Understanding animal and human activity through audio

Spectrograms of five soundscape classes. These spectrograms show time (x-axis) relative to sound frequency (y-axis) and sound amplitude (color) for 2-second sound recordings. Patterns unique to each soundscape class are shown by the black hashed boxes.

The Soundscapes to Landscapes (S2L) project focuses on using acoustic data collected across the hills and forests of Sonoma County, California, and linked with remote sensing data to better monitor biodiversity. The GEODE lab’s Colin Quinn, Patrick Burns, and Scott Goetz along with five other non-GEODE collaborators published work where they applied deep learning methods to classify animal and human activity patterns in 500,000 minutes worth of acoustic data, the equivalent of recording continuously for just under one year, from Sonoma County, California. The work demonstrates how a large amount of acoustic data can be classified into ecologically meaningful categories of activity and how patterns in acoustic activity relate to landscape characteristics such as land use and proximity to roads. These data will be used to map animal activity, human presence, and naturally quiet landscapes across Sonoma County using GEDI and other NASA remote sensing products. In addition to understanding broad patterns in sound, the S2L project focuses on classifying and mapping common bird species from the acoustic dataset with the help of countless citizen scientists. The potential impact of S2L work was described in a recent article in Bay Nature.