Identification of California, Oregon, and Washington Coastal Area Climate Divisions' Relative Anomaly Modes in Total Seasonal Precipitation with Characterization of Their Occurrence Probabilities Relative to El Nino, Neutral, and La Nina Episodes

In a recent study (Fisk, 2015), the existence and character of idealized relative statistical anomaly patterns or “modes” in July-June total precipitation were investigated for the seven NCDC California Climate Divisions, utilizing the 1895-96 thru 2013-14 period of record, K-means clustering analysis, and calculations of Bayesian probabilities of the patterns' likelihoods, given the particular ENSO phase (“La Nina”, “Neutral”, and “El Nino”) in place. Results resolved six modes, some of which for the two southernmost divisions displayed noticeable contrasts in relative anomaly character compared to those of the other five. The ENSO phases also showed different affinities for different clusters. Extending the scope to include the Pacific states collectively, Oregon and Washington as well as California, and refining the ENSO phase designations by creating the “Strong La Nina” and “Strong El Nino” subcategories, the following study repeats the objective of the previous study, incorporating in the process an additional season's data (2014-15). The expectation is that the contrasts in relative anomaly character and ENSO affinities would be quite pronounced across the three States' coastal divisions.

Twenty-six total climate divisions comprise the three (seven in California, nine in Oregon and ten in Washington). To parse the number of divisions down, keeping a near-coastal focus, and in the process simplying analysis interpretations, the selection is reduced to include those closest to the Pacific (and in the case of Oregon and Washington, essentially those west of the Cascade Mountains). This leaves 12 divisions, three in California, four in Oregon, and five in Washington. The K-Means Clustering methodology is integrated with the V-Fold Cross Validation Algorithm, an iterative training sample type procedure that optimizes the number of clusters created, depending on the choice of statistical distance metric (Euclidean, Squared Euclidean, etc.), percent improvement cutoff threshold (e.g., 5 percent), and other settings. In this study the K-means approach utilizes the Squared Euclidean metric combined with the 5 percent distance improvement cutoff threshold; also the precipitation data are normalized in advance, by division. Seven clusters are resolved.

Then, through referencing and processing of bi-monthly ranked statistics from the MEI ENSO Index Data base back to 1895-96, a Bayesian statistical analysis is performed which addresses the following questions: given the presence of a “Strong” El Nino, “Ordinary” El Nino, Neutral, “Ordinary” La Nina, or “Strong La Nina episode, what are conditional probabilities that each of the seven idealized anomaly modes would be realized for a given July-June rain season. Results are described and interpreted, with the Bayesian probabilities compared among episode types.