October 29, 2009 Meeting Summary
Mingjing Tong presented some of her recent work with the HWRF cycling system. For Mingjing's version of the HWRF, she created a new HWRF outer and inner domain based on the storm center position and filled it in with the GFS 3, 6, and 9h forecast. Then, she relocated the HWRF 3, 6, and 9h forecasted hurricane vortices and made the size correction and intensity and structure corrections optional. Thus, the background fields consisted of the environmental flow from the GFS forecast plus a modified hurricane vortex. Next, a GSI analysis was performed using ADR (radar data), TCP, conventional, and satellite data which then produced a 126h forecast. For her work, Mingjing focused on Hurricanes Ike and Gustav from 2008 and Hurricane Bill from 2009.
First off, Mingjing showed results from Hurricane Ike. For the track and intensity plots, the operational HWRF (H209) is in pink and Mingjing's experiment (called ADR1), which used relocation plus size correction plus data assimilation, is in green. For Ike, the track error for ADR1 is higher than for operational HWRF and increases with time. The intensity error peaks at 12h then decreases to less than that for H209 by 60h. Next Mingjing showed intensity plots for MSLP (on the left) and maximum 10m wind (on the right) for 3 cycles of Ike. In these plots operational HWRF is in red and Mingjing's experiments are represented by the blue, green and yellow lines. The 1st cycle included ADR (radar) data, and shows a much weaker storm from Mingjing's experiments than both operational and observations. The 2nd cycle, which included no radar data, showed a very weak storm from Mingjing's work initially in pressure but not in 10m wind intensity. The storm was weaker than operational HWRF throughout. Intensity for the 3rd cycle, which included radar data showed a similar pattern to that from the 2nd cycle.
Next, Mingjing presented results from Hurricane Bill. The track error for Mingjing's experiment is pretty close to that for operational HWRF. However, the intensity error is once again largest at 12h before decreasing to less than operational error values by 72h. After that, the ADR1 error increases again. The 1st cycle intensities for Bill with radar data show a much weaker pressure for Mingjing's experiments than for operational HWRF. While the ADR experiment 10m wind values are closer to operational values, they initialize with 170kt winds before decreasing to near 80kts by 6h. The 2nd cycle (no ADR data) of Bill ADR intensity plots show a big discrepancy from operational HWRF values, and a much weaker storm. For the 3rd cycle, with radar data, the ADR intensity values are closer to H209, however there is still a large difference between the two.
From these results, Mingjing explained that she wondered from where such a large difference in her experiments and operational HWRF was coming, so she conducted more experiments for each storm to find the error. For these plots, H209 (in red) is operational HWRF, GUES (in blue) is the HWRF 6h forecast vortex relocated to the correct position, QF01 (in green) is Qingfu's HWRF initialization with relocation and size correction and intensity correction, ADR3 (in purple) is Mingjing's experiment with relocation plus size correction plus data assimilation, ADR5 (in teal) is her experiment with relocation plus data assimilation, and ADR6 (in yellow) is Mingjing's experiment with HWRF initialization plus data assimilation. For Ike, the quick weakening seen with the previous experiments was alleviated when the entire initialization was used (QL01 and ADR6), and the intensities for these two experiments were closer to operational values. For Bill, QF01 is close to operational HWRF values of intensity. For this storm, Mingjing noted that there seemed to be an imbalance between pressure and intensity and the 10m wind values decreased greatly initially. For Gustav, there was no quick weakening of the storm shown in the 10m plot. Again QF01 intensity values were closest to H209.
Mingjing concluded by showing cross sections for each of the storms for the wind field (1st column), temperature perturbation (2nd column), and relative humidity (3rd column). The first row shows results from QF01, followed by ADR5 in the 2nd row and ADR6 in the 3rd. From these plots. Mingjing focused on the relative humidity for ADR5 being extremely low. Mingjing further explained that she believed this dry moisture field was her main issue, and that this could be a coding issue. Mingjing planned to further examine this and run more experiments to reach a conclusion.
Qingfu Liu then presented his work plan for HWRF initialization. The analysis cycle Qingfu will use includes a 9h forecast at times t-3, t, and t+3h, then a creation of a new HWRF domain and an interpolation of the GFS analysis onto the new HWRF grids. This is followed by replacing the hurricane vortex and modifying its structure and then running the GFS analysis to produce a 126h forecast. Qingfu then elaborated on the background field. He mentioned that it would consist of the interpolation of 6h GFS analysis fields onto the new HWRF grids, the removal of the GFS vortex, and then adding the modified HWRF 6h forecast vortex into the GFS environment. Qingfu then explained that all data in the area of the hurricane should be used in the 3D-var data assimilation because the GFS vortex is removed, meaning that all information in the area of the hurricane is lost. Thus, the filter domain could be as large as 2500 km in diameter. Also data from the area of the hurricane could lead to an improvement in the upper level storm structure.
Qingfu concluded with his work plan for initialization in 2010. First off Qingfu will increase the filter domain size for the GFS storm while reducing the filter domain size for the HWRF storm. This ensures that just the right amount is filtered out so that any storm remains from the GFS storm won't contribute to a stronger HWRF storm. Second, Qingfu will base the storm size correction on observations instead of an average between observations and the 6h model output. This correction is limited at 15%. The first and second steps in this work plan should lead to a weaker storm in the HWRF, which tends to over-intensify storms. The 3rd item in Qingfu's work plan included using cloud top information from satellite data to correct initial storm depth. This item is the least urgent of the three and will be something left for later on.