A comparison of elk-vehicle collision patterns with demographic and abundance data in the Central Canadian Rocky Mountains

Wildlife-vehicle collisions are a widespread phenomenon that are influenced by species behavior, abundance, and road and landscape interactions. The mortality rate of different age and sex classes can buffer or exacerbate how the population responds to vehicle collisions. We evaluated the demographic-specific patterns of elk-vehicle collisions in the Central Canadian Rocky Moun-tains. More females and adults were involved in collisions, but when compared to the sex and age of the population, males and subadults were more prone to collisions in the fall. The fat marrow content (condition) of elk was greater for road-and rail-kill than predator-killed elk indicating that vehicle collisions are an additive source of mortality. As traffic volumes increased elk collisions decreased because elk declined over the study period. Evaluation of long-term datasets can assist in designing mitigation that target the most vulnerable demographics of a population. For example, larger more open wildlife crossing structures have shown to be more suitable for vulnerable demographics such as female grizzly bears, male ungulates, and female ungulates traveling with young. When crossing structures are not practical, demographic-specific information can inform outreach and awareness programs that strive to elicit a favorable response from motorists ultimately avoiding collisions with animals on roads.


| INTRODUCTION
Roads are one of the most dominant forms of humancreated transformations of the world's terrestrial ecosystems.There are an estimated 21 million kilometers (km) of roads in the world and there may be an additional $4.7 million km more roads added by 2050 (Meijer et al., 2018).Approximately 56% of the terrestrial surface of the earth is within 5 km of a road (Ibisch et al., 2016).The effects of roads on ecosystems include mortality from collision with vehicles, the barrier effect created by impeding ecological flows (i.e., the movement of water, organisms, or propagules), and the facilitation of elevated levels of predation, disease, and invasive species (Brady & Richardson, 2017;Laurance et al., 2009).Roads facilitate human access to remote areas and instigate rapid land conversion to more urbanized and industrialized practices (Forman et al., 2003).The impacts of roads are thus greater than the footprint of the road surface and extend into surrounding habitats (Reijnen et al., 1996;Shanley & Pyare, 2011).
Many species of ungulates, for example, moose (Alces alces), elk (Cervus canadensis), and deer are fairly abundant in North America and Europe and their presence on roads pose a major safety hazard for motorists (Biggs et al., 2004;Bruinderink & Hazebroek, 1996;Conover et al., 1995).Huijser et al. (2009) for example estimated the average cost of a collision with a moose, elk, and deer to be US$ 30,760, $ 17,483, and $ 6617, respectively.These costs include damages to vehicle, injury or fatality to motorist, and other indirect costs such as towing, and loss of animal for hunting purposes.
The impacts of roads are notably more severe wherever suitable wildlife habitat, for example, protected areas occur adjacent to a major transportation corridor (Clevenger et al., 2001;Martinig & Bélanger-Smith, 2016).In the Canadian Rocky Mountains, highways bisect several protected areas, imposing a significant source of mortality and barrier effect on the wildlife community (Clevenger et al., 2015;Gunson et al., 2009).As a result, park managers began to recognize and document collisions between vehicles and wildlife, most notably ungulates in the mid-1970s (Banff Bow Valley Study, 1996;Damas & Smith, 1982;Holroyd, 1979).
Prior to installing exclusion fencing and wildlife crossing structures, elk comprised the second highest count of ungulate-vehicle collisions (27%) following mule deer (Odocoileus hemionus) and white-tailed deer (Odocoileus virginianus) (58% collectively) in the mountain parks (Gunson et al., 2009).Given their important ecological role as a top predator for wolves (Hebblewhite et al., 2002), and potential for increased damage to vehicles, elk were prioritized for mitigation solutions to reduce wildlife-vehicle collisions.In the 1980s, Parks Canada initiated the construction of phased wildlife crossing structures and wildlife-exclusion fencing along the Trans-Canada Highway (TCH) in Banff National Park (Ford et al., 2011).These measures have proven most successful in reducing vehicle collisions for deer and elk whose collision rates were reduced by up to 80% (Clevenger et al., 2001;Woods, 1990).The success of these measures led to an expansion of exclusion fence with dedicated crossing structures along highways in both Kootenay and Yoho National Parks (YNPs), and a new wildlife overpass is being designed and planned for along the TCH adjacent to Banff National Park (BNP) in the province of Alberta.
The success of mitigation measures in reducing impacts of roads on wildlife is influenced, in part, on the spatial and temporal patterns of wildlife-vehicle collisions.Collisions tend to be concentrated in spatial and/or temporal locations, or "hotspots" (Mountrakis & Gunson, 2009).These hotspots are formed through a combination of landscape structure near the road, wildlife population size, and traffic volume (Gunson et al., 2011) and may be comprised of a disproportionately portion of the animal population-such as females undergoing overland turtle migrations or inexperienced migrants and dispersers moving to new habitat (Clark et al., 2010;Mumme et al., 2000;Steen et al., 2012).Finally, patterns of wildlife mortality from collisions may be further complicated in populations experiencing density-dependent constraints associated with poor nutrition and vulnerability to predators (Gervasi et al., 2012;Hebblewhite et al., 2002;Olson et al., 2014).Restoration of landscapes impacted by roads must consider these detailed patterns of mortality in order to more accurately emulate natural processes for those individuals most impacted by roads following mitigation.
To support more informed design and implementation of mitigation strategies that are effective for entire wildlife populations impacted by roads, we developed the following hypotheses, predictions and explanations related to the role of sex, age, season, condition, population size, and traffic volume in explaining patterns in elkvehicle collisions (EVCs) (Table 1).We first looked at the demographic groups of elk that are most susceptible to EVCs by comparing these to the adjacent population structure.Next, we evaluated the condition of EVCs in relation to rail-and predator-killed elk.We then evaluated how abundance and traffic volume collectively and independently influence EVCs seasonally and annually.Our findings were then interpreted to support mitigation planning for wildlife in general including elk species evaluated in this study.

| Study area
Our research was carried out in the Central Canadian Rocky Mountains approximately 150 km west of Calgary in southwestern Alberta and southeastern British Columbia (Figure 1).The study area is comprised of the mountain landscapes in Banff, Kootenay, and YNPs, adjacent Alberta Provincial lands, and in the Treaty 7 area (Figure 1).The climate is continental and characterized by relatively long winters and short summers (Holland & Coen, 1983).The roads in the study traverse montane and subalpine ecoregions.Vegetation consisted of open forests dominated by Douglas fir (Pseudotsuga menziesii), white spruce (Picea glauca), lodgepole pine (Pinus contorta), Englemann spruce (P.englemannii), aspen (Populus tremuloides) and natural grasslands (Holland & Coen, 1983).
The geography of the central and eastern portions of the study area is dictated by the geology of the Rocky Mountains.This geography influences the distribution and movement of wildlife in the park.The few large valleys in this area, the Bow Valley being the most prominent, are recognized as critical not only in maintaining the regional-scale east-west movement of large animals (Hebblewhite & Merrill, 2009), but also in providing a vital link between the valleys nested among the front ranges of the park.
We selected highways within Banff, Yoho, and Kootenay National Park (KNP) and Alberta adjacent lands where reliable monitoring of EVCs and elk population sizes have occurred for decades (Table 2 and Figure 1).
The Alberta adjacent lands and Banff and YNPs are bisected by a nationally significant transportation corridor consisting of the Canadian Pacific Railway and the TCH.The TCH lies along the valley bottoms of this mountainous terrain, for approximately 162 km from the intersection of the TCH with Highway 40 to the western boundary of YNP.KNP is bisected by a north-south route called Highway 93 which connects Banff National Park southerly to Radium in the adjacent province of British Columbia.Annual Average Daily Traffic Volumes (AADTV) on all roads in the study area ranged from 2000 on Highway 93 to 16,960 east of BNP in the province in 2005 (Table 2; Parks Canada Highway Service Centre and Alberta Infrastructure, unpublished data).

| Data collection
We used elk mortality records collected year-round by Parks Canada (Banff, Yoho, KNPs) and Alberta Natural

Hypothesis
Prediction Explanation

H age
The proportion of young animals killed in EVCs will be greater than expected than the age structure of the surrounding elk population in the landscape surrounding the road.
Younger elk are inexperienced with roads and traffic (no reference known).

H sex
The sex ratio of EVCs will be the same as the sex ratio of the elk population in the landscape surrounding the road.
Vehicles are non-selective for sex: males and females are equally vulnerable to mortality from EVCs (Olson et al., 2014).

H body condition
Predator-killed elk have poorer health than transportation (i.e., road and railway)-killed elk.
Predators are more selective for body condition than vehicle mortalities (Huggard, 1993).
H season-F EVCs are more likely to occur in the fall than other seasons.
Rutting behavior increase the movement and activity of elk, making them more vulnerable to EVCs (Hubbard et al., 2000) and the addition of the spring calf cohort increases susceptibility of inexperienced individuals to EVCs.
H season-S EVCs are less likely to occur in the spring than other seasons After harsh winter conditions, elk numbers are low (Romin & Bissonette, 1996).
H season-W EVCs are more likely to occur in the winter than other seasons.
Elk have larger home ranges in the winter, increasing liklihood of a road encounter (Anderson et al., 2005).Adverse driving conditions may also make elk more vulnerable in the EVCs in the winter.
H traffic-neg Traffic volume negatively affects EVC frequency.Elk avoid roads because of traffic (Rost & Bailey, 1979).
H traffic-pos Traffic volume positively affects EVC frequency.Elk do not avoid roads (Forrest & St. Clair, 2009) and are killed more often with increasing vehicles (van Langevelde & Jaarsma, 2004).
H abundance Elk abundance positively affects EVC frequency.Elk do not avoid roads and are killed more often as elk abundance increases (Joyce & Mahoney, 2001).
F I G U R E 1 Location of study area and highways used to examine elk-vehicle collisions in the Central Canadian Rocky Mountains  (Larson & Taber, 1980).Body condition was estimated by measuring percent marrow fat content in the femur as measured by Neiland (1970), however a dehydrator was used instead of an oven.The location of the mortality was plotted as a Universal Transverse Mercator (UTM) coordinate to ±100 m on a 1:50,000 scale topographic map.All EVCs occurred on unmitigated sections of highway, therefore, within BNP, only EVCs that occurred along the TCH west of the Banff townsite to the western boundary (TCH in the west zone) where no mitigation (fencing or wildlife crossing structures) was present at the time of the collision were used.This section of highway corresponds to the "west zone" as defined by Hebblewhite et al. (2002) and Phases 3A and 3B during the highway construction periods (Figure 1; Clevenger & Barrueto, 2014).
We obtained AADTV data on national park roads from Parks Canada (Table 2; Parks Canada Highway Service Centre, unpublished data) and on provincial roads from the Province of Alberta (Table 2; Alberta Infrastructure, unpublished data) from 1986 to 1996.We used elk relative abundance estimates from annual classified ground counts in the national parks (KNP and BNP) conducted consistently in spring and fall for BNP and fall-only for KNP between 1986 and 1996 (Woods, 1991;Woods et al., 1996;Parks Canada, unpublished data).Counts were conducted along the road network, for example, TCH in the Bow Valley, which allowed the age and sex of elk to be observed.These counts captured 60% of the elk population (Woods, 1991), as evaluated by sightings of a known number of collared elk (Huggard, 1993).Spring counts occur after cows have calves, and fall counts occur when leaves have fallen to ensure optimal visibility.
Definitions of each variable used in the analysis, its source and range of years are given in Table 3. Season was defined as spring that occurred between March and May; summer between June and August, autumn between September and November and winter between December and February (Table 3).We analyzed data using Microsoft Excel, Statistica™ kernel release 5.5 statistical package (Statsoft ® 2000) and SPSS v13.0 (SPSS Inc., 2004) for all statistical analyses.We screened all data for outliers and normality prior to each analysis.

| Age (H age ) and sex (H sex ) patterns in elk-vehicle collision
We evaluated whether EVCs were biased by sex and age, and then compared the sex and age of EVCs with the adjacent elk population in the "west zone" of BNP where both spring and fall elk ground counts occurred.We tested for independence of sex and age (adult, subadult) classes in EVCs using Chi-squared analyses.
Subadults were defined as young-of-the-year (YoY) and yearlings, while adults were considered 2 years or older.We used records from the complete EVC dataset that had sex and age classes described from all regions between 1986 and 2000 (Table 2).In order to account  for population fluctuations, we grouped the EVC records into three 5-year periods from 1986 to 2000 and did a separate analysis for each period and again for all years.We performed three tests, one each for sex (2 Â 2), age (2 Â 2), and a combined analysis (4 Â 4) and assumed a 1:1 ratio among classes.We used a Chi-squared analysis to compare observed frequencies of EVCs for each sex and age class with expected frequencies derived from an average of spring and fall elk ground counts from 1986 to 2000.
The average of spring and fall elk ground counts would take into account differences in elk demographics from the elk herd migrating between the east and west zones of Banff National Park.Although individuals move between zones (east and west of the town of Banff ) movements are not permanent (McKenzie, 2001).Ground counts classified subadult elk as YoY and spike (yearling males), therefore only YoY and yearling males were included in the observed counts for EVCs.All unmitigated highway sections, that is, no fence or wildlife crossing structures.

| Seasonality of elk-vehicle collisions (H fall , H spring )
We evaluated when EVCs occurred by season and whether this was influenced by sex and age in the spring and fall.We calculated the proportion of total, male, female and unknown EVCs for each month.We then calculated the seasonal mean and standard deviation and tested for differences using a Kruskal-Wallis Analysis of Variance (ANOVA) test followed by a Tukey's range test.Autumn began in September, which is typically the start of the rut.
To test whether sex and age influenced EVCs differently by season, we used a Chi-square test to compare observed EVCs in the spring and fall with each respective seasonal elk ground count in the BNP west zone.

| Condition of elk killed on highways, railways, and by predators (H predation )
We tested for differences in the body condition of elk killed on highways, railways, and by predators in BNP.
Predator-killed elk were confirmed by park researchers, wardens and veterinary personnel.We used a Kruskal-Wallis ANOVA test to determine if percent marrow fat content differed between the three mortality types.

| Traffic volume (H traffic-neg ,
H traffic-pos ) and elk abundance (H abundance ) We evaluated how traffic volume influenced EVC rates per km on each highway collectively from 1986 to 2000 (TCH-province, TCH-west zone, TCH-YNP, and Highway 93 [KNP only]; Table 2) using a Kruskall-Wallis ANOVA.
We then compared the rates of EVCs with traffic volumes using a Pearsons Correlation test.
To evaluate how both elk abundance and traffic volume influence EVCs for each year we conducted a more detailed examination on the TCH in the west zone and Highway 93 where annual elk ground counts (fall only) were available from 1986 to 1996.We used an ANCOVA test to evaluate the influence of elk abundance and traffic volume separately on EVC rate using road-type as our categorical variable.Traffic volume data was log transformed to obtain normal distribution (Zar, 1999).
Elk abundance and traffic volume were negatively correlated to each other on the two roads combined (Spearman's rank correlation, both r > À0.80, p < .05; Figure 3).Therefore, to evaluate the influence of elk abundance independent of traffic volume on EVC rates we used an ANCOVA test with fall and spring as the categorical variable on the TCH in the west zone.Traffic volumes are similar during the spring and fall periods (t test, p > .05,[1996][1997][1998][1999][2000] therefore when using only these two seasons we have essentially controlled for the influence of traffic volume on EVC rates.This analysis was conducted for the TCH in the west zone only, where seasonal elk counts were available, therefore EVC frequencies, not rates were used.

| Seasonality of elk-vehicle collisions (H fall , H spring )
The monthly incidence of EVCs among female and male elk was relatively constant among winter, spring, and summer, but higher during the fall (Kruskal-Wallis test, F 3,44 = 3.48, p = .025;Figure 2).There were significantly more EVCs in fall compared to spring and winter (Tukey's HSD test, p = .0030and p = .0100respectively; Figure 2).
There was a significant interaction between the sex ratio of adult EVCs and season, when compared to the spring and fall population counts (χ 2 3 = 6.77, n = 64, p = .079).In the fall there were more male adult EVCs occurring than female adult EVCs as compared to the population (χ 2 2 = 3.81, n = 47, p = .051).There was a significant difference in the ratio of subadult:adult EVCs compared to the expected age ratio from the population during the spring and fall (χ 2 3 = 29.28,n = 133, p < .0001).The ratio of subadult: adult EVCs in the fall was 300% greater than in the spring.

| Condition of elk killed on highways, railways and by predators (H predation )
Between 1990 and 1998, 397 elk carcasses were collected in BNP from mortalities associated with highways (n = 102), railway (n = 133), and by predators (n = 162).There was a significant effect of elk condition on the three types of mortality (Kruskal-Wallis test, F 2,397 = 9.45, p < .0001).Percent marrow fat content of highway-and railway-killed elk was not significantly different, but both had a significantly greater percentage fat content than predator-killed elk (Tukey's HSD test, p < .05).

| Traffic volume (H traffic-neg , H traffic-pos ) and elk abundance (H abundance )
EVC rates were significantly different between each road type (Kruskal-Wallis test, H 3,60 = 14.99, p = .001;Table 2) and were positively correlated to traffic volume (Pearson's correlation = 0.99).The mean EVC rate ± SD from highest to lowest for all four roads was along the TCH in the province (7.26 ± 7.86); TCH in the west zone BNP (3.73 ± 11.57);TCH in YNP (2.06 ± 4.20) and highway 93, 1.26 ± 9.21 (Table 2).In a more detailed analysis, an opposite relationship was found between traffic volume and EVC occurrence.Both EVCs and elk population counts decreased on the TCH in the west zone and Highway 93 combined, by 89% and 83%, respectively, while traffic volume increased by 22% between 1986 and 1996 (Figure 3).
Both ANCOVA models evaluating elk abundance (p < .0001,R 2 = 0.75) and traffic volume (p < .0001,R 2 = 0.82) independently on EVC rate on Highway  93 and TCH in the west zone were significant (Table 7).
As elk abundance decreased, EVCs significantly decreased and this relationship was consistent between roads (Table 7 and Figure 4).As traffic volumes increased EVC rate significantly decreased and this relationship was more evident on highway 93 than on the TCH in the west zone (Figure 5).We found a significant positive influence of elk abundance on EVC occurrence in the seasonal model (Table 7; p < .0001)when traffic volume remained constant along the TCH in the west zone.More specifically, EVCs increased at a greater rate during the fall season when elk abundance numbers were significantly higher (t = À2.09,n = 16, p = .045;Figure 6).

| DISCUSSION
Our study provides novel and rare details on the links between the demographic structure of road mortalities and an adjacent large mammal population.While it is well established that roads can have negative impacts on biodiversity, it is less clear if road mortality is selective for particular individuals in a population.Some animals make greater contributions to population growth than others-for example-the survival of adult female ungulates is a critical parameter in wildlife management (Gaillard et al., 1998).Some animals may be vulnerable to mortality from multiple sources, that is, compensatory mortality, such that road-kill has minimal impacts on The interaction between elk population abundance (elk/ km) and road type for Highway 93 and Trans-Canada Highway in BNP.As abundance increases, EVCs increase and this relationship is not significantly different between roads.EVCs, elkvehicle collisions The interaction between annual average daily traffic volume (AADTV) and elk-vehicle collision rate on two roads: Highway 93 and Trans-Canada Highway in BNP.As traffic volumes increase EVC rate decreases and this is more apparent on Highway 93.AADTV, annual average daily traffic volume; EVCs, elk-vehicle collisions populations (Hebblewhite et al., 2002).Finally, some animals may have higher value for people than others, such as harvestable-aged male ungulates (Arnett & Southwick, 2015;Mysterud et al., 2002).If road mortality is selective for particular individuals, it may either exacerbate or buffer the effect of roads on a population and the perceived loss in ecosystem services to stakeholders (Knoche & Lupi, 2007).Here, we tested hypotheses related to the role of sex, age, season, population size, and traffic volume in shaping spatial and temporal patterns in EVCs.Given the disproportionate role that some individuals make in the growth of wildlife populationsparticularly adult females (Wilmers et al., 2020)-our study sheds important insights on mechanisms underlying the restoration and maintenance of biodiversity in human-occupied landscapes.

| Sex and age effects
Compared to males, we found significantly more female elk were killed on roads in the Rocky Mountains.We anticipate that this result is largely based on a femalebiased sex ratio in the local population that has been observed since at least the 1970s (Flook, 1970;Holroyd, 1979).Similar sex ratios have been found for studies that examined road-kill in mule deer and whitetailed deer in the United States (Allen & McCullough, 1976;Bellis & Graves, 1971;Olson et al., 2014).However, when we compared the proportion of female and male road-kill to the sex ratio in the population directly, there were more males being road-killed than expected by chance.Similar male-biased road mortalities have been found for moose in Newfoundland (Joyce & Mahoney, 2001).Male elk-like other ungulates-may be more vulnerable to road mortalities because of their larger home range sizes and greater vagility during particular seasons.Specific to the Bow Valley, male elk migrate more than females to access rutting grounds, mates, and foraging areas (Hebblewhite et al., 2002;Woods, 1991).
We found that more adult elk are killed on the highways than young animals.We anticipate that this result is largely based on an adult-biased age ratio in the local population (Parks Canada, unpublished data).However, when taking into account the age structure of the local population, subadults were more likely to be killed by vehicles.This pattern of age-selectivity is similar to earlier results found in the study area (Bottini, 1987) and for yearling male white-tailed deer in the lower Yellowstone River in Montana (Dusek et al., 1989).It is likely that subadults are more vulnerable to collisions than adults because they lack the necessary experience to navigate vehicular traffic.In addition, calves may be more prone to collisions during the mating season when cow movements are disrupted by males.

| Seasonal effects
Seasonal variations in ungulate-vehicle collisions have been noted since some of the earliest published records of road-related mortality of wildlife (Allen & McCullough, 1976;Bellis & Graves, 1971;Joyce & Mahoney, 2001;Pennsylvania Game Commission, 1969).The collision frequencies tend to increase from early spring to fall, peaking in October and November during the rut and/or hunting season (Bruinderink & Hazebroek, 1996;Creech et al., 2019;Hubbard et al., 2000;Neumann et al., 2012).
The interaction between elk abundance and elk-vehicle collision frequencies between spring and fall along the Trans-Canada Highway in BNP This fall peak has been attributed to increased ungulate activity level associated with migration and breeding behavior, which brings ungulates in contact with the road network more frequently (Ager et al., 2003;Woods, 1991).More EVCs occurred in winter in the Jemez Mountains plateau region of New Mexico and was attributed to migration behavior of elk during heavy snowfall in the adjacent higher terrain (Biggs et al., 2004).However, migration may not be an important factor in other areas.Creech et al. (2019) found little difference in seasonal WVC patterns between western Montana counties where ungulates tend to exhibit migratory behavior and eastern counties where migratory behavior is less common.
Notably, we found that subadults were killed more often in the fall than in the spring.This age-related pattern is likely attributed to changes in seasonal population demographics more than other factors such as traffic volume because the spring and fall traffic volumes are similar on the TCH in the Bow Valley.There is a doubling of the elk population in the fall due to the addition of a relatively inexperienced spring calf cohort into the population.This coupled with possible reduced population numbers after harsh winter conditions in the spring would further explain seasonal changes in sub-adult EVCs (Romin & Bissonette, 1996).

| Condition of elk
Elk populations have several sources of mortality, with predators and collisions with vehicles having significant effects in some populations (Gagnon et al., 2019;Hebblewhite et al., 2002).We found elk killed by vehicles and trains tended to be in better body condition than elk killed by natural predators similar to that found by Huggard (1993)-a pattern that is suggestive of additive mortality.When road mortalities occur for animals that would have otherwise been killed by natural predators or disease (i.e., because they have poor body condition), then the population is experiencing compensatory mortality (Boyce et al., 1999;Fowler, 1987).For example, O' Gara and Harris (1988), found that predators such as cougars and coyotes killed deer in good condition, while deer killed by vehicles were in poor condition in western Montana.They attributed this finding to sick and malnourished deer moving to and congregating in valley bottoms along roadsides clear of snow in winter months.
Additive mortality occurs when road-killed elk have a fairly high probability of survival and this may have been contributing to the decline of elk population size in Banff National Park seen in the early 2000s (Parks Canada, unpublished data).The recolonization of wolves to Banff and KNP in the 1980s is also a likely contributor to the decline in elk abundance and subsequently EVCs.Wolves reduced annual survival rates for elk from 0.89 to 0.61 from the central zone (no wolves) to the west zone in BNP (high wolf density), respectively (Hebblewhite et al., 2002;McKenzie, 2001).Following the installation of road mitigation measures in 1997 (phase 3A) and further west in 2009 (phase 3B), elk populations have stabilized and EVCs have declined to almost nothing (Clevenger & Barrueto, 2014).

| Effects of traffic volume and abundance on road mortality
EVCs were positively related to traffic volumes on a broad geographic scale in the Rocky Mountains, however when dissecting this further over time, the collision rate (mean number of EVCs per km of road) decreased as traffic volume increased.Unexpectedly, traffic volume negatively influenced EVC rates on both highways and this relationship was more evident on Highway 93.As expected, abundance positively influenced the occurrence of EVCs, especially in the fall.These findings indicate parameters such as traffic volume need to be statistically analyzed at different resolutions, and additional tests should evaluate how abundance and traffic volume interact and influence road-kill by road-type and region (Clevenger et al., 2015).
Wildlife-vehicle collisions could be assumed to increase with the abundance of animals and vehicles since these two forces, animal presence and vehicles on a road simultaneously create a collision (Fahrig et al., 1995;Philcox et al., 1999;Romin & Bissonette, 1996;Seiler, 2004;van Langevelde & Jaarsma, 2004).However, animals may avoid roads because of traffic (Gagnon et al., 2007) or because of the road surface (Ford & Fahrig, 2008;Shine et al., 2004) or other reasons.If road avoidance behavior co-varies with traffic (Seiler, 2005;Thurfjell et al., 2015), then increasing traffic volume may have minimal effects on road kill rates (Fahrig et al., 2001).
Another explanation for the negative effect of traffic volume on EVCs is that mortalities may eventually contribute to population declines (but see Munro et al., 2012).Other studies have shown that road-related mortality depressed populations to the point that road-kill rarely occurred (Eberhardt et al., 2013;Zimmermann et al., 2017) in areas of high traffic volumes.Given the corroborating evidence that EVCs are comprised of a high proportion of adult female elk in good body condition, the additive mortality hypothesis warrants further investigation as a partial explanation for the decline of elk.AND RECOMMENDATIONS We found that by incorporating population structure into the analysis of vehicle collisions with elk, new perspectives on the relative vulnerability of animals to mortality were revealed.These perspectives emerged from information on local (i.e., near the road) wildlife populations that are typically unavailable for most road mortality studies (Olson et al., 2014;Ramp et al., 2005).These insights add to the growing body of evidence that demographic-specific variation in behavior and vulnerability to roads needs to be considered in order to maximize populationlevel benefits of mitigation (Ford et al., 2017) especially for rare and declining populations vulnerable to additive mortality e.g., freshwater turtles (Steen & Gibbs, 2004).
To address demographic-specific vulnerabilities, mitigation measures such as crossing structures that are designed for vulnerable animal groups, for example, females traveling with young and male elk may be more effective for the population.There is a growing body of research that has showed greater frequencies of passages by wildlife assemblages at overpasses (Clevenger & Barrueto, 2014;Simpson et al., 2016), and that Grizzly bear (Ursus arctos) females and offspring in Banff National Park selected for more open crossing structures such as open-span underpasses and overpasses (Ford et al., 2017).In addition, a study in Ontario, showed that male moose and deer used a wildlife overpass more than expected when compared to the surrounding population (Eco-Kare International, 2020).
When more permanent measures such as crossing structures and exclusion fencing are not practical for a specific site, motorist-focused campaigns are often employed.These campaigns can be greatly improved with informed messaging that target specific animal demographics in specific regions.Campaigns for ungulates may be timed to fall periods when male elk and new calf cohorts are vulnerable to collisions.Freshwater turtle populations are particularly vulnerable to adult female-biased road-kill because of delayed sexual maturity (Congdon et al., 1993) and because females move overland during nesting migrations (Steen et al., 2012).Therefore, campaigns in northeastern North America, need to remind motorists to watch for, avoid and help turtles move across roads safely in June when adult female turtles move overland to nest.
Patterns of traffic volume and population abundance on WVCs, can help managers predict long-term viability of wildlife populations and assess when and where wildlife mitigation is most effective for targeted species.In the past road agencies have often installed mitigation where road-kill is the highest and this may ignore locations where road-kill has already depressed populations.Collective evaluation of long-term datasets can help tease out mechanisms for population declines and assist with appropriate recovery efforts prior to a local population becoming depleted near roads.

AUTHOR CONTRIBUTIONS
Kari Elizabeth Gunson led the data acquisition from Parks Canada and other researchers, conceptual design, data analysis, and write-up of the manuscript; Anthony Paul Clevenger contributed to data analysis and write-up of the manuscript.Adam Thomas Ford provided clear direction for analytical methods and data interpretation, and contributed peer-review and critical references for later drafts of the manuscript.
T A B L E 1 Hypothesis and predictions tested to explain elk-vehicle collisions in the Central Canadian Rocky Mountains a 2005 annual average daily traffic volume.Data from Parks Canada Agency, Banff National Park and Alberta Transportation, Edmonton, Alberta.b The 1999 summer average daily traffic volume.Data from Alberta Transportation, Edmonton, Alberta.

F
I G U R E 3 Standardized annual number of EVCs, with AADTV for both highways and estimated standardized elk abundance along highway 93 and the Trans-Canada highway in BNP, in the Central Canadian Rocky Mountains, 1986-1996.AADTV, annual average daily traffic volume; EVCs, elk-vehicle collisions T A B L E 4 The number of elkvehicle collisions by sex and age class in the Canadian Rocky Mountains from 1986 to 2000; assumed a 1The number of elkvehicle collisions by males and females compared to the expected frequencies from ground counts averaged for the spring and fall periods (Parks Canada, unpublished data) T A B L E 7 Analysis of covariance of elk abundance and traffic volume (continuous variable), and season and road (categorical variables) on elkvehicle collision rate on the Trans-Canada highway in the west zone and Highway 93 Source d.f.Type III SS F Pr. Type I error Elk abundance model (Figure 4); N = 22 observations; model R 2 = 0(Figure 5); N = 22 observations; model R 2 = 0.82 The location, general characteristics, and elk-vehicle collision rate from 1986 to 2000 on the four highways compared in the Central Canadian Rocky Mountains T A B L E 2