Sea surface temperature clustering and prediction in the Pacific Ocean based on isometric feature mapping analysis

Abstract

Isometric feature mapping (ISOMAP) is a nonlinear dimensionality reduction method and closely reflects the actual nonlinear distance by the view of tracing along the local linearity in the original nonlinear structure. Thus, the first leading 20 principal components (PCs) of low-dimensional space can reveal the characteristics of real structures and be utilized for clustering. In this study, a k-means algorithm was used to diagnose SST clustering based on ISOMAP. Warm and cold El Niño–Southern Oscillation events were subdivided into Central Pacific and Eastern Pacific types, and a two-dimensional cluster map was used to depict the relationship. The leading low-dimensional PCs of ISOMAP were considered as the orthogonal basis, and their trajectories demonstrated meaningful patterns that could be learned by machine learning algorithms. Predictions of SST in the Pacific Ocean were performed using support vector regression (SVR) and feedforward neural network (NN) models based on the low-dimensional PCs of ISOMAP. The forecast skills, the root-mean-square error (RMSE) and anomaly correlation coefficient (ACC), were comparable to those of current numerical models.