The traditional approach in seismic imaging assumes homogeneous layers with
constant seismic velocities. In the real earth, layers have numerous small
-scale variations of properties (e.g., the seismic P-wave velocity) which a
re so irregularly distributed, that they can no longer be described by dete
rministic models; statistical approaches have to be used instead. These var
iations can only be described by their stochastic properties. In this paper
, every distinctive layer is described by its mean velocity, its variance,
and a spatial autocorrelation function. This velocity model is transformed
into a model for the autocorrelation of the corresponding impulse response.
The parameters for this model are then estimated in the least-squares sens
e using stacked and migrated seismic sections as input. The method is valid
ated with synthetic data, and then applied to a real seismic data set where
additional borehole data is used to verify the estimation.