410 



Fishery Bulletin 105(3) 



whale (G. macrorhynchus) and long-finned pilot whale 

 (G. melas). However, the two could not be reliably dis- 

 tinguished at sea by the observers and therefore were 

 combined in this analysis. 



Determination of serious injury to marine mammals 



The MMPA requires that incidental mortality of and 

 serious injury to marine mammals during commercial 

 fishing operations be reduced below the PBR benchmark. 

 Serious injury has been defined as an injury likely to 

 result in mortality (Angliss and DeMaster, 1998). A 

 workshop of National Marine Fisheries Service (NMFS) 

 and nongovernmental experts was convened in 1997 to 

 evaluate the types of injuries occurring in commercial 

 fisheries and to develop guidelines to determine if a 

 given marine mammal observed interacting with com- 

 mercial fishing gear was seriously injured. For small 

 cetaceans, including pilot whales and other delphinids, 

 it was concluded that animals that ingested hooks, 

 that were released with significant amounts of trailing 

 fishing gear, that were swimming abnormally, or that 

 suffered severe external trauma would be considered 

 seriously injured (Angliss and Demaster, 1998). Serious 

 injury determinations were made by NMFS staff on a 

 case-by-case basis after reviewing the observations and 

 comments of fishery observers. 



Logistic regression analysis 



Logistic regression was used to evaluate the effects of 

 environmental conditions and fishing practices on the 

 probability of interactions with pilot whales and Risso's 

 dolphins. The vast majority (>85%) of interactions with 

 marine mammals involved only one individual with a 

 set. Therefore, the data were transformed to a binary 

 response variable indicating whether or not a marine 

 mammal interaction was observed. 



Only longline sets made along the U.S. Atlantic and 

 Caribbean coasts, including the Florida East Coast 

 (FEC), South Atlantic Bight (SAB), Mid-Atlantic Bight 

 (MAB), Northeast Coastal (NEC), and Caribbean (CAR) 

 fishing areas (Fig. 1), were included in this analysis 

 because interaction rates in most other areas were ex- 

 tremely low or zero. Although Risso's dolphin interac- 

 tions were observed during experimental fishing in 

 the NED, the fishing characteristics of these sets were 

 different from those of the normal commercial fishing 

 operations. Therefore, sets observed in the NED (experi- 

 mental and nonexperimental) were excluded from the 

 analysis. This included two observed interactions with 

 Risso's dolphins and none with pilot whales in nonex- 

 perimental fishing. The analysis focused on the Atlantic 

 fishing areas with the highest overall interaction rates 

 with pilot whales and Risso's dolphin and comprised 

 3187 observed longline sets. 



Explanatory variables were drawn from a broad suite 

 of data collected during each set by the fishery observer 

 (Table 1), or they were derived from the date and loca- 

 tion of the set (Table 1). Variables were categorized 



broadly as environmental conditions, space or time 

 variables, gear characteristics, fishing effort level, and 

 catch characteristics. However, each was considered 

 independently in developing the best fitting, most par- 

 simonious logistic regression model following strategies 

 for model selection outlined in Hosmer and Lemeshow 

 (1989). Briefly, each variable was examined individu- 

 ally to assess its explanatory power, and the subset 

 of significant single terms was included in the initial 

 model. Following this initial exploratory step, the single 

 terms and all two-way interactions were examined by 

 using stepwise selection. The explanatory power and 

 significance of each potential model term was examined 

 through chi-square tests and Akakie's information cri- 

 terion (AIC; Hosmer and Lemeshow, 1989; McCullagh 

 and Nelder, 1989). Terms were retained in the model if 

 they were significant (at a P-value <0.10) based upon 

 chi-square tests and if their addition resulted in a re- 

 duction in AIC. The resulting models were examined for 

 overdispersion by using the ratio between the residual 

 model chi-square and degrees of freedom (Burnham et 

 al., 1987; McCullagh and Nelder, 1989). For both the 

 pilot whale model and the Risso's dolphin model, this 

 ratio was approximately equal to one, which indicated 

 no significant problems with overdispersion and hence 

 accurate estimates of variance for model parameters. 



Results 



Observed marine mammal interactions 



Between 1992 and 2004, a total of 200 interactions 

 between marine mammals and pelagic longline gear 

 were observed. Of these, there were 10 observed mor- 

 talities and 94 observed serious injuries (Table 2). One 

 hundred of the observed interactions were with pilot 

 whales, 64 were with Risso's dolphin, and all other spe- 

 cies had six or fewer observed interactions. Other marine 

 mammal interactions of note occurred with a killer 

 whale, unidentified beaked whales (family Ziphiidae), a 

 pygmy or dwarf sperm whale (Kogia spp.), a northern 

 bottlenose whale (Hyperoodon ampullatus), and two 

 baleen whales (Table 2). 



The majority of marine mammal interactions were ob- 

 served near the shelf break along the U.S. Atlantic coast 

 between North Carolina and Georges Bank (Fig. 1). Pilot 

 whale interactions were concentrated in the MAB fishing 

 area between North Carolina and New Jersey, whereas 

 Risso's dolphin interactions primarily occurred in the 

 NEC region (Fig. 1). Thirteen Risso's dolphins, one pilot 

 whale, and eight additional animals of various species 

 interacted with fishing gear during experimental fishing 

 operations in the NED during 2001-03 (Fig. 1). 



The observed interaction rates (number of sets with 

 a marine mammal interaction/number of observed sets) 

 fluctuated across the time series. The highest rates 

 were observed during 1992-95 and in the most recent 

 years from 2000 through 2004 (Fig. 2A). There were 

 generally low interaction rates and very few pilot whale 



