January 29, 2009 Meeting Summary
Zhan Zhang presented his work on the high resolution HWRF experiment. The parameters for his experiment included: using a resolution of 0.09, 0.03 (instead of 0.18, 0.06); using a domain of 432x862, 118x198 (instead of 216x432, 60x100); using a time step of 27 and 9 seconds (instead of 54 and 18 sec.); and coupled with HYCOM every 9 minutes (with the same resolution). Due to the large CPU time needed for this experiment, the ability to use the restart option was required. HYCOM was used because it yields reproduceable results for the HWRF, while POM does not. For this experiment, Hurricane Rita from 2005092012 was used. The high resolution runs produced more visible smaller scale features and larger and stronger wind fields. The track and intensity of Rita were also significantly improved. However, run time for the hi-res HWRF is about 12 hours compared to 2 hours on Dew for lo-res experiments. Resolution parameters are also hard-coded in the HWRF.
Mingjing Tong showed some slides on her experiments in using radar data alone to correct storm intensity and structure and to move the storm to the correct position. The HWRF initialization used flow from the GFS analysis/forecast plus the modified HWRF 6h forecast hurricane vortex. Compared to previous experiments, the new experiments did not correct T, pressure or moisture within the HWRF initialization. These experiments were performed for Felix and Karen. Results showed that radar data alone was not enough to correct storm size or location, perhaps due to data coverage. Mingjing suggested obtaining more wind observations from different storm quadrants. Her future work includes performing these experiments with recently acquired Hurricane Katrina data.
Vijay Tallapragada presented results from the HWRF transition to Power 6. He mentioned that all scripts, source code, fix and parm files were transferred to Cirrus, and source codes were compiled using the same compiler options as those use on Dew. There were some changes to load leveler scripts due to different job submission guidelines on Cirrus. The HWRF run time on Cirrus is approximately 68 minutes for 80 processors and approx. 63 mins. using 98 processors. The coupler on Cirrus took 10 minutes compared to 5 minutes on Dew, and there were some minor differences in the results. Two test cases were run on Cirrus for Fay: 2008081518 and 2008081600. The tracks and intensities for the 2008081600 run were very similar for P5 and P6. The cold start values for track and intensity showed similar tracks but a deeper, stronger storm in P6 than in P5. It was mentioned that there could be an issue with the random number generator on Cirrus that could be looked into. Vijay mentioned he planned to run a storm all the way from start to finish and check statistic values for differences to further assess the situation.
Janna O'Connor showed a few slides based on HWRF runs she had completed using different options in the HWRF 2009 code. Only one run for Gustav was completed (2008082900), but she showed that the HWRF could successfully be run using a combination of different options. Her runs included: using HYCOM with INIT, GSI, and GWD; using HYCOM with INIT and GSI but no GWD, using HYCOM with INIT, GWD, and no GSI, using POM with INIT, GSI, and GWD, and using POM with INIT, GWD, and no GSI.