Improvement of Decomposing Results of Empirical Mode Decomposition and its Variations for Sea-level Records Analysis

Abstract

The performance of empirical mode decomposition (EMD) and its variations such as ensemble EMD (EEMD), complete EEMD with adaptive noise (CEEMDAN) and improved CEEMDAN (impCEEMDAN) are tested using artificial signal tests. In the artificial signal test, intrinsic mode functions (IMFs) are obtained using EMD, EEMD, CEEMDAN, and impCEEMDAN and then compared to prescribed oscillations in an artificial a priori known signal. In all cases, extra and redundant modes are found due to residual noises. Furthermore, the low frequency modes are generally distorted. To overcome this problem a novel approach for reconstructing IMFs is proposed, where low-energy redundant modes are merged to one common signal based on statistical significance tests by comparing the energy-density of IMFs with energy-density spread function of white noise with similar scale. Artificial signal tests illustrate that the mode reconstruction method works well in approximating the prescribed true modes. Overall, the impCEEMDAN performs best with a reasonable fit to the original components and statistically significant low-frequency modes. The mode reconstruction method can improve the decomposing and filtering capacity of EMD and its variations.