SVD-ROM: Scalable Reduced Order Modeling of Weather and Climate Data Using the Singular Value Decomposition

Despite their high dimensionality, datasets in fluid dynamics, weather, and climate often exhibit low-rank structure, where a small number of dominant patterns explain most of the variability. Singular Value Decomposition (SVD)-based methods provide efficient and interpretable tools for dimensionality reduction in such systems. Dynamic Mode Decomposition (DMD) extends this framework to time-resolved data by extracting coherent spatio-temporal structures and their associated dynamics. This enables the construction of low-dimensional, interpretable emulators of complex dynamical systems. However, its adoption is limited by the practical challenges of working with modern geophysical data volumes. In this work, we present SVD-ROM, an open-source Python package for SVD-based reduced order modeling, including DMD, designed to operate efficiently on datasets such as ERA5 using parallel and out-of-core computation on standard hardware.