RESUME / ABSTRACT  


Predicting forecast uncertainty with singular vectors and ensemble-derived analysis error covariances

par/by


Mark Buehner
Data Assimilation and Satellite Meteorology Division






Several important NWP applications require the accurate prediction of forecast uncertainty or its pre-cursor including: ensemble prediction, observation targeting, data assimilation and data quality control. Under certain conditions, the optimal estimate of forecast uncertainty is provided by singular vectors when the initial-time norm is defined by the inverse analysis error covariances. However, until now only the use of analysis error covariance estimates without correlations or lacking flow-dependence have been evaluated in an operational setting. In this talk, norms based on analysis error covariances from the ensemble Kalman filter and a simpler approach applied to an existing forecast-analysis system are evaluated in a quasi-operational setting. These approaches attempt to more realistically sample the probability distribution of analysis error by simulating, via a Monte Carlo approach, the error generated at each stage of the forecast-analysis process. Results from using these stationary or flow-dependent norms are compared with more common norms, including total energy. In the context of ensemble forecasting, the forecast error distributions derived obtained from both the singular vectors and the original EnKF analysis ensembles are compared.