A regional attentive empirical orthogonal function analysis: An application to North Atlantic Oscillation

Abstract

Empirical orthogonal function (EOF) analysis is widely adopted for identifying spatial patterns in regular latitude‐longitude gridded geographical data. However, a standard EOF analysis could underestimate variances from low latitudes due to the increasing grid cell areas when moving toward the equator. A broadly adopted compensating approach is the area‐weighted EOF, where the grid data are multiplied by a factor proportional to the grid area, thereby forcing each datum to represent identical areas. In this article, we revisit the area‐weighting scheme and discuss its potential drawbacks. In particular, we show that along with compensating the unequal areas, another unaddressed issue is which region of the data is more relevant to our focused problem. We propose a regional attentive EOF (RA‐EOF) method, which resolves area‐compensation and region selection simultaneously. We conduct case studies of the North Atlantic Oscillation (NAO) analysis and perform RA‐EOF to the sea level pressure variability in the North Atlantic sector. Experiments show that our method evidently detects the NAO pattern in the leading mode of sea level pressure variability and suggests the southwestward movement of the southern action centre of NAO during the summers. Our findings clarify the ambiguity about the summer NAO in previous studies. The proposed methodology provides a potent instrument for discerning the spatial distribution and analysing the temporal variability of atmospheric teleconnections and oscillations, particularly in cases characterized by weak signal strength.