In this section, we now seek to distinguish the geometric (specular) reflection contributions from the diffraction contributions to the total scattered response derived in (44). Without loss of generality, we define a local geometric/specular/Snell reflection wavefield such that

(45) |

The term is a geometric reflection coefficient, which may or may not be related to the Zoeppritz plane-wave coefficient, as discussed later. The unit vector is the direction of geometric reflection given by satisfying Snell's Law at the reflecting surface, as sketched in Figure :

(46) |

where is the unit normal to the reflecting surface at .Given this definition of , the following identity also relates to :

(47) |

Figure 1

We derive the following useful properties for :

(48) |

and

(49) |

Then, we evaluate the stress field associated with this geometric reflectivity as:

(50) |

Next, we decompose the total stress gradient into background and scattered stress field gradient components, as suggested by the Born analogy and physical interpretation of (36), (37) and (38):

(51) |

where is the gradient with respect to a smooth (reflectionless) background model, refers to the direct stress divergence wavefield in the smooth background medium, and refers to the scattered (reflected) stress divergence wavefield in the smooth background model. There is an implicit and as yet unexplored relationship between the Born model (36) and our Kirchhoff model (51):

(52) |

Our Kirchhoff decomposition (51) assumes a linear relationship between the total stress divergence in the true medium versus the direct and scattered stress divergences in the smooth background medium.

We evaluate the scattered stress divergence as:

(53) |

The terms of order vanish because the gradient is taken with respect to the smooth reflectionless background medium, and contains no source of scattering. Continuing to evaluate :

(54) |

where we have assumed that

(55) |

which implies that that *propagation* directions , traveltimes and spreading amplitudes of the scattered wavefield as it propagates
from the scattering point to the observation point, are approximately
the same in the smooth background as in the true background medium.
The divergence of the total stress field in the true medium is then:

(56) |

The body force equivalent with respect to is:

(57) |

Comparing (35) to (57) we see that

(58) |

and since

(59) |

the volumetric body force equivalent geometric reflection coefficient can be identified as

(60) |

Finally, we calculate the scattered field in terms of the geometric reflection coefficient using (39), (57), (45) and (40):

(61) |

For clarity, we define the angles shown in Figure :

(62) |

(63) |

(64) |

and

(65) |

Figure 2

With these definitions, (61) can be written in compact form as:

(66) |

To recap (66), the density at a
subsurface point is denoted , and the geometric reflection coefficient
at that point is . The amplitude terms *A*_{s} and *A*_{r} represent
the cumulative geometric spreading, transmission loss, Q-attenuation, etc.,
from the source and receiver to the subsurface point respectively.
The factor *A*_{r} also includes the vector component projection at the surface
location as defined by (42) and (43).
The term is the total traveltime from source at to the
subsurface point and back up to the receiver at .Finally, the diffraction weight
represents the angle between the anticipated geometric specular reflection
direction and the actual diffraction direction . In the case
of specular reflection when , and so .

We now make some general comments about the forward modeling theory given by (66).

- is in fact since we have removed the background field contribution by assuming a smooth background when solving for in (57).
- The scattered wavefield is
*linear*with respect to . This follows in that if the local reflector volume is subdivided into multiple smaller volumes, by linear superposition the volume integration result of (66) should not change. If the model is not ray valid for interfaces (i.e., interface topography varies faster than a spatial wavelength of the wavefield), then a linearized version of with respect to material property contrasts, such as (60), is required. Otherwise, if interfaces are ray valid, then by extension a coefficient which is nonlinearly related to material property contrasts may also be appropriate, such as Zoeppritz. - is symmetric with respect to and (reciprocity), but not .
- It is important to note that the volume integration (66)
models a single trace for any arbitrary source and receiver location.
There is no ``preferred'' acquisition geometry for this
synthesis; it works for shot gathers as well as random trace orientations.
In particular, (66) is perfectly adequate for
*constant offset*survey data, which will be the focus of the following inverse theory section.

This completes the forward modeling theory part of our paper that is necessary to proceed onward to the inverse estimation problem.

11/16/1997