Research Article
From calls to conservation: AnuraPAM, a ready-to-use acoustic classifier for Central European anurans
- HTWD University of Applied Sciences, Bioacoustics working group, Pillnitzer Platz 2, 01326 Dresden, Germany
- TUD Dresden University of Technology, Faculty of Environmental Sciences, 01062 Dresden, Germany
- Senckenberg – Leibniz Institution for Biodiversity and Earth System Research, Königsbrücker Landstraße 159, 01109 Dresden, Germany
- TUD Dresden University of Technology, Faculty of Biology, 01062 Dresden, Germany
Corresponding author:
Raffael Ernst (raffael.ernst@senckenberg.de)
Abstract
Freshwater ecosystems are among the most threatened habitats worldwide, and amphibians represent the most endangered vertebrate group globally. Effective conservation strategies therefore require scalable, cost-efficient monitoring methods. Passive acoustic monitoring (PAM), combined with automated call classification, has emerged as a promising approach, yet amphibians remain underrepresented compared to birds. Here, we present AnuraPAM, a flexible, ready-to-use collection of automated call classifier models for the Central European anuran fauna, trained using the deep-learning framework BirdNET. We developed four convolutional neural network-based classification models for 14 Central European amphibian species including one hybrid taxon (Pelophylax kl. esculentus), using different training data strategies. Model performance was evaluated using one-vs-rest receiver operating characteristic (ROC) and precision–recall (PR) curves. Most species achieved high classification performance, with ROC-AUC (area under the curve) values exceeding 0.9 across model variants. However, classification accuracy varied among species, particularly within the Pelophylax complex, where hybridization poses substantial challenges. Increasing the amount of training data did not consistently improve model performance, highlighting the importance of data quality and representativeness. This is supported by the fact that our best-performing model performed better than two freely available classifiers, FrogNET and Perch v2, despite being trained on smaller datasets. Our results demonstrate that automated acoustic identification of Central European anurans is feasible and reliable for most species. The developed classifiers provide a novel tool for large-scale amphibian monitoring and offer considerable potential for application in conservation practice and citizen science projects. All our models are available as ready-to-use classifiers for the application in free audio analysis software such as BirdNET-Analyzer, PAMalyzer and Raven Intelligence.
Keywords
- Amphibia
- Anura
- bioacoustics
- BirdNET
- convolutional neural networks
- passive acoustic monitoring
- conservation
- citizen science
Supplementary material
Cite this article
(2026): From calls to conservation: AnuraPAM, a ready-to-use acoustic classifier for Central European anurans. SALAMANDRA 62(4): 287–297.
HTML full text
The HTML full text is not yet available. Please use the original published PDF.