- Studies that confirm the presence of particular species are needed to understand and conserve them appropriately, but such studies are often costly, time-consuming and thus few and far between.
- Researchers from the University of Puerto Rico are deploying monitoring stations in the field, which record and process animal sounds to automate the collection and classification of audio data to determine species presence and compare animal communities.
- While the technology is limited to animals that call and requires some human expertise in identifying these sounds, it can provide detailed, long-term data for monitoring animal communities and informing wildlife management strategies
What is living where?
We collect weather data at sites all over the globe, so why not collect species data at those sites as well? This is the question that Mitch Aide and colleagues at the University of Puerto Rico (UPR) are asking.
Despite an explosion of research activity across the world, however, our understanding of which animals live where is still very limited. Studies that confirm the presence of particular species are often costly, time-consuming and thus few and far between.
Aide and his fellow UPR researchers want to fill the gap in our knowledge of animal communities by applying acoustic monitoring technology to record the daily calls and other sounds animals make everywhere. The team captures these calls using automated acoustic monitoring stations.
Aide and his colleagues began a Kickstarter campaign–ending this Thursday, October 6—to wire up research stations, protected areas, ecolodges, city parks, and other nature tourism centers across the globe with acoustic monitoring devices. By doing so, the researchers hope to address what he considers the “urgent need to increase the temporal and spatial coverage of ecological data collection.”
Why collect acoustic data?
“Acoustic monitoring can help us greatly improve our ability to monitor population change in thousands of species, but the ecological and conservation community has been slow in incorporating this technology into their monitoring and research projects,” Aide explained.
Over the past 10 years, his team has developed the Automated Remote Biodiversity Monitoring Network (ARBIMON) and shown how portable, inexpensive monitoring stations can collect a continuous stream of acoustic biodiversity data. Their aim with ARBIMON is to mimic the temperature and rainfall gauges that are standard equipment at most parks and field stations with gauges that can mechanically monitor and record the sounds of an animal community. Similar to camera traps that record the presence of animals based on the visual cue of their bodies passing by, automated acoustic monitoring devices would record animals’ presence through their sounds.
Identifying the species vocalizing at different points across a reserve or landscape would help scientists detect the presence of species that are endangered or previously unknown at the sites and would improve our knowledge of species’ distributions, relative abundances and community composition. Comparing the variety of sounds—“soundscapes”—of 10 tropical forest sites, for example, the UPR research team found that stations at sites with more mammal, amphibian and bird species also had a higher proportion of acoustic activity than sites with fewer species.
Long-term monitoring and identification of sounds could help reveal major changes in abundance or composition of species at a site. The UPR team has monitored soundscapes to assess changes in animal communities facing varying levels of human disturbance and found changes in bird and frog communities near illegal small-scale gold mines in southeastern Peru, compared to animal communities at abandoned mines and adjacent continuous forest.
More commonly, scientists have deployed automated acoustic monitoring gauges to detect the presence of target (often endangered) species such as whales nocturnal birds and bats and other animals difficult to survey with traditional techniques.
In a recent study, Marconi Campos-Cerqueira and Aide used acoustic sensors and models that automatically detect specific species’ vocalizations from the recordings to map the distribution of the elfin woods warbler, a rare, elusive bird endemic to upland forests in Puerto Rico. They discovered that the bird depends on Palo Colorado forest within an elevational band 600-900 meters above sea level, a result helpful for siting future management work.
Recording in the field
The permanent acoustic stations automatically record sounds 24/7 and store them in a web-based platform that enables users to manage, process and analyze acoustic data sets. Each remote acoustic monitoring station includes an Android phone for recording and communications in a waterproof case, a microphone and, in the permanent acoustic stations, a small solar panel. A permanent monitoring station, including 50,000 minutes of recordings on ARBIMON II and the solar panel costs US $2000 (excluding cellular or wifi connection and solar panel mounting hardware). Portable recorders cost roughly US$300.
The project team says these have held up well in hot rainy conditions, though the low-cost microphone may need to be replaced every 6 to 12 months to maintain high quality recordings. In forested areas, the station would also require a tower that extends above the canopy to ensure the solar panel faces full sun. To ensure that vegetation or cloud cover doesn’t cut solar power generation, the project developed an application that alerts the project team if a station is malfunctioning.
At sites with mobile phone coverage, each acoustic monitoring station can use the network to move recordings in real-time from the field to the project’s online analysis platform. At remote sites without cellular coverage, a field base station can store recordings on a memory card and then upload the recordings to the system every week or month.
Project teams can adjust the duration and frequency of recordings to fit research objectives. The UPR team typically sets each acoustic station to record for one minute every 10 minutes and send the sound data to a field base station, which relays the recordings in real-time to the project server.
The server processes and uploads the recordings to the online platform, where users can view, listen to and annotate recordings, displayed as spectrograms. Users can also upload, process and store the collected sound data in the cloud and share them with colleagues.
Aide considers each of these recordings a permanent record—“the equivalent of a museum specimen”—that allows researchers to both revisit an unidentified recording when new reference data are available and contribute to long-term species monitoring.
Users interested in detecting a particular species can create, test and validate species-specific identification models. The data resulting from this process can then be used, for example, in occupancy models that take into account the imperfect detectability of most species in the wild, to improve our understanding of animal spatial distribution.
Learning curve for humans and machines
While passive acoustic monitoring devices have existed for some time and can produce reams of sound data, they can, as Campos-Cerqueira and Aide state, “require substantial expert effort to extract useful information from the recordings.”
The amount of data a constantly recording device collects can even become overwhelming, requiring researchers to develop algorithms and models to automate sound identification, as has been done for numerous species of amphibians, bats, birds, cetaceans (whales and dolphins) and insects. In fact, each type of call for a given species (greeting, mating, warning) has a different signature and so requires a different model.
An algorithm analyzes the matrix of sounds to distinguish calls of interest from background noise, and the models help minimize the proportion of recordings that need to be validated manually by an expert. Validation requires the user to create a “training set” of sounds known to be from the target species (here is one of several tutorials). Validation also requires a reference data set consisting of recordings and spectrograms for which an expert has identified species presences and absences. These allow an expert to verify the models by correcting both false negatives, where the software misses a recording (due mainly to a faint call), and false positives, where the software incorrectly signals the animal is present (due mainly to misattributing a call).
One limitation to surveying using acoustic data is that an animal must call within the detection range of the acoustic recorders. Another is the need for reference recordings of the species of interest and a human expert to initially distinguish new vocalizations to provide training sets and validation data.
Regardless, taking action to conserve rare species or communities in the face of human activity and climate change requires knowing what animals live where. As Aide and colleagues explain in their 2013 paper, “these web-based tools greatly simplify the process of extracting useful results for researchers and managers from the raw data (i.e., recordings), which should help the users to improve and expand their ecological monitoring programs.”
References
Aide, T. M. (2016) Establishing a Global Acoustic Monitoring Network. FrogLog 24(1): 41–42.
Aide, T. M., C. Corrada-Bravo, M. Campos-Cerqueira, C. Milan, G. Vega and R. Alvarez (2013) Real-time bioacoustics monitoring and automated species identification. PeerJ 1:e103; DOI 10.7717/peerj.103.
Alvarez-Berrios, N., M. Campos-Cerqueira, A. Hernandez-Serna, A. Delgado, F. Roman-Dañobeytia, and T. M. Aide (2016) Impacts of small-scale gold mining on birds and anurans near the Tambopata Natural Reserve, Peru, assessed using passive acoustic monitoring. Tropical Conservation Science 9 (2): 832–851.
Campos-Cerqueira, M. and Aide, T. M. (2016) Improving distribution data of threatened species by combining acoustic monitoring and occupancy modelling. Methods in Ecology and Evolution. doi: 10.1111/2041-210X.12599.