|Seismic analysis horizon matching fault tracking marked point process stochastic annealing|
Seismic data are pictures showing subsurface seismic reflectivity. Seismic data interpretations concern with building geological models with the aim to describe relationship between the seismic data and a priori geological information. The models are used for hydrocarbon exploration or other geotechnical investigations. This thesis work is motivated by a demand of computer-assisted interpretation of seismic data. Manual interpretations take too long and are almost always non-unique. The computer-assisted interpretation has advantages in that it provides a faster interpretation framework and a consistent workflow. This thesis focuses on automating horizon matching across a fault surface. Horizons are visible boundaries between certain sediment layers in seismic data. Faults are discontinuity surfaces across which horizons are cut and displaced. Automation tools, such as auto-trackers, are widely being used to assist horizon interpretation. However, they fall short of tracking horizons across faults. Horizon matching across faults is done after defining the fault surface, and it is about establishing the pre-fault geological continuity of horizons. In order to define the fault surface, a new semi-automatic fault-tracking method has been developed. It involves fault enhancing through the use of a log-Gabor filter followed by fault tracking based on an active contour model. The thesis proposes a new Bayesian-based fully automated method for horizon matching across a fault. The new method exploits the existing 3D spatial information and the multi-resolution nature of sediment layer structures in seismic images. A stochastic model has been defined under a marked point process framework. It models spatial data correlations and geological continuity constraints about sediment layers. The optimal matching solution has been found as a parameter which maximizes the stochastic model conditioned on seismic data. This is done by using a multi-resolution stochastic annealing algorithm. Application of the proposed method is compared with real 3D seismic data interpretations. Tests were made for 20 different fault patches taken from 4 seismic data sets. Automated matching results on 16 faults were considered acceptable after comparing them with references obtained manually. The method works well for planar faults and relies on the correct definition of horizon geometries. Additional tests show that the new method is more robust than previous methods while providing coarse-fine scale horizon matching.