←Back

192 – Contributed Poster Presentations: Section on Statistics and the Environment

Identification of Contiguous-Hour Climatological Wind Modes Utilizing K-Means Clustering Analysis Combined with the V-Fold Cross-Validation Algorithm

The following describes and demonstrates a methodology for identifying and characterizing climatological adjacent-hour resultant wind patterns or "modes", a departure from more conventional explorations of this kind which generally focus on single-hour analysis units (i.e., stand-alone individual hours or solo consolidations of individual hours). Resolving such patterns can be a statistical clustering exercise, utilizing decomposed east-west ("u") and north-south ("v") wind components, and for this purpose, K-means Clustering Analysis integrated with an optimization tool, the V-Fold Cross-Validation Algorithm, is applied on daily series of hourly wind observations for five first-order Canadian weather stations with lengthy periods of records: Victoria, Vancouver, Calgary, Toronto, Montreal, and Victoria/Vancouver (two-station treatment). The resulting analysis units (hourly u and v centroid statistics) are then converted, utilizing the arctangent function, into discrete cluster-by-cluster progressions of adjacent hour resultant wind statistics (24 by cluster or mode, for a single station), including resultant wind directions/speeds, mean scalar speeds, and constancies (ratio of resultant wind speed to mean scalar wind speed times 100). The statistics are then depicted graphically using vector icons (representing the resultant winds), colored circular icons (depicting constancy magnitudes), and on the same layout, histograms, depicting the relative frequencies of the idealized patterns, by calendar month.