Fish stock recruitment and marine ecological prediction

Understanding and modelling the production of marine fish is essential for sustainable management of exploited fish populations. Marine fish are notoriously difficult to predict, where especially the recruitment is highly variable on a year-to year basis. Early life-stages and spawners are known to be vulnerable to environmental impacts, where small variations in the physical environment can cause a huge impact on a given cohort’s survival. Furthermore, the intra-species competition is high, causing density dependent effects within a given population. Recruitment is primarily modelled as a relationship between the spawning stock biomass and recruits. However, due to the early life-stage variability, this relationship is often difficult to identify and can cause large residual variance. While relationships between the environment and recruitment are known, previously established environment-recruit correlations tends to break when confronted with new data, making them ineffective in a management context.

However, fish is a primary source for protein in many parts of the world, and sustainable management can ensure the longevity of living marine resources. In most fish stocks, the management includes assumptions about recruitment in the control measurements, such as the total allowable catch and biological references points. For this reason, accurate modelling of recruitment using the available data and knowledge about the biological system can help progress the sustainable exploitation of marine fish.

In this thesis, I attempt to break with the notion of recruitment being unpredictable and argue that to improve our understanding of recruitment, the field has to keep producing and improving recruitment predictions. Here, I present a framework for modelling recruitment based on multi-model inference, where the model ensembles are evaluated on predictive power using methods commonly used in other disciplines, such as atmospheric science and meteorology.

This has been applied on the North Sea sandeel (Ammodytes marinus), an ecologically and economically important species, which currently sustain a large fishery. The framework was outperforming conventional stock assessment methods in two out of four analytically managed stocks. Relative value was quantified using a cost-loss decision model, and showed a large economic potential for industrial users to include recruitment predictions into decision-making.

Furthermore, the framework was applied on a global scale, where 176 stocks were analysed. Here, a majority of all stocks were found to show signs of predictability and potential economic value. The observed stock predictability relied on low-frequency noise within the time series, while high-frequency noise can limit the predictive potential.

Lastly, the explanatory potential of spawning stock biomass on recruitment was examined on 103 stocks in the North Atlantic and North Pacific. In a majority of stocks, environmental information and low-frequency variability explained more of the observed recruitment variability compared to spawning stock biomass. Additionally, life-history traits can have an effect on explanatory power, with especially increases in fecundity and maximum length relying on environmentally informed model. Furthermore, in stocks with habitats close to the thermal threshold of the species, temperature was observed as the primary model ensemble driver.