Hourly temperature readings, to go with winds, humidity, visibility, ceilings, etc., are a standard meteorological element measured at first-order weather stations. Due perhaps to their sheer numbers (744 readings for a single 31-day calendar month, many thousands for a 30-year climatic “normals'” period), relatively mundane character (usually), and the greater convenience and practicality of using summary-of-the-day temperature statistics like daily maxima and minima, detailed statistical analyses of hourly temperature data are less common. Hour-to-hour temperature time-series, though, like their summary-of-the-day counterparts, can have interesting variations in pattern which should be worthy of study. An array of twenty-four mean hourly climatological temperatures for July, for instance, can serve as that month's idealized “normal” diurnal pattern, in the same fashion as mean daily maxima and minima do for summary-of-the-day recordings, but identification by some objective methodology of the most anomalous hourly patterns that have actually occurred over midnight-to-midnight in July should be interesting and useful also, analogous to extracting all-time record absolute maximum and minimum temperature information. A previous study along these conceptual lines [Fisk, 2004] investigated Downtown Los Angeles extreme patterns in summary-of-the-day maximum/minimum temperatures (depicted as floating-bars), comparing inter-year “configurations” between the same calendar month, and between complete years. The same motivation is extended here involving a sixty-five year data base of midnight-to-midnight hourlies. Utilizing Los Angeles International Airport hourly (1939 and 1947-2010) temperature data, downloaded and processed from the ISH online site, the utility of Principal Components Analysis (Correlation and Covariance, each unrotated) is demonstrated. First component correlation loadings characterize "shape", first covariance loadings, "spread". The highest and lowest correlation/covariance loadings identify the most anomalous patterns in terms of these attributes, and for a set of calendar months (January, April, July, and October), graphs of the most extreme patterns are presented as illustrative examples. A few noteworthy examples from other months are also described.

Given the long history to work with, sample sizes are quite large, approaching 2000 cases in some instances. The analysis is also extended into 2-D (“shape”/”spread” scatterplots), using Kernel Ellipsoid contours to identify the most extreme patterns in this higher dimensional sense.