A pipeline for identification of bird and frog species in tropical soundscape recordings using a convolutional neural network

LeBien, Jack; Zhong, Ming; Campos‐Cerqueira, Marconi; Velev, Julian P.; Dodhia, Rahul; Ferres, Juan Lavista; Aide, T. Mitchell

doi:10.1016/j.ecoinf.2020.101113

Cited by 100 publications

(83 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Common sources of false positives for the Douglas' squirrel rattle call were wind, insects, and anuran calls. In spite of some noted issues, the performance of our CNN was broadly comparable to that of another recent CNN which achieved precision ranging from 0.13 to 1.00 and recall ranging from 0.25 to 1.00 at a detection threshold of 0.99 for 24 classes of avian and anuran vocalizations from Puerto Rico (LeBien et al 2020).…”

Section: Discussionsupporting

confidence: 59%

A Convolutional Neural Network and R-Shiny App for Automated Identification and Classification of Animal Sounds

Ruff

Lesmeister

Appel

et al. 2020

Preprint

View full text Add to dashboard Cite

The use of passive acoustic monitoring in wildlife ecology has increased dramatically in recent years as researchers take advantage of improvements in automated recording units and associated technologies. These technologies have allowed researchers to collect large quantities of acoustic data which must then be processed to extract meaningful information, e.g. target species detections. A persistent issue in acoustic monitoring is the challenge of processing these data most efficiently to automate the detection of species of interest, and deep learning has emerged as a powerful approach to achieve these objectives. Here we report on the development and use of a deep convolutional neural network for the automated detection of 14 forest-adapted species by classifying spectrogram images generated from short audio clips. The neural network has improved performance compared to models previously developed for some of the target classes. Our neural network performed well for most species and at least satisfactory for others. To improve portability and usability by field biologists, we developed a graphical interface for the neural network that can be run through RStudio using the Shiny package, creating a highly portable solution to efficiently process audio data closer to the point of collection and with minimal delays using consumer-grade computers.

show abstract

Section: Discussionsupporting

confidence: 59%

A Convolutional Neural Network and R-Shiny App for Automated Identification and Classification of Animal Sounds

Ruff

Lesmeister

Appel

et al. 2020

Preprint

View full text Add to dashboard Cite

show abstract

“…Acoustic monitors can be an effective tool for sampling bird communities and may be particularly effective and efficient for sampling loud species with distinctive songs (e.g., Lipaugus vociferans ; Ulloa et al, 2016 ). Further, when studies focus on a select set of species with distinctive vocalizations, automatic detection software allows processing of many hours of recordings in a much shorter period than needed to manually listen to recordings ( Acevedo et al, 2009 ; Aide et al, 2013 ; Ulloa et al, 2016 ; LeBien et al, 2020 ). Nonetheless, results from monitors also may provide a distinctly different perspective on overall community composition when compared to direct observations.…”

Section: Discussionmentioning

confidence: 99%

Acoustic monitors and direct observations provide similar but distinct perspectives on bird assemblages in a lowland forest of eastern Ecuador

Blake

2021

PeerJ

View full text Add to dashboard Cite

Bird communities in lowland Neotropical forests exhibit temporal and spatial variation in species composition and abundance at multiple scales. Detecting and explaining such variation requires adequate methods for sampling those bird communities but counting birds in highly diverse lowland forests of the Neotropics can be particularly challenging. Point counts are one of the most frequently used methods for counting birds in tropical forests but inter- and intra-observer variability in detecting and identifying sounds may cause problems. Acoustic monitors (passive acoustic monitors; autonomous recording units) provide an alternative and potentially effective method to sample bird communities by acting, in effect, as “point counts”, recording vocalizations at a given point for a set time. I used acoustic monitors to examine patterns of species richness, spatial distribution, and community composition of birds in a lowland forest in eastern Ecuador, one of the most diverse regions on earth. I deployed monitors at 25 locations, each separated by at least 200 m, on each of two 100-ha plots (Harpia, Puma) at Tiputini Biodiversity Station during January–February, 2013–2017. Monitors were set to record for 10 min followed by a 5-min break, from 0545 h to 0810 h (10 recording periods/morning). Recordings were later reviewed to identify species; no attempt was made to distinguish individuals or to estimate distance. Results were compared with contemporaneous direct observations along transects on the same plots. A total of 214 species were identified from recordings on both plots, combined, with slightly more on Harpia (208) than on Puma (188). Number per year ranged from 142 on Harpia in 2016 to 161 on Puma in 2015. Number per point was ~45 with an overall range of 29–68. Number of species detected in recordings was similar to but somewhat less than the number recorded during direct observations. Number of species recorded increased rapidly from the first period (0545–0555 h) to the third (0615–0625 h) but showed little subsequent change. Most species were recorded at relatively few points; the four most widely distributed species were the same on both plots (Patagioenas plumbea, Xiphorhynchus guttatus, Capito aurita, Ramphastos tucanus), all of which are relatively loud canopy or subcanopy species. Ordinations based on species composition illustrated differences between plots based on both recordings and direct observations; similarly, patterns of species composition differed between methods. Acoustic monitors can be an effective tool for sampling bird communities and may be particularly effective and efficient for sampling loud species with distinctive songs. Nonetheless, results from monitors may provide different perspectives on species composition when compared to direct observations. Which method is preferred likely will depend on the specific objectives of individual studies.

show abstract

“…All bird templates, except for Megascops nudipes (Puerto Rican screech owl), were applied to the diurnal playlist, and all frog templates and M. nudipes templates were applied to the nocturnal playlist. The classifications were based on a pattern matching procedure (LeBien et al 2020) that searches through each recording and detects regions that have a correlation with the template above a given threshold. For these analyses, we used a correlation threshold of 0.1.…”

Section: Methodsmentioning

confidence: 99%

Impacts of a drought and hurricane on tropical bird and frog distributions

Campos‐Cerqueira

Aide

2021

Ecosphere

Self Cite

View full text Add to dashboard Cite

During the last few decades, much attention has focused on how global change is affecting the environment and species distributions. Land-use change is still the major cause of species declines worldwide, but changes in species distributions have been documented even in pristine and protected areas. Here, we document the distribution dynamics of 26 species of frogs and birds within a Caribbean protected area between 2015 and 2019. Specifically, we document species occupancy and detectability in 59 sites along three elevational transects within the El Yunque National Forest in Puerto Rico. Species were sampled using acoustic recorders, species identification algorithms, and post-classification validation. The study period included a severe drought (2015-2016) and a category 4 hurricane (2017). The distribution of most frog species did not change significantly during the study period. In contrast, the distributions of the bird species contracted between the 2015 and 2016 pre-hurricane surveys. This contraction coincides with a severe drought that peaked after the 2015 survey. The response of bird species after H. Maria was heterogeneous. Our results suggest that for many species, particularly birds, the 2015-2016 drought in Puerto Rico had a greater negative effect than H. Maria. The difference in the response of the bird and frog communities is likely related to their abundance at the site level, their ability to disperse, and temporal patterns of reproduction. If a site is occupied by a frog species, it could easily include 100s of individuals, and even if condition worsens, over a relatively short period the species will continue to occupy the site. In contrast, a site may only be occupied by one or a few individuals of birds. Birds have also higher dispersal abilities than frogs. Furthermore, Puerto Rican frogs reproduce year-round, whereas birds concentrate their reproduction during a few months in the wet season. Future climate scenarios predict a decrease in total precipitation and an increase in droughts for the region, which emphasizes the need to consider the diversity of changes that will be associated with future climate change.

show abstract

A pipeline for identification of bird and frog species in tropical soundscape recordings using a convolutional neural network

Cited by 100 publications

References 12 publications

A Convolutional Neural Network and R-Shiny App for Automated Identification and Classification of Animal Sounds

A Convolutional Neural Network and R-Shiny App for Automated Identification and Classification of Animal Sounds

Acoustic monitors and direct observations provide similar but distinct perspectives on bird assemblages in a lowland forest of eastern Ecuador

Impacts of a drought and hurricane on tropical bird and frog distributions

Contact Info

Product

Resources

About