Seismic horizon mapping across faults with Growing Neural Gas

Abstract

Manual horizon mapping from 3D seismic data is a very time-consuming task. Most automated or semiautomated methods have undesired limitations, like the inability to find correlated horizon portions separated by interruptions, particularly seismic faults. We present a clusterization-based method to extract horizons that uses the Growing Neural Gas (GNG) algorithm. The sample representing each voxel is replaced by its local behavior in the vertical direction. Lateral similarity measures are implicitly stored in the sample distribution over the created clusters. The method was applied to real seismic data to extract horizons and performed well even when the surface was completely separated by seismic faults on volume data.