March 12, 2009 Meeting Summary
Vijay Tallapragada presented some slides on the final status of the HWRF 2009 T&E. HWRF 2009 T&E started on February 2nd and finished on February 28th. Almost 800 cases were run including 2008 Atlantic storms Bertha-Paloma, 2008 East Pacific storms Boris-Polo (minus Cristina and runs for Boris occurring before July 1), and Benchmark cases Dennis, Katrina, Rita, Wilma and Dean. Track error for all 2008 Atlantic storms shows H209 improvement over HWRF (operational) and H047 (control) through 72 hours with a degradation at 120 hours due to Hurricane Bertha's northward track bias. Compared to HWRF and H047, this H209 degradation at 120h was 14-17 nm loss in skill and and increase in westward bias. H209 showed as much as a 6% improvement over HWRF and H047 at 48h. Track error plots which included error values for the GFS showed that model improved after 84h while H209 showed a loss of skill after 96h. When Hurricane Bertha is removed from the track error calculation, the H209 track error values are almost identical to HWRF and H047 values. The intensity error plot for the 2008 Atlantic storms shows an H209 loss of skill at 48 and 72h with improvement at 96 and 120h compared to HWRF and H047. When quantified, the degradation at 48 and 72h is only on the order of 1-2 kts. There is an increased positive bias shown at all forecast hours for H209 showing H209 is producing stronger storms (especially seen in the cases of Fay, Gustav, and Hanna).
The track error values for H209 for 2008 East Pac storms showed a 2-30% improvement at all forecast times compared to HWRF error values. H209 also reduced the westward and northward model bias compared to HWRF. This was largely due to H209 forecasts for Norbert. The intensity error plot shows an H209 loss of skill at 48 and 96h, but this degradation is only on the order of 1-2kts compared to HWRF values.
For the Benchmark cases H209 showed a slight degradation in track at 96h, which was not statistically significant, followed by an improvement at 120h. Intensity error for H209 was lower than that for HWRF at all forecast times. While H209 still showed a negative bias, this bias was less than that for the Benchmark control (H41A).
In the future, individual storms may be selected to demonstrate the effects of individual upgrades included in H209. Currently, the HWRF is making the transition to Cirrus (P6). Runs for Fay, Gustav and Hanna are being compared to results obtained on P5. While track statistics show a very small difference between P5 and P6, intensity statistics showed a larger difference between P5 and P6 with better results from P6 for Gustav and Hanna and worse results for Fay. Vijay mentioned working with Carolyn Pasti on timing issues experienced on Cirrus. Vijay also explained a genesis mode experiment where the HWRF was run with and without initialization for two Invest cases that eventually developed into tropical storms and two Invest cases that did not. The runs without initialization do not strengthen the storm nearly as much as the HWRF and H209(with initialization) do.
Mingjing Tong presented some initial results from her work with H209 and the new GSI. She conducted the following experiments for Felix and Karen: CNT1 is the control using Qingfu's initialization and no radar data assimilation; ROLD uses the initialization and radar data assimilation using the 2008 HWRF and the old GSI with a radar data thinning grid of 9km x 9km, FOTO + FGAT; RNEW is the same as ROLD but uses the 2009 HWRF and the newest GSI; RDAW is the same as RNEW but with vertical velocity as an additional control variable; RD18 is the same as RDAW but the radar data thinning grid is 18 km x 18km; RSVL is the same as RDAW but with shorter vertical decorrelation length scale; RFGT is the same as RDAW but only uses FGAT; RRLC is the same as RDAW but only keeps Qingfu's relocation without intensity and size correction.
For Felix, the CNT1 showed a more northwesterly track compared to the rest of the experiments, and RRLC showed a much weaker intensity compared to the others. For Karen, most experiments show a more northwesterly track compared to what was observed except for the RRLC experiment. Mingjing mentioned that she wanted to use RDAW for future storms, and its performance for Felix and Karen was better than the CNT1 a majority of the time. Future work with this included examining the impact of assimilating other data, running more cycles for Karen and Felix, and running experiments for Katrina and Ophelia.