Four-dimensional variational data assimilation of heterogeneous mesoscale observations for a strong convective case

On 19 September 1996, a squall line stretching from Nebraska to Texas with intense embedded convection moved eastward across the Kansas-Oklahoma area, where special observations were taken as part of a Water Vapor Intensive Observing Period sponsored by the Atmospheric Radiation Measurement program. This provided a unique opportunity to test mesoscale data assimilation strategies for a strong convective event. In this study, a series of real-data assimilation experiments is performed using the MM5 four-dimensional variational data assimilation (4DVAR) system with a full physics adjoint. With a grid size of 20 km and 15 vertical layers, the MM5-4DVAR system successfully assimilated wind profiler, hourly rainfall, surface dewpoint, and ground-based GPS precipitable water vapor data. The MM5-4DVAR system was able to reproduce the observed rainfall in terms of precipitation pattern and amount, and substantially reduced the model errors when verified against independent observations. Additional data assimilation experiments were conducted to assess the relative importance of different types of mesoscale observations on the results of assimilation. In terms of the assimilation model’s ability to recover the vertical structure of moisture and in reproducing the rainfall pattern and amount, the wind profiler data have the maximum impact. The ground-based GPS data have a significant impact on the rainfall prediction, but have relatively small influence on the recovery of moisture structure. On the contrary, the surface dewpoint data are very useful for the recovery of the moisture structure, but have relatively small impact on rainfall prediction. The assimilation of rainfall data is very important in preserving the precipitation structure of the squall line. All the data are found to be useful in this mesoscale data assimilation experiment. Issues related to the assimilation time window, weighting of different types of observations, and the use of accurate observation operator are also discussed.