September 17, 2009 Meeting Summary
The first meeting presentation was given by John Ward and Bill Lapenta, and it detailed the GFS and HWRF FY10 testing and implementation schedule. John Ward first presented an overview of the proposed FY10 implementations for the global branch. These implementations could be broken down into three phases. The first phase, with an anticipated mid-December implementation date, consisted of adding new observation data sources, including tropical storm pseudo-MSLP observations. This phase also consisted of implementing improved techniques in the GSI analysis as well as unifying the two versions of the GFS post processor in production. Next, John showed hurricane track error plots for the experimental GSI/GFS bundle (shown in red) and operational GFS (in green). The track errors for the 2008 Atlantic storms show a reduction in track error by the GSI/GFS bundle at all forecast times. This is mostly true for the 2008 East Pacific storms, with the GSI/GFS bundle showing a reduction in error at 120h by about 100 nm. There is a slight degradation in track shown by slightly higher errors for the GSI/GFS bundle compared to the operational GFS for the 2009 Atlantic storms. However, John noted that Bill (03L) was the only long lived storm, so the track error values past 96h were mostly based on it. The track error plot for 2009 East Pacific storms showed reduced error values for the GSI/GFS bundle compared to operational GFS at all forecast hours except 96h. Next, John gave a brief overview of implementation phases 2 and 3. Phase 2’s new boundary layer, mass flux shallow convection, and deep convection schemes are expected to improve verification scores and reduce grid-point convective storms, as shown by the accompanying plots. Phase 3 consists of two resolution increases and many physics changes that have shown positive results so far. John concluded by saying that thus far, each phase is built off of the same baseline. Once the phases are based off of one another, improved results can be expected.
Next, Bill Lapenta presented a plan for the HWRF 2010 implementation. The HWRF team is preparing for this implementation by first working on the DTC code merge, involving HWRF 2009 physics and WRF version 3. This merged code will be configured as the operational HWRF08 or the baseline, and benchmark testing is currently ongoing with case testing to begin on or about September 21. A planned delivery date to DTC is October 31. The next step in preparations for the 2010 implementation is coupling the HWRF with HYCOM, which uses Atlantic RTOFS. This will be followed by work on the surface flux exchange coefficients, convection depth, and initialization. Bill concluded with a draft of the GFS and HWRF implementation schedule. In it, phase 1 of HWRF testing involves the DTC merged code, phase 2 includes the HWRF coupled to HYCOM with 2008 physics (defined as HWRF code with GWD and bug fixes turned off), and phase 3 includes the above phases with HWRF coupled to HYCOM using 2009 physics (GWD and bug fixes on) as well as any improvements involving the surface flux, convection depth, and initialization. Each HWRF round of testing will follow a GFS round of testing.
Next, Qingfu Liu gave a presentation titled “Determination of the Highest Cloud Top in the SAS Scheme and its Impact on Hurricane Forecasts.” Qingfu first presented a background on his work. Qingfu said that calculations of highest cloud top and mass flux at the cloud base have some of the biggest impact on hurricane track and intensity forecasts within the cumulus convection parameterization scheme (SAS) of the HWRF. In addition to being used to define vertical heating distribution, the above mentioned two parameters can be used to calculate vertical transports of water vapor and momentum, among other variables. Qingfu also explained that highest cloud top and cloud base mass flux calculations depend on assumptions of entrainment and detrainment within the rising parcel. Qingfu defined the highest cloud top as the level of neutral buoyancy or vanishing CAPE which allows overshoot.
Qingfu then gave a brief review of his previous presentation on the impacts of precipitation on parcel buoyancy, showing that parcel buoyancy increases due to precipitation. Qingfu also showed that cloud tops using pseudoadiabatic ascent (shallow convection removed) were higher than those using moist static energy and precipitation. Track and intensity plots for Hurricane Dean were shown with HWRF (green), added rainfall effect (blue), and added pseudoadiabatic effect (red). The pseudoadiabatic experiments produced a track closer to what was observed, especially later in the forecast. Compared to the other experiments, the pseudoadiabatic experiments also produced better intensity values.
Qingfu concluded by mentioning that his current work with Hurricane Fay has shown reductions in intensity bias, especially at the first 72h, with a slight track forecast improvement. More storms will be run in the future. Additionally, his future work includes using the new SAS scheme in the HWRF and adding to that shear impacts on parcel entrainment and detrainment.
Next, Hyun-Sook Kim presented some preliminary results from her HyHWRF simulations of 2009 Atlantic storms. For reference, Hyun-Sook defined HyHWRF as HYCOM coupled to the 2009 HWRF. Hyun-Sook then mentioned that for this work, NCO para RTOFS were used as initial and boundary conditions for the HYCOM.
Next, storm tracks were shown for all storms through Fred in the Atlantic. This was followed by simulation results for Bill (03L), Danny (05L), Erika (06L), and Fred (07L). For track and intensity error plots, the solid lines represent the mean error while the bars represent variances. For Bill, there were 31 cases. Overall, the HyHWRF system (represented by the red lines) had lower track and intensity errors compared to H209 (in black) and operational HWRF (represented by the green lines). The overall storm track and intensity for HyHWRF were also fairly close to observations for Hurricane Bill. For Danny (05L), which had 11 runs available, HyHWRF had lower track and intensity errors compared to H209 and operational HWRF. HyHWRF’s track was pretty close to observations, and its intensity was very good until 08/28. Erika (06L) had 12 runs available before it dissipated and once again, HyHWRF showed much lower track and intensity errors, especially intensity, compared to the other versions of HWRF. Finally Hyun-Sook showed simulation results for Fred (07L), which had 15 runs. Early in the forecast period, the HyHWRF track error was very similar to the values for HWRF and H209. From 72h through 96h, HWRF has lower track errors than HyHWRF and H209. From 102h on, HyHWRF again has the lowest track error values. With the exception of 90h, the HyHWRF intensity forecast error is lower than those for HWRF and H209. While the HyHWRF track is very close to the best track early on, it veers further north than what was eventually observed. The HyHWRF intensity is also weaker than what was observed for Fred.
To conclude, Janna O’Connor gave a brief presentation as a follow up to Hyun-Sook’s work that showed the HyHWRF track and intensity statistics for Bill (03L), Danny (05L), Erika (06L) and Fred (07L). All statistics were based on data from the H209-HYCOM runs conducted by Hyun-Sook. In all plots, the track error is on the left and intensity error on the right. Operational HWRF is represented by the red lines, H209 by the cyan lines, HyHWRF by the purple lines and GFDL by the green lines. For Hurricane Bill, HWRF, H209, and HyHWRF track errors are very close after 72h. Before that time, HyHWRF has lower track errors than the other two models and GFDL. For intensity error, HyHWRF shows a higher value through 12h, but has lower errors than HWRF, except at 72h, for the rest of the forecast period. H209 has the lowest intensity error from 24-60h for Bill. For Danny (05L), HyHWRF and H209 have very similar track and intensity errors, which are both lower than those for HWRF at all forecast times. For Erika (06L), HyHWRF has similar track error values compared to the operational HWRF and H209 through 36h, with lower error values to 48h. The HyHWRF intensity error is lower than HWRF and H209 for the entire forecast period. HWRF and HyHWRF track errors for Fred (07L) are very similar throughout the forecast period, while H209 has the highest track error from 48h to 96h. For intensity error, HyHWRF has lower values until 24h. At that time, HyHWRF and GFDL have higher error values than both HWRF and H209 (until 72h where H209 error rapidly increases).