A novel camera trap design for studying wildlife in mountain glacier ecosystems yields new insight for glacier biodiversity in the Pacific Northwest, USA

Context The global recession of glaciers and perennial snowfields is reshaping mountain ecosystems. Beyond physical changes to the landscape and altered downstream hydrology, the implications of glacier decline for biodiversity are poorly known. Before predictions can be made about how climate change will affect wildlife in glacier-associated ecosystems, a more thorough accounting of the role that glaciers play in species’ life histories is needed. However, typical approaches for documenting wildlife presence and behavior—remote camera traps—are difficult to use in glaciated terrain due to limited options for securing them (e.g., no trees) and dramatic seasonal changes in snowpack. Aims In this study, we sought to test a novel camera trap designed for glaciated mountain ecosystems. We also aimed to use this approach to gain insight into wildlife and human usage of a mountain glacier in western North America. Methods We deployed an elevational transect of uniquely designed camera traps along the western margin of the Paradise Glacier, a rapidly receding mountain glacier on the south side of Mount Rainier, WA, USA. Our simple camera trap design consisted of a wildlife camera attached to a camouflaged cylindrical cooler filled with snow and rocks. Key results Our camera design proved ideal for a mountain glacier ecosystem and from June to September 2021, we detected at least 16 vertebrate species (seven birds, nine mammals) over 770 trap nights using glacier-associated habitats. Humans, primarily skiers, were the most common species detected, but we also recorded 99 observations of wildlife (birds and mammals). These included three species of conservation concern in Washington: wolverine (Gulo gulo), Cascade red fox (Vulpes vulpes cascadensis), and white-tailed ptarmigan (Lagopus leucura). Conclusions Collectively, our results provide proof-of-concept for a novel camera trap design that is ideal for treeless, perennially snow-covered landscapes and revealed a rich diversity of wildlife using mountain glacier habitat in the Pacific Northwest. We highlight the global need for similar studies to better understand the true scale of biodiversity that will be impacted by glacier recession in mountain ecosystems.


Introduction 63
As climate change proceeds and mountain glaciers are lost, there is a pressing need to 64 understand how the loss of glacier ice will impact habitats and ecosystems (Hotaling et  Globally, 36 species of birds and mammals have been observed using glacier and perennial 81 snow habitat for some portion of their life history (Rosvold 2016). Mammals typically use these 82 habitats for relief from lower elevation conditions during warmer parts of the year, whereas birds 83 commonly forage on arthropods and seeds that have been atmospherically deposited on the ice 84 (Antor 1995;Hotaling et al. 2021;Rosvold 2016). However, closer links between wildlife and 85 glacier habitats have also been described. In western North America, the high abundance of 86 glacier ice worms (Mesenchytraeus solifugus)-which are commonly present at densities >200 87 per m 2 on glaciers during summer-may be a key resource subsidy for nesting birds (e.g., gray-88 crowned rosy finches, Hotaling et al. 2020). On glaciers of the high Andes in South America, the 89 white-winged diuca finch nests directly within crevasses (Hardy et al. 2018). Throughout higher 90 latitudes of the Northern Hemisphere, wolverines cache food in snow and ice, and the 91 persistence of spring snow cover and cache longevity has been hypothesized as a primary 92 reason underpinning their cold, snowy distributions (i.e., the "refrigeration-zone" hypothesis, 93 Inman et al. 2012). 94 95 While links between larger-bodied vertebrates and icy habitats have been made for a range of 96 species, these observations typically stem from focused species-specific studies and likely 97 underestimate the true scale of wildlife use in the cryosphere (the global collection of Earth's 98 frozen waters). Beyond researcher focus on specific species, this underestimation likely also 99 stems from two other, intertwined challenges. Mountain ecosystems, and particularly glaciers, 100 are poorly studied relative to more convenient habitats (e.g., those in close proximity to 101 universities and museums, Freitag et al. 1998). Moreover, mountain ecosystems have additional 102 challenges associated with their study. Indeed, for mountain research, typical field methods are 103 often difficult to implement due to, for example, the remoteness, rugged terrain, and harsh 104 weather conditions (and high rates of snowfall) associated with these habitats. For instance, 105 most camera trapping studies affix cameras to trees or T-Post stakes driven into soil (Rich et al. 106 2019; Rovero et al. 2010). In mountain glacier ecosystems, trees are typically absent, soil is 107 usually covered by deep snowpack or non-existent due to glacier activity scouring the 108 landscape down to bedrock, and affixing cameras to boulders limits locations for deployment 109 and timing (e.g., early season deployments might not be possible due to boulders still being 110 covered by seasonal snow). Thus, an easy to transport, flexible, but robust camera trap design 111 to ameliorate the many challenges of high mountain ecosystems is needed. 112

113
In this study, we sought to gain a more holistic understanding of human and wildlife usage of 114 alpine glacier habitat while also testing a simple camera trap designed to alleviate the difficulties 115 associated with high-mountain ecosystems. We deployed our camera-trapping array along an 116 elevational transect of a mountain glacier on Mount Rainier, Washington, USA. While we 117 expected our camera trap array to detect many species known to occur in these ecosystems 118 (e.g., marmot), we also expected to observe species not previously associated with glacier or 119 snow ecosystems. If detected, these results would underscore the likely underestimation of triggers. While our camera traps were largely reliable and stayed in working order throughout 153 the study, we did experience a memory card failure at C5 from 20 August -25 September. Due 154 to the variable terrain and study goals, we made no attempts to control for the amount of area 155 each camera could "see" which likely affected the comparability of our results across cameras. 156

157
To account for the challenges of the alpine glacier ecosystem, including the need to move the 158 cameras throughout the melt season, we employed a unique camera trap design (Figures 2a,b). 159 Cameras were mounted to 5-gallon (18.9 liter) cylindrical Igloo © coolers that were covered in 160 camouflage duct tape with small, camouflage ratchet straps. The coolers were filled with snow 161 and loose rocks from the trapping area and sealed with ⅛ " coated steel cable and a small U-162 clamp. Because we expected visitors to notice our cameras despite efforts to blend them into 163 the landscape, we added a small label to the top of each trap noting its purpose with contact 164 information. The low-profile but robust footprint of our camera traps when the coolers were filled 165 with snow/rock provided a stable platform that remained in place despite high winds and an 166 unprecedented heat wave that occurred during the course of our study and rapidly melted a 167 large amount of snowpack (Pelto et al. 2022). 168 169

Animal identification and data analysis 170
Images were visually inspected for any evidence of vertebrate detection (see example images in 171 Figure 3). When humans or wildlife (birds and mammals) were detected, they were identified to 172 the lowest taxonomic level possible by consulting guides for Mount Rainier National Park and 173 the Pacific Northwest as well as our own research team's expertise in mammals (A.A. and L.W., 174 specifically) and birds (J.B., N.A.P., and P.W., specifically).Given the proximity of our cameras 175 to glacier ice or perennial snow, we considered any species recorded in this study to have a link 176 to glacial habitats. This definition is slightly expanded relative to Rosvold (2016), the most 177 similar study to ours, which focused exclusively on birds and mammals that were directly on 178 snow and ice. 179 180 When multiple images from a camera burst included the same organism(s), it was only counted 181 once. The same is true for multiple bursts within minutes of each other. However, because our 182 camera array was arranged along an elevational transect, it is possible, and perhaps likely, that 183 some humans or wildlife were detected multiple times across different cameras. We did not 184 attempt to control for multiple identifications in space nor time beyond the methods described 185 above. Photos of the research team were also excluded from all analyses. Observations were 186 totaled and analyzed for each of the 14 weeks of the study. We note that the final "week" (week 187 14) contained two extra days due to where the study endpoint fell (25 September) relative to the 188 end of that particular week (23 September). 189

190
To test for a relationship between human and wildlife occurrences in our data set, we U a series 191 of statistical analyses in R v3.6.3 (R Core Team 2021). First, we grouped all human and wildlife 192 observations separately for each camera on a week-by-week basis. Then, we tested for 193 normality of the human and wildlife data sets using a Shapiro-Wilk normality test ("shapiro.test"). We identified at least 16 vertebrate species (seven birds, nine mammals) using glacier-202 associated habitats (Table 1). Aside from a single camera failure that was unrelated to the trap 203 itself, our low-cost, alpine camera trap design performed well. Across our data set, humans 204 were the most frequently observed taxon (N = 208), followed by American pipits (N = 18), 205 mountain goats (N = 13), hoary marmot (N = 13), and wolverines (N = 8; Table 1, Figure 4). Just 206 under one-fifth of birds could not be identified to species but for those that were identified, 207 American pipits and white-tailed ptarmigan accounted for more than half of the observations. 208 However, a number of nondescript brown and gray "blurs" were also recorded, of which many 209 were likely small birds (i.e. gray-crowned rosy finch, horned lark), and thus our results for birds 210 are likely underestimations. 211

212
We observed the most humans at the third-highest camera, C3 (N = 59), and the most wildlife at 213 the lowest camera, C6 (N = 36; Figure 4). Interestingly, C6 also recorded the fewest humans (N 214 = 8) and the most taxonomically diverse wildlife: five mammals (golden-mantled ground squirrel, 215 hoary marmot, human, mountain goat, wolverine), three identifiable birds (American pipit, gray-216 crowned rosy finch, white-tailed ptarmigan), and several unidentifiable passerines and other 217 birds. Because we did not control for the amount of area each camera could "see," comparisons 218 across cameras (and detection periods since cameras were adjusted throughout the study), 219 should be interpreted with caution. 220

221
We observed the highest number of detections during the first week of the study (18-24 June 222 2021) but this pattern was overwhelmingly driven by humans (Figure 4). Indeed, ~54% of all 223 human observations were within this period. For wildlife (mammals and birds), no clear temporal 224 pattern was present. At cameras where humans and wildlife were regularly observed (e.g., C4, 225 C5), we observed a temporal separation with humans being seen early in the season from mid-226 June to mid-July and other taxa appearing later (after mid-July; Figure 4). However, we did not 227 observe a correlation-positive or negative-between the presence of humans and wildlife 228 across all cameras and weeks of the study (P, Spearman's rho = 0.106). 229 230 During the study, we identified three taxa of conservation concern in Washington state: 231 wolverines (including two kits playing on the glacier), a Cascade red fox, and white-tailed 232 ptarmigan (Figures 3e-g). We observed wolverines at the three lowest cameras (C4-C6) in late 233 July/early August, a Cascade red fox at our C4 camera in mid-August, and white-tailed 234 ptarmigans at the two highest cameras (C1-C2) and the lowest camera (C6), including a pair of 235 ptarmigans together in mid-July at C2. In this study, we deployed an array of novel camera traps along the margin of a rapidly receding 251 mountain glacier to test our camera design and gain a better understanding of human and 252 wildlife use of these habitats. We found considerable success with our approach and identified 253 an array of birds and mammals, from songbirds to mesocarnivores, highlighting the ecological 254 complexity and potential importance of glaciers and alpine ecosystems to mountain biodiversity. 255 Indeed, many species observed in this study (e.g., American kestrel) have not been previously 256 linked to glacier habitats, raising questions about the degree to which components of their life 257 history have been overlooked. We also found limited, albeit non-significant, evidence for 258 interactions between human and wildlife usage of glacier ecosystems with little to no overlap 259 between the two groups in space and time. successful foraging in the area as wolverines are known to use rendezvous sites near food 274 caches, particularly when kits are young (Inman et al. 2012). Notably, we only observed 275 mesocarnivores (e.g., wolverines) where we also observed common prey species (e.g., 276 marmots; Table 1). 277 278 While our camera trap design and array in the Mount Rainier alpine was successful, it was not 279 without difficulties. Indeed, the challenging nature of camera trapping in a high-alpine landscape 280 should not be overlooked. In addition to highly variable, and often difficult, weather conditions, 281 the landscape does not lend itself well to traditional camera trapping approaches for three 282 reasons. (1) The physical landscape dramatically changes during summer due to snowmelt. 283 During 2021, seasonal snow depth at a substantially lower elevation than our study area peaked 284 in late winter at 5.36 m (213 inches; Paradise SNOTEL #679). This seasonal snow completely 285 melted by mid-July, effectively lowering the habitat surface by several meters during the first 286 month of our study (mid-June to mid-July). Thus, camera trap locations were limited either to 287 areas that were snow-free early in the melt season or required revisiting as the snow surface 288 changed. In our case, portability was important as we moved cameras small distances 289 throughout the season to track shifting ice margins. Changing snow depth also meant that 290 framing the view, and controlling for the field of vision across cameras, was nearly impossible. 291 For instance, one view that was completely snow-covered in early July was entirely rock and 292 talus within a matter of days. The end of the season offers another complication as seasonal 293 snow can accumulate rapidly, particularly in "sheltered" areas that are ideal for obscuring 294 camera traps. Thus, we recommend that similar future studies in the PNW or comparable areas 295 conclude by the end of September or that researchers devise a strategy for finding snow-296 covered cameras and/or keeping them free of snow. (2) Hiking and/or game trails are minimal or 297 non-existent above treeline. Thus, an exceptionally large and consistent survey area exists with 298 little to no mechanism for targeting areas where wildlife may be more abundant. To this end, a 299 denser camera array would have certainly yielded better quantifications of vertebrate presence 300 and interactions (e.g., links between carnivores and prey or humans and wildlife). Finally, (3) we 301 did not observe many songbirds on the glacier and in surrounding habitats despite well-known 302 connections between the two (e.g., Hotaling et al. 2020). This was likely due to a combination of 303 our cameras not being sensitive enough to detect small birds and images with potential birds in-304 flight being too blurry for identification. Indeed, a number of photographs included unidentifiable 305 brown or gray blurs that we could not confidently categorize as a bird or something else (e.g., a 306 flying insect). One way to overcome this, or at least improve the detection rate of small birds, 307 would be to pair cameras with acoustic recorders at each site so song could be used in 308 combination with (or in lieu of) imagery. 309 310

Conclusion 311
In this study, we showed that a simple, cost-effective camera trap design could reveal a rich 312 community of birds and mammals, including species of conservation concern, are using difficult- 0016. We thank Ben Lee for assistance in the field and park staff for their support during the 327 permitting process. We also thank David Sousa and the co-owners of Tatoosh    to unique sightings and the total observations for a given taxonomic group are given at the top 560 of the plot (N ALL ). The viewer should note the substantial scale difference of the y-axis for human 561