The authors have declared that no competing interests exist.
Conceived and designed the experiments: FA. Performed the experiments: FA. Analyzed the data: FA. Contributed reagents/materials/analysis tools: FA JT. Wrote the paper: FA CG SLP JT AC MSR. Responsible for the dataset analyzed: CG.
Efforts to reduce the negative impacts of roads on wildlife may be hindered if individuals within the population vary widely in their responses to roads and mitigation strategies ignore this variability. This knowledge is particularly important for medium-sized carnivores as they are vulnerable to road mortality, while also known to use available road passages (e.g., drainage culverts) for safely crossing highways. Our goal in this study was to assess whether this apparently contradictory pattern of high road-kill numbers associated with a regular use of road passages is attributable to the variation in behavioral responses toward the highway between individuals. We investigated the responses of seven radio-tracked stone martens (
The negative impacts of roads on wildlife have long been recognized
Measures to reduce WVC and mitigate the barrier effect are diverse
The life stage and state of an individual can affect its behavioral response to both the road and to the mitigation actions, such as transient individuals avoiding interactions with residents
This knowledge is particularly important for medium-sized carnivores that are especially vulnerable to the negative impacts of roads
Our goal in this study was to assess whether this apparently contradictory pattern of high road-kill numbers associated with a regular use of road passages is attributable to the variation in behavioral responses toward the highway between individuals. We reanalyzed the tracking data of seven stone martens (
Martens were tracked in the Mediterranean region of southern Portugal (39°38.154′N, 8°12.128′W), an area dominated by cork-oak woodlands (
B: Duration (2008–2009) of tracking nights for each marten (each bar is one night) with “LC” indicating loss of contact and “WVC” indicating a confirmed WVC (corpse recovered). Apparent home range overlap of F1 with M1 and M4, and F5 with F7 correspond to distinct periods.
The stone marten is a medium-sized carnivore occurring across parts of Asia and Europe
We selected seven individuals (two males and five females) that had sufficient data from the dataset used by
Marten | Number of tracking sessions | Tracking hours | Number of fixes | Mean time between relocations (min) |
F1 | 28 | 122 | 202 | 51±22 |
F3 | 28 | 190 | 300 | 44±44 |
M1 | 19 | 137 | 238 | 39±22 |
M4 | 5 | 36 | 64 | 37±13 |
F2 | 19 | 118 | 213 | 39±13 |
F5 | 19 | 124 | 205 | 41±16 |
F7 | 18 | 124 | 203 | 43±21 |
Tracked time and time between relocations includes only tracking sessions with at least two successful relocations. Individuals are sorted by whether they crossed the highway (F1, F3, M1, and M4) or not (F2, F5, and F7).
We estimated marten utilization distributions (UD) with biased random bridges (BRB), a movement-based kernel density estimation method
We used the nonlinear regression model described by
In these models, μ is set constant (i.e., independent of the animal-to-object distance,
In the ‘responsive’ model the decay of
We identified highway crossings as pairs of consecutive marten locations during the same tracking session recorded on opposite sides of the highway. For each marten, we counted the number of crossings and calculated the utilization distribution using these pairs of locations (UDcross) also with the BRB method, similar to the approach used by
We used a null model approach to determine the influence of the highway on marten utilization distribution and highway crossing patterns. Note that the models used to analyze the response angles already incorporate a comparison with a ‘no-response’ model. Null models are pattern-generating simulation models that deliberately exclude a mechanism of interest (for our purposes, the presence of a highway), and by using randomization procedures allow the user to test the importance of that mechanism in observed patterns
Simulated movements were parameterized with the attributes of the observed data (i.e., the number of tracking sessions, locations, step lengths, and utilization distribution boundary), but the simulated agent was naïve to the presence of the highway. For each tracking session, an agent (i.e., simulated marten) started from an observed resting site, chosen randomly, and then moved the same number of steps whose length followed the observed step lengths' sequence. The agents' successive location must fall within the home range boundary at a random direction from the previous location. Therefore, simulations follow a constrained random walk which has been successfully used in previous road ecology studies
For each response considered - utilization distribution, frequency and location of highway crossings - we performed a set of 1000 simulations, per marten. Each set of simulations was used to generate a frequency distribution, from which the confidence intervals of the observed response were estimated. Based on likelihood significance tests, we considered an effect of the highway if the observed parameter fell outside the 5–95% percentiles of the simulated parameter distributions. The model was built using NetLogo 4.1.3
Prior to analysis, we investigated marten habitat selection in the study area using a weighted compositional analysis as described by
For each marten home range, we calculated the sum of the probability values of all cells of the UD for the land cover classes ‘forest’ and ‘agricultural’, and considered these proportions to be the ‘available’ habitat. The ‘used’ habitat was estimated by calculating the proportion of locations that fell within forest or open per marten. The test was performed using the command ‘compana’ in the R package ‘adehabitatHS’(version 2.15.1)
We found no evidence for marten habitat selection (Λ = 0.72, p = 0.18) and so excluded land cover information from further analyses. This was not unexpected, as stone martens in the region are not forest-specialists
The UDs of martens whose home ranges overlapped with the highway (F1, F3, M1 and M4) revealed inconsistent patterns of space use near the highway (
Marten home-ranges were computed as the 95% isopleth of the UDs. Middle and right columns: black areas suggest areas where martens spent more and less time than expected by chance (95% or 5% of simulations), respectively. The highway is shown in each plot by the dotted line. Images are scaled (among martens). Individuals are sorted by whether they crossed the highway (F1, F3, M1 and M4) or not (F2, F5 and F7).
Marten responses to the highway were highly variable between individuals. The response angles of five of the seven martens were best predicted by the ‘responsive’ model, suggesting most martens showed a significant response to the highway (
Marten | θ1 | θ2 | ?2 | |
F1 | 0.24 | 0.33 | 0.000 | 8.7 (0.00) |
F3 | 1.16 | 1.32 | 0.045 | 2.6 (0.11) |
M1 | 2.30 | 0.84 | 0.013 | 2.3 (0.13) |
M4 | 1.61 | 1.09 | 0.008 | 5.6 (0.02) |
F2 | 2.96 | 2.55 | 0.004 | 6.3 (0.01) |
F5 | 3.08 | 10.75 | 0.004 | 9.3 (0.00) |
F7 | 2.62 | 6.66 | 0.005 | 15.1 (0.00) |
Last column stands for the comparison of movement responses to highway proximity, where ‘no-response’ and ‘responsive’ nonlinear models are compared by likelihood ratio test (degrees of freedom = 1). Between brackets is the p-value for the test. Individuals are sorted by whether they crossed the highway (F1, F3, M1, and M4) or not (F2, F5, and F7).
Marten crossing patterns varied between individuals, suggesting no general pattern in highway crossing frequency or crossing locations (
Grey areas represent the percentile (5–95%) envelope of reference from the simulated datasets. Dots outside of the percentiles suggest the individual crossed less often (left) or more often (right) than expected.
Bottom: the observed probability of crossing the highway at each road segment (UDcross at highway location; solid line). Grey areas represent the 5–95% percentile envelope of reference from the simulated datasets. White (black) arrows indicate highway segments with higher (lower) use than expected. Points indicate road passage location. For each marten, the highway segment in the upper-half of the figure is projected in the X axis from the bottom picture.
Our individual-based analytical framework improved our understanding of how martens respond to the presence of a highway. All martens demonstrated some level of influence of the highway proximity for each of the behavioral responses we considered. However, their responses were more variable than expected
We were able to provided new insights into the apparently contradictory results of previous work held in same study area, where martens were frequently killed on the highway
As previously described by
Overall, although the sample size of our dataset is less than ideal, we show that martens can exhibit a variety of responses to the highway, especially in their propensity to cross the highway and to use crossing structures to do so. We assume that martens F1, F2 and F3 were residents with well-established territories since they were tracked for long periods, having stable home range areas
We also believe that M4 was dispersing through the region as he was not detected before being captured, despite the continuous and intense trapping effort
To effectively mitigate the negative effects of roads at the population level we must understand the processes that affect the movements of individuals and the variability between individual responses to roads and existing mitigation
An important research question remains: how many individuals use the passages? If only a few individuals regularly use the same passage, as our results suggest, then the effectiveness of these structures could be overestimated
(DOCX)
(NLOGO)