3-D Geological Modeling from Legacy Seismic Data with Consideration of Uncertainties

Geological modeling is an essential component of reservoir characterization in geothermal exploration. A geological model aims to understand the spatial relation between geological features such as rock unit boundaries, horizons and discontinuities (unconformities, faults) at various scales. However, geological models can contain significant uncertainties – often due to limited information at depth. It is therefore imperative to use all available information, including legacy data. In the KarboEx2-project, legacy seismic data from former coal exploration in the region of North Rhine Westphalia are digitized and reprocessed with modern seismic processing workflows. In our contribution here, we investigate how uncertainties in the interpretation of this legacy data can be considered in subsequent geological modeling workflows.

In the context of model construction, this type of uncertainty relates to the real position of the input points, commonly associated to data uncertainties (seismic processing, picking and interpretation, etc.). Several position points result from the picking of the horizons on legacy seismic data. A simple way to address this type of uncertainty is to perform sampling from the data treating it as fully correlated (i.e., moving all points simultaneously) or fully uncorrelated (i.e., moving all points independently). However, geological errors are commonly correlated with distance. One possibility to consider spatial correlations is to generate a geological surrogate model with a lower-dimensional representation of modelled interface. In addition to accounting for different uncertainties in space, such a low-dimensional representation allows to perform inference, sensitivity analysis, etc. We explore here a workflow based on the application of a variational Gaussian process (VGP) model and universal co-kriging for implicit geological modeling from inducing points using two open-source Python packages (GeoML, GemPy).

Our results show that it is possible to create surrogate models efficiently for a range of geological settings – with a balance between the dimension (input points) of the surrogate model and the level of complexity of the original interface. In addition, due to a variational approach, uncertainties in the input data can also be represented in the surrogate model. In next steps, the generated surrogate models will then be integrated into geothermal exploration workflows, including the uncertainties in the legacy seismic data.