May 28, 2009 Meeting Summary
In today's meeting, Mingjing Tong presented some of her recent work on the impact of assimilating conventional data and comparing it with radar data assimilation. For her experiments, Mingjing mentioned that all runs start from the H209 6h forecast, which included Qingfu's initialization as a guess field for observations. For the RDA experiments, radar radial velocity is assimilated only. For the CNV experiments, only conventional data is assimilated, and for the FDA experiments, a full set of data (including radar radial velocity, conventional data, and satellite data) is assimilated.
For Hurricanes Felix and Karen, three plots of 1) surface pressure and 10m wind, 2) 850 hPa streamlines and isotachs, and 3) 500 hPa streamlines and isotachs were created for experiments with 1) no DA, 2) radar DA, and 3) conventional DA. For Felix, the radar DA produced decreased wind speeds at 850 hPa, while no DA showed reduced wind speeds at 500 hPa. Mingjing mentioned that the plots for conventional DA were very close to the tcvitals. For Karen, the radar DA showed reduced wind speeds at 850 hPa compared to the other experiments, but the surface pressure and wind were closer to observations.
Next, Mingjing looked at the values of observed minus forecast values for 1) the u,v flight-level wind reconnaissance and dropsonde profile, 2) NESDIS cloud drift and water vapor imager u- and v-wind, and 3) surface marine u- and v-wind with reported station pressure. To investigate the differences in these types of data and between observations and forecasts, Mingjing conducted sensitivity experiments. In the presentation, vertical velocity (W) was shown as the control variable, and the surface wind and pressure, 850 hPa streamlines and isotachs, and 500 hPa streamlines and isotachs were plotted. No significant differences were observed between assimilations without W and those with W. For the experiments that used radar data from 0.5 hr before and 0.5 hr after the analysis time, the wind speeds were not dampened as in the other experiments.
Track and intensity plots for Felix and then Karen were then presented with CNTL representing H209. For Felix, any kind of assimilation improved track forecasts. For intensity (maximum 10m winds), all assimilations, except CNV3, improved the intensity forecast, with a slight delay in improvement until 60h. The same is shown for the minimum pressure intensity plot. For Hurricane Karen, the track forecast was improved by assimilation in the first 12h only, and no consistent improvement was observed for intensity forecasts.
Mingjing concluded by saying that her future work includes further checking the quality of the airborne radar radial velocity data. She also planned to make parallel model runs for all 2009 hurricane cases with or without airborne radar data availability. If no radar data is available, the impact of assimilating conventional and other data will be examined.
Next, Vijay Tallapragada presented a few slides about the recent operational HWRF failure for the 2009052700 run of INVEST 91L. The issue was traced back to an abnormal SST value sent to the coupler by POM at the 67.2 hr point of the forecast. At this point, the temperature anomaly computation was equal to zero, which set the SST value to -99.0. This issue is still under investigation, and measures to take to prevent this occurrence are under discussion.