Identifying statistical interaction networks in marine communities using multivariate time series analysis: An application in the Gulf of Lions

The need for an ecosystem-based approach to fisheries management is widely recognized. Designing ecosystem models for management purposes requires the identification of key interactions and environmental forcing that drive the dynamics of fish stocks. This can be a very challenging task given the complexity of interactions, which determine the evolution of marine ecosystems. To overcome this difficulty, this study proposes a statistical approach based on multivariate time series analysis to identify the main biotic and abiotic interactions using as a case study of a complex and exploited marine ecosystem, the Gulf of Lions (GOL) in the Mediterranean Sea. To do so, first, pairwise Granger causality tests were performed to detect and select the strongest interactions and drivers, then followed by Multivariate Auto-Regressive (MAR) modelling techniques to evaluate the relevance of the selected causal relationships in a multivariate system. The results led to the identification of three statistical interaction networks (SINs) of moderated complexity. The first showed statistical interactions between blackbellied angler (Lophius budegassa), hake (Merluccius merluccius), grey gurnard (Eutrigla gurnardus), and John dory (Zeus faber), as well as the influence of phosphate concentration. The second focused on blackbellied angler, red mullet (Mullus barbatus), anchovy (Engraulis encrasicolus), under the combined influence of demersal trawlers, Sea Surface Temperature (SST) and nitrate concentration. Horned octopus (Eledone cirrhosa), capelan (Trisopterus capelanus), and sardine (Sardina pilchardus) were also investigated under the influence of nitrate concentration. These SINs can serve as a basis to build models of intermediate complexities to describe the dynamics of the main fish stocks of the GOL.