Bayesian hierarchical model of lake whitefish cohort strength from sparse trawl data

Recruitment indices for rare or intermittently recruiting fishes are needed to compare year classes and evaluate recruitment drivers, but sparse trawl data with many zero-catch observations complicate estimation. We used fall bottom trawl data from New York, Pennsylvania, and Ohio surveys in Lake Erie's central and eastern basins to estimate annual relative cohort strength of age-0 lake whitefish (Coregonus clupeaformis) from 1992 to 2021 and evaluate whether a Bernoulli-Bernoulli presence-absence model retained enough information for an annual relative cohort strength index compared with a Binomial-Poisson count model. We fixed detection probability at 0.31 in the primary analysis and refit both models using alternative fixed values in sensitivity analyses. Among 2879 tows, 173 were positive and 368 fish were collected, with positive catches ranging from 1 to 20 fish. Annual catch per unit area and both models recovered a similar recruitment pattern, with variable recruitment from 1992 to 2005, little to no recruitment from 2006 to 2014, and renewed recruitment in most years from 2015 to 2021. Cohort rank order was stable across fixed detection values (Spearman r_s = 0.996 to 1.000), and annual median estimates maintained high agreement with the primary analysis (Pearson r = 0.966 to 1.000). However, Bernoulli-Bernoulli estimates were not one-to-one with Binomial-Poisson estimates, and relative magnitude depended on assumed detection probability. These results indicate that the Bernoulli-Bernoulli simplification is adequate for recovering cohort strength patterns, but the Binomial-Poisson model is more appropriate for distinguishing relative cohort strength among years.