Seminar Jorge Alvarez

Natural hydrogen (H2) is rapidly emerging as a disruptive primary clean energy source. However, global exploration is currently hindered by severe spatial sampling bias, as historical discoveries overwhelmingly cluster near existing fossil fuel and mining infrastructures. This study develops the first high-resolution global prospectivity model for natural H2 by integrating 45 physicochemical and macrotectonic datasets with ensemble machine learning algorithms (Random Forest and XGBoost). To decouple true geological signals from anthropogenic noise, rigorous spatial buffering and pseudoabsence generation protocols were implemented. Both models demonstrated exceptional discriminative capabilities (ROC-AUC > 0.90), autonomously isolating distinct H2 generation pathways: Random Forest prioritised deep structural geometries and thermal regimes, whereas XGBoost highlighted radiolytic geochemical drivers. Furthermore, a novel spatial upscaling approach estimates total continental H2 emissions at approximately from 4.11 to 81.82 Megatonnes per year. This figure aligns seamlessly with the global theoretical baseline whilst revealing that the continental lithosphere’s generative capacity has been massively underestimated, shifting the exploration paradigm away from inaccessible oceanic ridges. By providing a robust, data-driven framework capable of identifying unexplored fairways, this research confirms that natural H2 possesses a quantifiable, commercially viable footprint to accelerate the global net-zero transition.