Modes of Optimal Persistence in SST and Height Fields

We describe a technique for determining a set of patterns in a time-varying field that optimizes the time required for the patterns to decorrelate. The decorrelation time of a pattern is measured by certain integrals of its lagged correlation function. Since these patterns typically remain correlated longer than individual EOFs, they may provide a more useful basis set for statistical prediction models. These patterns may also be of interest in predictability studies because they define the maximum lead-time of skillful empirical forecast models. We present examples of these patterns for monthly mean 500 hPa geopotential height fields and monthly mean SST fields. The height pattern with the maximum decorrelation time (1.5 months) tends to amplify during strong ENSO events and resembles the "SST-forced" teleconnection over North America reported in previous studies. Many SST patterns were obtained which remain correlated throughout the time span of the available time series.