Challenges and opportunities for bioacoustics in the study of rare species in remote environments

For many habitat specialists, data deficiencies are directly attributable to the rugged terrain they inhabit. Based on a recent survey effort for one such species, the threatened Yosemite toad (Anaxyrus canorus), we describe the prospects for acoustic monitoring of rare and declining species in remote areas. Passive acoustic monitoring can be a valuable way of increasing the scale at which vocally active populations are surveyed and the level of detail obtained, particularly in cases where direct observation of the species is impeded by cryptic behavior and inaccessible habitat. Challenges include limitations to hardware, teams' physical and mental constraints, and survey design. Teams must have substantial expertise about the species of interest, bioacoustics, and the terrain to be surveyed, and effective communication at all stages is essential. Particularly for globally endangered taxa like anurans, bioacoustics may be a powerful way to reduce widespread data deficiencies.


| BIOACOUSTICS AND AN ELUSIVE, DECLINING TOAD
Bioacoustics is widely touted as a means of scaling up population surveys of acoustically active species (e.g., Sugai et al., 2020), but the implementation of passive acoustic monitoring (PAM) at landscape scales can require substantial resources and logistical infrastructure. For example, the success of a recent invasive species removal was enabled by a PAM program implemented across an entire ecosystem (Hofstadter et al., 2022), which, in turn, was made possible in part by a vast road network. Yet many species of conservation concern live in regions that are not amenable to widespread, rapid access to survey locations (Neilson et al., 2013). Steep, rugged mountains tend to promote hotspots of endemism even within regions considered biodiversity hotspots (Noroozi et al., 2018), and also impede survey access to resident populations, causing data deficiencies for some species.
We illustrate the challenges and opportunities for PAM in the context of surveying rare but acoustically active species in remote environments using a case study featuring the Yosemite toad (Anaxyrus canorus), a declining amphibian endemic to the central Sierra Nevada, USA. Briefly, Yosemite toads typically breed at snowmelt in high-elevation meadows, when much of the toad's range is considered "inaccessible" (Brown & Olsen, 2013) because human surveys would require traveling on foot through snowy, mountainous terrain. Therefore, the USDA Forest Service's current Yosemite toad monitoring program predominantly conducts occupancy-oriented surveys for tadpoles and metamorphs during a 6-8-week period after the snow has melted (Brown et al., 2012;Brown & Olsen, 2013). Deploying autonomous recording units (ARUs) in breeding areas enables PAM of the extensive vocal activity produced by adult male toads during their breeding period, which could, in turn, yield insights into breeding phenology and possibly more detailed population data. In April 2022, we conducted a pre-breeding expedition to deploy ARUs along a transect of Yosemite toad breeding areas (Figure 1), which entailed a one-day gear caching effort (two people) followed by a five-day ARU deployment trip (five people) covering a total of 89 km with 4000 m of climbing, and a maximum elevation of 3670 m. The expedition was timed to maximize snow cover (facilitating efficient travel) and minimize snowpack instability (mitigating avalanche risk). The lessons learned from this project may inform future monitoring efforts for this species and other vocally active species living in remote areas.

| CHALLENGES OF REMOTE PASSIVE ACOUSTIC MONITORING
A prerequisite for PAM projects that require significant human-powered travel is the formation of a team with diverse skillsets. At a minimum, the team must have expertise about bioacoustics, the species of interest, and the terrain to be surveyed and relevant modes of travel. In our case, Connor Wood is a bioacoustics expert, Cathy Brown is a Yosemite toad expert, and Jason Champion was the lead guide of the deployment expedition. Accessing the ARU deployment locations entailed extensive travel through avalanche terrain via splitboards (snowboards that can be reconfigured to function as crosscountry skis; skis or snowshoes could have been used to conduct a similar survey effort) with occasional use of crampons and ice axes. All members of the deployment expedition (Connor Wood, Jason Champion, Will Brommelsiek, Isaac Laredo, and Roan Rogers) had formal training in avalanche safety and wilderness first aid, and Jason Champion is certified as a Splitboard Guide by the American Mountain Guides Association. Safe travel through remote areas, and alpine habitats in particular, can be equipment-and training-intensive, which can present a multi-faceted barrier to entry. Building a team with complementary strengths can help offset differences in experience and ability, and some universities have outdoor skills training and equipment libraries. In addition to the formation of the project team with suitable skills, there are several key challenges, aspects of which may be relevant to other researchers considering remote deployments of ARUs: (1) limitations to recording equipment, (2) physical and mental limitations to traveling through rugged terrain, and (3) making appropriate population inferences based on survey results.

| Hardware limitations
The effort required to reach remote survey locations incentivizes ARU deployments lasting many weeks or months. Power supply, hardware durability, and speciesspecific recording constraints will be the most salient limiting factors.
Power consumption is influenced by the recording schedule and sample rate, as well as intrinsic factors like circuitry. Less recording time per day and lower sample F I G U R E 1 Deploying autonomous recording units before the Yosemite toad's late-April-early-June breeding season requires extensive human-powered travel through rugged, snow-covered terrain in the Sierra Nevada, USA. Both photos: C. Wood. rates reduce power consumption but potentially reduce the chances of detecting the target species and recording higher frequencies, respectively. Compared to alkaline batteries, lithium batteries are lighter, provide much more power, do so at a near-constant high level until they are fully discharged, and can operate in a wider range of temperatures. However, lithium batteries are generally ten times as expensive as alkaline batteries and not all circuit boards can accommodate the higher voltage that lithium batteries provide. Voltage moderation can be achieved via switching supply (efficient but expensive and noisy) or linear supply (quiet and inexpensive but inefficient). It is unlikely that many teams will be able to choose their means of voltage moderation (if any) but understanding the basic mechanisms can inform power supply choices. Solar panels combined with rechargeable batteries are another option, but substantially increase per-unit cost and make the ARU much more vulnerable to damage. Finally, although most modern SD cards have some form of automatic write assist and memory fault protection, a combination of low input power for the SD card controller chip and power loss can interrupt the process of writing and allocating files to the card. If allocation interruption occurs, the entire SD card (not just the file in question) can be corrupted; unfortunately, largercapacity cards, which are more likely to be used in remote deployments, are more susceptible to this issue. Researchers can mitigate this risk by programming a schedule that ends before battery life is projected to expire and ensuring that ARU firmware can detect lowpower events and supports "graceful shutdowns".
Long-term deployments demand reliable physical construction of the ARUs and their components. We used SwiftOne recorders (Cornell Lab of Ornithology, Ithaca) housed in a Pelican case with one internal, omnidirectional microphone. A flush-mounted or internal microphone may sacrifice some recording performance, but it makes the unit more resilient to physical damage-likely a worthwhile trade-off given the effort required for remote deployments.
Finally, it is important to understand the recording range of the ARUs, the maximum distance at which a target signal can be recorded and reliably identified. Range is thus a function of the species (e.g., peak amplitude of the target signal), recording environment (e.g., dense or open vegetation, complex or open terrain) recording conditions (e.g., humidity, precipitation, wind, or other ambient sounds), and recording settings (e.g., gain and microphone quality). Recorder range and spatial distribution combine with focal species space-use in potentially complex ways to determine the spatial resolution of a study, a topic reviewed by Sugai et al. (2020). Teams should understand at least approximately how much area is being sampled by each ARU and thus how many units are needed to survey an informative area or proportion of the population. Yet the number of ARUs is not simply a question of study design.

| Physical and mental constraints
Batteries and ARUs are heavy, and human-powered expeditions can only carry limited amounts of equipment. The efficiencies gained by increasing expedition team size are finite, as are project budgets, meaning that "hiring more porters" is not an indefinite solution. The likelihood of compression of or damage to peripheral nerves increases with pack weight (Anderson et al., 2009), with 20%-30% of one's weight up to 13.6 kg (30 lbs) considered maximum pack weights to minimize the risk of discomfort or injury (Thomas, 2013). For context, our team's average base weight (pack, sleeping bag and pad, ice ax, boot-and ski crampons, and one person's proportion of the shelters and stoves) was roughly 7 kg (15.4 lbs). We also had 14 ARUs, which added 4 kg (8.8 lbs) per person, and with food, water, fuel, and clothing we were well over the advised limit at the start of the trip (Figure 2). Particularly in environments where winter gear and/or mountaineering gear is required, thus substantially increasing the base weight of all team members' packs, we suggest that pre-expedition gear caches be employed to the greatest extent possible such that selfsupported travel is limited to 3-to 4-d periods. The more weight each team member carries, the greater the daily physical demands and risk of injury, and the smaller the margin of safety if a serious injury is sustained.
Intense and sustained exertion, particularly when traveling in remote environments, exerts a psychological F I G U R E 2 Equipment weight is a major limitation to humanpowered expeditions. Pack weights substantially exceeded recommended limits during our trip, which contributed to slowerthan-expected travel, particularly at the highest elevations (4000 m). Photo: W. Brommelsiek. toll in addition to a physical one. Maintaining honest and respectful conversation among expedition team members is essential to safety (particularly in avalanche terrain), as well as ensuring that group dynamics remain positive (training on group dynamics and safety in remote settings is often included in advanced wilderness first aid curricula). For example, failure of communication can lead to expedition-ending injuries if team members do not feel comfortable sharing their limitations. Team leadership should also be aware that elevated physical and mental strain can amplify identity-based pressures (e.g., genderbased performance pressures [Demery & Pipkin, 2021]). We started each day by discussing our route plan, then maintained a continual dialogue throughout the day, including observations of snow condition or rockfall, individual energy levels, and mental state, and ended each day with a brief recap. We strongly advise that scientists participating in expeditions make a genuine commitment to themselves and a public commitment to the team that safety is a higher priority than scientific objectives.
During our deployment expedition, we substantially changed our route in response to the limits imposed by heavy packs and lower than expected snow levels. The lead guide (Jason Champion) and lead scientist (Connor Wood) had had many pre-expedition conversations about this possibility, so the decision was disappointing but not difficult. The creation of predeparture alternative ARU deployment locations made adjusting the plan easier; we further recommend that teams load a habitat suitability map on their GPS devices (if one is available) to further facilitate contingency planning. The physical and mental limits on deployment and retrieval expeditions imposed by the combination of the terrain, equipment needs, and team member preparations are unavoidable, and they are the ultimate deciding factor in where ARUs are deployed. Survey locations, in turn, have important implications for how the resulting data are used.

| Data interpretation
ARU deployment locations will be determined in part by what is possible, not always what is inferentially desirable. As a result, integrating expedition data with ongoing or future monitoring programs may be challenging. Statistically, the population that is being surveyed by expeditions will be defined in part by the limits of the deployment/retrieval teams. Appropriately interpreting non-detections will require an understanding of ARU recording range, especially if ARUs are deployed without knowledge of the target species' space use. Whether ARUs deployments are based on knowledge of the target species' space use also has important implications for how detections are interpreted .
Generating species detection/non-detection data are, of course, a precursor to the biological interpretations described above. For the Yosemite toad and >3000 species, most of them birds, the machinelearning algorithm BirdNET may be effective (Kahl et al., 2021). Yet rare species may not be included in premade tools. Custom detectors can be developed from limited archival recordings with software like Raven Pro 2.0 (Cornell Lab of Ornithology, Ithaca) (see Arvind et al., 2022) or with few-and zero-shot learning algorithms (Wang et al., 2021), and, for anurans in small-scale deployments, pulse detection tools may be effective (Lapp et al., 2021).

| OPPORTUNITIES FOR CONSERVATION SUCCESSES
Despite the challenges of data interpretation, and of even executing passive acoustic surveys in remote areas, any new data they yield may be valuable for endangered or data-deficient species (e.g., Rojas-Bracho et al., 2022). Thus, for rare, vocally active species, especially anurans, mammals, and birds, bioacoustics may be a powerful tool for conservation (Table 1) (Arvind et al., 2022;Neilson et al., 2013). The logistical challenges of accessing remote areas apply equally to active and passive surveys, and PAM may enable teams to visit more locations by alleviating the time pressure of conducting surveys exclusively during sometimes brief periods of species activity. Of course, ARUs must be retrieved, necessitating a second trip, but there is very little time pressure for such tripsand two trips might be an improvement over more intensive protocols (Neilson et al., 2013). Moreover, the value of audio data as a permanent archive of an actually or potentially endangered species can be quite high, Opportunities for public engagement via storytelling particularly if offers proof that the species inhabits a previously unknown location.

| Yosemite toad
There are several long-term possibilities for PAM of the Yosemite toad. Acoustic-based abundance indices would add a valuable dimension to ongoing occupancy-based monitoring (Brown et al., 2012), and could be combined with long-term, ongoing occupancy data via an integrated population model (Tempel et al., 2014). PAM could also be used to assess the occupancy status of breeding sites that would not have been otherwise surveyed in a given year. In either case, ARUs would need to be deployed before the breeding season, either in the early spring of that breeding season or in the preceding summer or fall.
Our early spring ARU deployment expedition resulted in the successful documentation of adult male Yosemite toads, via passive acoustic surveys and a newly expanded machine-learning algorithm, BirdNET. Accessing these areas before the breeding season was facilitated by the extensive experience of the local members of the deployment team (Jason Champion, Will Brommelsiek, Isaac Laredo, and Roan Rogers); retrieval of the ARUs was simpler, as sites were accessed via snow-free hiking trails during the summer and partially combined with ongoing Yosemite toad monitoring efforts. The extensive winter sports community in California could be a source of citizen scientists to conduct early-spring ARU deployments (and postbreeding retrievals), though such a plan would require careful consideration from relevant stakeholders. Long-term, climate change-driven decreases to California's snowpack are likely to be detrimental to both that recreational community and the toad itself, and public outreach to highlight that connection could help motivate participation in survey efforts or at least influence attitudes toward conservation. However, directional changes in the snowpack, as well as the complexity of pre-breeding ARU deployment trips, could also incentivize alternative deployment strategies.
In late fall 2021, we also deployed several ARUs that had been modified to accommodate 12 lithium batteries, as opposed to the standard six, and were programmed to remain on standby all winter and then record April-June. These units successfully recorded Yosemite toads, suggesting that overwinter deployments may be a viable alternative to the more logistically complex early spring deployments. An extension of the late fall deployment is a quasi-permanent deployment in which ARUs remain on-site indefinitely, receiving an annual visit to change batteries, SD cards, and check for damage. Although quasi-permanent deployments would reduce the demands on project personnel, a key uncertainty is how many years of extreme heat and cold the hardware could withstand. Both fall and quasi-permanent deployments require selecting a recording schedule without knowledge of the year's snowpack, which incentivizes reduced daily recording time to increase seasonal recording time. Regardless of deployment date, basing deployment locations on knowledge of likely breeding areas within putatively occupied meadows will be important.
Finally, rapid advances in multispecies sound identification algorithms such as BirdNET mean that audio collected "for" Yosemite toads can also be analyzed to determine the presence of sympatric amphibians like the Pacific chorus frog (Pseudacris regilla) and other meadow specialists like the Great Gray Owl (Strix nebulosa Yosemitensis).

| Global prospects
Among terrestrial vertebrates globally, amphibians have by far the greatest proportion of threatened or endangered species (41%) and the greatest proportion of data-deficient species (25.4%) (Bland et al., 2017). Datadeficient amphibians are more likely than fully assessed species to be threatened with extinction (Howard & Bickford, 2014), and data-deficient species with restricted geographic ranges have been suggested as priorities for population surveys (Morais et al., 2013). Careful survey work recently led to the rediscovery of at least one amphibian that had been classified as "possibly extinct" (Amor os et al., 2020).
Species that are data deficient, putatively range-limited, and acoustically active are ideal candidates for bioacoustic survey expeditions, and the biogeographic forces that gave rise to one rare species may well have shaped others as well (Noroozi et al., 2018). Fortunately, passive acoustic surveys are uniquely capable of providing data about multiple species (Wood et al., 2019). Steep elevational gradients in the tropics may harbor candidate anurans, birds (e.g., Williams & Fuente, 2021), and mammals (e.g., Neilson et al., 2013). Multispecies surveys pose their own ARU deployment location challenges , but coordination among conservationists to compile a pre-expedition list of vocally active priority species can maximize the potential benefits of bioacoustics expeditions. The adventurous aspects of remote survey expeditions could be leveraged for public-facing storytelling, an approach exemplified by the National Geographic Okavango Wilderness Project (https://www.nationalgeographic. org/projects/okavango/). Critically, however, not all surveys-even those targeting species in geographically remote areas-are likely to require physically strenuous field efforts or specialized training.
The success of passive acoustic survey expeditions to assess the distribution and population status of vocally active species in remote areas will hinge on partnerships. Teams must have expertise in bioacoustics, focal species biology, and in traveling through the relevant landscape. Partnerships with local communities can be both necessary and quite rewarding. Members of indigenous communities may even have knowledge of species distributions that "western" scientists have not documented. Ultimately, cultivating a team dynamic in which all members are comfortable sharing honest assessments about relevant limitations-from personal fitness to statistical inference-will be essential to the success of expedition-style bioacoustic surveys.

ACKNOWLEDGMENTS
We thank Nick Russell for assistance retrieving the recording units, Chris Tessaglia-Hymes for additional technical insights, and Barrett Wood for wilderness physiology input. Connor Wood secured funding and conceived the study; Connor Wood, Jason Champion, and Cathy Brown planned the deployment expedition; Connor Wood, Jason Champion, Will Brommelsiek, Isaac Laredo, and Roan Rogers conducted the deployment expedition; Connor Wood and personnel supervised by Cathy Brown conducted retrievals; Connor Wood wrote the first draft of the manuscript and all authors contributed to the final manuscript and gave their approval for submission.

FUNDING INFORMATION
This work was funded by the Explorers Club Discovery Expedition Grant program.