A method is presented for elevation (H) and spatial position (X, Y) decorrelation stretch of annual precipitation summaries on a 1-km grid for SW USA for the period 2003 to 2014. Multiple linear regression analysis of the first and second principal component (PC) quantifies the variance in the multi-temporal precipitation imagery that is explained by X, Y, and elevation (h). The multi-temporal dataset is reconstructed from the PC1 and PC2 residual images and the later PCs by taking into account the variance that is not related to X, Y, and h. Clustering of the reconstructed precipitation dataset allowed the definition of positive (for example, in Sierra Nevada, Salt Lake City) and negative (for example, in San Joaquin Valley, Nevada, Colorado Plateau) precipitation anomalies. The temporal and spatial patterns defined from the spatially standardized multi-temporal precipitation imagery provide a tool of comparison for regions in different geographic environments according to the deviation from the precipitation amount that they are expected to receive as function of X, Y, and h. Such a standardization allows the definition of less or more sensitive to climatic change regions and gives an insight in the spatial impact of atmospheric circulation that causes the annual precipitation.
Keywords: Annual precipitation; Daymet dataset; Environmental monitoring; Selective variance reduction; Spatial modeling.