15th Conference on Probability and Statistics in the Atmospheric Sciences

2.1

The use of principal component analysis for climatological inference

Michael B. Richman, Univ. of Oklahoma, Norman, OK; and R. H. Compagnucci

In recent years, the extensive use of eigentechniques has crossed beyond two important boundaries: pure data reduction and physical interpretation of each individual function. More recently, we have seen a third level of climatological inference, that of ascribing actual climatological or meteorological properties (e.g., wind direction and speed) directly to the output of the EOFs or PC loading gradients. While such inferences seem straighforward, the valdity of the interpretation remains unknown. In order to test the efficacy of these types of interpretations, we have created simple known climatological flow patterns, whose properties are fully known, a priori. These patterns can have direct inferences made and compared with those emerging from the eigenpatterns which arise from "typical" analysis procedures. By varying the types of patterns and their gradients, a set of results are obtained to encompass many types of analyses seen in the literature.

Results provide guidence for those cases which can be safely interpreted and those which con not. Moreover, various methodological options are investigated for their impact on the ability to draw valid inferences.

Session 2, Spatial and space-time statistics
Tuesday, 9 May 2000, 3:30 PM-4:49 PM

Next paper

Browse or search entire meeting

AMS Home Page