Three long-term datasets are used in this study of precipitation variability over the continental United States. The first is a 100-yr record of NCDC observations in 344 U.S. climate divisions. The second is NCEP's 40-yr (1958-97) reanalysis dataset. The third consists of a 13-member ensemble of 45-yr NCEP atmospheric GCM runs made with prescribed observed global SSTs for 1950-95. Because the probability distributions of precipitation are strongly skewed, the information is summarized in terms of fitted gamma distributions. There is generally good agreement between the datasets, though less so in summer.
The differences in the interannual variability of precipitation at different locations are associated with differences of both the scale parameter beta and shape parameter alpha (a measure of skew) of the fitted gamma distributions. This is reflected in large differences in the probability of extreme precipitation. To first order, the differences of alpha can be related to differences of the large scale
mid-tropospheric ascent and descent. Differences of available moisture are less important.
Substantial asymmetries are found in the precipitation signals for El Nino and La Nina years. These can also be understood, and put on a stronger footing, by the existence of consistent signals in the patterns of large scale ascent and descent. Another important result is that even in areas of weak mean ENSO signal, the gamma distributions can be significantly different for El Nino and La Nina years, with particularly marked differences in the probability of low precipitation. If correct, this result suggests that the occurrence of ENSO may have hydrological implications even in regions in which the mean ENSO signal is relatively weak