A machine learning model and biometric transformations to facilitate European oyster monitoring

• Published in: Aquatic Conservation: Marine and Freshwater Ecosystems
• Main Authors: Pineda‐Metz, Santiago E. A., Merk, Verena, Pogoda, Bernadette
• Other Authors: Bundesamt für Naturschutz
• Format: Article in Journal/Newspaper
• Language: English
• Published: Wiley 2023
• Subjects: North East Atlantic
• Online Access: http://dx.doi.org/10.1002/aqc.3912; https://onlinelibrary.wiley.com/doi/pdf/10.1002/aqc.3912; https://onlinelibrary.wiley.com/doi/full-xml/10.1002/aqc.3912

Abstract Ecosystem monitoring, especially in the context of marine conservation and management requires abundance and biomass metrics, condition indices, and measures of ecosystem services of key species, all of which can be calculated using biometric transformation factors. Following ecosystem restoration measures in the North Sea and north‐east Atlantic waters, European oyster ( Ostrea edulis ) restoration and its monitoring have substantially increased over the past decade. Restoration activities are implemented by diverse approaches and practitioners ranging from governmental conservation agencies, research institutions and non‐governmental institutions to regional groups, including citizen science projects. Thus, tools for facilitating data acquisition and estimation with non‐destructive techniques can support monitoring quantitatively and qualitatively. Weight‐to‐weight transformation factors for calculating dry weight of O. edulis from wet weight measurements are presented. Another important tool is the estimation of weight only from size measurements. The classical approach to achieve these transformation factors is the construction of allometric models, which, however, can greatly vary among regions and between years, making them extremely location/season specific. Alternative and more flexible models constructed using random forests are proposed. This algorithm is a machine learning technique that is increasingly used in ecology, and has been proven to outperform other predictive models. From biometric variable measurements of 1,401 O. edulis individuals, allometric models were used to estimate total, shell and body wet weights, and compare them with 15 random forest models. In general, the random forest models outperformed the allometric ones, with lower error when estimating weight. The developed random forest models can thus provide a tool for facilitating oyster restoration monitoring by increasing data acquisition without the need of sacrificing European oyster individuals. Their improvement can ...