Advanced Numerical Modeling And Data Assimilation Techniques For Tropical Cyclone Predictions PDF Download

Over the last decades, researchers have focused on improving tropical cyclone (TC) forecasts. Accurate TC predictions are very important in order to protect life and property. Scientists examine two important pieces regarding TC prediction: where the storm is going and how strong it will be in the future. These are referred as track and intensity forecasts. TC track forecast has improved tremendously over the last several decades. However, hurricane intensity forecasts continue to be a great challenge in operational and research communities. Previous studies have found that the lack of progress in intensity forecasts is partly due to the lag in the ability to specify the initial vortex in the numerical weather prediction (NWP) model, in addition to the lag in representing the observed inner-core storm intensity, structure and internal dynamics. Researches have introduced various data assimilation (DA) techniques to address the problem of determining the initial vortex. However, in order to better represent these features, there must be sufficient observations in the inner-core region along with a data assimilation method that can effectively use the data to accurately estimate the initial vortex. Some of the challenges in the TC data assimilation are: (1) scarcity of systematic data assimilation in the inner-core region, and/or, (2) absence of enough information about this region, and/or (3) the model resolution is inadequate to capture the structures at these smaller scales. This study examines the impact of assimilating high-resolution inner-core Airborne Doppler Radar (ADR) winds on two major hurricanes, Ike (2008) and Earl (2010). The primary objective is to understand its impact in the initial vortex structure and how it translates to the resulting forecasts. With the development of advanced data assimilation techniques, ADR data can improve the specification of the vortex and potentially improve intensity and structure forecasts. Nevertheless, there are two important factors that can affect the effectiveness of the method: (1) resolution on the grid where DA is performed and (2) the background error covariance used. This work focuses on improving the 4-Dimensional Variational (4DVar) data assimilation technique by using a high-resolution DA domain of 4-km in order to better represent convective scales features and by generating a new static background error covariance more suitable for the current DA experiment. This static error covariance includes the vortex structure information. The impacts of these two aspects were revealed by comparing the analyses and forecasts generated by 4DVar with relatively coarse resolution of 12-km that used the standard background error covariance file (that do not contain any vortex information), a 4DVar at 4-km that used the same background error covariance, and with a 4DVar at 4-km that used the newly generated covariance. This method is first applied on Hurricane Ike. The second experiment performed on Hurricane Earl only included one 4DVar setup: 4-km DA domain with the new static covariance that contains the vortex information. The results for Hurricane Ike experiment showed that increasing the resolution from 12-km to 4-km in 4DVar largely improved the initial vortex structure, enhancing the small eye and the inner-wind maximum. The newly generated vortex specific background covariance used in 4DVar helped to remove some unrealistic features in the wind field showed by the 4-km 4DVar that used the non-vortex static covariance. The adjustment in the initial condition brought the intensity and structure forecast to be in better agreement wit the observations. The mean errors of the maximum wind speed and track forecasts by both 4-km 4DVar experiments were smaller than those by the 12-km 4DVar. In contrast, the mean errors of the sea-level pressure forecast showed that the 12-km 4DVar produced a lower pressure at earlier stages of the forecast. This was attributed to the fact that in the higher-resolution 4DVar analyses, the model was not able to maintain the very small eye, double eyewall and strong pressure gradient features for a longer time. Detailed diagnostics of the surface structure revealed that the asymmetry was well maintained by all the 4DVar cases. However the 4-km 4DVar that used the vortex-specific background covariance gave a better fit with the observations. The control experiment in which no data was assimilated did not develop the inner-core structure and continuously over-estimated the storm intensity. Results from the second experiment performed on Hurricane Earl further demonstrated the advantages of using 4DVar to correct the initial conditions of a hurricane forecast model. For this case, the ADR winds were continuously assimilated during a period of 5 days. Overall the analyses showed that having continuous DA events better estimated the long-term intensity of the storm. The errors of the 4DVar intensity forecasts were evidently smaller than the forecasts with no DA (non-DA) initialized with GFS. The initial conditions were clearly adjusted to match the observed structure. Detailed verification of vertical structures showed that the 4DVar analyses constantly improved the inner-core structure reproducing the inner-wind maximum and maintaining the small eye during the intervening forecasts. This work also demonstrated one of the advantages of assimilating 3D winds in 4DVar since it was able to simulate the deepening and strengthening of the vortex during the rapid intensification event clearly observed by the ADR data.