In geological site investigations, often more than one method or data source is used
to measure the same property, but the distribution of the measurements and the
accuracy of each method may be different. In this case study pre-existing data are
used for geothermal exploration. The study aims to describe the architecture of a
geothermal reservoir as accurate as possible by a 3D structural stratigraphic model.
The reservoir is situated in the Lower Permian rocks of the NE German basin located
at the drill site Groß Schönebeck, 30 km north of Berlin. The rock of the reservoir
is faulted and located in around 4000 m depth. The used pre-existing data come from
former gas exploration and encompass 132 km of 2D seismic sections and data from 4
deep wells. Combining these data sets from various sources (different tools) and
various age (1964 to 1990) means to combine different levels of accuracy. The
interpolation of the seismic data and well data with a commonly used bi-cubic spline
function resulted in bumps and bows in the horizon grids at well data points. The
bumps and bows, however, do not reflect the real structure of a geological layer but
are the effect of a spatially not related interpolation of the differently accurate
data. An algorithm that deals with more than one variable at a time, is expected to
improve the estimated depth value. Therefore, the unknown between the scattered
horizon data is estimated by kriging with external drift defining the seismic data as
auxiliary variables in order to calculate an external trend model that is fitted to
the well data defined as target variable. Effectively, the bows and bumps around well
data are removed because the horizon grids are geostatistically corrected by the well
data depth values. The fault surfaces are processed with a 2D minimum tension
technique and the final structural model consists of 82 faults, 101 fault blocks and
16 horizons. It is assumed that external drift kriging is the appropriate algorithm
for interpolation of pre-existing data, provided that these data reflect different
degrees of accuracy.
|