Résumé : In a large number of geo-environmental applications, it is essential to model coupled processes that depend on several design parameters such as material properties and geometrical features. Thermo-hydro-mechanical (THM) processes are, among others, key effects to consider in critical applications such as deep geological repository of hazardous nuclear waste. This thesis proposes novel model order reduction strategies to evaluate the thermo-hydro-mechanical response of porous media, taking into account the complexities involved in coupled processes for such applications. An a-posteriori model order reduction approach using reduced basis methods is applied to tackle the high-dimensional parametric THM system. The reduction is based on an offline-online stage strategy. In the offline stage, reduced subspaces are constructed by a greedy adaptive procedure and in the online stage, multi-subspace projection is performed to quickly obtain the coupled THM response at any value of the design parameter. The effectiveness of the model reduction strategies is demonstrated through realistic parametrized problems in large-scale geomodels depicting the coupled processes in potential deep geological repository sites.