Surface Data Assimilation Using an Ensemble Kalman Filter Approach with Initial Condition and Model Physics Uncertainties

The assimilation of surface observations using an ensemble Kalman filter approach is evaluated with MM5 as the forecast model. Two cases are examined, in which the assimilation is performed from 1200 to 1800 UTC using hourly surface observations. We produce ensembles in three different ways, by using different initial and boundary conditions, by using different model physical process schemes, and by using both different initial and boundary conditions and different model physical process schemes. The three ensembles are compared in order to investigate the role of uncertainties in the initial and boundary conditions and physical process schemes in ensemble data assimilation. In the initial and boundary condition ensemble, spread is associated largely with the displacement of atmospheric baroclinic systems. This results in large spread in locations where the field is sensitive to the placement of such systems. In these regions of large spread, observations have a relatively large impact on the analysis produced by the filter. On the other hand, in the physics ensemble, the displacement of atmospheric systems among its members is small. Its spread comes from the difference in model physics, which results in large spread in thermodynamic variables such as potential temperature and dewpoint temperature, rather than in the dynamical variables such as the wind fields. The combined initial condition and physics ensemble has the properties of both of the previous two ensembles. It gives the largest spread and produces the best forecast for most of the variables, in terms of the rms difference between the ensemble mean forecast and the surface observations, and in the creation of mesoscale features important to the severe weather events of the day.