| "v1_text": "results : We conducted surveys for 737 km of highway (H) and 118.5 km of transects (T), that yielded the following number of track counts: deer = 970 H/887 T, elk = 575 H/152 T, moose 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t = 65 H/59 T, coyote = 58 H/111 T, bobcat = 6 H/11 T, cougar = 1 H/11 T, wolf = 0 H/10 T, fox = 3 H/2 T. No tracks were found for lynx, marten or wolverine (raw data and R model input files can be found in Data S1). methods : discussion : We determined that Hwy 3 posed a severe movement barrier to the local mammal community. Although each investigated species differed in the landscape variables associated with preferred and avoided crossing sites, we used a multi-scale approach to identify locations along the highway where mitigation measures may benefit all species in the large mammal community. Below we address our earlier questions and discuss the implications of our finding that multi-scale habitat assessments may be necessary to accurately predict the most effective locations for highway crossing structures (e.g., culverts and overpasses) or other mitigation measures. Permeability estimates for both carnivores and the majority of ungulate species considered were extremely low across the highway (Table 2), indicating that Hwy 3 likely acts as barrier to animal movement. Although permeability estimates for elk were comparatively high (likely due to herding behavior, whereas tracks for all other species tended to be solitary or in small groups), averaged estimates for all ungulates and the entire mammal community suggest that movement by large-bodied mammals is highly restricted across the highway. Likewise, track accumulation curves (Figure 2) indicate that for each species group considered, certain areas of the highway may rarely or never be crossed, posing large limitations to population connectivity across Hwy 3. This finding is consistent with previous estimates of wildlife permeability across a similar highway through the Rocky Mountain Range of Alberta, Canada (Alexander, Waters, & Paquet, 2005). Such low permeability across the highway suggests a severe threat of habitat fragmentation to the mammal community, which could result in decreased gene flow across the road barrier, and ultimately to lower 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t population viability in the region (Mader, 1984; Epps et al., 2005). These results indicate a need to accurately identify locations for potential mitigation measures along roads such as Hwy 3 to facilitate the movement of individuals across the highway and reduce this barrier effect (Harrison & Bruna, 1999; Haddad et al., 2003; Crooks & Sanjayan, 2006). By incorporating both highway and transect predictions simultaneously, we aimed to identify locations for potential mitigation measures that represent both preferred crossing sites as well as preferred approach habitat up to 1km from the highway. We determined that the landscape variables associated with preferred/avoided crossing sites differed for many of the mammal groups considered (Tables S1, S2). In all cases, noise generated from vehicles travelling on the highway could contribute to road avoidance by large mammals (Forman & Alexander, 1998; Jaeger et al., 2005; Barber, Crooks, & Fristrup, 2010). However, numerous studies on movement across roads by large and small mammals have found no consistent response to noise levels, and suggest that habitat characteristics surrounding crossing sites play a larger role in animal movement than individual tolerance to noise levels (McGregor, Bender, & Fahrig, 2008; Iglesias, Mata, & Malo, 2012). For instance, carnivores tended to avoid residential areas along the highway as well as open areas with low shrub cover (Tables S1, S2), consistent with previous studies (e.g. Mech, 1995). While elk and deer did not avoid these landscape features, these two species exhibited dissimilar patterns of habitat and crossing-site preference, consistent with their different habitat requirements (Johnson et al., 2000). These differing results per group indicate that a clear set of conservation goals for each species as well as the community as a whole must be established before mitigation measures are implemented to facilitate highway crossing (e.g. Beier, Majka, & Spencer, 2008). We used multi-model inference and model averaging to identify locations of preferred crossing sites for all mammal species considered, which would likely serve as the most effective locations for mitigation measures aimed at increasing mammal permeability across the highway. 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Cumulative scores of preferred/avoided landscape variables along both the highway and transect data sets indicate that preferred crossing sites tended to be within close proximity of water and longer stretches of unpaved road (Table 5). Crossing-specific scores indicate a preference for longer stretches of paved roads, and approach-specific scores suggest preference for areas of high crown cover with abundant broadleaf trees, respectively. Although this approach may reduce the efficiency of predicting highway crossing sites for certain focal species, community-level approaches are increasingly advocated as a more efficient means of implementing wildlife linkages across barriers such as major roads (Beier, Majka, & Spencer, 2008). To accomplish this goal, we applied an exhaustive model approach incorporating four separate distributions of abundance for each mammal group along Hwy 3. In only 4 of the 8 cases considered was preselection of the y-distribution successful, indicating that an exhaustive modeling approach incorporating multiple distributions may be necessary when the goal is to identify and predict preferred crossing sites based on limited data and uncertainties regarding which abundance distributions are most applicable to free-living animal populations. By adopting the approach described here, researchers may be able to extract more information from highway crossing data than could otherwise be gained from applying predefined and potentially inaccurate abundance distributions. Further, the best-supported distribution differed for each species; while ZINB and NB were the most commonly supported distributions, NB, ZIP and P each received the best support for at least one data set (highway versus transect). These results once again highlight the need for future studies to consider the unique habitat requirements of each species within mammal communities when developing mitigation strategies, but that those strategies which provide the greatest benefit to the largest number of species should be given priority for implementation. To establish conservation-based goals for large mammals along roads such as Hwy 3, further consideration must be given to whether the spatial scales at which habitat characteristics 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t are measured match the spatial scales at which the animals select preferred/avoided crossing sites. We determined that in 6/8 cases, a combined approach to modeling preferred crossing sites (incorporating remotely sensed and hand-digitized predictors) resulted in the best supported model. Further, utilizing multi-scale remote sensing-derived predictors always resulted in better model support than utilizing only hand-digitized predictors for each species and data set considered. Thus, our results indicate that while a combined approach may represent the most informative method for predicting landscape variables of preferred mammal crossing sites, freely-available macro-habitat data such as those generated through remote sensing may be more useful than labor-intensive micro-habitat assessments when time and budgetary constraints on data collection are imposed. Previous studies investigating habitat occupancy in birds have found similar results (e.g. McClure, Rolek, & Hill, 2012; Meiman et al., 2012), highlighting the increasing usefulness of remote sensing in evaluating localized questions in conservation and community ecology. The goal of our study was to identify locations along Hwy 3 where mitigation measures might increase connectivity across the highway for all species in the mammal community. Although we do not currently have data on which mitigation measures may be the most effective on increasing permeability in this system, previous studies investigating the costs/benefits of different mitigation strategies at the community level (e.g. Clevenger & Waltho, 2000, 2005) indicate that a diversity of crossing structures of different sizes may best serve large mammal communities. Because our permeability estimates were based on snow tracks and not on data for the entire year, there is the potential for our results to only be applicable for winter months. Further, because our permeability estimates are based on transects with a mean distance of 175 m from the highway, we likely overestimate permeability in certain cases by not considering the density of animals in areas further away from the highway. For instance, Dickson & Beier (2002) determined that cougars typically avoid high speed roads at a distance of 500m \u2013 1km and more 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t generally, mammal populations might be influenced by human infrastructure up to about 5km (Ben\u00edtez-L\u00f3pez, Alkemade, & Verweij, 2010). Although conducting further transects at a greater distance from the road may improve estimates of habitat preference for each species along Hwy 3, we believe our methods represent a realistic investigation of the types of habitat used by animals approaching and ultimately crossing the road, which may help inform strategies for implementing crossing structures. A potential limitation to our approach of determining the most appropriate locations for multi-species crossing structures is that preferred landscape traits differed among groups, indicating that some species would benefit less from crossing sites that serve the majority (for species specific preferences see Figures S1-S4). While the specifics of which species should be given priority in such an instance will depend on the conservation goals of managers, our method presents a potentially viable way of increasing highway permeability for multiple species, and ultimately improving connectivity and population viability for mammal communities along major roadways. Although our study was limited to one section of highway, its importance as a wildlife corridor suggests that our approach may be widely applicable to other areas where roads bisect important wildlife habitat. In situations where managers are capable of implementing mitigation measures aimed at increasing cross-road permeability for multiple mammal species, future studies should seek to evaluate the efficiency of this method over traditional single-species approaches. Specifically, to verify the effectiveness of our approach compared to a single- species mitigation strategy, managers would ideally implement our method in areas where traditional mitigation approaches have been in place for a number of years. By directly comparing permeability values before and after the implementation of a multiple-species mitigation approach, we may gain further insight into benefits of community-level conservation planning. Finally we would like to acknowledge that our modeling approach only constitutes one possible way of drawing inference about highway approach and crossing behavior of the 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t investigated mammal community. Here, we provide a flexible but somewhat restrictive framework for predicting animal abundance. Though there is always uncertainty surrounding model choice when using a multi-scale approach, extra caution should be used when basing model choice on \u2018stepwise\u2019 procedures and using p-values to exclude certain models from a set. The use of AIC to rank models is currently widely applied in the literature and is assumed to be valid, but this approach only gives a relative measure of fit for comparing models. AIC does not provide a measure for predictive ability of a model, which should ideally be tested against additional data. Finally, alternatives to model averaging such as a reversible jump MCMC approach (Green, 1995) could be employed to compare results and further improve robustness of analysis. study area : Our study was conducted along Southern Trans-Provincial Highway 3 (hereafter Hwy 3) between the towns of Creston and Cranbrook, in south-eastern British Columbia, Canada (Fig. 1). The study area is located in the Purcell Mountain Range, which ranges from 620m to 2,087m in elevation, and is comprised of Interior Cedar Hemlock and Interior Douglas Fir Biogeoclimatic zones (Meidinger & Pojar, 1991). We chose this study area for its ecological importance as a trans-boundary priority area (Yellowstone to Yukon Conservation Initiative, 2013) that connects small populations of carnivores such as grizzly bears (Ursus arctos horribilis) and Canada lynx (Lynx canadensis) along the Canada \u2013 USA border. Hwy 3 bisects this important corridor, possibly leading to negative effects on the connectivity of this movement corridor for mammal populations. The average annual traffic volume (AADT) for this highway section was 3050 cars/day in 2007, with a seasonal (December to March) average of 2020 vehicles/day (British Columbia Ministry of Transportation and Infrastructure 2010). 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t conclusion : Roads such as Hwy 3 represent severe barriers to animal movement and pose a major threat to wildlife habitat, but few studies investigate how or where to implement mitigation measures at the community level. We identified areas along the highway with habitat features of preferred crossing sites for eight species of large mammals, representing locations where mitigation measures may have positive effects for all species investigated. We determined that a combined approach incorporating both remotely sensed and hand-digitized landscape variables best predicted crossing site preference for most species, but that remote sensing data was always better than hand-digitized values when utilized separately. Our results indicate that a multi-scale approach may be necessary when identifying areas to implement mitigation strategies across roads, as differing habitat requirements for members of the mammal community may limit the usefulness of single-species, single-scale approaches. Acknowledgments 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t We thank D. Quinn for logistical support throughout data collection, S.M. Alexander for helpful advice on data collection and analysis, and W. Desch for initial methodological and statistical advice. F. Suppan, P. Beier, J. Jenness and K. Crooks provided feedback on GISanalyses, and A.E. Passmore helped edit a previous version of this manuscript. We thank G. Stewart, P. Beier, and two anonymous reviewers for comments and suggestions on an earlier draft of this manuscript. data collection : We monitored species movement through the study area by recording tracks in the snow where animals attempted to cross Hwy 3, as well as along ten transects approaching the highway, set back from any highway right-of-way (distance from transects to highway ranged from 10 to 900m, mean 175m). We pre-defined our transects as survey lines marked with flagging tape, roughly parallel to the highway. Highway and transect tracks were recorded over two winter seasons, January to March 2007 and December 2007 to February 2008 (all observations recorded by RS). Highway and transect track surveys were conducted using methods similar to Van Dyke, Brocke, & Shaw (1986), and Alexander, Waters, & Paquet (2005). Briefly, we conducted highway crossing attempt surveys along a 95km length of Hwy 3, at least 12 hours after the last snowfall. Each survey was conducted from a moving vehicle with a speed of approximately 10-15 km/h. When a track was observed, the investigator stopped the vehicle and conducted an on-foot inspection to identify the track. In total, we investigated tracks for 12 mammal species: coyote (Canis latrans), fox (Vulpes vulpes), wolf (Canis lupus), cougar (Puma concolor), bobcat (Lynx rufus), lynx, marten (Martes americana), wolverine (Gulo gulo), elk (Cervus canadensis), moose (Alces alces), white-tailed and mule deer (Odocoileus virginianus and Odocoileus hemionus, respectively). When we were uncertain of the identity of a track, we recorded track pattern measurements, took photos and later consulted field guides (Sheldon, 1997; Elbroch, 2003) for identification. Data at a total of 463 crossing sites were georeferenced with a handheld, Garmin eTrex Summit GPS receiver (WGS 1984, \u00b1 10 \u2013 40 m). If multiple tracks were found for one species at a single crossing area, we recorded the total track count. We also recorded the success of a crossing attempt, here defined as the presence of a continuing set of tracks on the opposite side of the road. When tracks of the 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t same species were found within 300 meters of a crossing site, it was not recorded as an individual crossing attempt, but rather as a possible repeat crossing of the same individual (Alexander, Waters, & Paquet, 2005). Surveys were suspended when continuous heavy snowfall covered tracks during data collection. Transects were established off-road in suitable areas close to the highway. Suitability was contingent upon minimal disturbance from residential areas, and no barriers to observer access (i.e., lakes, steep terrain, fences or private property). Seven transects had a linear distance of 1 km, while one was 2 km (Transect 6) and one was 5.4 km in length (Transect 10). Only the first kilometer of transect 10 was surveyed during the second season of data collection, and this was classified as Transect 9 for ease of data handling. We recorded tracks of the same species according to the protocol of the crossing attempt surveys, and georeferenced a total of 308 individual track locations along the transects. We surveyed transects between 12 and 96 hours after snowfall, usually starting the day following a road survey, with 5 to 7 km of transect being surveyed per day. Due to the limited number of tracks recorded for carnivores (coyote, bobcat, cougar, wolf, fox, lynx, marten, wolverine, see Results) we grouped all the above species into one category \u2018carnivores\u2019, while evaluating the remaining species of \u2018ungulates\u2019 (moose, elk, deer) separately for landscape variable preference models and predictive mapping, and both separately/combined for estimates of permeability across the highway. literature cited : Alexander, S., Waters, N., & Paquet, P. 2005. Traffic volume and highway permeability for a mammalian community in the Canadian Rocky Mountains. Canadian Geographer-Geographe Canadien 49(4):321\u2013331. Barber, J. R., Crooks, K. R., & Fristrup, K. M. 2010. The costs of chronic noise exposure for terrestrial organisms. Trends in Ecology & Evolution 25(3):180\u2013189. Barton, K. 2012. MuMIn: Multi-model inference 1.8.0, http://cran.rproject.org/package=MuMIn. Beaudry, F., deMaynadier, P. G., & Hunter, M. L. 2008. Identifying road mortality threat at multiple spatial scales for semi-aquatic turtles. Biological Conservation 141(10):2550\u20132563. Beier, P., Majka, D. R., & Spencer, W. D. 2008. Forks in the road: Choices in procedures for designing wildland linkages. Conservation Biology 22(4):836\u2013851. Beier, P., & Noss, R. F. 1998. Do habitat corridors provide connectivity? Conservation Biology 12(6):1241\u20131252. 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Ben\u00edtez-L\u00f3pez, A., Alkemade, R., & Verweij, P. A. 2010. The impacts of roads and other infrastructure on mammal and bird populations: a meta-analysis. Biological Conservation 143(6):1307\u20131316. Bennett, A. F. 1991. Roads, roadsides and wildlife conservation: a review. In D. A. Saunders & R. J. Hobbs (Eds.), Nature Conservation 2: the Role of Corridors. (pp. 99\u2013118). New South Wales: Surrey Beatty and Sons. Beyer, H. L. 2012. Geospatial Modelling Environment (Version 0.7.2.1). (software). URL: http://www.spatialecology.com/gme. Boyce, M. S. 1992. Population viability analysis. Annual Review of Ecology and Systematics 23:481\u2013506. Boyce, M. S., Mao, J. S., Merrill, E. H., Fortin, D., Turner, M. G., Fryxell, J., & Turchin, P. 2003. Scale and heterogeneity in habitat selection by elk in Yellowstone National Park. Ecoscience 10(4):421\u2013431. British Columbia Ministry of Transportation and Infrastructure. 2010. Traffic Data Program. Burnham, K. P., & Anderson, D. R. 2002. Model selection and multimodel inference: a practical information-theoretic approach. New York, NY: Springer Verlag. Clevenger, A. P., & Waltho, N. 2000. Factors Influencing the Effectiveness of Wildlife Underpasses in Banff National Park, Alberta, Canada. Conservation Biology 14(1):47\u201356. Clevenger, A. P., & Waltho, N. 2005. Performance indices to identify attributes of highway crossing structures facilitating movement of large mammals. Biological Conservation 121(3):453\u2013464. Crooks, K. R. 2002. Relative sensitivities of mammalian carnivores to habitat fragmentation. Conservation Biology 16(2):488\u2013502. 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Crooks, K. R., & Sanjayan, M. 2006. Connectivity conservation: maintaining connections for nature. In K. R. Crooks & M. Sanjayan (Eds.), Connectivity conservation (pp. 1\u201320). Cambridge: Cambridge University Press. Dickson, B., & Beier, P. 2002. Home-range and habitat selection by adult cougars in southern California. The Journal of wildlife management 66(4):1235\u20131245. Dickson, B. G., Jenness, J. S., & Beier, P. 2005. Influence of vegetation, topography, and roads on cougar movement in southern California. Journal of Wildlife Management 69(1):264\u2013276. Elbroch, M. 2003. Mammal tracks & sign: a guide to North American species. (M. Elbroch, Ed.). Stackpole Books, Pennsylvania, USA. Epps, C. W., Palsboll, P. J., Wehausen, J. D., Roderick, G. K., Ramey, R. R., & McCullough, D. R. 2005. Highways block gene flow and cause a rapid decline in genetic diversity of desert bighorn sheep. Ecology Letters 8(10):1029\u20131038. ESRI. 2009. ArcGIS 9.3. Economic and Social Reserach Institute Inc., Redlands, CA. http://www.esri.com/. ESRI. 2010. ArcGIS 10. Economic and Social Reserach Institute Inc., Redlands, CA. http://www.esri.com/. Fahrig, L., & Rytwinski, T. 2009. Effects of roads on animal abundance: an empirical review and synthesis. Ecology and Society 14:21. Fearer, T. M., Prisley, S. P., Stauffer, D. F., & Keyser, P. D. 2007. A method for integrating the Breeding Bird Survey and Forest Inventory and Analysis databases to evaluate forest bird\u2013 habitat relationships at multiple spatial scales. Forest Ecology and Management 243(1):128\u2013143. Forman, R. T. T., & Alexander, L. E. 1998. Roads and their major ecological effects. Annual Review of Ecology and Systematics 29:207\u2013231. 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Gonzalez, A., Lawton, J. H., Gilbert, F. S., Blackburn, T. M., & Evans-Freke, I. 1998. Metapopulation dynamics, abundance, and distribution in a microecosystem. Science 281(5385):2045\u20132047. Green, P. J. 1995. Reversible jump Markov chain Monte Carlo computation and Bayesian model determination. Biometrika 82(4):711\u2013732. Greene, W. H. 1994. Accounting for excess zeros and sample selection in Poisson and negative binomial regression models (p. 34). New York University Leonard N. Stern School of Business. Guisan, A., & Thuiller, W. 2005. Predicting species distribution: offering more than simple habitat models. Ecology Letters 8(9):993\u20131009. Haddad, N. M., Bowne, D. R., Cunningham, A., Danielson, B. J., Levey, D. J., Sargent, S., & Spira, T. 2003. Corridor use by diverse taxa. Ecology 84(3):609\u2013615. Harrison, S., & Bruna, E. 1999. Habitat fragmentation and large-scale conservation: what do we know for sure? Ecography 22(3):225\u2013232. Hilbe, J. 2008. Negative Binomial Regression (p. 251). Cambridge University Press New York. Holderegger, R., & Di Giulio, M. 2010. The genetic effects of roads: A review of empirical evidence. Basic and Applied Ecology 11(6):522\u2013531. Iglesias, C., Mata, C., & Malo, J. E. 2012. The Influence of Traffic Noise on Vertebrate Road Crossing Through Underpasses. Ambio 41(2):193\u2013201. Jackman, S. 2012. pscl: Classes and Methods for R Developed in the Political Science Computational Laboratory, Stanford University. http://pscl.stanford.edu/. Stanford, California. Jaeger, J. A. G., Bowman, J., Brennan, J., Fahrig, L., Bert, D., Bouchard, J., Charbonneau, N., Frank, K., Gruber, B., & von Toschanowitz, K. T. 2005. Predicting when animal populations are at risk from roads: an interactive model of road avoidance behavior. Ecological Modelling 185(2-4):329\u2013348. 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Johnson, B. K., Kern, J. W., Wisdom, M. J., Findholt, S. L., & Kie, J. G. 2000. Resource selection and spatial separation of mule deer and elk during spring. Journal of Wildlife Management 64(3):685\u2013697. Jones, M. E. 2000. Road upgrade, road mortality and remedial measures: impacts on a population of eastern quolls and Tasmanian devils. Wildlife Research 27(3):289\u2013296. Lambert, D. 1992. Zero-Inflated Poisson Regression, with an Application to Defects in Manufacturing. Technometrics 34(1):1\u201314. Lingle, S., & Wilson, W. F. 2001. Detection and Avoidance of Predators in White-Tailed Deer (Odocoileus virginianus) and Mule Deer (O. hemionus). Ethology 107(2):125\u2013147. Mader, H. J. 1984. Animal habitat isolation by roads and agricultural fields. Biological Conservation 29(1):81\u201396. Malo, J. E., Suarez, F., & Diez, A. 2004. Can we mitigate animal-vehicle accidents using predictive models? Journal of Applied Ecology 41(4):701\u2013710. Mayor, S. J., Schaefer, J. A., Schneider, D. C., & Mahoney, S. P. 2007. Spectrum of selection: new approaches to detecting the scale-dependent response to habitat. Ecology 88(7):1634\u20131640. McClure, C. J. W., Rolek, B. W., & Hill, G. E. 2012. Predicting Occupancy of Wintering Migratory Birds : Is Microhabitat Information Necessary? The Condor 114(3):482\u2013490. McGregor, R. L., Bender, D. J., & Fahrig, L. 2008. Do small mammals avoid roads because of the traffic? Journal of Applied Ecology 45(1):117\u2013123. Mech, L. D. 1995. The challenge and opportunity of recovering wolf populations. Conservation Biology 9(2):270\u2013278. Meidinger, D., & Pojar, J. 1991. Ecosystems of British Columbia. Victoria, BC: British Columbia Ministry of Forests. 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Meiman, S., Civco, D., Holsinger, K., & Elphick, C. S. 2012. Comparing Habitat Models Using Ground-Based and Remote Sensing Data: Saltmarsh Sparrow Presence Versus Nesting. Wetlands 32(4):725\u2013736. Noss, R. F., & Cooperrider, A. Y. 1994. Saving nature\u2019s legacy: protecting and restoring biodiversity. (R. F. Noss & A. Y. Cooperrider, Eds.). Island Press, Washington D.C., USA. Noss, R. F., Quigley, H. B., Hornocker, M. G., Merrill, T., & Paquet, P. C. 1996. Conservation biology and carnivore conservation in the Rocky Mountains. Conservation Biology 10(4):949\u2013 963. R Development Core Team. 2012. R: A language and environment for statistical computing 2.15.2, http://www.r-project.org. Vienna, Austria: R Foundation for Statistical Computing. Rawlings, J. O., Pantula, S. G., & Dickey, D. A. 1998. Applied Regression Analysis: A Research Tool (p. 657). Springer Verlag. Schuster, R., & Arcese, P. 2013. Using bird species community occurrence to prioritize forests for old growth restoration. Ecography 36(4):499\u2013507. Seiler, A. 2005. Predicting locations of moose-vehicle collisions in Sweden. Journal of Applied Ecology 42(2):371\u2013382. Selva, N., Kreft, S., Kati, V., Schluck, M., Jonsson, B.-G., Mihok, B., Okarma, H., & Ibisch, P. L. 2011. Roadless and Low-Traffic Areas as Conservation Targets in Europe. Environmental Management 48(5):865\u2013877. Sheldon, L. 1997. Animal Tracks of the Rockies. (L. Sheldon, Ed.). Lone Pine Publishing, Edmonton, Canada. Singleton, P., & Lehmkuhl, J. 1999. Assessing wildlfie habitat connectivity in the Interstate 90 Snoqualmie Pass Corridor, Washington. In G. L. Evink, P. Garrett, & D. Zeigler (Eds.), 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Proceedings of the third international conference on wildlife ecology and transportation (pp. 75\u2013 84). Florida Department of Transportation, Tallahassee, Florida, USA. Thompson, I. D., Davidson, I. J., Odonnell, S., & Brazeau, F. 1989. Use of track transects to measure the relative occurrence of some boreal mammals in uncut forest and regeneration stands. Canadian Journal of Zoology-Revue Canadienne de Zoologie 67(7):1816\u20131823. Trombulak, S. C., & Frissell, C. A. 2000. Review of ecological effects of roads on terrestrial and aquatic communities. Conservation Biology 14(1):18\u201330. Underhill, J., & Angold, P. 2000. Effects of roads on wildlife in an intensively modified landscape. Environmental Reviews 8(1):21\u201339. Van Dyke, F. G., Brocke, R. H., & Shaw, H. G. 1986. Use of road track counts as indices of mountain lion Felis Concolor presence. Journal of Wildlife Management 50(1):102\u2013109. Venables, W. N., & Ripley, B. D. 2002. Modern Applied Statistics with S (Fourth.). New York: Springer. Vuong, Q. H. 1989. Likelihood Ratio Tests for Model Selection and Non-Nested Hypotheses. Econometrica 57(2):307\u2013333. White, G. C., & Burnham, K. P. 1999. Program MARK: survival estimation from populations of marked animals. Bird Study 46(1 supp 1):120\u2013139. Wulder, M. A., White, J. C., Cranny, M., Hall, R. J., Luther, J. E., Beaudoin, A., Goodenough, D. G., & Dechka, J. A. 2008. Monitoring Canada\u2019s forests. Part 1: Completion of the EOSD land cover project. Canadian Journal of Remote Sensing 34(6):549\u2013562. Yellowstone to Yukon Conservation Initiative. 2013. Y2Y Priority Areas. http://y2y.net/ourwork/priority-areas. Accessed: 2013-March-26. Zeileis, A., & Croissant, Y. 2010. Extended Model Formulas in R: Multiple Parts and Multiple Responses. Journal of Statistical Software 34(1):1\u201313. 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Zeileis, A., Kleiber, C., & Jackman, S. 2008. Regression models for count data in R. Journal of Statistical Software 27(8):1\u201325. 614 615 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Table 1(on next page) Predictor variable description. Variables used in models predicting habitat variables of preferred and avoided crossing sites at 200 m, 500 m, and 1km spatial scales. Perceptual area polygons were only recorded at the 200 m scale and variables were hand-digitized from 1-m pixel photos. PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Variable name Variable decription Source Forest forested (forest + woodland) Perceptual area polygon Variable units: % of 200m radius area Shrub shrub Herb herbaceous (grassland + agriculture) Riparian riparian Freshwater river + lake Unvegetated non-vegetated (gravel, rock +dirt) Highway highway (+sholder) Road road/path Railroad railroad Residential residential + developed Disturbed disturbed habitat (e.g. excavation sites) Wetland PAP wetland Water Lakes, reservoirs, rivers, streams, or salt water. EOSD Variable units: area [m2] in spatial scale buffer around data point Exposed River sediments, exposed soils, pond or lake sediments, reservoir margins, beaches, landings, burned areas, road surfaces, mudflat sediments, cutbanks, moraines, gravel pits, tailings, railway surfaces, buildings and parking, or other non-vegetated surfaces. Low Shrub At least 20% ground cover which is at least one-third shrub; average shrub height less than 2 m. Wetland Land with a water table near/at/above soil surface for enough time to promote wetland or aquatic processes; Treed + Shrub + Herb Herbecous Vascular plant without woody stem (grasses, crops, forbs, gramminoids); minimum of 20% ground cover or one-third of total vegetation must be herb. Dense conifer forest Greater than 60% crown closure; coniferous trees are 75% or more of total basal area. Open conifer forest 26-60% crown closure; coniferous trees are 75% or more of total basal area. Open broadleaf forest 26-60% crown closure; broadleaf trees are 75% or more of total basal area. Gravel road length Road length within buffer (gravel) [m] TRIMPaved road length Road length within buffer (paived) [m] Buildings Number of buildings within buffer PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Table 2(on next page) Permeability values for track counts of the highway and transects along Hwy 3. Values are given for the community, ungulate and carnivore group levels as well as individual species for all tracks and individual crossings observed. A permeability value of 1.0 indicates no difference between off-road areas and the highway in terms of animal movement. PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t All species Ungulate s Carnivore s Deer Elk Moose all tracks 0.284 0.307 0.106 0.223 0.895 0.263 successful crossings 0.265 0.285 0.104 0.210 0.827 0.221 Bobcat Cougar Coyote Fox Wolf all tracks 0.123 0.019 0.121 0.286 0 successful crossings 0.123 0.019 0.118 0.286 0 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Table 3(on next page) Null model comparisons Results (likelihood ratio, AICc, and Vuong tests) for initial distribution testscomparing Poisson (Poiss), negative binomial (NB), zero-inflated Poisson (ZIP), and zero-inflated negative binomial (ZINB) distributions on highway andtransect abundance for all four mammal groups. Bold values represent the lowest AIC for each comparison. Likelihood ratio were performed between Poiss and NB (as well as there respective zero inflated equivilants). Vuong tests were performed between Poiss and ZIP as well as NB and ZINB. (p-values for both likelihood ratio and Vuong tests are shown in parenthesis.) PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t True Zeros logLik D f AICc pred Zero Likelihood ratio Vuong Carnivor es Hwy 404 Pois s - 204.67 7 1 411.354 400 NB - 203.20 4 2 410.407 404 2.947 (0.086) ZIP - 202.76 2 2 409.524 404 0.995 (0.160) ZIN B - 202.76 2 3 411.524 404 3e-04 (0.987) 1.156 (0.124) Tran s 211 Pois s - 300.18 4 1 602.367 193 NB - 282.54 9 2 569.097 213 35.27 (2.87e-09) ZIP -290.35 2 584.700 211 1.510 (0.066) ZIN B - 282.54 9 3 571.097 213 15.603 (7.81e05) -1.737 (0.041) Deer Hwy 181 Pois s - 1114.22 1 2230.43 8 57 NB - 899.05 8 2 1802.11 5 168 430.32 (< 2.2e-16) ZIP - 933.31 2 2 1870.62 4 181 7.654 (9.66e15) ZIN B -889.12 3 1784.24 0 181 88.384 (2.2e-16) 2.455 (0.007) Tran 110 Pois - 1 1890.75 16 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t s s 944.37 5 1 NB - 681.01 4 2 1366.02 8 100 526.72 (2.2e-16) ZIP - 744.00 8 2 1492.01 5 110 7.421 (5.79e14) ZIN B - 672.80 3 3 1351.60 6 110 142.41 (2.2e-16) 2.211 (0.013) Elk Hwy 181 Pois s - 975.88 8 1 1953.77 7 134 NB - 649.75 8 2 1303.51 7 297 652.26 (2.2e-16) ZIP - 650.43 5 2 1304.87 0 305 9.365 (< 2.2e16) ZIN B - 628.97 8 3 1263.95 7 305 42.913 (5.724e11) 3.293 (0.0001) Tran s 241 Pois s - 350.63 5 1 703.269 188 NB - 266.26 2 2 536.524 240 168.74 (2.2e-16) ZIP - 271.46 7 2 546.933 241 4.356 (6.63e06) ZIN B - 265.60 1 3 537.202 241 11.731 (0.001) 0.5704 (0.284) PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Moose Hwy 412 Pois s - 203.82 5 1 409.650 402 NB - 194.47 1 2 392.942 412 18.708 (1.524e-05) ZIP - 194.77 4 2 393.548 412 1.652 (0.049) ZIN B - 194.44 6 3 394.892 412 0.655 (0.418) 0.120 (0.452) Tran s 257 Pois s - 162.04 6 1 326.092 254 NB - 161.54 2 2 327.083 257 1.009 (0.315) ZIP - 161.25 3 2 326.506 257 0.662 (0.254) ZIN B - 161.25 3 3 328.506 257 3e-04 (0.987) 0.866 (0.193) PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Table 4(on next page) Landscape variable preference model results. Top ranked model AICc values from all model approaches used to determine landscape variable preference across six separate spatial approaches (columns) for all four mammal groups. Bold values represent the lowest AICc of the 4 distributions at one scale. Values in grey background represent the lowest AICc overall for a dataset (Hwy, Trans) and species combination. Values with an asterisk represent the approach used for creating predictive abundance maps for a dataset \u2013 species combination. PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t 200m 500m 1km 3 scales Digitized Combine d Carnivor a Hwy Poiss 406.72 393.99 393.44 393.44 398.10 384.89 NB 406.56 394.89 393.98 393.98 398.66 386.39 ZIP 405.82 394.26 384.43 378.92* 398.14 378.96 ZINB 407.86 396.29 386.55 381.03 400.20 374.81 Trans Poiss 581.72 584.91 584.22 563.26 570.09 546.26 NB 558.44 559.21 558.78 547.44* 551.32 541.26 ZIP 566.66 574.32 566.97 557.96 561.51 546.16 ZINB 560.53 561.29 557.56 571.18 553.44 571.18 Deer Hwy Poiss 2112.12 2113.99 2105.97 2058.87 2153.90 2046.83 NB 1768.82 1770.12 1766.99 1760.42 1787.08 1760.38 ZIP 1827.11 1812.58 1820.72 1807.55 1796.08 1768.06 ZINB 1750.03 1739.02 1739.43 1731.62* 1747.77 1707.97 Trans Poiss 1628.13 1523.81 1476.85 1417.29 1652.31 1394.78 NB 1311.95 1269.85 1250.39 1237.15 1303.81 1233.78 ZIP 1387.70 1321.88 1332.63 1320.64 1379.71 1309.27 ZINB 1287.61 1233.46 1231.36* 1238.64 1269.17 1235.47 Elk Hwy Poiss 1833.99 1859.39 1848.05 1754.72 1765.00 1648.51 NB 1286.84 1293.10 1287.29 1279.93 1285.68 1269.56 ZIP 1264.16 1273.91 1260.17 1241.21 1243.92 1213.37 ZINB 1238.56 1239.10 1225.10 1225.10* 1235.30 1219.42 Trans Poiss 627.19 575.37 569.61 543.48 589.83 496.10 NB 513.35 493.12 492.24 472.81 499.24 472.84 ZIP 530.49 496.65 508.39 476.01 510.29 448.41 ZINB 515.38 493.81 494.36 469.91* 500.70 459.12 Moose Hwy Poiss 396.80 373.96 371.63 353.20 353.53 318.84 NB 385.73 365.88 363.09 351.03 349.29 320.59 ZIP 373.22 351.55 342.48 331.93* 344.41 320.95 ZINB 375.33 354.47 344.59 334.06 351.33 327.95 Trans Poiss 303.36 291.82* 295.15 297.75 311.13 297.31 NB 313.82 293.94 297.25 299.82 313.22 299.38 ZIP 313.70 293.94 297.25 301.76 307.72 299.22 ZINB 315.80 295.53 299.35 309.62 309.91 305.06 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Table 5(on next page) Summed importance scores of predictor variables. The table shows how often a variable was included (as positive or negative predictor) in the eight remotely sensed modeling frameworks used to create predictive maps for Carnivora, Deer, Elk and Moose (marked with asterisks in Table 4). PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Highway Transect positiv e negativ e positiv e negativ e Water 2 0 2 1 Exposed 0 2 1 2 Low Shrub 0 0 0 4 Wetland 0 0 1 3 Herbecous 1 1 3 4 dense conifer forest 0 0 2 2 open conifer forest 0 0 1 2 open broadleaf forest 0 2 4 1 gravel road length 1 0 5 0 paved road length 3 1 1 2 number of buildings 0 3 1 1 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Figure 1 Study Area (Cranbrook 49 \u00b0 30\u2019 N, 115\u00b0 46\u2019 W). East Kooteney region, South eastern British Columbia, Canada. Also shown are the data collection points as well as the remote sensed (EOSD) class distribution that formed part of the model inputs. PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Figure 2 Cumulative track plots of successful crossing attempts by the four focal species groups. Areas of no increase indicate locations along the highway where the focal group rarely or never cross the highway. This shows that for some of the focal groups there is substantial stretches of highway that represent crossing barriers. PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Figure 3 Community crossing site preference (green) and avoidance (red) for highway approach and actual crossing predictions Crossing predictions are visible in inserts A and B as the polygons in the center within the highway outline. Results are based on averaged model results from the best remote sensed model framework for the carnivore group, deer, elk and moose. Individual model framework abundance predictions were split into 10 quantiles, multiplicatively combined and standardized by dividing by 1000 to create community scores between 0 and 10. None of our predictions approach the maximum of 10 as no location suits all species perfectly. Insert A shows an area with high overlap between approach and crossing scores. Insert B illustrates and area of high crossing scores but low approach scores. PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Authors: 1 Richard Schuster1,2, Neurobiology and Behavior Group, Department of Zoology, Karl- 2 Franzens-University, A-8010 Graz, Austria 3 Heinrich R\u00f6mer, Neurobiology and Behavior Group, Department of Zoology, Karl- 4 Franzens-University, A-8010 Graz, Austria 5 Ryan R. Germain, Centre for Applied Conservation Research, Department of Forest & 6 Conservation Sciences, 2424 Main Mall, University of British Columbia, Vancouver, 7 BC, V6T 1Z4, Canada 8 9 10 1Corresponding author: 11 Richard Schuster, Centre for Applied Conservation Research, Department of Forest & 12 Conservation Sciences, 2424 Main Mall, University of British Columbia, Vancouver, 13 BC, V6T 1Z4, Canada. Phone: +1 604 822 1256. Email: mail@richard-schuster.com 14 15 2Current Address: Centre for Applied Conservation Research, Department of Forest & 16 Conservation Sciences, 2424 Main Mall, University of British Columbia, Vancouver, 17 BC, V6T 1Z4, Canada.18 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Introduction 19 As human-induced fragmentation of wildlife habitats continues to increase, there is a growing 20 need to both investigate the effects of such fragmentation on animal communities and to 21 present possible solutions to help mitigate these effects (e.g. Gonzalez et al., 1998; Crooks, 22 2002). Roads are a major contributor to the fragmentation of wildlife habitat around the 23 world (e.g, North America: Trombulak & Frissell, 2000; Underhill & Angold, 2000; Europe: 24 Holderegger & Di Giulio, 2010; Selva et al., 2011; Australia: Jones, 2000), and their 25 construction and maintenance are one of the most widespread forms of human-based habitat 26 modification (Bennett, 1991; Noss & Cooperrider, 1994). Major effects of roads on wildlife 27 can include traffic mortality, modification of animal behavior (e.g., road avoidance), and 28 alteration of the physical and chemical environment leading to barrier effects and habitat 29 fragmentation (reviewed in Trombulak & Frissell, 2000; Jaeger et al., 2005). Movement 30 barriers such as roads can affect wildlife at several different levels; in addition to lowering 31 individual fitness through restricted access to resources and increased mortality risk 32 (reviewed in Fahrig & Rytwinski, 2009), roads may also reduce gene flow between 33 fragmented habitats and contribute to the creation of smaller subpopulations which are more 34 vulnerable to stochastic events (Boyce, 1992; Forman & Alexander, 1998; Jaeger et al., 35 2005). For example, road fragmentation is implicated as a major contributor towards the 36 extirpation of carnivorous mammals in the Rocky Mountains of western North America 37 (Noss et al., 1996). Thus, there is a clear need for research on predicting areas of preferred 38 animal crossing sites to both identify appropriate locations for mitigation measures and help 39 reduce the negative effects of roads on wildlife communities. 40 Most studies investigating how to apply practical mitigation measures (e.g., crossing 41 structures such as overpasses) aimed at reducing the effects of roads on animal communities 42 focus on predicting the landscape features of animal-vehicle collision sites (e.g. Malo, 43 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Suarez, & Diez, 2004; Seiler, 2005). Although current funding for mitigation measures is 44 often allocated to sites along roads where collisions have previously been reported, collision 45 sites may not necessarily represent the areas preferentially used by wildlife to cross roads 46 (Alexander, Waters, & Paquet, 2005). Consequently, identifying the landscape features 47 surrounding roads which represent both preferred and avoided animal crossing sites may help 48 inform mitigation design and optimize animal movement between sub-populations, thereby 49 reducing the effects of habitat fragmentation (Singleton & Lehmkuhl, 1999; Alexander, 50 Waters, & Paquet, 2005). 51 Previous studies on the efficiency of mitigation strategies indicate that different 52 mammal species can be highly variable in their tolerance to human structures, suggesting that 53 the effects of barriers such as roads and the success of mitigation strategies will also likely 54 vary by species (Beier & Noss, 1998; Trombulak & Frissell, 2000). Studies investigating 55 mitigation strategies for high-traffic areas should therefore incorporate multiple focal species 56 and predict spatial linkages across roads at the community level (Beier, Majka, & Spencer, 57 2008). In particular, modeling animal movement across multiple spatial scales may aid our 58 understanding of preferred habitat use along roads when considering multiple species of large 59 mammals, which may each differ in terms of habitat requirements, home range sizes, and 60 sensitivity to road disturbance. Animals may also select movement habitat at multiple scales, 61 as shown in migratory birds, reptiles, and large mammals (e.g. Boyce et al., 2003; Beaudry, 62 deMaynadier, & Hunter, 2008; McClure, Rolek, & Hill, 2012). Therefore, studies which 63 incorporate several spatial scales into the same analytical framework, and compare results of 64 predicted crossing sites across multiple spatial scales may prove particularly useful in 65 planning mitigation strategies. Because micro-habitat assessments are often costly and labor-66 intensive (e.g. Fearer et al., 2007), direct comparisons of the validity of predictive models 67 generated from micro-habitat data versus macro-habitat assessments from remotely sensed 68 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t data may aid future research in allocating more time and funding to the most efficient 69 methods. 70 Here, we characterize preferred and avoided crossing sites of eight large-bodied 71 mammal species along a 95km length of highway through the Purcell Mountain Range of 72 North America. We use a multi-scale approach comparing high-resolution, manually-73 digitized habitat metrics with remote sensing-derived metrics at three spatial scales (200m, 74 500m, and 1km) to investigate the potential drawbacks of each method in implementing 75 mitigation measures. Our goals for this study include identifying the habitat variables 76 (\u2018predictors\u2019) of preferred and avoided crossing sites for each mammal species along this 77 highway, and evaluating the efficiency of using macro-habitat predictors derived from freely 78 available remote sensing data versus manually-digitized micro-habitat maps to predict such 79 crossing sites. To address these goals we ask the following specific questions: 1) does the 80 highway present a movement barrier to a multi-species community of mammals, 2) do 81 species show preference in their choice of crossing sites towards predefined landscape 82 predictors, 3) are there preferred crossing areas for species or species groups along the 83 highway that could potentially serve as mitigation sites, 4) are preferred versus avoided 84 crossing sites better predicted by habitat variables generated at the macro-scale, micro-scale, 85 or a combination of both? 86 87 Methods 88 STUDY AREA 89 Our study was conducted along Southern Trans-Provincial Highway 3 (hereafter Hwy 90 3) between the towns of Creston and Cranbrook, in south-eastern British Columbia, Canada 91 (Fig. 1). The study area is located in the Purcell Mountain Range, which ranges from 620m to 92 2,087m in elevation, and is comprised of Interior Cedar Hemlock and Interior Douglas Fir 93 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Biogeoclimatic zones (Meidinger & Pojar, 1991). We chose this study area for its ecological 94 importance as a trans-boundary priority area (Yellowstone to Yukon Conservation Initiative, 95 2013) that connects small populations of carnivores such as grizzly bears (Ursus arctos 96 horribilis) and Canada lynx (Lynx canadensis) along the Canada \u2013 USA border. Hwy 3 97 bisects this important corridor, possibly leading to negative effects on the connectivity of this 98 movement corridor for mammal populations. The average annual traffic volume (AADT) for 99 this highway section was 3050 cars/day in 2007, with a seasonal (December to March) 100 average of 2020 vehicles/day (British Columbia Ministry of Transportation and Infrastructure 101 2010). 102 103 DATA COLLECTION 104 We monitored species movement through the study area by recording tracks in the 105 snow where animals attempted to cross Hwy 3, as well as along ten transects approaching the 106 highway, set back from any highway right-of-way (distance from transects to highway ranged 107 from 10 to 900m, mean 175m). We pre-defined our transects as survey lines marked with 108 flagging tape, roughly parallel to the highway. Highway and transect tracks were recorded 109 over two winter seasons, January to March 2007 and December 2007 to February 2008 (all 110 observations recorded by RS). 111 Highway and transect track surveys were conducted using methods similar to Van 112 Dyke, Brocke, & Shaw (1986), and Alexander, Waters, & Paquet (2005). Briefly, we 113 conducted highway crossing attempt surveys along a 95km length of Hwy 3, at least 12 hours 114 after the last snowfall. Each survey was conducted from a moving vehicle with a speed of 115 approximately 10-15 km/h. When a track was observed, the investigator stopped the vehicle 116 and conducted an on-foot inspection to identify the track. In total, we investigated tracks for 117 12 mammal species: coyote (Canis latrans), fox (Vulpes vulpes), wolf (Canis lupus), cougar 118 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t (Puma concolor), bobcat (Lynx rufus), lynx, marten (Martes americana), wolverine (Gulo 119 gulo), elk (Cervus canadensis), moose (Alces alces), white-tailed and mule deer (Odocoileus 120 virginianus and Odocoileus hemionus, respectively). When we were uncertain of the identity 121 of a track, we recorded track pattern measurements, took photos and later consulted field 122 guides (Sheldon, 1997; Elbroch, 2003) for identification. Data at a total of 463 crossing sites 123 were georeferenced with a handheld, Garmin eTrex Summit GPS receiver (WGS 1984, \u00b1 10 124 \u2013 40 m). If multiple tracks were found for one species at a single crossing area, we recorded 125 the total track count. We also recorded the success of a crossing attempt, here defined as the 126 presence of a continuing set of tracks on the opposite side of the road. When tracks of the 127 same species were found within 300 meters of a crossing site, it was not recorded as an 128 individual crossing attempt, but rather as a possible repeat crossing of the same individual 129 (Alexander, Waters, & Paquet, 2005). Surveys were suspended when continuous heavy 130 snowfall covered tracks during data collection. 131 Transects were established off-road in suitable areas close to the highway. Suitability 132 was contingent upon minimal disturbance from residential areas, and no barriers to observer 133 access (i.e., lakes, steep terrain, fences or private property). Seven transects had a linear 134 distance of 1 km, while one was 2 km (Transect 6) and one was 5.4 km in length (Transect 135 10). Only the first kilometer of transect 10 was surveyed during the second season of data 136 collection, and this was classified as Transect 9 for ease of data handling. We recorded tracks 137 of the same species according to the protocol of the crossing attempt surveys, and 138 georeferenced a total of 308 individual track locations along the transects. We surveyed 139 transects between 12 and 96 hours after snowfall, usually starting the day following a road 140 survey, with 5 to 7 km of transect being surveyed per day. Due to the limited number of 141 tracks recorded for carnivores (coyote, bobcat, cougar, wolf, fox, lynx, marten, wolverine, see 142 Results) we grouped all the above species into one category \u2018carnivores\u2019, while evaluating the 143 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t remaining species of \u2018ungulates\u2019 (moose, elk, deer) separately for landscape variable 144 preference models and predictive mapping, and both separately/combined for estimates of 145 permeability across the highway. 146 147 ROAD BARRIER EFFECT 148 We standardized the highway crossing attempt and transect survey data by the 149 number of 12 hour periods that had elapsed since the time of the last snowfall to correct for 150 time effects (Thompson et al., 1989). For calculation of the road barrier effect, we 151 standardized survey data for the highway and transects by kilometers surveyed: 152 \ud835\udc36\ud835\udc5f\ud835\udc5c\ud835\udc60\ud835\udc60\ud835\udc56\ud835\udc5b\ud835\udc54\ud835\udc60 \ud835\udc5d\ud835\udc52\ud835\udc5f \ud835\udc58\ud835\udc5a = \ud835\udc47\ud835\udc5c\ud835\udc5f\ud835\udc4e\ud835\udc59 \ud835\udc5b\ud835\udc62\ud835\udc5a\ud835\udc4f\ud835\udc52\ud835\udc5f \ud835\udc5c\ud835\udc53 \ud835\udc61\ud835\udc5f\ud835\udc4e\ud835\udc50\ud835\udc58\ud835\udc60 \ud835\udc47\ud835\udc5c\ud835\udc61\ud835\udc4e\ud835\udc59 \ud835\udc59\ud835\udc52\ud835\udc5b\ud835\udc54\ud835\udc61\u210e \ud835\udc5c\ud835\udc53 \ud835\udc60\ud835\udc62\ud835\udc5f\ud835\udc63\ud835\udc52\ud835\udc66\ud835\udc60 We then calculated the permeability of the highway by standardizing the crossings per km of 153 highway with the crossings per km of transect: 154 \ud835\udc43\ud835\udc52\ud835\udc5f\ud835\udc5a\ud835\udc52\ud835\udc4e\ud835\udc4f\ud835\udc56\ud835\udc59\ud835\udc56\ud835\udc61\ud835\udc66 = \ud835\udc3b\ud835\udc56\ud835\udc54\u210e\ud835\udc64\ud835\udc4e\ud835\udc66 \ud835\udc50\ud835\udc5f\ud835\udc5c\ud835\udc60\ud835\udc60\ud835\udc56\ud835\udc5b\ud835\udc54\ud835\udc60 \ud835\udc5d\ud835\udc52\ud835\udc5f \ud835\udc58\ud835\udc5a \ud835\udc47\ud835\udc5f\ud835\udc4e\ud835\udc5b\ud835\udc60\ud835\udc52\ud835\udc50\ud835\udc61 \ud835\udc50\ud835\udc5f\ud835\udc5c\ud835\udc60\ud835\udc60\ud835\udc56\ud835\udc5b\ud835\udc54\ud835\udc60 \ud835\udc5d\ud835\udc52\ud835\udc5f \ud835\udc58\ud835\udc5a We also constructed track accumulation curves along the 95km of highway for all four 155 species groups to identify areas of the highway with greater crossing intensity for each 156 mammal group. 157 MULTI-SCALE LANDSCAPE VARIABLES 158 To develop our micro-habitat assessments, we imported the collected GPS data into 159 ArcGIS 9.3 (ESRI, 2009). The GPS points from the highway surveys and the transect surveys 160 were set on top of a georeferenced (WGS 1984, UTM Zone 11N) orthophotograph layer from 161 2004, with a spatial resolution of 1m, provided by a Web Map Service (WMS) of GeoBC 162 (http://www.geobc.gov.bc.ca). For each GPS point, we created a circular buffer of 200m to 163 represent the perceptual area of the animal directly influenced by the surrounding landscape 164 predictors (e.g. Lingle & Wilson, 2001), which we define as \u2018perceptual area polygon\u2019. For 165 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t each buffer area, we digitalized polygons for predefined landscape predictors and 166 georeferenced them using the orthophotograph layer and Google Earth, as the latter provided 167 more recent images of the research area. We used the following landscape predictors, adapted 168 from Dickson, Jenness, & Beier (2005): forested (forest + woodland), shrub, herbaceous 169 (grassland + agriculture), riparian, water, non-vegetated (gravel, rock +dirt), highway 170 (+shoulder), road/path, railroad, residential, developed, disturbed and wetland (Table 1). We 171 then calculated the percentage of each buffer area overlapped by each landscape predictor. 172 Because large mammals might respond to both fine and coarse scale habitat features 173 (e.g. Mayor et al., 2007), we developed a series of variables describing macro-habitat 174 landscape features at three spatial scales: 200m, 500m, and 1 km. For modeling species 175 abundance along the highway and transects, we chose candidate predictor variables based on 176 their ability to predict species abundance at site and landscape levels in similar studies (e.g. 177 Malo, Suarez, & Diez, 2004; Guisan & Thuiller, 2005). All remotely sensed predictors (Table 178 1) were derived from the following sources: Terrain Resource Information Management 179 (TRIM, Province of BC 1992) and Earth Observation for Sustainable Development 180 Landcover (EOSD LC 2000, Wulder et al., 2008). Our dataset comprised 12 predictor 181 variables from the perceptual area polygons and 11 from remote sensing on 3 scales (200m, 182 500m, 1km; Table 1), derived at each of 463 highway locations and 308 transect locations. 183 All remote sensing predictors were created using Geospatial Modelling Environment (Beyer, 184 2012) in conjunction with ArcGIS 10 (ESRI, 2010) and R v. 2.15.2 (R Development Core 185 Team, 2012). Due to their widely varying scales, all predictors were standardized to mean=0, 186 sd =1 to ensure that their importance was not driven by measurement scale (White & 187 Burnham, 1999). 188 189 LANDSCAPE VARIABLE PREFERENCE MODELS 190 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Since predetermining the appropriate data distribution for our count data from ecological 191 knowledge alone was not possible, we modeled abundance incorporating both a Poisson 192 distribution (P) and negative binomial (NB) distribution to account for potential 193 overdispersion (e.g. Zeileis, Kleiber, & Jackman, 2008). Because of the large proportion of 194 zero values included in our data-set, we also applied zero-inflated models (ZIP, ZINB; 195 Lambert, 1992), which are mixture models that combine both count data and a binomial 196 model. To determine which of these distributions best represented our species data, we 197 visually inspected the data and compared the log Likelihood, AIC, and number of correctly 198 predicted zeros for each distribution model fits using intercept-only models. To test for 199 differences among distribution functions, we used likelihood ratio tests to compare the 200 Poisson and negative binomial distributions, since the Poisson distribution is a restriction of 201 the more general negative binomial distribution (Hilbe, 2008). We tested H0 for no difference 202 between the two and H1 that the negative binomial was a better fit to the data. We tested the 203 same hypothesis using the zero-inflated Poisson and zero-inflated negative binomial. Next, 204 we used a Vuong test (Vuong, 1989; Greene, 1994) to evaluate whether the zero-inflated 205 models were a statistically better fit to the data than their base model (Hilbe, 2008). The 206 Vuong test is generally formulated as: 207 V = \ud835\udc60\ud835\udc5e\ud835\udc5f\ud835\udc61(\ud835\udc41) \u2217 \ud835\udc5a\ud835\udc52\ud835\udc4e\ud835\udc5b (\ud835\udc5a) \ud835\udc60\ud835\udc5a \u2217 \ud835\udc5a \ud835\udc5a = \ud835\udc59\ud835\udc5b ( \ud835\udf071 \ud835\udf072 ) Where \u03bc1 = predicted probability of y for the zero-inflated model, \u03bc2 = predicted probability 208 of y for the base model, sm = standard deviation of m, and N = number of observations in 209 each model, where both must use the same observations. The test statistic V is asymptotically 210 normal. If V > 1.96, the zero-inflated model is preferred; if V< -1.96, the base model is 211 preferred; and if the value of V is between -1.96 and 1.96 neither model is preferred (Hilbe, 212 2008). To perform these tests we used the function vuong in R package pscl v.1.04.4 (Zeileis, 213 Kleiber, & Jackman, 2008; Jackman, 2012). 214 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t We compared our measures of model selection (AIC, LogLik, predicted zeros, Vuong) 215 for all four distributions (P, NB,ZIP, ZINB) throughout each of the model building stages of 216 this study to avoid bias in predetermining the distribution with intercept-only models. For 217 each of the four mammal groups (deer, elk, moose, carnivores), we compared models of 218 predicted habitat preference using each distribution (P, NB, ZIP, ZINB) and data set 219 (Highway, Transect) across six separate spatial approaches: i) 200m scale, ii) 500m scale, iii) 220 1km scale, iv) all 3 scales combined; v) perceptual area polygons; vi) all scales and 221 perceptual area polygons combined. For spatial approaches iv) and vi), we created an 222 iterative model fitting procedure that starts with an intercept only base model, individually 223 adds predictor variables, records the results of each fitted model, and retains all top-ranked 224 models (\u0394AICc \u2264 2) at each iteration as base models for subsequent iterations as long as there 225 is reduction in AICc (R function in R-Code S1). The fitting procedure constitutes an 226 extension of a more restrictive routine that only included the top ranked model for each 227 iteration in subsequent iterations (Schuster & Arcese, 2013). We opted for the iterative 228 approach because creating all possible models for approach iv) and vi) would have resulted in 229 2^45 models each. For the remaining approaches, we created models for each possible 230 combination of predictors. For both ZIP and ZINB we further expanded our model lists using 231 a) intercept only models for the zero-inflation component, while using predictors in the count 232 component and b) including the same predictors for both zero-inflation and count 233 components. In post-processing we reduced the candidate set of models from each approach 234 based on the statistical significance of all predictors, using p-values as a general and liberal 235 criterion for retaining models. We selected a cutoff value of p = 0.15 as it serves as the 236 default for many stepwise model selection approaches (e.g. Rawlings, Pantula, & Dickey, 237 1998). We chose this approach to reduce the probability of including non-informative models 238 (i.e., those stuck at local maxima for parameter estimation) in subsequent model averaging. 239 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Additionally we checked for and removed models with unrealistically high (>50) parameter 240 estimates and/or Standard Errors, which would indicate lack of model fit. For model 241 selection, we ranked all remaining candidate models by AICc and averaged those with \u2206AICc 242 \u2264 2 from the top ranked model (Burnham & Anderson, 2002). All analysis were conducted 243 using R v.2.15.2 (R Development Core Team, 2012); package MuMIn 1.8.0 (Barton, 2012) 244 was used for AICc calculations; package MASS v. 7.3-22 (Venables & Ripley, 2002) for NB 245 models; and packages pscl and Formula v.1.1-0 (Zeileis & Croissant, 2010) for ZIP and 246 ZINB models. 247 To determine the distribution and scale that provided the best relative fit to the data for 248 each mammal group we compared and ranked the models with the lowest AICc of each 249 approach and determined the approach resulting in the overall lowest AICc value. We further 250 contrasted these results with the initial results from the intercept only models to determine 251 whether initial models were sufficient to identify the error distributions that were most 252 appropriate for a given data set or whether predictor variables had to be included first. 253 254 PREDICTIVE MAPS OF PREFERRED CROSSING SITES 255 Using the previous model results we created predictive maps for each species/group 256 for both the abundance of animals approaching the highway (transect models) and the 257 abundance of animals, having reached the highway, crossing it (highway models). For each 258 map we chose the approach with the lowest AICc values out of the 4 remote sensing-derived 259 frameworks (200m, 500m, 1km, 3 scales), as landscape level data was not available for the 260 perceptual area polygons. For predictive polygons, we used 30x30m polygons to follow the 261 EOSD resolution. For the highway predictions we created polygons around the highway (line 262 feature buffered 15m on each side of highway) resulting in predictions for 7374 polygons. 263 For transect predictions we expanded the buffer around the highway to 1km, resulting in 264 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t 213442 polygons. Next, we generated a predictor set for each polygon centroid that was 265 identical to those used for survey points, and then estimated abundance based on our 266 averaged models for each of the focal species groups. We combined individual mammal 267 group abundance estimates into 10 quantiles to consolidate focal species maps into an index 268 of site preference, multiplied those scores for each polygon and standardized them by 269 dividing by 1000, resulting in community site preference scores between 0 (being the lowest 270 preference) and 10 (highest preference). 271 272 Results 273 We conducted surveys for 737 km of highway (H) and 118.5 km of transects (T), that 274 yielded the following number of track counts: deer = 970 H/887 T, elk = 575 H/152 T, moose 275 = 65 H/59 T, coyote = 58 H/111 T, bobcat = 6 H/11 T, cougar = 1 H/11 T, wolf = 0 H/10 T, 276 fox = 3 H/2 T. No tracks were found for lynx, marten or wolverine (raw data and R model 277 input files can be found in Data S1). 278 279 ROAD BARRIER EFFECT 280 Highway permeability values for the majority of groups were extremely low (where a 281 value of 1 indicates full permeability across the highway and 0 represents no permeability), 282 indicating that Hwy 3 likely acts as a barrier to mammal movement (Table 2). The 283 permeability values for carnivores were only one third those of ungulates (moose, elk, deer, 284 combined) on the investigated section of Hwy 3 (Table 2), indicating that deer, elk, and 285 moose were much less affected by the highway in terms of movement than carnivores. Track 286 accumulation curves for all mammal groups indicate that in all four cases there were areas of 287 the highway where the focal group rarely or almost never crosses the highway (Figure 2). 288 289 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t LANDSCAPE VARIABLE PREFERENCE MODELS 290 Based on intercept only model comparisons using likelihood ratio tests, Vuong test 291 and AIC ranking, the best supported distributions for each mammal group were: ZINB (Deer-292 Hwy, Elk-Hwy, Deer-Trans), NB (Moose-Hwy, Elk-Trans, Carnivora-Trans), ZIP 293 (Carnivora-Hwy) , P (Moose-Trans), indicating that ZINB and NB were the most commonly 294 supported distributions (Table 3). In 6 out of 8 cases, the modeling approach which included 295 predictors from all three scales and the digitized polygons was selected as the top model 296 based on AICc (Table 4). In only two cases for the transect data did other approaches result in 297 lower AICc values: Deer (1km scale) and Moose (500m scale). In direct comparison between 298 predictors derived from remotely sensed data and hand digitized data, the remotely sensed 299 model framework resulted in lower AICc values in all eight cases. When comparing the 300 remotely sensed data approach at the same scale as the digitized (200m) data, digitized 301 predictors resulted in lower AICc values in 7/8 cases. A comparison of the extended 302 modeling results (Table 4) with the initial distribution tests (Table 3) indicates that in 4/8 303 cases the results from the initial tests were rejected and different distributions formed the 304 basis of models with the lowest AICc values. For Highway and Transect data from each 305 mammal group, model coefficients describing preferred (positive values) and avoided 306 (negative values) habitat variables from the best supported model (above) are depicted in 307 Tables S1 and S2. Summed values of preferred and avoided landscape variables for the entire 308 mammal community are presented in Table 5. 309 310 PREDICTIVE MAPS OF PREFERRED CROSSING SITES 311 To map preferred crossing sites, we used averaged model results for each mammal 312 group based on the framework with the lowest AICc value out of the 4 remotely sensed 313 model sets (Cells marked with an asterisk in Table 4; maps and shapefiles in Figures S1-S4 314 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t and Data S2 respectively). The combined predictions of preferred (green) and avoided (red) 315 crossing sites for all investigated species within the mammal community are illustrated in 316 Fig. 3. Based on predictions generated from landscape variables (above), certain regions of 317 the study area exhibited high preference scores from both approach (transect) and crossing 318 (highway) models (e.g., Fig. 3 insert A), indicating that these locations likely represent areas 319 of high priority when implementing mitigation measures for all species considered in our 320 study. Conversely, certain regions of the study area exhibited high preference scores for one 321 of the model sets (crossing vs. approach), but not the other (e.g., Fig. 2 insert B), indicating 322 that these may represent less-ideal locations to implement mitigation measures such as 323 crossing structures. Areas of unambiguous preference for particular crossing sites (i.e., those 324 where crossing and approach preference scores overlap) differ for each mammal group 325 considered in our study (Figures S1-S4), indicating that mitigation strategies aimed at 326 mammal communities may differ substantially from those aimed at a target species. 327 328 Discussion 329 We determined that Hwy 3 posed a severe movement barrier to the local mammal 330 community. Although each investigated species differed in the landscape variables associated 331 with preferred and avoided crossing sites, we used a multi-scale approach to identify 332 locations along the highway where mitigation measures may benefit all species in the large 333 mammal community. Below we address our earlier questions and discuss the implications of 334 our finding that multi-scale habitat assessments may be necessary to accurately predict the 335 most effective locations for highway crossing structures (e.g., culverts and overpasses) or 336 other mitigation measures. 337 Permeability estimates for both carnivores and the majority of ungulate species 338 considered were extremely low across the highway (Table 2), indicating that Hwy 3 likely 339 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t acts as barrier to animal movement. Although permeability estimates for elk were 340 comparatively high (likely due to herding behavior, whereas tracks for all other species 341 tended to be solitary or in small groups), averaged estimates for all ungulates and the entire 342 mammal community suggest that movement by large-bodied mammals is highly restricted 343 across the highway. Likewise, track accumulation curves (Figure 2) indicate that for each 344 species group considered, certain areas of the highway may rarely or never be crossed, posing 345 large limitations to population connectivity across Hwy 3. This finding is consistent with 346 previous estimates of wildlife permeability across a similar highway through the Rocky 347 Mountain Range of Alberta, Canada (Alexander, Waters, & Paquet, 2005). Such low 348 permeability across the highway suggests a severe threat of habitat fragmentation to the 349 mammal community, which could result in decreased gene flow across the road barrier, and 350 ultimately to lower population viability in the region (Mader, 1984; Epps et al., 2005). These 351 results indicate a need to accurately identify locations for potential mitigation measures along 352 roads such as Hwy 3 to facilitate the movement of individuals across the highway and reduce 353 this barrier effect (Harrison & Bruna, 1999; Haddad et al., 2003; Crooks & Sanjayan, 2006). 354 By incorporating both highway and transect predictions simultaneously, we aimed to 355 identify locations for potential mitigation measures that represent both preferred crossing 356 sites as well as preferred approach habitat up to 1km from the highway. We determined that 357 the landscape variables associated with preferred/avoided crossing sites differed for many of 358 the mammal groups considered (Tables S1, S2). In all cases, noise generated from vehicles 359 travelling on the highway could contribute to road avoidance by large mammals (Forman & 360 Alexander, 1998; Jaeger et al., 2005; Barber, Crooks, & Fristrup, 2010). However, numerous 361 studies on movement across roads by large and small mammals have found no consistent 362 response to noise levels, and suggest that habitat characteristics surrounding crossing sites 363 play a larger role in animal movement than individual tolerance to noise levels (McGregor, 364 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Bender, & Fahrig, 2008; Iglesias, Mata, & Malo, 2012). For instance, carnivores tended to 365 avoid residential areas along the highway as well as open areas with low shrub cover (Tables 366 S1, S2), consistent with previous studies (e.g. Mech, 1995). While elk and deer did not avoid 367 these landscape features, these two species exhibited dissimilar patterns of habitat and 368 crossing-site preference, consistent with their different habitat requirements (Johnson et al., 369 2000). These differing results per group indicate that a clear set of conservation goals for 370 each species as well as the community as a whole must be established before mitigation 371 measures are implemented to facilitate highway crossing (e.g. Beier, Majka, & Spencer, 372 2008). 373 We used multi-model inference and model averaging to identify locations of preferred 374 crossing sites for all mammal species considered, which would likely serve as the most 375 effective locations for mitigation measures aimed at increasing mammal permeability across 376 the highway. Cumulative scores of preferred/avoided landscape variables along both the 377 highway and transect data sets indicate that preferred crossing sites tended to be within close 378 proximity of water and longer stretches of unpaved road (Table 5). Crossing-specific scores 379 indicate a preference for longer stretches of paved roads, and approach-specific scores 380 suggest preference for areas of high crown cover with abundant broadleaf trees, respectively. 381 Although this approach may reduce the efficiency of predicting highway crossing sites for 382 certain focal species, community-level approaches are increasingly advocated as a more 383 efficient means of implementing wildlife linkages across barriers such as major roads (Beier, 384 Majka, & Spencer, 2008). To accomplish this goal, we applied an exhaustive model approach 385 incorporating four separate distributions of abundance for each mammal group along Hwy 3. 386 In only 4 of the 8 cases considered was pre-selection of the y-distribution successful, 387 indicating that an exhaustive modeling approach incorporating multiple distributions may be 388 necessary when the goal is to identify and predict preferred crossing sites based on limited 389 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t data and uncertainties regarding which abundance distributions are most applicable to free-390 living animal populations. By adopting the approach described here, researchers may be able 391 to extract more information from highway crossing data than could otherwise be gained from 392 applying predefined and potentially inaccurate abundance distributions. Further, the best-393 supported distribution differed for each species; while ZINB and NB were the most 394 commonly supported distributions, NB, ZIP and P each received the best support for at least 395 one data set (highway versus transect). These results once again highlight the need for future 396 studies to consider the unique habitat requirements of each species within mammal 397 communities when developing mitigation strategies, but that those strategies which provide 398 the greatest benefit to the largest number of species should be given priority for 399 implementation. 400 To establish conservation-based goals for large mammals along roads such as Hwy 3, 401 further consideration must be given to whether the spatial scales at which habitat 402 characteristics are measured match the spatial scales at which the animals select 403 preferred/avoided crossing sites. We determined that in 6/8 cases, a combined approach to 404 modeling preferred crossing sites (incorporating remotely sensed and hand-digitized 405 predictors) resulted in the best supported model. Further, utilizing multi-scale remote 406 sensing-derived predictors always resulted in better model support than utilizing only hand-407 digitized predictors for each species and data set considered. Thus, our results indicate that 408 while a combined approach may represent the most informative method for predicting 409 landscape variables of preferred mammal crossing sites, freely-available macro-habitat data 410 such as those generated through remote sensing may be more useful than labor-intensive 411 micro-habitat assessments when time and budgetary constraints on data collection are 412 imposed. Previous studies investigating habitat occupancy in birds have found similar results 413 (e.g. McClure, Rolek, & Hill, 2012; Meiman et al., 2012), highlighting the increasing 414 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t usefulness of remote sensing in evaluating localized questions in conservation and 415 community ecology. 416 The goal of our study was to identify locations along Hwy 3 where mitigation 417 measures might increase connectivity across the highway for all species in the mammal 418 community. Although we do not currently have data on which mitigation measures may be 419 the most effective on increasing permeability in this system, previous studies investigating 420 the costs/benefits of different mitigation strategies at the community level (e.g. Clevenger & 421 Waltho, 2000, 2005) indicate that a diversity of crossing structures of different sizes may best 422 serve large mammal communities. Because our permeability estimates were based on snow 423 tracks and not on data for the entire year, there is the potential for our results to only be 424 applicable for winter months. Further, because our permeability estimates are based on 425 transects with a mean distance of 175 m from the highway, we likely overestimate 426 permeability in certain cases by not considering the density of animals in areas further away 427 from the highway. For instance, Dickson & Beier (2002) determined that cougars typically 428 avoid high speed roads at a distance of 500m \u2013 1km and more generally, mammal 429 populations might be influenced by human infrastructure up to about 5km (Ben\u00edtez-L\u00f3pez, 430 Alkemade, & Verweij, 2010). Although conducting further transects at a greater distance 431 from the road may improve estimates of habitat preference for each species along Hwy 3, we 432 believe our methods represent a realistic investigation of the types of habitat used by animals 433 approaching and ultimately crossing the road, which may help inform strategies for 434 implementing crossing structures. A potential limitation to our approach of determining the 435 most appropriate locations for multi-species crossing structures is that preferred landscape 436 traits differed among groups, indicating that some species would benefit less from crossing 437 sites that serve the majority (for species specific preferences see Figures S1-S4). While the 438 specifics of which species should be given priority in such an instance will depend on the 439 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t conservation goals of managers, our method presents a potentially viable way of increasing 440 highway permeability for multiple species, and ultimately improving connectivity and 441 population viability for mammal communities along major roadways. 442 Although our study was limited to one section of highway, its importance as a wildlife 443 corridor suggests that our approach may be widely applicable to other areas where roads 444 bisect important wildlife habitat. In situations where managers are capable of implementing 445 mitigation measures aimed at increasing cross-road permeability for multiple mammal 446 species, future studies should seek to evaluate the efficiency of this method over traditional 447 single-species approaches. Specifically, to verify the effectiveness of our approach compared 448 to a single- species mitigation strategy, managers would ideally implement our method in 449 areas where traditional mitigation approaches have been in place for a number of years. By 450 directly comparing permeability values before and after the implementation of a multiple-451 species mitigation approach, we may gain further insight into benefits of community-level 452 conservation planning. 453 Finally we would like to acknowledge that our modeling approach only constitutes 454 one possible way of drawing inference about highway approach and crossing behavior of the 455 investigated mammal community. Here, we provide a flexible but somewhat restrictive 456 framework for predicting animal abundance. Though there is always uncertainty surrounding 457 model choice when using a multi-scale approach, extra caution should be used when basing 458 model choice on \u2018stepwise\u2019 procedures and using p-values to exclude certain models from a 459 set. The use of AIC to rank models is currently widely applied in the literature and is assumed 460 to be valid, but this approach only gives a relative measure of fit for comparing models. AIC 461 does not provide a measure for predictive ability of a model, which should ideally be tested 462 against additional data. Finally, alternatives to model averaging such as a reversible jump 463 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t MCMC approach (Green, 1995) could be employed to compare results and further improve 464 robustness of analysis. 465 466 Conclusion 467 Roads such as Hwy 3 represent severe barriers to animal movement and pose a major 468 threat to wildlife habitat, but few studies investigate how or where to implement mitigation 469 measures at the community level. We identified areas along the highway with habitat features 470 of preferred crossing sites for eight species of large mammals, representing locations where 471 mitigation measures may have positive effects for all species investigated. We determined 472 that a combined approach incorporating both remotely sensed and hand-digitized landscape 473 variables best predicted crossing site preference for most species, but that remote sensing data 474 was always better than hand-digitized values when utilized separately. Our results indicate 475 that a multi-scale approach may be necessary when identifying areas to implement mitigation 476 strategies across roads, as differing habitat requirements for members of the mammal 477 community may limit the usefulness of single-species, single-scale approaches. 478 479 Acknowledgments 480 We thank D. Quinn for logistical support throughout data collection, S.M. Alexander 481 for helpful advice on data collection and analysis, and W. Desch for initial methodological 482 and statistical advice. F. Suppan, P. Beier, J. Jenness and K. Crooks provided feedback on 483 GIS-analyses, and A.E. Passmore helped edit a previous version of this manuscript. We thank 484 G. Stewart, P. Beier, and two anonymous reviewers for comments and suggestions on an 485 earlier draft of this manuscript. 486 487 488 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Literature cited 489 Alexander, S., Waters, N., & Paquet, P. 2005. Traffic volume and highway permeability for a 490 mammalian community in the Canadian Rocky Mountains. Canadian Geographer-Geographe 491 Canadien 49(4):321\u2013331. 492 Barber, J. R., Crooks, K. R., & Fristrup, K. M. 2010. The costs of chronic noise exposure for 493 terrestrial organisms. Trends in Ecology & Evolution 25(3):180\u2013189. 494 Barton, K. 2012. MuMIn: Multi-model inference 1.8.0, http://cran.r-495 project.org/package=MuMIn. 496 Beaudry, F., deMaynadier, P. G., & Hunter, M. L. 2008. Identifying road mortality threat at 497 multiple spatial scales for semi-aquatic turtles. Biological Conservation 141(10):2550\u20132563. 498 Beier, P., Majka, D. R., & Spencer, W. D. 2008. Forks in the road: Choices in procedures for 499 designing wildland linkages. Conservation Biology 22(4):836\u2013851. 500 Beier, P., & Noss, R. F. 1998. Do habitat corridors provide connectivity? Conservation 501 Biology 12(6):1241\u20131252. 502 Ben\u00edtez-L\u00f3pez, A., Alkemade, R., & Verweij, P. A. 2010. The impacts of roads and other 503 infrastructure on mammal and bird populations: a meta-analysis. Biological Conservation 504 143(6):1307\u20131316. 505 Bennett, A. F. 1991. Roads, roadsides and wildlife conservation: a review. In D. A. Saunders 506 & R. J. Hobbs (Eds.), Nature Conservation 2: the Role of Corridors. (pp. 99\u2013118). New South 507 Wales: Surrey Beatty and Sons. 508 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Beyer, H. L. 2012. Geospatial Modelling Environment (Version 0.7.2.1). (software). URL: 509 http://www.spatialecology.com/gme. 510 Boyce, M. S. 1992. Population viability analysis. Annual Review of Ecology and 511 Systematics 23:481\u2013506. 512 Boyce, M. S., Mao, J. S., Merrill, E. H., Fortin, D., Turner, M. G., Fryxell, J., & Turchin, P. 513 2003. Scale and heterogeneity in habitat selection by elk in Yellowstone National Park. 514 Ecoscience 10(4):421\u2013431. 515 British Columbia Ministry of Transportation and Infrastructure. 2010. Traffic Data Program. 516 Burnham, K. P., & Anderson, D. R. 2002. Model selection and multimodel inference: a 517 practical information-theoretic approach. New York, NY: Springer Verlag. 518 Clevenger, A. P., & Waltho, N. 2000. Factors Influencing the Effectiveness of Wildlife 519 Underpasses in Banff National Park, Alberta, Canada. Conservation Biology 14(1):47\u201356. 520 Clevenger, A. P., & Waltho, N. 2005. Performance indices to identify attributes of highway 521 crossing structures facilitating movement of large mammals. Biological Conservation 522 121(3):453\u2013464. 523 Crooks, K. R. 2002. Relative sensitivities of mammalian carnivores to habitat fragmentation. 524 Conservation Biology 16(2):488\u2013502. 525 Crooks, K. R., & Sanjayan, M. 2006. Connectivity conservation: maintaining connections for 526 nature. In K. R. Crooks & M. Sanjayan (Eds.), Connectivity conservation (pp. 1\u201320). 527 Cambridge: Cambridge University Press. 528 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Dickson, B., & Beier, P. 2002. Home-range and habitat selection by adult cougars in 529 southern California. The Journal of wildlife management 66(4):1235\u20131245. 530 Dickson, B. G., Jenness, J. S., & Beier, P. 2005. Influence of vegetation, topography, and 531 roads on cougar movement in southern California. Journal of Wildlife Management 532 69(1):264\u2013276. 533 Elbroch, M. 2003. Mammal tracks & sign: a guide to North American species. (M. Elbroch, 534 Ed.). Stackpole Books, Pennsylvania, USA. 535 Epps, C. W., Palsboll, P. J., Wehausen, J. D., Roderick, G. K., Ramey, R. R., & McCullough, 536 D. R. 2005. Highways block gene flow and cause a rapid decline in genetic diversity of desert 537 bighorn sheep. Ecology Letters 8(10):1029\u20131038. 538 ESRI. 2009. ArcGIS 9.3. Economic and Social Reserach Institute Inc., Redlands, CA. 539 http://www.esri.com/. 540 ESRI. 2010. ArcGIS 10. Economic and Social Reserach Institute Inc., Redlands, CA. 541 http://www.esri.com/. 542 Fahrig, L., & Rytwinski, T. 2009. Effects of roads on animal abundance: an empirical review 543 and synthesis. Ecology and Society 14:21. 544 Fearer, T. M., Prisley, S. P., Stauffer, D. F., & Keyser, P. D. 2007. A method for integrating 545 the Breeding Bird Survey and Forest Inventory and Analysis databases to evaluate forest 546 bird\u2013habitat relationships at multiple spatial scales. Forest Ecology and Management 547 243(1):128\u2013143. 548 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Forman, R. T. T., & Alexander, L. E. 1998. Roads and their major ecological effects. Annual 549 Review of Ecology and Systematics 29:207\u2013231. 550 Gonzalez, A., Lawton, J. H., Gilbert, F. S., Blackburn, T. M., & Evans-Freke, I. 1998. 551 Metapopulation dynamics, abundance, and distribution in a microecosystem. Science 552 281(5385):2045\u20132047. 553 Green, P. J. 1995. Reversible jump Markov chain Monte Carlo computation and Bayesian 554 model determination. Biometrika 82(4):711\u2013732. 555 Greene, W. H. 1994. Accounting for excess zeros and sample selection in Poisson and 556 negative binomial regression models (p. 34). New York University Leonard N. Stern School 557 of Business. 558 Guisan, A., & Thuiller, W. 2005. Predicting species distribution: offering more than simple 559 habitat models. Ecology Letters 8(9):993\u20131009. 560 Haddad, N. M., Bowne, D. R., Cunningham, A., Danielson, B. J., Levey, D. J., Sargent, S., & 561 Spira, T. 2003. Corridor use by diverse taxa. Ecology 84(3):609\u2013615. 562 Harrison, S., & Bruna, E. 1999. Habitat fragmentation and large-scale conservation: what do 563 we know for sure? Ecography 22(3):225\u2013232. 564 Hilbe, J. 2008. Negative Binomial Regression (p. 251). Cambridge University Press New 565 York. 566 Holderegger, R., & Di Giulio, M. 2010. The genetic effects of roads: A review of empirical 567 evidence. Basic and Applied Ecology 11(6):522\u2013531. 568 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Iglesias, C., Mata, C., & Malo, J. E. 2012. The Influence of Traffic Noise on Vertebrate Road 569 Crossing Through Underpasses. Ambio 41(2):193\u2013201. 570 Jackman, S. 2012. pscl: Classes and Methods for R Developed in the Political Science 571 Computational Laboratory, Stanford University. http://pscl.stanford.edu/. Stanford, 572 California. 573 Jaeger, J. A. G., Bowman, J., Brennan, J., Fahrig, L., Bert, D., Bouchard, J., Charbonneau, 574 N., Frank, K., Gruber, B., & von Toschanowitz, K. T. 2005. Predicting when animal 575 populations are at risk from roads: an interactive model of road avoidance behavior. 576 Ecological Modelling 185(2-4):329\u2013348. 577 Johnson, B. K., Kern, J. W., Wisdom, M. J., Findholt, S. L., & Kie, J. G. 2000. Resource 578 selection and spatial separation of mule deer and elk during spring. Journal of Wildlife 579 Management 64(3):685\u2013697. 580 Jones, M. E. 2000. Road upgrade, road mortality and remedial measures: impacts on a 581 population of eastern quolls and Tasmanian devils. Wildlife Research 27(3):289\u2013296. 582 Lambert, D. 1992. Zero-Inflated Poisson Regression, with an Application to Defects in 583 Manufacturing. Technometrics 34(1):1\u201314. 584 Lingle, S., & Wilson, W. F. 2001. Detection and Avoidance of Predators in White-Tailed 585 Deer (Odocoileus virginianus) and Mule Deer (O. hemionus). Ethology 107(2):125\u2013147. 586 Mader, H. J. 1984. Animal habitat isolation by roads and agricultural fields. Biological 587 Conservation 29(1):81\u201396. 588 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Malo, J. E., Suarez, F., & Diez, A. 2004. Can we mitigate animal-vehicle accidents using 589 predictive models? Journal of Applied Ecology 41(4):701\u2013710. 590 Mayor, S. J., Schaefer, J. A., Schneider, D. C., & Mahoney, S. P. 2007. Spectrum of 591 selection: new approaches to detecting the scale-dependent response to habitat. Ecology 592 88(7):1634\u20131640. 593 McClure, C. J. W., Rolek, B. W., & Hill, G. E. 2012. Predicting Occupancy of Wintering 594 Migratory Birds : Is Microhabitat Information Necessary? The Condor 114(3):482\u2013490. 595 McGregor, R. L., Bender, D. J., & Fahrig, L. 2008. Do small mammals avoid roads because 596 of the traffic? Journal of Applied Ecology 45(1):117\u2013123. 597 Mech, L. D. 1995. The challenge and opportunity of recovering wolf populations. 598 Conservation Biology 9(2):270\u2013278. 599 Meidinger, D., & Pojar, J. 1991. Ecosystems of British Columbia. Victoria, BC: British 600 Columbia Ministry of Forests. 601 Meiman, S., Civco, D., Holsinger, K., & Elphick, C. S. 2012. Comparing Habitat Models 602 Using Ground-Based and Remote Sensing Data: Saltmarsh Sparrow Presence Versus 603 Nesting. Wetlands 32(4):725\u2013736. 604 Noss, R. F., & Cooperrider, A. Y. 1994. Saving nature\u2019s legacy: protecting and restoring 605 biodiversity. (R. F. Noss & A. Y. Cooperrider, Eds.). Island Press, Washington D.C., USA. 606 Noss, R. F., Quigley, H. B., Hornocker, M. G., Merrill, T., & Paquet, P. C. 1996. 607 Conservation biology and carnivore conservation in the Rocky Mountains. Conservation 608 Biology 10(4):949\u2013963. 609 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t R Development Core Team. 2012. R: A language and environment for statistical computing 610 2.15.2, http://www.r-project.org. Vienna, Austria: R Foundation for Statistical Computing. 611 Rawlings, J. O., Pantula, S. G., & Dickey, D. A. 1998. Applied Regression Analysis: A 612 Research Tool (p. 657). Springer Verlag. 613 Schuster, R., & Arcese, P. 2013. Using bird species community occurrence to prioritize 614 forests for old growth restoration. Ecography 36(4):499\u2013507. 615 Seiler, A. 2005. Predicting locations of moose-vehicle collisions in Sweden. Journal of 616 Applied Ecology 42(2):371\u2013382. 617 Selva, N., Kreft, S., Kati, V., Schluck, M., Jonsson, B.-G., Mihok, B., Okarma, H., & Ibisch, 618 P. L. 2011. Roadless and Low-Traffic Areas as Conservation Targets in Europe. 619 Environmental Management 48(5):865\u2013877. 620 Sheldon, L. 1997. Animal Tracks of the Rockies. (L. Sheldon, Ed.). Lone Pine Publishing, 621 Edmonton, Canada. 622 Singleton, P., & Lehmkuhl, J. 1999. Assessing wildlfie habitat connectivity in the Interstate 623 90 Snoqualmie Pass Corridor, Washington. In G. L. Evink, P. Garrett, & D. Zeigler (Eds.), 624 Proceedings of the third international conference on wildlife ecology and transportation (pp. 625 75\u201384). Florida Department of Transportation, Tallahassee, Florida, USA. 626 Thompson, I. D., Davidson, I. J., Odonnell, S., & Brazeau, F. 1989. Use of track transects to 627 measure the relative occurrence of some boreal mammals in uncut forest and regeneration 628 stands. Canadian Journal of Zoology-Revue Canadienne de Zoologie 67(7):1816\u20131823. 629 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Trombulak, S. C., & Frissell, C. A. 2000. Review of ecological effects of roads on terrestrial 630 and aquatic communities. Conservation Biology 14(1):18\u201330. 631 Underhill, J., & Angold, P. 2000. Effects of roads on wildlife in an intensively modified 632 landscape. Environmental Reviews 8(1):21\u201339. 633 Van Dyke, F. G., Brocke, R. H., & Shaw, H. G. 1986. Use of road track counts as indices of 634 mountain lion Felis Concolor presence. Journal of Wildlife Management 50(1):102\u2013109. 635 Venables, W. N., & Ripley, B. D. 2002. Modern Applied Statistics with S (Fourth.). New 636 York: Springer. 637 Vuong, Q. H. 1989. Likelihood Ratio Tests for Model Selection and Non-Nested Hypotheses. 638 Econometrica 57(2):307\u2013333. 639 White, G. C., & Burnham, K. P. 1999. Program MARK: survival estimation from populations 640 of marked animals. Bird Study 46(1 supp 1):120\u2013139. 641 Wulder, M. A., White, J. C., Cranny, M., Hall, R. J., Luther, J. E., Beaudoin, A., 642 Goodenough, D. G., & Dechka, J. A. 2008. Monitoring Canada\u2019s forests. Part 1: Completion 643 of the EOSD land cover project. Canadian Journal of Remote Sensing 34(6):549\u2013562. 644 Yellowstone to Yukon Conservation Initiative. 2013. Y2Y Priority Areas. http://y2y.net/our-645 work/priority-areas. Accessed: 2013-March-26. 646 Zeileis, A., & Croissant, Y. 2010. Extended Model Formulas in R: Multiple Parts and 647 Multiple Responses. Journal of Statistical Software 34(1):1\u201313. 648 Zeileis, A., Kleiber, C., & Jackman, S. 2008. Regression models for count data in R. Journal 649 of Statistical Software 27(8):1\u201325. 650 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t road barrier effect : We standardized the highway crossing attempt and transect survey data by the number of 12 hour periods that had elapsed since the time of the last snowfall to correct for time effects (Thompson et al., 1989). For calculation of the road barrier effect, we standardized survey data for the highway and transects by kilometers surveyed: 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t Crossings per km=Toral number of tracks Total length of surveys We then calculated the permeability of the highway by standardizing the crossings per km of highway with the crossings per km of transect: Permeability= Highway crossings per km Transect crossings per km We also constructed track accumulation curves along the 95km of highway for all four species groups to identify areas of the highway with greater crossing intensity for each mammal group. MULTI-SCALE LANDSCAPE VARIABLES To develop our micro-habitat assessments, we imported the collected GPS data into ArcGIS 9.3 (ESRI, 2009). The GPS points from the highway surveys and the transect surveys were set on top of a georeferenced (WGS 1984, UTM Zone 11N) orthophotograph layer from 2004, with a spatial resolution of 1m, provided by a Web Map Service (WMS) of GeoBC (http://www.geobc.gov.bc.ca). For each GPS point, we created a circular buffer of 200m to represent the perceptual area of the animal directly influenced by the surrounding landscape predictors (e.g. Lingle & Wilson, 2001), which we define as \u2018perceptual area polygon\u2019. For each buffer area, we digitalized polygons for predefined landscape predictors and georeferenced them using the orthophotograph layer and Google Earth, as the latter provided more recent images of the research area. We used the following landscape predictors, adapted from Dickson, Jenness, & Beier (2005): forested (forest + woodland), shrub, herbaceous (grassland + agriculture), riparian, water, non-vegetated (gravel, rock +dirt), highway (+shoulder), road/path, railroad, residential, developed, disturbed and wetland (Table 1). We then calculated the percentage of each buffer area overlapped by each landscape predictor. Because large mammals might respond to both fine and coarse scale habitat features (e.g. Mayor et al., 2007), we developed a series of variables describing macro-habitat landscape 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t features at three spatial scales: 200m, 500m, and 1 km. For modeling species abundance along the highway and transects, we chose candidate predictor variables based on their ability to predict species abundance at site and landscape levels in similar studies (e.g. Malo, Suarez, & Diez, 2004; Guisan & Thuiller, 2005). All remotely sensed predictors (Table 1) were derived from the following sources: Terrain Resource Information Management (TRIM, Province of BC 1992) and Earth Observation for Sustainable Development Landcover (EOSD LC 2000, Wulder et al., 2008). Our dataset comprised 12 predictor variables from the perceptual area polygons and 11 from remote sensing on 3 scales (200m, 500m, 1km; Table 1), derived at each of 463 highway locations and 308 transect locations. All remote sensing predictors were created using Geospatial Modelling Environment (Beyer, 2012) in conjunction with ArcGIS 10 (ESRI, 2010) and R v. 2.15.2 (R Development Core Team, 2012). Due to their widely varying scales, all predictors were standardized to mean=0, sd =1 to ensure that their importance was not driven by measurement scale (White & Burnham, 1999). LANDSCAPE VARIABLE PREFERENCE MODELS Since predetermining the appropriate data distribution for our count data from ecological knowledge alone was not possible, we modeled abundance incorporating both a Poisson distribution (P) and negative binomial (NB) distribution to account for potential overdispersion (e.g. Zeileis, Kleiber, & Jackman, 2008). Because of the large proportion of zero values included in our data-set, we also applied zero-inflated models (ZIP, ZINB; Lambert, 1992), which are mixture models that combine both count data and a binomial model. To determine which of these distributions best represented our species data, we visually inspected the data and compared the log Likelihood, AIC, and number of correctly predicted zeros for each distribution model fits using intercept-only models. To test for differences among distribution functions, we used likelihood ratio tests to compare the Poisson 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t and negative binomial distributions, since the Poisson distribution is a restriction of the more general negative binomial distribution (Hilbe, 2008). We tested H0 for no difference between the two and H1 that the negative binomial was a better fit to the data. We tested the same hypothesis using the zero-inflated Poisson and zero-inflated negative binomial. Next, we used a Vuong test (Vuong, 1989; Greene, 1994) to evaluate whether the zero-inflated models were a statistically better fit to the data than their base model (Hilbe, 2008). The Vuong test is generally formulated as: V =\u221a ( N )\u2217mean ( m ) sm\u2217m m= ln ( \u03bc1 \u03bc2 ) Where \u03bc1 = predicted probability of y for the zero-inflated model, \u03bc2 = predicted probability of y for the base model, sm = standard deviation of m, and N = number of observations in each model, where both must use the same observations. The test statistic V is asymptotically normal. If V > 1.96, the zero-inflated model is preferred; if V< -1.96, the base model is preferred; and if the value of V is between -1.96 and 1.96 neither model is preferred (Hilbe, 2008). To perform these tests we used the function vuong in R package pscl v.1.04.4 (Zeileis, Kleiber, & Jackman, 2008; Jackman, 2012). We compared our measures of model selection (AIC, LogLik, predicted zeros, Vuong) for all four distributions (P, NB,ZIP, ZINB) throughout each of the model building stages of this study to avoid bias in predetermining the distribution with intercept-only models. For each of the four mammal groups (deer, elk, moose, carnivores), we compared models of predicted habitat preference using each distribution (P, NB, ZIP, ZINB) and data set (Highway, Transect) across six separate spatial approaches: i) 200m scale, ii) 500m scale, iii) 1km scale, iv) all 3 scales combined; v) perceptual area polygons; vi) all scales and perceptual area polygons combined. For spatial approaches iv) and vi), we created an iterative model fitting procedure that starts with an intercept only base model, individually adds predictor variables, records the results of each 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t fitted model, and retains all top-ranked models (\u0394AICc \u2264 2) at each iteration as base models for subsequent iterations as long as there is reduction in AICc (R function in R-Code S1). The fitting procedure constitutes an extension of a more restrictive routine that only included the top ranked model for each iteration in subsequent iterations (Schuster & Arcese, 2013). We opted for the iterative approach because creating all possible models for approach iv) and vi) would have resulted in 2^45 models each. For the remaining approaches, we created models for each possible combination of predictors. For both ZIP and ZINB we further expanded our model lists using a) intercept only models for the zero-inflation component, while using predictors in the count component and b) including the same predictors for both zero-inflation and count components. In post-processing we reduced the candidate set of models from each approach based on the statistical significance of all predictors, using p-values as a general and liberal criterion for retaining models. We selected a cutoff value of p = 0.15 as it serves as the default for many stepwise model selection approaches (e.g. Rawlings, Pantula, & Dickey, 1998). We chose this approach to reduce the probability of including non-informative models (i.e., those stuck at local maxima for parameter estimation) in subsequent model averaging. Additionally we checked for and removed models with unrealistically high (>50) parameter estimates and/or Standard Errors, which would indicate lack of model fit. For model selection, we ranked all remaining candidate models by AICc and averaged those with \u2206 AICc \u2264 2 from the top ranked model (Burnham & Anderson, 2002). All analysis were conducted using R v.2.15.2 (R Development Core Team, 2012); package MuMIn 1.8.0 (Barton, 2012) was used for AICc calculations; package MASS v. 7.3-22 (Venables & Ripley, 2002) for NB models; and packages pscl and Formula v.1.1-0 (Zeileis & Croissant, 2010) for ZIP and ZINB models. To determine the distribution and scale that provided the best relative fit to the data for each mammal group we compared and ranked the models with the lowest AICc of each approach 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t and determined the approach resulting in the overall lowest AICc value. We further contrasted these results with the initial results from the intercept only models to determine whether initial models were sufficient to identify the error distributions that were most appropriate for a given data set or whether predictor variables had to be included first. Highway permeability values for the majority of groups were extremely low (where a value of 1 indicates full permeability across the highway and 0 represents no permeability), indicating that Hwy 3 likely acts as a barrier to mammal movement (Table 2). The permeability values for carnivores were only one third those of ungulates (moose, elk, deer, combined) on the investigated section of Hwy 3 (Table 2), indicating that deer, elk, and moose were much less affected by the highway in terms of movement than carnivores. Track accumulation curves for all mammal groups indicate that in all four cases there were areas of the highway where the focal group rarely or almost never crosses the highway (Figure 2). LANDSCAPE VARIABLE PREFERENCE MODELS Based on intercept only model comparisons using likelihood ratio tests, Vuong test and AIC ranking, the best supported distributions for each mammal group were: ZINB (Deer-Hwy, Elk-Hwy, Deer-Trans), NB (Moose-Hwy, Elk-Trans, Carnivora-Trans), ZIP (CarnivoraHwy) , P (Moose-Trans), indicating that ZINB and NB were the most commonly supported distributions (Table 3). In 6 out of 8 cases, the modeling approach which included predictors from all three scales and the digitized polygons was selected as the top model based on AICc (Table 4). In only two cases for the transect data did other approaches result in lower AICc values: Deer (1km scale) and Moose (500m scale). In direct comparison between predictors derived from remotely sensed data and hand digitized data, the remotely sensed model framework resulted in lower AICc values in all eight cases. When comparing the remotely 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t sensed data approach at the same scale as the digitized (200m) data, digitized predictors resulted in lower AICc values in 7/8 cases. A comparison of the extended modeling results (Table 4) with the initial distribution tests (Table 3) indicates that in 4/8 cases the results from the initial tests were rejected and different distributions formed the basis of models with the lowest AICc values. For Highway and Transect data from each mammal group, model coefficients describing preferred (positive values) and avoided (negative values) habitat variables from the best supported model (above) are depicted in Tables S1 and S2. Summed values of preferred and avoided landscape variables for the entire mammal community are presented in Table 5. predictive maps of preferred crossing sites : Using the previous model results we created predictive maps for each species/group for both the abundance of animals approaching the highway (transect models) and the abundance of animals, having reached the highway, crossing it (highway models). For each map we chose the approach with the lowest AICc values out of the 4 remote sensing-derived frameworks (200m, 500m, 1km, 3 scales), as landscape level data was not available for the perceptual area polygons. For predictive polygons, we used 30x30m polygons to follow the EOSD resolution. For the highway predictions we created polygons around the highway (line feature buffered 15m on each side of highway) resulting in predictions for 7374 polygons. For transect predictions we expanded the buffer around the highway to 1km, resulting in 213442 polygons. Next, we generated a predictor set for each polygon centroid that was identical to those used for survey points, and then estimated abundance based on our averaged models for each of the focal species groups. We combined individual mammal group abundance estimates into 10 quantiles to consolidate focal species maps into an index of site preference, multiplied those scores for each polygon and standardized them by dividing by 1000, resulting in community site preference scores between 0 (being the lowest preference) and 10 (highest preference). To map preferred crossing sites, we used averaged model results for each mammal group based on the framework with the lowest AICc value out of the 4 remotely sensed model sets (Cells marked with an asterisk in Table 4; maps and shapefiles in Figures S1-S4 and Data S2 respectively). The combined predictions of preferred (green) and avoided (red) crossing sites for all investigated species within the mammal community are illustrated in Fig. 3. Based on predictions generated from landscape variables (above), certain regions of the study area exhibited high preference scores from both approach (transect) and crossing (highway) models (e.g., Fig. 3 insert A), indicating that these locations likely represent areas of high priority when implementing mitigation measures for all species considered in our study. Conversely, certain regions of the study area exhibited high preference scores for one of the model sets (crossing vs. approach), but not the other (e.g., Fig. 2 insert B), indicating that these may represent less-ideal locations to implement mitigation measures such as crossing structures. Areas of unambiguous preference for particular crossing sites (i.e., those where crossing and approach preference scores overlap) differ for each mammal group considered in our study 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 PeerJ reviewing PDF | (v2013:04:385:1:0:ACCEPTED 8 Oct 2013) R ev ie w in g M an us cr ip t (Figures S1-S4), indicating that mitigation strategies aimed at mammal communities may differ substantially from those aimed at a target species.", |
