2019 AAPG Annual Convention and Exhibition:

Datapages, Inc.Print this page

Enhancement of Anomalies in Time-Lapse Seismic Data by Means of Principal Component Analysis


Time-lapse seismic is part of a process that has become a crucial tool for the integrated management of hydrocarbon reservoirs. After proper seismic processing considering repeatability and cross-equalization methods, two 3D seismic images are subtracted to reveal anomalies of dynamic changes in the reservoir. However, this method might not be effective due to low signal-to-noise ratio (SNR) caused by poor repeatability or other factors. In order to improve results, a different approach is proposed to detect anomalies rather than applying plain algebraic difference between base and monitor cubes. The 4D seismic data was analyzed using a multivariate statistics tool, principal component analysis (PCA), which is a simple non-parametric numerical method for extracting relevant information from mixed-nature datasets. For this paper, two post-stack seismic volumes from different years of Teal South field data at a shallow sandstone reservoir were used. Base seismic was acquired in July 1997 and monitor acquisition was performed in April 1999. First, amplitude balance was applied to compensate for different levels of energy. Then a waveform correction was conducted which adjusts the spectrum of the entire monitor to match the spectrum of the base. Finally, the major seismic reflection events present in the monitor cube were precisely vertically shifted to match the same time position as those in the base cube. After this data conditioning process, a subtraction between cubes is carried out to get the conventional 4D anomaly. The PCA method was applied to this standard workflow on base and monitor data after conditioning, generating two other seismic volumes, PC1 and PC2, in which the former can be interpreted as the seismic response of the structural geology and the latter as the anomaly due to fluid and pressure changes. Numerical experiments of PCA method application on real data showed good results in the identification of the time-lapse anomaly when compared to the conventional algebraic difference. In the anomaly data (PC2) the SNR is increased while the anomaly still remains, showing that the PCA method can be seen as an expansion of the conventional method of plain algebraic difference. Additionally, the proposed methodology seems more suitable for a screening analysis of 4D datasets with more than one monitor because all the anomalies can be summarized automatically in the second component.