Assessing the value of acoustic indices to distinguish species and quantify activity: A case study using frogs

1. Soundscapes can provide information about a wide range of habitats and species through the recording of vocalisations over long temporal scales. Because of the large volumes of data collected, computational approaches, such as the application of acoustic indices, are required to extract useful information from long-duration recordings.
2. Acoustic indices summarise various soundscape features into frequency ranges over defined time intervals and can aid in the visual exploration, detection, and analysis of species vocalisation patterns. Here, we examine the performance of combinations of three acoustic indices commonly used in visual exploration, the acoustic complexity index, the temporal entropy spectrum index, and the event spectrum index, and assess their ability to distinguish species and describe acoustic features commonly used to detect species and analyse activity. Our case study focuses on three frog species with distinct call structures from Bickerton Island, Northern Territory, Australia. Call structure was categorised based on the number of pulses and harmonics.
3. We summarised acoustic activity by calculating acoustic indices in 256 equal-sized bins over the entire the frequency spectrum, for 30-s intervals, and found that acoustic index values could be used to distinguish species and describe acoustic features. The acoustic complexity index was the most effective index for distinguishing species. To describe acoustic features, we examined correlations between acoustic index values and summarised acoustic features, including call rate, total duration, loudness and signal-to-noise ratio. In single-pulsed species with no harmonics, we found spectral index values were significantly and sometimes strongly correlated with acoustic features. In comparison, species with harmonics were found to be weakly and less frequently correlated with acoustic features even if sampled calls were loud and have high signal-to-noise ratio. We suggest that acoustic indices have the potential to describe acoustic features in single-pulsed species but are limited in those with harmonics.
4. We conclude that acoustic indices can be a useful tool to distinguish some anuran species and to broadly understand specific acoustic features used to analyse calling activity over long periods of time.
5. Further research is required to better understand the relationships between acoustic indices and acoustic features to determine the general utility of indices to detect and distinguish audible species and to identify other acoustic features of various taxa.

     

Acoustic monitoring reveals diversity and surprising dynamics in tropical freshwater soundscapes

1. Freshwater systems are globally threatened and in need of enhanced monitoring and assessment. We applied soundscape recording and analysis—which presents an opportunity for long-term, high-resolution animal community monitoring and assessment—to a freshwater context to better understand the acoustic diversity and dynamics of these systems.
2. We recorded the aquatic soundscape of a Neotropical freshwater swamp in Costa Rica for 23 days in January and February 2015 during the dry season. We classified biological sound types in these recordings and developed measurements of richness and occupancy based on this classification. We also calculated six complementary acoustic indices to assess soundscape diversity and daily and longer-term soundscape dynamics, and we examined correlations between these acoustic indices and sound type metrics.
3. We found rich soundscapes in which biological sounds were almost always present, and we classified 18 sound types that we attribute to aquatic insects. These sound types showed distinct daily patterns and exhibited temporal and spectral acoustic niche partitioning. Sound type richness was most correlated with the number of peaks index (correlation = .36; p < .001), while sound type occupancy was most correlated with the Bioacoustic Index (correlation = .92; p < .001). In contrast to generally high levels of acoustic activity, there were brief (approximately 1 hr), unexpected quiet periods around dawn and dusk.
4. This study represents an early attempt to comprehensively describe tropical freshwater soundscapes in a systematic and quantitative manner. We demonstrate that sound type classification and the quantification of acoustic occupancy capture aspects of soundscape diversity and dynamics that are complementary to those assessed by acoustic indices. Our analyses reveal that the soundscapes of this tropical wetland were diverse and exhibited daily dynamics that differed from those found in other ecosystems.

     

Ecoacoustics can detect ecosystem responses to environmental water allocations

1. Globally, water abstraction for human consumption and irrigated agriculture leads to significant changes in aquatic ecosystems. To counter these detrimental effects, water releases—often termed environmental water allocations—restore overbank flow or are delivered to artificially disconnected wetlands. While a suite of monitoring methods is available, few programmes track continuous change in biota, mainly because repeated remote site visits can be prohibitively expensive.
2. In this paper, we propose a new approach to environmental flow monitoring, using ecoacoustic methods. We test acoustic monitoring of frog and waterbird responses to environmental water deliveries in the Goulburn Broken, a valley in the southern Murray–Darling river system. Response to three major environmental water deliveries within 2 years was monitored at four sites along Reedy swamp. Every 2 weeks, 30 s were recorded every 30 min, for a total of 24 hr.
3. We used two analysis strategies—manual counts of bird calls, as well as ecoacoustic indices, which describe the sonic properties of the acoustic spectrum at a site. Manual counts demonstrated that water-dependent birds were clearly responding to environmental water deliveries, whereas non-water-dependent species did not show any increases in activity. After restricting the analysis to the dawn chorus of birds and frogs, two acoustic indices (the median amplitude and the acoustic complexity index) showed responses to watering events.
4. Ecoacoustic methods show promise for continuous response monitoring to environmental water allocations. However, the first strategy—manual annotation of calls—might be too labour intensive for standard monitoring programmes. The second strategy—index-based approaches—can also detect ecological responses, although further investigation using control sites is needed. Automated call classifiers are an alternative that is currently being developed for endangered species. We also encourage simultaneous monitoring of the soundscape above and under water.

     

Automated acoustic monitoring of endangered common spadefoot toad populations reveals patterns of vocal activity

1. In the context of global amphibian decline, monitoring and restoration programmes are important. Acoustic monitoring is a possible approach for underwater vocalising species like the rapidly declining European common spadefoot toad (Pelobates fuscus). In this study, our aim was to design a dedicated software detector to be used in combination with programmable audio recorders to process the large amount of data generated by long-term acoustic monitoring and to use it for investigating the seasonal and circadian patterns of P. fuscus vocal activity.
2. The software detector targets advertisement calls of the species. Based on acoustic analysis of that call, we developed a detector that utilises both frequency and time features of the calls. Data collected during three breeding seasons in four known or potential P. fuscus breeding sites of north-eastern France were used to build a ground truth in order to test the performance of the detector. Then, we used the detector for analysing four acoustic monitoring campaigns conducted in two different sites over two breeding seasons to gain insight into the seasonal and circadian patterns of vocal activity of this species.
3. Evaluation of the P. fuscus call detector against a ground truth returned false-positive rates below 1.5% and true-positive rates ranging from 53% to 73%. These figures are compatible with long-term monitoring of the presence of the species. Running the software detector on standard hardware, the computation time for post-processing the 360 hr of a typical 3-month monitoring campaign was less than 1 day.
4. The seasonal pattern of P. fuscus underwater vocal activity is more complex than previously recognised. Over the whole ostensible 3-month breeding season, the actual time window for vocalising and breeding can last from a few days up to several weeks and may be split into clearly distinct episodes. When vocalisations occurred at both night- and daytime, the circadian vocal activity of P. fuscus occasionally proceeded uninterrupted for 24 hr but usually a several hour lull occurred immediately prior to sunset. When vocalisations occurred at both night- and daytime, the vocal activity pattern followed a bimodal distribution with a nocturnal highest peak of activity and a second peak occurring in the morning.
5. Our results demonstrate that it is feasible to monitor presence of P. fuscus in north-eastern France using a dedicated software detector combined with programmable audio recorders. Based on the outcomes of the detector applied to long-term audio data sets, we reveal temporal patterns of the vocal activity of the species and subsequently provide recommendations for attended and unattended acoustic monitoring.

     

Looking for love under the ice: Using passive acoustics to detect burbot (Lota lota: Gadidae) spawning activity

1. Burbot (Lota lota: Gadidae) is a difficult species to manage effectively due to its preference for deep-water habitats and under-ice spawning behaviour, resulting in a poor understanding of its reproductive activity. However, the use of acoustic signalling by burbot as part of their mating system has recently been described and this behaviour may provide a means of investigating questions regarding the spatial and temporal distribution of spawning aggregations using passive acoustic monitoring.
2. We used audio and video recording to confirm that burbot vocalise and that these vocalisations can be detected under field conditions as well as to characterise the relationship between burbot acoustic signalling and spawning behaviour. We also evaluated the feasibility of locating and monitoring burbot spawning aggregations in real time using passive acoustics.
3. Burbot vocalisations were difficult to identify with only about 6% of the recordings containing calls being successfully identified as such in the field. Burbot vocalised more often between sundown and sunrise than during daylight hours. Calls recorded at night tended to be lower frequency, longer duration, and have lower bandwidth than those made during the day.
4. Burbot vocalisations could not be recorded in conjunction with video recordings of spawning activity, indicating that burbot may not call during active spawning, but may use acoustic communication to signal the onset of reproductive readiness and to form pre-spawning aggregations.
5. While burbot calls were readily identifiable, observers had a difficult time identifying burbot calls in real time under field conditions. Passive acoustic monitoring demonstrates considerable potential as a management tool to locate burbot spawning grounds and identify periods of activity, but may not be an appropriate technique for monitoring spawning activity in real time.

     

SILIC: A cross database framework for automatically extracting robust biodiversity information from soundscape recordings based on object detection and a tiny training dataset

• 1.Passive acoustic monitoring (PAM) offers many advantages comparing with other survey methods and gains an increasing use in terrestrial ecology, but the massive effort needed to extract species information from a large number of recordings limits its application. The convolutional neural network (CNN) has been demonstrated with its high performance and effectiveness in identifying sound sources automatically. However, requiring a large amount of training data still constitutes a challenge.
• 2.Object detection is used to detect multiple objects in photos or videos and is effective at detecting small objects in a complex context, such as animal sounds in a spectrogram and shows the opportunity to build a good performance model with a small training dataset. Therefore, we developed the Sound Identification and Labeling Intelligence for Creatures (SILIC), which integrates online animal sound databases, PAM databases and an object detection-based model, for extracting information on the sounds of multiple species from complex soundscape recordings.
• 3.We used the sounds of six owl species in Taiwan to demonstrate the effectiveness, efficiency and application potential of the SILIC framework. Using only 786 sound labels in 133 recordings, our model successfully identified the species' sounds from the recordings collected at five PAM stations, with a macro-average AUC of 0.89 and a mAP of 0.83. The model also provided the time and frequency information, such as the duration and bandwidth, of the sounds.
• 4.To our best knowledge, this is the first time that the object detection algorithm has been used to identify sounds of multiple wildlife species. With an online sound-labeling platform embedded and a novel data preprocessing approach (i.e., rainbow mapping) applied, the SILIC shows its good performance and high efficiency in identifying wildlife sounds and extracting robust species, time and frequency information from a massive amount of soundscape recordings based on a tiny training dataset acquired from existing animal sound databases. The SILIC can help expand the application of PAM as a tool to evaluate the state of and detect the change in biodiversity by, for example, providing high temporal resolution and continuous information on species presence across a monitoring network.

     

Temporally adaptive acoustic sampling to maximize detection across a suite of focal wildlife species

相似文献   

Singing streams: Describing freshwater soundscapes with the help of acoustic indices

1. Understanding soundscapes, that is, the totality of sounds within a location, helps to assess nature in a more holistic way, providing a novel approach to investigating ecosystems. To date, very few studies have investigated freshwater soundscapes in their entirety and none across a broad spatial scale.
2. In this study, we recorded 12 freshwater streams in South East Queensland continuously for three days and calculated three acoustic indices for each minute in each stream. We then used principal component analysis of summary statistics for all three acoustic indices to investigate acoustic properties of each stream and spatial variation in their soundscapes.
3. All streams had a unique soundscape with most exhibiting diurnal variation in acoustic patterns. Across these sites, we identified five distinct groups with similar acoustic characteristics. We found that we could use summary statistics of AIs to describe daytimes across streams as well. Most difference in stream soundscapes was observed during the daytime with significant variation in soundscapes both between hours and among sites.
4. Synthesis and Application. We demonstrate how to characterize stream soundscapes by using simple summary statistics of complex acoustic indices. This technique allows simple and rapid investigation of streams with similar acoustic properties and the capacity to characterize them in a holistic and universal way. While we developed this technique for freshwater streams, it is also applicable to terrestrial and marine soundscapes.

     

Monitoring the acoustic activity of an aquatic insect population in relation to temperature,vegetation and noise

1. Acoustic population monitoring is a noninvasive method that can be deployed continuously over long periods of time and at large spatial scales. One of the newly discovered threats acting on biological diversity is anthropogenic noise. High levels of anthropogenic noise occur in aquatic environments, yet their effects on animals living in freshwater habitats have very rarely been investigated.
2. Here, we used acoustic monitoring and automatic detection to assess the acoustic activity of a population of a soniferous freshwater insect.
3. The sounds emitted by the corixid Micronecta scholtzi were recorded in a Mediterranean pond with an array of 12 hydrophones. An automatic analysis based on a measure of the amplitude found in the frequency band of M. scholtzi was developed to assess the level of acoustic activity. We used functional linear models, accounting for the periodicity of the calling behaviour, to estimate the possible effect of temperature, vegetation and a noise due to an immersed engine.
4. The automatic analysis was validated as an efficient method to measure the acoustic activity. The monitoring revealed a clear 24-hr pattern in the acoustic activity of M. scholtzi and three peaks of activity during the morning. Functional linear models revealed negative effects of both temperature and vegetation and showed that an engine noise, played back for 2 hr during the night, elicited an increase in the level of acoustic activity of the population. Moreover, a cross-correlation procedure showed that noise delayed the acoustic activity of the population.
5. Our results suggest that acoustic survey and automatic detection are efficient methods to monitor the acoustic activity of an insect population especially in response to an anthropogenic disturbance.

     

Calling at the highway: The spatiotemporal constraint of road noise on Pacific chorus frog communication

Loss of acoustic habitat due to anthropogenic noise is a key environmental stressor for vocal amphibian species, a taxonomic group that is experiencing global population declines. The Pacific chorus frog (Pseudacris regilla) is the most common vocal species of the Pacific Northwest and can occupy human‐dominated habitat types, including agricultural and urban wetlands. This species is exposed to anthropogenic noise, which can interfere with vocalizations during the breeding season. We hypothesized that Pacific chorus frogs would alter the spatial and temporal structure of their breeding vocalizations in response to road noise, a widespread anthropogenic stressor. We compared Pacific chorus frog call structure and ambient road noise levels along a gradient of road noise exposures in the Willamette Valley, Oregon, USA. We used both passive acoustic monitoring and directional recordings to determine source level (i.e., amplitude or volume), dominant frequency (i.e., pitch), call duration, and call rate of individual frogs and to quantify ambient road noise levels. Pacific chorus frogs were unable to change their vocalizations to compensate for road noise. A model of the active space and time (“spatiotemporal communication”) over which a Pacific chorus frog vocalization could be heard revealed that in high‐noise habitats, spatiotemporal communication was drastically reduced for an individual. This may have implications for the reproductive success of this species, which relies on specific call repertoires to portray relative fitness and attract mates. Using the acoustic call parameters defined by this study (frequency, source level, call rate, and call duration), we developed a simplified model of acoustic communication space–time for this species. This model can be used in combination with models that determine the insertion loss for various acoustic barriers to define the impact of anthropogenic noise on the radius of communication in threatened species. Additionally, this model can be applied to other vocal taxonomic groups provided the necessary acoustic parameters are determined, including the frequency parameters and perception thresholds. Reduction in acoustic habitat by anthropogenic noise may emerge as a compounding environmental stressor for an already sensitive taxonomic group.     

Automated auditory detection of a rare,secretive marsh bird with infrequent and acoustically indistinct vocalizations

Autonomous recording units (ARUs) provide a non-invasive and efficient method for acoustic detection of elusive species across large temporal and spatial scales. However, species with indistinct vocalization structures can be a considerable challenge for automated signal recognizers. We investigated the performance of ARUs and signal recognizers in identifying the broadband, short-syllable, pulsed calls of a secretive, threatened marsh bird, the King Rail Rallus elegans. Other sympatric species in the same habitat also have repetitive calls within the same frequency range that can be difficult to distinguish. Following serial ARU deployments at specified sites in known breeding habitat, we conducted standardized callback surveys and nest searches to provide an independent measure of breeder density. To analyse recordings, we developed a signal recognizer based on user-input training files to detect two common call types, kek and grunt. Detections that remained following manual review of recognizer output revealed a previously undescribed seasonal decline and crepuscular diel pattern in calling rate. The rate of the grunt call also predicted density. These patterns emerged despite the recognizer's low precision and high false-positive rate, which were largely due to misclassification of other species' calls, although ambient noise and effective detection radius also limited the detectability of King Rail calls. We demonstrate that with informed ARU scheduling, improved ability to manipulate user-specified parameters within signal detection software, and attention to quality control, even the simplest call structures can be located consistently in a diverse acoustic landscape. Our behavioural findings will inform improvements to auditory surveys and to management of King Rails across their range.     

Mutualism promotes site selection in a large marine planktivore

1. Mutualism is a form of symbiosis whereby both parties benefit from the relationship. An example is cleaning symbiosis, which has been observed in terrestrial and marine environments. The most recognized form of marine cleaning symbiosis is that of cleaner fishes and their clients.
2. Cleaner species set up cleaning stations on the reef, and other species seek out their services. However, it is not well understood how the presence of cleaning stations influence movements of large highly mobile species. We examined the role of cleaning stations as a driver of movement and habitat use in a mobile client species.
3. Here, we used a combination of passive acoustic telemetry and in‐water surveys to investigate cleaning station attendance by the reef manta ray Mobula alfredi. We employed a novel approach in the form of a fine‐scale acoustic receiver array set up around a known cleaning area and tagged 42 rays. Within the array, we mapped structural features, surveyed the distribution of cleaner wrasse, and observed the habitat use of the rays.
4. We found manta ray space use was significantly associated with blue‐streak cleaner wrasse Labroides dimidiatus distribution and hard coral substrate. Cleaning interactions dominated their habitat use at this site, taking precedence over other life history traits such as feeding and courtship.
5. This study has demonstrated that cleaning symbiosis is a driver for highly mobile, and otherwise pelagic, species to visit inshore reef environments. We suggest that targeted and long‐term use of specific cleaning stations reflects manta rays having a long‐term memory and cognitive map of some shallow reef environments where quality cleaning is provided. We hypothesize that animals prefer cleaning sites in proximity to productive foraging regions.

     

Estimating population density of insectivorous bats based on stationary acoustic detectors: A case study

1. Automated recording units are commonly used by consultants to assess environmental impacts and to monitor animal populations. Although estimating population density of bats using stationary acoustic detectors is key for evaluating environmental impacts, estimating densities from call activity data is only possible through recently developed numerical methods, as the recognition of calling individuals is impossible.
2. We tested the applicability of generalized random encounter models (gREMs) for determining population densities of three bat species (Common pipistrelle Pipistrellus pipistrellus, Northern bat Eptesicus nilssonii, and Natterer's bat Myotis nattereri) based on passively collected acoustical data. To validate the results, we compared them to (a) density estimates from the literature and to (b) Royle–Nichols (RN) models of detection/nondetection data.
3. Our estimates for M. nattereri matched both the published data and RN‐model results. For E. nilssonii, the gREM yielded similar estimates to the RN‐models, but the published estimates were more than twice as high. This discrepancy might be because the high‐altitude flight of E. nilssonii is not accounted for in gREMs. Results of gREMs for P. pipistrellus were supported by published data but were ~10 times higher than those of RN‐models. RN‐models use detection/nondetection data, and this loss of information probably affected population estimates of very active species like P. pipistrellus.
4. gREM models provided realistic estimates of bat population densities based on automatically recorded call activity data. However, the average flight altitude of species should be accounted for in future analyses. We suggest including flight altitude in the calculation of the detection range to assess the detection sphere more accurately and to obtain more precise density estimates.

     

声景生态学研究进展和展望

声景生态学以景观中的声音为研究对象, 探讨其在不同时空维度上的分布和变化模式, 从而揭示自然环境、野生动物和人类活动的相互作用关系。本文通过系统检索声景生态学研究的相关文献, 回顾了该学科的研究框架和研究方法, 总结了目前常用的声学指标, 重点归纳了声景生态学的研究内容, 包括声景组成和各组分间的相互作用, 声景的时空格局, 以及声景生态学在生物多样性监测中的应用。目前, 声景监测中存在的问题主要包括监测的生态系统类型和物种类群有限、声学指标效力有待提高等。建议未来着重推进建立系统性的声景监测网络和数据管理平台, 开发和完善音频数据采集、分析方法和评估指标, 并重视声景数据的采集, 将声景视作一种资源进行研究和保护。     

Spatio-temporal heterogeneity in river sounds: Disentangling micro- and macro-variation in a chain of waterholes

1. Passive acoustic monitoring is gaining momentum as a viable alternative method to surveying freshwater ecosystems. As part of an emerging field, the spatio-temporal replication levels of these sampling methods need to be standardised. However, in shallow waters, acoustic spatio-temporal patchiness remains virtually unexplored.
2. In this paper, we specifically investigate the spatial heterogeneity in underwater sounds observed within and between waterholes of an ephemeral river at different times of the day and how it could affect sampling in passive acoustic monitoring.
3. We recorded in the Einasleigh River, Queensland in August 2016, using a linear transect of hydrophones mounted on frames. We recorded four times a day: at dawn, midday, dusk, and midnight. To measure different temporal and spectral attributes of the recorded sound, we investigated the mean frequency spectrum and computed acoustic indices.
4. Both mean frequency spectrum and index analyses revealed that the site and diel activity patterns significantly influenced the sounds recorded, even for adjacent sites with similar characteristics along a single river. We found that most of the variation was due to temporal patterns, followed by between-site differences, while within-site differences had limited influence.
5. This study demonstrates high spatio-temporal acoustic variability in freshwater environments, linked to different species or species groups. Decisions about sampling design are vital to obtain adequate representation. This study thus emphasises the need to tailor spatio-temporal settings of a sampling design to the aim of the study, the species and the habitat.

     

Soundscape analysis and acoustic monitoring document impacts of natural gas exploration on biodiversity in a tropical forest

Natural resource extraction is increasing rapidly in tropical forests, but we lag behind in understanding the impacts of these disturbances on biodiversity. In high diversity tropical habitats, acoustic monitoring is an efficient tool for sampling a large proportion of the fauna across varied spatial and temporal scales. We used passive acoustic monitoring in a pre-montane forest in Peru to investigate how soundscape composition and richness of acoustic frequencies varied with distance from a natural gas exploratory well and with operational phase (construction and drilling). We also evaluated how anuran and avian species richness and vocal activity varied with distance and between phases. Soundscape analyses showed that acoustic frequency similarity was greatest among sites closer to (≤250 m) and farther from (≥500 m) the platform. Soundscapes revealed more frequencies were used during construction and showed a weak trend of increasing frequency richness with increasing distance from the disturbance. Avian species richness and detections increased with distance from the platform, but anuran richness and detections declined with distance. Operational phase did not play a significant role in overall richness or activity patterns of either group. Among birds, insectivore detections increased with distance from the platform, and nectarivores were detected more frequently during the drilling phase. Results demonstrate that acoustic monitoring and soundscape analyses are useful tools for evaluating the impact of development activity on the vocalizing community, and should be implemented as a best practice in monitoring biodiversity and for guiding specific mitigation strategies.     

Surface temperature and shrub cover drive ground beetle (Coleoptera: Carabidae) assemblages in short-rotation coppices

1. Increasing demand for biomass has led to an on-going intensification of fuel wood plantations with possible negative effects on open land biodiversity. Hence, ecologists increasingly call for measures that reduce those negative effects on associated biodiversity. However, our knowledge about the efficiency of such measures remains scarce.
2. We investigated the effects of gap implementation in short rotation coppices (SRCs) on carabid diversity and assemblage composition over 3 years, with pitfall traps in gaps, edges and interiors. In parallel, we quantified soil surface temperature, shrub- and herb cover.
3. Edges had the highest number of species and abundances per trap, whereas rarefied species richness was significantly lower in short rotation coppice interiors than in other habitat types. Carabid community composition differed significantly between habitat types. The main environmental drivers were temperature for number of species and abundance and shrub cover for rarefied species richness.
4. We found significantly higher rarefied species richness in gaps compared with interiors. Hence, we argue that gap implementation benefits overall diversity in short rotation coppices. Furthermore, the differences in species community composition between habitat types through increased species turnover support carabid diversity in short rotation coppices. These positive effects were largely attributed to microclimate conditions. However, to maintain positive effects, continuous management of herb layer might be necessary.

     

The effect of soundscape composition on bird vocalization classification in a citizen science biodiversity monitoring project

There is a need for monitoring biodiversity at multiple spatial and temporal scales to aid conservation efforts. Autonomous recording units (ARUs) can provide cost-effective, long-term and systematic species monitoring data for sound-producing wildlife, including birds, amphibians, insects and mammals over large areas. Modern deep learning can efficiently automate the detection of species occurrences in these sound data with high accuracy. Further, citizen science can be leveraged to scale up the deployment of ARUs and collect reference vocalizations needed for training and validating deep learning models. In this study we develop a convolutional neural network (CNN) acoustic classification pipeline for detecting 54 bird species in Sonoma County, California USA, with sound and reference vocalization data collected by citizen scientists within the Soundscapes to Landscapes project (www.soundscapes2landscapes.org). We trained three ImageNet-based CNN architectures (MobileNetv2, ResNet50v2, ResNet100v2), which function as a Mixture of Experts (MoE), to evaluate the usefulness of several methods to enhance model accuracy. Specifically, we: 1) quantify accuracy with fully-labeled 1-min soundscapes for an assessment of real-world conditions; 2) assess the effect on precision and recall of additional pre-training with an external sound archive (xeno-canto) prior to fine-tuning with vocalization data from our study domain; and, 3) assess how detections and errors are influenced by the presence of coincident biotic and non-biotic sounds (i.e., soundscape components). In evaluating accuracy with soundscape data (n = 37 species) across CNN probability thresholds and models, we found acoustic pre-training followed by fine-tuning improved average precision by 10.3% relative to no pre-training, although there was a small average 0.8% reduction in recall. In selecting an optimal CNN architecture for each species based on maximum F(β = 0.5), we found our MoE approach had total precision of 84.5% and average species precision of 85.1%. Our data exhibit multiple issues arising from applying citizen science and acoustic monitoring at the county scale, including deployment of ARUs with relatively low fidelity and recordings with background noise and overlapping vocalizations. In particular, human noise was significantly associated with more incorrect species detections (false positives, decreased precision), while physical interference (e.g., recorder hit by a branch) and geophony (e.g., wind) was associated with the classifier missing detections (false negatives, decreased recall). Our process surmounted these obstacles, and our final predictions allowed us to demonstrate how deep learning applied to acoustic data from low-cost ARUs paired with citizen science can provide valuable bird diversity data for monitoring and conservation efforts.     

Contributions of MIR to soundscape ecology. Part 3: Tagging and classifying audio features using a multi-labeling k-nearest neighbor approach

Scientists are using acoustic monitoring to assess the impact of altered soundscapes on wildlife communities and human systems. In the soundscape ecology field, monitoring and analyses approaches rely on the interdisciplinary intersection of ecology, acoustics, and computer science. Combining theory and practice of each field in the context of Knowledge Discovery in Databases (KDD), soundscape ecologists provide innovative monitoring solutions for ecologically-driven research questions. We propose a soundscape content analysis framework for improved knowledge outcome with assistance of the new multi-label (ML) concept.Here, we investigated the effectiveness of a ML k-nearest neighbor algorithm (ML-kNN) for labeling concurrent soundscape components within a single recording. We manually labeled 1200 field recordings for the presence of soundscape components and extracted ecological acoustic features, audio profile features, and Gaussian-mixture model features for each recording. Then, we tested the ML-kNN algorithm accuracy with well-established metrics adapted to ML learning.We found that seventeen unique acoustic features could predict a set of biophonic, geophonic, and anthrophonic labels for a single field recording with average precision of 0.767. However, certain labels were predicted incorrectly depending on the time of day and co-occurrence of that label with another label, suggesting further refinement is needed to improve the accuracy of predicted labels.Overall, this ML classification approach could enable researchers to label field recordings more quickly and generate an “alert” system for monitoring changes in a specific sound class. Ultimately, the adaptation of the ML algorithm may provide soundscape ecologists with new metadata labels that are searchable in large databases of soundscape field recordings.     

Post-stocking movement and survival of hatchery-reared bloater (Coregonus hoyi) reintroduced to Lake Ontario

1. Determining the movement and fate of fishes post-stocking is challenging due to the difficulty in monitoring them, particularly immediately after release. Bloater (Coregonus hoyi; Salmonidae) is a deepwater cisco that has been extirpated from Lake Ontario for several decades and is presently the focus of binational restoration stocking efforts; however, there is limited information to evaluate the efficacy of these efforts. The aim of this study was to examine the initial post-release survival, 3D movement, and behaviour of hatchery-reared bloater stocked in Lake Ontario to expand knowledge of post-stocking ecology of fish and inform stocking practices for deepwater ciscoes.
2. In total, 74 hatchery-reared bloater were tagged with acoustic transmitters with depth and temperature sensors in 2016, 2017, and 2018 and passively monitored on an array of 105 69-kHz acoustic receivers deployed in north-eastern Lake Ontario. Several spatial metrics analysed movements after release to investigate immediate post-stocking survival and behaviour for the first time in a pelagic freshwater forage fish.
3. Estimated survival for tagged bloater was low (≤42%) and detection periods of live bloater ranged from 0.2 to 12.1 days (mean ± SD: 2.9 ± 2.9 days). Following release, tagged bloater dispersed quickly and exhibited an association with deeper water (>40 m). Despite overlap in space use for some bloater, there was no evidence of schooling behaviour. Bloater underwent extensive diel vertical migration from near bottom to within metres of the surface. These results demonstrated that, despite high initial mortality, some hatchery-reared bloater survived the initial stress of release and displayed characteristic behaviour of the species.
4. This study demonstrated the value of acoustic telemetry in restoration efforts and revealed survival and behaviour of bloater that has never been observed at this resolution, providing novel information for the management of reintroduced species. Establishment of a self-sustaining population of bloater will help restore fish native to Lake Ontario thus increasing prey fish diversity, improving ecological integrity and resilience, and serving as a model for the reintroduction and management of other native species throughout the Great Lakes.