Dr. Xuyang Ge from Penn State University presented some of his work titled, "Towards Dynamical Initialization of Tropical Cyclones." Dr. Ge introduced his work by describing his motivation, which was to provide a more realistic structure for the initial vortex with the hopes of improving hurricane intensity/structure prediction. He then mentioned common approaches to this which included using a bogus vortex, like in HWRF, and data assimilation, like 3dvar and EnKF. In a comparison of analytical solution versus model simulation, Ge mentioned that a model is more able to capture the vortex structure.

Next, Dr. Ge described part 1 of his presentation: tropical cyclone dynamic initialization (TCDI) for COAMPS. In this method, a Rankine vortex is first spun up in a quiescent environment (using the research model TCM3). Next, the first guess field is split into a TC vortex and an environment field. Finally a nonlinearly balanced TC vortex is merged into the environmental field using three conditions: if r is less than 100km, if r is between 100 and 300 km (called a blending zone), and if r is greater than 300 km. The blending zone has a linearly outward slope for altitudes above 500 hPa. Ge then described the TCDI system in COAMPS-TC. Starting with the NOGAPS/NCEP analysis, 3dvar data assimilation is performed. Then the TC vortex is removed and another vortex is inserted, which was generated from TCDI (which uses a nudging method on MSLP and not 10m max winds). Finally the forecast model is run.

Ge illustrated performance of TCDI by showing plots comparing the control (CTL), taken from COAMPS operational runs which used 3dvar-assimilated synthetic observations, and TCDI. These results are for Hurricane Bill from 2009. In the surface pressure pattern, TCDI better describes the inner core showing a stronger storm. The same is true for the vortex structure plots. The most significant difference between CTL and TCDI can be seen in the warm core plots. For the convective structure comparison, TCDI is better able to capture the more organized eyewall and spiral bands than CTL. Intensity error and bias plots, provided by NRL, show significantly smaller TCDI errors when compared to CTL. However, track error is not improved using TCDI. Future work with TCDI includes nudging wind to correct storm size/intensity, using consistent research and operational models to reduce bias, looking at the environmental influence, and physics initialization in the spin-up vortex.

Part 2 of Dr. Ge's presentation involved preliminary results on a pseudo-ensemble hybrid 3dvar (PE3DVAR) system for TC initialization. The goals of this work included estimating the flow-dependent background error covariances (B) for TC initialization and comparing PE3DVAR to standard 3DVAR. Ge next discussed the ensemble method of creating a background error covariance, which uses statistics of ensemble deviations to estimate model error. For this method, a TC vortex library was established for use in vortex replacement. For this library, the WRF-ARW 3.1 was run using different initial conditions such as initial vortex size, perturbation sounding field, latitude effect, and physics options. The goal was to obtain as many different vortices as possible. These vortices were then binned based on their maximum wind speed. Next, Ge described the PE3DVAR system, which began using the GFS analysis which was used to generate ensemble members (Ge mentioned there were currently 48 members). Then, the TC vortex was replaced using the vortex library. This was then used to generate pseudo-ensemble members which then created a flow-dependent multivariate background error covariance which was then put into 3dvar.

Next, Ge described the experiments he compared for part 2 of his work. They were the GFS analysis, 3DVAR (which was the GFS after 3dvar using the NMC method B), and PE3DVAR (which was GFS after 3DVAR using the new ensemble-based B). The NMC method uses statistics of forecast error in the short-term forecast range. The first case for these experiments uses data from a 3-D dual Doppler radar wind assimilation. Looking at the inner core wind structure comparisons for these experiments, Dr. Ge noted that the RMW for PE3DVAR is very contracted in the analysis. Looking at the 3DVAR and PE3DVAR fields with GFS subtracted, an increase in the inner core structure can be seen. For the warm core structure comparisons, the PE3DVAR field with GFS subtracted shows an amplitude increase not present when GFS is subtracted from the 3DVAR field. Dr. Ge concluded that both 3DVAR and PE3DVAR improved wind structure but only PE3DVAR improved the warm core structure. The inner-core moisture field shows no signal from 3DVAR but enhanced moisture in the PE3DVAR field. Looking at track plots, there are only slight differences between the three experiments. Intensity plots for MSLP show PE3DVAR slightly improved over GFS and 3DVAR. However, results are more mixed for max wind. The second case uses data from other inner-core observations such as from the Multi-platform Tropical Cyclone Surface Wind Analysis (MTCSWA), which can resolve a stronger inner core wind. Ge presented comparisons of the warm core structure for this case from each experiment. Plots showing PE3DVAR with GFS subtracted again show an improvement in warm core structure.

Dr. Ge concluded by presenting his summary for part 2 of his presentation. First, preliminary results indicate using PE3DVAR can improve the inner-core temperature and moisture fields of TCs during analysis. Also, PE3DVAR performs better than 3DVAR using the multivariate, anisotropic covariance ensemble-based method yet maintains a similar computational cost to 3DVAR. Ge's future work with PE3DVAR includes continual evaluation of the PE3DVAR system and improvement to the TC vortex library.