Evaluation of time-area closures to reduce incidental sea turtle take in the Hawaii-based longline fishery: generalized additive model (GAM) development and retrospective examination

Generalized additive models (GAMs) of sea turtle take in the Hawaii-based longline fishery were developed at the National Marine Fisheries Service (NMFS), Pacific Islands Fishery Science Center (PIFSC) to identify time-area closures that would effectively reduce interactions with sea turtles while minimizing hardship to longline fishermen. Detailed observations gathered by NMFS Southwest Region (NMFS-SWR) and NMFS Pacific Islands Region (NMFS-PIR) observers assigned to the longline fleet were used to develop the GAMS. The GAMs were then used in predictive mode to estimate turtle take over the entire longline fleet using federally mandated logbook data. High resolution environmental data were merged with the fishery data in an attempt to find useful covariates of turtle take. Computer simulation was used to assess the impact of seasonal (monthly resolution) closures or spatial (whole degrees of latitude/longitude resolution) closures over a systematic grid of 361,194 possible closure scenarios.
Leatherback turtles were of primary concern because of their endangered status. Immediate impacts to the fishery were measured by predicting the fraction of the fleet displaced spatially or temporally by the proposed management action. Long-term and financial impacts were also estimated using models of fishing effort reallocation and predicted catch rates of the displaced fishing effort coupled with market revenue data. Variability was addressed by the randomization procedure called bootstrapping. "Efficient frontier" analysis was used to visually determine the efficacy of proposed management scenarios. This approach is used primarily in Modern Portfolio Theory but has wide applicability for the identification of optimal solutions in a complex setting.
Due to the widespread patterns of leatherback turtle take (primarily in space, but also in time), it was difficult to define an optimal management scenario that could substantially reduce leatherback takes with a minimal impact to the fishery. However, the Emergency Closure (November 1999) of the fishery was shown to be quite distant from the efficient frontier. The GAM results were evaluated by using a substantially larger (4.5X) database of more recent observer data, facilitated by the increased rate of observer coverage of the fleet as mandated by the federal court. These findings indicate that the initial time/area closure analysis was robust with respect to general patterns of turtle take in time and space for loggerheads and leatherbacks, the two species primarily encountered by the fishing fleet.