Abstract
A network of 248 tree-ring chronologies in the conterminous United States is assembled and analyzed by rotated principal components analysis (RPCA) to delineate “regions” of common tree-growth variation during the period 1705–1979. Spatial continuity of the tree-ring data is summarized by variogram analysis, and tree-ring data are gridded before RPCA to reduce effects of site clustering. Principal component drought information is evaluated by comparing PC scores and primary pattern coefficients with Palmer Drought Severity Index (PDSI) data from instrumental records.
High PC pattern coefficients group geographically into regions coinciding roughly with nine drought regions delineated by RPCA of PDSI by other researchers. The drought signal as measured by the correlation between tree-ring PC scores and July PDSI, 1929–79, is strongest in the South and the interior West (r>0.7), and weakest in the Northeast and Pacific Northwest (r<0.16). A count of years with large negative PC scores in multiple regions marks the 1950s as the extreme in widespread drought across the southern United States to 1705.
Tree-growth regions are sensitive to whether tree-ring data are gridded before RPCA. Principal components on ungridded tree-ring data tend to center on dense clusters of sites. The importance of site density is most noticeable in the RPCA results for the southeast, where the gridded data yield a PC centered on a group of climate-sensitive but widely spaced bald cypress chronologies. Cross-validation indicates that gridding of tree-ring anomalies over different species for drought reconstruction is more appropriate in the semiarid southwest than in cooler, moister regions—especially the northeast and the Pacific Northwest. Our results endorse the large-scale chronology network as a long-term proxy for the spatial and temporal patterns of past drought across the United States.