Self-optimization of Model Parameters with the LETKF: from Idealized Experiments to a Real-world Application

The state augmentation approach to optimizing model parameters is explored using the local ensemble transform Kalman filter (LETKF) with the T30L7 SPEEDY AGCM with simplified physics and with the Weather Research and Forecasting (WRF) model. The parameter variables are augmented to the state vector; the LETKF accounts for ensemble-based correlations between the parameter variables and observed variables, and estimates optimal parameters that give better fit to observations. Usually parameters are not directly observed, so that the ensemble-based correlations play an essential role. Experiments with the SPEEDY model revealed sensitivity to the inflation method for the parameter ensemble spread. We developed an approach to finding an appropriate covariance inflation factor for parameter variables and improved filter stability. Following a number of idealized experiments with the SPEEDY model, we tested the parameter-estimation method with the Weather Research and Forecasting (WRF) model in the real-world case of Typhoon Sinlaku (2008). The LETKF successfully found optimal parameter values of the sea-surface latent-heat and momentum exchange coefficients and improved the Tropical Cyclone analyses and forecasts.