Preliminary results of forecasting of the loop current system in Gulf of Mexico using robust principal component analysis

Abstract

Long term forecasting of the Loop Current (LC) and its eddies, also called the Loop Current System (LCS) in Gulf of Mexico (GoM) region is crucial for the GoM communities to take adequate preparations to avoid undesired outcomes of this natural phenomena. In this paper, a new approach is developed to forecast the LC and its eddies. The sea level anomaly (SLA) data of the GoM is utilized as observations. The key element of the proposed approach is based on a time series data decomposition strategy, Robust Principal Component Analysis (RPCA). The time components of SLA data obtained by RPCA are fed to a recurrent neural network, the Long Short-Term Memory (LSTM) algorithm, which predicts the timing of eddy separations from the extended LC, and their positions. In the experimental study, observations of sea surface height variations during a period of 23 year were used to train the LSTM network and observations from two additional two years to validate the performance of the prediction model. As shown in the paper, the proposed model can predict the movements of the LCS six weeks in advance.