Study Area

We conducted the research within Kouchibouguac National Park, in eastern New Brunswick, Canada. The area is characterized by mixed forest dominated by black spruce (Picea mariana), white pine (Pinus strobus), red pine (P. resinosa), balsam fir (Abies balsamea), maple (Acer spp.), and birch (Betula spp.). Wetlands are prominent in the park, in the form of peatlands, rivers, brooks, ditches, and permanent and temporary ponds.

Field Survey

The field study consisted of night-driving surveys (sensu Shaffer and Juterbock, 1994) on a 20-km road segment of Route 117 going through the park. The two-laned road is ca. 8-m wide, and has an average nocturnal traffic load of approximately 13.6 vehicles/h during spring and summer (Mazerolle, 2003). We conducted the surveys from May to September 2001–2004 after dark when the road surface was humid (during or following rain) between 2000 and 0400 h. A single 20-km survey was conducted on any given night. Driving along the road at 10–20 km/h, we stopped each time an amphibian was seen on the road. At each stop, we identified the individual to species and noted its behavior (i.e., moving or immobile) when it was first seen on the road from inside the moving car. Because we were limited by suitable weather conditions (i.e., rainy nights), we only conducted five surveys in 2001 and four in 2002. However, wetter conditions prevailed in 2003 and 2004, during which we realized 27 and 14 surveys, respectively. For each species, we computed the proportion of live individuals encountered on the road that had moved at the approach of the vehicle during each survey night and used these proportions as the basis for our analyses (see “Statistical Analyses” below).

Field Experiment

We selected the six amphibian species most common in the study area for the field experiment: spring peeper (Pseudacris crucifer), spotted salamander (Ambystoma maculatum), blue-spotted salamander (Ambystoma laterale), American toad (Bufo americanus), green frog (Rana clamitans), and wood frog (Rana sylvatica). A total of 91 adult and juvenile amphibians were captured during night-driving surveys on the same stretch of road described above or at a nearby pond. These consisted of (mean snout–vent length ± 1 SD): 20 mole salamanders (4.3 ± 1.4 cm), 11 American toads (3.8 ± 1.7 cm), 18 spring peepers (2.7 ± 0.8 cm), and 42 ranid frogs (4.3 ± 1.6 cm). Each animal was kept in captivity in an individual, moist container in a dark cool room (ca. 14 C) for no more than 24 h before the trials.

All treatments were conducted on a 62-m stretch of single-lane paved road (4.8 m in width) leading to the parking lot of Kelly's Bog (46° 58′ 59″ N, 64° 57′ 12″ W) within Kouchibouguac National Park. We selected this road mainly because the very low traffic allowed us to safely conduct the experiment. Furthermore, because the road was unlit (i.e., no street lights), we were able to control all artificial light sources. During the experiment, we poured well water over the road surface to maintain its humidity and simulate wet road conditions such as those encountered during rainy nights.

We conducted the experiment under similar weather conditions on the nights of 31 May, 30 June, 11 July, 10 August, and 13 August 2001. We noted the air temperature before starting a set of trials for a given individual. We conducted all observations between 2200 h and 0430 h, at least 1 h after sunset. On a given night, we randomly selected each amphibian from the group of individuals captured less than 24 h earlier. We submitted each individual to four treatments in a random sequence. We determined the response (i.e., moving vs. immobile) of each individual following exposure to the sound of the car, the headlights, or a combination of both, relative to a control. Standing at least 5 m from the individuals, the observer used a headlamp fitted with a red filter to monitor amphibians during the trials.

The first treatment (i.e., headlight and motor treatment) simulated the encounter of a car and the individual on the road surface. For the second treatment (i.e., motor-only treatment), we assessed the response of amphibians to the sound associated with a car. We completely covered the headlights of the car with a thick opaque fabric to prevent any light from being emitted. The third treatment consisted of measuring the behavior of amphibians relative to headlights without the sound of the motor (i.e., headlights-only treatment). Because it would have been impossible to move the car during the headlights-only treatment (i.e., no motor), we used a car surrogate. To do so, we fastened two strong flashlights (Mag Lite™ with 3-D cells) at a height and width identical to that of the car onto a children's wagon (Radio Flyer™). The intensity of the flashlights was ca. 39 lux at 5 m (Corben and Fellers, 2001) and equates to that of car headlights at approximately 50 m. This treatment provided a means to isolate the effect of light and the sound of the motor on amphibian behavior. The wagon was pushed from behind to simulate the approach of a car. The fourth treatment (i.e., control) consisted of watching the behavior of individuals under the red light for 30 s without any other disturbance.

Every treatment was preceded by a 2-min habituation period during which the individual was placed in the middle of the road under a 500 ml container to minimize disturbance. At the start of a given trial, the container was removed. The car or wagon then started 62 m from the position of the individual and approached at a speed of 5–10 km/h, coming to a stop 1 m in front of the animal. The observation period began as soon as the vehicle started its approach and was terminated when the vehicle was 1 m from the individual (ca. 30 s later). We recorded the behavior of the individual during the observation period as moving (i.e., moving/resuming activity) or immobile (i.e., remains in same place). At the end of each of the four trials, we rinsed, with water, the surface on which each individual was placed to minimize the possible effect of secretions on the next individual's behavior. All individuals were measured to snout–vent length (SVL) and released at their point of capture the day following their use in the experiment.

Statistical Analyses

In both the field survey and the field experiment, we recorded whether each amphibian moved (i.e., a binary variable reflecting the probability of fleeing from an approaching car). This type of data is best suited for logistic regression analysis and is analogous to linear regression in its structure and interpretation (Hosmer and Lemeshow, 1989). Because each individual in the field experiment was submitted to four treatments, we used an extension of logistic regression for repeated measures (i.e., generalized estimating equations) to account for repetitive observations on the same individual (Diggle et al., 1994; Horton and Lipsitz, 1999; Stokes et al., 2000). Due to low numbers of certain species and to facilitate comparisons between the survey and the experiment, we pooled the mole salamanders together (A. laterale and A. maculatum), and the ranid frogs (R. clamitans, R. pipiens, and R. sylvatica).

We chose a modelling approach based on the second-order Akaike Information Criterion (AIC_c), as it prioritizes estimation of the effect, its associated precision, and confidence intervals instead of relying on hypothesis testing and P-values (Burnham and Anderson, 2002; Mazerolle, in press). It is especially well-suited for model selection (i.e., finding the best model or variables) because it is not subject to the problems of the more typical model-building procedures such as stepwise, backward, or forward elimination (Burnham and Anderson, 2002). For the field survey data, we considered a set of six models differing in complexity to explain the behavior (i.e., moving or immobile) of amphibians on the road at the approach of a vehicle. We included the categorical variable species (i.e., with ranid frogs as the reference level), as well as the categorical covariates year and period of survey (i.e., May–June vs. July–September) to account for potential temporal variation in amphibian behavior. We also considered the species × period interaction, as the behavior of certain groups might differ with changing weather conditions (e.g., air temperature) across seasons.

For the field experiment, we considered a total of 14 different models. Some of these involved the continuous variables snout–vent length (SVL), air temperature, and time at which the trials of each individual were conducted (log of min after sunset). Other models included the categorical variables car treatment (i.e., with the control as the reference level), species group (i.e., with ranids as the reference level), as well as previous car treatment. The latter was to test whether the effect of a given treatment influenced the response of individuals later during the experiment (i.e., carry-over effects, see Diggle et al., 1994). Finally, some models included interactions (i.e., species × treatment, species × previous treatment), or combinations of the variables mentioned earlier (e.g., air temperature, SVL).

All analyses were conducted with SAS 8.02, using the LOGISTIC and GENMOD procedures. The Hosmer-Lemeshow goodness-of-fit statistic suggested the logistic regression models were appropriate for the data. We computed the AIC_c (or where there was evidence of overdispersion, the QAIC_c) and measures derived from it (i.e., delta AIC_c and Akaike weights) to assess the strength of each model and the effect of each explanatory variable in the models (Anderson et al., 2000; Burnham and Anderson, 2002; Pan, 2001). We obtained the estimates and standard errors for each variable with model-averaging techniques, as described in detail in Burnham and Anderson (2001, 2002) and Mazerolle (in press).

Field Survey

Most of the 2767 live amphibians observed during night-driving surveys remained immobile during encounters with the car (Fig. 1). The model which included the species × period interaction had the most support, being more than 400 times better (i.e., based on evidence ratio of Akaike weights: 0.9966/0.0024) than the model ranked in second place (Table 1). Indeed, the behavior of amphibians differed considerably across species and period. Spring peepers, mole salamanders, and ranid frogs on the road were less likely to remain immobile at the approach of the vehicle early in the season (May–June) than later (July–September), but the response was weaker in spring peepers than in ranid frogs, whereas it was stronger in mole salamanders than in ranid frogs (Table 1). Based on the data for the amphibians observed on the road, the mean predicted probability ± 1 SE of amphibians remaining immobile when encountering a car on the road (i.e., probability calculated from the logistic regression equation; SE obtained from 10,000 bootstrapped samples) was 0.82 ± 0.01: individuals had, on average, an 82% chance of remaining immobile at the approach of a car.

Field Experiment

Overall, the 91 individuals used in the experiment tended to remain immobile in the presence of car-associated stimuli (Table 2, Fig. 2). The mean predicted probability ± 1 SE (calculated as above) of amphibians to remain immobile when faced with the motor and headlights treatment, the motor only, the headlights only, or the control treatments was 0.92 ± 0.01, 0.80 ± 0.01, 0.84 ± 0.01, and 0.48 ± 0.02, respectively. The response of individuals of different species did not vary across the four treatments (i.e., no effect of species × treatment interaction). The car treatment (combination of headlights and sound of motor) elicited the strongest response for all amphibians, with very few individuals moving on the road, followed by the headlights-only treatment, and the motor-only treatment. Amphibians moved more often under the control than any of the car-associated stimuli. Spring peepers were more likely to flee during trials than the other species tested, but behaved similarly relative to the four treatments.

Large individuals were less likely to move at the approach of a car (Table 2). The other covariables did not influence the behavior of individuals. Air temperature and time after sunset were not good predictors of the response of amphibians. Finally, preceding treatments did not alter individuals' behavior later during the experiment (i.e., no carry-over effect).

Behavioral Patterns Across Species

Though our field experiment revealed that spring peepers are more likely to move than other amphibians regardless of the treatment, the data from the field survey indicate that this behavior can change across the season. Indeed, differences may stem from a shift in the motivation of the individuals across the season, as well as body size. During the breeding season, most individuals encountered on the road were undergoing migrations, whereas later in the summer, some individuals may have been either migrating or foraging. Andrews and Gibbons (in press) observed that the motivation of snakes to cross or avoid roads varied across species, and that smaller individuals had a greater tendancy to avoid roads. They also reported that the probability of remaining immobile at the approach of a car and time spent on the road varied greatly across snake species.

All else being equal, we would expect the probability of mortality to decrease as the individual spends less time on the road. Mazerolle (2004) found a decreasing number of spring peepers dead on the road with increasing traffic intensity. He attributed this effect to either an avoidance of busy roads by individuals or a sampling artefact (i.e., small-sized hylid carcasses repeatedly crushed by traffic are difficult to detect, see also Dodd et al., 2004). The behavior of spring peepers we report here may partly explain Mazerolle's (2004) results. This suggests that during certain periods, the species is less susceptible to road mortality than ranid frogs and slow-moving species such as toads and mole salamanders (Hels and Buchwald, 2001).

Road Traffic and Amphibians

Hels and Buchwald (2001) estimated that the probability of amphibians getting killed while crossing a secondary road ranged between 0.34 and 0.61, but increased to 0.89 and 0.98 while crossing a motorway. Amphibian mortality on roads will depend on the traffic intensity (Fahrig et al., 1995; Mazerolle, 2004; Palis, 1994), but also, as our study indicates, on the behavior of individuals relative to car-associated stimuli. This is especially relevant as the probability of amphibians remaining immobile when faced with an approaching vehicle is 0.82–0.92, based on our field experiment and 4-yr survey.

The avoidance or attraction of amphibians towards roads undeniably requires further investigation. Though recent efforts have been deployed to study road-crossing in snakes (Andrews and Gibbons, in press; Shine et al., 2004), birds (St. Clair, 2003), and mammals (Clevenger et al., 2001), comparable data for amphibians remain scarce. For animals already on the road, the pause in activity at the approach of a vehicle will increase the time spent on the road, thus increasing the chance of mortality. In some cases, however, staying on the road could be a better strategy than fleeing at the approach of a car, depending on where the amphibian stops on the road. For instance, a hopping frog might be hit by the bumper or hot exhaust pipe of a passing car.