Diverse integrated ecosystem approach overcomes pandemic-related fisheries monitoring challenges

The pandemic forced widespread cancellation of marine fishery-independent surveys in 2020, creating critical data gaps for stock and ecosystem assessments. Such gaps are especially problematic amid increasing climate variability affecting coastal ecosystems. Resource agencies have highlighted the need to plan for unavoidable survey effort reductions and to develop tools that can maintain reliable indicators when sampling is curtailed. The Rockfish Recruitment and Ecosystem Assessment Survey (RREAS) is a long-running mid-water trawl survey in the California Current Large Marine Ecosystem used to track recruitment and forage species abundance. Anticipating loss of ship time in 2020, the authors asked whether an integrated set of empirical models, auxiliary observations, and simulations could update key ecosystem indicators, quantify uncertainty due to reduced sampling, and provide actionable ecosystem status information despite data scarcity.

The paper situates the work within a context of: (1) global reports of COVID-19 impacts on fisheries surveys and management; (2) prior workshops and guidance on adapting to unavoidable survey effort reductions, including simulation-based tradeoffs between sampling effort and index accuracy; (3) decades of RREAS-based ecosystem monitoring informing recruitment indices, forage species dynamics, and links to predator responses and ocean-climate variability; and (4) advances in analytical approaches such as delta-GLMs for index standardization, predator (seabird) diet as ecosystem indicators, and species distribution models (SDMs) integrating environmental drivers (e.g., upwelling, ROMS products) to predict krill distribution and abundance. The authors leverage these strands to design a diversified, integrative framework for a data-poor year.

Study system and survey: The NOAA Rockfish Recruitment and Ecosystem Assessment Survey (RREAS) samples epipelagic micronekton and YOY groundfish annually (May–June) using night mid-water trawls at fixed stations across central and broader California waters. CPUE per trawl is computed for key taxa. In 2020, the NOAA vessel was unavailable; a commercial charter was used following COVID-19 protocols. Only 15 trawls were completed in the core area (vs. ~62 average, 2004–2019), biased inshore, with later timing (8–25 June). No oceanographic profiles, hydroacoustics, or visual predator surveys were collected. Gear and standardized tow protocols were maintained with temperature-depth recorders verifying headrope depth and duration.

Indicator estimation: For 1990–2020, indices for YOY rockfish (Sebastes spp.), YOY sanddabs (Citharichthys spp.), YOY Pacific hake (Merluccius productus), adult northern anchovy (Engraulis mordax), market squid (Doryteuthis opalescens), myctophids (Myctophidae), octopus (Octopoda), and total krill (Euphausiidae) were estimated using a delta-GLM (hurdle) approach. Separate binomial (presence/absence) and lognormal positive-catch models with spatial covariates (station, region, depth, region×depth; and for YOY rockfish, Julian day bin) were fit; year effects were estimated in a Bayesian framework with vague priors using rstanarm in R 3.6.3/RStudio 3.5.3. Annual indices (proportion positive × mean positive) were back-transformed, log(x+1)-transformed, and standardized (z-scores) to report anomalies.

Survey effort reduction simulations: To evaluate bias and precision under reduced effort, for each year 1990–2019 the index was recomputed using only one trawl from each of the 15 stations sampled in 2020 (with all other years’ data retained) and compared to the full-sample index to quantify relative bias. Relationships between sample size (n), standard error (SE), sample-size–corrected SD, and mean log index were analyzed. SE scaling with effort used SE ∝ SD/√n and was validated by random subsampling (including realistic consecutive-trawl removal) across focal years.

Seabird diet regression models: Common murre (Uria aalge) chick provisioning at Southeast Farallon Island (1983–2020) provided independent predator diet data. Prey items were identified visually during standardized feeding watches; proportions by prey category were derived and converted to mass. Linear regressions relating the mean proportion of YOY rockfish in murre diet to the YOY rockfish index (standardized to age 100 days) for 1983–2019, and the proportion of anchovy in diet to the anchovy index for 1990–2019, were updated using the delta-GLM indices. 2020 diet proportions (33% rockfish; 61% anchovy) were used to predict 2020 indices and compared with survey-based indices.

Krill species distribution models (SDMs): For Thysanoessa spinifera (coastal/neritic) and Euphausia pacifica (offshore/slope), boosted regression tree SDMs previously trained on RREAS CPUE (2002–2018) and environmental drivers (winter preconditioning upwelling from remote sensing and ROMS 4D-Var products, depth, distance to shore, and spring mesoscale conditions) were used to predict 2019–2020 distributions and mean ln(CPUE+1) at core stations. SDM-derived time series were compared to delta-GLM indices; spatial anomalies (relative to 2002–2018 mean) were mapped. Patchiness was assessed by comparing observed among-station CVs with model predictions.

• Reduced-effort, spatially biased 2020 sampling rendered traditional average log(CPUE+1) indices biased; model-based (delta-GLM with spatial covariates) indices showed no systematic bias when applied to equivalently subsampled past data and were adopted for 1990–2020 reporting.
• 2020 model-based indices: total YOY rockfish and sanddabs were the second lowest on record, continuing declines from highs during the 2014–2016 marine heatwave; Pacific hake, myctophids, and octopus were below average; adult northern anchovy remained persistently high; market squid was below average; total krill (delta-GLM) was lower than average with likely underestimated uncertainty due to patchiness.
• Uncertainty: For most taxa, 2020 had the greatest SE in the time series. SEs were >3× the long-term mean for rockfish, hake, myctophids, and octopus; less than double for sanddabs and krill. Adult anchovy had lower-than-average uncertainty due to high abundance/frequency of occurrence. Error roughly doubled when trawls decreased from ~62 to 15; reducing to 40 increased relative error by just under 25%; increasing to 90 decreased relative error by ~16%. Error scaled with abundance in a taxon-specific way (e.g., ~2× at lowest vs. highest abundance for rockfish, sanddabs, hake, squid; >4× for anchovy and octopus; modest scaling for myctophids and krill).
• Seabird diet models: Linear relationships between murre diet and indices were strong (YOY rockfish r^2 ≈ 0.70; anchovy r^2 ≈ 0.58; both p<0.001). 2020 diet-based predictions were consistent with survey-based indices within 95% CIs (rockfish slightly higher, anchovy slightly lower from the murre perspective), corroborating limited trawl results.
• Krill SDMs vs. delta-GLM: SDMs indicated 2020 T. spinifera above average and E. pacifica below average; delta-GLM suggested both below average. Time series coherence between approaches prior to 2020 was moderate (E. pacifica r=0.64, p<0.01; T. spinifera r=0.66, p<0.01). Independent evidence (improved Cassin’s auklet breeding success; unusual aggregation of ~50 blue whales feeding near Farallones) supported higher krill availability in 2020, particularly T. spinifera. SDMs cannot capture fine-scale patchiness at fixed stations, contributing to residual uncertainty.

The study addressed whether an integrated framework could sustain credible ecosystem indicators during a year of sharply reduced survey effort. By switching to spatially explicit delta-GLMs, validating bias and precision via retrospective subsampling, and leveraging independent data streams (seabird diets) and predictive models (krill SDMs), the authors produced 2020 indices with quantified uncertainties and cross-validation. Findings show: (1) spatially aware model-based indices mitigate bias from unbalanced sampling; (2) explicit effort–uncertainty tradeoffs inform planning (e.g., diminishing returns above ~60 hauls, steep uncertainty increase below ~40–60); (3) predator diet data provide robust, low-cost validation or proxy indicators in data-poor years; and (4) SDMs trained on historical conditions can complement sparse observations, though caution is needed under novel environmental regimes and when patchiness is high. The integrated results maintained strategic advice for ecosystem-based management in the California Current during the COVID-19 era, identifying below-average YOY groundfish and squid, persistently high anchovy, and likely recovery of coastal krill relative to 2019. This diversified approach enhances resilience of monitoring programs to shocks and informs communication of uncertainty to decision-makers.

The paper demonstrates a diversified, integrated ecosystem approach to overcome pandemic-driven monitoring gaps. By combining reduced but targeted trawl sampling, spatially explicit delta-GLMs, retrospective effort-reduction simulations, seabird diet regressions, and krill SDMs, the authors recovered key 2020 ecosystem indicators with quantified uncertainty and independent corroboration. The framework is broadly transferable as a stopgap for data-poor years, not a replacement for surveys. The authors recommend: planning for unavoidable effort reductions via power/effort analyses; building partnerships to integrate top predator observations; incorporating predictive ecosystem models; and exploring autonomous platforms (e.g., gliders with acoustics) to augment surveys. Future research should examine consequences of missed or partial surveys for stock assessments, develop multi-observation Bayesian models that merge trawl and predator data, and expand model ensembles to handle patchiness and novel ocean conditions.

• 2020 sampling effort was only ~25% of average in the core area (15 vs. ~62 trawls), spatially biased to inshore and conducted later in the season, potentially missing peak YOY rockfish abundance.
• No concurrent oceanographic profiles, hydroacoustics, or visual predator/mammal surveys were collected due to platform constraints, limiting environmental context and fine-scale biomass mapping.
• Krill assessments are affected by high patchiness; delta-GLM uncertainty may be underestimated and SDMs cannot resolve fine-scale swarm structure at fixed stations.
• SDM performance may degrade under novel environmental conditions; models were trained on 2002–2018 conditions.
• The approach is a temporary substitute for lost data and depends on continuity of historical time series; it does not determine whether missing a survey materially biases stock assessments.
• Broader spatial time series beyond the core region were not updatable in 2020, limiting geographic inference.