The effect of random and density‐dependent variation in sampling efficiency on variance of abundance estimates from fishery surveys

Abstract

Abundance indices (AIs) provide information on population abundance and trends over time, while AI variance (AIV) provides information on reliability or quality of the AI. AIV is an important output from surveys and is commonly used in formal assessments of survey quality, in survey comparison studies, and in stock assessments. However, uncertainty in AIV estimates is poorly understood and studies on the precision and bias in survey AIV estimates are lacking. Typically, AIV estimates are “design based” and are derived from sampling theory under some aspect of randomized samples. Inference on population density in these cases can be confounded by unaccounted process errors such as those due to variable sampling efficiency (q). Here, we simulated fish distribution and surveys to assess the effect of q and variance in q on design‐based estimates of AIV. Simulation results show that the bias and precision of AIV depend on the mean q and variance in q. We conclude that to fully evaluate the reliability of AI, both observation error and variability in q must be accounted for when estimating AIV. A decrease in mean q and an increase in the variance in q results in increased bias and decreased precision in survey AIV estimates. These effects are likely small in surveys with mean q ≥ 1. However, for surveys where q ≤ 0.5, these effects can be large. Regardless of the survey type, AIV estimates can be improved with knowledge of q and variance in q.

Diverse integrated ecosystem approach overcomes pandemic-related fisheries monitoring challenges

“…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 predators 24 .…”

“…Consideration of the impacts of unavoidable survey effort reduction is an increasingly high priority for marine resource management agencies worldwide, as research budgets shrink or hold constant and the costs associated with such surveys increase 11 – 13 . Perhaps serendipitously, the most recent effort to consider strategies for adapting to such impacts was held at the earliest stages of the 2020 pandemic 14 .…”

“…Using several modeling approaches, we demonstrate how we updated ecosystem indicators using limited data and evaluated their robustness so as to maintain confidence in the trends and variability of key ecosystem components. We applied models that better account for spatially biased sampling and used resampling of past data to estimate changes in the accuracy and precision of estimates given observed (and potential) loss-of-survey effort 13 , 14 , 19 . We developed new models, trained on past survey data, including seabird diet observations, to evaluate abundance of young-of-the-year (YOY) rockfish ( Sebastes spp.)…”

Improving estimates of species distribution change by incorporating local trends

“…We expect that the predictions in our example application in this study are robust to the sensitivity of the method to spatiotemporal variation because the estimated spatiotemporal variance is much lower than the spatial variance for groundfish species in this system. Observation error in trawl surveys can include a wide range of values as a result of variance in sampling efficiency (Kotwicki and Ono 2019), but relating such values to the observation error scale parameter evaluated in our simulations requires additional research. Additional ways to constrain the variance parameters, such as developing informative Bayesian priors from similar surveys might extend the detectability of local trend structure over the models used here.…”

Reducing uncertainty in survey abundance estimates by considering alternative designs and estimators: a case study with 3 species in the Gulf of Alaska

“…Until recently, it has been difficult to make direct comparisons between different estimators and sampling designs because of a lack of an appropriate simulation framework. However, Kotwicki and Ono (2019) developed a framework for simulating distributions of fish population density with a spatiotemporal model conditioned on historical catch and environmental survey data for a number of commercially important species in the GOA. We used this operating model to simulate the "true" distribution of abundance and the results of simulated sampling with different designs.…”