Analysis and prediction of satellite altimetric sea level anomalies in the tropical Pacific Ocean

  • 陳 怡靜

Student thesis: Master's Thesis


Sea level rise one of the consequences of global warming has a substantial impact on economic and living environment to even damage human life and property Therefore continuous monitoring and analyzing sea level signals are extremely important for understanding ocean characteristics while accurate predicting of sea level changes is also indispensable for coastal management to prevent or reduce disaster resulting from sea level rise In this research three kinds of Empirical Orthogonal Functions (EOFs) including conventional EOF Complex EOF and Trend EOF were applied to decompose monthly gridded sea level anomalies data derived from satellite altimetry to extract dominant signals in spatial and corresponding temporal domains for finding out the ocean phenomena After examining the leading corresponding principle components (PCs) derived from the conventional EOF or Complex EOF with ENSO index we discover that ENSO signal is the dominant phenomena in the Tropical Pacific Ocean and the correlation coefficients of the PCs and Multivariate ENSO Index (MEI) are up to roughly 0 9 Moreover complex EOF can show information about the signal propagation in the decomposed modes Trend EOF can directly extract the trend which is presented in first mode from altimetric sea level anomalies with the estimated rate of 2 6 mm/yr during 1993-2013 in tropical Pacific Ocean Not only focus on analyzing the ocean signal but also interested in sea level prediction Autoregressive Integrated Moving Average (ARIMA) model and Support Vector Regression (SVR) were used to predict one-year time series of the first five PCs derived from the conventional EOF decomposition Afterwards we reconstructed gridded sea level anomalies using the combination of predicted PCs and spatial modes The one-year predicted results in the tropical Pacific Ocean indicate that SVM demonstrates a better performance in predicting with the root mean square error (RMSE) of differences between the SVR predicted and observed mean sea level anomalies at 1 mm when RMSE differences of ARIMA predicted and observed sea level is 1 5 mm Sixty-two percent of the study areas can be perfectly predicted by SVR with the coefficient of determination larger than 0 9 while ARIMA just can predict well in the 42% of the area
Date of Award2015 Aug 26
Original languageEnglish
SupervisorChung-Yen Kuo (Supervisor)

Cite this