Reduced order models (ROMs) of the nonlinear geometric response of structures with or without coupling to a similar representation of the temperature have been steadily developed over a series of years in the context of hypersonic vehicles. They are now capable of handling structures of notable complexity subjected to various types of loads, including thermal ones. One particular challenge of these models is the a priori selection of the basis within which the response (structural and/or thermal) is represented. The nonlinearity of the structural response, its added dependence on the applied thermal loading, and the complexity of multiphysics interactions between aerodynamics, structural, and thermal problems render the a priori selection of a reliable basis a difficult task. While processes to construct this basis have been formulated and validated, they are likely to lead to large bases, and thus computationally expensive ROMs, for sizable/complex structural components. In this light, the focus of the proposed investigation is on the development of a data-driven reduction process of structural-thermal ROMs into “reduced ROMs” (RROM) of much smaller but potentially evolving bases that lead to predictions of the structural response and temperature with an accuracy similar to that of the original, “full”, ROMs at a much reduced computational cost.