NOAA Privacy Policy | NWS Disclaimer
N.O.A.A. logo HWRF banner image National Weather Service logo

February 14, 2011 Meeting Summary

Tomi Vukicevic presented a seminar on "Hurricane Ensemble Data Assimilation with HWRF Experimental System: Successes and Challenges". Tomi first described the motivation for this work, which involved HFIP's goal of improving track and intensity hurricane forecasts. Then, she mentioned that at HRD/AOML, airborne, as well as other, observations are assimilated into the experimental version of HWRF (HWRF-X) with the objective to reduce forecast errors due to a poor representation of the initial conditions on TC scales. For the work presented, HRD's Hurricane Ensemble Data Assimilation System (HEDAS) was used. HEDAS uses HWRF-X as the forecast model with 2 nested domains (at 9/3 km resolution and 42 vertical levels) with a static inner nest and explicit microphysics. The data assimilation uses a square-root ensemble Kalman filter, and the ensemble system was initialized from GFS-EnKF ensemble, which included 30 ensemble members. For the 2010 hurricane season, HEDAS was only run when Doppler radar wind data was available from NOAA P-3 flights resulting in 19 cases. Types of observations assimilated include flight level and dropsonde wind, temperature, humidity and pressure and radial wind superobs. Tomi noted that most of these observations were located in the core of the storms. Of the 19 cases run, most of them came from Earl 07L, with a few from the early stages of Karl 13L, one from Richard 19L, and three from Tomas 21L. Most of the observations assimilated came from the Doppler radar data.

Next, Tomi presented some results comparing HEDAS analysis to best track for all 19 cases. For most periods, HEDAS had similar intensity values to best track, except for where Earl was its most intense. For RMW, HEDAS overestimated for weak systems, and HEDAS had a tendency to underestimate the storm radius. When looking at a snapshot of storm structure, HEDAS underestimated intensity for all storms, but was closest to Tomas' H*Wind analysis intensity. HEDAS also captured Earl's storm size very well. In the plots of overall forecast performance for the 2010 cases, HWRF-X is represented by the red line, HEDAS by the blue line and operational HWRF by black. Overall, HEDAS had lower intensity errors than HWRF and HWRF-X and was comparable with track errors.

Tomi then presented some of the problems identified by these HEDAS experiments. As previously mentioned, HEDAS significantly underestimated maximum surface wind values, especially when a TC was strong or intensifying (as for Earl). To try to understand this underestimation problem, a diagnostic procedure was developed using cylindrical coordinates and the azimuthal average of tangential wind, radial wind, vertical wind, vorticity, surface pressure, etc. Plots of the max axisymmetric tangential velocity for different TC intensities (TS, Cat 1 hurricane, cat 3 hurricane, and cat 4 hurricane) were then shown. For the stronger intensities, the forecast systematically reduces vortex strength while the analysis increases it, creating a very up-down pattern. The same pattern can be seen in axisymmetric vorticity but with an overall increase. The MSLP decreases due to the influence of observations. However, Tomi noted that this result shows that the dropping pressure and vorticity spin-up are not balanced with the wind evolution, which is a big problem with the system.

Tomi next presented some diagnostics that could be used to pinpoint the main source of the previously mentioned problem. For the vortex structure, there was a significant phase discrepancy in the radial wind between the prior and posterior states below 3km which is indicative of a large discrepancy between forecasts and observations in the PBL. Looking at the depth of the inflow layer in the PBL, a bimodal pattern was present in the posterior state, which became more pronounced for a stronger storm, and was not seen for the prior state. This indicated the forecast model had a strong tendency to develop erroneous PBL structure. Tomi summed this up with a flow chart showing the coupling between model and analysis errors. For the start up forecast, the model PBL errors create a large difference between the background forecast and observations of wind at less than 3 km in the storm core. The EnKF analysis is updated with observations correcting the winds, perhaps too much. In the short forecast (1h), the effect of the updated observations is reversed while PBL height increases, and this is passed back to the EnKF analysis which continues this cycle much like a feedback loop.

Tomi concluded by pointing out that the 2010 HEDAS experiments produced a skilled 3-D atmospheric state analysis with the potential for significant improvements to the intensity forecast. Future work includes representing model error in the ensemble by physics perturbations, getting more observations, and developing a hybrid data assimilation system for HWRF. Creating forecast diagnostics and verification with satellite observations was also mentioned.

Please e-mail comments, questions, or suggestions about the contents of this webpage to Janna O'Connor, at

Home | HWRF Main Page
EMC | NCEP | National Weather Service | NOAA | Department of Commerce