The successful management of large herbivores requires the monitoring of a set of indicators of ecological change (IEC) describing animal performance, herbivore impact on habitat and relative animal abundance. Roaring counts during the rut have often been used to assess the abundance of red deer Cervus elaphus populations, but a formal evaluation of this method is still lacking. In this paper, we examined the usefulness of the number of red deer recorded during roaring counts for managing red deer populations. Using standardised spotlight counts applied for the monitoring of red deer at La Petite Pierre, France, as a reference method, we found that roaring counts did not correlate with spotlight counts. Moreover, we did not find any evidence that roaring counts decreased with increasing number of male and female red deer harvested in the reserve during the previous hunting season. We thus conclude that managers should not rely on roaring counts for managing red deer populations.
Deer populations have increased dramatically in size and distribution over recent decades both in Europe and North America (Andersen et al. 1998, Fuller & Gill 2001). As a result, in many areas, management goals are changing from protection to population control (Milner et al. 2006). Controlling deer populations is required to prevent damage to farming and forestry, deer-vehicle collisions or the spread of diseases (Rooney & Waller 2003, Côté et al. 2004, Gordon et al. 2004, Seiler 2004). In areas where large predators have been eliminated, deer populations are usually controlled through hunting (Langvatn & Loison 1999, Festa-Bianchet 2003, Milner et al. 2006). To define the hunting quotas, managers typically use estimates of population size obtained from counting methods (e.g. Williams et al. 2002).
Counting methods have repeatedly been shown to be neither accurate nor precise (CV > 30%; Andersen 1953, Gaillard et al. 2003) except when a high proportion of individuals are marked (Strandgaard 1967, Vincent et al. 1991). Andersen (1953) showed, for example, that counts in a population of roe deer Capreolus capreolus can lead to a threefold underestimation of the population size. Moreover, the absence of repeated counts within a year prevents the assessment of the precision of population estimates. Lastly, even the best estimate of population size does not provide the required information to set hunting plans because it does not provide information to managers about the state of the plant-herbivore system (Morellet et al. 2007).
In this context, researchers have developed new tools for managing ungulate populations, which are based on the monitoring of a set of indicators of ecological change (IEC; Cederlund et al. 1998, Morellet et al. 2007). An IEC is an easily measured parameter, sensitive to changes in the relationship between the population and its habitat (Cederlund et al. 1998). IEC are based on the concept of density-dependence (Bonenfant et al. 2009), from which wildlife managers should be able to monitor deer populations without population size estimates. Morellet et al. (2007) proposed to simultaneously monitor three categories of IEC describing 1) animal performance (Bonenfant et al. 2002, Zannèse et al. 2006, Garel et al. 2011), 2) habitat quality (Morellet et al. 2001, Chevrier et al. 2012) and 3) animal abundance (Vincent et al. 1991, Loison et al. 2006, Garel et al. 2010). It is important to note that only the complementarity of different IEC monitored over several years allows for the measurement of population-habitat relationships.
Among indices of red deer Cervus elaphus abundance, spotlight counts are commonly used by managers (e.g. Gunson 1979, McCullough 1982, Fafarman & DeYoung 1986). Recently, Garel et al. (2010) showed that, after controlling for observation conditions, spotlight counts can provide a reliable IEC of relative abundance for the monitoring and management of red deer populations living in forested areas. Spotlight counts are inexpensive and allow for rapid cover of large areas (McCullough 1982). However, this method is restricted to areas with good network of roads and tracks. An alternative option could be to carry out roaring counts from fixed spots (Albaret et al. 1989).
Counting red deer during the period of reproduction (rut) in autumn is often used in addition to spotlight counts, or when spotlight counts are difficult to apply (Bobek et al. 1986, Albaret et al. 1989), to assess trends in population abundance. Mature red deer males, like many other ungulates, commonly spend most of the year in areas separated from those occupied by the females except in the breeding season (Clutton-Brock & Albon 1979, Clutton-Brock et al. 1982). At this particular time of the year, stags use vocalisations (roars) to monopolise a number of hinds. With their highly conspicuous behaviour, red deer are easily detected during the rut based on acoustic and visual clues. Hence, roaring counts have two components: first, counting the number of roaring stags, and second, counting hinds and stags seen in rutting areas, both being used for the calculation of the abundance index. However, Ciucci et al. (2009) have suggested that roaring counts are poor indicators of red deer abundance because the timing of the survey is not consistently synchronous with the roaring peak. Overall, the roaring counts method has never been formally validated.
In our study conducted at La Petite Pierre, France, we examined the usefulness of roaring counts for the management of red deer populations. Managers of our study site have monitored red deer using both spotlight and roaring counts for 31 years. Providing that roaring counts allow for reliable assessment of the abundance of red deer resident populations, we expected a positive correlation between roaring and spotlight counts because spotlight counts provide a suitable measure of annual variation in abundance of the resident population (Garel et al. 2010). Second, as variation in hunting pressure accounted for most annual changes in population size in the studied population (Bonenfant et al. 2002, Richard et al. 2010; see also section Material and methods), the number of counted red deer during rut should decrease with the number of deer harvested during the previous hunting season (Garel et al. 2005b).
Material and methods
Study site and population
La Petite Pierre National Reserve (PPNR) is a 2,674 ha unfenced forest located in the Vosges mountain range, northeastern France (48.82°N, 7.34°E; Fig. 1). The PPNR is characterised by a succession of small hills and steep-sided valleys ranging between 200 and 400 m a.s.l. in elevation (see Fig. 1). The climate is continental with oceanic influences, leading to cold winters and cool summers (average January and July temperatures are 0.6 and 18.4°C, respectively). The PPNR consists of a balanced mix of broadleaved, mainly European beech Fagus sylvatica, and coniferous trees such as silver fir Abies alba, Norway spruce Picea abies and Douglas fir Pseudotsuga douglasii. The sandstone substrate produces acidic and poor soils, resulting in a vegetation of low nutritive quality for herbivores. Roe deer and wild boar Sus scrofa are also present within the reserve. All three ungulates are managed through hunting, either with quotas (deer) or without quotas (wild boar). Since 1978, the red deer population has been hunted on an annual basis between 1 August and 1 February by both professional and sport hunters (more details on hunting practice in France can be found in Maillard et al. 2010). A temporal variation in hunting pressure controlled by the Office National de la Chasse et de la Faune Sauvage and the Office National des Forêts occurred throughout the study period (see Results), leading to marked variation in population size over the past 30 years (Richard et al. 2010). This management context provided us with a quasi-experimental manipulation of red deer density to assess the relevance of roaring counts as a management tool of red deer (Sinclair 1989). Every deer shot was sexed and aged. Our study area was free of predators that could have any influence on deer behaviour or habitat use.
We analysed roaring counts of red deer performed during 31 consecutive years from 1978 to 2008. Each year, four surveys have been conducted (N = 124) at the time of the rut from mid-September to mid-October (Loe et al. 2005; mean date of counts was 25 September). Red deer were counted at 47 observation spots within the PPNR (see Fig. 1), with one or two observers at each spot. Counts were performed at dusk and the next morning at dawn, twice a year and required approximately three hours to be completed. Observations included both red deer seen and heard. The animals seen during counts were classified into three age-sex classes (calves, stags and hinds) according to antler and body size characteristics. For seen red deer, observers recorded the exact location, the direction and time of arrival and departure, the activity (roaring or silent) and the shape of antlers for males. For heard red deer, the approximate location, the start time and end of roar were recorded. Similarly to Ciucci et al. (2009), we used triangulation to locate red deer males more accurately using bearings taken by different observers. All this information was simultaneously analysed (by J-L. Hamann) to correct as far as possible for double counting and distinguish sighted stags from roaring stags. Hence, roaring males corresponded to stags that were not seen.
For each survey, we calculated an abundance index as the total number of red deer observed (AI-R). Poisson regression is often used to model count variables (Agresti 2002). However, overdispersion often occurs when a Poisson regression is used, resulting in an underestimated variance of the regression model parameters (Poortema 1999). Here, the overdispersion parameter Φ, calculated as a ratio of the deviance to degrees of freedom, was >> 1 for AI-R (Φ = 4.35), providing evidence for overdispersion. Thus, similarly to Garel et al. (2010), we used generalised linear models (GLM) with a log-link and negative binomial distribution to obtain annual estimates of AI-R. The model included year as a factor (31 modalities) and both daily rainfall and daily mean temperature, which are likely to influence roaring counts (Ciucci et al. 2009). We coded the rainfall variable as a two-level factor (i.e. presence/absence of rainfall). Climate data were obtained from a Météo-France weather station at Danne-et-Quatre-Vents (48.7°N, 7.29°E) located < 5 km from our study area. To obtain corrected values of annual AI-R, we used back-transformed year-specific coefficients of the fitted model.
AI-R included both acoustic and visual components, which did not necessarily have the same detection probabilities, and these ones may be influenced by different factors. Consequently, we derived two other abundance indices following the procedure used for AI-R; one based on acoustic detection of stags (AI-Ra) and the other based on visually detected individuals (AI-Rv).
Estimates of population size and spotlight counts
Spotlight counts of red deer were conducted from February to April each year from 1979 to 2009 (N = 495). The method consisted of driving three independent routes (median length of 33 km) within the PPNR twice a month and counting every deer seen with a powerful spotlight. We used a generalised linear model with a log-link and negative binomial distribution to obtain the total number of red deer observed during annual spotlight counts (AI-I). The model included year as a factor (31 modalities), the route length (log-transformed) as an offset covariate and a two-level factor describing the conditions of observation (‘good’ (i.e. clear sky) vs ‘bad’ (i.e. occurrence of rainfall, snowfall and/or fog); see Garel et al. 2010 for details). As hunting occurred between roaring counts and spotlight counts five months later, we added the number of deer killed in the PPNR between the two counts as a variable. Indeed, the number of deer killed during the hunting season varied strongly among years and affected the number of red deer counted during spotlight counts negatively (β = −0.028 ± 0.010, P = 0.005), supporting that hunting pressure accounted for most annual changes in population size (Bonenfant et al. 2002, Richard et al. 2010). More details about spotlight counts and its relationship with population size can be found in Garel et al. (2010). Here, we only used spotlight counts as a reference because spotlight counts were available for 31 years, whereas Capture-Mark-Recapture (CMR) estimates were available for only 16 years (Garel et al. 2010).
Testing roaring counts as an index of abundance
We first examined the relationship between AI-R estimated from the generalised model (see section Roaring counts) and AI-I, using a standard linear model. Then, we tested for the negative relationship between AI-R and the number of red deer shot within the PPNR during the previous hunting season using a standard linear model. We log-transformed all these variables to satisfy the homoscedasticity hypothesis of linear models (see also Loison et al. 2006, Garel et al. 2010). We tested the statistical significance of correlation coefficients (r) by examining whether zero was included within coefficient confidence interval at 95%.
Roaring counts (inter-annual CV = 22%) led to an average of 69.0 ± 1.9 animals during the study period. Visual and acoustic detections accounted for 70% and 30% of the total number of observed red deer, respectively. Temperature ranged from 7.3 to 20.6°C (averaging 13.3°C across study years) and was negatively correlated with AI-R (β = −0.024 ± 0.010, P = 0.016). AI-R tended to decrease with the occurrence of rainfall (β = −0.083 ± 0.049, P = 0.095). Once the conditions of observation were accounted for, AI-R was highly variable from year to year (χ2 = 121.8, df = 30, P < 0.001) ranging from 46.0 ± 5.6 animals in 1991 to 108.4 ± 12.1 animals in 1997 (Fig. 2). Since the value of deviance divided by the number of degrees of freedom was close to 1 (Φ = 1.41), the model fitted the data satisfactorily.
In contradiction with our first prediction, the log-transformed AI-R was not correlated to log-transformed AI-I (r = −0.01, CI = −0.36 - 0.34, N = 31 years; Fig. 3). We also assessed the relationship between the two components of AI-R (AI-Ra and AI-Rv) and AI-I, but our results remained unchanged with or without distinguishing between visual and acoustic detection of deer (see Appendix I).
On average, 43 red deer were shot each year between 1978 and 2007 in the PPNR, but the hunting effort varied considerably among years, with 78 deer shot during the 1979 hunting season and 24 during the 1996 hunting season. Contrary to our second prediction, log-transformed AI-R was not influenced by the log number of male and female red deer killed during the previous hunting season (r = −0.27, CI = −0.58 - 0.09, N = 30 years).
Roaring counts are commonly used in addition to spotlight counts by managers in European countries to monitor red deer populations (Ciucci et al. 2009). For instance, the method has been systematically used since the 1970s to monitor the endangered Corsican red deer C. e. corsicanus in Sardinia (Lovari et al. 2007). Indeed, at the time of the mating season, red deer gather and are easily detected, particularly roaring stags, due to movements and sounds. Although roaring counts might be a useful alternative to counting methods such as spotlight counts in mountainous areas (Bobek et al. 1986, Albaret et al. 1989), a formal evaluation of its reliability has been lacking. In our study, we found no relationship between the total number of red deer observed during roaring counts (AI-R) and spotlight counts, suggesting that roaring counts do not capture the variation of abundance of the resident deer population.
Bias can occur in analysis of counts when data contain sources of variation other than the changes in the size of the population. Potential sources of variation include weather, habitat or differences between observers in their ability to detect animals (Garel et al. 2005a). These effects should be taken into account to get reliable measures of population change (for an example in red deer, see Garel et al. 2010). Here, we reported a negative relationship between temperature and AI-R, which can be explained by a reduction of animal activity when temperature increases. Garel et al. (2005b) found similar results on ground and aerial counts during summer in mouflon Ovis gmelini musimon. Roars can usually be heard from a long distance in good weather conditions, and therefore, sampling points in the field are generally separated by several hundred metres. Heavy rainfall decreases the vocal activity of stags (Pépin et al. 2001). However, the negative effect of rainfall on AI-R we reported here reflected the difficulty for observers to see animals under rainy conditions rather than a difficulty to hear roars (see Appendix I).
In addition to weather conditions, the proportion of roaring stags in a population depends on several factors such as condition of the stags, age structure, population density, quality and spatial distribution of the resources (Clutton-Brock & Albon 1979, Carranza 1995, Pépin et al. 2001, Yoccoz et al. 2002). Moreover, differential visibility of the various age and sex classes introduces additional sources of uncertainty in the population abundance estimates (Ciucci et al. 2009). Deer activity during the rut is fundamentally different between sexes, with stags roaring and moving a lot, whereas females remain rather elusive during this period. The detection probability is thus higher and less variable in males than in females which explains that 62% of the red deer recorded during the rut in the PPNR were males.
In mountainous areas, the altitudinal movements of red deer from a low-elevation winter range to a high-elevation summer range is a common pattern of migration (Mysterud et al. 2001). In the PPNR, due to a very low range in elevation (see section Material and methods), such an altitudinal migration to benefit from high quality food did not occur. However, a marked sex-specific seasonal migration occurred in relation to mating opportunities. While females tend to occupy the same particular area during the whole year, in autumn, males move to rutting areas that are up to 30 km from the area occupied during the rest of the year (Bonenfant 2004:226). This form of partial migration (sensu Dingle & Drake 2007) has also been reported in a red deer population in Sweden where an average distance of 14 km and a maximum distance of 47 km were recorded between rut and summer/winter observations (Jarnemo 2008). Consequently, a high proportion of the stags that are observed during roaring counts are not resident of PPNR. On the contrary, red deer counted and monitored during spotlight counts that occurred from February to April were resident deer of PPNR. A combination of seasonal migration and unreliability of roaring counts are likely to explain the lack of a positive relationship between AI-R and spotlight counts found in our study. Although we can not prove that roaring counts were unreliable because the population size in autumn was unknown, the observed effects of hunting provided further support to our interpretation. Indeed, we found an absence of a relationship between AI-R and the number of red deer harvested in the reserve during the previous hunting season, whereas hunting quotas led to a decrease in the number of red deer counted during spotlight counts.
Successful management of large herbivores requires the monitoring of IEC describing animal performance, herbivore impact on habitat and animal abundance (Morellet et al. 2007). Monitoring temporal changes in these indicators provides a basis for setting hunting quotas to achieve specific management objectives. In this sense, spotlight counts provide a reliable index of the abundance of resident populations in a given area for red deer (Garel et al. 2010). On the other hand, roaring counts cannot be interpreted as a reliable IEC because they do not track abundance of the resident population of red deer despite a much larger sampling effort compared to spotlight counts (average annual number of 184 observers for roaring counts vs 72 for spotlight counts). Based on a least-cost approach for achieving management objectives, roaring counts should not be used for managing resident red deer populations.
we are grateful to the Office National des Forêts and to all volunteers for their participation during roaring counts. We would also like to thank Météo-France for the climatic data and Aurélie Barboiron for Figure 1. We also thank Stefano Focardi and an anonymous reviewer for insightful comments on a previous draft of the manuscript.
Appendix I. Relationship between spotlight counts (AI-I) and the acoustic (AI-Ra) and visual (AI-Rv) components of roaring counts
We used generalised linear models (GLM) with a log-link and negative binomial distribution to obtain annual standardised estimates of AI-Ra and AI-Rv. The models included year as a factor (31 modalities), and the daily rainfall and daily mean temperature which were likely to influence the results of roaring counts. We coded the rainfall variable as a two-level factor (i.e. presence/absence of rainfall). We examined the relationship between AI-Ra and AI-Rv estimated from the generalised model and AI-I, using a standard linear model. We tested the significance of correlation coefficients (r) by examining whether zero was included within coefficient confidence interval at 95%.
We found no evidence of any influence of climate on AI-Ra (all P-values > 0.40) while AI-Rv decreased with increasing temperature (β = −0.030 ± 0.013, P = 0.02) and rainfall (β = −0.130 ± 0.060, P = 0.04). Neither the log-transformed AI-Ra nor the log-transformed AI-Rv was correlated with the log-transformed AI-I (LN AI-Ra: r = −0.35, CI = −0.69 - 0.11, N = 31 years; LN AI-Rv: r = 0.14, CI = −0.33 - 0.55, N = 31 years).