Assessing predictability with a Local Ensemble Kalman Filter

Istvan Szunyogh
Department of Atmospheric and Oceanic Sciences, University of Maryland, College Park

In this presentation, the spatio-temporally changing nature of predictability in the National Centers for Environmental Prediction (NCEP) Global Forecast System (GFS) is dicussed. Atmospheric predictability is assessed for the perfect-model scenario, where forecast uncertainties are entirely due to uncertainties in the estimates of the initial states. The imperfect initial conditions are obtained by assimilating simulated noisy observations of the ``true" states with the Local Ensemble Kalman Filter (LEKF) data assimilation scheme.

For this specific choice of the model and data assimilation system, the forecast errors grow exponentially in the extra-tropics and linearly in the tropics. The analysis errors are the smallest in the regions, the extratropical storm tracks, where the growth of the forecast errors is the fastest. This seemingly paradoxical result can be explained by the strong anti-correlation between the local dimensionality and the error variance explained by the LEKF ensemble. This strong anti-correlation makes the LEKF algorithm extremely efficient in estimating the analysis and forecast uncertainties in the regions of local low dimensionality, which coincide with the regions fastest error growth. The efficient estimation of the space of uncertainties enables the LEKF to produce very accurate analyses and very accurate estimates of the forecast uncertainties. It is conjectured that the results presented here could be reproduced with any suitably formulated ensemble-based Kalman filter data assimilation scheme.

Last Update: September 30, 2005