An invasive species model and dataset for bioacoustic monitoring of common brushtail possum

Passive acoustic monitoring (PAM) is a critical tool in the monitoring and conservation of native species but until now its use in the detection of invasive species has been under-utilised. We present the first publicly available dataset of invasive common brushtail possum (Trichosurus vulpecula) vocalisations including 3500 annotated field recording segments. This study presents an automatic classification model designed and fine-tuned to detect the presence/absence of possums, achieving 98.4% test set accuracy and F1 score of 0.983. To our knowledge, this is the first model of its kind applied to the target taxa. We also discuss the development of computational tools in the context of invasive species detection, conservation potential, and critical challenges such as vocalisation frequency and feature sparsity. All data and code are publicly available.