Climate variability significantly impacts marine resources like fish stocks, with consequences for socio-economic systems. In the Anthropocene, understanding climate's impact on marine ecosystems is crucial, especially at interannual to decadal timescales, which can modify long-term climate change effects. Integrating climate information into resource modeling is necessary for both ecological understanding and forecasting future states. Decadal forecasting is fundamental for timely management interventions ensuring sustainable resource use. However, applying climate models for decadal-scale ecosystem predictions remains challenging. While climate influences fish through various mechanisms (recruitment, food, fecundity, growth, migration), climate variables are rarely included in management-oriented fish population forecasting. This is partly due to the historically large impact of fishing mortality and the transient, non-stationary nature of climate impacts. Further complicating factors include the cumulative effects of different drivers (fishing and climate) leading to non-linear dynamics and potential system collapses. Anthropogenic climate change necessitates including environmental variables in fish stock modeling and forecasting. A key limitation is the inadequate representation of shelf seas in global circulation models (GCMs), which provide future climate information. Existing approaches using bioclimate envelope models, detailed ecosystem models, or combining GCM output with high-resolution shelf-sea models focus on long-term changes and do not provide decadal forecasts. The high computational costs associated with these approaches warrant exploration of novel methods. Decadal prediction of fish stocks, emerging from the predictability of the physical environment, particularly in the predictable North Atlantic, offers a promising avenue, using statistical climate-fisheries models to translate GCM-based decadal climate predictions into reliable fisheries forecasts. This study focuses on two Atlantic cod stocks: North Sea cod (near its upper temperature limit, overexploited) and Northeast Arctic cod (near its lower temperature limit, at record-high biomass). Predicting their biomass is vital for sustainable management decisions.

Numerous studies have explored the impact of climate on fish stocks, establishing empirical relationships between climate and various aspects of fish biology, including recruitment, growth and migration. However, these climate variables are rarely incorporated into management-oriented models, due in part to the significant influence of fishing mortality and the complex, non-stationary nature of climate's effects on fish populations. Existing climate-based predictions of fish stocks either focus on longer time horizons, exceeding decadal scales, or are computationally intensive, limiting their applicability for practical management purposes. Furthermore, coupled Global Circulation Models (GCMs) often lack the resolution and complexity necessary to accurately represent shelf sea dynamics, further hindering the integration of climate information into fisheries forecasts.

To provide decadal predictions of cod stocks, the study employs a linear regression model that transforms dynamical predictions of sea surface temperature (SST) into predictions of total stock biomass (TSB). The dynamical SST predictions are provided by 10-year initialized forecasts (and hindcasts) from a decadal prediction system based on the Max Planck Institute Earth System Model (MPI-ESM). Initial conditions for decadal hindcasts are taken from an assimilation experiment incorporating observed atmospheric and oceanic information. A non-initialized historical simulation is also analyzed to isolate prediction skill due to external forcing. Three fishing mortality scenarios are considered. For North Sea cod, the model utilizes North Sea SST, while for Northeast Arctic cod, the subpolar gyre (SPG) SST is used. Various cross-validated statistical models (simple linear regression with temperature and fishing mortality separately, and multiple linear regression with both) are assessed to determine retrospective skill. The models are tested for their ability to predict the time series of TSB. The MPI-ESM's prediction skill for North Sea and SPG temperature is evaluated, noting that skill degrades with increasing prediction horizons. However, high prediction skill is shown to persist in North Sea temperature until year 10. The study then combines the dynamical temperature predictions with the statistical temperature-cod relationships to generate dynamical-statistical predictions of TSB for both cod stocks. The simplest model, using temperature as the sole explanatory variable, is chosen for its extended predictability horizon. For the forecast period (2020-2030), forecasts from the combined fishing and temperature model are included. Anomaly correlation coefficient (ACC) and mean square error skill score (MSESS) are used to assess hindcast and forecast skill. The uncertainty in regression coefficients is estimated using bootstrapping methodology, providing confidence intervals for predictions. The study accounts for potential issues with using ICES stock assessment outputs as observations (being model outcomes, and not entirely independent).

Analysis reveals key differences in the two regions' SST and cod stock dynamics. North Sea SST shows an increasing trend and a negative correlation with cod biomass (temperature increase corresponds to biomass decrease). Linearly detrended North Sea temperature maintains the negative effect on TSB. Fishing mortality of 2-4-year-old cod shows weak correlation with TSB. In contrast, Northeast Arctic cod exhibits multi-annual to decadal variability, positively correlated with SPG SST (with a 7-year lag). Removing trends maintains high correlation, indicating decadal variability dominance. Fishing mortality of 5-10-year-old cod shows strong correlation with TSB (peaking at a 2-year lag). For North Sea cod, the linear model using fishing mortality alone lacks predictive power; combining fishing and temperature yields comparable skill to the temperature-only model. For Northeast Arctic cod, while the fishing-only model fits better than the temperature-only model, the skill difference is not statistically significant. The model combining temperature and fishing has the best fit for both but the temperature only model is chosen for longer prediction horizons. Analysis of MPI-ESM decadal prediction skill reveals high skill in North Sea temperature predictions until year 10, largely attributable to the long-term trend. In contrast, SPG temperature skill is robust irrespective of trends. The dynamical-statistical prediction model shows robust skill in simulating both North Sea and Northeast Arctic cod biomass. For North Sea cod, forecasts suggest continued unfavorable conditions, with no significant recovery under various fishing mortality scenarios. For Northeast Arctic cod, forecasts predict a climate-driven biomass decline in the coming decade compared to present levels, although this decline is influenced by previous high fishing mortality years in cold periods. The study emphasizes the delayed impact of 2010-2016 SPG cooling on Northeast Arctic cod.

This study demonstrates the successful decadal-scale prediction of cod stock abundance using climate predictions from MPI-ESM, fulfilling two key conditions: a robust cod-environment relationship and multiyear physical environment predictability. The strong negative correlation between temperature and North Sea cod biomass is explained by the stock's nonlinear dynamics. Ocean warming negatively affects cod through low recruitment and prey availability changes; fishing restrictions might be insufficient to counteract warming's effects. The long-term trend in North Sea surface temperature explains a large part of cod biomass variance, thus hindcast skill stems largely from the trend. The 2020-2030 forecast for North Sea cod primarily reflects the long-term trend, not year-to-year variations. For Northeast Arctic cod, the positive temperature-biomass correlation is due to temperature's effects on life history traits. The pronounced decadal variability in the SPG, which influences Atlantic water volume entering the Barents Sea, plays a crucial role. The study highlights the importance of considering the specific time lags between SST and cod biomass and how SPG temperature better explains Northeast Arctic cod biomass variability than local SST. The dynamical-statistical model shows significant skill, and provides predictions using a 16-member ensemble and considering fishing scenarios. The contrast in skill sources between North Sea and Northeast Arctic cod highlights the importance of understanding regional climate dynamics in fishery predictions.

This research successfully demonstrates decadal-scale prediction of cod stocks using a combined dynamical-statistical approach. The findings highlight the importance of incorporating climate predictions into fisheries management, allowing for adjustments to future catch targets to account for climate-driven fluctuations in productivity. Future work should refine the models by incorporating more nuanced environmental and ecosystem considerations and investigate the potential of using other ecologically relevant variables, such as primary production, for improved predictions.

The study acknowledges several limitations. Firstly, the low-frequency nature of climate variability influencing Northeast Arctic cod biomass relies on the assumption that the training period is representative of future variability. Secondly, the use of ICES stock assessment outputs, which are model outcomes, introduces dependence and uncertainty. Thirdly, the linear models might not capture complexities like temperature's impact on carrying capacity and reproductive output. Lastly, the models implicitly account for ecosystem processes, potentially overlooking the importance of species interactions.