Horizon Detection in Seismic Data: An Application of Linked Feature Detection from Multiple Time Series

Seismic studies are a key stage in the search for large scale underground features such as water reserves, gas pockets, or oil fields. Sound waves, generated on the earth’s surface, travel through the ground before being partially reflected at interfaces between regions with high contrast in acoustic properties such as between liquid and solid. After returning to the surface, the reflected signals are recorded by acoustic sensors. Importantly, reflections from different depths return at different times, and hence the data contain depth information as well as position. A strong reflecting interface, called a horizon, indicates a stratigraphic boundary between two different regions, and it is the location of these horizons which is of key importance. This paper proposes a simple approach for the automatic identification of horizons, which avoids computationally complex and time consuming 3D reconstruction. The new approach combines nonparametric smoothing and classification techniques which are applied directly to the seismic data, with novel graphical representations of the intermediate steps introduced. For each sensor position, potential horizon locations are identified along the corresponding time-series traces. These candidate locations are then examined across all traces and when consistent patterns occur the points are linked together to form coherent horizons. 1. Introduction There are many applications with multiple time-series data recorded at high frequency, such as environmental science, geophysics, financial trading, and internet marketing. Often the aim of the analysis is to identify events which occur across many data series but not necessarily at the same time. Rapid acquisition means that a full analysis may be impractical and so simple yet reliable methods are needed. An exemplary application is geophysical surveying where large datasets are obtained but only limited questions need answering, such as the following: is there an oil field? And if so, what is its location? In such cases, there is no need to perform a full 3D reconstruction of the study area and then interpret the reconstruction. Instead, it is possible to produce an answer directly from the data. A seismic study will form motivation for the proposed approach, but the method could be equally applied to other situations where the aim is to identify coherent features across multiple time series. A seismic dataset consists of records of reflected signals measured by a number of seismic sensors placed at the nodes of a lattice on the earth surface. Figure 1 presents a diagram of