This study investigates natural hydrogen (H2) occurrences in the Paris Basin, using Optical Character Recognition (OCR) technology to analyze an extensive, yet historically underexploited, well database that contains older drilling records. With the growing demand for carbon-free energy, natural hydrogen, produced through processes like serpentinization and water radiolysis, offers a promising alternative to fossil fuels. However, its potential has been largely unexplored in conventional oil and gas wells. Utilizing the BEPH (Office of Exploration and Production of Hydrocarbons) French database, which includes well logs, mudlogs, and End Drilling Reports (EDRs) in PDF image format, we applied the Tesseract-OCR Engine to convert these documents into searchable formats for efficient data analysis. Our analysis revealed several H2-bearing wells across the French sedimentary basins. The hydrogen occurrences in the Aquitaine Basin correlate with the geological context, but those in the Paris Basin present an anomaly, as their H2 occurrences do not align with the expected geological factors. In the Paris Basin, H2 has been detected in four main formations: the Lusitanian aquifer, Dogger aquifer, Triassic aquifer, and the basement. The highest hydrogen concentration (52 vol%) was found in the Dogger formation. These wells are primarily located along the Bray fault and thrust, indicating a geological influence on H2 distribution. This research demonstrates the effectiveness of OCR in reprocessing historical drilling data for natural hydrogen exploration, highlighting the need for comprehensive exploration methodologies in this emerging field.

Ambient noise surface wave tomography is an environmentally friendly and cost-effective seismic technique for subsurface imaging. However, noise sources acting from preferential azimuths may introduce bias in the Green’s function reconstruction and in the resultant velocity models. This study, focused at the deposit scale, investigates how to correctly merge the different phase velocity measurements at various frequencies, in order to fill the gap between natural and anthropogenic noise sources while adjusting the bias caused by changes in the azimuth of the source. The target is the Marathon PGE-Cu deposit (Ontario, Canada), an alkaline intrusion containing gabbros and syenites (ø = 25 km). Mineralisation is hosted by gabbros close to the inward-dipping footwall of the intrusion. The country rocks are Archaean volcanic breccias. 1024 vertical-component receivers were deployed for 30 days in two overlapping grids: a 200 m x 6040 m dense array with node spacing of 50 m, and a 4000 m x 2500 m sparse array with node spacing of 150 m. Beamforming analysis of the recorded data indicates variations in the distribution of noise. Below 5 Hz, the Lake Superior (SSW) is the dominant source of noise, while above 12 Hz, noise from the Canadian Pacific Railway and Trans-Canada highway (SW) is prominent. In the 5 - 12 Hz frequency band, surface-wave energy is dominant, and it comes from the Lake Superior and vehicle traffic. Between 12 Hz and 20 Hz, the signal is characterized by body-wave energy combined with less energetic surface waves, while above 20 Hz the imprint of body waves is dominant. We retrieved the fundamental mode of Rayleigh wave propagation from the recorded data set. The signal was down-sampled to 50 Hz, divided into segments of 30 minutes, cross-correlated and stacked. Surface wave dispersion curves were extracted from 2-km-long arrays. Besides, various phase velocity measurements were applied. Phase-velocities were inverted to S-wave velocity structures using different probabilistic approaches. The overall results show a high-velocity shallow anomaly, probably related to the gabbro intrusion hosting the mineralization, as well as other structures consistent with the geological model inferred from surface mapping and drill logs.