Enhancing Deep Geothermal Reservoir Characterization in Paris via Structural and CNN-Driven Quantitative Interpretation Methods

The GEOSCAN campaign, led by ADEME, Île-de-France Region, and the BRGM, aims to develop geothermal in the underexplored western and southern of Paris. This program involves reprocessing of over 1600 km of vintage seismic data, acquisition of 280 km of new high-resolution 2D seismic and petrophysical measures to improve undergoing knowledge. The dataset were analyzed and interpreted using advanced technique involving convolutional neural networks (CNNs) to perform quantitative interpretation of seismic date and predicted key reservoir properties like porosity and clay volume, revealing detailed reservoir architecture and facies variations, especially in the Oxfordian formation. A 3D velocity model was developed for accurate time-to-depth conversion, using checkshot surveys, sonic logs, and 2D migration velocity profiles. This enabled precise structural modeling, identifying faults and anticlines, which significantly influence reservoir geometry and thickness, particularly in the Dogger and Triassic formations. All geothermal reservoirs were mapped in detail over the 1900 km² area of interest, revealing structural transitions from west to east. These insights, unprecedented in resolution, support the creation of a 3D reservoir property model to assess geothermal potential and guide future development.