636 Identification and Analysis of Contiguous Hours' Climatological Wind Pattern Modes Utilizing K-Means Clustering Analysis combined with the V-Fold Cross Validation Algorithm

Wednesday, 13 January 2016
Charles J. Fisk, Naval Base Ventura County, Point Mugu, California
Manuscript (638.4 kB)

Climatological wind variability is an important meteorological element to be considered in planning, forecasting, and decision-making activities in which wind conditions are crucial on some level. Wind rose diagrams, for example, can provide insights into the wind character for individual hours of interest by depicting the most favored compass directions and associated speeds. Resultant wind calculations can be valuable in producing distilled single-value statistics derived from many different individual observations.

Potentially of enhanced usefulness but more challenging to produce should be depictions concerning the most prominent contiguous hour-to-hour wind patterns that occur climatologically (encompassing, for example, midnight-to-midnight) over the course of the year. In the same manner as there are favored individual hourly directions and related speeds, there are undoubtedly preferred, multiple, adjacent hour-to-hour patterns, or “modes”. This could be extended further into the spatial realm, relating simultaneous hours' wind patterns for more than one station.

Resolving adjacent hours' wind patterns statistically can be a clustering exercise, utilizing u and v components of individual wind observations, and in this study, K-Means Clustering Analysis integrated with a special add-on capability, the V-Fold Cross-Validation Algorithm, will be applied. The V-Fold Algorithm is an iterative, automated, training sample type procedure that optimizes the number of clusters created, depending on the advance choice of statistical distance metric (Euclidean, Squared Euclidean, etc.), percent improvement cutoff threshold (e.g., 5 percent), and other settings. The user also has the option of fixing the number clusters created. This wind clustering methodology has already been applied with good results on many stations' data in the Southern California area, both for operational and purely informative purposes, including those of several NOAA Buoys in the Southern California Bight area, local military bases and NWS Offices, plus automatic weather stations. Among the individual cluster centroid outputs are, by hour, mean vector wind directions, mean vector wind speeds, mean scalar wind speeds, and mean vector wind constancy statistics. Frequently, the hourly constancy magnitudes (ratio of mean vector wind speed to mean scalar wind speed times 100) are in the 80's and 90's, indicative that the methodology effectively “distills” wind observations of a very similar character into separate clusters, the mean vector wind direction centroids in these cases, for practical purposes, being mean scalar wind directions.

As a demonstration, a set of six midnight-to-midnight hourly wind pattern clusters are created (using the squared Euclidean distance metric) and analyzed for an individual station (Point Mugu, CA), including comparisons of cluster percent frequency statistics, month-by-month, as well as for sub-month to sub-month intervals. Following this is a similar two-station analysis incorporating the California High Desert stations of Daggett CA, and Victorville, CA.

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner