Filter assisted bi-dimensional empirical mode decomposition: a hybrid approach for regional-residual separation of gravity anomaly Animesh Mandal ⁎, Shankho Niyogi Department of Earth Sciences, Indian Institute of Technology, Kanpur 208016, India abstract article info Article history: Received 11 April 2017 Received in revised form 10 August 2018 Accepted 4 September 2018 Available online xxxx Adequate separation of regional-residual components from observed gravity anomaly is always a challenging task in gravity interpretation. Several techniques have been developed for effective regional-residual separation, however, no single approach can perfectly accomplish the job. In this work, a hybrid approach has been proposed with an objective to enhance the performance of Bi-dimensional Empirical Mode Decomposition (BEMD) in decomposing gravity anomaly by jointly employing low pass filter and BEMD. The paper discusses the efficacy of this hybrid approach in gravity anomaly separation from noisy synthetic and field data. Synthetic studies involving forward modelling of asymmetrically placed shallow and deeper spherical bodies with added Gaussian noises of different levels demonstrate that the proposed approach is more efficient in separating the regional-re- sidual components from observed gravity anomaly compared to the individual application of filtering or BEMD. It has an added advantage of suppressing the unwanted noises significantly. After successful test with synthetic data the proposed approach has been applied to field data and satisfactory results are obtained. Thus, the pro- posed hybrid approach is more effective in delineating gravity signature related to complex near surface features even from noisy gravity data. © 2018 Elsevier B.V. All rights reserved. Keywords: Gravity anomaly Regional-residual Low pass filter Bi-dimensional Empirical Mode Decomposition (BEMD) Hybrid approach 1. Introduction Observed gravity anomalies at different surface locations are due to the superposition of anomalies associated with lateral and vertical mass (density) distribution of geological features at different depths. Separa- tion of gravity anomalies caused by widespread deep seated (regional) geological features from that due to shallow subsurface (residual) mass distribution is a crucial step in quantitative interpretation of gravity anomalies. This can be accomplished using various standard ap- proaches, e.g.,graphical smoothing (Telford et al., 1990), second vertical derivative (Henderson and Zietz, 1949), trend surface analysis (Agocs, 1951; Merriam and Harbaugh, 1964; Beltrao et al., 1991; Roach et al., 1993), filtering (Griffin, 1949; Zurflueh, 1967; Spector and Grant, 1970; Pawlowski and Hansen, 1990) to name a few. In recent years, sev- eral new approaches have been introduced, e.g.,based on 3D inversion algorithm (Li and Oldenburg, 1998), Wavelet and spectrum analysis (Fedi and Quarta, 1998; Xu et al., 2009) etc. However, the separation of regional and residual components is a difficult task and it does not have an absolute solution. Thus, there always remain scope for improve- ment, e.g.,the problem of spectral overlapping of regional and residual anomalies in Fourier domain spectral filtering has been overcome by the use of wavelet domain decomposition (Zhang et al., 2009). Wavelet analysis can simultaneously provide the time and frequency informa- tion of a signal, thereby capturing the non-stationarity aspects of the data. However, the requirement of a predefined basis wavelet for all data throughout a signal makes wavelet analysis inefficient in capturing the non-linear behaviour of the data (Huang et al., 1998; Hassan and Pierce, 2008). These methods are unable to adaptively handle the input signals and are, therefore, not suitable for handling data character- ized by both non-linear and non-stationary nature (Huang et al., 1998; Hassan, 2005; Huang, 2006). Therefore, researchers have very recently applied new self-adaptive techniques, namely, Empirical Mode Decom- position (EMD) and its variant in two dimensions, i.e.,Bi-dimensional Empirical Mode Decomposition (BEMD) to satisfactorily decompose the non-stationary and non-linear geophysical data into inherent com- ponents. The superiority of these techniques is a result of their adaptive nature while handling the data and having no prior basis for the decom- position process (Huang et al., 1998, 2003). EMD decomposes an origi- nal one dimensional signal into constitutive components termed Intrinsic Mode Functions (IMFs) and can express the original signal as a sum of IMFs. Extending the concept of EMD in two dimension, the BEMD was developed; and like EMD it also decomposes a two dimen- sional signal into constitutive Bi-dimensional Intrinsic Mode Functions (BIMFs) (Huang et al., 2010). One major disadvantage of this technique is the non-continuity of constitutive signals and mixing up of the energy Journal of Applied Geophysics 159 (2018) 218–227 ⁎ Corresponding author. E-mail address: animeshm@iitk.ac.in (A. Mandal). https://doi.org/10.1016/j.jappgeo.2018.09.003 0926-9851/© 2018 Elsevier B.V. All rights reserved. Contents lists available at ScienceDirect Journal of Applied Geophysics journal homepage: www.elsevier.com/locate/jappgeo