Critical summer foraging tradeoffs in a subarctic ungulate

Abstract Summer diets are crucial for large herbivores in the subarctic and are affected by weather, harassment from insects and a variety of environmental changes linked to climate. Yet, understanding foraging behavior and diet of large herbivores is challenging in the subarctic because of their remote ranges. We used GPS video‐camera collars to observe behaviors and summer diets of the migratory Fortymile Caribou Herd (Rangifer tarandus granti) across Alaska, USA and the Yukon, Canada. First, we characterized caribou behavior. Second, we tested if videos could be used to quantify changes in the probability of eating events. Third, we estimated summer diets at the finest taxonomic resolution possible through videos. Finally, we compared summer diet estimates from video collars to microhistological analysis of fecal pellets. We classified 18,134 videos from 30 female caribou over two summers (2018 and 2019). Caribou behaviors included eating (mean = 43.5%), ruminating (25.6%), travelling (14.0%), stationary awake (11.3%) and napping (5.1%). Eating was restricted by insect harassment. We classified forage(s) consumed in 5,549 videos where diet composition (monthly) highlighted a strong tradeoff between lichens and shrubs; shrubs dominated diets in June and July when lichen use declined. We identified 63 species, 70 genus and 33 family groups of summer forages from videos. After adjusting for digestibility, monthly estimates of diet composition were strongly correlated at the scale of the forage functional type (i.e., forage groups composed of forbs, graminoids, mosses, shrubs and lichens; r = 0.79, p < .01). Using video collars, we identified (1) a pronounced tradeoff in summer foraging between lichens and shrubs and (2) the costs of insect harassment on eating. Understanding caribou foraging ecology is needed to plan for their long‐term conservation across the circumpolar north, and video collars can provide a powerful approach across remote regions.


| INTRODUC TI ON
Climate change in the arctic and subarctic (hereafter, arctic) region is unfolding faster than anywhere else on Earth, resulting in alterations of ecosystem structure and function (Box et al., 2019;Hinzman et al., 2005;IPCC, 2014). Vegetation communities are experiencing abrupt and lasting changes resulting from warming temperatures, increased precipitation and more frequent and severe wildfires (Berner et al., 2020;Loranty et al., 2016;Myers-Smith et al., 2011;Walker et al., 2006;Wang et al., 2020). Some plant functional types, like shrubs, are expanding their distribution in response to warming temperatures and increased precipitation (i.e., rain) and outcompeting previously dominant functional groups (lichen; Berner et al., 2018;Myers-Smith et al., 2011).
Changes in vegetation communities are expected to affect ecological carrying capacity through changes to the availability and timing of forage resources (e.g., phenology; Post & Forchhammer, 2008) for herbivores across the circumpolar north (Joly et al., 2012;Post, 2013;Yu et al., 2017). Changing vegetation directly alters the composition, biomass and quality of available forages for large herbivores (Rickbeil et al., 2018;Stark et al., 2021;Zamin et al., 2017).
Previous studies have demonstrated the key role of summer nutrition, especially for arctic ungulates who experience short growing seasons (Barboza et al., 2009;Cook et al., 2004;Shively et al., 2019).
Following the forage maturation hypothesis for large herbivores (Fryxell, 1991;Hebblewhite et al., 2008), caribou transition from a diet dominated by low-quality lichen (winter) to a diet dominated by higher-quality green vegetation (i.e., graminoids and shrubs) to meet the digestible energy and protein requirements for fetal growth (spring) and lactation (summer; Barboza et al., 2018;Crête & Huot, 1993;Denryter et al., 2020). However, caribou experience nutritional deficiencies due to reproductive costs of lactation and inadequate nutrition for energetic demands in many land cover types in boreal forests (Denryter et al., 2018). Further supporting the nutritional deficiency hypothesis, researchers have shown the highest rates of natural adult mortality for caribou in July and August (Cook et al., 2021;Gurarie et al., 2019;McLoughlin et al., 2003). Thus, identifying tradeoffs between foraging for high-quality foods and behaviors that inhibit eating, like those resulting from insect harassment and movement, are key to understanding nutritional implications for caribou during summer.
Observational studies of caribou have shown insect harassment reduces the time caribou spent foraging in summer and increases energy expenditures (e.g., movement) that could result in consequences for body weight and thus, reproduction, calf recruitment and survival (Colman et al., 2003;Toupin et al., 1996;Witter, Johnson, Croft, Gunn, & Gillingham, 2012;Witter, Johnson, Croft, Gunn, & Poirier, 2012). Therefore, climate change has the potential to increase both the benefits of foraging, by increasing the availability of high-quality foods like shrubs, and the costs, through changes to energy budgets from insect harassment. However, measuring foraging ecology of remote caribou in the Arctic remains challenging.
Animal-borne video cameras provide an exciting opportunity to study large herbivore nutritional ecology especially in remote regions. Animal-borne video cameras have improved our understanding of foraging ecology for marine, avian and terrestrial species (Kane & Zamani, 2014;Lavelle et al., 2015;Seminoff et al., 2006).
Large herbivores are unique in that they spend a great deal of their time foraging, upwards of 14 h every day (e.g., Sukumar, 1989).
Animal-borne cameras have recently been applied to large herbivores across remote regions of Mongolia and Canada (Kaczensky et al., 2019;Vuillaume et al., 2021). Previous studies using video collars have measured foraging and diet, grooming and reproduction across cervids (e.g., Lavelle et al., 2015;Thompson et al., 2012;Viejou et al., 2018). One challenge with any new method, such as animal-borne video collars, is the calibration with existing methods, for example, to study diet. Previous studies used a variety of diet methods including behavioral observations in the wild (Fortin et al., 2004;Schaller, 1998), captive and/or tame animals (Shipley et al., 1999), harvested animals (Helle & Tarvainen, 1984), stomach diet analyses (Skoog, 1956) and fecal diet analyses (Russell et al., 1993).
These diverse methods measure diet at different stages in the foraging process, that is, intake rate (behavioral observations of foraging), in vivo (stomach) or following digestion (fecal samples). They also use different metrics, such as percent composition, frequency, number of bites or intake rate in grams/bite (Robbins et al., 1987;Thompson & Barboza, 2014). Thus, comparing diet estimates from different methods is challenging. Many previous methods, including observations and fecal diet sampling, and newer methods like metagenomics are often limited by sample sizes and are costly to implement in remote arctic regions. Animal-borne camera collars can, however, provide finer-scale details of foraging behavior and diet for remote ungulates (e.g., Kaczensky et al., 2019;Thompson et al., 2015;Viejou et al., 2018).
We used animal-borne GPS video-camera collars (hereafter, "video collars") to study behavior and diets of a migratory population of caribou in the subarctic during spring and summer. Caribou are an important cultural, socioeconomic and ecological resource across the circumpolar north (Hummel & Ray, 2008). We focused on adult female caribou during summer because females drive population dynamics (Cook et al., 2021;Roff, 1992). The Fortymile Caribou Herd in central Alaska, USA and Yukon, Canada, is a population that has undergone intensive management for over 50 years (Gronquist et al., 2005;Macdonald et al., 2009). Recent population growth of the Fortymile Caribou Herd (Boertje et al., 2017) has led to questions about deteriorating range conditions and food limitation, for which there is growing evidence for migratory caribou (Bergerud et al., 2008;Crête & Huot, 1993;Schaefer et al., 2016).
Due to this, understanding foraging behaviors and summer diets of caribou remains central for managing migratory populations around the globe (Video 1).
Using videos collected from collars, we first characterized behavioral activities of caribou and quantified insect avoidance behaviors, while considering individual variation among caribou, and tradeoffs between eating and insect avoidance behaviors. To test for individual variation, we also tested for differences in behavioral activities among individual caribou to understand individual-level variability in behavior. Second, we tested if insect avoidance behaviors reduced the time caribou spent eating (Colman et al., 2003).
We predicted the already short summer foraging period would be further restricted by insect harassment. Third, we estimated diet at two levels of taxonomic resolution, the forage functional type (i.e., plants like forbs and shrubs, plus lichen and mushrooms) and the finest taxonomic resolution "species, genera or family" obtained from videos. In the context of the forage maturation hypothesis (Fryxell, 1991), we predicted caribou would switch from a lichenbased diet in late spring to one of higher protein, green vegetation in summer, ostensibly to replenish protein and fat reserves.
We then expected caribou to return to lichen in autumn with the senescence of green vegetation. Finally, we compared diet estimates from video collars to results from fecal pellet microhistology (Dearden et al., 1975) for the Fortymile Caribou Herd, after adjusting for plant digestibility. Addressing our research questions required data classification from video collars, citizen-science volunteer training, data management and coordination with trained botanists specialized in arctic species to classify plants consumed by caribou. We summarize our protocols and data processing steps (Box 1, Appendix A) because of the growing interest in the application of video collars for arctic wildlife.

| Study area
The Fortymile Caribou Herd is a migratory population of caribou spanning a 105,200 km 2 region across east-central Alaska and northcentral Yukon (Canada; Figure 1). The Fortymile Caribou Herd has increased from around 52,000 in 2010 to >84,000 in 2017 (Figure 2; Boertje et al., 2017;Harvest Management Coalition, 2019), spurning concerns regarding deteriorating summer range conditions and nutritional limitation. The bioclimate is characterized by long, cold winters (minimum temperatures = −50°C) and short, warm summers (maximum temperatures = 37°C). Precipitation is light in summer (mean 300-600 mm) and moderate in winter (average 1.5 m as snow), and fires are frequent and widespread (Jorgensen & Meidinger, 2015). Vegetation types include subalpine spruce (Picea spp.) forests, deciduous forests, shrubland and herbaceous tundra (Wang et al., 2020). Treeless herbaceous and tussock alpine tundra dominate landscapes above 800 m that also provide important habitats for calving, post-calving and late summer aggregations that help minimize insect harassment (Boertje et al., 2017).

| Ethics statement
All animal captures were conducted by the Alaska Department of Fish and Game and approved in accordance with animal welfare standards (IACUC permit numbers through ADFG 0002-2018 and 0002-2019).
Caribou were then fitted with a GPS-Iridium collar integrated with a camera and pre-programmed with a drop-off mechanism programmed to release on September 10 each study year (VERTEX Plus Iridium V 3.0, Vectronic Aerospace GmbH, Germany).
Video collars were programmed to record videos during daylight hours (14-18 h/day). For all programming periods from May to September, collars recorded a 9-s video and GPS location every 20 min during daylight hours (Appendix A). Videos were processed using a two-phased approach. First, trained volunteers classified a random subset of videos to classify caribou behavior (see Box 1,in blue; Appendix A). Second, videos classified as "eating" were viewed by five botanists with subarctic classification experience to identify species of forage(s) consumed by caribou (Box 1, in green).

| Caribou behavior
We classified caribou behavior from videos into states of eating, ruminating, travelling, stationary awake, napping and others. We explored differences in behavior between/across (1) individuals, (2) years and (3) months, and contrasted frequencies of videos classified into different behaviors using one-way Chi-square goodness-offit contingency tests (GOF; Sokal & Rohlf, 1995). We used one-way tests as an initial simple analysis step to explore temporal and individual behavioral differences. We could not consider two-or threeway tests (e.g., to account for year/month by individual differences) because we radiocollared different individuals between years. We acknowledge that such one-way tests likely commit type I error but used these as an initial exploratory step to focus subsequent statistical analyses of the main behavioral axis, changes in foraging. We also quantified insect avoidance behaviors observed in videos (e.g., shook head, scratched, sought snow patch, kept muzzle to ground and huddled; Morschel & Klein, 1997;Witter, Johnson, Croft, Gunn, & Gillingham, 2012;Witter, Johnson, Croft, Gunn, & Poirier, 2012; see Appendix A).
To test for the effects of insect harassment on eating in videos, we used generalized mixed-effect models (GLMER, lme4 package in R, R Core Team, 2020) with a binomial (logit) link (Bates et al., 2015).
We tested for the effects of the presence of insect avoidance behaviors (binary) on eating (binary) by female caribou in each video.
Eating and insect avoidance behaviors were treated as events, suitable for analysis of frequencies (Altmann, 1974). We considered a random intercept to test for variation in eating between individuals and, in so doing, treated the individual as the sampling unit for all video-based GLMER analyses. We also tested for a random coefficient for individual caribou and their individual variable responses to insect harassment (random coefficient; Appendix B Table B2).

BOX 1
Flow chart of our data collection process using caribou video collars. We excluded video recordings that malfunctioned were shorter than 8 s and confirmed videos recorded on schedule for the duration of the study for each caribou. Using R, we created folders of randomly selected videos (with an equal number of videos per study animal). To improve efficiency, we classified videos using two phases. In the first phase (in blue), volunteer observers (citizen scientists) viewed videos to identify caribou behaviors and other supplemental information (see Appendix A). This first phase required approximately 2 min of time per observer to classify a one 9-s video from caribou. In the second phase (in green), botanists who were specialized in arctic flora viewed videos classified as eating from the first phase to identify forage items consumed by caribou. Botanists identified forages to the most refined taxonomic level possible with the highest level of confidence. It took each botanist about 4 min of time to classify forages consumed by caribou in a one 9-s video. Volunteer observers and botanists were required to review protocols and complete evaluations using training videos where we then could calibrate responses prior to starting data collection. Observers could also flag ambiguous videos for expert review.
Random subsampling and data quality assurance and control procedures were developed and included for consistency.
Model selection was performed using BIC selection criterion (Brewer et al., 2016).

| Diet composition using video collars
Botanists experienced in arctic plant classification identified forages consumed to the most refined taxonomic level possible while still maintaining a high level of confidence (e.g., Salix spp., Salix pulchra; Box 1). If forage identification was uncertain, then videos were reviewed for a second opinion to confirm forage(s) selected by caribou.
We calculated diet for each taxonomic unit as binary (yes, no) for each video and estimated diet as the percentage of videos classified as "eating" for that taxonomic unit. Diet composition estimated from video cameras is expressed as absolute percentages, as the sum of the percentages from different forage types could exceed 100% (because more than one forage type could be consumed in a one 9-s video).

| Diet composition using microhistological analysis
We collected fecal samples across the summer range of the identifying species with >5% occurrence) or fine (A150; identifying all species occurrences ≥ trace levels) scale because of budget fluctuation. We removed rare forage types (those making up <4.0% of composite sample) and reported the mean diet of major plant classes (genera, species) averaged across each month from 2011 to 2018.
Diet composition estimated from fecal microhistological analysis is expressed as a relative percentage, as the sum of percentages from different forage types sum to 100%.

| Taxonomic resolution
We tested the taxonomic resolution between diet composition estimates from video collars and microhistology. We focused on the seven forage functional types (FFT) that occurred across both video collar and fecal data sets: Equisetum spp., forb, graminoid, lichen, moss, mushroom and shrub. We excluded forage types estimated as unknown or represented broader classes (e.g., ground-cover vegetation).

| Correcting fecal diet samples for digestibility
We measured apparent dry-matter digestibility (DMD in %;Van Soest, 1982) for plants consumed by caribou to correct fecal samples for digestibility to facilitate comparison to video-collar-derived diet estimates. We collected plant samples across the summer range of the Fortymile Caribou Herd from May to September for two summers concurrent with video collar deployment (2018 and 2019; Figure 2). Plant samples were air dried, weighed and stored in paper bags. Samples were dried in a ventilated drying oven at 65°C for 48 h (to a constant weight) and analyzed for detergent fibers (Van Soest, 1982), crude protein and tannin concentrations with bovine serum albumin (BSA; Martin & Martin, 1982) at the Wildlife Habitat and Nutrition Laboratory (Pullman, Washington, USA). We calculated DMD and adjusted for tannin content using Equations (1) and (2) of Hanley et al. (1992). For those forage functional types not assessed for forage quality by our team, we used DMD values estimated for the nearby Denali Caribou Herd (Boertje, 1990).

| Correlation of methods
Because we observed no differences in the frequency of eating between years from our initial Chi-square tests, we lumped all years together. To test for similarities in diet composition estimated from video collar and fecal samples, we first applied the correction factor to our microhistological results to account for digestibility using F I G U R E 1 A female caribou of the Fortymile Caribou Herd (Rangifer tarandus granti) strips and consumes leaves from a Salix pulchra shrub. We classified behavioral and foraging activities for caribou during summer as observed from 9-s videos recorded from GPS video-camera collars across Alaska, USA and Yukon, Canada (2018 and 2019) our values for DMD (see details in Appendix B Table B4). We then compared, for each month, the six FFTs in the diet shared by video collar and fecal estimates; thus, we dropped the FFT for mushrooms because of their absence in microhistological analysis. We included May-August, as fecal samples were not collected in September.
Forages that made up small portions (<1%) of the diet, as estimated by microhistological analysis, were removed. Next, we compared proportions of forage functional types between methods using Chisquare tests. Finally, because of their large prevalence in the summer diet (see Section 3), we tested for correlations between the proportions of lichen and shrubs estimated by video collars and fecal pellets. good or excellent in 91% of video clips, poor in 8% and extremely obstructed in 1%. In most of the "extremely obstructed" videos, data could reliably be collected; most obstructions (71%) occurred as caribou foraged on ground-level vegetation, neck or jaw fur obstructing the view, or as caribou napped (11%).

| Caribou behavior
Caribou partitioned their behavioral activities into eating (mean = 43.5%), ruminating (25.6%), travelling (14.0%), being stationary awake (11.3%), napping (5.1%) and others (0.5%; e.g., drinking, F I G U R E 2 Study area for female caribou of the Fortymile Caribou Herd (Rangifer tarandus granti) across central interior Alaska, USA and North-central Yukon, Canada. Caribou were outfitted with animal-borne GPS video-camera collars (n = 30) over two summers (2018 and 2019). Citizen scientist volunteers classified videos into categories based on caribou behavior (n = 18,134 videos). Circles represent the spatial distribution of all classified video locations for caribou, and colors highlight behaviors classified as eating (green; n = 5,549) and not eating (purple; ruminating, travelling, stationary awake, napping or others) licking soil for minerals and wading; Figure 3a). Summer behavioral activities for caribou did not differ between years (χ 2 = 7.55, df = 5, p = .18); therefore, we lumped data between years. Behavior did vary across months (χ 2 = 512.9, df = 20, p < .001) and individual females (χ 2 = 444.2, df = 145, p < .001; Figure 3b). We acknowledge the lack of independence of individual caribou in the Chi-square GOF tests casts doubt on the strength of the p-values.
Nevertheless, they helped confirm that the main state driving changes in behavioral activity of caribou seemed to be the reduction in eating in July and not differences between individuals or years (Table 2, Figure 3). Subsequently, we thus focused on exploring foraging.

F I G U R E 3
The proportion of videos (%) where caribou were observed (a) in different behavioral activities and (b) eating for each individual caribou throughout the summer season. We monitored female caribou (n = 30) of the Fortymile Caribou Herd (Rangifer tarandus granti), Alaska, USA and the Yukon, Canada during summer daylight hours, May-September 2018-2019 Insect avoidance behaviors increased through July and were associated with reductions in the frequency of eating (Figure 4; Appendix B Figure B1). Our most parsimonious model (Table 1) showed a strong negative effect of insect harassment on the probability of eating for caribou (β = −2.02, p < .001; Table 2). The standard deviation (SD = 0.1) of the random effect suggests responses among individual females did not vary strongly. The second ranked model (Table 1) was the same as the top model without a random effect for individual. These results collectively support our Chi-square analyses above showing minimal individual-level variation in behavior and eating ( Figure 3b), and the consistency in the tradeoff between insect avoidance behaviors and eating. These conclusions are also supported by the tradeoff at weekly eating scales (see Appendix B Figure B1).

F I G U R E 4
The relationship between the probability of eating and insect avoidance behaviors observed within 9-s videos for female caribou of the Fortymile Caribou Herd (n = 30; Rangifer tarandus granti), Alaska USA and Yukon, Canada, 2018 and 2019. As the probability of insect avoidance behaviors increased, the probability of eating by caribou decreased. The probability caribou reduced eating while displaying insect avoidance behaviors varied across months TA B L E 1 The five most parsimonious models, based on ∆BIC values, from a set of candidate binomial generalized linear models of the effects of insect harassment on the frequency of foraging events observed in videos throughout the summer months for caribou of the Fortymile Caribou Herd ( Figure B7).

| Taxonomic resolution
We identified 63 species in 70 genera in 33 families of summer forages consumed by caribou using video collars (Appendix B Figure B9).
Microhistological analysis identified plants to 12 species in 24 genera in six families using plant fragments found in fecal pellet samples.

| Correcting fecal diet samples for digestibility
We measured apparent dry matter digestibility (% DMD) for 167 plant samples across four forage functional types: shrubs (58.2%, n = 85), lichen (75.1%, n = 37), graminoids (72.9%, n = 37) and forbs (77.2%, n = 8; Appendix B Table B4). The concentration of tannins (mg BSA/ mg forage) was calculated for 118 caribou forage samples. We then adjusted DMD for tannin precipitate, as tannins cause reductions in forage digestibility for ruminants. We considered Equisetum spp. highly digestible and used our DMD value for forbs (77.2%; sensu Boertje, 1990). For mosses, we used DMD values determined by Boertje (1990; 7%), as mosses have been shown to have poor digestibility (Ihl & Barboza, 2005). Our DMD values were highly correlated to Boertje's (1990), which allowed us to use their values with accuracy when needed (Appendix B Figure B8). Our shrub samples included some woody stems and therefore likely underestimated shrub digestibility and the resulting proportion of shrub in the corrected diet estimates.

| Correlation of methods
We found a positive correlation between the proportions of forage functional types estimated across months (r = 0.79, p < .01; Appendix B Figure B10) from video collar and digestibility-adjusted microhistological methods (Figure 7). The relationship between summer diet estimates was marginally statistically significant (r = 0.79, p = .06).
Diet estimates for monthly lichen (r = 0.81, p = .18) were not correlated between the video collar and microhistological methods; however, estimates for monthly shrub (r = 0.93, p = .07) were marginally statistically significant.

| DISCUSS ION
Animal-borne video collars provided a powerful new tool to remotely assess behavioral and foraging patterns for large herbivores TA B L E 2 Coefficient table from the most parsimonious logistic regression model explaining the probabilities of caribou (Rangifer tarandus granti) eating that included fixed effects for insect avoidance behaviors and month and a random effect for individual caribou of the Fortymile Caribou Herd, Alaska, USA and Yukon, Canada, 2018 and 2019 Note: Included are the model predictions for the amounts of instantaneous (in 9-s videos) probabilities for females eating (%) with and without insect avoidance behaviors. Also included are comparisons to the frequencies of eating and insect avoidance behaviors (%) from counts of the raw video footage averaged over the month.
across remote regions. This tool allowed us to identify behavioral and nutritional tradeoffs that were previously difficult to detect with field observations and/or fecal plant fragment analysis. Behavioral activities for caribou varied strongly across the summer and were strongly driven by insect avoidance behaviors. Using video collars, we identified (1) higher dietary diversity by discerning forage types at finer taxonomic levels than fecal sampling and (2) a strong temporal tradeoff in the consumption of lichen and shrubs. Our work demonstrates video collars are useful, especially in remote regions like the arctic, to document behavior and diet.
We found managing and classifying videos took significant amounts of effort (Mattern et al., 2018). Recruiting and retaining volunteers were time intensive, and only 30% expressing interest completed the training to become observers. We incentivized student engagement with undergraduate independent research credits.
Training volunteers, using data entry forms and evaluation processes, provided consistency in data collection. Out of 91 volunteer observers that completed training and collected data, few (n = 14) classified >300 videos. Similar to Thompson et al. (2015), hiring arctic plant experts to classify foraging videos provided the necessary tion as an automated process, it will be difficult to automate accurate diet classification from videos, and researchers should be prepared to allocate resources to processing diet data.

| Caribou behavior
Our work demonstrates video collars can quantify behavioral activities across a variety of temporal scales: daily (e.g., Appendix B Figure B1), weekly, monthly, seasonally and yearly. Caribou spent an average of 45% of daylight hours eating in summer (Table 2).
Our results are also consistent with the foraging ecology of large herbivores in summer. Because summer forages are more digestible, ungulates reduce gut retention and rumination time, and increase intake rates (Barboza et al., 2009;Van Soest, 1982). As a result, passage rates become the limiting factor in ungulate nutrition during summer. Caribou spent just 25% of their time ruminating in summer, similar to previous summer studies (Maier & White, 1998;Russell et al., 1993), but much lower than winter when rumination accounts 40%-50% of the activity budget (Russell et al., 1993). Video collars also documented the evident tradeoff between eating and other behaviors, like insect avoidance and movement, foundational to mechanistic ungulate foraging models (e.g., Hobbs et al., 2003;Spalinger & Hobbs, 1992).

| Foraging behavior and insect harassment
Our results show interior populations of migratory caribou reduce eating when exposed to insect harassment as predicted and based on other studies. Reductions in the probability of eating by caribou correlated strongly with increased probability of insect avoidance behaviors ( Figure 4) and temperatures in July and were not correlated with the increase in shrub consumption (Appendix B Figure   B2). Caribou reduced their frequency of eating from 48% in May to 34.5% in July ( Figure 3, Table 2). These reductions in eating are similar to observations of coastal populations of migratory caribou.
Caribou summering on the coastal plains of Alaska and the Yukon (Russell et al., 1993), as well as in alpine tundra (Morschel & Klein, 1997 (2012) and Toupin et al. (1996) found caribou fed only 30%-38% of the time in the presence of oestrid (e.g., bot fly) insect harassment. Similarly in Norway, semi-domesticated migratory reindeer reduced their feeding to 23% under insect harassment (Colman et al., 2003). Although fewer studies have quantified foraging reductions for interior populations in Alaska (Boertje, 1985;Maier & White, 1998;Morschel & Klein, 1997), our work shows that Diet composition was estimated as the mean proportion for the six forage functional types found in both methods for individual caribou (sampling unit for video collars = "video collars") and composite fecal sample (sampling unit for microhistological analysis = "fecals"). Diet composition estimates from video collars are expressed as absolute percentages (purple circles), and estimates from microhistological analysis are expressed as relative percentages (green circles) Witter, Johnson, Croft, Gunn, & Poirier, 2012). The avoidance behaviors we frequently observed (e.g., muzzle to the ground, head shaking, stomping and scratching), however, suggest harassment by oestrids (Oestridae) and tabanids (Tabanidae). In addition, caribou collar temperature (an indicator of oestrid insect activity; Appendix B Figure B2) had a strong negative correlation with the frequency of eating. As temperatures rise due to climate change, insect activity is predicted to increase across the arctic (Koltz & Culler, 2021;Mörschel, 1999;Witter, Johnson, Croft, Gunn, & Gillingham, 2012;Witter, Johnson, Croft, Gunn, & Poirier, 2012), potentially further reducing summer foraging (Appendix B Figure B2).
For example, the Western Arctic Caribou Herd moved nearly twice as much during insect harassment periods .
These increased movements can decrease foraging opportunities.
Instead, caribou in mountainous areas travel from nutrient-dense lower-elevation habitats to high-elevation, nutrient-poor vegetation communities in alpine to seek relief from insects on wind-blown ridgelines (Appendix B; Figure B3; Russell et al., 1993;Anderson et al., 2001).
The joint effects of reduced foraging and increased movement can lead to high energetic costs. Caribou may be unable to compensate or replenish energy reserves lost from reduced foraging (Colman et al., 2003) especially during summer, the critical time female ungulates improve body condition for lactation and year round nutrition (Cook et al., 2004(Cook et al., , 2021White et al., 2013). We studied the effects of insect harassment on females, but juveniles experience immediate and more severe consequences than adult females from increased stress, low weight gain and, in rare cases, death (Helle & Tarvainen, 1984;Weladji et al., 2003). In the future, researchers could pair accelerometers with foraging and insect data from videos to calculate the true energetic costs of extra movement across age and sex classes (Williams et al., 2014). Our estimates of tradeoffs between eating and insect avoidance behaviors could be also used in energetics models (e.g., White et al., 2014) to understand consequences of changes in insect harassment to populations.
There are several caveats to consider in analyzing complex behavioral responses across time, space and individuals. First, we acknowledge behavior is obviously an explicitly multivariate process, and our bivariate analyses of tradeoffs between insect avoidance behaviors and eating likely overlooked this multivariate process.
However, we used random effects for each individual female caribou, with new individuals radiocollared each study year, to account for individual heterogeneity in foraging behavior (Gillies et al., 2006). Thus, we choose to account for the sample unit of individual animals in the GLMM with a random effect for individual instead. This demonstrated weak individual-level variation, for example, a key finding of our study. It is also important to acknowledge the temporal sampling scale of our behavioral activity within 9-s videos, a near-instantaneous foraging scale (e.g., on average, we classified 4.8 videos/day/caribou for behaviors and 1.5 videos/ day/caribou for identifying foraging items). This instantaneous scale likely overestimated the tradeoff between eating and insect avoidance behaviors at daily or longer foraging scales, following theory on upscaling foraging of ungulates (Fryxell, 1991;Spalinger & Hobbs, 1992). For example, in Table 2, the probability of eating while also being harassed by insects was 17.4% in July in 9-s videos. But, averaged over 1 month, insects reduced the frequency of eating by 10.5% (Table 2, Appendix B Figure B1). However, the tradeoff between eating and insect avoidance behaviors was evidenced not only within 9-s videos but also when looking at means across all temporal scales. And our estimates from instantaneous scales were similar to previous studies that demonstrated reductions in foraging activity from direct observations (e.g., Witter, Johnson, Croft, Gunn, & Gillingham, 2012;Witter, Johnson, Croft, Gunn, & Poirier, 2012).
Throughout the boreal forest, caribou and elk show similar responses to insects (Gates & Hudson, 1981;Raponi et al., 2018). Insect harassment is critical not only for caribou summering along the arctic coasts but also for interior subarctic populations in alpine tundra, as our results show, and for large herbivores around the world. Many components of herbivore ecology and evolution are driven by insect harassment, so much so that zebra (Equus burchelli or E. quagga) evolved stripes to confuse and prevent flies from landing and probing for blood (Caro et al., 2019). Global changes in environmental conditions may alter the distribution and abundance of parasitic insects in ways that reduce nutritional condition of large herbivores, especially in arctic regions . Future studies could similarly use video collars to investigate insect-herbivore ecology.

| Summer diets
We found video collars provided greater taxonomic resolution of diet that correlated with traditional methods (Lavelle et al., 2015;Newmaster et al., 2013;Parrish et al., 2005). We identified >60 species from videos but only 12 species from fecal samples (Appendix B Figure B9). Some taxonomic groups were difficult to identify from cameras, like those we lumped into the "ground-level vegetation" category. But it remained challenging to discern forages at levels finer than the forage functional type or genera level using microhistological analysis. Furthermore, the finer the taxonomic level, the greater the discrepancy between diet from video collars and microhistological analysis (Appendix B Figure B9). Newmaster et al. (2013) and Thompson et al. (2015) first used video collars to document seasonal diets of woodland caribou, noting some of these same discrepancies but did not account for digestibility when comparing fecal results to videos. Accordingly, Newmaster et al. (2013) found summer diets estimated from fecal samples to be <15% cor- Despite the methodological challenges, the strong tradeoff we observed with videos between shrubs and lichen has important implications for caribou nutritional ecology. Caribou clearly eat shrubs in summer to accumulate fat, because of their relatively high digestibility properties and nitrogen content (Boertje, 1984;Murie, 1935;Skoog, 1956;White et al., 2013). The diet estimates we obtained from video collars support our predictions and match nearly a century of a broad array of different types of studies from Alaska and Canada (Boertje, 1990;Murie, 1935;Russell et al., 1993;Skoog, 1956;Thompson & McCourt, 1981) that documented tradeoffs between shrubs and lichens between seasons and, in our study, within summer. Forbs accounted for small portions of the diet but increased gradually as the growing season advanced.
The tradeoff observed from lichen to shrubs occurred when shrubs green up in early summer (June; Figure 5). However, the decline in shrub consumption we observed in July may arise because of insect-induced shifts in resource selection where caribou select higher elevations, forcing animals to suboptimal habitats where shrub biomass is reduced (Russell et al., 1993;Appendix B Figures B1 and B3). In the future, we can assess how spatial covariates affect diet estimated from video collars; something we have never been able to do with fecal samples. Combined with the evident bias against shrubs in microhistological samples, which are critical for summer protein replenishment (White et al., 2013), we conclude that video collars provide researchers a powerful tool to study changes in caribou diet over time and at fine spatial scales.

| Significance and conclusions
High abundance and declining indices of nutritional condition (Boertje et al., 2012)  to plan for their long-term conservation across the circumpolar north, given the accelerated effects of climate change in these regions and the uncertain future of many caribou herds.

CO N FLI C T O F I NTE R E S T
The authors have no conflicts of interests to declare.    • Phase 2 focuses on caribou diet and requires observers able to identify the Alaska/Yukon flora to the genus taxonomic level. The data collected will then be combined with GPS locations from the video collars and results from field-based sampling of forage quality and biomass.

Learn this protocol and evaluate your proficiency (2-3 h in total)
3. Questions/feedback as needed

Requirements
• Computer; speakers/headphones can help detect foraging activity.
• Good internet connection and mainstream web browser.
• Split screen: for consistency and efficiency, view the clips on one side of the screen (in one browser window) and the data entry form on the other side (in a separate browser window). Regardless of the device used, the "video window" should be equivalent to at least half the size of a typical laptop screen. The "form window" can be made narrower without losing functionality. Please contact the project manager for any help.

Viewing video clips
Each qualifying observer is assigned a folder containing a unique set of 90 video clips (six random clips from each of the 15 study animals).

Clips are viewed online via a custom link to UM Box (University of
Montana's cloud-based storage). You may need to view each clip more than once to focus on the different types of data to be collected. To navigate to the next clip, hover the cursor over the image and click on the arrow. Video file names contain the animal ID, date, and time: "ID#_YYYYMMDD_HHMMSS".
If you have completed your folder and are still available, please contact the project manager and a new folder will be assigned to you. If you are unable to complete your folder, please notify the project manager and the remaining clips will be reassigned.

Entering data
Data are entered in Google Forms online. A link to the live form will be provided along with your assigned folder. In the meantime, please follow the link below to preview the form while learning the protocol.

• Form Preview: tinyu rl.com/y3y9avap
Use the "NEXT" and "BACK" buttons at the bottom to navigate across the three sections of the form (please avoid using your browser buttons). Upon submitting a form, you may choose to edit your response, fill another form or close the window until your next data entry session. Please keep track of your progress in order to prevent duplicates or missed entries. If you lose track of your progress, contact the project manager and you will be pointed in the right direction.
The data collected pertain to the individual caribou wearing the video collar. Please refer to the video examples and field descriptions below. If uncertainty remains, enter your best response and then flag the form for review in the last section of the form. You may additionally contact the project manager for a quicker response.
If you realize along the way that you have been misinterpreting a question or have not entered the best possible response in previous submissions, let us know and steps will be taken to edit those responses. Please keep in mind that some video clips are ambiguous and the observer's best assessment is usually sufficient! However, for reoccurring uncertainties regarding foraging activity or calf identification, please contact us for further guidance.
• Video Examples: tinyu rl.com/yc9r67zz More video examples will be added as we go, so please refer to this folder often through the data collection process. Video file names in the examples indicate the correct assessment for each type of data collected.

Observer Name
For quick navigation through the list: click "Choose", then scroll down or type the first letter of your first name (keep pressing that same letter to navigate to your name) and press enter.

File name
This is the most important entry of the form.
1. Locate the file name (top-left of the video window), select it by double-clicking (no need to include the file extension, but it can also be included), then press ctrl-C (Mac: command-C).

Paste into the form: ctrl-V (command-V).
3. Please ensure that the file name has copied correctly.

Video quality
This is a quick, somewhat arbitrary assessment. See video examples linked on page 2. Camera lens obstruction may consist of long fur, condensation, water drops, dirt etc.
• EXCELLENT -excellent image • FAIR to GOOD -most clips fall in this category; allows easy observation, partial to no camera lens obstruction • POOR -some data can be collected but the image is problematic (e.g., significant lens obstruction, low light, problematic camera angle and blurry image) • EXTREMELY OBSTRUCTED -the image is obstructed the entire time (often by the chin or fur while eating) and a botanist would not be able to identify any of the vegetation present.

Foraging Status
During summer days, caribou spend almost half of their time eating and a quarter of their time ruminating. Please view the video examples linked on page 2.
• RUMINATING -Caribou are ruminants (like cattle) and spend a lot of time chewing their cud (food that is regurgitated from their first stomach compartment to be chewed a second time). If the caribou is chewing while bedded or resting, it is almost certainly ruminating.
They can also ruminate while walking if they get disturbed. If you see "swollen" cheeks or the bolus going up the esophagus, the caribou is definitely ruminating. If the cheeks are not bulging, the caribou may nonetheless be ruminating, please view the video examples! • CHEWING -Chewing food but did not take a bite during the video recording (only took a bite before the start of the recording; e.g., chewing while searching for food).
• EATING -"Took a bite" of a food item. Select "eating" even if you cannot identify the food item consumed (when the caribou eats, fur from its neck can obstruct the camera; having the sound on can help identify eating activity).

• DRINKING
• LICKING -Licked the soil/rock for minerals but did not take a bite of a food item.
• None of the above

State of Locomotion
This field may be ambiguous for some clips, and your best assessment is sufficient (do not flag for review or comment). Please select the first applicable option in the list.
• Wading/Swimming • Running • Walking • Stationary Awake: standing or lying, but awake • Napping: head on the ground, minimal movement (breathing, twitching), may see curled up legs/hooves or sideways camera angle

Is a calf visible?
If age determination is not obvious, please flag for review in the last section of the form.
• Yes -her own: calving evidence (placenta/sac, wet neonate) or maternal behavior (suckling, licking/grooming, being near a very young calf or very close to a young calf) • Yes -possibly her own: no maternal behavior detected, but the calf is not with another cow • Yes -calf is with another cow

• No
Calf identification tips: • Look for a smaller body, shorter ears and shorter face with a narrower snout.

• Look at the timestamp (YYYYMMDD) in the video file name. Calves were born around May 19-28, so identification is easier in May-June and becomes progressively more difficult. The example videos can be sorted by date and include non-calves as a comparison.
• Caribou color is highly variable and not reliable for age determination.
• An antlerless caribou is not necessarily a calf.
• Small antlers (spikes) may be visible on calves by late summer.

Other caribou visible (excluding own calf)?
• Yes -herd (about 10 or more caribou) • Yes -one to a few individuals

Does the cow have antlers?
It is sometimes possible to confirm the presence or absence of antlers when the caribou's shadow is visible or to confirm the presence of antlers through a direct glimpse of the top of the caribou's head. There is no need to spend time on assessing this outside the period of May to early June (see date stamp in the file name). From

mid-June to September, you can simply select "Not relevant".
Antler presence in May to early June provides an additional clue that a caribou was pregnant, as pregnant cows drop their antlers later than non-pregnant cows (retaining antlers helps defend food patches later into the season). By fall, all caribou have grown new antlers.

• No
• Can't see/Not sure/Not relevant (most clips fall in this category)

Potential insect harassment behavior (select all that apply)
• Shook its head • Kept its nose still AND on the ground (to prevent parasitic flies from laying eggs in the nostrils) • Scratched (may use mouth or hoof) • Sought snow patch (lying/standing on a snow patch, as opposed to just walking or foraging through snow) • Huddled

What part of the habitat is visible?
• Ground and immediate surroundings (a good glimpse of the habitat is sufficient, as long as the predominant vegetation type around the caribou can be identified)

What is the PREDOMINANT vegetation?
Quick assessment of the main vegetation type present near the caribou. Any category (including poor visibility) may be selected on their own or in concurrence with another. Select only the predominant categories (preferably 1 or 2, but can be up to 3).

Habitat features visible (select all that apply)
• Snow cover 1%-50% (in the vicinity, ignore mountain tops and faraway snow) • Snow cover 50%-100% (in the vicinity, ignore mountain tops and faraway snow) • Water (e.g., river and puddle) • Burn area visible (at any successional stage; e.g., burn scars and sooty snags/logs) • Human signs: any sign of human presence (e.g., human activity, roads, buildings and other structures) • None of the above FORM -Section 3 of 3

Other species detected?
Enter the type of animal detected (e.g., mammal, canine, bird and bird of prey) or finer taxonomic level if known (e.g., wolf and golden eagle).

FLAG for review? "There was uncertainty in my response(s) regarding…"
Some footage may be difficult to interpret, and a second opinion will help determine the best response(s). Please note that the observer's best assessment is usually sufficient without need for review. However, for reoccurring uncertainties, particularly regarding eating or calf identification, please communicate with us for further guidance.

FLAG as favorite? "This clip is an outstanding example of…"
Please select all reasons that apply. More details or categories can be added under "Other".
• Potential predation attempt (rare video capture, please flag!) • Interesting/rare behavior or interaction • Interesting vegetation/habitat feature • Visually appealing video clip (e.g., scenery, herd, calf and habitat) • Sounds (e.g., caribou call and other species). Please do not flag ruminating sounds and sounds of vegetation rubbing against the collar.
• Other: (additional comments can be added here)

Note (Please use very sparingly)
This field may be used to relay pertinent information not otherwise included in the form. Please be concise, use key words and avoid repeating information already entered. Almost always leave this field blank!

Tips for writing notes: It is important to only write a comment in this
section if there is something particularly extraordinary or peculiar and leave it blank otherwise. The bulk of the data needed is already included in the form.

TR A I N I N G
The training procedure is conducted online and through communication with peers or project contacts. We aim to ensure consistency and efficiency among observers, generate high quality data and provide a platform for questions and feedback, which may help improve the data collection process.
Once you have read the field descriptions above and viewed the video examples, please study each pre-filled form below and read the "practice notes" at the end of each form. Questions and feedback are welcome at any time.
• Videos for prefilled forms: tinyu rl.com/y2wsgj6q Once the task has been completed, please notify the project manager to discuss your results and receive your assigned folder. Thank you for your interest in being part of this project!