The Impact of Different Data Fields on Storm-Scale Data Assimilation

Many data assimilation techniques have been developed to assimilate radar velocity and reflectivity data into storm scale numerical prediction model. Although these research efforts have been generally successful, a clear understanding of which data fields have the strongest impact on storm scale data assimilation is lacking. Such knowledge is very important for improving storm-scale observing systems and developing new data assimilation procedures. In this study, the issue is addressed by assimilating the pseudo- observation for different model variables separately to determine the impact of different data fields on storm-scale data assimilation. This is accomplished by first creating a control simulation of a thunderstorm evolving from a thermal bubble, then extracting different type of observations from the control run. The observational data are assumed perfect, exist at every model grid point, and thus the assimilation is simply performed by inserting the data into model's initial condition for each data assimilation cycle. By using the different assimilation cycles (every, 3 minutes, 5 minutes, 10 minutes), the different types of pseudo observations, and the different amount of pseudo- observations, the impact of each data fields on the data assimilation can be quantitatively evaluated by monitoring a variety of different error statistics. It is found that horizontal velocity and water vapor exert the greatest impact on the storm evolution. Horizontal wind fortunately can be derived from Doppler radar; however, water vapor field is not easily observed for storm scale. It is also found that more observations are assimilated, the shorter spin-up time the model needed to build up accurate precipitation field, and shorter data assimilation cycle contributes to more accurate result.