High-resolution ensemble data assimilation applied to Hurricane Katrina (2005)

Although several groups are performing high-resolution simulations of tropical cyclones (TC), the initial conditions for these models are often obtained from coarse model data that contains a symmetric representation of the storm. The lack of storm-scale details in the initial conditions may prevent the model from producing an accurate forecast of the TC's intensity. One method of improving a model's initial conditions is to perform data assimilation, whereby observations are combined with a short-term forecast to produce the best-estimate of the tropical cyclone. Most operational data assimilation schemes spread observation information onto model grid points using an averaged error covariance structure, which is not always appropriate for TCs, especially at fine spatial scales. As a consequence, these groups employ either a TC bogusing or vortex repositioning scheme near the TC; however, these initial conditions do not necessarily represent the best estimate of the TC or surrounding environment. The ensemble Kalman filter (EnKF) is an attractive alternative for TC data assimilation because this technique assimilates observations using flow-dependent covariances, and it can output an ensemble of equally-likely analyses that are available for use in TC ensemble forecasting.

In this study we create an analysis ensemble on a coarse fixed domain and a vortex-following nest using a Weather Research and Forecasting (WRF) model during the intensification phase of Hurricane Katrina (2005). This 96 member ensemble is updated each three hours using conventional observations including RAINEX dropsondes (no satellite radiances). Comparison of the EnKF analyses to NHC best track data and satellite observations indicate that the high-resolution nest is able to capture the gross features of the storm (minimum SLP and maximum wind) and certain storm-scale structures. On the fixed coarse grid, assimilating observations near the storm often leads to shifting the position of the TC. In contrast, observation assimilation in the moving nest can adjust storm-scale structures since all ensemble members are on a storm-centered grid. We will also discuss how these high-resolution ensemble forecasts can be used to estimate initial condition sensitivity and where additional observations could improve TC intensity forecasts.