Résumé
Growing fisheries datasets, together with ever-improving records of environmental variability, allow for increasingly useful descriptions of fish-environment relationships. These relationships can provide valuable insights into the behaviour, ecology and habitat preferences of resource species. Historical (1981-2001) purse-seine catches of skipjack (Katsuwonus pelamis) and yellowfin tuna (Thunnus albacares), from French and Spanish fleets operating in the Indian Ocean, were compared to concomitant records of environmental variables using Generalised Additive Models (GAMs). The majority of environmental fields were provided from hindcasts of a state-of-the-art coupled bio-physical ocean model (NEMO-PISCES). A delta approach was used to handle the zero-inflated data, by which presence-absence models were evaluated, and thereafter, catch-per-unit-effort (CPUE) was modelled on condition that catches were positive (present). Results showed important differences between relationships obtained from catches made underneath fish aggregating devices (FADs), compared to those made on free-swimming schools. FADs efficiently congregate tuna from the surrounding environment, thereby seemingly reducing 'catchability' effects and allowing relationships that can be interpreted in terms of tuna abundances or habitat, to come to the fore. Results suggest that deeper thermocline depths translate to greater skipjack and yellowfin tuna abundances. In addition, tuna densities seem to react to a mesozooplankton proxy of prey concentrations, by a Holling type II functional response, which has been assumed in modelling studies but not yet shown empirically from fisheries data. As well as the insights gained into the ecology and behaviour of tuna, and how these interact with the fishery, such tuna-environment relationships are potentially valuable resource management tools. Applied to quasi real-time data, they could map changing tuna habitats and inform decisions on the management of mobile no-take zones in the pelagic realm. Applied to forecast models, they could further aid management decisions by providing predictions of how tuna habitat might change under future climate variability.