ICES Journal of Marine Science

Accurate estimation of post-release survival is fundamental to fisheries stock assessment and effective management. Conventional tag-return studies and acoustic telemetry are commonly used to estimate this probability, yet each approach has limitations when applied independently. Using gag (Mycteroperca microlepis) as a case study, we integrated data from a large-scale conventional tagging program and an acoustic telemetry experiment within a discrete-time statistical modeling framework that links relative recapture risk with telemetry-derived fate. This approach enabled estimation of post-release survival across a broad gradient of capture depths representative of recreational fishing conditions. Estimated survival was high in shallow waters ({approx}97%) but declined with increasing capture depth, consistent with depth-related barotrauma. Applying model predictions to depth distributions from the recreational fishery yielded annual and monthly post-release survival probabilities. Annual estimates were consistent with values assumed in recent stock assessments, while monthly values highlighted seasonal patterns potentially relevant for management. This integrated framework advances post-release survival estimation by combining the extensive sample sizes and environmental coverage characteristic of conventional tagging data with the direct fate observations provided by acoustic telemetry, and offers a transferable approach for other highly targeted fisheries.

The development of Artificial Intelligence (AI) presents novel opportunities for tackling complex marine resource management challenges. Among AI models, neural networks are a powerful class of tools capable of learning nonlocal and lagged patterns from fisheries data as well as approximating nonlinear relationships among multiple latent variables using estimation methods that automatically implement statistical shrinkage. This gives them potential to effectively handle data obtained from fisheries populations subject to dynamic environments. We highlight two flexible subclasses and one application of neural networks: Long Short-Term Memory (LSTM) and Convolutional Neural networks (CNNs) and policy discovery via Reinforcement Learning. LSTMs are designed to handle sequential data by allowing prediction from past values at both short and long time-lags. CNNs are not explicitly designed to handle temporal information, but can interpolate a spatial latent variable based on its value within a geographic neighborhood, and can be combined with LSTM models for this purpose. This "Food for Thought" paper introduces and applies these neural network approaches, both alone and in combination, to demonstrate their potential application for several foundational topics in fisheries science: 1) the forecasting of population weight-at-age, 2) the standardization of spatio-temporal indices of relative abundance, and 3) the discovery of harvest policies to optimize catches and maintain spawning biomass. Each section provides a simple, simulated example and describes the tradeoffs - particularly the lack of inferential capability - presented by using neural networks over traditional approaches for each topic. We then outline medium-term research questions that may clarify, facilitate or qualify the applicability of these tools to fisheries management science. Finally, we discuss how future combinations of these approaches could result in simplified ways to estimate and forecast stock biomass and provide harvest advice.

Catch inequality--the disproportionate distribution of catch across anglers-- is a fundamental but overlooked driver of recreational fisheries dynamics. Here, we use 11 years (2012-2022) of compulsory angler report cards to characterize long-term catch dynamics in the specialized recreational steelhead (Oncorhynchus mykiss) fishery in California, U.S.A. Spatialized catch data reveal the fishery is principally supported by wild fish, despite evidence of widespread hatchery straying. California steelhead appear to represent the most catch-unequal recreational fishery studied yet, exhibiting a statewide Gini coefficient of 0.81. Across basins, inequality varies substantially but remains relatively stable over time and flow conditions; high inequality is primarily driven by significant proportions of zero-catch anglers. We find the relationship between sample size and inequality measures is especially influential in fisheries data. Hence, we develop a three-prong approach for identifying minimal sample sizes required for robust Gini estimation. Across basins and years, an average minimum of 77 report cards were required for the present fishery. Collectively, these findings demonstrate the necessity of considering catch inequality in fisheries management, particularly when utilizing angler data. Graphical AbstractN.a.

Habitat degradation and biodiversity loss in the Black and Azov Seas necessitate improved tools for spatially explicit conservation planning. We employed stacked species distribution modelling (SSDM) to assess habitat quality for the three resident cetacean species, the common dolphin (Delphinus delphis ponticus), the bottlenose dolphin (Tursiops truncatus ponticus), and the harbour porpoise (Phocoena phocoena relicta), which serve as apex predators and indicators of ecosystem health. Occurrence data were compiled from the Global Biodiversity Information Facility (GBIF), and ensemble species distribution models (ESDMs) were constructed using nine algorithms within the SSDM framework, with eight environmental predictors extracted from Bio-ORACLE v3.0. Individual ESDMs demonstrated excellent predictive performance (AUC: from 0.82 to 0.83; TSS: from 0.65 to 0.67; prop.correct: from 0.82 to 0.83). However, the initial continuous stacking method (pSSDM) yielded low community-level prediction success (0.36), prompting evaluation of three correction approaches. The Probability Ranking Rule (PRR) substantially improved performance (prediction.success = 0.459, sensitivity = 0.704, Jaccard = 0.465), effectively mitigating the overprediction bias inherent in stacked models. Species richness mapping identified multi-species hotspots along the southwestern Black Sea shelf, the Crimean coast, the Kerch Strait, and parts of the eastern coast, while the deep central basin exhibited the lowest richness. Variable importance ranking revealed bathymetry as the primary community-level driver (41.2%), followed by dissolved oxygen (13.8%), sea surface temperature (11.9%), and salinity (10.4%). Species-specific importance patterns confirmed ecological niche segregation, with common dolphins favouring deeper offshore waters and bottlenose dolphins and harbour porpoises associated with shallower shelf environments. The moderate richness observed in the highly productive northwestern shelf, despite high nutrient inputs, may reflect a combination of natural factors (elevated turbidity, reduced salinity) and anthropogenic pressures (fisheries bycatch, shipping, coastal development, and military activity) that limit species co-occurrence. Our findings demonstrate that PRR-corrected SSDM provides a robust framework for mapping cetacean habitat quality and identifying conservation priorities in the Black and Azov Seas, offering an evidence-based tool to inform ecosystem-based management in this ecologically unique and increasingly pressured marine region.

Despite increasing calls for sustainability and ecosystem objectives to manage fishing gear interactions with bottom habitats there are few quantitative approaches for assessing risks from bottom contact fishing. Risk assessments for bottom longline fisheries are particularly challenging due to a lack of information for estimating bottom contact areas from longline gear. In this paper, we demonstrate how data sensors and video cameras deployed on fishing gear can be used to quantify the bottom contact area for longline trap and hook fishing gear from the British Columbia Sablefish fishery. Our bottom contact estimates indicate that Sablefish fishing risks to bottom habitat are low in the majority of fishing areas, since 91.8% of the area fished is expected to have had zero bottom contact over the last 17 years. For the other 8.2% of Sablefish fishing areas that experience some contact from fishing gear, the majority are only contacted once. This indicates that most habitats contacted by Sablefish gear can be expected to have a minimum of 17 years to recover between subsequent bottom contact events. We demonstrate an approach for estimating fisheries bottom contact that can be widely implemented across longline fisheries. Our findings address key data gaps in bottom impacts research for longline gear fisheries, allowing fishing risks to be quantified over fine spatial scales. Such quantitative approaches for habitat risk assessment can provide essential information for management decisions aimed at determining acceptable trade-offs between habitat preservation and fishery benefits.

In the Northeast Atlantic, advice for many fish stocks follows the ICES MSY approach, where a zero catch will be recommended if the stock is below its limit reference point, Blim, and cannot rebuild in the short-term. How-ever, zero catch advice are rarely implemented by managers. This study used medium-term stochastic forecasts with harvest control rules (HCRs) to investigate the consequences of allowing reduced fishing below Blim. We applied the method to western Baltic herring and North Sea cod, two contrasting species currently estimated below Blim. We show that the minimum rebuilding probability of 95% required by the MSY approach could be impossible to reach in the short-to medium-term. When this is the case, a lower probability may need to be considered instead in the short-term. Recruitment is the largest source of uncertainty in stock response to management, and can exceed differences between HCRs. Reference points should be estimated in accordance with current recruitment levels if they are to be used for short-term advice or as realistic rebuilding targets. For both stocks, it is possible to keep fishing at reduced levels for similar cumulative catch, SSB and risk on the stock in the medium-term compared to no catch below Blim. Medium-term trade-offs between stock conservation and fisheries considerations may be needed when fishery closure cannot be implemented in practice.

The hockey-stick (HS) stock recruitment relationship (SRR) has been widely used as an empirical alternative to conventional SRRs such as the Beverton-Holt (BH) and Ricker (RI) models. However, the management performance and risks associated with estimating maximum-sustainable-yield (MSY) reference points (RPs) based on HS remain insufficiently understood. This study first defines deterministic and stochastic MSY RPs under the HS model and provides an overview of their properties. We then conduct simulation experiments to investigate the bias and management consequences that arise when MSY RPs are estimated from the HS model (HS-derived MSY RPs) rather than from the true SRR (e.g., BH) across a range of biological and stochastic parameters, with particular focus on scenarios with insufficient data contrast. Our results show that HS-derived MSY RPs tend to exhibit higher bias but lower variance than MSY RPs derived from the true SRR. Management strategy evaluation simulations further reveal that management procedures combining HS-derived MSY RPs with adaptive model learning and some precautionary measures gradually reduce this bias and achieve average spawning biomass and yield that are comparable to those obtained under management based on the true BH SRR. We also show that the management effectiveness of the precautionary measures depends on life-history traits and recruitment variability. These findings indicate that although HS-derived MSY RPs may be biased and require cautious use, combining them with appropriate precautionary measures allows management to remain robust while limiting variability and yield losses. This broadens the range of management options that are available for supporting sustainable fisheries management.

Global climate change has increased the frequency and severity of stressful temperatures that freshwater fishes experience, necessitating rapid and sensitive methods to monitor wild populations. Tissues used to measure transcriptional responses traditionally involved invasive or lethal sampling, which may be undesirable for imperilled species. Epidermal mucus offers a non-lethal and minimally invasive alternative, but whether thermal thresholds can be detected in mucus to identify fish experiencing thermal stress is unclear. Bull trout (Salvelinus confluentus) are a legally protected salmonid and cold-water specialist, generally occupying waters 12 {degrees}C and below, with higher temperatures resulting in cellular stress. Therefore, we measured a suite of 56 genes using high-throughput qPCR to compare machine learning classifiers developed with transcriptional profiles of epidermal mucus, gill, liver, and muscle to classify laboratory reared juvenile bull trout as below (9 {degrees}C, 12 {degrees}C) or above (15 {degrees}C, 18 {degrees}C) cellular thermal thresholds. Mucus profiles most resembled gills but represented an intermediate transcriptional response to all tissues. A reduced biomarker panel of 10 genes in mucus assigned fish to stress categories with 94.1% (95% CI = 71.3-99.9%) accuracy, which was comparable to gill (100.0%, CI = 82.4- 100%), liver (95.0%, CI = 75.1-99.9%), and muscle (100.0%, CI = 80.5-100.0%). Sex-specific temperature effects were evident in all tissues, but less pronounced in mucus and gill than in liver and muscle. Our findings demonstrate that transcriptional profiling of mucus can reliably identify individuals experiencing thermal stress, highlighting the promise of this non-lethal approach for monitoring at-risk species.

We developed an agent-based model (ABM) to assess how area-based controls on fishing methods can reduce fishing mortality and population declines. The model incorporates the behavior and distributions of dolphins and fishing vessels, and realistic displacement of fishing effort when protection is extended. Our case study is New Zealand dolphin - Hectors and Maui dolphins. The model was designed and calibrated using pattern-oriented modeling. Our results show that mortality due to entanglement in fishing gears has been reduced thanks to a gradual increase in dolphin protection. However, current protection is not as effective as previously thought, and scarce populations are negatively affected by Allee effects. Neither national nor international goals for reducing bycatch are met by current dolphin protection. The IUCN has recommended banning gillnet and trawl fisheries in New Zealand waters < 100m deep. For most New Zealand dolphin populations, this would be effective in achieving national and international goals for reducing bycatch. Only two populations would require additional protection. This modelling approach is also suitable for assessing impacts of bycatch and ship strikes for other marine species, making it suitable for informing management decisions in many regions.

Development of offshore wind energy resources has led to growing concerns for marine wildlife. However, significant uncertainty remains regarding the technologys potential to impact species of interest that may occupy planned development sites. This is further compounded by the difficulty of monitoring highly migratory or data-poor species in marine waters, making practical assessment of site- or species-specific threats that could require additional management intervention particularly problematic. Here, I identify a highly generalizable framework to inform species interactions in marine habitats allocated for offshore resource exploitation, using telemetry-derived artificial intelligence species distribution models. Results from a case study of the federally protected Atlantic Sturgeon (Acipenser oxyrinchus) demonstrate excellent discriminatory capacity (i.e., AUC [&ge;] 0.9) at a relatively fine scale (raster resolution = 1 km2), while providing critical information on predicted occurrence over a broad swath of unmonitored marine habitats (i.e., the Atlantic OCS region of the US; area > 620,000 km2). Furthermore, ensemble map products developed from these models are readily scalable to ongoing management needs and, when overlaid with offshore wind energy lease areas, can feed directly into management strategies to inform best practices for potential habitat influences on Atlantic Sturgeon, as well as other species of commercial or conservation interest.

Effective cetacean conservation depends on robust monitoring, yet traditional visual and passive-acoustic surveys have constraints. We evaluated environmental DNA (eDNA) metabarcoding coupled with species distribution modeling (SDM) as a tool to study habitat use of cetaceans along the Washington State coast, USA. Seawater was collected at the surface and at a 50 m depth from 43 sites (86 samples) during the 2019 U.S.-Canada Integrated Ecosystem & Acoustic-Trawl Survey. A partial section of the mitochondrial control region was amplified with cetacean - specific primers, sequenced on an Illumina MiSeq, and taxonomically assigned with a curated reference database. Nine species were detected; we modelled the three most frequent: Pacific white-sided dolphin (Lagenorhynchus obliquidens) (10 detections), humpback whale (Megaptera novaeangliae) (8 detections), and Rissos dolphin (Grampus griseus) (6 detections). Binomial generalized additive models related presence-absence to bathymetry, distance to shore, longitude, slope, and sea-surface temperature; model performance was assessed with stratified five-fold cross-validation. SDMs explained 17-51% of null deviance and presented high specificity ([&ge;] 0.80). The Pacific white-sided dolphin showed the highest eDNA presence probabilities offshore, beyond the shelf break. Humpback whale eDNA presence probabilities showed hotspots along the shelf break with secondary high-probability patches in near-shore waters. Rissos dolphin eDNA presence probabilities were elevated in offshore zones characterized by steep bathymetric gradients, particularly northwest of the sampled transect. These spatial patterns are consistent with historical visual-acoustic records, suggesting that eDNA-informed SDMs can capture cetacean habitat use. This proof of concept indicates that combining eDNA detections with flexible SDMs could provide a cost-effective, non-invasive complement to conventional surveys and may offer a scalable pathway for marine-mammal monitoring and spatial planning.

Ship strikes are a significant and growing threat to marine megafauna, yet few mitigation measures are implemented at scale due to perceived economic costs to shipping. Here, we present a proof of concept for integrating biodiversity considerations into commercial voyage optimisation, using priority aggregation sites for the endangered whale shark (Rhincodon typus) as a case study. We simulated eight port-to-port voyages for two vessel classes--a crude oil tanker and a container ship--under three routing scenarios: baseline optimisation, speed reduction to 10 kts within core habitats, and complete avoidance of these areas. Across routes, fuel-use changes ranged from -0.13% to 9.65%, with minimal impacts (<1%) for most long-distance voyages. Results indicate that speed reduction is the more efficient mitigation for short voyages, while area avoidance is preferable for longer voyages, with impacts varying by vessel type and operational constraints. Incorporating dynamic, species-specific habitat layers into voyage planning could enable targeted ship-strike mitigation with negligible disruption to global trade. Adoption of such measures - supported by improved data pipelines, real-time forecasting, and integration into regulatory and incentive frameworks - offers a scalable pathway to align biodiversity conservation with decarbonisation goals in the maritime sector.

Antarctic krill (Euphasia superba) are a key component in the Southern Ocean ecosystem, especially in the Atlantic sector, where the majority of the population is concentrated. The Norwegian commercial krill fishery exclusively targets three subareas in the Antarctic: the western Antarctic Peninsula, and the northern shelves of both the South Orkney Islands and South Georgia. Given its reliance on oceanic transport from other regions and the potential impact of rising sea temperatures on the northern habitat, the South Georgian krill population is particularly sensitive to altering environmental conditions. The relative distance from the peninsular regions to South Georgia means that choosing to trawl in this region implies a higher risk, which is why it is exclusively targeted in winter when extensive sea-ice makes peninsular regions unsafe and inaccessible to commercial fishery operations. In this article, we show that relative to operations at South Orkney and the Antarctic Pensinsula, average catches at South Georgia have been lower with higher variability over the past 15 years. Using a Lagrangian modelling approach, we illustrate that variability in advection from source regions in the Antarctic Peninsula are correlated with proceeding catch values at South Georgia. This was not the case for source release sites at the South Orkney Islands. The dominant transport pathways for krill were strongly determined by position of regional fronts and the source sites of recruits to South Georgia were related to the position of fronts at both the Antarctic Peninsula and South Orkney Islands. This study highlights the importance of advective patterns on the variability in krill fishing activity and supports the hypothesis that South Georgia is a sink region for krill in the Southern Ocean while the western Antarctic Peninsula is a central source site.

The precautionary approach to fisheries management advocates for risk-averse management strategies that include biological reference points as well as decision rules and account for scientific uncertainty. In this regard, two approaches have been recommended: (i) harvest control rules (HCRs) with threshold reference points to safeguard against low stock biomass, and (ii) the P* method, a probability-based HCR that reduces the catch limit as a function of scientific uncertainty (i.e. process, model, and observation uncertainty). This study compares the effectiveness of these precautionary approaches in recovering over-exploited fish stocks with various life-history traits and under a wide range of levels of scientific uncertainty. We use management strategy evaluation based on a stochastic, age-based operating model with quarterly time steps and a stochastic surplus production model. The results show that the most effective HCR includes both a biomass threshold as well as the P* method, and leads to high and stable long-term yield with a decreased risk of low stock biomass. For highly dynamics stocks, management strategies that aim for higher biomass targets than the traditionally used BMSY result in higher long-term yield. This study makes the case for probability-based HCRs by demonstrating their benefit over deterministic HCRs and provides a list of recommendations regarding their definition and implementation.

The yellownose skate (Dipturus chilensis) is an endangered skate with a narrow distribution in the southeastern Pacific, facing intense fishing pressure and potential climate threats. Using a species distribution model, we projected the current and future distribution of D. chilensis under contrasting climate change scenarios (SSP1-2.6, SSP2-4.5, and SSP5-8.5) for mid-century (2050) and end-of-century (2100). Our models, which demonstrated robust predictive performance significantly better than random expectations, identified maximum temperature and minimum oxygen as the primary environmental drivers of habitat suitability. Projections revealed a consistent poleward range shift towards the Channels and Fjords of Southern Chile ecoregion across all scenarios. While localized habitat loss was projected in Central Chile and Araucanian ecoregions, particularly under high emissions (SSP5-8.5), these losses were outweighed by southern expansions, leading to a net increase in total suitable habitat by 2100. These findings underscore the critical need for climate-adaptive management strategies, including the protection of emerging southern refugia and dynamic fisheries regulations, to ensure the long-term persistence of D. chilensis.

Small-scale, multispecific fisheries in the Gulf of California face significant challenges including limited species-specific catch data, uncertainty about climate change impacts, and insufficient biological information needed for traditional deterministic models. These knowledge gaps hamper efforts to forecast future conditions and develop appropriate management strategies accurately. The complexity of these multi-species fisheries, combined with data scarcity for many target species, creates substantial barriers to quantifying and addressing climate vulnerability. Deep learning approaches offer a promising alternative by leveraging available data to identify patterns and project trends despite these limitations, providing valuable insights for fisheries management in data-poor contexts. Here, we apply a Mixture of Expert, a deep learning forecasting models for small-scale, multi-specific fisheries in the Gulf of California under future climate change scenarios. Results show varied responses across marine habitats, with reef and benthic fish projected to experience substantial declines (-12.46% and -9.37%) during the 2050s-2060s, followed by recovery in the 2070s-2080s. Economic implications are significant, with reef fish facing projected losses of $1.2 million by the 2050s before recovering by the 2080s. Shapley Additive Explanations (SHAP) analysis was applied to evaluate the importance of features for each predictive model, the analysis revealed the effects of the temperature in different depths for each fishery, and the sensitive analysis pointed to the magnitude of the effect. Our findings suggest that climate impacts will not be uniform across the Gulf, necessitating region-specific management approaches and highlighting the value of maintaining diverse fishing portfolios to enhance resilience against climate-driven changes.

Warming temperatures elicit shifts in habitat use and geographic distributions of fishes, with uneven effects across life stages. Spawners and embryos are particularly sensitive to environmental conditions, with direct impacts of temperature on spawning habitat, as well as indirect connections between their population dynamics and fisheries effort, productivity, and management. Here, we ask how changing environmental conditions and thermal sensitivities of developing embryos confer spatiotemporal variability of thermally-suitable spawning habitat for Pacific cod in the eastern Bering Sea. Specifically, we use bottom temperature values from regionally downscaled global climate models coupled with an experimentally-derived relationship between hatch success and temperature to predict how the extent, mean latitude, and consistency of suitable spawning habitat has changed in the past and may change into the future. We then validate our predictions of suitable spawning habitat with distributions of adults and larvae and examine whether thermal habitat availability relates to recruitment success into the adult cod into the population. We find that the extent and mean latitude of suitable spawning habitat increase over time, particularly if no climate change mitigation occurs in the future. Hotspots of suitable spawning habitat are consistent across shorter time periods but do shift across the Bering Sea shelf by the end of the century. Finally, we find no correlation between the availability of suitable spawning habitat and annual estimates of recruitment. Collectively, our results suggest that as temperatures warm, the availability of suitable spawning habitat will increase and expand spatially and, thus, is not likely to limit recruitment. This work highlights the importance of coupling experimental data with climate models to identify the complex and mechanistic dynamics among temperature, life histories, and ecology, and offers a pathway for examining life stage-specific changes in habitat use and distribution with continued climate change.

Using linear inverse ecosystem modeling as a data assimilation tool, we compare spawning grounds of Atlantic and Southern Bluefin Tuna (ABT and SBT, respectively) based on results from field campaigns in the Gulf of Mexico (GoM) and eastern Indian Ocean off northwest Australia (Argo Basin). Both regions are warm, stratified, low-nutrient waters dominated by cyanobacteria (Prochlorococcus). Despite these similarities, the Argo Basin is more productive, with [~]1.5X higher net primary production and nearly 2X higher production of top trophic levels in the model (tuna larvae, planktivorous fish, and predatory gelatinous zooplankton). Higher primary production in the Argo Basin is mainly driven by higher N2 fixation and storm mixing of new nutrients in the upper and lower euphotic zone, respectively. Increased ecosystem efficiency (secondary production of top trophic levels / primary production) results from differences in plankton food web organization. In the GoM, protistan zooplankton are the direct consumers of nearly all phytoplankton production. In contrast, higher rates of herbivory by crustaceans feeding on nanophytoplankton combines with a higher impact of appendicularians on cyanobacteria to convert plankton production into larval tuna prey more efficiently in the Argo Basin. Despite similarities in the proportions of phytoplankton production mediated by cyanobacteria and other picoplankton in both systems, food web pathways to larval tuna and other planktivorous fish are substantially shorter in the Argo Basin. Our results highlight the impact of distinct zooplankton ecological niches on ecosystem efficiency and suggest a need for better inclusion of plankton food-web structure in models simulating climate impacts on fisheries production. HIGHLIGHTSO_LIDeveloped food web models of tuna spawning habitat (Indian Ocean & Gulf of Mexico) C_LIO_LISpawning habitats in the Argo Basin and Gulf of Mexico (GoM) are both oligotrophic C_LIO_LIArgo Basin had higher net primary production in part as a result of nitrogen fixation C_LIO_LIArgo Basin had higher rates of direct herbivory by metazoan zooplankton C_LIO_LIThis resulted in greater ecosystem efficiency in the Argo Basin. C_LI

Larval dispersal models are central to mapping and predicting ichthyoplankton dynamics in the ocean, yet despite decades of refinement they remain fundamentally limited by their ability to represent adaptive behaviors, relying instead on static trait parameterizations. This deficiency constrains our capacity to design effective restoration and mitigation strategies in an increasingly stressed ocean. SWARM (Simulating Waterborne Agent Routes for Marine connectivity) overcomes this barrier by integrating Large Language Model (LLM)-based behavioral agents with a standard biophysical model to simulate active decision-making during the pelagic larval stage. In both idealized and realistic conditions focusing on Red Snapper larvae in the Gulf of Mexico, agents develop adaptive behaviors that improve settlement and generate explainable vertical distribution patterns. SWARM demonstrates that LLMs can overcome long-standing limitations in dispersal modelling by explicitly representing behavioral drivers of movement, opening new pathways for predicting connectivity and designing effective marine-ecosystem restoration.

Projections of future conditions from Earth systems models (ESMs) are necessary to understand and predict effects of changing environmental conditions on biological systems. Such projections suffer from biases, or mismatches between model output and observations. While adjusting or bias-correcting model output is common, many methods exist with little understanding of their effects on forecasts of biological change. Here, we explore the bias-correction process and its effects on downstream predictive biological models. As an example, we use the Bering 10K, a downscaled ESM for a productive and economically important subarctic ecosystem. We first characterize existing biases for three categories of variables exhibiting different scales and challenges: bottom temperature, sea ice, and net primary production. We then apply eight bias-correction approaches to six indices generated from the three categories and quantify sources of uncertainty in the trajectories of these ecosystem variables. Finally, we demonstrate how different bias-correction approaches affect downstream biological models using three case studies: 1) fish thermal spawning habitat suitability, (2) predicted zooplankton abundance, and (3) match-mismatch of phytoplankton and zooplankton bloom timing. We find that biases manifest in absolute values over time and in the timing of seasonal events. Time series of all six indices differed depending on bias-correction method, differences that were propagated to downstream biological models. For a given year and simulation, depending on method, thermal spawning habitat suitability and zooplankton abundance differed up to 149% and 151%, and match-mismatch increased or did not change. Our work highlights that bias correction reduces mismatches between observations and model output but choosing an approach requires careful consideration as to not amplify and propagate bias in downstream biological models. To that end, we identify best practices for bias correcting global or regional ESMs, including a decision tree to help improve forecasts of the effects of climate change on biological systems.