Modeling CPUE series for the fishery along northeast coast of India: A comparison between the Holt- Winters, ARIMA and NNAR models

Mathematical as well as statistical models not only help in understanding the dynamics of fish populations but also enables in short-term predictions on abundance. In the present study, three univariate forecasting techniques viz., Holt-Winters, Autoregressive Integrated Moving Average and Neural Network Autoregression were used to model the CPUE data series along northeast coast of India. Quarterly landings data which spans from January 1985 to December 2014 was used for building the model and forecasting. The accuracy of the forecast was measured using Mean Absolute Error, Root Mean Square Error and Mean Absolute Percent Error. Based on the comparison of the model, performance of Holt-Winter’s model was found to provide more accurate forecasts than the Autoregressive Integrated Moving Average and Neural Network Autoregression model. A Holt-Winters model with smoothing factors α = 0.172, β = 0, γ = 0.529 was found as the suitable model. The presence of seasonality in the series is evident from gamma value. An ARIMA model with one non-seasonal moving average term combined with two seasonal moving average terms was found to be suitable to model the CPUE series based on the Akaike Information Criteria. Among the Neural Network Autoregression models used to fit the CPUE series, a configuration of 13 lagged inputs and one hidden layer with 7 neurons provided the best fit