Introduction

Modern exploration for hydrocarbon requires more advanced techniques for each steps of seismic acquisition, processing, and interpretation (Chen et al., 2014a; Qu et al., 2015). The key step that connects geophysics and geology is interpretation. A clean pre-stack dataset can aid AVO-related inversion. A clean post-migrated image will help geologists make right decisions and reduce the interpretation risks (Xue et al., 2014; Yang et al., 2015b; Chen and Ma, 2014; Chen et al., 2015b). Due to different reasons, e.g. unpredictable noise or sampling issues, the raw seismic data might not be clean or coherent enough.

To solve this problem, we can not use the traditional random noise attenuation techniques to make the seismic reflections coherent, because discontinuous noise is not randomly spreaded across the whole profile. The traditional random noise attenuation technique (Chen et al., 2015a; Yang et al., 2015a), such as the $f-x$ deconvolution (Canales, 1984), $f-x$ singular spectrum analysis (Oropeza and Sacchi, 2011), Karhunen-loeve (KL) or singular value decomposition (SVD) filtering (Jones and Levy, 1987), will harm much of the useful reflections. The discontinuous problem of some seismic dataset can be solved using a smoothing operator. A mean filter is usually used to smooth the seismic reflections instead of solely removing random noise. However, mean filter requires exactly flattened reflections, otherwise much of the useful reflections will be lost. In addition, the parameters setting when utilizing the mean filter requires some effort in practice. Hale (2011) proposed a bilateral filtering approach for improving the smoothness and coherency along the local structure of seismic data. Considering the damages to useful energy in many smoothing and denoising algorithms, Chen and Fomel (2014,2015) proposed a signal-compensation approach for reducing the damages to useful reflections during the initial denoising processing.

In this paper, we propose a novel approach for enhancing seismic signal either in pre-stack dataset or in post-stack (-migrated) image. The novel approach utilizes the good smoothing ability of the empirical mode decomposition (EMD) based denoising approaches. Instead of using EMD to remove random noise (Chen et al., 2014b), here we only use EMD to enhance seismic reflections and to make seismic events more coherent. Because those EMD based denoising approaches are best used in horizontal reflections, a plane-wave flattening process is implemented before the traditional EMD based approach. Even though, the EMD based filtering approach does not require exactly flattened events. Unlike 1D mean and median filters, EMD based filtering can also preserve useful reflections with small dip angle. What is even more attractive is that EMD based denoising approach is adaptive. The only parameter we need to define is the number of dip components. Considering that, in practice, we commonly choose to remove the first EMD component in order to remove the highest oscillating components, the EMD based filtering is non-parametric. Because of the adaptivity and the superior performance of the EMD based smoothing in field seismic data processing, more and more researchers are turning to use this techinique as a blind-processing tool in order to deal with the rapidly increasing data size in modern seismic data processing (Gan et al., 2015b; Han and van der Baan, 2015). We use both pre-stack common midpoint (CMP) gather and post-stack image to demonstrate the performance of the proposed approach. Results show that the amplitude and the coherency of seismic reflections can be tremendously enhanced.