Automated event picking in prestack hyperspace

Abstract

Seismic data mining is part of an interactive processing and interpretation workflow. The extraction of information will often have the prerequisite of picking reflection events. Methods that aid in automatically extracting information are required when handling large volumes of data. Migrated 3-D seismic data in prestack form (which includes the offset dimension) creates a 4-D hyperspace. An algorithm for tracking prestack reflection events in that hyperspace will be presented. The algorithm combines a range of techniques including supervised learning. Results of automated picking will be presented for migrated, prestack, field 3-D data. The algorithm was able to track a nominated reflection event in prestack hyperspace from a single seed pick. The results are superior to those produced using a 2-D gather-based approach and a correlation autopicker. A small number of manual picks are used to train a probabilistic neural network, which assigns each sample an event probability. These probabilities are updated using a set of flow features that propagate seed picks through the hyperspace. Flow features constrain possible picking locations based on inter-relationships with nearby picks and event probabilities in 4-D. The combination of the global 4-D event probability distribution and localised 4-D flow feature updates, creates a highly constrained algorithm. Evolution of a picked event is controlled by quantitative assessment of previously made picks. The algorithm provides a quantitative measure of the reliability of each pick.