Qingfu Liu presented some of his preliminary results involving cloud physics impacts in HWRF. Before showing results, Qingfu provided some background explanation for his work. Based on his test results, Qingfu mentioned that evaporation of larger-sized raindrops has a small impact on hurricane track and intensity forecasts (new code provided by Brad Ferrier). A possible explanation for this was that rain evaporation from the sub-grid and grid scale quickly saturated the atmosphere below deep convection. Qingfu also detailed the following impacts of deep cumulus convection on storm structure: First, the horizontal pressure gradient is relaxed at the storm's upper levels, and grid scale flow transports heat upward and outward. Thus, the eyewall size increases with height, perhaps too much, which could cause a larger eyewall size at the surface. Second, Qingfu's tests showed that the model top only played a minor role in lack of track and intensity skill. Qingfu attributed eyewall tilt mostly to the cloud top in the Simplified Arakawa-Schubert (SAS) scheme. Additionally, sub-grid scale heating may have too low of a distribution in the eyewall in HWRF, and heat is unable to be transported to the top by the grid-resolved scale flow. Finally, HWRF-produced storms are not well organized at upper levels causing the upper level steering to have less of an impact on their movement. This could be shown by the fact that Easterly winds decrease with height causing less westward movement.

Qingfu compared three sensitivity experiments: The control or H209 (2009 operation HWRF), QL03 (which used the highest cloud top in the SAS scheme with no random number method), and QL04 (which used the lowest cloud top and no SAS convection scheme). Only Hurricane Bertha was run, and results are shown for early in the storm's lifetime. E-W cross sections for each of the three experiments show higher clouds on the west side of the storm for both QL03 and QL04. For the track and intensity plots, the QL03 and QL04 tracks are closer to best track values than H209, for the dates shown (2008070306-2008070406). Track statistics show that both QL03 and QL04 have similar values and lower track errors than H209 throughout the 126h forecast period. However, intensity errors showed mixed results with H209 having lower values at 12-24h and 96h. Qingfu then presented a discussion on deep cumulus convection in the HWRF mentioning sub-grid scale parameterization and grid scale condensation as the two physical processes that compensate one another in deep convective areas. Thus, if the cloud top was set to be randomly distributed between the highest and lowest cloud, the air parcel might have less energy to move upward compared to a case with no SAS scheme. More work on this topic will be done in the future, including running cases from the list of high resolution cases.

Vijay Tallapragada then presented an update on the HWRF transition from P5 to P6. He mentioned that all HWRF codes on the P6 were recompiled and this led to a faster execution time on P6 compared to P5. Specifically, there was a 10% gain for a 126h forecast. Due to limitations with RSL in HWRF, further improvements in runtime may not be possible with the current configuration. However, results from P5 and P6 show better agreement. A wall-time chart comparing P5 and P6 coupled HWRF runs showed that P6 had a faster execution time (excluding wait times) by almost an hour over P5. Some of the biggest gains were seen at the nmm-si, ocean-init, and relocation steps. An ATCF file for a Hurricane Gustav 12h forecast (cycled at 2008090118) showed almost identical values for P5 and P6.