Underground hydrogen storage in a depleted underground gas storage field in Austria: Forecasting under uncertainty and pilot test design

Underground Hydrogen Storage (UHS) is a promising solution for balancing fluctuating renewable energy supply and demand. Reservoirs previously used for Underground Gas Storage (UGS) are particularly well-suited for UHS due to their established infrastructure and proven capacity for high-rate gas injection and withdrawal. However, limited field experience and high geological complexity introduce substantial uncertainty into predicting UHS performance. This paper presents a simulation case study of converting an Austrian UGS reservoir into UHS applying a comprehensive stochastic workflow based on Approximate Bayesian Computation (ABC) to quantify subsurface uncertainty. A fully integrated modelling framework is established, incorporating 74 geological, petrophysical, and dynamic parameters defined from historical field data and literature. Machine learning proxy modelling is applied to test over 5 million subsurface model realizations and reduce computational cost during the history matching process. A posterior ensemble consisting of 48 history matched models was used to forecast key UHS performance metrics: working and cushion gas volume as well as hydrogen purity in the back-produced gas stream. Furthermore, the ensemble informed the design of a pilot test involving injection of a 10–90% hydrogen–methane mixture, aimed at reducing uncertainty in the most influential parameters governing UHS performance. The analysis reveals that volumetric properties, permeability, aquifer parameters, and fault transmissibility exert the strongest influence on history matching of production and pressure data. Forecasting results highlight substantial uncertainty in all major UHS metrics despite extensive production history and prior UGS operations. Forecasts demonstrate variability in hydrocarbon concentrations in the back-produced gas stream, ranging from 15% to 25% after the first production cycle, which is a critical factor due to separation and handling requirements. Prediction of working and cushion hydrogen volumes range ± 25% from P50 case. Trapped gas saturation and reservoir rock quality distribution dominate working gas volume predictions, whereas mechanical dispersion, molecular diffusion, and advection properties primarily govern hydrogen-methane mixing. A pilot injection test is shown to be representative for predicting hydrogen purity during full-field operation but insufficient for reducing uncertainty in working-gas capacity. This study demonstrates, for the first time, the applicability of ABC-based uncertainty quantification to UHS in depleted gas reservoirs, offering a robust framework for decision making, value of information assessment, and optimization of future hydrogen storage development.