Direct assimilation of radar reflectivity using an ensemble 3DEnVar approach to improve analysis and forecasting of tornadic supercells over eastern China

An ensemble three‐dimensional ensemble‐variational (3DEnVar) data assimilation (En3DA) approach that directly assimilates radar reflectivity was developed based on the Weather Research and Forecasting model data assimilation system. This system adopts radar reflectivity as the control variable to avoid the need for a tangent linear and adjoint of the observation operator. Flow‐dependent covariance was introduced via ensemble forecasts updated by a group of 3DEnVar. The performance of the En3DA system was examined for two selected cases of high‐impact severe tornadic supercells over China. Results for both cases indicated that the structure of the storms in terms of intensity, coverage, and associated low‐level mesocyclones were analysed more accurately when using the En3DA approach than when adopting the 3DVar method. Hydrometeor analysis showed that En3DA provided a more physically reasonable increment of hydrometeors compared to 3DVar, especially for the graupel mixing ratio. Furthermore, the En3DA forecast was better than the 3DVar forecast throughout the forecast period for both studied cases. En3DA produced smaller errors in terms of intensity and location for supercell forecasts with respect to reflectivity and reflectivity swaths. Furthermore, the quantitative forecast skill of radar reflectivity was improved using En3DA. Errors in the wind, temperature, and water vapor forecast fields produced by En3DA were also reduced compared to those of 3DVar. Diagnostics revealed that En3DA predicted an enhanced low‐level cold pool and stronger outflows in the forward‐flank downdraft and the rear‐flank downdraft regions, which are important for tornadogenesis.