Résumé
In order to be managed by adequate policies, fish populations are regularly moni-tored thanks to scientific surveys. Depending on the target species, two main sampling techniques can be used: the direct capture of animals (for instance through trawling) at discrete sampling locations or the use of continuous acoustic devices to integrate the amount of fish echoes under the vessel track. More and more, those two types of measurements are collected simultaneously, whereas the estimations of the abundance indices use the two kinds of data separately. The aim of the study is to combine them into a relevant and operational bivariate model. In geostatistics considerations, the type of sampling pattern generated by this type of survey is a bivariate heterotopic configuration since acoustic data are available both between and at trawling locations. The acoustic measurements play the role of the auxiliary variable while the fish capture is the main variable. In the particu-lar case where the cross structure between the two variables is proportional to the structure of the auxiliary variable, the cokriging can be simplified avoiding a tedi-ous modelling phases of allowed bivariate geostatistical models. As a matter of fact, in this case, the auxiliary variable is spatially uncorrelated with the residual of the linear regression of the main variable by the auxiliary one. This leads to a model that can be factorized with the auxiliary variable and the residual as fac-tors (Markov-type model). The cokriging of the target variable (the catch) reduces to the sum of the ordinary kriging of each factor. In this paper, we analyse the application of such Markov-type models in a series of annual surveys to evaluate the contribution of the acoustic in the accuracy of the abundance estimations, in terms of details of the kriging maps, local variance and cross-validation. We show how the implementation of simple bivariate mod-els in a high heterotopic context with skew data can raise difficulties and requires cautious.