An adaptive linear-mode decomposition for effective separation of linear and nonlinear seismic events, ground roll, and random noise

Abstract

Ground roll and random noise usually mask primary reflections in land seismic data. Different sets of signal processing methods are used to suppress these two noises based on statistical and/or transformation filtering. Among these methods, linear-mode decomposition (LMD) decomposes linear and nonlinear seismic events into amplitude-frequency modulated modes using the Wiener filter. Different combinations of these decomposed linear modes then can be used to represent different seismic events. However, LMD requires predefining the level of decomposition that must be selected carefully to avoid suboptimal binning, which can influence the fidelity of the decomposed seismic modes. To that end, we introduce an adaptive LMD (ALMD) that optimally separates seismic events, ground roll, and random noise. ALMD uses the correlation between the decomposed modes and the input data to determine the decomposition level. Consequently, an optimum decomposition divides the data into linear modes with minimum mixing. In addition, unlike conventional ground roll suppression methods, ALMD does not require estimating the slope or the frequency bandwidth of the ground roll. Moreover, ALMD automates the random noise segregation by separating modes as the signal, noise, and mixed modes, based on the permutation entropy and kurtosis criteria. ALMD iteratively decomposes mixed modes with remnant random noise until a signal or noise criterion is met. Using synthetic and real data examples, we demonstrate that the proposed ALMD is an effective method for separating desired linear and nonlinear events, unwanted ground roll energy, and random noise from the seismic data.