Pest control is a global issue for agriculture, health, biodiversity conservation and economy. Anticoagulant rodenticides are used over large areas to control rodent pests and can cause widespread poisoning of nontarget wildlife. In France, bromadiolone is the only pesticide authorized to control the water vole Arvicola terrestris Scherman, in grasslands. Since 2001, legislation has been in place to replace curative treatments by preventive ones and limit the quantity of rodenticide used. As the legislation took effect over time, the impact on red fox Vulpes vulpes populations was monitored.
Fox populations and bromadiolone treatments were monitored in the Doubs Department (5000 km² area), France. Fox counts were carried out during spring, and vole control was primarily conducted in autumn. Relative fox densities (Kilometric Abundance Index: KAI) obtained per commune for year n (2004–2009) were related to treatments achieved during year n−1 (2003–2008). Treatments from year n−2 were used to investigate possible delayed responses in fox populations.
Kilometric Abundance Index of foxes was significantly related to treatment intensities in years n−1 and n−2. The impact was greatest in a large area (>1000 km²), where intensive treatments were achieved in 2003. Fox KAI generally remained dramatically low in this area until 2005, after which a partial recovery was observed.
The same area was treated again from 2006 to 2008 but with only half the amount of bait per hectare that was used in 2003. These treatments were followed by a moderate decrease in fox populations.
Synthesis and applications: We have established, for the first time on a regional scale, the negative impact of a rodenticide on fox populations. We have shown that a shift to preventive treatments with reduced anticoagulant rodenticide use is less harmful to fox populations. However, to approach a zero impact, treatments should be reduced further by limitation of bait quantities authorized per hectare and per commune and using alternative methods to chemical control. Long-term monitoring of wildlife populations using index methods can provide valuable information about the adverse effects of pesticides; therefore, we recommend their inclusion in the assessment of pest management practices.
Pest control is a global issue for health, biodiversity conservation and economy. Indeed, pest species can play a key role in the transmission of infectious diseases to humans and livestock, invasive species can threaten native species, and over half of world-wide food production is lost due to pests (Pimentel 2007). Rodents are among the most damaging pests (Singleton et al. 2010). Anticoagulant rodenticides (ARs) are the main pesticides used to control them. ARs are one of the most frequently reported causes of wildlife poisonings in Europe (Berny 2007) inhibiting the vitamin K cycle and causing death by haemorrhage. As rodents became resistant to first-generation ARs (e.g. chlorophacinone), second-generation ARs (e.g. bromadiolone and brodifacoum), which are more potent, more persistent and consequently more hazardous for nontarget fauna, were developed (Erickson & Urban 2004).
Anticoagulant rodenticides have been used across large areas (thousands of ha) to eradicate invasive rodents from islands or to control agricultural pests. Although systematic eradication of invasive rodents from islands using these pesticides prevented species extinctions and restored ecosystems (Howald et al. 2007), significant impacts of ARs on nontarget populations have also been reported. For instance, decreases in bird populations (20–100%) were reported following a brodifacoum-baiting programme in New Zealand (Empson & Miskelly 1999).
Thus, land managers have to plan delivery methods (e.g. type and timing) of rodenticides to mitigate potential unintended impacts on other species (Howald et al. 2007). When islands are small, intensive monitoring of nontarget populations can be achieved and their size can be known with precision. For instance, the North Island saddleback Philesturnus carunculatus rufusater of Mokoia Island (135 ha) were all colour banded such that survival could be monitored during a poisoning operation (Davidson & Armstrong 2002). Mitigation techniques for vertebrates (e.g. birds, reptiles, rodents) have been applied successfully (e.g. temporary caging) on small islands. On large islands, which are biologically more complex, the monitoring of nontarget impacts remains a challenge (Howald et al. 2007).
In the case of agricultural pest control, land managers may have to deal with the potential impacts of ARs on a broader scale than on islands. In Eurasia, for instance, outbreaks of rodents may occur on a regional scale (3000–10 000 km2) and cause heavy damage to various crops, including cereals and grasslands (Singleton et al. 2010). Management decisions to control rodent outbreaks can be poorly informed and irrational treatments occur over large areas. Between 2006 and 2007, an outbreak of the common vole Microtus arvalis occurred in agricultural areas of north-western Spain. Rodenticides were applied over hundreds of thousands of hectares (270 000 ha in 2007), but most vole subpopulations had already collapsed before the third poisoning campaign began. These treatment campaigns resulted in tonnes of toxic bait remaining accessible to nontarget species, causing massive mortality (Olea et al. 2009). In many parts of Laos during the last decade, the use of rodenticides – such as coumatetralyl – has become chronic and has drastically reduced the number of domestic animals in villages. Presumably native wildlife also suffers nontarget mortality, but nothing is known about the long-term impacts on these species (Singleton et al. 2010). ARs may also have contributed to the decline of certain at risk species, such as the red kite Milvus milvus (Berny & Gaillet 2008). In most of these cases, the actual impacts of ARs on nontarget species were not assessed at the population level, mainly because they could not be easily monitored over such large areas.
In France, bromadiolone is the only pesticide authorized for use to control outbreaks of the water vole Arvicola terrestris Scherman in grasslands (Ministère de l'Agriculture et de la Pêche 2002). In the 1990s, hundreds of predators (mainly common buzzard Buteo buteo, red fox Vulpes vulpes and red kite) died after large-scale treatments (60 000 ha treated) in the Doubs Department (Giraudoux et al. 2006). During vole outbreaks, those predators feed almost exclusively on water voles and common voles, and both prey species eat bromadiolone baits (Sage et al. 2008). This leads to secondary poisoning of predators in treated areas (Berny et al. 1997). Foxes temporarily specialize on grassland voles when those prey are available (Raoul et al. 2010) and are relatively sedentary compared with raptors. Therefore, this species appears to be a relevant sentinel species to assess the effects of rodenticide field treatment on predator populations. The French wildlife monitoring scheme, the SAGIR network, reported hundreds of lethal intoxications, which included 164 foxes from 1997 to 2002 (SAGIR, unpublished data: Traitements bromadiolone et évolution de la mortalité de la faune non cible dans le département du Doubs de 1997 á 2008, Office National de la Chasse et de la Faune sauvage). From 1989 to 2000, fox populations were monitored using an index method on a local scale of approximately 100 km2 (Raoul et al. 2003). A sharp decrease in the population index was observed after the 1997–1998 treatments (almost 80 000 ha treated), and the population remained extremely low until 2000. Although the temporal dimension of the decrease was clear, its range and spatial dynamics could not be observed. Since 2001, legislation has prohibited AR application when vole density estimates surpass 200 individuals ha−1; farmers have been encouraged to control voles as early as possible during the early stage of population growth (Ministère de l'Agriculture et de la Pêche 2002). By shifting control efforts to the initial lower density phase of a population outbreak, reduced bromadiolone quantities can be used and control efficiency remains high (90–95% with bait at 0·5 kg ha−1; Delattre & Giraudoux 2009). In the early 2000s, many farmers continued AR applications in areas with high vole population densities, but increasing numbers of farmers adapted their practices over time. Treatment reporting is compulsory and so farmers must report bromadiolone quantities and treatment locations to governmental agencies (Préfecture du Doubs 2007). A data base has been set up since 1991 that maintains records of the bromadiolone quantities delivered each year at the resolution of the commune (a French administrative division of approximately 10 km²). In addition, since 2004, as part of a game management scheme, fox populations have been monitored by the Federation des chasseurs du Doubs (a federation of hunting associations in the Doubs Department) over a 5000 km² area. The time span covered by the data bases encompasses two successive A. terrestris cycles and includes the period when farmers shifted from the treatment of high population densities to the treatment of low population densities.
This context provides us with a quasi-experimental design at a regional scale that has allowed us to characterize the impacts of bromadiolone field treatments on red fox populations. We addressed the following questions:
Is the impact of treatment on fox populations dependent on bromadiolone quantities used, and, if so, were the measures implemented to reduce bromadiolone use sufficient to mitigate the impacts?
Is there a delayed impact on fox populations and what is their recovery pattern?
Materials and methods
Study area and control of rodent outbreaks
The study area is the Doubs Department, an administrative unit of 5234 km² in eastern France (47·11°N, 6·24°E), divided into 594 communes. Cyclic outbreaks of water voles occur every 5–6 years, mainly on the mid-altitude plateaus of the Jura Mountains (400–900 m altitude, Fig. 1). High vole population densities (200–1000 voles ha−1) persist locally for 1–3 years, dramatically reducing grass production. To control vole populations, field treatments with bromadiolone baits are carried out by farmers, mainly in autumn and, to a lesser extent, in spring and summer. Wheat baits were industrially prepared at a constant concentration of 50 mg kg−1 during the study period. Baits were distributed in artificial linear burrows (details in Sage et al. 2008). Legislation limits the quantity of baits used per treated area to a maximum of 20 kg ha−1. Giraudoux et al. (1997) reported a spatial asynchrony between population peaks with a subsequent time-lag in bromadiolone treatments between different altitudinal areas. The treated area ranged from 2400 to 17 900 ha year−1 during the study period (2003–2008).
Assessment of bromadiolone treatments
Bromadiolone treatments are carried out by farmers but require official permission from local authorities (Direction Régionale de l'Agriculture, de l'Alimentation et de la Forêt; DRAAF). Subsequently, the Fédération Régionale de Défense contre les Organismes Nuisibles (FREDON, farmer pest control association) delivers bait to farmer groups and records data on bromadiolone quantities used, areas treated and their locations. The quantity of bait (tonne, t) was used to quantify the intensity of bromadiolone treatment per commune per year. The quantities of bait used are highly accurate because the DRAAF monitors the bait quantities delivered to farmers and these data match the FREDON data.
Red fox data
Data were provided by the Fédération des Chasseurs du Doubs, a game management association.
Since 2004, fox populations have been monitored using night spotlight counts. Counts were performed along 472 geo-referenced transects, 1–2 km in length, distributed over the Doubs Department (Fig. 1). Every spring, counts were performed along each transect over 3–4 nights. These counts were performed at least 1 h after sunset in a car (maximum speed of 20 km h−1) and by 4 people: a driver, a data recorder and 2 observers. Observations were made using 100-W spotlights and binoculars for species identification. Double counting was unlikely because transects were relatively straight and observers were careful about animal movements. A Kilometric Abundance Index (KAI) was calculated for each transect as the maximum number of red foxes recorded km−1 (thus providing a lower limit for the number of foxes present).
Culling can substantially depress fox numbers (Heydon & Reynolds 2000). Consequently, we investigated whether culling and bromadiolone poisoning could be confounded as causes of decline. Shot fox numbers were collected from hunter reports from each shooting estate in the department from 2003 to 2007 (no data were available for 2008 because a different data collection process was used). Each shooting estate was assigned to a commune, and we computed the (minimum) number of red foxes shot commune−1 year−1 assuming that biases due to unreported shootings, if any, were constant over years.
The red fox is a generalist feeder and its population densities are not influenced by vole population densities in the Jura Mountains (Weber et al. 2002). Consequently, we did not include prey density data in our analysis. Moreover, the SAGIR network did not report any major disease outbreaks associated with high fox mortality during the study period.
Relationships between fox KAI and bromadiolone treatments, expressed through bait quantities (tonnes commune−1 year−1), were studied at a commune resolution over the whole department. Ordinary kriging was used to assign a KAI value per commune: KAI values from transects were interpolated on the centroids of communes. Ordinary kriging uses variogram models of spatial autocorrelation to provide spatially weighted estimators that are known to be useful for interpolation (Bivand, Pebesma & Gómez-Rubio 2008). For kriging, KAI were log-transformed to provide a closer approximation to normality. Omnidirectional empirical variograms were computed each year with log-transformed KAI values [log (KAI+1)]. Variographic envelopes, which test for potential absence of spatial autocorrelation, were computed by permuting geographical coordinates among KAI records (n = 999 simulations), re-calculating the empirical variogram for each simulation and plotting the maximum and minimum semi-variance obtained at each lag. Empirical variograms were fitted, using weighted least squares (Cressie 1985), with three correlation functions (exponential, gaussian and spherical) chosen via visual inspection. The variogram model providing the smallest residuals was selected. Kriged interpolations were mapped on a regular grid (2-km mesh).
Fox counts were carried out in spring (March to April) and water vole control treatments were primarily conducted in autumn (September to November). Fox densities (kriged log-transformed KAI, klKAI) for years n (from 2004 to 2009) in a commune were related to the number of shot foxes in years n−1, rodenticide use in years n−1 (B1) and rodenticide use in years n−2 (B2), thus investigating possible time-lag effects of secondary poisoning.
First, we assessed the possible confound of foxes shot (S1) on klKAI. We fitted a linear mixed effects model with klKAI of year n as response variable, and the commune and S1 as independent variables. The year was included as a random effect accounting for repeated sampling of fox densities (in the period 2003–2008). Residuals were normal and spatially independent.
The correlation between klKAI and S1 was positive. The number of foxes shot and bromadiolone poisoning could not be confounded as causes of decline. Therefore, we excluded S1 and modelled the klKAI of year n against B1 and B2. We fitted a panel data model designed to account for spatial dependence between communes in spatial time-series data (Kapoor, Kelejian & Prucha 2007). This approach uses a neighbourhood graph to obtain spatial weights that provide a basis for fitting spatial random effects (Bivand, Pebesma & Gómez-Rubio 2008; Millo & Piras 2012). Residuals were checked for normality and empirical variograms were used to test for residual spatial structure.
For mapping bromadiolone quantities, each commune was coloured depending on bromadiolone quantities used.
Statistics and graphical displays were computed using R 2.12.1 (R Development Core Team 2010), the packages sp (version 0.9-91, Pebesma & Bivand 2005), maptools (version 0.8-10, Lewin-Koh & Bivand 2012), nlme (version 3.1-105, Pinheiro et al. 2012) and splm (version 1.0-00, Millo & Piras 2012).
The intensity of bromadiolone treatments
Two distinct time periods were observed in the treatment intensity data, each characterized by a first year of intensive treatment followed by 2 years of less intensive treatment which were not necessarily in the same communes (Fig. 3a). In the first period, 2003–2005, 207 tonnes of bait were deposited over 23 811 ha (Table 1). From 2006 to 2008, bait use decreased with 143 tonnes used over a larger area of 27 755 ha. During the study period, treatment was maximal in 2003, with annual bait quantities used over the whole department being at least 3 times greater than during the 2004–2008 period and both total treated area and the number of treated communes being 1·5 times larger (Table 1). Bait quantities used per commune showed the same trend. In 2003, treatments reached up to 13·57 tonnes per commune (Fig. 3a) and 50% of communes were treated with doses larger than 0·69 t. In later years, treatments reached a maximum of 8·30 t per commune (2009) and the median quantity of used bait never exceeded 0·41 t (2006). The least intensive treatments (in terms of both bait quantities and treated surface area) occurred in 2004 and 2005 (Table 1).
Table 1. Bait quantity, surface area treated and number of communes where bromadiolone treatment occurred in the entire Doubs Department from 2003 to 2008
Bait Quantity (tonnes)
Area treated (ha)
Number of communes treated
The quantity of bait used per hectare treated was 1·67 times lower during the second period than the first (Table 1): it was at a maximum in 2003 (10 kg ha−1) and decreased to 5 or 6 kg ha−1 in later years. In 2003, 50% of communes were treated with doses higher than 8·7 kg ha−1; in 2004 and 2005, approximately 30% were treated with such doses; and from 2006 to 2008, fewer than 5% were so treated.
Impacts of the quantity of used bromadiolone on red fox population densities at the communal level
If fox shooting was a driver of fox density, fox density should decrease when shooting increased. Thus, since the number of shot foxes reported was positively and weakly correlated with klKAI (coefficient = 0·004, P <1·10−3) (Fig. 2a), we did not include shooting in further analysis. Communal klKAI were weakly significantly and negatively related to rodenticide use in year n−1 (coefficient = −0·069 ± 5·10−3, P <1·10−3) and rodenticide use in year n−2 (coefficient = −0·062 ± 5·10−3, P <1·10−3) (Fig. 2b). Thus, fox populations were negatively affected by bromadiolone treatments, and these effects were detectable more than one year later. The panel data model provided a good fit to the observed values with a pseudo-R2 of 0·524 (Fig. 2c). The fixed effects accounted for 10·8% of the total variance.
In the area that received the most intensive treatment during 2003 (south-west of the department), klKAI were less than 0·5 (i.e. 0·6 foxes km−1) across more than 1000 km² in 2004 (Fig. 3). In 2005, bait use declined yet klKAI remained lower than 0·5 in a central area of more than 500 km². The KAI needed at least 2 years to recover, and in 2006, all klKAI values in the treated area were greater than or equal to 0·5. During the second treatment period (2006–2008), klKAI of the treated sector was less than 0·5 in 2007 over a 100-km² area where bromadiolone was used. Fox densities did not decrease to the same extent as during the 2003–2006 period. In this 100-km² area, the quantity of used bait did not exceed 8·4 kg ha−1 over 96% of the treated area.
We have presented correlative evidence that bromadiolone field treatment reduced red fox population densities in the Doubs Department. While the correlative nature of our analyses limits our ability to identify the processes involved in the observed fluctuations in fox density, we can reasonably rule out the influence of some potential confounds. One confound might be that the baiting reduced vole populations, thereby depleting food availability for foxes and thus indirectly affecting fox populations via bottom-up control. However, such a numerical response of fox populations to prey densities, notably water voles, was not detectable in earlier studies carried out in the Jura Mountains (Delattre & Giraudoux 2009). Moreover, the treated area within a commune rarely exceeds a nominal part of the total grassland area (1–10%); thus, prey availability at this scale is not at stake. Fox culling and/or epidemics may also substantially depress fox numbers. In our study, fox culling was positively correlated with KAI, indicating that shooting was not a driver of fox decline. Furthermore, the SAGIR network did not report any major disease events associated with elevated mortality rates during the study period. Finally, no significant quantities of anticoagulants were used to control other rodent species in the studied area.
Visual examination of our maps clearly shows that KAI variation was spatially and temporally structured and indicates considerable similarity between the spatial structures in KAI and treatment intensity, yet the partial R2 associated with the fixed effect remained low (0·108). Moreover, the spatial random effects, fitted using the neighbourhood graph, explained more of the variation in KAIs than bromadiolone treatments at a communal resolution. This suggests that the spatial autocorrelation model might have absorbed some of the variance originating from the bromadiolone treatment. The low partial R2 also suggests that important factors other than bromadiolone treatment influenced KAI over this time period. According to studies carried out in our study area or in regions with similar landscapes (mosaics of farmland and woodland), fox feeding behaviour and detectability during night counts should not lead to variability in KAI. Indeed, in the Swiss Jura, Meia & Weber (1995) showed that daily distances travelled by foxes did not change despite drastic changes in A. terrestris population densities and thus the availability of preferred prey. Moreover, the proportion of time spent by foxes in open and closed habitats should not vary in space and time because foxes are strongly attracted by open areas and grassland small mammal species (Giraudoux 1991; Doubs department; Weber & Meia 1996; Swiss Jura). Ruette, Stahl & Albaret (2003) measured fox detectability in north-eastern France with a protocol similar to ours (spotlighting along transects of approximately 2 km). Despite strong variations in the percentage of forested cover among sites, they concluded that detection rate corrected KAIs were of little interest when spotlighting foxes. Thus, we can reasonably exclude that the high variability in KAI is due to heterogeneity in fox behaviour.
The long-term monitoring of the impacts of ARs on populations of rodent predators on a regional scale, and in a spatially explicit context, is not a general practice in Europe. To our knowledge, impacts at the population level have only been reported for the red kite after a large-scale control of common vole with chlorophacinone and bromadiolone in Spain (Mougeot, Garcia & Viñuela 2011). In the UK, Shore et al. (2003) reported that current usage of ARs did not and would not prevent polecats Mustela putorius from further expanding their range in eastern areas of England. Furthermore, no direct relationship was evidenced between the decline of the national populations of the barn owl Tyto alba (Shore et al. 2005), or the tawny owl Strix aluco (Walker et al. 2008), and their exposure to ARs in the UK.
The monitoring of AR impacts often relies on carcass collection (e.g. WIIS in UK or SAGIR in France). Due to a short data acquisition time frame, carcass collection can signal the need to urgently cease treatment in response to massive mortality. For instance, bromadiolone treatments were temporarily stopped during the winter 2011/2012 in the Auvergne region (France) after 28 red kites and 16 common buzzards were found poisoned (LPO 2012). However, it is important to question whether the number of poisoned animals collected accurately reflects the impact of ARs on a population. During our study period, 22 foxes were found poisoned in 2003 (SAGIR, unpublished data), the year of most intensive bromadiolone use which was followed by the lowest observed klKAI values in 2004. However, in subsequent years, treatment intensity decreased and no simple pattern linking poisoned fox numbers to treatment intensity emerges: just 5 foxes were found in 2007 and 2 in 2006, while in 2006 more communes were treated with greater bait quantities. In most cases, carcasses were not actively sought, they were essentially discovered circumstantially, resulting in low numbers and a high noise to signal ratio. Reports of carcasses may depend on the mobilization of local observers and on carcass detectability (Decors & Mastain 2010). The SAGIR network mainly records acute lethal exposures, and chemical analyses are performed if poisoning is suspected (Berny 2007). Complementary to carcass collection, the type of population monitoring we carried out enabled the detection of an integrated effect of chemicals (i.e. both lethal and sublethal effects). Concerning ARs, if lethal poisoning is evident, sublethal impacts on wildlife are rarely documented. Negative correlations between anticoagulant concentrations and body condition suggest that exposure to anticoagulants adversely affects the fitness of stoats Mustela erminea and weasels Mustela nivalis (Elmeros, Christensen & Lassen 2011); and several raptor species (common kestrel Falco tinnunculus, tawny owl and common buzzard) (Pereira 2010). Moreover, some positive relationships between rodenticide exposure and pathogens prevalence have been evidenced for bobcats Lynx rufus (Riley et al. 2007) and common voles (Vidal et al. 2009).
Moreover, our long-term monitoring demonstrates that populations were affected for more than one year, which supports two nonexclusive hypotheses: delayed poisoning and/or a time of longer than 1 year is required for fox population recovery. Regarding delayed poisoning, following the extinction of a local vole population, new vole colonizers may be intoxicated 3 months after bromadiolone treatment and some voles with bromadiolone residues may even be trapped after 6 months (Sage 2008). However, the number of foxes found poisoned 1 year after intensive treatments was low (1 fox found in 2004, whereas 22 were found in 2003, SAGIR, unpublished data), suggesting that delayed exposure is marginal. The other hypothesis relies on the demographic processes involved in fox population recovery. These processes may involve higher recruitment rates into impacted populations due to decreased intraspecific competition. Heydon & Reynolds (2000) showed that a moderate breeding suppression in rural fox populations was associated with a higher density: in an area with 0·79–2·76 foxes km−2, females produced an average of 4·49 cubs year−1, whereas in an area with densities of 0·14–0·60 fox km−2, females produced 6·24 cubs year−1. Another process involved in fox population recovery may be the recolonization of foxes in the impacted area from neighbouring areas that were treated less intensively. This recolonization is a strong possibility because foxes are able to disperse over considerable distance and, consequently, might swiftly recolonize areas from which they have been removed (Gentle, Saunders & Dickman 2007). We could not determine the time necessary for a total recovery of fox populations because bromadiolone treatments never stopped within the department.
Here, we showed how geo-referenced monitoring of both pesticide use and wildlife populations can permit the assessment of mitigation measures and, thus, may help to improve these measures. In 2011, the use of bromadiolone was re-authorized to control water voles for plant protection until 31 May 2015 (Directive 2011/48/EU), but the European commission requested additional information about the effectiveness of proposed mitigation measures to reduce the risk to wildlife. Even if red fox is not specifically mentioned, we show here that this species is a relevant sentinel to assess the effectiveness of mitigation measures such as those applied in our study area. Since 2002, French legislation has gradually moved to bromadiolone control of voles during subthreshold population density periods (i.e. surface indices should not exceed 50% of the intervals on line transects performed according to protocols adapted from the study by Giraudoux et al. (1995)) and a maximum of 20 kg of bait ha−1 year−1. Since 2005, a contract for sustainable control has been proposed to farmers by FREDON. This contract aims to maintain vole population densities at low as possible levels by combining rodenticides and alternative methods such as trapping, habitat destabilization via ploughing rotations for grasslands, control of moles Talpa europaea, (Delattre et al. 2006), hedgerow creation for favouring predators and reducing grassland connectivity, and protection of typically unprotected predator species such as foxes (see e.g. Delattre & Giraudoux 2009 for details). During our study period, these methods combined and permitted reductions of at least 66% in total bait quantity applied to fields across the department. Moreover, treatment intensity was decreased from 10 to 5 or 6 kg of bait ha−1. This shift from intensive treatment at relatively high vole densities to treatment at lower densities and the growing use of alternative methods limited the impact on fox populations, which demonstrates the efficiency of the mitigation measures taken to date.
However, field treatments with bromadiolone still damage red fox and other nontarget species (Plaquin & Clerc 2009; SAGIR, unpublished data), and we propose additional mitigation measures to sustain the trend towards a zero impact on wildlife. The first measure would be to reduce the maximum bait quantity authorized per ha. Sage et al. (2008) and Coeurdassier et al. (2012) demonstrated that treatments performed at bait densities of 20 kg ha−1 (i.e. the highest quantity authorized by regulation) presented a high risk to predators of voles. In the present study, we showed that fox populations were affected even if less than 8·4 kg ha−1 of bait was used across 96% of the treated area. Just 0·5 kg ha−1 is sufficient to control water voles at very low densities, and, based on bromadiolone residues in vole tissue, this level of treatment led to a calculated risk of zero for the fox (Defaut et al.; Coeurdassier et al. in Delattre & Giraudoux 2009). However, the low water vole density threshold above which foxes readily concentrate on this prey (Raoul et al. 2010) suggests that foxes might be exposed to rodenticides while vole densities and treatments are low. Moreover, when present on treated plots, common voles will consume bromadiolone baits (Sage et al. 2008) making secondary exposure of foxes likely. This finding supports drastic reductions in treatment intensity to levels lower than those currently permitted by legislation. Moreover, whereas legislation only limits the quantity used per agricultural parcel, our results show that the bait quantities used over the whole commune have negative consequences on fox densities and, thus, should also be considered in legislation. Finally, bromadiolone (and other ARs) is only one of several options available for rodent pest control (Singleton et al. 2010). Because bromadiolone is authorized for use in Europe until at least 2015, we believe that it is essential to promote alternative methods of control, including those presented above, and that AR field applications should be limited to farmers contractually engaged in a sustainable integrated management of vole outbreaks.
Ecologically sensitive rodent management has been developed in various agricultural systems, particularly rice-based systems in Asia (Singleton et al. 2010). The proposed alternative control strategies rely on the same principles as in the system studied here: collective rodent culling actions at key times in their population cycles and habitat manipulation. Control should be preventive, whereby rodent populations are maintained at manageable levels, and alternatives to rodenticides should be developed. In some countries, this has led to rodenticide reduction (e.g. 50% in Vietnam), but potential benefits for nontarget animals were not measured. Simply decreasing rodenticide use does not guarantee the removal of nontarget impacts – poisoning of endangered species remains possible (SAGIR, unpublished data). Moreover, rodenticides can kill rodent predators (wild and domestics) (Berny et al. 1997; Singleton et al. 2010) and thus impair biological control of rodents. We consequently recommend implementation of monitoring of appropriate nontarget species populations within those systems. Even implementing simple dead body collection protocols can provide valuable complementary information that can improve understanding of how ARs act as drivers of population decline.
Monitoring red fox population dynamics using a simple index method is relevant for assessing the long-term impacts of bromadiolone field treatments for A. terrestris control. In the present study, annual monitoring was performed over a large area courtesy of effective collaboration between research institutions, game management associations and farmer organizations. The next step is to propose more ambitious mitigation measures seeking to achieve zero impact on wildlife. This monitoring provides an integrated indicator of treatment impacts, which, as a complement to the carcass collection based monitoring scheme, will continue to improve the control of vole outbreaks based on AR use and/or nonchemical methods.
We thank volunteer hunters for assisting with roadside counts and our financial partners, the Franche-Comté Region and the Agence Nationale de la Recherche, who sustained the RODENT Program (Agreement n°2009CESA00801). We thank Olivier Mastain, Anouk Decors and Betty Plaquin from the Office National de la Chasse et de la Faune Sauvage for providing us with data from the SAGIR network in the Doubs Department. We are most grateful to Roger Bivand and Gianfranco Piras for their kind guidance in the use of the package splm and general estimation with panel data models.