Karst is a type of solutional caves that are usually formed in soluble rock limestone when an aggressive fluid dissolves the rock. Subsurface karsting can produce a range of different structures, such as widened fissures and different types of cavities at all scales. These may locally be partly or completely filled by speleothems, sediments deposited by streams or breccia derived from stoping or collapse of the cave roof and walls (Nordli, 2009). Commonly known karst features are fluid-enhanced faults (Nissen et al., 2005), eroded caves, and sinkholes. The karst features are usually related with big petroleum reservoirs (Ford and Williams, 1989). Understanding the spatial distribution of the karsts is of great importance to the discovery of hydrocarbons and their optimal recovery from these reservoirs. However, this can be very difficult due to several reasons. One reason is that the karsts are not uniformly distributed and exhibit an extreme spatial variability. Another reason is that the karst features are usually below seismic resolution, which makes them difficult to map. In this paper, I propose a high-resolution characterization approach for mapping the karst features by utilizing time-frequency decomposition of 3D seismic data.

Time-frequency decomposition offers extra information when analyzing seismic data, which has been extensively used in seismic data processing and interpretation (Rodriguez et al., 2012; Fomel, 2013; Han and van der Baan, 2013; Liu et al., 2016b; Chen et al., 2015; Zhang et al., 2015a,2016; Reine et al., 2009). One of the most commonly used applications of time-frequency analysis is that deeper channels are usually of stronger amplitude at lower frequency in the time-frequency map, and the shallower flank of the channel has larger amplitude at higher frequencies in the time-frequency map (Cheng et al., 2016; Liu et al., 2016c,a). In addition, time-frequency analysis can be used in estimating attenuation, pore-pressure prediction, detecting seismic unconformities, and some implementation of seismic chronostratigraphy (Lin et al., 2013). Over the past decades, time-frequency decomposition has found a panoply of applications in seismic data processing and interpretation (Liu and Marfurt, 2007). Most time-frequency decomposition approaches are used in detecting low-frequency shadow (Castagna et al., 2003), detecting paleochannels (Liu et al., 2011a), reservoir characterization (Zhang et al., 2015b), structural and stratigraphic delineation (Chen et al., 2016; Liu et al., 2016c), arrivals picking of P or S wave modes (Pinnegar and Mansinha, 2003). In this paper, I propose an application of the time-frequency decomposition in probing the subsurface karst features. Four approaches with different analyzing resolutions are compared to offer more practical guidelines for utilizing the proposed approach.

Some researchers have mentioned the application of time-frequency decomposition approaches in delineating the subsurface karst feature, but most applications are not introduced in detail (Chopra and Marfurt, 2007; Fisher et al., 2010). The karst features are depicted based on the criteria that they are discontinuous in spatial domain and show frequency anomalies. The challenge in delineating karst feature lays in the fact that high-resolution delineation may result in many spurious frequency anomalies while low-resolution delineation may fail to depicted such valuable details. A method that can offer us freedom to control the characterization resolution and fidelity is in strong demand.