threaten the public's sense of security. Both our research and the research of others show that people who live in wolf areas are less positive towards wolves than people who live in areas where wolves do not occur (Ericsson & Heberlein 2003, Ericsson et al. 2006, Karlsson & Sjöström 2007). Increasing wolf numbers can lead to more negative experiences which can lead to more negative attitudes.

Furthermore, lack of experience may be even more important for changing attitudes than experience itself. The results of our meta-analysis showed that in most studies >30% of the respondents reported no strong attitudes towards wolves (Williams et al. 2002). We also found this level of disinterest in our 2001 Swedish survey. But a single negative event could make this group change from neutral to negative. For example, Duda et al. (1998) found that the support for a proposed wolf restoration in the Adirondack Mountains region in the state of New York, USA, dropped from 76 to 46% in a single year following a proposal to reintroduce wolves. Subsequent studies performed by researchers at Cornell University showed that these attitudes remained negative two years later (Enck & Brown 2002).

The most dramatic change in Sweden has been a decline in hunter support for wolves. In 1976, 3/4 of the hunters and the public agreed that it was important to do something for wolves, and 60% of both of these groups supported artificial reintroduction of wolves (Andersson et al. 1977). At that time, the hunters were more positive than the general public in their support for a free-ranging wolf population (63 vs 51%), and in their support for an unrestricted population of wolves (59 vs 51%). Today, when real wolves have returned to Sweden, we found that only 40% of the hunters said they liked wolves compared to 61% of the general public (Ericsson & Heberlein 2003). So, today, hunters are much less likely than the general public (40 vs 71%) to say that the wolf population should increase. We believe that these changes occurred because 30 years ago hunters anticipated that wolves would show up in the mountains and in the reindeer Rangifer tarandus areas in the north rather in the southern forests and in the moose Alces alces hunting areas where the restoration actually happened. While hunters in Sweden compose about 3% of the total population and hunters in the wolf area compose 0.1% of the population between 16-65 years old, they represent an important interest group when it comes to wolves as they are directly affected. Hunters' annual licensing fee also helps fund wildlife research. Hunting is symbolically important in Sweden and hunters have nationwide political influence. Though small in numbers, this is not a group that can be easily ignored.

Attitude studies usually show that urban residents are more favourable in their attitudes towards wolves than are rural residents (Williams et al. 2002). This is not the case in Sweden where we found no statistically significant difference between rural and urban residents in their support for wolves (Ericsson & Heberlein 2003). In 2005, we (Heberlein & Ericsson 2005) took a closer look at urban residents. We asked if respondents living in cities were multigenerational urban residents. It turned out that those who were sons and daughters of parents who themselves had been born and were raised in cities had the most negative attitudes towards wolves (Heberlein & Ericsson 2005). Those urbanites that had the fewest contacts with the countryside (through recreation or visits to second homes) also had most negative attitudes towards wolves. It is possible that if urbanisation continues and more of the cities' citizens become multigenerational urbanites, support for wolves will decrease. Thus it is not clear whether future urbanisation will lead to more positive attitudes towards wolves as would be expected by the simple rural-urban comparison in studies done outside Sweden.

Educating the public is a widely proposed solution to cope with all sorts of social problems. It turns out that the basic assumption that underlies this approach is not met in the wolf attitude research. Our data as well as those of other studies (e.g. Kellert & HBRS 1990) showed no positive correlation between knowledge and positive attitudes towards wolves (Ericsson & Heberlein 2003). Actually, in Sweden, the groups who knew the least about wolves liked wolves most. This association would imply that learning more about wolves would make people less positive about wolves. This paradox is more apparent than real when we look at the specific groups. Hunters living in areas with wolves had the most accurate objective knowledge about wolves but consistently the most negative attitudes. Simply ‘educating the hunters’ about wolves would not make them as positive as the general public. It is their experience with predation and their role as hunters that affect their attitude more than their general knowledge. We did find that within each of the groups, hunters, the general public and the public living in the wolf areas, those who had more knowledge were more in favour of wolves (Ericsson & Heberlein 2003). A major barrier to a successful education programme is that attitudes towards wolves among the general public are not very strong so therefore people are not likely to look for information. People who either love or hate wolves will be most likely to look for any information about wolves, but will be the least likely to change their minds. Those who are changeable, i.e. those with neutral attitudes, do not care enough about wolves to read pamphlets or take notice of information campaigns. So, trying to increase the knowledge of and information to the public in the long run may be helpful, but it should not be regarded as a silver bullet for making attitudes towards wolves more positive.

What appears to be an age cohort effect in the Swedish and international data will lead to a decline in negative attitudes over time. Our studies and the meta-analysis consistently show that the elderly have the most negative attitudes towards wolves. We believe that this is because they learned these attitudes during an earlier time and have carried them on through life. As the human population ages, these elderly people who learned their more negative attitudes at a previous period in time will make up a smaller and smaller proportion of the whole population. This should lead to an increase in the more positive attitudes in the human population. We do not expect the young people of today who have positive attitudes towards the wolf population to become less in favour as they grow older. However, this topic deserves future research using the same people in panel studies.

The discovery that contact of urban people with the countryside (either growing up in the countryside or making visits for recreation) is associated with more favourable attitudes towards wolves (as well as to hunting and wildlife; Ericsson & Heberlein 2005) presents some possibilities to increase support for wolves. Based on this, we would argue that programmes that support rural development and which get urban people out into the countryside will be likely to lead to more positive attitudes towards wolves, hunting and nature in general.

The most important problem right now is that wolves affect hunters negatively in as far as they kill their dogs, compete for the same prey, and provide few recreational benefits. Serious thought needs to be given to ways in which wolves can provide recreational or other benefits. The obvious possibility would be to allow sport hunting of a limited number of wolves every year when the wolf population level allows it. Hunters have a long tradition of becoming protectors of species that they are allowed to hunt.

In North America and Scandinavia, wolves become symbols of urban dominance in rural areas where wolves are restored. The major reason for the shift in attitudes from positive to negative in the state of New York was that local politicians reframed the issue from ‘wolf restoration’ to ‘outsiders telling us what to do’. The powerlessness and hopelessness that many people living in rural areas feel leads to an antipathy for symbols of urban dominance and the wolf has become one such symbol whether it be in Wyoming, Wisconsin or Dalarna in Sweden (Sharpe et al. 2001, Ericsson et al. 2008). But change is possible. Wolves in Yellowstone National Park have had demonstrable positive tourism effects (Bioeconomics 2005). Rather than being a threat, wolves have become interesting to some ranchers (Bass 1992). Making wolves a game species even in a limited number might make wolves part of the utilitarian culture of wildlife and provide rural residents with a greater sense of control. Recent changes in Sweden that allow land owners to shoot wolves even outside fences should help reduce human senses of powerlessness.

In Sweden today, attitudes towards wolves among the general public are positive and stable. This fits well with international data. The arrival of real wolves that do wolf-like things is likely to lead to more negative attitudes towards wolves, particularly in areas where the wolves return. The weak attitudes among the general public mean that large swings, likely negative, are possible. ‘Educating the public’ is not likely to offset such negative tendencies. Also the increase in the multigenerational urban population suggests the possibility of a more negative attitude in the future. The aging of the human population should lead to a more positive attitude over time. Programmes which increase the rural contacts of urban human populations might help maintain the current positive attitudes towards wolves in Sweden. Efforts should be put into making wolves valuable to hunters and reduce their symbolic status.

1.

2.

3.

4.

5.

6.

7.

8.

9.

10.

11.

12.

13.

14.

15.

16.