December 3, 2009 Meeting Summary
Young Kwon was the first presenter with his talk titled: "Sensitivity of Air-Sea Exchange Coefficients (Cd and Ch) on Hurricane Size and Intensity: Part VII - Results of Combining the New Surface Physics with New Vortex Initialization". First Young reminded everyone about his experiment design. The Cd values were taken from Powell's 2003 paper while the Ch values were from the 2007 Jun et al. CBLAST paper. For this work, Young added what he called a new vortex initialization procedure which basically involved adding more data to the GSI portion of the HWRF initialization. Young used 4 storms from the 2008 hurricane season: Fay (06L), Gustav (07L), Hanna (08L), and Ike (09L). For all plots in his presentation, the H48N (in blue) represents the operational HWRF with the new GFS, H5_5 (in red) is the same as H48N but with new surface physics, and HIST (in brown) is H48N with the new surface physics and the new vortex initialization procedure.
Next, the intensity (top plot) and track (bottom plot) errors were shown for all 4 storms. For the intensity, HIST and H5_5 were fairly similar with each experiment alternating the lowest error values, and for track, HIST produced a roughly 9% improvement in error compared to H5_5 and H48N, which were very similar. Next, bar charts of bias (top plot) and standard deviation (bottom plot) were shown for intensity values for all storms. For bias, HIST shows a reduction in positive bias from 48h onward, however, it has a larger negative bias than both H48N and H5_5 early in the forecast period. The same was true for stddev values, with HIST producing slightly higher values before 48h.
Young concluded by showing track and intensity error plots for individual storms. For Fay, all three experiments had similar intensity errors through 36h, then H5_5 and HIST alternated between having the lowest values. For track error, HIST had the lowest errors, especially after 36h. For Gustav, H5_5 and HIST are again very similar for intensity error with HIST showing slightly lower errors from 60h onward. All three experiments had very similar track errors for Gustav. For Hanna, H5_5 and HIST were again very similar in intensity errors with HIST slightly better later on in the forecast in track error. For Ike, H5_5 showed lower intensity errors than HIST except at 36 and 96h. For track error, HIST was again slightly better. Young explained that overall, the intensity improvement likely came from the new Cd/Ch values while track error improvement was likely the result of using more data in the GSI portion of the initialization.
Next, Zhan Zhang and Bob Tuleya presented their study on the positive intensity bias for hurricanes over/near land in the H209 version of the HWRF. H209 can be defined as the suggested and subsequently rejected HWRF upgrade for 2009. Zhan explained that the issue that he and Bob wanted to address was the negative impact by H209 on track and intensity forecast skill for storms from the 2008 hurricane season, specifically Fay, Hanna, Gustav, and Ike. Differences in the H209 configuration compared to the 2008 version of HWRF include initialization changes, inclusion of gravity wave drag, new GFS, and bug fixes, most importantly, an irregular radiation calculation. Zhan and Bob proposed 2 solutions for this irregular calculation, first by calling the frequency (rad1/rals) change in the namelist (which was the solution used by Bob and Zhan in this work), and second by making no radiation calculation when the nest domain is moving. Next, Zhan outlined the difficulty he and Bob faced when trying to decide what their benchmark should be. Choosing a benchmark is difficult because there have been several changes to the HWRF forecast system including the use of a new GFS, switching from P5 to P6 and two JIF emergency bug fixes implemented in HWRF 2009 production. For the experiments in this work, HWRF represents the 2008 operational version of the HWRF run on P5 and using the 2008 GFS, HBEN is the control or benchmark using the current operational HWRF with the 2009 GFS. HBFX represents HBEN plus bug fixes and the 2009 GFS. Bob and Zhan also ran the HP5L (HBFX plus a 50% initial vortex moisture reduction over land) and HMM1 (HBFX plus a momentum mixing coefficient equal to 1 instead of 0.5) experiments to fix the overintensification. It should be noted that HBEN, HBFX, HP5L, and HMM1 are all run using identical GFS data on a P6 machine with all the same code other than what is indicated. These experiments were run for Fay, Hanna, Gustav, and Ike from the 2008 hurricane season.
The background of the issue with H209 was then presented using examples from the storms selected. The surface temperature was first shown for Fay at 36h for HBEN (top plot) and HBFX (bottom plot). The top plot shows a very close land surface temperature while the bottom plot's temperature values are more realistic (note: the temperature scales are different). Fay's intensity error values for HBEN (blue), HWRF (pink), and HBFX (red) were then shown. HBEN had the lowest error values from 24h onward while HBFX had the highest error values. A plot of the surface temperature difference for Fay at 36h between HBEN and HBFX was then shown, with differences as high was 24K over land. A cross section (lower plot) was then shown to illustrate the low level temperature difference likely due to the bug fix perhaps explaining the overintensification seen with H209. Then plots of the latent heat flux difference (between HBFX and HBEN) (top plot) and the sensible heat flux difference (bottom plot) were shown for the 36h forecast for Fay. Zhan noted that there was a latent heat max difference of 600 W/m2 over the entire storm region(which was over water), and there was a positive surface sensitive heat flux difference over land. For Hanna, a plot of intensity error was shown for the HWRF, HBEN, and HBFX experiments. HBFX had the highest error values through 84h while HWRF and HBEN had fairly similar error values. To investigate what could be causing the overintensification of Hanna, the surface temperature difference between HBFX and HBEN was examined (top plot). It was shown that there was a zero difference for most of the storm area with a few areas of negative values around the boundary of the storm and a line of positive values, which Zhan and Bob guessed could be due to the storm's wake. A cross section of the temperature difference for Hanna was also examined (lower plot) and Zhan mentioned that the differences seen could possibly be caused by the change in the frequency of the radiation calculation, however, Bob noted that more investigation is needed.
Plots of intensity errors for Gustav (top left), intensity errors for Ike (top right), intensity errors for all storms (bottom left) and track errors for all storms (bottom right) were then shown. For Gustav, HWRF has the lowest errors through 72h when HBEN then has lower errors. HBFX showed a positive bias for Gustav. For Ike, HWRF showed some of the lowest error values, while HBFX again had the highest. For all storms, HWRF had the lowest intensity errors, while track errors were pretty similar for all experiments with HBFX showing slightly lower values from 36h-96h. Some proposed solutions to the overintensification issue included reducing the initial storm moisture over land and tuning model parameters such as the momentum mixing coefficient. Both of the these proposed changes would make HWRF more similar to the GFDL model. Next, Zhan and Bob showed a plot of intensity errors for Fay featuring the HWRF, HBEN, HBFX, and HP5L (in purple) experiments. In the first 36h, HP5L (featuring moisture reduction) had the lowest error values, however, by 48h, HP5L was similar to HBFX, which had the highest intensity errors. Plots of intensity and track error were then shown for Fay including the HMM1 experiment (in green). Here, HMM1 has the lowest track and intensity errors throughout. The same positive impact was seen for Hanna (not shown).
Bob and Zhan presented the following conclusions about their work: First the positive intensity bias for Fay is most likely caused by the radiation bug fix. Also, smaller intensity differences occurred even over water (as seen for Hanna), and the positive intensity bias for Hanna could be caused by the change in the frequency of the radiation calculation. Finally, the differences between the 2008 (HWRF) and 2009 HWRF (HBEN) showed as large differences as those seen between the HBEN and HBFX experiments, and by reducing the initial moisture over land and tuning the momentum mixing coefficient, improvements were seen in forecast intensity skill.
Lastly, Hyun-Sook Kim presented some of her most recent work on a comparison of simulations between two versions of HWRF. For this work, Hyun-Sook explained that there were 4 models used: H41A (uses POM and the 2008 operational HWRF) shown in green, H209 (uses POM and the 2009 version of HWRF) shown in black, HyH08 (uses HYCOM and the 2008 operational HWRF) shown in red, and HyH09 (uses HYCOM and the 2009 version of HWRF) shown in blue. Hyun-Sook ran each of these model configurations for Fay, Gustav, Hanna and Ike from the 2008 hurricane season. Next, track bias and intensity error plots are shown for each storm and a combination of storms. In the following plots, the vertical bars represent variance and the horizontal line represents the mean standard value.
Hyun-Sook characterized her results for Fay as atypical because HyH08 had lower track bias values and lower intensity errors comparable to H41A. For Gustav, HyH09 showed lower track bias values through 72h, while HyH09 was comparable to H41A and H209 for intensity errors. For Hanna, track biases were similar for all models through 72h, then HyH09 had lower values. For intensity error, H41A had lower values through 72h, then HyH08 had the lowest error values while HyH09 overintensified the storm. For Ike, HyH09 showed similar track bias values to H41A and H209, while HyH08 had lower intensity error values from 54h onward.
Next, Hyun-Sook presented track bias and intensity error plots for all 2008 storms (Fay, Hanna, Gustav, and Ike). For track, HyH09 is comparable to H41A and H209 throughout the forecast period. H41A has lower intensity error values, and HyH09 has the highest error values. Hyun-Sook skipped the subjective skill score slide and instead discussed the 2009 parallel simulations for the H41A, H209, and HyH09 (in red) models. Track bias and intensity error plots were shown for Hurricane Bill from 2009. HyH09 showed comparable track biases and intensity errors (sometimes slightly lower than H209 and H41A) throughout the forecast. For all storms from the 2009 season, there were two groups of results due to several weak and short-lived storms and only a few stronger and longer-lived storms (like Bill, Fred, and Ida). Here, HyH09 showed lower track biases and intensity error values.