MOST PREDICTABLE STRUCTURES IN MEDIUM-RANGE HYDRODYNAMIC FORECASTS - PRINCIPAL COMPONENT AND SINGULAR DECOMPOSITION APPROACHES

The empirical orthogonal function (EOF) and singular value decompositi on (SVD) approaches apply to medium-range dynamic forecasts for the su mmertime Northern Hemisphere to reveal the most predictable regions an d atmospheric structures. Global fields of anomalies in analyses and f orecasts of 500-hPa heights for periods from 1 to 20 days were obtaine d from the data of dynamic extended-range forecasting experiment DERF- 90. The EOF technique applies separately to analysis and forecast fiel ds. For ten-day forecasts, projections of the forecasts and correspond ing analyses on the principal EOF mode of analysis correlate in the No rthern Hemisphere at an anomaly correlation (AC) of 0.65, which is muc h greater than the AC in forecasting complete fields (about 0.20). Thi s most predictable mode includes the Pacific/North-American oscillatio n. The matrices of cross correlations (analysis-forecast) were calcula ted to apply the SVD to analysis and forecast fields. As expected, cor relations between SVD modes of forecast and analysis are higher than t hose between EOF modes. Maps of time correlations between grid-point v alues of analysis and forecast fields and the first two EOF and SVD co efficients of forecast fields (EOF and SVD heterogeneous correlation m aps) were used to determine the regions of the highest verification sc ore. It is shown that among the regions of rather high verification sc ore are the Pacific/North-American, western and southeastern Pacific, southern Atlantic, and Indian Ocean regions. These results are support ed by estimates of complete forecast fields.