ADAPTIVE MANAGEMENT OF SPATIALLY REPLICATED GROUNDFISH POPULATIONS

Assessment techniques that recognize information shared among substocks can help reduce uncertainty about stock size and productivity, compared with methods that treat each subpopulation independently. We develop this methodology using time series of catch and relative abundance data for six subpopulations of yellowtail flounder, Limanda ferruginea, from the Northwest Atlantic. The Deriso/Schnute delay-difference production model was fit to the data, but the model parameters associated with population productivity and absolute population size were very uncertain. Estimates of parameter uncertainty were used to identify a set of models that are consistent with the observed data and imply different management policies. The value of these different policies was estimated with a stochastic simulation program that uses a Kalman filter to predict stock size for each model and Bayesian updating to assign a probability of each model being correct. With parameters shared among stocks and a shared response to external influences, in the order of 10-20 yr was needed to distinguish the correct model. Adaptive management simulations indicated that long-term yield was greater in simulations with adaptive learning and that a lower initial exploitation rate should be applied than would be the case if adaptive learning were ignored.