All the flight we cannot see: Using passive acoustic monitoring to track the world's largest bumble bee, Bombus dahlbomii

Abstract Bumble bees ( Bombus spp.) are declining worldwide, creating an urgent need for rapid and non‐lethal sampling of their distributions and abundances. The endangered Patagonian bumble bee, Bombus dahlbomii (Hymenoptera: Apidae) Guérin‐Meneville, 1835, threatened by invasive congeners, exemplifies this trend. We deployed passive acoustic monitoring (PAM) to survey geographically distant populations of B. dahlbomii in Chile and Argentina. PAM stations in flowering stands recorded flight buzzes of B. dahlbomii and invasive species, Bombus terrestris (Hymenoptera: Apidae) Linnaeus, 1758 and Bombus ruderatus (Hymenoptera: Apidae) Fabricius, 1775. Recordings were paired with visual observations for ground‐truthing. Flight buzzes of native and invasive congeners exhibit unique harmonic structures that differ in their fundamental frequency. We used a convolutional neural network (CNN) model to identify bumble bee buzzes in soundscapes, and heuristic classification to assign these buzzes to either native B. dahlbomii or invasive congeners. Both CNN and heuristic models exhibited accuracy ≥95%. However, the error rate for the CNN algorithm increased by ~20% when data from the test site were omitted in training, as would be the case for sampling a novel location. Anthropogenic background noise reduced the correlation between CNN output and observer sightings and increased errors in recognizing B. dahlbomii buzzes. Overall, the use of PAM shows promise in identifying drivers of the decline of B. dahlbomii and particularly the role of the invasive congeners.