Improving seismic chimney detection using directional attributes

The chimney cube is a seismic body detection technique to facilitate the interpretation of seismic chimneys. The method uses multi-attributes and a supervised neural network to produce a chimney cube, i.e. a three-dimensional (3D) volume of the chimney probability. In the original method, single and multi-trace attributes are extracted in three vertically aligned windows. Within each window the energy and several trace-to-trace similarities are computed. Similarities are computed without correcting for structural dip. Here we discuss improvements made to the original algorithm by taking local dip and azimuth information into account. The similarity attributes are now calculated by following the local direction (dip and azimuth) from trace to trace. A technique we call “phase-locking” fine-tunes the calculated times at which trace segments are extracted for similarity calculations. The new network with these similarities and a new attribute called dip-variance yields a lower misclassification and a significantly higher resolution chimney cube as compared to the original algorithm.