A methodology for the computationally efficient CFD simulation of hydrogen-air explosions (including transition to detonation) in large volumes is presented. The model is validated by means of the largest ever conducted indoor DDT experiments in the RUT facility. A combination of models is proposed with a particular focus on the influence of flame-instabilities, especially of thermal-diffusive nature, which are crucial for very lean mixtures. Excellent agreement is achieved in terms of flame acceleration. The quality of DDT predictions itself depends on the underlying mechanism. Whereas DDT by shock-focusing is successfully simulated on under-resolved meshes, DDT by local explosions in the vicinity of the turbulent flame brush remains a challenge. Adaptive mesh refinement therefore emerges as a key technique to resolve more of the essential phenomena at reasonable computational costs affordable by industry. Finally, a generic case demonstrates the influence of mixture inhomogeneity, which can promote flame acceleration and ultimately DDT. (C) 2014 Elsevier Ltd. All rights reserved.