We present a procedure for producing a Bayesian DHI for low frequency passive seismic (LFPS) data. The approach utilizes two LFPS attributes to classify and determine the likelihood of hydrocarbon presence in the subsurface. The attributes are based on strength and variability of the empirically observed hydrocarbon tremor. An improved, more robust tremor energy measure based on the temporal characteristics of the signal is presented and used. An interpreter-driven Bayesian classification is employed both to accommodate uncertainties in the data and to provide a risk estimate. Prior knowledge from wells or structural information from active seismic can be incorporated into the analysis through interpretative interaction.
The process is tested over four fields with known surface projection of the oil-water contact (OWC). Correct and false hydrocarbon (HC) indication rates are shown to be statistically significant. Using a randomness test, it is established that the rates are very unlikely to be due to chance. The approach provides a rigorous method for producing hydrocarbon probabilities based on LFPS data.
Possible applications for the method include: (1) more informed drilling decisions over fields with limited or problematic active seismic data, (2) expanding production into areas near existing wells (exploitation).
Ambient seismic noise recordings made with broadband seismometers over hydrocarbon-bearing structures have been observed to exhibit spectral anomalies in the frequency range of about 1 to 6 Hz, most notably in the vertical displacement (Dangel et al., 2003; van Mastrigt and Al-Dulaijan, 2008; Saenger et al., 2009; Lambert et al., 2009). Although these anomalies have been found to be associated with hydrocarbons, there is no accepted theory for the generating mechanism. A preliminary theory states that the anomaly is a microtremor associated with the presence of a multiphase fluid: brine and hydrocarbons (Saenger et al., 2009). The tremor is visible as redirected spectral energy over the seismic background.
The microtremor strength was shown to be indicative of hydrocarbon (HC) presence in the subsurface (Saenger et al., 2009; Lambert et al., 2008). Both short-term (e.g. transient car noise) and long-term (e.g. daytime anthropogenic activities) interferences in the LF band require a careful editing of the time intervals over which the recording is analyzed. Using the quietest time windows the tremor energy is calculated by summing the short-time power spectral densities (PSDs). The energy is used as a low frequency (LF) attribute plotted on a map or in profile. Even when using minimum noise-affected data, the LF attributes often exhibit unexpected variability and the resulting maps are difficult to interpret. This is likely a consequence of using a sum of PSDs of a non-stationary signal.
Empirical evidence and forward modeling experiments suggest that tremor variability can also be indicative of HC presence (van Mastrigt and Al-Dulaijan, 2008; Lambert et al., 2009). The statistical process presented here for LFPS data employs two attributes: tremor strength and variability. Both attributes are based on the tremor energy distribution in time. We will show that distribution based attributes are more robust compared with the previously used methodology.