First automatic passive acoustic tool for monitoring two species of procellarides (Pterodroma baraui and Puffinus bailloni) on Reunion Island, Indian Ocean
Introduction
Between 1996 and 2015, the number of threatened species of birds increased by 13.43% (IUCN Red List version 2015.2: Table 1). Policy makers are expected to make decisions to mitigate or manage the threats of climate change and the high rates of species loss (Aide et al., 2013). Unfortunately, they rarely have the information needed to make informed decisions. The highly aggregated distribution of information limits our ability to understand large-scale ecological processes and to properly manage fauna in large areas (Condit, 1995, Gentry, 1993, Porter et al., 2005, Porter et al., 2009, Terborgh et al., 1990, Underwood and Inouye, 2005).
An alternative to physical surveys of seabirds presence is the use of acoustic monitoring (Aide et al., 2013). Acoustic monitoring requires very little time in the field. Autonomous Recording Units (ARU) can be easily deployed and retrieved. They can record animal populations on a very long time (from whole days to several years), in multiple stations across a variety of habitats, facilitating spatial and temporal comparisons of activity (Acevedo et al., 2009, Acevedo and Villanueva-Rivera, 2006, Aide et al., 2013, Hoeke et al., 2009, Lammers et al., 2008, Scott Brandes, 2008, Sueur et al., 2008, Tricas and Boyle, 2009). Acoustic monitoring is especially advantageous for monitoring nocturnal seabirds, due to their conspicuous and nocturnal vocalisations (Oppel et al., 2014, Robb et al., 2008). Colonies of nocturnal burrowing seabirds are noisy places, where vocalisations replace visual displays (Brooke, 1986). Moreover, nocturnal vocalisations have been used as an indicator of general activity (Gaston et al., 1988, Jones et al., 1990) and have been suggested to be a useful indicator of seabirds status and relative abundance (Bradford, 2005, Whittington et al., 1999). Population size assessments of nocturnal burrow-nesting seabirds are logistically challenging. These species are active in colonies only during darkness. And they often nest on remote and hilly islands where manual inspections of breeding burrows are not feasible (Oppel et al., 2014). However, ARU collect an overwhelming amount of data, challenging data analysis (Villanueva-Rivera and Pijanowski, 2012). Researchers regularly develop algorithms to automate species identification from vocalisations of bats, whales, dolphins, insects, amphibians and birds (Acevedo and Villanueva-Rivera, 2006, Anderson et al., 1996, Kogan and Margoliash, 1998).
Barau's petrel (Pterodroma baraui) and tropical shearwater (Puffinus bailloni) are two pelagic seabirds breeding on Reunion Island. Barau's petrel is an endemic species classified as endangered since 2008 and its population is known to decrease (IUCN). Tropical shearwater (P. bailloni) is (1) a subspecies of the Puffinus assimilis lherminieri complex and (2) regionally endemic of the Malagasy region (Austin, 2004, Safford and Hawkins, 2013). Petrels in general, and tropical species in particular, are notoriously difficult to locate and census (Bretagnolle et al., 2000, Day and Cooper, 1995). They both spend daytime offshore to feed and come back to the breeding colony by night. Barau's petrel breeding colonies are accessible only after several hours of walking. The breeding colonies of tropical shearwater on Reunion Island are unreachable since restricted to cliffs (Bretagnolle et al., 2000). Thus, the monitoring of these two species is currently limited by the extreme difficulty of reaching colonies, environmental conditions (wind, rain and waterfall) and timing (darkness, moon phase, season). Pinet et al. (2009) suggested the extinction of Barau's petrel in fewer than 100 years in the absence of cat predation control at breeding colonies. Corre et al. (2002) also highlighted how Barau's petrel fledglings are attracted by urban lights. This phenomenon causes massive and seasonal light-induced mortality. Effective conservation strategies of Barau's petrel and tropical shearwater by the recent National Park of Reunion Island are constrained by the low number of big-scale spatial and temporal data available concerning these two species (Pinet et al., 2009).
For these reasons, here are proposed two automatic detectors of Barau's petrel and tropical shearwater vocalisations (we defined a segment as an uninterrupted sound in temporal and frequency domains) in noisy audio recordings (1) trained with a low number of positive training instances, and (2) whose performances would be the highest possible. To do so, we examined a random forest machine learning algorithm combined with some classical methods of acoustic signal characterisation (Buxton and Jones, 2012, Ericsson, 2009, Fagerlund, 2004, Harma, 2003, Stowell and Plumbley).
Section snippets
Recording system
Autonomous digital recording units (ARU) called Song Meters (Model SM2+, Wildlife acoustics Inc., 3 Hz high-pass filter, 44.1 kHz sampling rate, 16 bits quantification resolution) were used to collect sounds on Reunion Island. Gain was set to 36 dB. The detection range of vocalisations depends on background noise levels, wind speed, topography, vocalisations properties, etc. Song Meters can detect vocalisations at least up to 50 m away under ideal conditions (Buxton and Jones, 2012 and personal
Results and discussion
Mean F1 scores and standard deviation vary depending on the target species, the acoustic features, the number of training and testing instances. Standard deviation values are low. This indicates that mean F1 scores are reliable. Best scores were reached with N1 = 500 for both species and all features. Increasing the number of negative instances used for training from two hundred to five hundred allowed to increase performances of all classifiers. This was a predictable result. Increasing the
Conclusion
Our objective was to propose automatic detectors of Barau's petrel and tropical shearwater vocalisations in noisy audio recordings (1) trained with a low number of positive training instances, and (2) whose performances would be the highest possible. Audio recordings were collected by autonomous recording units equipped with omnidirectional and standard-quality microphones in harsh, windy and wet environments. Signal-to-noise ratios and untargeted sounds were variable and diversified
Acknowledgments
PhD funds of 1st author were provided by Agence De l’Environnement et de la Maîtrise de l’Energie ([email protected]) and by BIOTOPE company (Mr. Buzon, [email protected], R&D Manager).
References (40)
- et al.
Automated classification of bird and amphibian calls using machine learning: a comparison of methods
Ecol. Inform.
(2009) Research in large, long-term tropical forest plots
Trends Ecol. Evol.
(1995)- et al.
A direct lda algorithm for high-dimensional data with application to face recognition
Pattern Recogn.
(2001) - et al.
From the field: using automated digital recording systems as effective tools for the monitoring of birds and amphibians
Wildl. Soc. Bull.
(2006) - et al.
Real-time bioacoustics monitoring and automated species identification
PeerJ
(2013) - et al.
Template based automatic recognition of birdsong syllables from continuous recordings
J. Acoust. Soc. Am.
(1996) A global molecular phylogeny of the small Puffinus shearwaters and implications for systematics of the little-Audubon's shearwater complex
Auk
(2004)Status of Xantus's murrelet and its nesting habitat in Baja California, Mexico
Mar. Ornithol.
(2005)- et al.
Audubon's shearwaters Puffinus lherminieri on Reunion Island, Indian Ocean: behaviour, census, distribution, biometrics and breeding biology
Ibis
(2000) The Vocal Systems of two Nocturnal Burrowing Petrels, The White-Chinned Procellaria aequinoctialis and the Grey P. cinerea
(1986)
Measuring nocturnal seabird activity and status using acoustic recording devices: applications for island restoration
J. Field Ornithol.
Light-induced mortality of petrels: a 4-year study from Réunion Island (Indian Ocean)
Biol. Conserv.
Patterns of Movement of Dark-rumped Petrels and Newell's Shearwaters on Kauai
PLP and RASTA (and MFCC, and inversion) in Matlab, online web resource
Automatic Speech/Music Discrimination in Audio Files
Automatic Recognition of Bird Species by Their Sounds
Monitoring ancient murrelet breeding populations
Colon. Waterbirds
Four Neotropical Rainforests
Automatic identification of bird species based on sinusoidal modeling of syllables
Coral reef ecosystem integrated observing system: in-situ oceanographic observations at the US Pacific islands and atolls
J. Oper. Oceanogr.
Cited by (3)
A review of automatic recognition technology for bird vocalizations in the deep learning era
2023, Ecological InformaticsCitation Excerpt :Thus, bird vocalizations can be used for monitoring bird species and even bird individuals. With the ongoing developments in automatic recording technology and equipment manufacturing capabilities, automatic recording devices with long battery life, large storage capacity, and reasonable price have become widely used in ecological research (Gibb et al., 2019; Dufour et al., 2016; Enari et al., 2019; Kalan et al., 2015). Automatic recording devices can be deployed in the wild to capture field sounds 24 h a day, 7 days a week.
Impact of climate change on migratory birds in asia
2021, Pertanika Journal of Science and Technology