October 8, 2009 Meeting Summary
Bill Lapenta started off the meeting with an update on the status of the GFDL/HWRF testing for GFS phase 1 implementation. Bill started on slide 3 of his presentation with the GSI changes. He noted that tropical storm pseudo-sea-level pressure observations had the potential for a significant impact on GFS vortex initialization and forecasts and thus, initialization and forecasts in the GFDL and HWRF. Bill intends to build upon this presentation and provide updates to TPC on a regular basis. Next, Bill focused on the summary of GFS tropical cyclone simulations in the GFS/GSI phase 1 implementation on slide 9. For the complete 2008 Atlantic season, there was an overall reduction in track error, with an 8% reduction at 120h. For the complete 2008 East Pacific season, track error was also reduced, specifically by 24% at 96h and 28% at 120h. For a partial 2009 Atlantic season, with no storms beyond September 15, there is not a change in track error through 48h. Also, for the 2009 season, Bill (03L) is the only storm to last beyond 72h, thus far. For a partial 2009 East Pacific season, there was a reduction in track error through 120h, except at 96h. At 96h, track error increased by 2% while at 120h, the error was reduced by 28%. Bill noted that there have not been enough long-lived storms thus far in the Atlantic for 2009, so errors at later forecast times are dominated by one storm.
Next, Bill presented graphs showing track and intensity errors for the operational GFDL versus the GFDL with the phase I GFS versus GFDL with phase I GFS and no bogus. For the GFDL with phase I GFS, there was a small increase in track error through 96h (<3.5%) and a 2% increase in track error at 120h with intensity error showing a 6% improvement at 72h and a 6% degradation at 96h for the 2008 Atlantic complete season compared to the operational GFDL. For the 2008 East Pacific, which was an incomplete set of only 5 storms, the GFDL with phase I GFS showed an increase in track error through 120h. When the GFDL with phase I GFS and no bogus was run for the 2008 Atlantic season, there was a reduction in track error from 24-120h, and this track error was lower than the operational GFDL at 36, 72, and 96h.
Next, Bill asked Vijay Tallapragada to give an update on the HWRF with GFS phase I (H048) runs for the 2008 Atlantic and East Pacific cases. H048 was very comparable to operational HWRF in track error for the 2008 Atlantic storms, except at 96h where H048 reduced track error. H048 intensity error for the 2008 Atlantic showed only slight improvements and degradations. For the 2008 East Pacific cases, there was a 5% increase in track error but also big improvement, especially at 48h. Most of this improvement could be attributed to Norbert (15E). The intensity error for the 2008 East Pacific HWRF cases still showed a high bias with H048 producing stronger storms.
Next, Morris Bender presented some additional results from HWRF and GFDL model runs with and without bogusing. The results shown were for Hurricane Ike (09L) initialized at 2008090400. In the plots shown, for the HWRF cross sections, HWRB (HWRF with new GFS and with bogus) plots are in the top panels and HWNB (HWRF with no bogus) are in the bottom panels. For the GFDL cross sections, GFD1 (GFDL with new GFS and with bogus) are in the top panels and GFNB (GFDL with no bogus) are in the bottom panels. Morris mentioned that the goal of this work was to try to understand the difference in HWRF and GFDL. Slide 4, showed the HWRF cross sections for HWRB and HWNB for Ike at 2008090400. Morris noted the broader HWRF vortex. Slides 6 and 7 showed 24h forecasts for Ike. Here, the GFDL with no bogus (GFNB bottom panel) has started to develop the storm while the HWNB has not. In fact, Morris noted that the HWNB didn't develop a storm until around 72h. Morris mentioned that he had run several cases confirming this trend, and potential cause could be that GFDL does not filter in the upper levels while HWRF does. Morris concluded by presenting the environmental field for Elida 2008071400 on slide 14. The operational environmental field at 850 mb is in the top panel and environmental field from the new GFS at 850 mb is in the bottom panel. Since the vortex is clearly visible in the bottom panel, Morris suggested that an increase in filtering may be necessary to completely remove the vortex in the new GFS.
Then, Young Kwon presented part 5 of his work with the sensitivity of air-sea exchange coefficients (Cd and Ch) on hurricane size and intensity. For his most recent work, Young added a new Cd from Powell's 2007 paper. It is represented by opt_6, and was an addition to the Cd from Powell's 2003 paper. Also, Young changed the surface physics code to use the 10m wind speed for Cd/Ch instead of the lowest model level wind speed, which was at 37m. Next, Young ran a whole cycle of Hurricane Erika (2009) using Cd values from Powell's 2003 and 2007 papers. Then, Young presented a brief description of the Cd (as calculated in Powell's 2007 paper) and Ch (from Jun Zhang et. al's 2007 work) values he used for this work.
The 2007 JHT results plot shows Cd as a function of wind speed. The red symbols represent the profile for the 20-160m layer and blue symbols represent the 10-160m layer. Here we see that Cd decreases with wind speed up to 60 m/s. Young then explained that wind speed (U) is proportional to the log (Zo). Basically, if there is enough wind speed at the surface then the slope will be related to U* and the intercept will be related to Zo. Using the equations in slide 4, Cd can be found to be a function of Zo only. Next, Young showed how the data points could be divided using mean boundary layer wind speed. Cd profiles versus wind speed were then shown, with the Cd from Powell's 2007 paper in maroon. Compared to the Cd value from Powell's 2003 work (in purple), the Cd value for Powell 2007 is pretty large, and decreases more rapidly past 40 m/s than the other profiles. Past 60m/s Powell's 2007 Cd levels off.
Next Young explained that his Ch values were obtained from Jun's paper on the CBLAST experiment (2002-2004). The figure on slide 8 shows the CBLAST flight path through hurricane Isabel (2003). The flux was directly measured and the equation for Ch was used. Compared to the HEXOS data from 1996, CBLAST data is pretty close, and the plot on slide 9 shows that there's not much evidence that Ch will increase with wind speed. This is the reasoning behind using Jun's Ch values for this work. Next Young showed a graph comparing Ch/Cd profiles. The red line represents the operational Ch/Cd, the purple line represents Jun's Ch over Powell's 2003 Cd, and the maroon line represents Jun's Ch over Powell's 2007 Cd. The red line is mostly greater than 1 meaning Ch is greater than Cd. The purple line drops below 1 near 30-40m/s, and the maroon line drops significantly below 1 at 40m/s before rapidly increasing. The red dotted line on this graphs represents a finding from Emanuel that suggests Ch/Cd values greater than 0.75 if a storm is to be maintained. Young then showed a bar graph of intensity bias for Erika (2009). Operational HWRF is shown by the red bar, yellow is the CBLAST Ch and blue uses Powell's 2007 Cd. For the 2007 Cd, there is not much change in bias early on when the storm is weak then a slight reduction later on. Young then showed a bar graph for Bertha, Hanna, Gustav, and Ike that doesn't include results with Powell's 2007 Cd value. Since the intensity bias was greatly reduced using Jun's Ch value and Powell's 2003 Cd value (yellow bar), there is a good chance using Powell's 2007 Cd value will show further improvement.
Young concluded with his plan for future work. Young will next test the 10-160m surface layer Cd estimation to avoid a Ch/Cd ratio lower than the critical value. Sea spray effects will also be added since spray amount is a function of 10m wind speed. Finally, Young will conduct PBL sensitivity tests for next year's HFIP work.