Dynamic Initialization of Hurricanes in the Ocean-Land-Atmosphere Model (OLAM)

Initial atmospheric conditions for a numerical hurricane forecast model should be as consistent as possible both with available observations and with the numerical framework of the model, including grid resolution, formulation of the dynamic core, and parameterization of physical processes. Meeting both criteria simultaneously remains a major challenge, but is recognized as essential for improving the accuracy of hurricane forecasts.

We present a dynamic initialization approach to this problem that assimilates in-situ aircraft and dropwindsonde observations using a Kalman Filter in a series of short-term forward integrations of the forecast model. At the end of each forward integration, which spans approximately 2-4 hours, the simulated vortex is analyzed in terms of its dynamic, thermodynamic, and moisture structure relative to its undisturbed tropical environment. The diagnosed structure is then relocated back to its initial position, superimposed on the undisturbed tropical environment, and re-combined with the same in-situ observations as before in the next forward integration cycle. The series of assimilation and integration cycles results in a vortex that is balanced in the model numerical framework, not as a simple dry vortex but with a fully developed field of liquid and ice hydrometeors. This enables the actual forecast to commence without undergoing a period of spin-up or adjustment that would otherwise result from physics processes suddenly being switched on, as in some other existing methods of initialization. This dynamic initialization approach somewhat resembles the Running-In-Place (RIP) technique proposed by Kalnay and Yang (2010), but its purpose here is very different. Instead of providing sufficient time for differences in ensemble members to develop adequately, the forward integrations in our case enable the simulated vortex to achieve approximate balance and equilibrium with respect to all modeled processes.

The dynamic initialization procedure is carried out with the Ocean-Land-Atmosphere Model (OLAM), which is a unique global model with an unstructured hexagonal mesh that enables seamless local mesh refinement to a high degree. This framework is ideally suited to hurricane forecasting because it can match the high resolution of nested regional models within its own self-consistent global domain, and it carries a detailed microphysics model that is applicable to both highly convective and stratiform environments.

Kalnay, E. and S-C Yang, 2010: Accelerating the spin up of Ensemble Kalman Filtering, Q. J. R. Meteorol. Soc., 136, 1644-1651.