Abstract

This study developed a hybrid four-dimensional (4D) ensemble–variational (4DEnVar) radar data assimilation (DA) system for the Weather Research and Forecasting model. The 4DEnVar approach incorporated ensemble covariances at multiple time levels to assimilate observations distributed in the assimilation windows. By approximating the evolution of the background error using 4D ensemble covariance, use of the tangent linear and adjoint models was avoided. The impact of 4DEnVar radar DA on convective-scale analyses and forecasts was examined through comparison with 3DVar and 3DEnVar methods for the case of a squall line that occurred over southeastern China. In comparison with the other methods, 4DEnVar produced both smaller root mean square innovations for radar reflectivity and radial velocity and better analysis of the vertical structure of reflectivity. The corresponding relative humidity and vertical wind in convective regions were strengthened. Ultimately, 4DEnVar produced a substantially improved forecast, including improved quantitative precipitation and reflectivity forecast skill, and better representation of the squall line in terms of both areal coverage and intensity. In contrast, 3DEnVar improved the analysis and forecast modestly in comparison with 3DVar. Furthermore, sensitivity experiments indicated that a moderate assimilation window and a stronger ensemble weighting factor used in 4DEnVar could produce superior forecast results. The wind, temperature and water vapor were also improved by 4DEnVar, with the largest bias reduction for water vapor at low and middle levels. The improvements of 4DEnVar were further verified and shown effective using a mesocale convective system case and a local convection case.