An EEMD and BP neural network hybrid approach for modeling regional sea level change

摘要

Sea level prediction is essential and complicated in the context of climate change. Conventional methods developed for the prediction are still considered insufficient due to the complexity of the nonstationary and nonlinear sea level change. To improve the modeling accuracy of the sea level, this paper proposed a methodology combining the ensemble empirical mode decomposition (EEMD) and the back propagation (BP) neural network for monthly mean sea level record modeling in South China Sea. The results show that the EEMD can extract the signals with physical meanings according to their unique frequencies. The inputs of the BP, defined by the preprocessing of the original time series, turn out to be smoother and more regular, influencing the modeling in a positive way. The good performance of the hybrid method, with higher correlation coefficient (R = 0.89) and lower root square mean error (RMSE = 28.16 mm) between the modeling and the observed data, suggests an improved accuracy on sea level modeling than using the BP directly (with R = 0.76 and RMSE = 36.74 mm). This hybrid method can be further applied to sea level modeling in another region. The results of the study also suggest that the preprocessing of the original time series such as smoothing and denoising is significantly improving the modeling.