Filling the gap: Improving the spatio-temporal coverage of small pelagic fish surveys through modelling approaches

Abundance indices from fisheries-independent surveys provide key information for fish stock assessments, serving as a foundation for sustainable resource management. Changes in survey effort, e.g., caused by vessel breakdowns or bad weather conditions, can lead to areas not being covered in certain years. These spatio-temporal gaps in survey coverage increase uncertainty or can even lead to discontinuation in time series. The current approach for imputing missing strata in Baltic pelagic clupeid surveys uses area-corrected abundances from higher-level subdivision units (baseline model), which does not account for stratum-specific effects and requires at least partial subdivision coverage. We tested three alternative imputation methods: Linear mixed-effects models (LMMs), Generalized additive models (GAMs), and Gradient boosted trees (XGB). Our results suggest that modelling abundance variations across strata, as implemented in the LMMs, improved abundance estimates compared to the baseline model, particularly when only few areas were not covered, i.e., under typical annual variations in survey coverage. When larger areas were not covered, LMMs, relying on explicit spatial strata, performed worse, whereas the XGB models and, in particular, the spatial GAM remained robust and could still be applied even when entire subdivisions were not surveyed. In conclusion, advanced data imputation techniques can enhance the robustness of abundance indices and should be considered as standard practice in survey groups when survey effort varies between years.