Résumé
Demersal fish populations are regularly monitored with scientific bottom trawl surveys, during which it is now routine to also collect acoustic information. Correlation analyses often fail at describing the link between acoustic and trawl data partly because such statistics are relevant for linear relationships. This data driven study looks for different ways to combine acoustic and trawl data. Twenty bottom trawl surveys with simultaneous acoustic recordings collected either on station (i.e. while trawling) or underway (i.e., while shipping between stations) are used. Backscattering energies are integrated over bottom referenced layers: ten layers of 1-meter height and layers of 10 meters height afterwards. For each study area (North Sea, Barents Sea or Irish Sea), we define four classes of vertical profiles (two classes of shape combined with two classes of intensity) for which the average proportions of a given (group of) species are significantly different. This classification is used to assign a shape index and an intensity index to the acoustic profiles collected underway. The two indices are kriged to provide probability maps (e.g. probability to be of shape 1). The probability maps to belong to each of the 4 classes are obtained by multiplying these two maps and their complementaries. These maps are finally weighted by the mean proportions of the species for each class and sum up to generate a distribution map for the target species. The quality of the model in terms of variance and cross-validation is studied and the possibility to come back to abundance from the proportions estimation is discussed.