Open AccessForecasting World Tuna Catches with ARIMA-Spline and ARIMA-Neural Networks Models
Author(s) -
Boonmee Lee,
Suhartono Suhartono,
Apiradee Lim,
Sung K. Ahn
Publication year - 2021
Publication title -
walailak journal of science and technology
Language(s) - English
Resource type - Journals
eISSN - 2228-835X
pISSN - 1686-3933
DOI - 10.48048/wjst.2021.9726
Subject(s) - autoregressive integrated moving average , univariate , tuna , box–jenkins , econometrics , time series , goodness of fit , statistics , geography , computer science , fishery , mathematics , multivariate statistics , fish <actinopterygii> , biology
Tuna is a renewable resource that has been managed regionally, but its worldwide capacity for regeneration is still little known. A time-series dataset of tuna catches was used to develop nonlinear univariate models for monitoring the sustainability of tuna catches. Two approaches were compared: 1) fitting an ARIMA-spline model to the volume of annual tuna catches and 2) combining neural networks with an ARIMA model to fit the annual changes in volume. These models offer competitive forecasting performance with small percentage errors. By averaging results of the best model developed in each of these approaches, our ensemble forecast predicts that world tuna catches will reach the optimal level of 5.09 million tons in 2025, remain stable thereafter until 2033, and start decreasing about 0.78 % annually. These models could be used by regional fishery management groups to discover discrepancies between such projections and other science-based estimations of the maximum sustainable output.HIGHLIGHTSAnARIMA-spline model is practical for forecasting time series with uncertainties and complex interaction of variablesThe plausibility of forecasts is essential as the goodness of fit for statistical model validationThe ensemble forecasts of results from modelling both catches and the changes of catches offer an alternative view for monitoring trend of fishery practicesGRAPHICAL ABSTRACT