Utilizing Principal Components Analysis (PCA), the existence, character, and prominence of month-to-month modes in July-June precipitation variability were investigated for the complete set of seven NCDC California climatic divisions. Period of record was the 1895-96 through 2000-2001 seasons. Preliminary processing of the data by creation of combined July through November and April through June series (to go with December, January, February, and March), addition of an .01 constant (“start”) to all members in a given series if at least year had a zero observation, application of the Box-Cox Power transformation procedure, and standardizations enabled the analysis to be carried out with no compromises to the multivariate-normal assumption.

Results resolved physically sensible, but modest (as to eigenvalue magnitude and differentiation) components for each division, their character dependent on general latitude. The first three (out of six) modes typically explained about 60% of the variance. For example, “primary” mode for divisions 6 and 7 (“South Coast Drainage” and “Southeast Desert Basins”, respectively) was a contrast in eigenvector coefficient signs between January, February, and March (each significantly, positively signed) with those of the other periods. For divisions 4 and 5 (“Central Coast Drainage” and “San Joaquin Drainage”) the primary mode contrasted highly positive coefficient signs for December and January with relatively undistinguished coefficients for the other periods. Interestingly, Division 6's and 7’s primary mode was 4's and 5’s secondary mode; and 4's and 5’s primary was 6's and 7’s secondary. For divisions 1, 2 and 3 (“North Coast Drainage”, Sacramento Drainage”, and “Northeast Interior Basin”, respectively), the primary mode contrasted highly positive coefficient signs for December, January and April-June with the others, the secondary mode contrasting highly positive coefficient signs for February and March with those of the other periods.

In addition, time-series plots of component scores were constructed division-by-division, and identification of the individual most anomalous yearly patterns were done for each division using multivariate statistical distances.