Large-scale blending in an hourly 4D-Var framework for a numerical weather prediction model

The Met Office limited-area model (LAM) data assimilation (DA) system is based on an hourly 4D-Var method with lateral boundary conditions (LBCs) provided by the Met Office global model (GM). This set-up was introduced operationally in summer 2017. The work presented here improved the representation of large-scale dynamics in the LAM DA, which is now integrated on an extended domain and therefore also needs to represent larger scales than were present in 2017. To improve the representation of the large-scale motions in the convective-scale system, we choose to take advantage of the better estimation of these scales from the host model. The method presented here is called large-scale blending (LSB), a spectral nudging solution which is applied to the Met Office LAM (UKV). Spectral nudging allows the LAM evolution to be constrained by a host model with a better representation of the larger scales while preserving the smaller scales predicted by the LAM. In this article, we investigate different options for the implementation of LSB, such as cut-off wavelength and update frequency. The LAM forecast is typically improved for various variables over a forecast range from about T+10 to T+20 by the introduction of LSB. We also see a reduction in the internal gravity-wave activity generated when new lateral boundary conditions are introduced to the LAM from the latest GM forecast. This research proves the benefits of a better representation of large-scale motions for LAM forecasts.