Time series analysis and forecasting techniques applied on loliginid and ommastrephid landings in Greek waters

Time series analysis techniques (ARIMA models), artificial neural networks (ANNs) and Bayesian dynamic models were used to forecast annual loliginid and ommastrephid landings recorded from the most important fishing ports in the Northern Aegean Sea (1984–1999). The techniques were evaluated based on their efficiency to forecast and their ability to utilise auxiliary environmental information. Applying a “stepwise modelling” technique, namely by adding stepwise predictors and comparing the quality of fit, certain inferences concerning the importance of the predictors were made.The ARIMA models predicted the test data very precisely (high R2), especially if the target time series contained a strong autoregressive character, after they were first differenced to obtain stationarity (R2 > 0.96). The disadvantage of the ARIMA, as with most statistical models, is their assumption that the relationships and system parameters remain the same across the observation and forecasting periods.The influence of temperature on catches was mainly investigated by applying neural models, which predicted the monthly landings with high precision (R2 = 0.89), even when incorporating in the model exclusively monthly SST descriptors. Similarly, ANN models of annual landings containing monthly mean temperatures provided high precision (R2 = 0.87) and valuable inference concerning the possible effect of the SST in certain months.Bayesian dynamic models also provided a high precision (R2 = 0.96). They combined the information of both environmental and landing time series, namely the monthly mean temperatures and the monthly seasonality of the landings. The impact factors estimated from the model have the form of time series representing the temperature effect.The results reveal that both the monthly and the annual landings can be predicted and that the Bayesian model is the best performer overall, characterised by a higher number of stable forecasts, and forecasts with higher precision and accuracy, than the other methods. It is evident, from application of the “stepwise modelling” technique, that the incorporation of temperature descriptors can significantly improve the model performance.