Autumn and winter daily precipitation areas in Wales, 1982–1983 to 1986–1987

Abstract

Daily precipitation data from 121 sites for five autumn-winter seasons (September to January, 1982January, -1983January, to 1986 have been analysed to determine discrete precipitation areas within Wales, using S-mode principal components (PCA) and cluster analyses. Using unrotated PCA, much of the variation in daily precipitation is explained by the first five PCs, representing an accumulated variance of 75.7 per cent. The first four PCs appear to relate respectively to: altitude (608 per cent of the total variance explained), an exposure effect relative to north or north-east winds (6.0 per cent), a 'westerly exposure' effect (4.2 per cent) and an 'easterly exposure' effect (2.9 per cent). The fifth PC, with high scores only over Ynys M6n and Gwent, could not be attributed to any particular influence. Oblique rotation produced a better simple structure, enhanced the apparent importance of exposure to wind from the main compass directions, but lessened that of altitude. Three methods of clustering PCs were subsequently tried on the Oblimin solution: centroid, median and Ward. The precipitation areas produced by each were broadly similar, and the boundaries between them often coincided with clear topographic units, influenced by exposure and shelter effects. The Ward technique yielded the areas best defined in terms of topography and exposure, and best satisfied criteria determined by the nature of the data themselves.

Comprehensive analysis of the climate variability in the eastern Mediterranean. Part I: map‐pattern classification

“…Also it should be recalled that each component represents two opposite phases, differing in the sign of the PC loadings, thus two symmetric thresholds should be used to identify the most characteristic days of both modes. A search in the literature offers some advice regarding thresholds for the PC loadings (Bonell and Summer, 1992;Richman and Gong, 1999) but 'optimal' PC score thresholds have rarely been studied. The selection criteria are rather dependent on the aims of a particular project.…”

Evaluation of the NCEP–NCAR reanalysis in terms of synoptic‐scale phenomena: a case study from the Midwestern USA

“…However, the PC loadings matrices (representing the relationships between each variable and each PC) and PC scores matrices (representing the relationships between each observation and each PC) are also valuable tools in the interpretation of the classification results (see Section 4.1). Correlation matrices are used as the dispersion matrices for the PCA to weight the input variables evenly, and the resulting PCs are rotated using (orthogonal) Varimax rotation (Richman, 1986;Bonell and Sumner, 1992;Brinkmann, 1999a). The number of PCs to retain for rotation and interpretation is evaluated using a number of methods, including scree plots (Catell, 1966), logarithmic eigenvalue plots (Craddock and Flood, 1969) and statistical tests, such as the N rule (Overland and Preisendorfer, 1982).…”

Summer temperature extremes in northeastern Spain: spatial regionalization and links to atmospheric circulation (1960–2006)

“…Romero et al, 1999, DeGaetano, 2001). This algorithm is an ANOVA-type approach which 270 explicitly minimizes the within-group similarity and maximizes the between-group 271 similarity (Bonell and Summer, 1992 (Everitt and Dunn, 1991 in which each observation is assigned to a unique partition (cluster) and cannot be 297 reassigned to alternative cluster whenever more appropriate (Gong and Richman, 1995).…”