Stochastically-Perturbed Parameterizations (SPP)

Description

For both regional and global ensembles, sufficient initial condition and physics uncertainty representation is critically important to ensemble performance. Otherwise, members may tend toward similar solutions, resulting in forecasts not capturing the full envelope of possible outcomes. To improve the effectiveness of the ensemble, stochastic physics can be employed. Stochastically-Perturbed Parameterizations (SPP) provides one method through which unknown or uncertain physical processes can be represented in ensemble forecasting. As it is currently employed in CCPP (Beck et al.(2022) [14]), SPP can be applied to the following physics parameterizations: MYNN-EDMF Boundary Layer and Shallow Cloud Scheme, MYNN Surface Layer Scheme, GFS RRTMG Shortwave/Longwave Radiation Scheme, GFS Unified UGWP Scheme,GSL Drag Suite Scheme and Thompson Aerosol-Aware Cloud Microphysics Scheme (note that both RUC Land Surface Model and GFS Noah Land Surface Model can also be run with stochastic parameter perturbations). A list of the parameters perturbed within each scheme can be found below:

• MYNN-EDMF Boundary Layer and Shallow Cloud Scheme
  • Eddy diffusivity
  • Viscosity
  • Background water vapor
  • Lateral entrainment rate
• MYNN Surface Layer Scheme
  • Thermal roughness length
  • Moisture roughness length
  • Aerodynamic roughness length
• GFS Unified UGWP Scheme and GSL Drag Suite Scheme
  • Subgrid-scale terrain variations
  • Froude number
  • Wind speed tendencies
• GFS RRTMG Shortwave/Longwave Radiation Scheme
  • Effective cloud water, snow, and ice radii
• Thompson Aerosol-Aware Cloud Microphysics Scheme
  • Graupel intercept parameter
  • Cloud droplet shape parameter
  • Vertical velocity used in CCN activation
  • Ice number concentration

Magnitudes, time decorrelation lengths, and other namelist settings can be modified to control the perturbations for each scheme; however, for now, all parameters are perturbed identically within each scheme. For detailed information on how to activate and modify SPP settings for these schemes, the reader is referred to the SRW App Users Guide.

Each parameter listed above was chosen through consultation with physics experts, often the author of the parameterization, in order to target fields that could benefit from the application of SPP. However, there are many other parameters which could be effectively perturbed to improve uncertainty representation and ensemble spread generation. In most of the perturbed parameterizations, the perturbation magnitudes are multiplied by coefficients to generate reasonable parameter values; however, users should be careful when making changes to the magnitudes provided as a guideline in the SRW App documentation, since it is still possible to generate unphysical parameter values. Default magnitudes and decorrelation lengths in the SRW App documentation were the result of extensive testing with 3-km CONUS simulations. If the user would like to apply SPP at other resolutions, it is recommended that retrospective simulations be conducted over the area of interest to ensure reasonable application of the perturbations.

GFS Surface Parameter Perturbation

Land surface perturbation (Gehne et al. (2019) [69]) has been recently introduced into UFS. This treatment is based on the hypothesis that one of the major causes of the insufficient spread in current global NWP model,especially near the surface, is a lack of treatment of uncertainty in the soil state and in the associated model parameters. It allows for land surface parameters such as surface albedo, vegetation fraction, soil hydraulic conductivity, leaf area index (LAI), surface roughness lengths for heat and momentom to vary in space. These parameters and variables have been shown to impact forecasts of 2m temperature, 10m wind and precipitation. Based on the parameter or variable,different strategies to perturb are necessary.