March 18, 2010 Meeting Summary
Vijay Tallapragada presented an overview/update on the status of FY2010 HWRF experiments conducted in the past three months. Specifically, this presentation focused on HWRF results using GFS phase-2 data. The different versions of HWRF compared in the results presented included H050, the new baseline runs which included bug fixes and the use of GFS phase-1 data; H054, which was H050 plus changes to the initialization plus surface physics changes of Cd/Ch values calculated based on observations plus gravity wave drag parameterization; H056, which was H054 with GFS phase-2 data (which included upgraded physics and an increased resolution); and H56C, which was H056 with surface physics corrections. As a reminder, Vijay mentioned that the test cases included in these experiments involved Cristobal, Dolly, Fay, Gustav, Hanna, Ike, and Omar from the 2008 Atlantic season; Elida, Fausto, Genevieve, Marie and Norbert from the 2008 East Pacific season; Bill, Claudette, Danny, Erika, Fred, Henri and Ida from the 2009 Atlantic season; and Felicia, Guillermo, Hilda, Ignacio, Jimena, Linda, Olaf and Rick from the 2009 east Pacific season for a total of about 610 cases. Vijay then presented an updated version of the HWRF testing schedule for FY2010. He noted that the H56C experiments, in red, were just concluding while the HYC4 and HYC5 experiments, in red, were ongoing. The rest of the experiments, in black, had concluded.
Next, Vijay addressed the corrections made to the momentum and heat flux computations in the surface physics module resulting in experiment H56C. Thanks to Hyun-Sook Kim, a coding error was discovered in the surface physics module that affected the computation of heat and momentum fluxes. This was particularly true at low wind speeds. Once this error was corrected, all 610 cases for Atlantic and East Pacific basins were rerun using GFS phase-2 data. The resulting track and intensity forecasts indicated a positive impact from the correction. To illustrate the issue with the heat flux computations, Vijay presented figures showing a scatter plot of wind speed versus latent heat flux, a latent heat flux contour plot, and a wind speed contour plot for a 96h forecast for Bill (03L) first for H056 (with the error) then for H56C (without the error). The H056 scatter plot shows a discontinuity with two groupings of points while the latent heat flux plot shows a big hollow ring of high flux values circling the center of the storm. By comparison, the H56C scatter plot shows no discontinuity or hollow ring of extremely high values in the latent heat contour plot. The storm size has also been reduced. As for the scatter plot outliers from 0-15m/s at 0 W/m^2, Vijay said more investigation as to their cause was needed, but it was perhaps a boundary issue. Vijay then explained the main differences between the H056 and H56C executables by first stating that executables for H056 and H56C both have the correct Cd and Ch values based on work from Powell (2003) and CBLAST (2007), respectively. Vijay went on to say that the problem arose when one of the experimental executables, designed to use the same enthalpy fluxes as that of the operational HWRF for wind speeds under 30m/s, was used for the H056 test suite instead of the CBLAST Ch value. The new executable used in H56C is correct with CBLAST Ch values used over all wind speeds.
Then, Vijay presented track and intensity error statistics for the GFS phase-2 experiments for the 2008/2009 Atlantic basin. In these plots, H050 is in orange, H054 is in red, H056 is in purple, and H56C is in green. The track error plot shows H56C with reduced error values over H056, especially at 120h. Vijay pointed out that for this plot, H050 had the lowest error values which could be due to the fact that it used GFS phase-1 data. Some issues with GFS phase-2 data, used for H054, H056, and H56C, were observed for Bill and Gustav, which affected the results. The intensity error plot again shows H56C improved over H056 overall, except at 120h which Vijay mentioned could be attributed to Gustav here. The intensity bias plot for the Atlantic storms showed a negative bias on the order of 5-6 knots out to 96h for H56C, which Vijay mentioned seemed appropriate at the model resolution used here.
Next, track and intensity error statistics were presented for 2008/2009 East Pacific storms. The color scheme used in these plots is the same as for the Atlantic statistic plots. The East Pacific track error plot showed H56C with the lowest error values throughout the forecast period. H56C especially showed a reduction in error by 20-25 kts at 72 and 96h. There was also a large increase in skill from H056 to H56C at 120h where H056 skill was degraded. The intensity error plot again shows an improvement in skill by H56C compared to H056 and the other experiments, and the intensity bias plot again showed H56C's negative bias early in the forecast period with bias values closer to zero later on.
Vijay concluded by presenting plots showing frequency of superior performance for track and intensity in both basins as well as an overview of work in progress. For the superior performance plots, H56C, in green, consistently showed more instances of better performance than H056 for both Atlantic and East Pacific basins in track and intensity. Current HWRF work in progress included an evaluation of the results presented here through additional diagnostics for a select set of cases. Specifically, Vijay mentioned using James Franklin's verification software involving the isolation of land points and re-doing of intensity verification for land-falling cases as well as the verification of metrics for storm size calculations. Additional HWRF diagnostic output included simulated GOES brightness temperatures for water vapor and infrared channels and simulated radar reflectivity. Vijay also mentioned current work on generating RFC's for the proposed implementation and an upcoming evaluation of HWRF-HYCOM results.