The impact of Doppler radar data on rainfall forecast: a case study of a convective rainband event in Mississippi Delta using WRF 3D-Var

The assimilation of Doppler radar observations into mesoscale models remains a challenging problem. The impact of these observations on quantitative precipitation prediction was studied before but is still a question to be answered. In this study, we use a convective rainband system occurred during 29-30 April, 2005 in Mississippi Delta to further evaluate the impact of radar radial velocity on precipitation forecast and to examine the sensitivity of the forecasts with respect to forecast background statistics. In particular, we seek to answer the following questions: first, whether an ensemble-based background error statistics can improve the rainfall forecasts; second, if a 3-hour cycling with radar data assimilation can help the generation of new storms in front of the existing storm; third, if thinning (or sueprobserving) radar data relative to model resolution is necessary for optimizing their impact in the analysis field.

To address the first question, we calculate and analyze three background error covariances using 31 members of ensemble forecasts. Each ensemble member is generated by random perturbation that has Gaussian distribution with zero mean and unit standard deviation in the control variable space of WRF 3D-Var. To answer the second question, a 3-hour analysis and forecast cycling is applied and compared with a no-cycling experiment. The third question arises because radar data has higher resolutions and the distribution in space is uneven. We compare the impact of different resolutions of the radar data on the rainfall forecast. Experiments show that the background error statistics calculated using the 6-hour ensemble forecast results in a better analysis and 3-hour rainfall forecast. The radar data assimilation helps the generation of new storms but it forces the rainband to move faster than the observation. Thinning radar data has positive impact in rainfall forecast, and it also reduces the computational cost.