Abstract

In fisheries science, two main types of data are commonly available: fishery-independent data, obtained through standardized research surveys, and fishery-dependent data, collected from commercial fishing activities. Research surveys are conducted periodically over broad spatial domains using controlled sampling schemes, resulting in relatively sparse but systematically collected observations. In contrast, commercial fleet data are typically more frequent and spatially dense, but are subject to preferential sampling, as fishing effort is concentrated in locations with higher expected abundance.

While these data sources provide complementary information, their joint use requires modelling approaches capable of accommodating heterogeneous sampling mechanisms. Classical geostatistical methods are well suited for standardized sampling designs but are not designed to handle the preferential nature of commercial data. Ignoring this feature may lead to biased inference and misleading predictions of species distribution.

This work presents a spatio-temporal modelling framework that jointly integrates fishery-independent and fishery-dependent data while explicitly accounting for preferential sampling. The proposed approach combines elements from spatial statistics and hierarchical modelling to capture both the underlying ecological process and the observation mechanisms associated with each data source. In addition, the framework incorporates strategies to handle zero-inflation, a common feature in survey data arising from the absence of species in many sampled locations.

Preliminary results demonstrate that the proposed joint modelling approach improves the characterization of species distribution patterns by leveraging the strengths of both data sources. In particular, it enhances predictive performance and provides a more coherent representation of spatial and temporal dynamics compared to models based on a single data source. The integration of preferential sampling mechanisms further contributes to reducing bias and improving inference reliability.

Overall, this work contributes to the development of flexible and robust statistical tools for analysing complex fisheries data. The proposed framework provides a promising basis for improving species distribution modelling and supports more informed decision-making in marine resource management.