Soundscapes & deep learning enable tracking biodiversity recovery in tropical forests

Three members of RFCx’s science team are co-authors on a recent Nature Communications study examining how acoustic monitoring and AI can track forest restoration in the Choco region of Ecuador. The study involved deploying passive acoustic recorders across a restoration gradient, from active pastures and cacao plantations to abandoned pastures and plantations being naturally regenerated and finally to undisturbed forests. The authors combined expert identification of animal calls along with two automated methods, including one that utilized deep learning AI models. 

Results from each of the automated methods showed strong correlations with the restoration gradient. Species diversity was lowest in actively-used lands and highest in undisturbed forests. Sound-based species composition, rather than species richness, served as an accurate indicator of tropical forest recovery. Regenerating plantations recovered vocal species diversity faster than regenerating pastures, but then converged during later restoration.  Importantly, these sound-based results also correlated with insect diversity data obtained through DNA metabarcoding; this shows that vocal species metrics can reflect broader ecosystem-wide trends as well. 

The study demonstrates that automated ecoacoustic monitoring can be used to track forest recovery, even beyond vocalizing vertebrates, suggesting its broad use to assess restoration outcomes. Read the full paper here!