The ensemble Kalman filter has the potential to provide probabilistic analyses and forecasts at any scale. In particular, this technique could provide a new tool to study mesoscale phenomenon through an ability to assimilate high spatio-temporal resolution observations such as Doppler radar and ACARS data. Existing studies have focused on implementing ensemble Kalman filters for models with periodic boundary conditions. Mesoscale, limited area applications of this technique introduce new problems, most notably how to generate lateral boundary conditions for each member. It is unclear how to apply boundary conditions that maintain an appropriate amount of ensemble variance, without synthetically disrupting the mass-wind balance.

Here we investigate several different techniques for generating lateral boundary conditions as applied to limited area ensemble Kalman filters. These techniques include using tendencies from a global ensemble, using lagged forecasts and drawing random time series from climatology. These different methods are tested on the Lorenz-Emanuel 40 variable model and the WRF model under perfect model assumptions. Furthermore, we will discuss an efficient algorithm for initially populating an ensemble for real-data applications. Initial tests with the WRF model show an asymmetry in RMS error between the west and east boundaries consistent with the expected impact of observations.