Fisheries-independent surveys provide some of the most important data used to inform stock assessments and management decisions (Gunderson, 1993; Chen et al., 2003). Indices of abundance generated from survey data are typically assumed to be proportional to the true abundance of target populations, a relationship which is expressed via the catchability parameter (q) in assessment models (Quinn and Deriso, 1999). The relationship between the survey index and population size is assumed to be constant in most cases, violation of which can lead to spurious estimates of stock dynamics (Pope and Shepherd, 1985; Wilberg et al., 2009; Thorson et al., 2013; Kotwicki et al., 2014). Consequently, changes to survey designs and operations that affect catchability may disrupt the stationarity of this relationship and confound true population trends if not accounted for (Godo, 1994; Kimura and Somerton, 2006; Cadigan et al., 2022). Changes in gear selectivity and effort allocation across space may also shift the composition of ages and sizes sampled, potentially affecting estimates of population age and size structure (Ono et al., 2015; Kotwicki et al., 2017; O’Leary et al., 2020). As such, to the extent possible, fisheries monitoring agencies typically avoid altering survey designs and sampling procedures in order to minimize the impacts such changes could have on data products and population estimates.

While long-term time series of fisheries-independent survey data with minimal interruptions or modifications are desirable (Godo, 1994; Stompe et al., 2020), circumstances may arise where changes are unavoidable or beneficial. For instance, sections of historically surveyed areas may become inaccessible due to marine protected area designations or wind energy development (Field et al., 2006; Methratta et al., 2020; Hare et al., 2022). Similarly, hazardous weather, vessel breakdowns, staffing and budget shortages, as well as pandemics, international conflicts, and economic turmoil can disrupt monitoring agencies’ ability to complete surveys (ICES, 2020; Santora et al., 2021; ICES, 2023). Changes may also offer advantages, such as modernizing fishing gear or improving the efficiency of sampling designs (Brown et al., 2007; Oyafuso et al., 2021; Oyafuso et al., 2022). Moreover, marine ecosystems are dynamic, and updating survey designs may be necessary to adapt to shifting species distributions, abundance trends, and management priorities (Dulvy et al., 2008; Pinsky and Mantua, 2014; Maureaud et al., 2021; DeFilippo et al., 2023). While some fisheries-independent survey designs can readily accommodate changes in sampling density (e.g., stratified random), others are less flexible (e.g., systematic) (Cochran, 1977) and may require more detailed evaluation of the consequences of such changes.

The U.S. National Oceanic and Atmospheric Administration (NOAA) eastern Bering Sea (EBS) bottom trawl survey provides critical data in support of stock assessment and management for some of the world’s most commercially valuable fish and crab stocks. The EBS survey follows a stratified systematic design composed of 350 rectangular grid cells which are each sampled as part of annual surveys (Lauth et al., 2019). However, near St. Matthew Island and the Pribilof Islands the corners of the grid cells are sampled in addition to the centers. The rationale for finer-scale sampling near St. Matthew and the Pribilof Islands has been to improve monitoring and data products for blue king crab (Paralithodes platypus) stocks in these areas, which historically supported valuable commercial fisheries. Blue king crab exhibit a sparse and patchy distribution, resulting in highly uncertain abundance estimates. Sampling these corner stations was instituted to increase the probability of encountering concentrated patches of blue king crab and reduce uncertainty in survey data products. However, both the Pribilof and St. Matthew stocks of blue king crab have declined substantially and are now closed to fishing with little sign of rebuilding to harvestable levels in the near future (Palof et al., 2020; Stockhausen, 2021a). Consequently, it is unclear if the effort and funds required to continue sampling the corner stations is justifiable, or if resources might be better allocated to other priorities, such as improving data quality for other species.

Here, we present an approach for evaluating the consequences of reduced sampling effort in systematic survey designs and apply it to estimate the effects of removing the St. Matthew and Pribilof Island corner stations from the EBS survey. While changes to survey fishing gear and sampling protocols can be resolved by paired fishing and intercalibration studies (e.g., Cadigan and Dowden, 2010; Miller, 2013; Kotwicki et al., 2017; Cadigan et al., 2022), there is less guidance for evaluating changes in sampling effort (but see Zimmermann and Enberg, 2017; ICES, 2020; ICES, 2023), particularly for systematic survey designs. Using existing survey time-series and simulation analyses, our approach estimates the impacts of effort reduction on the precision and accuracy of survey data products, as well as stock assessment output and biological and management reference points. As fisheries management agencies are tasked with monitoring changing marine ecosystems under static or declining budgets, tools for addressing effort reduction and reallocation decisions will be essential for optimizing survey efficiency and ensuring reliable data products (ICES, 2020; ICES, 2023). The approach we present here is generic and flexible and can be widely applied to other species and regions by scientists and managers facing survey effort reduction decisions.

The crab stocks for which corner station removal had the greatest impacts on data quality were those whose distributions are concentrated around St. Matthew and the Pribilof Islands. However, it is important to note that the uncertainty in these stocks’ data products was often substantial even with the corner station data included. The difficulty of producing high quality biomass estimates for the St. Matthew and Pribilof Island king crab stocks may be partially due to these species’ associations with hard-bottom habitats that the EBS bottom trawl survey cannot sample effectively (e.g., female blue king crab; Vølstad, 2012; Zacher et al., 2020), and/or limited ranges and low abundance levels (e.g., Pribilof red king crab; Zacher et al., 2020). At such levels of uncertainty, the contributions these data make to stock assessment models are limited and further declines in precision from corner station removal may not represent a meaningful erosion of their information content. For instance, the annual survey biomass estimates for Pribilof blue and red king crab are considered too variable and uncertain for use in stock assessment models even with the corner station data, such that these assessments rely on post hoc smoothing of the survey biomass time series (Stockhausen, 2021a; Szuwalski, 2022). In some cases the declines in data precision and accuracy arising from corner station removal were mitigated to some extent by using a model-based versus design-based estimator.

The impacts of corner station removal on both the precision and accuracy of biomass estimates were minor for all groundfish species that we investigated except for yellow Irish lord, which exhibits high concentrations around the Pribilof Islands ( Figure S2 ). The other groundfish species we examined exhibit broader distributions, either due to lack of association with the St. Matthew or Pribilof Island habitat areas or because their distributions are more strongly driven by dynamic environmental conditions than by static habitat characteristics. For instance, Alaska skate are widely distributed throughout the outer EBS shelf area ( Figure S3 ), and yellowfin sole are broadly associated with shallow, inshore areas of the shelf ( Figure S4 ). Moreover, many groundfishes are mobile and alter their distributions in response to environmental conditions. For example, walleye pollock and Pacific cod actively avoid a mass of <2° C subsurface water (the cold pool) that occurs on the EBS shelf (Wyllie-Echeverria and Wooster, 1998; Kotwicki and Lauth, 2013). As such, static habitat associations are likely a less important component of these species’ distributions. Importantly, with the exception of yellow Irish lord (which does not have a stock assessment and is of no commercial importance) it is unlikely that the removal of corner stations from the EBS survey would meaningfully impact estimates of stock status or management recommendations for any of the groundfish species we examined.

Our analysis of the effects of corner station removal on stock assessment outcomes showed negligible effects for Tanner crab. The Tanner crab assessment model predictions were nearly identical regardless of whether or not the corner station data were included. The only appreciable changes in model output arising from corner station removal were marginally greater estimates of biomass-related reference points and the OFL, likely attributable to the slightly lower estimates of natural mortality produced without the corner station data. Such minor changes in model predictions and reference points seem unlikely to affect decision-making for this stock, indicating that assessment and management of Tanner crab would be robust to removal of the corner stations.

Not surprisingly, we found somewhat greater effects of corner station removal on stock assessment output for St. Matthew blue king crab. Assessment model predictions made without the corner station for this stock indicated slightly lower mature male biomass and recruitment over time compared to baseline model output that included the corner stations. However, the trends in biomass and recruitment were nearly identical regardless of corner station inclusion. Despite the minor differences between time-series produced with versus without the corner stations, the coherence of trends across data scenarios suggests that ability to detect stock recovery would not be impaired without the corner station data. Similarly, the estimated ratio of current biomass relative to B_MSY (B/B_MSY), the reference point used for tracking stock rebuilding, was not affected by removal of the corner station data. As such, despite the substantial effects of corner station removal on survey data products for St. Matthew blue king crab, it does not appear that there is a correspondingly appreciable effect on assessment outcomes for this stock, at least with respect to monitoring stock rebuilding. This may be partially due to the fact that the EBS bottom trawl survey data is not the only source of fisheries-independent data that are used in the St. Matthew blue king crab stock assessment, the other being the Alaska Department of Fish and Game (ADFG) pot survey (Palof, 2022).

While our analysis focuses on the effects of survey effort reduction on stock assessment inputs (survey data products) and outputs (biological/management reference points), there are other objectives that can be important to fisheries and ecosystem management. For instance, there may be benefits to finer-scale sampling of ecologically important areas – beyond those to stock assessment outcomes – in improving understanding of crucial habitats. Indeed, the Pribilof Islands are important habitat for many fish, invertebrate, seabird and marine mammal species (Craighead and Oppenheim, 1985; Gentry, 1998; Ferrero et al., 2000) and one of the most productive regions in the Bering Sea (Cooney and Coyle, 1982; Coyle and Cooney, 1993), with oceanographic and ecological dynamics that are distinct from the rest of the EBS shelf (Hunt et al., 2008; Ciannelli et al., 2004). However, it is difficult to quantitatively assess the value of finer-scale survey observations for understanding ecological and oceanographic processes. Nonetheless, objectives beyond utility to stock assessment such as ecosystem considerations may be important to managers, and it is important to note that such concerns cannot be assessed using the approach we present here.

We found that the stocks for which corner station removal caused substantial declines in data quality were also those with the lowest quality data to begin with. This result suggests that while sampling the corner stations may improve data quality for some stocks, the extent of improvement may not be sufficient to produce satisfactorily informative data products. The limited data quality for Pribilof and St. Matthew king crab and yellow Irish lord may simply reflect the fact that the EBS bottom trawl survey is not optimized for sampling certain species, a challenge that naturally arises in fisheries-independent survey designs with multispecies objectives (Cochran, 1977; Godo, 1994; Oyafuso et al., 2021). Data quality for these stocks may be more effectively improved by alternative approaches beyond increased bottom trawl sampling density, such as expanded collection of other forms of fisheries-independent data (e.g., pot surveys for king crab; Gish and Vanek, 2010; Palof, 2022), or building capacity for generating and using model-based indices (Thorson, 2019). While we found that model-based biomass estimates were more robust to corner station removal than their design-based counterparts, adequate model performance was difficult to achieve in some cases (e.g., St. Matthew blue king crab, Pribilof red and blue king crab) and stock assessments for Bering Sea crab still rely entirely on design-based abundance and compositional estimates (Stockhausen, 2014; Szuwalski et al., 2014; Szuwalski and Turnock, 2016; Palof et al., 2020; Stockhausen, 2021a; Stockhausen, 2021b; Szuwalski, 2022). Continued innovation in model-based index production methods for crab, as well as expanded capacity to use model-based indices in crab assessment models may lead to improved inference on stock status and mitigate adverse consequences from potential survey effort reduction. Model-based indices can also incorporate environmental covariates to facilitate extrapolation to unsampled areas (Thorson, 2019; O'Leary et al., 2022) which may also help offset the impacts of reductions in sampling effort.

While stock assessment models are sensitive to the precision of input data, reductions in data quality beyond a certain point may not exert a corresponding effect on their information content. This is due to the integrated design of many contemporary stock assessment models (including those for Tanner and St. Matthew blue king crab), in which multiple data sources contribute to parameter estimates via a joint likelihood (Fournier and Archibald, 1982; Maunder and Punt, 2013). In such integrated models, the greater the uncertainty in one data source, the less information it will contribute to the model, whether by explicit data weighting procedures (e.g., Francis, 2011) or via the joint likelihood itself (e.g., DeFilippo et al., 2021). In our analysis, this effect is best demonstrated by St. Matthew blue king crab, for which corner station removal had a minor effect on assessment model outputs despite exerting a large influence on the input survey data products. This result emphasizes the importance of propagating the impacts of survey effort reduction through stock assessment models to understand the impacts of proposed survey design changes. Indeed, contemporary stock assessment models are complex, with multiple contributing (and often conflicting) sources of information that can interact to affect parameter estimates in unpredictable ways (Maunder and Punt, 2013; Ichinokawa et al., 2014; Peterson et al., 2021).

The approach we demonstrate here is flexible and can be used to evaluate the potential effects of removing any set of existing stations from a systematic fisheries-independent survey. The corner stations we considered for removal in our analysis were selected because they represent an area of higher sampling density than the rest of the survey grid and because the stocks that motivated their implementation have remained below harvestable levels for a prolonged period of time. However, there are other situations in which a set of stations may be identified a priori as candidates for removal. For instance, a particular area within a survey’s boundaries may be under consideration for wind energy development or MPA designation, or become inaccessible due to political (e.g., international strife) or logistical (e.g., prohibitive fuel costs for reaching remote areas) reasons (Field et al., 2006; ICES, 2020; Methratta et al., 2020; Hare et al., 2022; ICES, 2023). In such scenarios, scientists could follow the general approach outlined here to quantify the expected impacts of removing the stations in question on data quality and stock assessment outcomes. These results could be used to understand and predict the impacts of the effort reduction on scientific inference, and/or be used to develop calibration factors to correct for these effects in the survey data time series. Conversely, there may be a need to reduce sampling effort more generally (e.g., due to budgetary and staffing shortages), without a specific subset of stations indicated as leading options for removal (ICES, 2020; ICES, 2023). In response to such a need to reduce sampling effort without obvious candidates for station removal, scientists may use our approach to evaluate a range of alternative scenarios (e.g., thinning the sampling density of the entire survey grid by a specified fraction, removing stations from various areas based on expert opinion), and compare the impacts of each effort reduction scenario on data quality and stock assessment output to inform decision-making.

While evaluating the effects of station removal on all survey data products and stock assessment outcomes is optimal, scientists may prioritize certain stocks to focus effort on (e.g., crab in our case study) and conduct more limited investigations on others for which expected impacts are lower (e.g., groundfish in our case study). As shifting ecosystems and species distributions place growing demands on fisheries-independent monitoring amid rising costs and limited institutional budgets, unavoidable reductions in survey effort are likely to become increasingly common (ICES, 2020; ICES, 2023). Consequently, it is imperative that scientists can evaluate the effects of survey effort reductions to inform decision-making on such actions and understand their consequences. The analysis we present here demonstrates a generic set of steps for evaluating reductions in sampling density for systematic survey designs that can be applied in similar situations elsewhere to provide a quantitative basis for decision-making.

Publicly available datasets were analyzed in this study. This data can be found here: https://www.fisheries.noaa.gov/foss/f?p=215%3A28.

The manuscript presents research on animals that do not require ethical approval for their study.

SK and LD conceived the study. LD designed and executed the analyses and wrote the paper. WS and KP conducted and helped interpret the stock assessment model runs. JR and ML contributed crab data and input on their use and interpretation. LB, LD, and SK contributed input on interpreting study results. All authors contributed to manuscript writing. All authors contributed to the article and approved the submitted version.