A data-driven method that extracts spatiotemporal modes from nonlinear dynamical systems without prior filtering.Nonlinear Laplacian spectral analysis (NLSA) is a computational technique for finding patterns in complex, time-varying data — such as climate records, ocean currents, or fluid flows — without assuming the system behaves linearly. Traditional methods like Fourier analysis or principal component analysis assume that the underlying dynamics are linear and stationary, but many natural systems are not. NLSA instead uses ideas from geometry and spectral theory to extract modes that capture the system's intrinsic, often nonlinear, behavior.
NLSA was introduced around 2010 by Dimitrios Giannakis and Andrew J. Majda, who applied it to climate data to extract modes of variability such as the Madden–Julian oscillation.
The method builds a graph from the data points, where each point represents a state of the system at a given time. Points that are close in state space are connected, and the graph's Laplacian — a matrix that encodes local connectivity — is computed. By analyzing the eigenvectors of this Laplacian, NLSA identifies smooth, slowly varying patterns that correspond to the system's dominant dynamical modes. Crucially, it does this without first filtering the data, so it can separate meaningful signals from noise and capture intermittent or anharmonic oscillations that linear methods miss.
NLSA is used in climate science to identify and track modes of variability in large datasets, such as sea-surface temperature or wind fields, and it has also been applied to problems in fluid dynamics and neuroscience.
NLSA can detect patterns that repeat irregularly in time, like the Madden–Julian oscillation, which standard Fourier analysis would smear into a broad, indistinct peak.














