Identification of intra-month daily mean temperature modes using principal components analysis
Charles J. Fisk, Naval Base, Point Mugu, CA
A widely used statistic for summarizing temperatures is the “daily mean temperature”, usually the arithmetic average of the extreme maximum/minimum readings over a 24-hour period (generally midnight-to-midnight for first-order weather stations). Long-term smoothed climatological averages of same-day means serve as “normals”, enabling calculation of “departures from normal”, and determining anomaly character.
Climatological daily mean temperatures are, of course, statistical idealizations, and considered as time-series for calendar month periods, their smoothed, linear-like form does not typlify “normal” patterns in day-to-day mean temperatures that actually occur (more likely quasi-sinusoidal). In this regard, a month-by-month pattern climatology of day-to-day mean temperatures might provide a better appreciation of the nature and relative frequencies of those that do in fact happen. These “modes” could be identified and characterized using Linear Principal Components Analysis.
Using a Correlation PCA with no rotations, the nature and hierarchy of daily mean temperature modes were explored, by month, for the 85-year Downtown Los Angeles daily temperature record (1921-2005). For eleven of the twelve months, results resolved 14 eigenvectors (or modes) with eigenvalues greater than equal to one (September had 13). Eigenvector one's standardized scores suggested propagation of long-waves and, occasionally, seasonal trend as the most important intra-month daily-mean temperature pattern. Highest percent-of-variance-explained statistic for Eigenvector one was 30.7% for June. Other modes, in descending rank order, seemed to describe progressively higher frequency quasi-sinusoidal patterns with phase shifts, suggestive of a gradation in relative importance from long-waves to short-waves.
For illustrative purposes, plots of selected months' eigenvector standard scores are provided along with samplings of actual months' temperature patterns that conformed particular well to certain modes.
Extended Abstract (336K)
Joint Poster Session 1, Observation and Datasets-Part I (Joint between the 16th Conference on Applied Climatology and the 14th Symposium on Meteorological Observations and Instrumentation)
Monday, 15 January 2007, 2:30 PM-4:00 PM, Exhibit Hall C
Previous paper Next paper
Browse or search entire meeting
AMS Home Page