Diverse integrated ecosystem approach overcomes pandemic-related fisheries monitoring challenges

Conducting an ecosystem survey during a pandemic

Cancellation of the survey aboard its primary National Oceanic and Atmospheric Administration (NOAA) survey vessel was overcome through acquisition of a charter for a commercial fishing vessel, following all COVID-19 guidelines (Supplementary Figs. 1 and 2). Initial plans were for 15 days at sea, rather than the 45 typically conducted. This lower effort, along with adverse weather and vessel constraints, resulted in only 25% of the average number of mid-water trawls being collected in the long-term core survey area (Fig. 1 and Supplementary Fig. 1). Despite the data reduction, this effort was one of the only fisheries independent surveys to occur on the US West Coast after the first lockdown in March 2020, furthering the need to evaluate impacts of reduced sampling and provide a robust synthesis of survey results for fishery management. Here we provide updated indices for a selection of ecologically and commercially important species that are critical for assessing ecosystem status.

The 2020 sampling was spatially biased towards inshore (shallow) stations (Fig. 1) and thus the previously used method for calculating abundance indices (averaging log-transformed catch-per-unit-effort (CPUE), across all sampled stations) was expected to result in biased indices, in particular for species with strong nearshore (e.g., market squid Dorytheuthis opalescens, anchovy) or offshore (YOY Pacific hake Merluccius productus, myctophids Myctophidae, octopus Octopoda, krill) habitat associations (Supplementary Fig. 3). We confirmed that this bias does indeed occur by recomputing indices for the past 30 years, but using only 1 trawl from each of the 15 stations that were sampled in 2020, and comparing these indices to those using all available trawls (Fig. 2 and Supplementary Fig. 4). In contrast, model-based indices computed from equivalently subsampled past data did not show systematic bias due to the incorporation of spatial covariates (Fig. 2). Thus, although the average log CPUEs were well correlated with model-based indices for well-sampled years (1990–2019), average log CPUEs were determined to be inappropriate for 2020 reporting, and the model-based results were used to develop indices for all taxa for years 1990–2020.

The 2020 model-based indices for total rockfish and sanddab (Citharichthys spp.) were the second lowest on record and continued a decline from record high abundance levels observed during the 2014–2016 marine heatwave (Fig. 1)^22,23. Pacific hake, myctophids, and octopus were also below average. In contrast, the 2020 index for adult northern anchovy continued a multi-year period of persistently high abundance (Fig. 1). Market squid indices were below average, following a mostly positive trend over the past 7 years. Following the steep decline in 2019, the krill index in 2020 was lower than average (Fig. 1); however, as discussed below, uncertainty may be underestimated for this highly patchy taxonomic group. As a consequence of the low sample sizes, a more rigorous evaluation of the trade-off between sample size (trawls) and uncertainty was conducted, as well as further evaluation of trends through application of existing ecosystem science tools.

Quantifying uncertainty by resampling the past

For most taxa, the uncertainty associated with the 2020 relative abundance estimate was the greatest in the time series, an intuitive result of the sparse sampling for that year (Figs. 1b and 2). The SE was estimated to be over three times the long-term average SE for rockfish and Pacific hake, myctophids, and octopus, and the largest (but less than double the long-term mean) for sanddabs and krill (Fig. 2a). By contrast, the uncertainty associated with the adult anchovy index was lower than the long-term average, due to the great abundance and high frequency of occurrence of anchovy in 2020, compared to years in past decades. This reflects the general trend of uncertainty (on the log scale) being greater for a given taxon when abundance is lower, which generally held for all taxa except krill in our explorations (Fig. 2 and Supplementary Fig. 5). Through time, the relative bias of the subset of stations (2020) vs. the full sample size is also consistently lower for the model-based solution compared to using the average estimate (Fig. 2 and Supplementary Fig. 6). There is also a strong relationship between the number of trawls conducted and the resulting error for each point estimate, with the error essentially doubling when the number of trawls is reduced from the long-term average of 62 to the 15 that were conducted in 2020 (Fig. 2a). By contrast, reducing the total number of trawls from 62 to 40 increases the relative error by just under 25%, while increasing the number of trawls from 62 to 90 only decreases the relative error by 16%. The extent to which the mean relative abundance scales that error up or down, regardless of sample size, is taxon specific. There is an approximate doubling of the error at lowest abundance levels relative to the highest levels for rockfish, sanddabs, hake, and market squid, an increase of more than fourfold over the same range for anchovies and octopus, and relatively modest scaling of the error for myctophids and krill (Fig. 2). This trade-off between survey effort and the error of the ecosystem indices provides critical guidance for future survey planning with respect to the complex trade-off between effort and uncertainty in the face of highly variable interannual catch rates.

A seabird’s perspective

The Farallon Islands (National Wildlife Refuge) are located in the center of the survey region and host the largest breeding colony of common murre (Uria aalge) in the region (Fig. 1). Interannual variability of Farallon Island seabird population dynamics, reproduction, and foraging ecology are well understood and also track RREAS observations^6,17. In particular, patterns such as alternating cycles of forage species occurrence and subsequent reproductive output are known to be linked to regional ocean and climate conditions^17,20. Long-term observations of seabird diets in the Farallon Islands were fortunately not impacted by the pandemic. As common murre feed their chicks predominantly either juvenile rockfish or northern anchovy (Supplementary Fig. 7), and common murre prey selection is known to covary with prey abundance in the surrounding ecosystem^17,20, these observations provide a critical data stream for evaluating 2020 rockfish and anchovy abundance index estimates from the limited trawl sampling. We updated regression models relating the proportion of rockfish and anchovy in murre diets, respectively, to model-based abundance indices for rockfish and anchovy using past data (Fig. 3). Linear models provided the best fit for YOY rockfish and anchovy, (r² = 0.70; r² = 0.58, respectively, both p < 0.001). During 2020, common murre diet was mixed, with 33% rockfish and 61% anchovy (Supplementary Fig. 7). Application of the seabird regression model produced 2020 index predictions that were largely in agreement with the 2020 indices generated from limited trawl data. From the common murre’s perspective, rockfish was slightly higher and anchovy was slightly lower than suggested by the trawl survey, but estimates were within the 95% confidence intervals. This new seabird tool can be applied in the event of future survey cancellations and time series estimates can be used in stock assessment and food-web studies. Many seabird population and diet data sets are available throughout the world²⁴ and effort should be made to derive similar models with fishery-independent data sets. Further, data streams such as seabird diets could be incorporated directly into multi-observation models, e.g., fish abundance modeled as a latent variable sampled by multiple observation processes (i.e., trawls and birds) in a Bayesian framework²⁵. This diversified data integration approach should allow for more robust estimates and strengths in one data stream may make up for deficiencies in another.

Krill species distribution modeling

Relative abundance of krill is a critical ecosystem indicator that is used to monitor the health and functioning of the coastal and offshore marine food webs^9,26,27. Due to their high abundance and tendency to form dense aggregations (hotspots), reduced offshore sampling likely impacts the assessment of overall krill abundance and regional distribution patterns in 2020 (Fig. 1 and Supplementary Fig. 8). Application of SDMs that are parameterized and trained on historical, environmental, and biological observations are potentially important tools for predicting krill species abundance during reduced sampling (Fig. 4 and Supplementary Fig. 9). The 2020 indices were highly uncertain due to limited sampling, so we applied the delta-generalized linear model (GLM) approach and the new krill SDM to predict relative abundance¹⁸.

Although the delta-GLM index from the trawl survey suggested that total krill abundance in 2020 continued to be low following a steep decline in 2019 (Fig. 1b and Supplementary Fig. 7), the SDMs revealed different patterns for the two dominant krill species (Fig. 4). Relative to long-term averages, the SDM approach indicated that the coastal species, Thysanoessa spinifera, was higher in abundance, whereas Euphausia pacifica, the numerically dominant and offshore species, was lower than average (Fig. 4a). The delta-GLM index indicates E. pacifica and T. spinifera were lower than average. Overall, there is coherence between the two time series derived from both modeling approaches (E. pacifica, r = 0.64, p < 0.01; T. spinifera, r = 0.66, p < 0.01), but the limited 2020 sampling likely impacted delta-GLM approach; the SDM approach is independent of the 2020 observations. Spatially, the SDM predicted strong positive anomalies for T. spinifera throughout the coast of California and near average conditions for E. pacifica (Fig. 4b).

The delta-GLM estimates from the 2020 survey were thus inconsistent with the predictions from the T. spinifera SDM, which suggested a return to higher abundance that may be attributed to favorable upwelling conditions during the previous winter^18,28. Other observations also support that krill abundance in 2020 was likely higher rather than lower. The breeding success of a krill-dependent seabird, the Cassin’s auklet (Ptychoramphus aleuticus)²⁹ increased in 2020 following a sharp decline in 2019 that coincided with low regional krill abundance (Fig. 4 and Supplementary Figs. 7 and 8), and the 2020 survey coincided with an unusual early occurrence of ~50 blue whales (Balaenoptera musculus) aggregated near the Farallon Islands that were feeding on high-density krill swarms³⁰. Although the SDM and delta-GLM estimates were largely in accordance prior to 2020, it’s important to note that the SDM by design cannot capture fine-scale patchiness (fixed station data), which likely introduces some uncertainty in predictions (Supplementary Fig. 9). Similarly, the lack of a relationship between relative abundance and SE described previously (Fig. 2) suggests that patchiness (high variability among observations) does not abate with increasing abundance and alternative modeling and model consensus approaches may benefit future efforts³¹.

Implications for ecosystem monitoring and assessment

Our synthesis provides an optimistic outlook for coping with the loss of ecosystem science data streams needed for fishery assessments. A diversified approach involving application of models and auxiliary observation data streams is promising for monitoring the ecosystem status and maintaining strategic advice for stakeholders during the COVID-19 era. Further, the synthesis and methodology can be easily extended to assess future survey planning when faced with either reduced budgetary constraints or the need to optimize existing survey effort, while maintaining robust estimates of ecosystem conditions. The three-pronged modeling approach involving survey effort simulations, seabird (or other predator) indicators, and species distribution modeling can potentially be extended to any fishery and ecosystem survey that monitors status trends of ecosystem indicators. Notably, we emphasize that our modeling approach works because of the continuous data stream (i.e., no previous cancellations) and should be considered a stopgap measure for a data-poor year and not a replacement for data collection. Furthermore, given the scope of the survey and its use in various stock assessment and ecosystem monitoring frameworks, we did not determine whether or not it truly matters if a survey is missed or if partial data collection may result in increased uncertainty. Future fishery stock assessments should examine this issue and conduct simulations to better understand survey effort reduction or unforeseen cancellations¹². We maintain that limited survey effort, combined with other ecosystem modeling tools (e.g., from seabirds and SDMs), provides important context for monitoring the health and state of a marine ecosystem during data-poor situations and can be applied to optimize other surveys.

Survey effort simulations provide a powerful tool for assessing strength and uncertainty of ecosystem indicators and can inform future modeling studies and strategic sampling design^12,13. Signals from seabirds are informative and should be explored to provide important context on connections between fished resources and dependent predators²⁴. In our case study, new seabird diet models provided increased confidence of estimates of fish abundance and are powerful, especially during data-poor sampling years. SDMs are helpful for filling information gaps, but often are used to predict species habitat suitability (i.e., probability of occurrence) and not abundance³², but they can have limited ecosystem-assessment capabilities if the model performs poorly on novel environmental conditions, such as recent ocean-climate warming events³³. However, our krill modeling approach was parameterized and trained on surveyed abundance data and historical ocean conditions, thereby providing a novel source of spatio-temporal information for tracking conditions of a critical food-web species.

Planning for potential unavoidable and limited sampling effort should be a priority for all long-term fishery and ecosystem-assessment surveys. The COVID-19 pandemic left a significant mark on marine ecosystem monitoring studies and many ecosystem indicators designed to inform fishery management will likely not be updated, at least without a number of caveats⁴. On the US West Coast, several major surveys were canceled³ and the loss of 2020 data may result in increased uncertainty in future stock and ecosystem assessments^4,14. For example, cancellation of the coastwide groundfish bottom survey, Pacific hake survey, and spring and summer coastal pelagic species surveys leaves a large information gap³. The COVID-19 pandemic has also impacted fishery landings due to safety and reduced economic demand^5,34. In the US West Coast region, fishery landings during March–July 2020 were 31% lower than the previous 5-year period (2015–19 median) and total commercial ex-vessel revenue through October 2020 was reportedly 14% lower^5,35. Therefore, to better understand these potential fishery economic losses, ecosystem indicators, such as those quantified here (e.g., groundfish, anchovy, and market squid), may benefit fishery assessments by providing strategic advice on marine ecosystem state during the COVID-19 era.

In addition to providing recruitment indicators for a suite of commercially important taxa such as groundfish and market squid^36,37, our ecosystem survey has played a pivotal role for informing salmon management, understanding unusual marine mammal mortality events, and unraveling impacts from heatwaves to understand and mitigate whale entanglements in fishing gear^27,38. Thus, losing ecosystem monitoring data may be directly related to loss of insight for informing dynamic ocean management, leading to increased uncertainty⁷. The ecosystem status indicators presented here may be the only information available during 2020, to assess trends and variability of epipelagic forage species in this region. For example, abundance of YOY rockfish, sanddabs, hake, and market squid are below average, whereas the multi-year persistence of high anchovy abundance continues to dominate. Although reduced sampling likely impacted krill assessments, SDM approaches indicate krill species abundance (e.g., T. spinifera) should be higher than average, signaling a recovery from the 2019 large marine heatwave⁹. Ecosystem impacts from long-term change and recent unprecedented ocean-climate variability makes it difficult to lose an observation year; because of higher frequency fluctuation of environmental conditions, loss or monitoring data will reduce predictability of ecosystem state^6,27,39. To better prepare for this uncertainty, evaluation and parameterization of species interactions, coupled with ocean ecosystem models and sampling simulation studies, will contribute to strategic ecosystem-based fishery management.

Updates of many ecosystem indicators will be permanently lost due to the pandemic⁴. Our synthesis provides a basic framework and example for attempting to both recover and validate indicators informed by sparse data, and highlights the need for integration across available data streams, predictive ecosystem modeling, and ultimately a lesson in survey preparedness. Although NOAA fishery-independent surveys were mostly canceled during 2020^3,4, we were fortunate in our ability to be flexible and respond quickly to develop a contract for chartering a commercial fishing vessel to recover some survey effort. The inspiration from the recent unavoidable survey effort reduction workshop¹⁴ was invaluable for planning the limited survey and developing a power analysis to evaluate sensitivity and uncertainly of indicators derived from imperfect sampling. We recommend that other fisheries and ecosystem surveys conduct an analysis of the implications of unavoidable survey effort reduction, develop partnerships with researchers investigating top predator population and foraging ecology, and incorporate predictive ecosystem models to their best ability¹⁰. Seabirds are excellent ecosystem indicators and can contribute substantial information content in the absence and or minimization of fishery and ecosystem survey effort. Further, monitoring of seabirds and marine mammal populations at their colonies occurs in remote field locations via small research teams, making them ideal monitoring tools during a pandemic. In addition, as the charter vessel was unable to deploy oceanographic equipment, we also designed alternative survey plans using robotics, such as oceanographic gliders, equipped with acoustics to monitor fine-scale distribution of krill and forage fish. Although we were ultimately not able to deploy these devices in a time-effective manner, autonomous vehicles may offer substantial opportunities for integrating additional options into existing surveys and could have benefited ecosystem assessments during the COVID-19 era. We urge ecosystem scientists and resource managers to review what existing tools are available to inform ecosystem condition in the absence and or minimization of observational data. Although preparing for the unexpected is difficult, a diversified modeling approach may help overcome fishery management challenges attributed to the pandemic and future reductions of ecosystem monitoring.

Source: Ecology - nature.com