Week Ahead Time Series Prediction of Sea Surface Temperature Using Nonlinear Autoregressive Network with and Without Exogenous Inputs

Abstract

The seas and the oceans have existed since the origination of the Earth and play a very significant role in balancing the ecosystem. Coincidentally, they are the primary and early indicators of global imbalances that maliciously creep in the nature. Sea surface temperature is the foremost parameter that is observed to have a direct or indirect relationship to such variations. Literature surveys have revealed linkages of the sea surface temperature with various bioparameters that further affects not only the aquatic flora and fauna but the entire global biome, at times causing drastic changes. In the Pacific Ocean, a window defined by +5° N latitudes and +180° E longitudes is selected. The time series has data collected by the array of Tropical Atmosphere Ocean (TAO) across the equatorial Pacific Ocean from 1980 to 1998 using around 70 moored buoys. National Oceanic and Atmospheric Administration’s (NOAA)’s Pacific Marine Environmental Laboratory (PMEL) has developed moorings that measure air temperature, relative humidity, sea surface temperatures, and surface winds readings. A nonlinear autoregressive network is first optimized, and then using exogenous inputs, the values of sea surface temperature is forecasted for a week and compared. As many as 10 different error parameters are tabulated to provide insight into the error analysis.