American Bullfrogs (Lithobates catebeianus), native to eastern North America, are among the globe’s most widespread and harmful invaders. They are implicated in native amphibian declines and disease spread on multiple continents. Males advertise during breeding season with a deep, booming vocalization, and scientists are capitalizing on this. A multi-year study uses acoustic monitoring to identify new invasions into previously uninvaded habitats and assess effectiveness of eradication efforts.

Media

Sources/Usage: Public Domain. View Media Details

Bullfrogs have been widely introduced across the western USA, an area that hosts endemic and declining amphibians, including the Oregon spotted frog (OSF; Rana pretiosa). Bullfrogs pose a threat to OSF, and land managers are concerned by increasing observations of bullfrogs in OSF habitats. Researchers with the USGS Forest and Rangeland Ecosystem Science Center (FRESC) partnered with agencies, non-profits, and private landowners to deploy autonomous recording units (ARUs) at 90 wetland sites in 10 hydrographic basins across Oregon and Washington. This broad-scale monitoring network is targeting vulnerable OSF habitats along bullfrog invasion fronts. 

The use of passive monitoring and automated call detection offers a cost effective and efficient way to improve early detection and rapid response to new invasions of bullfrogs. ARUs are deployed at field sites and programmed to record at standard intervals during potential bullfrog breeding season. At the end of the field season, researchers remove the ARUs and download the audio data. As you might imagine, there’s a lot of data – thousands of hours of audio each year. Researchers couldn’t possibly listen to all of it. Enter: machine learning.

Media

Sources/Usage: Public Domain. View Media Details

USGS FRESC has teamed up with USGS Core Science Systems’ (CSS) Advanced Research Computing group to process data using the Tallgrass supercomputer and funding from the Ecosystem Mission Area’s Amphibian Research and Monitoring Initiative (ARMI) and the Biological Threats and Invasive Species Research Program. Researchers use a convolutional neural network (CNN), which is a type of machine learning model, that extracts important acoustic signals from bullfrog call waveforms and learns to recognize bullfrog calls from the soundscape. Yes, researchers are using AI technologies, not too unlike the technologies that inform facial recognition applications. Researchers trained the CNN on thousands of 2.5 second clips of bullfrog calls converted into Mel spectrograms, which are visual representations of the sound on a human-audible scale. The model was tuned and optimized, resulting in a final recognizer that was 96% accurate in predicting bullfrog calls in new audio from the ARUs.

The CNN is also a proof-of-concept for another collaboration between ARMI, CSS, and the Vermont Cooperative Fish and Wildlife Research Unit. Using CSS High Performance Computing data storage and computing resources, ARMI researchers and partners are developing a framework to process large volumes of amphibian call data through AMMonitor, an open-source R package. This project is developing the means to use cloud-based data storage and analysis to facilitate collaboration with organizations across North America – an effort that could expand to citizen-science data collection. 

Work is ongoing, but preliminary results from the CNN suggest regional variation in call patterns (onset, duration, and periodicity), with onset of breeding beginning as early as April in some low elevation sites in the Willamette Valley, OR. Researchers are using model output to assess environmental predictors of peak call activity. Proposed future work includes coupling the acoustic monitoring with a new predictive model to improve the efficiency of Early Detection and Rapid Response efforts that limit bullfrog spread. The rapid advancement of new computing technologies provides opportunities to process massive amounts of data and more efficiently detect invasive species and native species of conservation concern so that managers can act quickly on the ground.