Model Order Reduction
The main idea behind reduced order modeling is to significantly reduce the complexity of the high-fidelity physical model of a dynamical system while retaining a large fraction of the information about the underlying processes (variance). Model order reduction can be achieved by projecting the data onto a low-order subspace of the data structure. Several methods exist for achieving this, the most common being proper orthogonal decomposition. The application of interest in the context of spaceflight is the modeling of the thermosphere (upper atmosphere), which can effect the orbit of a space object through atmospheric drag. In the video below, the true global mass density field (top left) is recreated using a low-dimensional sub-space (bottom right) using minimal computational resources compared to the original high-fidelity model.
For more information, read my paper "A methodology for reduced order modeling and calibration of the upper atmosphere".