October 1, 2009 Meeting Summary
Morris Bender discussed some of his recent results with the GFDL model with regards to the GFS Phase I Implementation. Morris mentioned that he started to re-run the 2007 and 2008 hurricane season using the GFDL with the new GFS. He managed to complete 56 cases thus far: all of Fay (06L) and Ike (09L), most of Gustav (07L) and half of Hanna (08L) at 00Z and 12Z. With the new GFS, he saw a very different and stronger vortex with a slight degradation. All of the plots shown used the 56 cases. The first plot Morris presented shows track skill for operational GFDL (in black), GFDL with new GFS (in red), operational GFS (in green) and the new GFS (in blue). From 12 through approximately 84 hours of the forecast, both GFDL models showed more skill than the GFS, with the operational GFDL slightly better than the GFDL with the new GFS. After 84h, the GFS models showed more forecast skill, and the new GFS showed more skill than the operational GFS throughout. Next, Morris discussed the track forecast skill for the operational GFDL, the GFDL with the new GFS, and a no bogus GFDL with the new GFS (in green). Morris explained that the no bogus GFDL started from the GFS fields with no relocation. In the track forecast skill plot, the no bogus GFDL started off with the least skill, but it quickly improved and showed more skill than the GFDL with the new GFS between 24 and 36 hours to 72h, from which it showed the most skill. Then, a plot of intensity forecast skill was described. Again, the no bogus GFDL started off with the least skill, but showed more skill than the other two versions of the GFDL between 24 and 36h onward. Morris noted that the GFDL with the new GFS showed intensity skill very similar to the operational GFDL with a slight degradation at 96h onward. Morris's discussion concluded with the decision that the rest of the 2008 season would be run with no bogus. This would include Paloma (17L) which was a small but intense storm that formed in November of last year.
Next, Bob Tuleya presented some of his recent work on reducing initial storm moisture over land in the HWRF. Bob began by presenting a slide from Vijay Tallapragada's previous presentation discussing the HWRF model sensitivity in landfalling storms. Essentially Bob reminded everyone that the inclusion of the radiation and land surface temperature bug fixes in the H209 code, as well as other changes, showed a negative impact on track and intensity forecasts. The bug in the operational HWRF eliminates the surface moisture supply over land and occurred because radiation wasn't being called at regular intervals. Since that bug was "fixed" for H209, H209 may enhance the intensity bias over land. To assess the impact of the bug fixes one could reduce the moisture supply for storms initialized over land, much like what is done with the GFDL initialization. For Bob's experiments, he reduced the atmospheric moisture of the storm vortex over land by 50% in the inner nest if any perturbations were seen over land at all (referred to by HP5L in the subsequent discussion). Currently, the HWRF uses the same storm vortex over land as it uses over the ocean, and this change would put the HWRF more in line with the GFDL.
Bob then showed a plot of intensity error for Fay and Hanna (on the left) and Gustav and Ike (on the right). In these plots, H209 is represented by the red line and operational HWRF by the blue line. Here, we see that the H209 showed much higher intensity errors for Fay and Hanna compared to HWRF. While these errors were higher compared to HWRF for Gustav and Ike, the disparity between the two was not as large. Bob next presented an intensity forecast featuring an early attempt to reduce the positive land bias by changing the momentum field. This also involved modifying or eliminating the composite vortex (represented by HXCS and the yellow line), but it is difficult to do over land. The track forecast showed an HXCS track very similar to H209 and HWRF, which is not really an improvement.
Next, Bob presented some of his results by first showing the moisture field of the GFDL (on the left) compared to H209 (on the right) for Fay at 2008082000. Overall, H209 showed a very wet center making it hard to tell the storm's location based on moisture alone. The GFDL showed a very reduced amount of moisture where Fay is located. The increased moisture in H209 could be due to the bogus/cycling carrying the moisture forward in time. Then, a plot of the moisture field for the GFDL (on the left) compared to HP5L (on the right) was shown for Fay. Here, we see reduced moisture where the storm is located which more closely resembles the results for GFDL. A plot showing a cross section of the initial wind field for Fay was then shown. Again, GFDL is on the left and HP5L on the right. Here, HP5L shows a tighter and deeper storm which might be the result of some of the composite storm showing through. Next, Bob showed a plot of cycling's effects on the initial wind structure for H209 (on the left) and HP5L (on the right). Bob proposed that the moisture reduction in HP5L could be causing the initial storm to be less compact, shown by the larger storm size of HP5L. Bob noted that the relative angular momentum was not reduced but higher, and that this issue has the potential to decrease further into the integration.
Next, intensity plots were shown for Fay at 2008081818 (on the left) and 2008081918 (on the right). These two plots show that the impact of reducing the atmospheric moisture could produce a very dramatic impact throughout the forecast (as seen on the left by HP5L forecasts much closer to observations) or only affect day one or two of the forecast (as seen on the right where HP5L is slightly better than H209 through about 36h). Then a plot of intensity error was shown using all cycled results for H209 (in red), HWRF (in blue), and HP5L (in green). HP5L showed a large error reduction compared to H209 throughout the forecast period, and was similar to HWRF through about 36h and again after 84h. Bob concluded with a summation stating that using the HP5L configuration would make the HWRF more in line with the GFDL, which is practical, but it only affects land cases. Overall, HP5L increased the intensity skill for 18 cases of Fay. Next Bob proposed running a complete suite using HP5L for Hanna, Gustav, and Ike. The impact of using HP5L on Hanna would probably not be as large as for Fay due to Hanna's more limited interaction with land.
Bob then showed a plot of track and intensity for Erika (06L) on 2009090212. For track, the HWRF with the operational GFS (HGFS) is in red, and for intensity, it is shown by the yellow line. The intensity forecast shows no intensification of the HGFS until day 4. It was noted that the forecast looks like the NCAR model forecast while the HWRF was too cyclogenic. From this Bob commented that two models using each others initial conditions might be useful.