(Talk) Pooling fishery-dependent and -independent data to model species spatio-temporal dynamics - a framework for data boosting and multiple bias correction

The monitoring and assessment of fishery resources rely on two main datasets, namely: commercial fisheries (fishery-dependent) and scientific survey (fishery-independent) data. The monitoring includes mapping the fishery-resources to assess their spatio-temporal dynamics, procedure often conducted through statistical models that usually focus on one data type at a time. Due to their different aims, the sampling design of these data provides distinct levels of bias and consequently hampers their coupling in a joint-likelihood modelling approach. However, if these biases are properly accounted, it is hypothesized that a complete picture and more robust abundance estimates could be achieved by such a model. The main objective hereby was to develop a flexible statistical framework that can compare and integrate these datasets while accounting for their relative bias contributions. As such, a Negative Binomial Cox Process model (NBCP) was designed in Template Model Builder (TMB), where several abundance-related response-variables can be specified and differences in fishing catchability and effort, trawled distance and spatio-temporal correlations can be accounted. Special attention was given to correct the preferential-sampling nature of the fishery-dependent data. To demonstrate its broad applicability, the model was applied on data-rich and -poor stocks from the Kattegat-Western Baltic Sea. For each case study, the NBCP model was tested on (i) commercial, (ii) survey, and (iii) coupled data to assess their differences and evaluate the improvements in abundance estimates. The results revealed that each data source supplied different, yet complementary, information on the species dynamics. Moreover, they confirmed that more precise abundance estimates were obtained by the integrated set-up, especially for the data-poor stocks and where the preferential-sampling bias was considered. Although the fishermen’s prevailed sampling behaviour can still be improved, the proposed NBCP model will likely support future stock management and is particularly a benchmark for data-poor stocks due to the boosting of information.