The response of damaging winds of a simulated tropical cyclone to finite-amplitude perturbations of different variables

Four-dimensional variational data assimilation (4DVAR) is an established data assimilation method that finds the finite-amplitude perturbation that best fits observations consistent with a priori information and model dynamics. The response of a simulated tropical cyclone to specially designed finite perturbations of selected model variables was studied with a modified version of 4DVAR. The usual goal of minimizing data misfits was replaced with a goal of reducing damaging surface winds at the end of six hours of forecast time. For this purpose a property value cost function based on topography was defined. The case studied was a 20-km simulation of a hurricane approaching the Hawaiian Islands. Each prognostic variable in turn—temperature, winds, humidity, vertical velocity, and perturbation pressure—and all prognostic variables at once were used as the control vector for the optimization problem. Of all prognostic variables examined, temperature and the horizontal wind were the most effective at reducing damaging surface winds. The wind-only perturbation was very similar to the wind component of the perturbation calculated when all prognostic variables were used at once. Calculated perturbations had scales of 0.25°C or 1 m s−1, but changes at a few grid points near the center of the storm were an order of magnitude greater. Vertical velocity and humidity perturbations alone were ineffective at reducing damaging winds. The perturbation pressure experiment failed to converge but did substantially reduce the damaging winds.