Electrofacies classification using supervised machine learning for Permian layered evaporitic sequences of the Delaware Basin, USA: implications for hydrogen storage potential

Evaporitic sequences offer significant potential for large-scale underground hydrogen storage (UHS) in salt caverns, but their geological heterogeneity poses challenges for cavern development. This study applies supervised machine learning for electrofacies classification in the Castile and Salado formations of the Delaware Basin, USA, to assess their suitability for UHS. The support vector machine, random forest and extreme gradient boosting (XGBoost) algorithms were assessed using conventional well logs calibrated with core data, with XGBoost achieving a superior performance for both formations. The models effectively captured distinctive mineralogical characteristics: the Castile Formation characterized by thicker, halite- and anhydrite-dominated beds, while the Salado Formation exhibited greater heterogeneity with interbedded non-evaporite rock layers. Core-log integration revealed well-log resolution limitations in capturing thin beds and subtle transitions. Integrated analysis of critical factors such as halite thickness and depth demonstrated that the Castile Formation's thicker, more uniform halite beds provide more favourable conditions for cavern development compared to the more complex Salado Formation, where heterogeneity and differential dissolution risks are greater. This study underscores the importance of electrofacies prediction in assessing layered evaporite sequences suitability for UHS. The developed workflow offers a scalable screening tool to support informed decision-making for site selection and risk assessment in diverse evaporitic settings worldwide.