Regional-scale Wind Field Classification Employing Cluster Analysis

A cluster analysis classification scheme can potentially lend itself to the evaluation of wind fields at a specific location to assist event planning as well as consequence and risk analysis, and to help address overall atmospheric transport uncertainty issues. Cluster analysis is a statistical tool that divides multivariate data in such a way that objects of the same cluster resemble each other and objects of different clusters do not, or are “dissimilar”. For the clustering presented here, an automated k-means approach with a Euclidean distance measure of dissimilarity is used. K-means clustering lends itself towards the rapid analysis of a large amount of data and allows for the clustering of that data according to user-specified criteria. For example, spatially varying wind observations at five different altitudes can be grouped together for hourly clustering, or spatially varying observations over 24-hour intervals can be grouped for daily clustering. A clustering classification scheme is introduced for a collection of regional-scale (~100km by ~100km) prognostic wind fields where an entire year’s data can be classified with distinct probability distributions for both wind speed and direction. The wind data was generated using historical COAMPS model simulations of a specific location for one continuous year (December 2002 to November 2003) on hourly intervals at altitudes up to 200mb (12km). The k-means clustering scheme is applied first to two distinct months of hourly data in 2003: January and July. For hourly near-surface observations, the clustering scheme successfully separates January from July wind field clusters. For example, when two clusters of the data are generated using the k-means method, the hourly data is grouped as a distinct cluster of high northwesterly winds (most of January observations, no July observations) and as a distinct cluster of low southeasterly winds (all of July observations, some January observations).

This work was performed under the auspices of the U.S. Department of Energy by the University of California, Lawrence Livermore National Laboratory under Contract No. W-7405-Eng-48. UCRL-ABS-203280