Survey coverage, recording duration and community composition affect observed species richness in passive acoustic surveys

Abstract

Bioacoustic assessments of species richness are rapidly becoming attainable, but uncertainty regarding the optimal acoustic survey design remains. Selecting the duration of recording and the number of recording units are critical decisions, and we used both simulated and empirical data to quantify the trade‐offs those choices present. We evaluated the performance of 30 hypothetical acoustic survey designs (e.g. continuous recording, every other 5 min, etc.). Simulated bird species' (n ≤ 60) abundance across the study area, probability of daily availability and time‐dependent probability of vocal activity varied randomly within ranges of realistic values. Field data, collected in central New York, USA (747 hr) and in the northern Sierra Nevada, USA (1,090 hr), was analysed with a novel machine‐learning algorithm, BirdNET. All three datasets were subsampled at 5‐min intervals, observed species richness was compared across survey designs, and detection probability was calculated for each species. Observed species richness increased with survey coverage (number of recording units) and with recording duration in all three datasets. The impact of differences in survey coverage decreased as recording duration decreased. Species' detection probabilities were negatively affected by reducing the number of days of recording and by reducing the daily recording duration. The more rare species a community had, the more species richness was underestimated as survey coverage decreased. Rarefaction curves indicated that increasing recording time has diminishing marginal utility but that the asymptote varies among communities. The cost per species observed decreased with increasing recording duration. Discontinuous and reduced‐coverage sampling may still yield fairly accurate assessments of biodiversity but reducing recording duration or coverage will result in different species remaining undetected. Whether the performance of a study design is ‘good’ or ‘bad’ depends on researchers' constraints and scientific questions to be answered. More hardware and longer recording durations are not always better, but we caution researchers against doing the bare minimum required for their present needs without pressing financial reasons to do so.

Simultaneous passive acoustic monitoring uncovers evidence of potentially overlooked temporal variation in an Amazonian bird community

“…3). That 23% (27/120) of the species in our study were featured on just one recording site, and 32% and 27% at one site and day, respectively, suggests that ARU performance should be expected to improve continuously over a larger sample (Wood et al 2021). Annotating many short-duration time blocks, representing a broader range of temporal windows than contiguous dawn-hours, can reduce requisite analysis time or improve results, including higher detection rates of uncommon and rare species, more complete estimates of alpha and gamma diversity, and fewer false absences (Shonfield & Bayne 2017, Metcalf et al 2021.…”

“…3). That 23% (27/120) of the species in our study were featured on just one recording site, and 32% and 27% at one site and day, respectively, suggests that ARU performance should be expected to improve continuously over a larger sample (Wood et al . 2021).…”

Optimizing landscape‐scale monitoring programmes to detect the effects of megafires

“…If multiple megafires occur in a single season, affected monitoring sites from multiple fires could be pooled (within reasonable context‐dependent limitations) to effectively increase sample size and move beyond a single‐fire focus. Alternatively, comparing site occupancy to fire history for all sites in a monitoring programme could allow researchers to infer colonization rates based on the duration of post‐fire vacancy, which, in turn, could be used to develop mechanistic understandings of focal species and even whole communities’ responses to megafires (Wood et al, 2021). The proliferation of landscape‐scale monitoring programmes, especially in the western United States, represents a further opportunity to increase sample sizes: if acoustic, camera or other data from multiple monitoring programmes can be pooled and analysed in a unified framework, regional‐scale understandings of species’ responses to fire and other disturbances may be attainable (Rich et al., 2019).…”

“…By virtue of their spatial extent, landscape‐scale monitoring programmes should be well‐suited to provide survey coverage within the footprints of future megafires, potentially allowing for before‐after/control‐impact (BACI) studies. Data from camera traps, hair snares and autonomous recording units (ARUs) collected at landscape scales can support statistically powerful multispecies occupancy and mark–recapture analyses to assess population trends (Banner et al., 2018; Ellis et al., 2014; Magoun et al., 2011; Steenweg et al., 2016; Wood et al, 2021; Wood et al., 2019). However, the statistical power of these types of non‐invasive monitoring programmes has primarily been evaluated in the context of population trends affecting many or all sites, rather than the effects of localized disturbances (e.g., Banner et al., 2019; Ellis et al., 2014; Steenweg et al., 2016; Wood et al., 2019).…”

Overview of LifeCLEF 2021: An Evaluation of Machine-Learning Based Species Identification and Species Distribution Prediction

“…Passive acoustic monitoring can be a valuable sampling tool for habitat assessments and the observation of environmental niches which often are endangered. However, manual processing of large collections of soundscape data is not desirable and automated attempts can help to advance this process [57]. Yet, the lack of suitable validation and test data prevented the development of reliable techniques to solve this task.…”