| description abstract | The objective of this study was to classify 500 mb height anomaly patterns for North America using principal component analysis with oblique rotation. Two versions of the oblique rotation, oblimax and direct oblimin, were applied to two gridded networks of different point densities and areal extents. The larger network contained 188 points (N188) in an equal-area grid for the area bounded by 10°E to 170°W longitude, and latitudes north of 20°N within North America and neighboring oceans. The density of the 65-point network (N65) was half that of N188, with a smaller longitudinal domain restricted to the continent and small area of ocean. Daily data were obtained from the National Center for Atmospheric Research for the 1946?79 period. A rotated screen test suggested that 30 and 22 components, for the larger and smaller networks, respectively, be extracted for oblique rotation and subsequent plotting of map types. Since each component actually represents two map types 60 and 44 map types were analyzed for the two networks, respectively. The oblimax rotation was applied to N188, resulting in 78% of all days being classified. An 85% clarification rate was obtained using direct oblimin on N65. More significantly, inter-component corrections were much lower with direct oblimin, resulting in fewer redundant map patterns. Computed anomaly patterns from both networks were compared to observed patterns for selected 5-day periods, obtained from Hay and Knox. From a combined sample of 801 computed anomaly centers, approximately 65% were spatially matched with observed anomalies (61.5 and 67.0% for the larger and smaller networks respectively). In all the examples of map patterns, there were some observed anomaly centers that were not identified. However, the overall level of performance seemed to indicate that oblique rotations showed some potential to become useful tools in the classification of spatial anomaly patterns of 500 mb heights. | |
