Communication Dans Un Congrès Année : 2025

Advancing 3D otolith shape analysis for stock identification

Analyse 3D de la forme des otolithes pour l'identification des stocks

Résumé

Accurate spatial stock delineation is crucial for sustainable fisheries management, yet traditional methods often struggle to account for the spatial complexity of fish populations. Otoliths, calcified structures in the inner-ear of fish, serve as natural biological markers that encode valuable information about an individual’s life history, habitat, and population structure. Otolith shape is widely used for stock identification because it reflects genetic and environmental influences, allowing researchers to distinguish between fish populations. However, conventional otolith shape analysis relies primarily on two-dimensional (2D) imaging, which captures only a single plane of these inherently three-dimensional structures. This approach can introduce biases due to orientation inconsistencies and the loss of crucial morphological details. To overcome these limitations, we applied an innovative, three-dimensional (3D) otolith shape analysis to red mullet (Mullus barbatus) stocks in the Mediterranean Sea, comparing it to traditional 2D methods. The Elliptical/Spherical Fourier Descriptors were applied in both 2D and 3D data. Using 316 sagittal otoliths from 16 geographical sub-areas, we used various supervised and unsupervised machine learning techniques to assess stock structure. The results show that 3D shape analysis provides a more accurate and robust stock delineation with both supervised and unsupervised methods, capturing the full spatial complexity of otolith morphology. The estimated stock structure was compared with stock delineations based on expert-defined references and environmental data, showing a closer match when using 3D analysis. Unsupervised classification suggested an optimal division into two major stocks, supervised learning confirmed that 3D descriptors outperform 2D in stock identification. 3D-bases hierarchical clustering has a higher stock determination rate (Adjusted Rand index over 0.6) than those obtained from 2D data. Moreover, 3D-based supervised classifiers, especially Random Forest as the best model fitted the data, achieved an accuracy gain of around 10% compared to the 2D data with the same approaches. By offering a more precise and scalable approach to stock identification, 3D otolith shape analysis presents a powerful tool for improving spatial stock assessments.

Session_F_Andrialovanirina.pdf (5.85 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)
Licence

Dates et versions

hal-05299732 , version 1 (06-10-2025)

Identifiants

  • HAL Id : hal-05299732 , version 1

Citer

Nicolas Andrialovanirina, Émilie Poisson Caillault, Rémi Laffont, Sébastien Couette, Kélig Mahé. Advancing 3D otolith shape analysis for stock identification. ICES Annual Science Conference 2025, ICES CIEM (International Council for the Exploration of the Sea - Conseil International pour l'Exploration de la Mer), Sep 2025, Klaipeda, Lithuania. 12 p. ⟨hal-05299732⟩
84 Consultations
26 Téléchargements

Partager

  • More