February 4, 2010 Meeting Summary
S. G. Gopalakrishnan (Gopal) gave a presentation titled "Recent Developments with the HWRFX System at HRD." Gopal began by acknowledging the model development team, including collaborators, and presenting the mission summary. He mentioned that the HWRFX model was meant to serve as a research tool to improve forecast guidance and accomplish four goals. The first goal, which had been started, was to understand environmental and vortex-scale processes at a high model-grid resolution of 1-3km. The second goal was to incorporate into the model a better representation of physical processes in the tropics, which was dependent upon the first goal. The third goal was to incorporate observations on a vortex scale, and the fourth goal was to advance initialization procedures. Gopal mentioned that the 2-4th goals are yet to be achieved. Next, Gopal described the wish list and developments associated with HWRFX. The rest of the presentation was focused on these: advanced diagnostic software for inner core analysis, a suite of idealized cases to understand the hurricanes modeled, a simple 1-dimensional mixed-layer ocean model for impact assessment, and a multiple moving nest for understanding inner core processes down to a 1km resolution (which is in the testing and evaluation phase). The three plots on this page show the Doppler divergence taken from P-3s (top), the simulated divergence from HWRFX (middle), and the secondary circulation from an ideal case (bottom).
Next, Gopal described the HWRFX system, which is an experimental version of the HWRF designed specifically to study the intensity change problem at a very high model resolution of 1km. First, there is data acquisition and preparation which automatically takes data from NCEP and uses HWRF,GFS, and GFDL grib data. Next, the HWRFX is run. It uses most of the physics that HWRF does with the NCAR radiation scheme. HWRFX also uses the WRF-NMM core with a multiple moving nest, which is still in testing. For the HWRFX, there is no cycling. The HWRFX output goes into visualization and diagnostics, which are based off of Java code. Gopal noted that it is possible to do inter-model comparisons and use ensembles. Next, a list of 2009-2010 milestones was presented for HWRFX. This list included the running of 69 cases for the HFIP high-resolution test plan at an HWRFX resolution of 9km (parent domain)/3km (nest domain)/1km (innermost nest) as well as a real-time demonstration of the HWRFX system at 9km/3km. Making the 3rd moving nest and the vortex initialization scheme valid for the 1km resolution, evaluation of the visualization tools to aid analysis of the high resolution simulations, alternating physics valid for tropical predictions, and transitioning research enhancements to DTC/EMC are also on the list.
Next, Gopal presented some results from the high-resolution HFIP runs. For track forecasts, the HWRFX at a higher resolution improved track forecast for lead times 24-48h. Meanwhile, the higher resolution HWRFX improved intensity forecasts for lead times 0-30h. Then, some results from the 2009 hurricane season were presented. For these runs, Gopal explained that HWRFX was run at 9km/3km with no ocean coupling and no vortex initialization and no cycling. HWRF initial and boundary conditions were used. HWRFX was also run in real-time, twice a day at 00 and 12Z, and 126h forecasts were produced for 78 cases. The ATCF files were provided to the multi-model regional ensemble, and using the NJET to run HWRFX came out to 2.5 hours for simulation time and 50 mins. for post-processing, on average. Gopal noted that HWRFX real-time demo simulations could be viewed at: https://storm.aoml.noaa.gov/hwrfx/. Next, a plot of track forecast error was presented for storms Ana through Erika for the 2009 Atlantic hurricane season. The HWRFX (in red) is comparable to HWRF (in blue) and GFDL (in green). In the plot showing 10m wind forecast error (line graph) and bias (bar graph), the HWRFX intensity bias and intensity error are both high indicating the need for evaluating the use of HWRF initial conditions and physics in the HWRFX. From the 2009 Atlantic season, Gopal presented plots for Hurricane Bill and Erika. For Bill, a 10m max. wind speed plot is shown for HWRFX using HWRF initial conditions (in red) and GFS initial conditions (in green). Gopal explained that this plot was used to better understand what was going on with HWRFX. The HWRF init. conditions produce stronger wind speeds for Bill. For Erika, the HWRF init. conditions again produce a stronger storm, but using GFS init. conditions also produces in increased intensity than what was observed. Gopal also mentioned that the storm tracks were also affected.
Next, a vortex scale evaluation was described where axisymmetric vortex structures were compared with airborne Doppler composites. The Doppler composites (observations) in the left column were compared with the 3km HWRFX runs (in the middle column) and the 9km HWRFX runs (right column) for Katrina. For the tangential wind, the 3km HWRFX better simulated what was observed. This was also true for the radial wind, as the 9km HWRF deepened this field too much. The higher resolution HWRFX also did a better job simulating vertical wind, vorticity, and divergence compared to the 9km HWRFX. Gopal then presented simulations of the evolution of an isolated vortex in a uniform easterly flow, which was an ideal case for the HWRFX. Here, the 9km and 3km model are compared and even the 9km model does a decent job with this simulation. More examples of diagnostics were presented in the next few slides and they included an example of how a warm core forms with rapid intensification, the relationship between structure and intensity changes (which were consistent with theory), and modeled mean structure of a hurricane compared to Doppler observations (where the HWRFX proved to be consistent with theory and observations). Gopal then provided an update on the innermost nest work. He mentioned that at the higher resolution, mesovorticies were occurring in the HWRFX. To determine the center in these cases, a centroid was used in the nest motion algorithm. To address the issue of projection and interpolation issues with multiple moving domains, a rotated latitude-longitude grid was used for all computations in the HWRFX.
Gopal concluded by first presenting a list of questions involving HWRFX issues and ongoing discussions. They included assessing the lessons learned from the 69 test cases in relation to storm structure and intensity beyond statistics, determining the role of shear in the 2009 hurricane season and whether a higher resolution helped to predict shear-vortex interactions, looking at the outcome of the sensitivity experiments for Erika and Danny from the 2009 Atlantic hurricane season, and re-evaluating whether surface layer properties are consistent with observations. Finally, a list of issues and future research was shown. Gopal noted that the resolution was needed to understand the vortex of the storm to prevent tuning of the model for the wrong reason. Also, while the 1-3km resolution was needed to understand the multi-scale prediction problem, it wasn't a solution to it. Lastly, Gopal questioned the physics suite for high resolution, asking how the right physics should be evaluated and whether we should move past a maximum wind at the lowest model level.