Machine learning aided 3D modeling of Buntsandstein Formations for potential analysis of hydrothermal systems

Hydrothermal systems offer significant potential for green energy and heat production. The Buntsandstein Formations provide aquifers that are applicable for hydrothermal heat (or power) production. The Geological Survey of Brandenburg (LBGR) will improve its data base on Middle Buntsandstein aquifers using machine learning algorithms on archived exploration data.

Data stored in the archive of the LBGR contains drilling information for some 200 boreholes penetrating the Buntsandstein Subgroup. The LBGR has access to some thousand 2D seismic profiles from exploration surveys, typically reaching the Zechstein Group or the deeper-lying sandstone aquifers in the Upper Rotliegend Subgroups.

Drilling logs such as gamma ray or density logs provide valuable information for stratification and petrophysical rock property analysis. Available drilling logs will be digitized, and machine learning algorithms will be trained on exemplary data sets such as drilling E Hr 1/68. Machine learning algorithms will help to stratify the Buntsandstein sections of the available drillings. The stratification based on machine learning is further validated on newly interpreted seismic and drilling results, such as the cores stored at the drilling core and sample archive of the LBGR. The base will be in consideration of depositional gaps and tectonics. Also, a well-to-seismic-tie is performed at selected locations, and synthetic seismic profiles are generated.

If possible, seismic lines are (semi-)automatically evaluated to identify Hardegsen, Detfurth, and Volpriehausen Formations. Interpretation results are integrated in a 3D-geological model to build a regional-scaled Buntsandstein model of Brandenburg, allowing parameterization and calculation of reservoir temperature or heat in place.