Extended Abstract (196K)JP1.7
Automated, supervised synoptic map-pattern classification using recursive partitioning trees
Alex J. Cannon, MSC, Vancouver, BC, Canada; and P. H. Whitfield and E. R. Lord
An automated method for producing synoptic map-pattern classifications is introduced. The procedure uses recursive partitioning trees to separate gridded synoptic circulation data into discrete groups. Unlike traditional unsupervised classification algorithms, the recursive partitioning trees make use of surface weather-element data to guide the formation of classes. This results in classifications that are better able to represent links between the synoptic scale circulation and local weather conditions. Statistically unique classes that have little climatic relevance are avoided; infrequent but climatically important classes are more likely to be resolved using this technique. Application of the classification procedure requires few user decisions and guidance is provided to help the user determine the appropriate number of map-patterns.
The classification procedure is used to generate a synoptic climatology for southwestern British Columbia, Canada. Gridded sea-level pressure and 500-hPa geopotential height data covering the north Pacific Ocean and British Columbia are chosen as inputs to the recursive partitioning algorithm. Surface weather-element data from Vancouver are used to generate target variables for the recursive partitioning trees. The suitability of principal component analysis (PCA), nonlinear PCA, and unsupervised clustering procedures for representing the weather-element target is investigated. Classification performance is measured by applying the synoptic classifications to a number of environmental scenarios. Results are compared against traditional unsupervised clustering procedures.
Joint Poster Session 1, Ensemble Forecasting and Other Topics in Probability and Statistics (Joint with the 16th Conference on Probability and Statistics in the Atmospheric Sciences and the Symposium onObservations, Data Assimilation,and Probabilistic Prediction)
Wednesday, 16 January 2002, 1:30 PM-3:00 PM
Previous paper Next paper