Assessing causal links in fish stock–recruitment relationships
|Author(s)||Pierre Maud1, Rouyer Tristan2, Bonhommeau Sylvain3, Fromentin Jean-Marc2|
|Affiliation(s)||1 : Irstea, UR EABX, 50 Ave Verdun, F-33612 Cestas, France.
2 : Ifremer Stn Sete, UMR MARBEC, Ave Jean Monnet,CS 30171, F-34203 Sete, France.
3 : IFREMER, Delegat Ocean Indien, Rue Jean Bertho,BP 6097822, Le Port, La Reunion, France.
|Source||Ices Journal Of Marine Science (1054-3139) (Oxford Univ Press), 2018-05 , Vol. 75 , N. 3 , P. 903-911|
|WOS© Times Cited||11|
|Keyword(s)||causality, forecasting models, marine|
Understanding whether recruitment fluctuations in fish stock arise from stochastic forcing (e.g. environmental variations) rather than deterministic forces (e.g. intrinsic dynamics) is a long standing question with important applied consequences for fisheries ecology. In particular, the relationship between recruitment, spawning stock biomass and environmental factors is still poorly understood, even though this aspect is crucial for fisheries management. Fisheries data are often short, but arise from complex dynamical systems with a high degree of stochastic forcing, which are difficult to capture through classic modelling approaches. In the present study, recent statistical approaches based on the approximation of the attractors of dynamical systems are applied on a large dataset of time series to assess (i) the directionality of potential causal relationships between recruitment and spawning stock biomass and potential influence of sea-surface temperature on recruitment and (ii) their performance to forecast recruitment. Our study shows that (i) whereas spawning stock biomass and sea surface temperature influence the recruitment to a lesser extent, recruitment causes also parental stock size and (ii) that non-linear forecasting methods performed well for the short-term predictions of recruitment time series. Our results underline that the complex and stochastic nature of the processes characterizing recruitment are unlikely to be captured by classical stock–recruitment relationships, but that non-linear forecasting methods provide interesting perspectives in that respect.