Pre-emptive forecasts from an ensemble Kalman filter

Ensemble Kalman Filters estimate the error covariance matrix of a model first guess field using ensembles. An ensemble Kalman Filter can be used for data assimilation – using the ensemble based error statistics to interpolate the difference between observations and the first guess field through out the model domain, creating an ‘increment' to the first guess field. An ensemble Kalman Filter can be used for targeting – propagating error statistics forward in time. Combining these two uses of an ensemble Kalman filter – the creation of an increment and the propagation of error statistics, a ‘pre-emptive forecast' can be generated. In a pre-emptive forecast, the increment to the first guess field at the analysis time is, using ensembles, propagated to some future time.

In an OSSE, a barotropic vorticity model was run to produce a 300-day ‘nature-run'. The same model, run at lower resolution and with a different relaxation scheme, served to forecast the nature-run. The forecast model was re-started every 24 hours, assimilating observations of the nature-run using a hybrid Ensemble Kalman Filter / 3D-Var data assimilation scheme. The forecast model produced 24-hour and 48-hour forecasts for each of the 300 days. In addition, 24-hour forecasts were made by using the ensemble Kalman Filter (EnKF) to re-weight members of a 64-member forecast ensemble to produce a ‘pre-emptive forecast'. The pre-emptive forecasts were more accurate than the 48-hour forecast, though not as accurate as the conventional 24-hour forecast. Issues of covariance localization and model error will be addressed.