Statewide averages of temperature and precipitation, from January 1895 to April 1981, were interpolated to a grid and Palmer Drought Severity Indices (PDSI) calculated for each of the grid points. Principal component (PC) analysis was performed on the gridded values of PDSI. By rotating the PC, nine readily identifiable patterns of drought are delineated. Each PC relates to different regions of the country, and is characterized by a distinct annual oscillation of monthly average precipitation. A high degree of confidence is placed in the state gridded data as patterns compare favourably to those derived using smaller averaging areas.
Univariate and multivariate autoregressive spectral estimation procedures together with the traditional spectral estimation techniques were used to investigate the spectra and coherence of the PC scores. No significant cycles were found in the 86-year (1036 months) data set. Coherence was very low (<0.30) for all frequencies between the various patterns depicted by the PC.
Both the spectra of PC scores and drought duration statistics point to the major difference in droughts across the United States; droughts persist much longer in the interior portions of the country than in areas closer to the coasts. Sensitivity tests on the PDSI indicate that this is not directly attributable to the spatially varying available soil water capacity parameter used in the PDSI calculations. This is not to say that soil moisture evaporation does not indirectly influence drought duration, but merely that the differing lengths of droughts in the United States are real and not an artefact of the mechanics in the PDSI calculations.