<p>The estimation of flame acceleration induced by thermo-diffusive and hydrodynamic instabilities is a key step toward developing accurate models for lean hydrogen combustion. One way to estimate this acceleration is to perform Direct Numerical Simulations (DNS) of laminar unstable flames. However, previous studies have shown that the 3D nature of instabilities significantly enhances the flame propagation speed compared to the 2D predictions. Systematically performing 3D simulations of laminar unstable flames remains out of reach due to prohibitive computational costs. In this work, a series of 2D and 3D DNS under various conditions are performed to establish a link between 2D and 3D flame wrinkling and stretch factors. Models are proposed that accurately estimate the 3D quantities from their 2D counterparts, leading in particular to an improved prediction of the 3D flame consumption speed. The robustness of these models is assessed using simulations performed with another chemical mechanism, and using published data obtained with another CFD solver. The prediction of the wrinkling and stretch factors is found to be highly robust. Overall, the resulting 3D flame consumption speed predictions are very accurate. These findings pave the way for improved subgrid-scale models of flame instabilities in RANS and LES frameworks.</p>