FORECAST-OBSERVATION PATTERN RELATIONSHIPS IN NCEP MEDIUM-RANGE FORECASTS OF NON-WINTER NORTHERN-HEMISPHERE 500-MB HEIGHT FIELDS

Empirical orthogonal function (EOF) and singular value decomposition ( SVD) analyses are applied to the Dynamical Extended Range Forecasting 90 (DERF90) dataset to find the most predictable areas and patterns in the non-winter Northern Hemisphere 500-mb height fields. Global 500-m b height forecast and analysis fields covering 4 months of 1990 are us ed The EOF method is applied to the forecast and analysis fields separ ately A close correspondence between the forecast and analysis modes i s found for 1-day lead time. The correlation relations between these m odes rapidly decrease with increasing forecast lead time. At 10-day le ad the projections of the forecasts and corresponding analyses onto th e strongest Northern Hemisphere EOF mode of the analyses correlate at 0.65, which is which higher than the general forecast-analysis correla tion of about 0.20. This particularly predictable mode consists of a P acific/North America-like pattern. Joint forecast vs. analysis correla tion matrices were used to apply the SVD method to forecast and analys is fields. By design, the correlations between the forecast and analys is SVD modes are stronger than between the forecast and analysis EOF m odes. Maps of temporal correlations between gridpoint values of analys is fields and the first and second EOF as well as SVD expansion coeffi cients of the forecast fields (i.e., heterogeneous EOF and SVD correla tion maps) are used to find the regions of the best prediction skill. It is shown that the Pacific/North American, western and southeastern Pacific the southern Atlantic and Indian ocean areas have relatively b etter prediction skill.