Do available products to mask human scent influence camera trap survey results?

D. Mu ñ oz (dmunoz@elon.edu), Dept of Environmental Studies, Elon Univ., Elon, NC 27244, USA. Present address: 435 Forest Resources Building, University Park, PA 16802, USA. – J. Kapfer, Dept of Biological Sciences, Upham Hall Rm 352, Univ. of Wisconsin-Whitewater, Whitewater, WI 53190, USA. – C. Olfenbuttel, Division of Wildlife Management, North Carolina Wildlife Resources Commission, 1239 Laurel Bluff s, Pittsboro, NC 27312, USA

Camera traps (i.e. remotely, motion or heat triggered cameras) can be used to examine a variety of ecological factors ranging from the activity patterns (Oliveira-Santos et al. 2008, Harmsen et al. 2010 to habitat selection (Fedriani et al. 2000, Kelly andHolub 2008) of wildlife. Camera traps are also used to estimate abundance, density and the distribution of secretive or rare species (Karanth and Nichols 1998, Trolle and K é ry 2003, Larrucea et al. 2007). Camera traps may be preferable to traditional mammal trapping techniques for various reasons, such as greater eff ectiveness for cryptic species (Sanderson andTrolle 2005, De Bondi et al. 2010) and high detection rates (Silveira et al. 2003, Gompper et al. 2006, Balme et al. 2009).
Despite their widespread use, there are still questions regarding appropriate protocols for the use of camera traps (Rowcliff e et al. 2011, Hamel et al. 2013; reviewed by Rovero et al. 2013). For example, little eff ort has been placed on assessing how disturbance associated with researcher presence at cameras during maintenance (i.e. checking batteries and memory cards) infl uences capture rates. Past studies have noted that animals can learn to avoid camera locations or are generally wary of camera traps, which may be due to either camera fl ash or disturbance associated with researcher activity (Cutler and Swann 1999, Sequin et al. 2003, Lyra-Jorge et al. 2008. Even minimal researcher disturbance may result in deposition of scent that may alarm wary species and result in avoidance of the area. Despite the need for further research on this topic, there has been no attempt to rigorously examine the role of human scent, or the masking of human scent, on camera trap eff ectiveness. Our goal was to ascertain if the number of wildlife detections diff er when human scent is masked versus unmasked while researchers perform regular camera maintenance. We hypothesized that wildlife capture rates would diff er based on the presence of researcher scent, and we predicted captures would be higher on cameras where human scent was masked. We also believed that if no treatment eff ect was observed, two possible conclusions could be made: 1) the target species in the study region does not alter behavior due to human scent at camera trap stations and/or 2) employing available scent-masking products does not improve capture rates during camera trap surveys. Th e results of this study have potentially broad implications for the utility of this increasingly common survey technique.

Methods
Research was conducted at two sites in the Piedmont region of North Carolina, USA. We selected sites based on their similar habitat characteristics, remote location, and limited human traffi c. Th e fi rst site, referred to as the ' Haw River ' site, is 16.2 ha of privately owned land in Alamance County, North Carolina. Th e property borders the Haw River and is predominantly alluvial forest habitat as described by Spira (2011). Th is rural locality is mainly exurban with minor components of agriculture nearby. Th e second site, referred to as the ' Rocky River ' site, is located in Chatham County, North Carolina. Th e 12.9-ha study site is bordered by the Rocky River, and is comprised of oak -hickory forest (Spira 2011). Th is locality has been mostly uninhabited with only scattered agriculture close by. During the course of our study, human activity was documented once at the Haw River site (hiker) and twice at the Rocky River site (one hiker, one hunter). Th e Haw River site was surveyed outside of the recreational hunting season, and such activities are generally not permitted on this property. However, it is known that adjacent property owners hunt regularly. At the Rocky River site, hunting was also not permitted. On one occasion hunters were seen onsite, and on other occasions, hunters were heard in the surrounding area.
To test our hypothesis, we deployed eight camera traps at each site in total. We randomly applied one of two treatments, ' scent unmasked ' or ' scent masked ' , to each of the eight cameras at each site, which resulted in four replicates of each treatment per site. Camera traps were placed semi-randomly within each study site, using randomly generated initial locations from a geographic information system (GIS). We chose the nearest appropriate location within 15 m of the predetermined random locations, e.g. along a game trail, for camera deployment to ensure the highest chance of detection (Moen and Lindquist 2006, Rowcliff e et al. 2008, Brown and Gehrt 2009, O'Connell et al. 2011. We locked cameras in steel boxes and affi xed them to trees at 23 -27 cm above ground level to better capture both small and large animals (Kelly 2008). Cameras were oriented north to avoid issues associated with receiving direct sunlight (Brown and Gehrt 2009). We attempted to space cameras in order to reduce possible interaction between treatments. Given the limited size of our study sites, cameras were located at least 75 -150 m from their nearest neighbor. Cutler and Swann (1999) suggested that traditional white fl ash cameras may infl uence mammal activity. Although it is unclear if cameras with ' no glow ' infrared fl ash off er signifi cant benefi ts (reviewed by Rovero et al. 2013), we used a ' no glow ' camera to minimize potential confounding eff ects. Cameras were set to take fi ve pictures when their passive infrared detectors were triggered (trigger speed: 0.3 s). Camera sensitivity was set to high with no recovery time between triggers. Once cameras were placed, they remained in the same location during the duration of the study.

Experimental treatments
Prior to treatment application, cameras were deployed for one month without researcher disturbance. Th is was done to limit the impacts of scent deposition during camera installation. Pictures obtained in this baseline period were not directly comparable to data collected during treatment application period. Th us, pictures of wildlife from the baseline period were excluded from analyses, although patterns detected between pre-and post-treatment were used to make suggestions for future experimental design. Once treatment application began, we visited cameras for maintenance every two weeks (hereafter referred to as sampling period). Camera maintenance occurred over two subsequent days. All cameras in the scent unmasked treatment were visited on day one, while all cameras in the scent masked treatment were visited on day two. We used a GPS to map routes to each camera so that a maximum possible distance (100 -150 m) was maintained from adjacent cameras in the opposite treatment.
For cameras in the unmasked treatment, normal fi eldwork clothes were worn to mimic how a typical researcher might visit a camera location. For cameras in the scent masked treatment, we used a suite of commercially available scent-masking products shown to minimize or eliminate human-scent output (e.g. Pickering v. A.L.S. Enterprises 2012). We chose common commercial brands because we were interested in investigating the eff ectiveness of readily available items that researchers may employ. Scent-masking clothes include carbon fabric layers that reportedly bind to odor molecules, adsorbing them and preventing their release. Clothing consisted of scent-controlling boots, pants, socks, shirts, jacket, head-wear and gloves. In between site visits, these items were stored in a scent-obscuring bag that contained leaves and twigs from the respective study sites to further help mask unusual odors that may persist on clothing.
Th e outfi t was washed every six weeks with an odoreliminating detergent, following manufacturer ' s guidelines. Additionally, prior to maintenance of scent-masked cameras, researchers bathed with shampoo, conditioner, and soap meant to obscure scent, and they applied scent-masking deodorant. A set of non-specialized clothing was washed in detergent to act as a ' transition outfi t ' during travel to the research site. Once on-site, the specialized outfi t was carried to the edge of the study area and then adorned. Before entering the study area, the researcher lightly misted the outfi t with a spray to further ensure that any incidental scent accidentally transferred to the outfi t was eliminated. Before handling cameras, scent blocking spray was re-applied to the outside of the gloves to remove any incidental scent from accidental contact with researcher skin or hair. After each camera was handled, a fi ne mist of scent-blocking spray was applied to the camera's security enclosure. Upon leaving the site, the researcher changed back into the transition outfi t and the scent-obscuring clothes were appropriately stored. Th e products include proprietary formulas, so no information is available on their active ingredients. From the company website, these products rely on converting odor molecules, oxidation, bonding of molecules, and neutralization of odor ( Ͻ www.hunterspec.com/products/all/all/ Scent-A-Way Ͼ ; accessed 3/25/2014). While the effi cacy of such scent-masking products is hotly debated, product testing has shown scent-masking clothes can adsorb up to 99% of produced odors (Pickering v. A.L.S. Enterprises 2012).

Data analysis
Many complications can arise when analyzing camera trap images, such as multiple individuals captured in a single image, a long image series taken of a single individual, or false trigger events (Royle et al. 2009). Th e variable of interest during this study was the number of wildlife detections during each sampling period (capture rate). Th erefore, if multiple animals were captured in a single image, they were each counted as separate detection events. In addition, if an animal spent extended amounts of time in front of the camera (based on time stamp), dozens of images would only count as one detection event. Likewise, if an animal left the fi eld of view and returned from the same side it originally departed within two minutes, we did not consider it a new detection event. Th erefore, each sampling period generated a count of animal detections (hereafter referred to as ' captures ' ), which we statistically analyzed.
We compared captures between treatments with a general estimating equation (GEE) analysis ( α ϭ 0.05). GEEs are a ' semi-parametric ' extension of the generalized linear model (GLM) and allow for analysis of repeated measures in a similar fashion to a repeated-measures analysis of variance (ANOVA). However, GEEs are particularly robust to data that break assumptions of normality and independence (Nelder andWedderburn 1972, Ballinger 2004). Furthermore, GEEs are highly appropriate for analysis of count data due to their quasi-likelihood method of estimation (Zeger et al. 1988, Ballinger 2004. Because wildlife are not uniformly distributed across a landscape, captures were highly associated with camera location, so we selected an exchangeable correlation structure to account for this.
Standard error was calculated with a model-based estimator which performs better for data with few subjects and many repeated measures (Hardin and Hilbe 2002). Th e model selected for analysis was Poisson-loglinear (Gardner et al. 1995, O'Connell et al. 2011 . Considering many mammal species exhibit seasonal variation in activity, we included survey period as another explanatory variable. For all species analyzed we assessed the likelihood of our models (treatment only, survey period only, both treatment and survey period) with quasi-Aikake's information criterion (QIC; Pan 2001). All statistical analyses were conducted with SPSS 20.

Results
Camera traps were active at the Haw River site from early February to mid-July 2011, with each camera simultaneously in operation for 158 trap-nights. In July 2011, three cameras were stolen from this property, and research at the site immediately ceased. Th e study resumed at the Rocky River site and ran from September 2011 to the beginning of March 2012, with each camera simultaneously in operation for 210 trap-nights.
We obtained 2085 and 3358 mammal captures at the Haw River and Rocky River site respectively. Between both sites, 11 mammal species were observed (Table 1). Th e most frequently captured species at both sites were white-tailed deer Odocoileus virginianus , eastern gray squirrel Sciurus carolinensis and raccoon Procyon lotor (Table 1). Several species were omitted from statistical analyses due to low capture rates (Table 1), including the eastern cottontail rabbit Sylvilagus fl oridanus at the Rocky River site.
Our analyses revealed that treatment is a likely factor explaining diff erences in white-tailed deer captures at the Haw River site only (p ϭ 0.013; QIC ϭ 0.376.8; mean scent masked captures/survey period ϭ 5.23 Ϯ 0.63 SE; 95% CI ϭ 4.1 -6.6; mean unmasked captures/survey period ϭ 3.38 Ϯ 0.49 SE; 95% CI ϭ 2.5 -4.5; Fig. 1). Although not formally compared, captures of white-tailed deer at the Haw River diff ered between baseline and treatment period. For example, we obtained fewer deer captures by cameras that would receive the scent-masked treatment (4.75 Ϯ 2.1 SE) than captures by cameras that would receive the unmasked treatment (16.75 Ϯ 15.5 SE) during the baseline period (Fig. 1). Similar ' switch ' patterns were seen for raccoon and eastern cottontail at the Haw site, and opossum at the Rocky site. For the other species at both sites, survey period (seasonality) was a stronger predictor of captures (Table 2, 3). However, when examining the treatment eff ects for species at both sites, there are general trends that imply scent (or scent masking products) cannot be ruled out as aff ecting mammal activity at camera locations (Table 4). Even though the use of scent-masking products did not signifi cantly aff ect model selection for other species, we include the confi dence intervals of the scent eff ect sizes to give a sense of the strength of our results, as recommended by Steidl et al. (1997) and Johnson (2002).

Discussion
Our hypothesis and prediction (i.e. that capture rates would diff er and be higher at scent masked cameras) were only statistically supported for white-tailed deer at the Haw River site. Other species did not exhibit signifi cant treatment responses, yet the treatment related eff ect sizes for most species indicate that scent-masking could have a more subtle eff ect than our study was able to detect (Table 4). Many had an average GEE slope parameter (Beta) that indicated higher captures at cameras where scent was masked. Th us, the impacts of human scent and scent-masking products on wildlife activity and survey eff ectiveness appear to be complex.
Our focal species are generally wary of human activity, with the exception of habituated individuals in more urban or suburban areas. Our inability to detect a diff erence based on treatment may seem unexpected for several of the species that we captured, which are often considered scentmotivated (e.g. raccoons). Yet, it is likely that scent motivation in these species relates to food rather than aversion to humans. It is not surprising that white-tailed deer showed a response based on treatment type. Th is species is hunted recreationally in North Carolina. Hunting pressures are known to cause changes in home range size, movement and activity patterns of white-tailed deer (Kilpatrick and Lima 1999). Th is may be the result of an anthropogenic ' landscape of fear ' created by hunters. As reviewed by Laundr é et al.
(2010), it is benefi cial for prey species to maintain a baseline level of fear of predation. Without such fear, prey species may undertake risky behavior that could lead to mortality. Much like the fear that elk Cervus canadensis feel due to the threat of predation by wolves Canis lupus , human hunters may instill fear in various game species in the vicinity of our While white-tailed deer showed a treatment eff ect at the Haw River site, this trend was not detected at the Rocky River site. Th ere was more documented human activity at the Rocky River site, and given the presence of hunters nearby on several occasions, it is possible that undocumented trespassing could have confounded treatments in certain survey periods. Additionally, it is possible that potential diff erences in the density of white-tailed deer per site, subtle variation in habitat composition at each site, diff erences in the surrounding landscape matrix, or the time of year during which each survey occurred precluded any treatment eff ect at the Rocky River site. Th is last factor may play the biggest role. Th e rate of human scent deposition and the duration of scent retention in the area surrounding camera traps is likely to be higher in the hotter months of the spring and summer in North Carolina. Th e autumn and winter months in which surveys took place at the Rocky River site would be when researcher odor output, due to sweating, would be at a minimum. Additionally, during the months of study at the study sites, such as white-tailed deer (Laundr é et al. 2001, Ciuti et al. 2012. As a result, game species may associate human scent with increased risk of mortality in our study areas and shift activity away from camera locations where scent was unmasked. We found evidence to support this notion by descriptively comparing the capture rates of white-tailed deer during the baseline period to the treatment period. For example, during the baseline period at the Haw site we had more white-tailed deer detections at cameras that would be scent unmasked versus scent masked (February; Fig. 1). Once treatment application began, human scent may have suffi ciently altered activity levels at unmasked camera locations. For species other than white-tailed deer, weaker scent eff ects may be due to lower hunting pressures or to an inconsistent response to human scent. Th e ' landscape of fear ' concept requires that animals identify human activity as a threat. Low hunting pressure may reduce the perceived threat, resulting in a weaker response to human scent. Figure 1. White-tailed deer captures among survey periods ( Ϯ SE) at the Haw River site (Alamance County, NC). Grey represents cameras in the unmasked treatment, and white represents cameras in the scent-masked treatment. February is baseline data and was excluded from formal analyses. work to theirs because the perceived ' rewards ' associated with scent in these studies diff er (e.g. a camera trap versus a prey item in a nest). Our study is the fi rst to examine whether scent-masking products aff ect camera trap surveys, and our results are useful as a spring-board for future research on this topic. Th ere are several potentially confounding variables inherent in a study of this nature that we attempted to address. First, persistence and range of detection of olfactory stimuli are extremely diffi cult to assess. We visited cameras only every two weeks to help control for this, but it is unknown how long our scent remained on-site. It was also diffi cult to determine a priori how far apart cameras should be spaced to avoid cross-treatment contamination. We attempted to off set this by placing cameras as far apart as our study site boundaries would allow and by mapping widely-spaced travel routes for servicing cameras. However, we recognize the size of our study areas may have confounded treatment eff ects. We recommend that future research on human scent and camera traps include a greater number of large study areas that allow greater spacing of cameras, a higher number of camera traps, longer camera deployment periods, and simultaneous surveys at multiple sites. Given the ' switch ' in some species ' capture rates between the baseline and treatment periods, we believe that a crossover design would provide stronger evidence to support an eff ect from human scent. It may also be benefi cial to add a third treatment, including items that are well-saturated with human scent (such as unwashed socks) to provide stronger negative stimulus for passing wildlife. Th is could potentially allow for a better determination between a scent response and a scent-masking product response.
Haw River site, a regional drought occurred, whereas precipitation levels were higher during research at the Rocky River site. Scent is harder to track during frequent precipitation, and environmental conditions play a large role in scent detection and persistence (Regnier and Goodwin 1977) as seen in search-dog studies (Shivik 2002). Th us, the likelihood that researcher scent deters wildlife from a given area may be higher in warmer, drier seasons.
Several past camera trap studies have commented on the need for research to assess the infl uence of human activity and scent on camera trap eff ectiveness (Cutler and Swann 1999, Lyra-Jorge et al. 2008, Rowcliff e et al. 2008. While there have been camera trap studies that address the response of wildlife to olfactory cues as attractants (Monterroso et al. 2011) and the impact of human activity on wildlife (Griffi ths and van Schaik 1993, Ngoprasert et al. 2007, Ohashi et al. 2013, none have explicitly addressed the subtler eff ect of researcher scent. In the latter studies, mammals in areas of high human disturbance were more active at night, when humans were less so. Several noncamera trap studies report that the infl uence of human scent and scent-masking products had variable eff ects on nest or seed predation by mammalian predators. Some found that masking human scent decreased predation (Whelan et al. 1994, Duncan et al. 2002 whereas others found no scent-related eff ect (Skagen et al. 1999, Donalty andHenke 2001). Furthermore, Shivik (2002) reported that the eff ectiveness of trained search-dogs at fi nding targets was not altered by scent-masking clothes. By simultaneously using hygiene, clothing, and chemical products, our methods may address scent-eff ects more comprehensively than the aforementioned studies; however, we cannot directly compare our Our results indicate that selection of appropriate camera locations and implementation of surveys during seasons of high wildlife activity may be more important than masking human odor to conduct eff ective camera trap surveys for species that show no strong aversion to human activity. In other words, researcher use of commonly available scent-masking products may not substantially increase camera trap capture rates for many Piedmont mammals. It is important to note that our results may not be representative of other locales or species. For example, canids and felids are generally wary of human activity (Sequin et al. 2003), and we would expect them to be more sensitive to scent at camera locations. Unfortunately, our data did not yield enough captures to include them in our analysis. Furthermore, the Piedmont region of North Carolina possesses a high human population, so there are few areas where animals have not experienced some level of human activity. Despite this, we were able to detect that human scent (or the masking of human scent) potentially aff ected the activity of a species that adapts well to suburban areas, where human scent should be prevalent. It seems likely that human scent could have a larger impact on camera trap surveys for species that 1) exist in lower densities than whitetailed deer, 2) are in areas where human scent is less common (and may therefore be perceived as a novel threat), or 3) are extremely cautious and do not acclimate well to human scent or activity. Because our results cannot defi nitively support that some species exhibit scent-related eff ects, it might be benefi cial for camera trap studies to take some scentmasking precautions to maximize eff ectiveness. Th is is again particularly true if wary species (i.e. canids and felids) are the focus. In as much as camera traps are used for monitoring rare, declining, or endangered species, ensuring that camera trap surveys capture and detect animals eff ectively is important for accurately informing conservation decisions. Th erefore, we believe our study provides a good starting point for further research that addresses the role of human scent and scent-masking products in camera trap surveys.