Implementation and testing of WRF Digital Filter Initialization (DFI) at NOAA/Earth System Research Laboratory

Without some type of special treatment, a model's initial mass and momentum fields are typically sufficiently out of dynamical balance to initiate spurious gravity waves at the start of a model forecast. These waves can seriously contaminate the model's solution during the first few hours. The digital filter initialization (DFI) was developed as a simple, conceptually straightforward procedure to reduce these initial imbalances and likewise reduce the amplitude of spurious gravity waves produced at the forecast beginning. The DFI acts as a low-pass time filter, in effect creating initial conditions without imbalances associated with short-period (less than a few hours) time oscillations. Experience with the Rapid Update Cycle (RUC) model has shown that the DFI is very effective in producing 1-h forecast background fields that have substantially less spurious gravity-wave small-scale "noise". The use of the DFI has thus been critical to the success of the 1-h cycling in the RUC forecasts.

Accordingly, it was regarded as essential to have a similar capability for WRF, as it will be used as the forecast model for the Rapid Refresh (RR) 1-h cycle that is expected to replace the RUC in NCEP operations in 2010. At NOAA/ESRL (ESRL hereafter) the DFI has been implemented in the RR developmental version of WRF-ARW, and through collaboration with NCAR/MMM, was also included as part of the official version 3 release. To allow utility and flexibility in use of the DFI, the WRF model was modified through the addition of a DFI namelist and DFI related variables to the registry. In addition, calls to a DFI module as well as time keeping information were added to higher-level WRF routines. This allows a user to run the DFI and subsequently a WRF simulation using a single executable and namelist input file.

The presentation will discuss the work performed by ESRL and NCAR on implementing the DFI into WRF. Results from test simulations conducted at ESRL both with and without the DFI will be presented. These results reveal that the current implementation of the DFI is both computationally fast enough for operational use as well as effective at creating a more balanced initial state for use in the Rapid Refresh.