Quantitative prediction of fluvial sandbodies by combining seismic attributes of neighboring zones

The geological and geophysical characterization of hydrocarbon-bearing sandstones of fluvial origin is a challenging task. Channel sandbodies occurring at different stratigraphic levels (i.e., in a reservoir interval of interest as well as in overlying and underlying stratigraphic intervals) but overlapping in planview usually cause significant seismic interference due to limitations in seismic resolution: this can produce significant error in the prediction of sand location and thickness using seismic attributes. To mitigate the effect of seismic interferences by zones neighboring a target reservoir interval, a new method is proposed that combines multiple seismic attributes of the target interval and of its interfering neighboring zones, implemented by a supervised machine learning algorithm using support vector regression (SVR). Since the thickness of neighboring intervals causing seismic interference has a constant value of a quarter of a wavelength (1/4 lambda), the stratal slice corresponding with the top horizon of the target interval is taken as the base of a window of 1/4 lambda. to calculate seismic attributes for the overlying zone; similarly, the stratal slice corresponding with the bottom horizon is taken as the top of a window of 1/4 lambda. to calculate seismic attributes for the underlying zone. The proposed method was applied to a subsurface dataset (including a 3D seismic dataset and 255 wells) of the Chengdao oilfield, in the Bohai Bay Basin (China). The interval of interest is located in the Neogene Guantao Formation, whose successions are interpreted as fluvial in origin. This application demonstrates how the proposed method results in remarkably improved sandstone thickness prediction, and how consideration of multiple attributes further improves the accuracy of predicted values of sandstone thickness.

Automated summary

The study proposes a new method to improve seismic interference and enhance the prediction accuracy of sandstone thickness. By combining multiple seismic attributes and implementing supervised machine learning using support vector regression, the effectiveness of the method was demonstrated in an oilfield in China.