
A statistical technique that finds orthogonal spatial patterns of variability and their time series.Empirical orthogonal functions (EOFs) are a way to break down a complex, multidimensional dataset—like global temperature maps over time—into a few dominant patterns. Each EOF is a spatial pattern that varies coherently, and it comes with a time series showing how that pattern strengthens or weakens over time. The method is essentially principal component analysis applied to geophysical data, but the name 'EOF' is standard in atmospheric and oceanographic sciences.
The technique was popularized in meteorology by Edward Lorenz in the 1950s, though the underlying linear algebra was known earlier.
You take a data matrix where rows are time points and columns are spatial locations. The EOFs are the eigenvectors of the covariance matrix of this data, ordered by how much variance they explain. The first EOF captures the most variance, the second captures the most of the remaining variance while being orthogonal (uncorrelated) to the first, and so on. The corresponding time series (principal components) are obtained by projecting the original data onto each EOF.
Widely used in climate research to identify patterns like El Niño–Southern Oscillation from sea surface temperature data, and to reduce dimensionality before further analysis.
EOFs are not unique in sign—you can flip the pattern and its time series together and still get the same result, which sometimes leads to confusion when comparing studies.






