Comparison of design- and model-based approaches for estimating relative abundance of American horseshoe crab in the Delaware Bay area

Objective The American horseshoe crab Limulus polyphemus is harvested for use in bait and biomedical industries. The Atlantic States Marine Fisheries Commission began managing the Delaware Bay stock in 1998. We developed two-stage hurdle models for this stock to evaluate the influence of external factors on CPUE, estimate relative abundance, and compare model- and design-based estimates. Methods We conducted stratified random trawl surveys along the mid-Atlantic coast (New Jersey–Virginia) annually since 2003, with strata based on bottom topography and distance from shore. Hurdle models (delta-lognormal) were developed for all six demographic groups (immature males, newly mature males, mature males, immature females, newly mature females, and mature females) to estimate year effects. Variable selection utilized forward stepwise regression, and quantile intervals for year effects were calculated using a 1,000-iteration bootstrap analysis. Predictive power was assessed through k-fold cross validation. Results Design- and model-based approaches produced similar relative abundance estimates across demographic groups, differing significantly in only a few years. Longitude, average depth, distance from shore, time of day, and month consistently affected CPUE. September yielded the highest mean CPUE for most groups. Night trawls generally resulted in higher abundance estimates. Relative abundance fluctuated over time. Conclusions Model- and design-based estimates of relative abundance were similar, although model-based estimates were typically more precise. Relative abundance varied over time, with a downward trend for newly mature females since 2016. We recommend using model-based estimates for future management decisions. Future research should focus on improving survey stratification and ensuring balanced sampling across months and times of day to validate this study’s findings.