Hierarchical models of fishing behavior by factory trawlers in a midwater-trawl fishery for Pacific hake (Merluccius productus).

摘要

The fishing behavior of factory trawlers in the Pacific hake (Merluccius productus) fishery at different spatio-temporal scales was studied using several modeling techniques. In Chapter 1, a Markov decision process model was developed for the scheduling of fishing operations on a factory trawler. Stochastic dynamic programming was used to obtain the optimal controls for setting and retrieving the net. The optimal controls generally consisted of a bin threshold that signals the vessel to start fishing and a catch threshold that signals the vessel to stop fishing. A range of simple "rule of thumb" strategies generated nearly as much net revenue as the optimal control, indicating that the reward surface is flat in the region of the optimal control.; In Chapter 2, a statistical analysis was conducted of mesoscale (5-50 km) movement patterns of factory trawlers. Generalized additive models (GAM) were used to examine influence of the catch rates of prior hauls on the distance between successive hauls. Results of GAM models suggested that deviations from the expected catch rate influence the decision to move from a local foraging area, as expected from optimization models of animal foraging. These decisions appeared to be based on relatively short time frames, such that information from only the most recent 1-2 hauls in the area are utilized. In addition, results indicated that the presence of other fishing vessels operating nearby reduces the probability that a vessel will leave an area.; In Chapter 3, a simulation model was developed for an individual factory trawler. The model integrated the results of the above analyses in a hierarchical model of decision-making at different spatio-temporal scales. Decision-making occurred at two scales: (1) choosing an area within which fishing will be conducted, and (2) scheduling haul setting and retrievals while fishing within an area. A novel aspect of the model was a procedure, based on the Kalman filter, for modeling information about local fish densities gained by searching and fishing. The simulation model was used to identify the optimal decision rules, and to evaluate the usefulness of indices derived from factory trawler catch data to monitor population abundance trends.