Sensitivity of ensemble Kalman filter assimilation of hurricane inner core observations to the background error specification in nested domains

In the research and operational practice, multi-level nested domains are commonly used to achieve the high-resolution numerical assimilations of the hurricanes. Therefore, it is a very common way to initialize the high-resolution model domain by data from the coarse-resolution domains or global models. Because of this, the underlying high-resolution mesoscale evolution is generally lacking in the background term for the regional data assimilation at the high-resolution regional domain. As a consequence, spin-up and spin-down problems are commonly experienced in short-range forecasting at the high-resolution domain.

In this study, we test the sensitivity of ensemble Kalman filter assimilation of hurricane inner core observations to the background error specification in nested domains. Specifically, we examined the effects of coarser-resolution versus high-resolution background error covariance on data assimilation in high-resolution domains for hurricane vortex initialization. Numerical experiments are performed with an advanced research version of the weather research and forecasting (WRF) model and its ensemble data assimilation system developed by the NCAR Data Assimilation Research Testbed (DART, Anderson et al. 2009).

The preliminary numerical experiments were conducted to improve the initial conditions for Hurricane Katrina (2005) at 1800 UTC 25 August 2005. In a two-level nested domain configuration, in order to achieve initialization for a higher-resolution nested domain (inner domain, 9 km horizontal grid spacing), two sets of experiments were performed. In the first set, the background error covariance for inner domain was generated from outer domain (27 km horizontal resolution). In the second numerical experiment, the background error covariance for inner domain was provided by the statistics from a set of 2-h ensemble forecasts prior to initial time (for data assimilation). The airborne Doppler radar–derived 3-dimensional u and v components over the vortex inner-core region were assimilated into the WRF model with the DART EnKF system. It is found that the analysis increment in the experiment with the background error term generated by ensemble forecasting in the inner domain reproduces a more organized hurricane inner core, outer rainbands, and detailed small-scale features of the hurricane vortex. As a consequence, the subsequent forecast from this type of background error term results in a more intense and more realistic hurricane in terms of the minimum central sea level pressure, compared with the experiment that uses the background error term from the outer domain.

Based on the results from this study, an effective way to generate background terms for nested domains are suggested.