Data Assimilation and Initialization of Hurricane Prediction Models

Abstract

This paper investigates the problem of initializing operational hurricane models with several types of real data. Imbalances in real data generate inertia-gravity waves with periods that vary strongly in different regions of the hurricane domain. The energy of these waves is removed by propagation out of the domain, by the horizontal diffusion process, and by the truncation errors associated with the Matsuno time-differencing scheme. Several initialization schemes are tested with a symmetric hurricane model. Random and bias errors superimposed on perfect data produce imbalances that lend to significant errors in short-range forecasts. A general dynamic initialization scheme that is suitable for diabatic, viscous models yields very promising results. The dynamic initialization technique is utilized in an effort to determine the types of data that will be most useful in initializing operational hurricane models. In general, observations are most useful near the center of the storm at low levels. Temperature and wind observations are about equally effective in reducing initial analysis errors. Specific humidity observations, on the other hand, seem less important. Finally, the sensitivity of the initialization method is tested with observations that include rather large bias errors.