Although the forecast capability of numerical weather prediction models has improved significantly over the past decades, there still exists significant issues related to model representation of the complex dynamics of the boundary layer (BL), which impede the realization of turbine-height wind forecast accuracy. This study is an effort to revisit the basic theory of the Mellor, Yamada, Nakanishi, and Niino (MYNN) BL scheme with a focus on its function as posed for the stably stratified environment that supports the onset of a low-level jet (LLJ), a mechanism that can often result in wind ramp events, which are of special concern for the wind power industry.

The MYNN BL scheme approximates the turbulence covariance variables, which define turbulence momentum and heat flux as well as turbulent kinetic energy (TKE). These approximations are derived from the Reynolds-averaged Navier-Stokes equations and consist of a system of interdependent diagnostic expressions with terms involving gradients of the mean flow and turbulent fluxes. The influence of each term is affected by a set of weighting factors known as "closure parameters" (CPs), which have been empirically derived, as well as a diagnosed mixing length.

In this study the MYNN scheme is modified in three ways. First, an updated set of CPs are formulated specifically for a stable BL that exhibits LLJ development and associated wind ramps. A large-eddy simulation model with spatial resolution of 3-4m is used to simulate turbulence response and effect for such cases. These data provide the means to generate Reynolds-averaged values for explicit representation of covariance variables and TKE, which in turn, provide the basis for calculating new MYNN closure parameters for mesoscale numerical forecasts of wind ramps. Second, a new means of calculating the turbulent mixing length is formulated, by which vertical mixing is enhanced across the BL above that predicted by Monin-Obukhov theory when wind speeds exceed a given wind threshold. Third, a new approach for calculating turbulent fluxes is implemented within the MYNN framework, which accounts for the effects of anisotropy and turbulent potential energy (TPE).

All three modifications are evaluated using a set of 15 wind ramp cases as identified in tall tower data from Iowa in the U.S. and near Hamburg, Germany. The WRF model is used to generate 24-hour wind forecasts, which are evaluated relative to observations at 100m height. It is found that invoking the new set of CPs provides marked forecast improvement only when used in conjunction with the new mixing length formulation and only for cases that are originally under- or over-forecast. For these cases the MAE of wind forecasts at 100m on average is reduced by 17%. Reduction in average MAE by 26% is realized for these same cases when invoking the method that accounts for anisotropy and TPE along with the new mixing length. This last method results in an average reduction in MAE of 13% across all 15 cases.