BIOINDIC-IA

Deep learning for automatic image-based biomonitoring of aquatic ecosystems

Inspiration

The intensification of land use and exploitation of aquatic resources has led to a rapid increase in anthropogenic pressures on ecosystems, altering biological communities and impairing ecosystem functioning across spatial and temporal scales. These changes threaten the delivery of essential ecosystem services and call for more advanced and reliable ecological diagnostic tools.

Current biomonitoring approaches, implemented under regulatory frameworks such as the Water Framework Directive and the Marine Strategy Framework Directive, rely largely on taxonomic and biotic indices derived from indicator species like benthic diatoms (microscopic aquatic algae) and benthic foraminifera (marine amoeba-like organisms with shells). While effective, these methods are time-consuming, expertise-dependent, and sometimes limited in their ability to diagnose specific environmental pressures.

Recent advances highlight the value of trait-based approaches, which consider morphological, physiological, and behavioural characteristics of these organisms to better identify stressors such as eutrophication or contamination. However, collecting such data manually remains a major bottleneck, emphasizing the need for automated, scalable solutions.

Innovation

BIOINDIC-IA is an international collaboration (PIs: Martin Laviale and Carlos Wetzel) funded by the Agence National de la Recherche (ANR) and by the Luxembourg National Research Fund (FNR) through the INTER programme (Grant reference: 18964279). This project leverages artificial intelligence and deep learning to transform aquatic biomonitoring into a high-throughput, automated, and trait-informed process. Using convolutional neural networks (CNNs), the project aims to simultaneously:

Perform species-level taxonomic classification
Quantify key morphological traits directly from microscope images

The approach is applied to two complementary biological models:

Freshwater benthic diatoms, widely used in river monitoring
Marine benthic foraminifera, an emerging indicator group for coastal ecosystems

Detection of diatoms using a pre-trained YOLO model showing predicted labels and associated probabilities

A major innovation lies in the development of an end-to-end automated pipeline, including:

Automated image acquisition workflows
High-quality, curated, and labelled image datasets
Machine learning (ML) models trained for both taxonomy and trait extraction

By integrating taxonomy and traits into multi-metric, ML-based bioindicators, the project moves beyond traditional methods and enables more precise identification of environmental pressures affecting aquatic ecosystems.

Impact

The project will significantly improve the efficiency, scalability, and diagnostic power of aquatic biomonitoring. By automating key steps such as image acquisition, species identification, and trait measurement, it reduces reliance on time-intensive manual analyses while maintaining high accuracy.

Key expected impacts include:

Enabling faster and standardized ecological assessments across freshwater and marine systems
Enhancing the detection of specific environmental stressors through trait-based indicators
Supporting regulatory monitoring under the Water Framework Directive and Marine Strategy Framework Directive
Providing scalable tools for large-scale, high-resolution environmental monitoring

More broadly, the project contributes to the transition toward AI-driven environmental management, where automated image analysis and ecological expertise combine to deliver actionable insights for ecosystem conservation and restoration in a rapidly changing world.