Modeling a Very Rare Event to Estimate Sea Turtle Bycatch: Lessons Learned

Estimation of sea turtle bycatch in the Hawaii-based pelagic longline fishery is discussed in the context of modeling a very rare event using heirarchical catch data collected by longline vessel captains and NMFS observers. Problems in bycatch model formulation, identification of efficient predictor variables, model selection, and model diagnostics are explored in detail. Models to predict bycatch of leatherback, olive ridley, and loggerhead sea turtles are developed using a variety of statistical tools including classification trees, generalized linear models, and generalized additive models. Prediction intervals for bycatch are derived using a nonparametric bootstrap algorithm. The statistical methods are applied to estimate annual bycatch and corresponding prediction intervals for all three turtle species in the years 1994-1999. Problems encountered in all aspects of the research and their resolution are discussed at length. Unresolved statistical issues are identified and suggestions for improving turtle bycatch estimation methods are offered.