Prediction of silver hake distribution on the Northeast U.S. shelf based on the Gulf Stream path index

Abstract

Over the past ~40 years, the distribution of silver hake on the Northeast U.S. shelf is found to be significantly correlated with changes in the latitude of Gulf Stream path. The correlation coefficient between the fall Gulf Stream position and the center of biomass of spring silver hake reaches 0.75 when the Gulf Stream leads the silver hake for 6 months. Based on this lead-lag relationship and low-frequency variability of Gulf Stream position with a dominant periodicity of ~9–10 years, the Gulf Stream position is used as a predictor for the center of biomass of silver hake in linear autoregressive (AR) models. The goal of this study is then to optimize the AR model for the prediction of silver hake based on the observed changes in Gulf Stream position. Fall Gulf Stream position is first predicted out to 5 years using a 5th order AR model and the observed Gulf Stream position in preceding years. An optimization process is proposed to choose best AR coefficients based on a newly proposed combined skill parameter. Furthermore, the robustness of our Gulf Stream prediction is verified by comparing the observed Gulf Stream path index data from 2009 to 2012, which are not used for optimizing the AR model, and the predicted Gulf Stream path values for the same time period. We then use this predicted Gulf Stream position to further predict the center of biomass of silver hake in the subsequent spring. Three different methods are used and compared for the silver hake prediction. The predicted silver hake time series can explain as much as 69% of the variance of the observation for the 1st year prediction and 41% for the 5th year prediction. Our results indicate that including Gulf Stream as a predictor produces better prediction skills of silver hake center of biomass than the AR model prediction solely based on the observed silver hake time series.