Introduction
Phenology, the seasonal timing of life history events, is increasingly relevant in the framework of global change studies (Cohen et al., 2018; Staudinger et al., 2019; Horton et al., 2020). In general, species are predicted to exhibit phenological shifts across wide geographical scales in response to climate change. Phenological responses are predicted to be particularly important for high elevation communities (Hodkinson, 2005; Stewart et al., 2019), wildlife at northern latitudes (Berger et al., 2018) and cold-adapted species (Sarmento et al., 2019) prone to seasonal mismatches, for example in coat colour (e.g., white snowshoe hares on a brown background; Mills et al., 2013, Zimova et al., 2014, Pedersen et al., 2017; Zimova et al., 2018) or arrival at calving grounds (e.g., caribou arriving after spring vegetation flush; Post & Forchhammer, 2008).
Many mammals of temperate zones experience high seasonal variance in exposure to ambient temperature; pelage that offers heat retention in winter is shed in summer, but for most species little is known about rates of shedding or the extent it varies by sex or across broad latitudinal or altitudinal gradients. Among these species are mountain goats (Oreamnos americanus ), inhabitants of mountainous terrain in northwestern North America (White et al., 2018). Mountain goats use high, windy slopes above the treeline during summer (Chadwick, 2002) and make use of snow patches for cooling (Sarmento et al., 2019). If they did not shed their thick, two layered winter coats, which can grow over ten centimeters long (Foresman, 2012), these would likely pose a liability in summer. Across the late Holocene, their ranges diminished in accordance with warming temperatures from northern Mexico to Idaho (Festa-Bianchet & Cote, 2008).
Much remains to be learned about the triggers of shedding and differences between the sexes in mammals, despite the noted high visibility of massive chunks of hair hanging from species like bison (Bison bison ) and muskoxen (Ovibos moschatus ) (Wilkinson, 1974; Berger and Cunningham, 1994). Most mammals moult to replace worn out hairs and provide different summer and winter coats (Ryder, 1965), including moose (Alces alces ), bison, elk (Cervus canadensis ), and thinhorn sheep (Ovis dalli ). While moult is a relatively understudied life history event (Beltran et al., 2018), it is well-established that photoperiod and, to a lesser extent temperature, control moult onset (Murray, 1965; Lincoln and Ebling, 1985; Mo et al., 2006; Zimova et al., 2018). Moult is known to occur latest in lactating mountain goat females (Déry et al., 2019) likely because, as documented in red deer (Cervus elaphus ), the costs of milk production affect female body condition even though food may be most abundant in summer (Clutton-Brock et al., 1982).
Mountain goats offer an unusual opportunity to examine ecological predictions about moult because they occur along latitudinal gradients that vary in ambient condition including elevation. Because of their stature as an iconic mammal of the mountains, there is likely to be high interest from community scientists. Further, they occur in captive settings including at the far northern extent of their range providing opportunity for individual-level and repeated observation and comparison with southern, better-photographed populations.
Community generated datasets offer a promising way to test hypotheses about moult across broad geographical ranges in part because some historical data are available and because volunteer monitoring is growing in popularity (Taylor et al., 2019). This community-based approach is intended to complement rather than replace long-term research. For instance, community science data have been combined with satellite data to examine how bird migration (arrival time at breeding grounds) responds to advancing vegetation green-up dates (Mayor et al., 2017). Community science data have also been explored across multiple projects to assess climate change effects on American pika (Ochotona princeps ) with reasonably reliable results (Moyer-Horner et al., 2012). In addition to mammals and birds, community-contributed photographs have been used to document glacial retreat, and show promise to shift climate change conversations and enhance public education and engagement (Mullen et al., 2013).
Here, we assess the utility of community photography, un underutilized resource, to detect phenological patterns of moult in mountain goats within and between a geographical gradient and between the sexes. We employed a comparative study design involving detailed fieldwork of our own that concentrated on captive and adjacent wild populations, with decades of community scientist photos of moulting goats along gradients of latitude and elevation (Figures 1-2). If the community science data are commensurate with those from our own focal research, then it seems plausible to explore long-term phenological patterns of moult across the entire geographical range of a species using photographs. Collation of a more complete photographic data base, and then using it to quantify long-term and spatially extant patterns of shedding, provides a barometer for detecting patterns and developing inferences regarding impacts of climate change (Hetem et al., 2014; Vieira et al., 2017). We might also expect greater community engagement with climate science outcomes when communities are actively in involved in data collection.
However, there are issues to consider when relying on community sourcing of data that may be influenced by variation in identification skills, sampling effort and efficiency (Dickinson et al., 2010). Large data sets and applying appropriate statistical models may help account for potential bias (e.g., including observer error in assigning sex to an animal and variation in sampling effort). Data are also often missing from community science datasets (e.g., sampling some locations but not others). The best way to deal with missing data often associated with community science projects has received relatively little attention, and the most common approach is to simply filter out such data (Dickinson et al., 2010), despite it being well known that non-random filtering of data can lead to bias and poor inference (Nakagawa, 2015). However, we might expect that patterns in observed data provide information about likely values of missing data (i.e., information is shared across observations) (see Nakagawa, 2015).
Missing data is an issue for our study because animal sex is not always clearly distinguishable in community sourced photographs nor is whether or not a female is associated with a kid. Here, we develop a statistical model of coat shedding that infers the most likely state of an animal when unknown, and so, even the photographs with incomplete information are still included in the model fit. This general approach could be applied to other community science data sets, reducing the need to remove seemingly uninformative data. Using this approach, we show how community science data can help identify important environmental predictors of shedding rates (e.g., elevation and latitude), the role of the state of the animal (e.g., sex, whether caring for young), and quantify the extent of geographical variation.