Large-Scale Blending in the Rapid Refresh Forecast System Version 1

The Rapid Refresh Forecast System (RRFS) is a high-resolution ensemble prediction system built upon a foundation of a frequently (i.e., hourly) updated data assimilation framework using hybrid 3DEnVar and EnKF. The RRFS covers all of North America at 3-km grid-spacing, and is poised to become a critical tool for convective-scale weather prediction in the U.S. The RRFS leverages a partial cycling approach for both its ensemble and deterministic control configurations. The land states are fully cycled and the atmospheric states are refreshed twice per day from global ensemble perturbations in the case of the EnKF or the full atmospheric state is replaced by that from the global model in the case of the EnVar system. In this talk we focus strictly on the ensemble reinitialization portion of the partial cycle framework and compare two different ensemble reinitialization strategies: the current partial cycling technique and a large-scale blending (LSB) technique. The LSB method employs a low-pass filter (Raymond 1988) and allows for the selection and extraction of only the large-scale information from the global ensemble which is then combined with the RRFS EnKF members. Such an approach eliminates the discarding of RRFS perturbations in the present reinitialization approach while maintaining the valuable periodic injection of large-scale perturbations from the global ensemble. One of the most important parameters of the LSB is the cut-off wavelength that determines the point of scale separation and is used by the low-pass filter on the host model. Various cut-off wavelengths will be compared and results will be discussed.