BatSpot: A retrainable neural network for automatic detection and classification of bat echolocation and detection of buzzes and social calls

Bats are a taxonomically and behaviourally diverse group that includes many keystone species, increasingly at risk from habitat loss and human–wildlife conflict. Despite their ecological importance, bats remain understudied not least because their nocturnal behavior and ultrasonic echolocation challenges data collection. Advances in passive acoustic monitoring now enable large-scale datasets, yet data processing to detect and classify vocalisations remains a bottleneck. Existing tools are often commercial, geographically limited, and focus on echolocation search-phase calls. Here, we present BatSpot, a comprehensive software tool with a convolutional neural network at its core. BatSpot detects search-phase calls, feeding buzzes, and social calls, and classifies search-phase calls to species(−complex) level. It includes a graphical user interface that enables users to retrain or transfer-train models and validate performance for specific applications. We evaluated BatSpot against commercial and open-source alternatives under consistent settings and found improved performance (search-phase file-level F1: 0.97 vs 0.96; buzz detector F1: 0.95 vs 0.11). Retraining with only 59 recordings from a new region substantially increased performance of the search-call detector (F1: 0.48 to 0.79), demonstrating strong adaptability. Currently trained on data from Denmark, Germany and Panama, BatSpot enables global application through retraining. By incorporating social call and buzz detection – features largely absent from existing tools – BatSpot provides ecologically relevant insights into mating and foraging activity, supporting habitat identification, sensitivity mapping, and conservation management.