Developing predictive models for bycatch in tuna fisheries
Historically, when people have considered the impact of fisheries on marine resources, their concerns have largely focused on the population status of the target species. In the last decade, however, the issue of bycatch reduction has been getting more attention.
[Bycatch problem in tuna fisheries]
In the eastern Pacific Ocean, purse-seine nets are used to catch tunas. In this fishery, tunas are detected in three ways: in association with floating objects ("floating object" sets), in association with herds of dolphins ("dolphin" sets), and as free-swimming schools visible at the surface. Incidental mortality of dolphins, sharks, sea turtles, and other species can occur during fishing operations for tunas. The incidental mortality of dolphins in this fishery was the first bycatch problem that attracted public attention. In 60's, hundreds of thousands of dolphins are estimated to have been killed annually incidental to fishing operations in dolphin sets. With the development of dolphin-release techniques by fishermen, national legislation and international agreements establishing quotas on incidental mortalities, and the implementation of a seminar program designed to educate fishermen on methods for avoiding dolphin mortalities, incidental mortality of dolphins has declined to less than three thousand animals annually since 1998.

[Prediction of Shark bycatch]
Although dolphins are rarely killed incidental to fishing operations in floating object sets, large amounts of bycatch of many other species can occur in these sets. We are working with Dr. Cleridy Lennert-Cody of the Inter-American Tropical Tuna Commission to analyze shark bycatch data in floating object sets. Annually, shark bycatch occurs in more than one third of the floating object sets. We are exploring new statistical techniques for the prediction of the occurrence of shark bycatch, including boosting methods, which are recently proposed discrimination methods. Our preliminary results suggest that boosting techniques give more stable predictions than existing discrimination methods for these data.
[Analysis of shark bycatch counts]
One notable characteristic of shark bycatch data is that there are many sets with zero bycatch, yet sets with large amounts of bycatch can also occur. Thus, we are developing a zero-inflated negative binomial regression model for the shark bycatch data. The zero-inflated negative binomial regression model assumes that there are two states: an "complete" state in which bycatch never happens, and an "incomplete" state in which bycatch might occur. In the incomplete state, bycatch counts are assumed to follow a negative binomial regression model. We have fit this model to shark bycatch data from floating object sets. The distribution of the predicted bycatch was found to be quite similar to that of the observed bycatch suggesting that the zero-inflated negative binomial is a reasonable model for these data. We will use this model to explore trends in shark bycatch rates.

Members
Mihoko Minami (The Institute of Statistical Mathematics)
Shinto Eguchi (The Institute of Statistical Mathematics)
Masanori Kawakita (The Graduate University for Advanced Studies)
Cleridy Lennert-Cody (Intra-American Tropical Tuna Commission)