Predicting hydrogen solubility in water with gradient boosting and LightGBM

Hydrogen (H2) is expected to become a key component of the global energy mix, both through engineered underground storage and the exploitation of naturally occurring subsurface deposits. A critical parameter governing the feasibility of these strategies is the solubility of H2 in water, which directly influences its reactivity, migration, and storage capacity in porous media. Experimental determination of H2 solubility under reservoir conditions is, however, costly, hazardous, and limited in scope. Here, we present two advanced machine learning models—gradient boosting (GB) and light gradient boosting machine (LightGBM)—trained on a dataset of 545 experimental measurements of H2 solubility in pure water across wide ranges of pressure and temperature. The GB model achieves superior performance (R2 = 0.9974; RMSE = 0.000165 mol mol−1), outperforming LightGBM. Trend analysis confirms that the models capture the underlying thermodynamic behavior of H2 solubility with respect to temperature and pressure, while leverage analysis using William’s plot indicates that fewer than 2% of the data points fall outside the model’s applicability domain. These results demonstrate that boosting-based algorithms can deliver highly accurate, rapid, and cost-effective predictions of H2 solubility, offering a robust alternative to conventional laboratory measurements. The validated GB model provides a practical tool for assessing hydrogen reactivity in geological systems and has immediate implications for the design and optimization of underground hydrogen storage and natural hydrogen production.