Abstract
We examine the persistence of outgoing longwave radiation (OLR) anomalies in the tropics on many different time scales during 1974–86. We calculate “one-lag autocorrelations” by constructing nonoverlapping 1-, 15-, and 60-day averages and calculating the correlation at every grid point between every time average and the following average for the entire dataset. One-day averages produce the 1argest local autocorrelations everywhere except over the equatorial Pacific. Large autocorrelations using 15-day averages are confined to the equatorial Pacific, but large autocorrelations based on 60-day averages extend eastward from the eastern Indian Ocean through South America We attribute the increase in autocorrelation in some areas as the averaging period increases to the presence of the 30–60 day oscillation in those areas The spatial match between the autocorrelation and the standard deviation of OLR is best for 60-day averages and worst for 15-day averages.
We then calculate pattern correlations over a domain that extends along the equator from the eastern Indian Ocean through the central Pacific. When plotted as a time series the one-lag pattern correlations for two-month means are seen to vary wildly, although they are generally positive. There are some extended periods, however, during which the pattern correlation remains 1arge, most notably during the 1982–83 ENSO event.
The average one-lag pattern correlation is plotted for many different time averages. They decrease until a minimum at 20-day averages, beyond which they slowly increase as the averaging 1ength is increased.
The average one-lag pattern correlations using one-day averages are smallest during the mid-year months, but using 60-day averages they are largest during these months. The seasonality, however, is not large.
Finally, we identify eastward propagation of OLR anomalies with at least two distinct phase-speeds in addition to a quasi-persistent signal. It is suggested that forecasts of OLR anomalies might be improved over simple, local persistence by a multiple regression technique.
