Impact of JEDI LETKF Assimilation with Radar Observations on Ensemble Forecast for Landfalling Hurricane Ida (2021)

Radar is known being an effective observation platform capable of providing valuable information on hurricane structures at high temporal and spatial resolutions; its benefits for improving model forecasts of hurricane track, intensity, and hurricane-induced precipitation through various data assimilation (DA) techniques have also been broadly documented in previous studies. This study is the first to investigate the impact of convective-scale radar DA on ensemble hurricane analysis and forecast using JEDI LETKF, coupling with FV3 limited area model utilizing the physics suite consistent with the operational Hurricane Analysis and Forecast System (HAFS). The studied hurricane Ida made a landfall at Port Fourchon, LA (category 4) on the 29^th of August 2021. Ensemble forecasts at a 3-km resolution are initialized from 30-member Global Ensemble Forecast System (GEFS) analyses and serve as the background for DA experiments. A 3-hour DA window beginning ~5 hours ahead of landfall is applied in which conventional data are assimilated hourly. Additional MRMS reflectivity (Z) or/and quality-controlled WSR-88D level II radial velocity (Vr) data are assimilated at a 30-min or 1-hr interval and their respective impacts are examined.

Results show that assimilation of Vr observations, particularly from the coastal radars, can effectively improve the hurricane intensity forecast that is underpredicted in the background. The ensemble-averaged hurricane-centered azimuthal-mean analysis at the final cycle shows a significantly upright inner-core structure with much deeper and tighter circulations and a stronger warm core extending to higher altitudes for the Vr DA experiment. When verified against the best track data, the experiment assimilating both Z and Vr shows the most improved ensemble forecasts on both hurricane track and intensity up to a 15-hr lead time; assimilation of Vr alone can produce very comparable forecast performance. More frequent radar DA (i.e., 30-min as compared to 1-hr) can help fast deepen the minimum center pressure of the hurricane during the cycles while results in no major further improvement on the ensuing forecasts. On the rainfall prediction, the radar DA shows the largest benefit in improving 0- to 12-hr accumulated rainfall forecasts in terms of ETS computed at short neighborhood radii. Assimilating Z appears to slightly lessen the rainfall overforecast, which is found to be an extensive issue for all experiments. Furthermore, assimilation of the wind analysis retrieved by a Generalized Velocity-Track Display (GVTD) technique is also tested and its impact is compared with that of Vr DA. More evaluation will be presented at the conference.