TABMON – real-time acoustic biodiversity monitoring across Europe

Benjamin Cretois, Carolyn Rosten, Julia Wiel, Cynthia Barile, Ben McEwen, Corentin Bernard, Michiel Boom, Gerard Cabau Bota, Lluís Brotons, Eva Serrano Davies, Hervé Glotin, W. Daniel Kissling, Ricard Marxer, Cristian Pérez-Granados, Dan Stowell, Dani Villero, Jelle S. van Zweden, Sarab S. Sethi2026 · DOI

TL;DR

The TABMON project has deployed a novel, real-time acoustic monitoring network across Europe, using 97 autonomous sensors to continuously collect bird vocalizations. This initiative addresses critical gaps in biodiversity data by providing a standardized, scalable, and automated system for monitoring avian populations across national borders, paving the way for improved biodiversity assessment and reporting.

What Problem Does This Paper Address?

Traditional ecological surveys are often fragmented, expensive, and limited in scope, leading to significant knowledge gaps about biodiversity. While Passive Acoustic Monitoring (PAM) offers a promising automated solution, existing networks struggle with scalability, standardization, and data management, hindering their ability to provide truly novel insights. This paper tackles these challenges by presenting TABMON, a large-scale, transnational network designed to overcome these limitations and deliver continuous, real-time biodiversity data across Europe.

Methods

TABMON employs a sophisticated data pipeline that begins with autonomous acoustic sensors (Bugg devices) recording 5-minute audio files continuously at a 44.1 kHz sampling rate. These recordings are compressed using MP3 VBR0 and transmitted in near real-time via 4/5G mobile internet to Google Cloud. From there, data is transferred to a Norwegian Infrastructure for Research Data (NIRD) server for partner access and then to a dedicated LIS server at the University of Toulon for long-term archival and large-scale analysis.

The core of the analysis relies on the BirdNET model, an AI tool capable of identifying over 6500 bird species. This model is executed via the AvesEcho service, optimized for parallel processing. To improve accuracy and address potential biases, the pipeline incorporates an active learning strategy. This involves experts annotating a subset of BirdNET detections, which are then used to retrain and refine the models, focusing on clips that provide the most information gain. This iterative process aims to enhance model performance, especially for European species where BirdNET might be less accurate, and to create valuable expert-annotated datasets.

Datasets

The primary dataset introduced by this paper is the TABMON dataset, comprising continuous audio recordings from 97 autonomous sensors deployed across Norway, the Netherlands, France, and Spain. This dataset is collected over a two-year period, covering two complete annual migration cycles. While the raw audio data is not shared due to privacy considerations, the project commits to publicly sharing BirdNET detections and expert annotations. The paper also references the EBBA2 occurrence map for filtering BirdNET predictions based on known breeding ranges.

Key Insights and Ideas

  • Scalability is Key: TABMON demonstrates that it's possible to deploy and manage a large-scale, transnational acoustic monitoring network, overcoming previous limitations in scale and standardization.
  • Real-time is Achievable: The system's ability to upload and process data in near real-time allows for immediate quality control and adaptive sampling, moving beyond traditional batch processing.
  • AI + Human Expertise = Better Models: The integration of an active learning pipeline, combining automated AI analysis with expert annotation, is crucial for improving model accuracy and creating valuable, validated datasets.
  • A Blueprint for the Future: TABMON provides a methodological template and a practical blueprint for future large-scale, multimodal biodiversity observatories, integrating acoustic data with other sensor types.
  • Bridging the Gap: The project effectively bridges the gap between conceptual frameworks for continental biodiversity monitoring and their technical realization, addressing long-standing fragmentation issues.

Key Results

  • Network Deployment: A standardized network of 97 autonomous sensors was successfully deployed across Norway, the Netherlands, France, and Spain, spanning a latitudinal gradient of 3,649 km.
  • Data Ingestion and Analysis: A robust data pipeline was established for near real-time audio upload, storage, and automated analysis using the BirdNET model.
  • Active Learning Implementation: An active learning pipeline was implemented to iteratively refine AI models through expert annotation, aiming to improve accuracy and create valuable labeled datasets.
  • Dashboard for Transparency: A publicly accessible dashboard (https://tabmon.nina.no) was developed to visualize network status, data availability, and sensor deployment information, fostering transparency and collaboration.

Limitations and Future Work

The authors acknowledge several limitations and suggest future directions:

  • Raw Audio Privacy: Due to data privacy concerns, raw audio data cannot be shared publicly. However, BirdNET detections and expert annotations will be made available.
  • Connectivity Challenges: In more populated southern areas, connectivity and equipment interference presented challenges. The authors suggest a flexible monitoring architecture that combines networked, solar-powered sensors with lower-cost, periodically serviced units.
  • Manual Annotation Bottleneck: Manual data annotation remains a limiting factor, even with active learning. Future work will focus on optimizing annotation strategies and exploring further improvements in AI models.
  • Multimodal Integration: The TABMON infrastructure is designed to be expandable and could be coupled with other sensor types (e.g., automated camera traps, eDNA samplers) to create multimodal networks for capturing a wider range of ecological processes and supporting new Essential Biodiversity Variables (EBVs).
  • Integration with Existing Platforms: Future work will explore integrating TABMON outputs with existing European initiatives and platforms for enhanced biodiversity reporting and forecasting.

Summary updated

Abstract

1. Ecological surveys are often fragmented, costly, and limited in scale, leading to large and long-standing knowledge gaps which threaten our ability to properly safeguard biodiversity. 2. Passive Acoustic Monitoring (PAM) has promised to deliver automated biodiversity monitoring, but networks are rarely deployed on scales that can offer truly novel insights due to scalability and standardisation challenges around collecting, managing, analysing, and sharing data. 3. Here we present the Transnational Acoustic Biodiversity Monitoring Network (TABMON), a standardised deployment of 97 autonomous sensors across Norway, Netherlands, France and Spain along a continental bird migration route. Audio is recorded continuously, uploaded in near real-time, and analysed with expert-validated artificial intelligence models to deliver Essential Biodiversity Variables compatible with established European monitoring efforts. 4. TABMON offers a blueprint for ambitious deployments of autonomous sensors that traverse national boundaries to transform biodiversity assessment and reporting globally.
BibTeX
@misc{cretois2026tabmon,
  author = {Benjamin Cretois and Carolyn Rosten and Julia Wiel and Cynthia Barile and Ben McEwen and Corentin Bernard and Michiel Boom and Gerard Cabau Bota and Lluís Brotons and Eva Serrano Davies and Hervé Glotin and W. Daniel Kissling and Ricard Marxer and Cristian Pérez-Granados and Dan Stowell and Dani Villero and Jelle S. van Zweden and Sarab S. Sethi},
  title = {{TABMON – real-time acoustic biodiversity monitoring across Europe}},
  year = {2026},
  doi = {10.32942/x2236j},
}
Figure