Comparison of Neural Network and Linear Regression Models for Predicting El Nino Events

摘要

The El Nino phenomenon of the tropical Pacific Ocean consists of major climatic changes which produce significant changes in rainfall, storm systems over much of the two Americas, and various effects on the fisheries off the coast of the Americas. Sea surface temperature is a significant indicator of El Nino events, so this paper analyzes the effectiveness of two methods of forecasting it. One method is the use of a neural network, and the other is the use of linear regression. The methods are developed and tested using data from the Comprehensive Ocean-Atmosphere Data Set, which contains ocean surface temperature, air pressure, and wind data from 1884 to the present. The continuation of El Nino events is also forecast with some success. Comparisons of the prediction skills of the two techniques gave mixed results, with the realization that each forecasting method reacts differently to changes in input data representations.