Abstract
Joint interpretation of models from seismic tomography and inversion
of magnetotelluric (MT) data is an efficient approach to determine
the lithology of the subsurface. Statistical methods are well established
but were developed for only two types of models so far (seismic P
velocity and electrical resistivity). We apply self-organizing maps
(SOMs), which have no limitations in the number of parameters considered
in the joint interpretation. Our SOM method includes (1) generation
of data vectors from the seismic and MT images, (2) unsupervised
learning, (3) definition of classes by algorithmic segmentation of
the SOM using image processing techniques and (4) application of
learned knowledge to classify all data vectors and assign a lithological
interpretation for each data vector. We apply the workflow to collocated
P velocity, vertical P-velocity gradient and resistivity models derived
along a 40 km profile around the geothermal site Groß Schönebeck
in the Northeast German Basin. The resulting lithological model consists
of eight classes covering Cenozoic, Mesozoic and Palaeozoic sediments
down to 5 km depth. There is a remarkable agreement between the litho-type
distribution from the SOM analysis and regional marker horizons interpolated
from sparse 2-D industrial reflection seismic data. The most interesting
features include (1) characteristic properties of the Jurassic (low
P-velocity gradients, low resistivity values) interpreted as the
signature of shales, and (2) a pattern within the Upper Permian Zechstein
layer with low resistivity and increased P-velocity values within
the salt depressions and increased resistivity and decreased P velocities
in the salt pillows. The latter is explained in our interpretation
by flow of less dense salt matrix components to form the pillows
while denser and more brittle evaporites such as anhydrite remain
in place during the salt mobilization.
Users
Please
log in to take part in the discussion (add own reviews or comments).