Tropical cyclone dynamics deduced from ensemble state estimation

Among the important problems concerning tropical cyclones, it is widely recognized that issues related to genesis and rapid intensity changes are least well understood. Traditional investigation approaches, such as observational and numerical modeling studies, are frustrated by limited in situ observations of the storm and its environment. Moreover the relatively small, heterogeneous, sample of observed cases makes it difficult to draw general conclusions. Although numerical modeling studies may be used to conduct sensitivity studies, comparisons with available observations are typically ad hoc. Here we propose an alternative technique using ensemble-based state estimation (data assimilation) to generate very large samples of independent realizations for individual storms. These samples are consistent with available observations, and may be used to estimate statistically the processes responsible for rapid intensity changes and vortex genesis.

We use an ensemble Kalman filter to assimilate all available observations for hurricane Katrina (2005), including RAINEX dropsonde data, but excluding radar and satellite radiance data. An advantage of this technique is that flow-dependent covariance information is used to directly assimilate the observations in a tropical cyclone environment without the need for vortex bogusing. We then use the ensemble sample of independent analyses and forecasts to statistically determine the processes and locations responsible for rapid intensity changes; we also plan to explore the genesis problem for Katrina. Preliminary results for the spatial patterns that covary with cyclone central pressure during the rapid deepening phase of hurricane Katrina show familiar patterns, including a warmer core and a shorter radius of maximum wind. The results also show less obvious relationships, including an azimuthal wavenumber-one gyre favoring intensification along the northern side of the storm.