Relationships between Systematic Errors in Medium Range Numerical Forecasts and Some of the Principal Modes of Low-Frequency Variability of the Northern Hemisphere 700 mb Circulation

Abstract

The relationship between the existence of low-frequency 700 mb height anomalies in the initial conditions of NMC's MRF global spectral model and subsequent 5-, 7-, and 10-day forecasts of 700 mb height from 1982 to 1988 is explored. Low-frequency 700 mb flow regimes are specified in each of four two-month seasons by performing a rotated principal component analysis (RPCA) on 38 or 39 year time series of daily, low-pass filtered 700 mb height analyses. In a given season, the amplitude time series (ATS) for each mode is used to decide which MRF forecast error maps should be used in forming a composite map corresponding to either the ?+? phase or the ??? phase of the given mode. Several methods, including Monte Carlo simulations, are used to evaluate the statistical significance of the composite maps. Many modes, including the Pacific North American (PNA) mode in winter and the leading summer mode, are found to be related to either unusually strong or unusually weak systematic error signature. Two different modes, one in spring and one in autumn, corresponding to quasi-stationary patterns over the United States and the North Atlantic, respectively, are related to unusually strong forecast error signatures. A statistically significant number of such modes is found in each of the four seasons, with the number of such results being smallest in autumn, and 1argest in spring. The results also indicate that the MRF model response to the presence of low-frequency regimes in the initial conditions is such that composite error signatures have a component with opposite phase and amplitude for opposite phases of a given mode (linear response). The overall results demonstrate the feasibility of using this technique to identify mode-linked forecast error signatures, and provides a potential opportunity to correct forecasts in the MRF, and possibly in other models, by removing the appropriate systematic error signatures.