Fishery Forecasting Based on Singular Spectrum Analysis Combined with Bivariate Regression

Abstract

The fishery of anchovy and sardine has a great importance in the economy of Chile; they are important resources used for internal consumption and for export. The forecasting based on historical time series is a fishery planning tool. In this paper is presented the forecasting of anchovy and sardine by means of the monthly catches in the Chilean northern coast ($18^{\circ }S - 24^{\circ }S$), during the period January 1976 to December 2007. The forecasting strategy is presented in two stages: preprocessing and prediction. In the first stage the Singular Spectrum Analysis (SSA) technique is applied to extract the components interannual and annual of the time series. In the second stage the Bivariate Regression (BVR) is implemented to predict the extracted components. The results evaluated with the efficiency metrics show a high prediction accuracy of the strategy based on SSA and BVR. Besides, the results are compared with a conventional nonlinear prediction based on an Autoregressive Neural Network (ANN) with Levenberg-Marquardt; it was demonstrated the improvement in the prediction accuracy by using the proposed strategy SSA-BVR with regard to the results obtained with the ANN.