Assessment of Covariance Inflation Options for a Continuously-Cycled Mesoscale EnKF Analysis System

An important design consideration when configuring a continuously-cycled ensemble Kalman filter (EnKF) data assimilation system is made in the selection of a method to cope with sampling deficiencies in the analysis system. Multiplicative inflation is now the most widely used method to combat these deficiencies in system performance that arise from model error, sampling flaws in the observing network, and errors in covariance statistics from the prior state estimates. Yet, among commonly used multiplicative inflation methods, little guidance is available on how different options and settings impact the ensemble analysis and subsequent ensemble forecasts. In this study, several options are explored for a mesoscale, continuously-cycled, regional EnKF data assimilation system. Inflation options examined include spatially and temporally adaptive covariance inflation, where multiplicative inflation is applied to the prior, posterior, and both prior and posterior states (Anderson 2009). Also, a new technique referred to as “spread restoration” is described and demonstrated in combination with the above mentioned prior adaptive inflation method. Finally, a flavor of relaxation to prior spread is considered (Whitaker and Hamill 2012), which is an alternate form of posterior inflation. Among these algorithm options, several parameters are considered. How these options and parameter settings impact select aspects of the analysis such as prior and posterior spread, mean analysis bias, and utilization of available observations will be shown. To illuminate the impact of inflation choices on ensemble forecast skill and reliability, convection-permitting ensemble forecasts are initialized daily during a month-long period from ensemble analyses that employ these varied configuration options.