On the combination of rotated principal component analysis regionalization technique and linear regression in seasonal rainfall prediction
Abstract This study considers the selection of predictors for regional rainfall based on dynamical considerations; for this reason, a regionalization technique that can preserve the underlying physics of rainfall was used in obtaining landmasses and local oceanic domains that are spatially coherent. For the study region (Africa, south of the equator), the adjacent oceans play a vital role in the seasonal rainfall variability at the landmasses; thus uncovering the complex nature of the multivariate relationship between rainfall coherent landmasses and local oceanic domains will enhance the construction of oceanic indices as predictors of seasonal rainfall at specific landmasses using linear regression analysis. Among different cluster analysis techniques, the rotated principal component analysis (PCA) is both fuzzy and allows the overlapping of the classified data set, which makes it a better choice for geophysical research that aims to regionalize continuous data such as rainfall. 10 regions with spatially homogeneous austral summer monthly rainfall totals were classified using the rotated PCA; some classified regions featured landmasses that are spatially coherent with the adjacent ocean, which qualifies them to be further considered on how rainfall anomaly and other physical parameters related to rainfall (e.g. convergence, relative vorticity, and sea level pressure) at the adjacent oceans explain the variations in austral summer rainfall anomaly at the homogeneous landmasses. The analysis of the physical mechanisms associated with the time development of the selected rainfall regions reveals that at the west-central equatorial rainfall region, variations in relative vorticity and convergence are associated with the development of the rainfall region; whereas at the central domains of southern Africa, variations in the patterns of sea level pressure, relative vorticity and convergence at the landmasses, the tropical and the southwest Indian Ocean can be associated with the development of the distinct rainfall sub-regions. The predictability of austral summer rainfall anomaly at the homogeneous landmasses using appropriate predictors at the adjacent local oceanic domains was relatively more accurate at the deep tropics, possibly due to the dominating mechanism of convergence in controlling the tropical rainfall.