Spatiotemporal models highlight influence of oceanographic conditions on common dolphin bycatch risk in the Bay of Biscay

Gilbert Lola, Rouby Etienne, Tew-Kai Emilie, Spitz Jerome, Quilfen Victor, Peltier Helene, Authier Matthieu

What is this web page for ?

This Shiny App complements the article Spatiotemporal models highlight influence of oceanographic conditions on common dolphin bycatch risk in the Bay of Biscay 1 (Gilbert et al. 2021 ) in order to facilitate data and results visualization .

This work was a collaboration between Pelagis (French Observatory of marine mammals, in La Rochelle ) and the Shom (French Naval Hydrographic Service, in Brest ).

Context : For the past decades, winters in the Bay of Biscay (Fig.1) have been characterized by an increase of common dolphin strandings showing bycatch evidence. The goal of the study was to explore the influence of the environment on bycatch events at sea.

Main hypothesis : Oceanographic processes may induce a structuring in the availability of preys creating areas prone to harmful interactions between fishermen and common dolphins.


1 Gilbert L, Rouby E, Tew-Kai E, Spitz J, Quilfen V, Peltier H, Authier M (2021) Spatiotemporal models highlight influence of oceanographic conditions on common dolphin bycatch risk in the Bay of Biscay, Marine Ecology Progress Series, 679, 195-212, https://doi.org/10.3354/meps13894




Fig. 1. Map of the study area : Bay of Biscay & Celtic Sea.

1. Data

The study is based on a monthly analysis of the dataset, from 2012 to 2019 . The data tool will help one visualize data with maps and histograms.

Oceanographic data was provided by the Shom and was output of the circulation model Hybrid Coordinate Ocean Model (HYCOM). Bycatch mortality information was based on the calculation of a Mortality Index (MI) from stranding data. The MI allows to identify potential mortality areas at sea and to quantify the intensity of mortality events.

2. Results

We conducted a modelling study with 12 monthly hierarchical spatio-temporal bayesian models inferred with INLA (Integrated Nested Laplace Approximations).

  • Oceanographic covariates effects were considered as linear. However, linear coefficients were allowed to vary annually, resulting in a temporal series of each monthly coefficient from 2012 to 2019.

  • A confounfing factor was integrated with a linear coefficient: the stranding probability .

  • A spatial field was inferred to capture all spatial variations of the response variable not taken into account by oceanographic covariates.

  • Fitted values were mapped over the study area to allow comparison with values of the dataset ('observed').

  • The sum of all monthly spatial MI is equal to the number of stranded dolphins showing bycatch evidence and integrated in the dataset. We summed fitted values over each month to compair the two temporal series.

  • We assessed the performance of our models through a cross-validation procedure, predicting values of MI for year 2019 Both maps of predicted MI and annual series of total monthly predicted MI are available to allow comparison with values of the dataset.

  • We further tested seven repetition scenarios, still predicting values of MI for year 2019 but considering it as a repetition of one of the previous years of the study period instead of considering it as a new year (as for the cross-validation procedure). For each scenario, both maps of predicted MI and annual series of total monthly predicted MI are available to allow comparison with values of the dataset.

Additionnal results are presented in the article, like results of variance partioning conducted to assess how much of MI variations are accounted for with each of the monthly model components.











For any additional information, please contact Lola Gilbert : lola.gilbert at univ-lr.fr