2.1. Background

Directivity is a concept in which dip and azimuth information is used to improve attribute accuracy and object detection power. For example, let us consider the calculation of a similarity attribute. This attribute compares two or more trace segments by measuring their distance in a normalized Euclidean space. Two identical trace segments will yield an output value of one, while two completely dis-similar trace segments will return the value zero. In a horizontally layered earth this will work nicely, but in a dipping environment the results will deteriorate. So, instead of comparing two horizontally extracted trace segments we should follow the local dip to find the trace segments that should be compared. The process of following the dip from trace to trace is called steering and requires a steering cube as input. The steering plugin for OpendTect supports two different modes of data-driven steering: Central steering and Full steering. In Central steering the dip / azimuth at the evaluation point is followed to find all trace segments needed to compute the attribute response. In Full steering the dip / azimuth is updated at every trace position. The difference between 'no steering', 'central steering' and 'full steering' is shown in the following figures. Note that these figures show the 2D equivalent of steering, which in actual fact is a 3D operation.

A steering cube is computed in OpendTect using a sliding 3D Fourier analysis technique. A small (typically 7x7x7) cube is transformed to the Fourier domain where the maximum is determined. The maximum value corresponds to the dip, which is stored in the steering cube in two components: inline dip and crossline dip.

Directivity also plays a role in defining attribute sets that are tuned to a particular seismic object. For example, in chimney detection we utilize the knowledge that chimneys are vertical bodies with a certain dimension by selecting attributes in three vertically oriented attribute windows. Open the default chimney attribute set and notice that all attributes occur three times with different time windows (-120,-40 / -40,+40 / +40,+120). When we feed these attributes to a neural network, the network learns that the responses above, around and below the evaluation point are similar when the location belongs to a chimney but are not the same when the location is not a chimney.

For chimneys the windows obviously must be arranged vertical. Similarly you can argue that for flat spot detection one should use horizontal windows. But can we also use this concept for e.g. fault detection? The answer is yes. You should steer the attributes along the fault plane directions. The problem of course is that you do not know the fault plane direction, neither can it be calculated directly from the seismic data. However, we have successfully calculated faultplane directions from a predicted faultcube and used these directions to improve TheFaultCube®. The process is called iteration and can be performed in OpendTect as follows:

1. Create a fault cube in the same way as you would create a chimney cube.

2. Filter the first generation fault cube (FC-1) with a velocity fan filter (this is a 3D-kf filter that must be defined in the attribute definition window).

3. Create a steering cube from the dip-filtered FC-1.

4. Extract new attributes from the original data steered along the faultplanes where needed and possibly extract attributes from FC-1 to create FC-2 using the same process as in step 1.