Résumé
Rapid changes in marine ecosystems highlight the need to account for time-varying productivity in stock assessments used to support fisheries management. Common approaches incorporate annual variation or regressing processes such as recruitment, natural mortality, or growth on environmental variables. While the latter represents a step toward biological realism, it often fails to account for interactions among variables and may yield biased inferences when key drivers are correlated or unmeasured. We introduce a novel framework, Structural Causal Enhanced Stock Assessment Modelling (SCEAM), which integrates a Dynamic Structural Equation Model (DSEM) into a state-space stock assessment method. SCEAM encompasses and extends the full range of existing time-varying approaches within a single framework, enabling direct comparison among them. We applied SCEAM to walleye pollock (Gadus chalcogrammus) in the Gulf of Alaska to improve recruitment forecasting, comparing three causal models of increasing complexity to recruitment modelled as random deviations around a mean, a first order autoregressive process, or regressed on a single variable. We found that a causal model with intermediate complexity best balanced fit, parsimony, and predictive skill. This configuration reduced unexplained variance of recruitment by 69% and improved one-year-ahead forecasts. Key variables included juvenile body condition and juvenile and larval catch rates. Our study represents the first application of a structural causal model embedded within a fisheries population model. SCEAM offers a unified, hypothesis-driven approach for integrating multiple non-independent variables. We therefore propose that SCEAM serve as a general scientific and statistical framework for conducting next-generation ecosystem- and climate-linked fisheries stock assessments.