Mapping risk of common dolphin (Delphinus delphis) bycatch in the Bay of Biscay

The strandings of common dolphin (Delphinus delphis) have been significantly increasing along the French coast since 2016, putting at risk the whole population. These strandings occur especially in winter, despite the temporal variability remains high. Therefore, it is essential to understand the distribution of this species on a fine scale to propose solutions to reduce accidental captures. The suite of models SEAPODYM was used to map the spatio-temporal distribution of dolphins and their preys.

Highly mobile, dolphins respond to environmental variations and move towards conditions that best match their physiology and the presence of their prey. Using biophysical data (CMEMS) as an input of the SEAPODYM-LMTL (Low and Mid trophic Levels) model, we have modelled the spatio-temporal distribution of the dolphins’ preferred prey: sardines (Sardina pilchardus), anchovies (Engraulis encrasicolus), mackerel (Scomber scombrus), horse mackerel (Trachurus trachurus) as well as a mesopelagic fish, Kroyeri’s lanterfish (Notoscopelus kroyeri). Then we modelled dolphin’s habitat with SEAPODYM-MASS (Migratory Aged Structured Stock) model, function of the presence of the preys and their accessibility. The density energy of the prey was taken into account, as it varies across seasons, to compute the preferred habitat of dolphins.

Besides modelling the spatial distribution of the common dolphin in the Bay of Biscay, it is crucial to understand the distribution of the fleet's effort using fishing gears that are implicated in the strandings to identify the areas where dolphins are the most at risk to interact with fishing gears at concern. To achieve this goal, we trained machine learning algorithms to recognize fishing activities along trajectories of vessels using those gears. For each fishing occurrence identified by our algorithm, environmental conditions like the type of sediment, sea temperatures at various depth, currents or phytoplankton concentration are extracted and compared to the conditions over the entire Bay of Biscay through a Species Distribution Model (SDM). During the training process, SDM performance are evaluated by cross-validation. Once trained, the SDM can predict the probability of fishing activity across the entire region given the current environmental conditions. The output of the SDM being maps of probabilities it can overlapped with the dolphin habitat maps to characterize in near real time the risk of interaction between dolphin populations and fishing activities.

Key words: mechanistic habitat modelling, small pelagic, mesopelagic fish, energy density, fishing effort distribution.