February 19, 2009 Meeting Summary
Vijay Tallapragada presented a summary of the preliminary results from the 2009 HWRF pre-implementation testing. As a reminder, the 2009 HWRF (H209) has the following features: new GFS/GSI, new initialization, gravity wave drag, land-surface temperature bug fix in the nested domain, and radiation bug fix in the nested domain. Currently, 40-50% of HWRF testing is finished, and completed storms include: Bertha 02L, Cristobal 03L, Dolly 04L, Fay06L, Josephine 10L, Kyle 11L, Laura 12L, Elida 06E, Hernan 09E, Julio 11E, Karina 12E, Dennis 04L, and Katrina 12L. In the results shown for Bertha and Fay, the following conventions apply: H209 is the 2009 version of HWRF, HWRF is the operational version of HWRF, H047 is the new GFS/GSI or ctl, AVNO is the operational GFS, PRDO is the new GFS/GSI parallel, H29B is H209 without the land-sfc T bug fix, and H47N is the 2008 HWRF with Qingfu's new initialization.
For Bertha, we see an increase in track error after 72h compared to HWRF, H047, GFDL, AVNO, and PRDO, which could be attributed to error from initial cycles. The intensity errors are lower for H209 when compared to GFDL, HWRF, and H047.
For Fay, there is a strong positive bias, and we once again see an increase in track error around 72h with intensity error for this storm the highest compared to the other models. To pinpoint where this loss of skill originated, plots for track and intensity error for Fay were created that included values for H29B and H47N experiments. While there were not large differences in track error between models, the H29B and H47N runs had lower intensity error values when compared to H209, by as much as 8 kts at 48 and 72h. This suggests that the bug fixes could be making the storm worse. Bob Tuleya suggested examining the wetness coefficient to see what kind of an impact it may have on a storm like Fay.
Track and intensity errors were shown for Cristobal, Dolly, Josephine, Kyle, Laura, and Dennis. For the most part, these plots showed some improvement in error values for H209 compared to HWRF. Katrina had the lowest values of H209 track and intensity errors compared to other models. This could be due to the HWRF negative bias for Katrina that may have been "corrected" by H209.
Side by side plots of HWRF and GFS 500 mb geopotential height were shown for Bertha to see if the large scale pattern in GFS was replicated in the HWRF. At the 30h forecast, HWRF shows Bertha taking a northern turn while the GFS shows a more westerly motion. The difference field between GFS and HWRF geopotential heights does not show an extended area where GFS values are larger than HWRF values. Compared to 30h, the difference values at 78h show a much larger area where GFS values are larger than those from HWRF, and this area is even larger at 114h. Intensity values for HWRF show Bertha at a category 1 storm when it should be a category 3, according to best track values. Vijay mentioned that he is running some experiments with an expanded domain (by 20 degrees in E-W) to diagnose this issue.
Additional issues are are mentioned below: For Bertha, the initial location in PRDO (as well as AVNO) for several cycles early on (between 2008070300 and 2008071200) was off by as much as 100nm compared to observations. This issue was due to the failture of relocation (lack of tcvitals) in GFS operations as well as PRDO. Also, the storm size for H209 storms is much larger when compared to H047, and Fay intensifies over land, perhaps due to the land-sfc T bug fix? Currently, H209 runs are halted but expected to resume in the next few days.