Premium
Rapid, short‐term ensemble forecast adjustment through offline data assimilation
Author(s) -
Madaus L. E.,
Hakim G. J.
Publication year - 2015
Publication title -
quarterly journal of the royal meteorological society
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.744
H-Index - 143
eISSN - 1477-870X
pISSN - 0035-9009
DOI - 10.1002/qj.2549
Subject(s) - data assimilation , ensemble forecasting , forecast verification , ensemble kalman filter , covariance , forecast error , quantitative precipitation forecast , range (aeronautics) , kalman filter , computer science , forecast skill , numerical weather prediction , term (time) , global forecast system , meteorology , statistics , mathematics , econometrics , machine learning , extended kalman filter , geography , precipitation , materials science , physics , quantum mechanics , composite material
Rapid updates of short‐term numerical forecasts remain limited by the time it takes to assimilate observations and run a dynamical model to produce new forecasts. Here we use an ensemble‐based statistical method to adjust a user‐defined subspace of forecast grids rapidly as observations become available. Specifically, an ensemble Kalman filter is used to adjust forecast variables based on covariances between the ensemble estimate of the observation and the forecast variables. This approach allows rapid adjustment of forecast fields, or functions of those fields, ‘offline’, without the expense or time of running the full dynamical model. Furthermore, by updating an ensemble, forecast uncertainty is also adjusted. The technique is tested using operational ensemble forecasts from the European Centre for Medium‐Range Weather Forecasts and Canadian Meteorological Centre. Results show that the method is effective at reducing forecast errors in surface pressure at least 18–24 h after the observation time, with a maximum impact of 9–15% for 12 h forecasts. Results for surface temperature show an error reduction 6–12 h after the observation time. Incorporating time‐lagged ensembles provides even greater reduction in error and a novel covariance localization technique that operates in space and time based on statistical significance is evaluated.