Detecting Atmospheric Regimes Using Cross-Validated Clustering

Low-frequency variability in geopotential height records of the Northern Hemisphere is a topic of significance in atmospheric science, having profound implications for climate modeling and prediction. A debate has existed in the atmospheric science literature as to whether or not “regimes” or clusters exist in geopotential heights, and if so, how many such clusters. This paper tells the “detective story” of how cross-validated mixture model clustering, a methodology originahy described at the 1996 KDD conference (Smyth, 1996), h as recently provided clear and objective evidence that three clusters exist in the Northern Hemisphere, where each of the detected clusters has a direct physical interpretation. Cross-validated mix^-^ture modeling has thus answered an iiiip0i;tAK.i upcu scientific question.