Estimates of marine mammal, sea turtle, and seabird bycatch in the California large-mesh drift gillnet fishery : 1990-2018

Bycatch of marine mammals, sea turtles, and seabirds in the California swordfish drift gillnet fishery during 1990-2018 is estimated using random forest regression trees. Tree estimates are compared with annual ratio estimates generated from the same observer data. Biases associated with ratio estimators (systematic under-and overestimation of bycatch) are notable when observed bycatch is rare, bycatch rates are inferred from within-year data, and observer coverage is low. Estimates from regression trees result in more stable annual bycatch estimates with better precision, because estimates are informed by all available data. Even in years without observed bycatch, expected values from regression trees are typically positive (sometimes fractions of animals) and include estimates of error, whereas corresponding ratio estimates are zero and lack error estimates. Regression tree bycatch models include oceanographic, location, and gear variables used as predictors to estimate bycatch at the fishing-set level. Variables used in models were identified with ‘balanced random forest’ classification trees that deliberately oversample sets with observed bycatch to overcome zero-inflated data signal-to-noise challenges. This method was previously validated with a simulated rare bycatch dataset where significant predictor variables were correctly identified in most cases, even when simulated bycatch events represented <1% of all data (Carretta et al. 2017). 2020 NMFS (National Marine Fisheries Service) SWFSC (Southwest Fisheries Science Center) Submitted https://doi.org/10.25923/t49n-9h37 Public Domain 1864