Correlation Length Scale and Data Assimilation of Radar and Satellite Observations in Convection-Permitting Ensemble Simulations of a Severe Thunderstorm Event

Data assimilation of Doppler radar observations using EnKF has been utilized for over a decade and proved to be of great value to both research and operational forecasts. The newly-launched GOES-16 satellite also provides a great opportunity for improving the prediction of severe weather as a complementary to current ground-based radars, with proof-of-the-concept studies already demonstrated the potential impact of assimilating all-sky infrared radiance observations of GOES-16 ABI on numerical simulation of convective storms using EnKF. However, detailed structure of correlation between radar or radiance observations and other model prognostic variables and hence a physically-based knowledge of the radius of influence of such observations remain largely unexplored.

This study employed a kilometer-scale EnKF data assimilation system and explored the correlation length scales between synthetic radar radial velocity, reflectivity, infrared radiance and various dynamical and thermodynamical variables both in the horizontal and vertical directions. As expected, radar reflectivity is strongly correlated with water condensates, and the absolute correlation with respect to distance drops to quasi-constant values within several tens of kilometers away from the observation. The correlation between reflectivity and water vapor or wind is much weaker, while wind is strongly correlated with itself. For satellite radiance, vertical distributions of correlation between radiance of ABI channel 8 (upper-troposphere water vapor channel) and various water condensates show very strong peaks exceeding -0.5 at upper troposphere to tropopause, and iced-phased condensates have stronger correlations with radiance than water-phased particles. Although not as significant as for radar reflectivity, the absolute horizontal correlation between radiance and ice condensates also experienced a persistent drop with respect to the increase of distance.

These results indicate not only the correlation length scales of radar reflectivity, radial velocity and satellite infrared radiance with various dynamical and thermodynamical variables, but also the spatial length scales of storm structures and associated variables that are depicted by different observations. As such, they may experience different error structure, and a variable-specific localization scheme might be beneficial for the simultaneous assimilation of radar and satellite observations.