Use of Stochastic Physics Approach in North American Rapid Refresh Ensemble (NARRE)

The main focus of the pre-NARRE effort is to work toward the development and implementation of the hourly updated North American Rapid Refresh Ensemble (NARRE) forecasting system. This new system is planned for implementation into NCEP operations as an extension of the Short Range Ensemble Forecasting (SREF) system in 2017. This new system will provide improved probabilistic forecasts for aviation and other short-range applications. The work is conducted collaboratively by the Global Systems Division, Environmental Modeling Center, and Developmental Testbed Center staff. Current plans for NARRE address model-related uncertainty by using a combination of multi-dycore (ARW and NMMB) and multi-physics, where the physics suites are those used in current operational systems (RAP and NAM) and variations of these well-tested suites. An ensemble system based on multiple physics suites comes at the price of needing to maintain multiple physics packages. To minimize ensemble system maintenance, the ultimate goal is to move towards representing model uncertainty through a stochastic physics approach. Before moving forward with such a transition, the value added by applying a stochastic physics approach must be demonstrated. For this purpose we plan to compare impacts on the ensemble performance produced by use of multi-physics and stochastic physics approach. The focus will be on the WRF-ARW core. The latest version of the ARW core includes an option for the generation of a random perturbation field that can be used for parameters in physics parameterizations. The stochastic pattern is centered on zero and has a user-prescribed standard deviation. With this option, the user can specify random perturbation length-scale, temporal de-correlation of the randomly perturbed field and vertical structure of the random perturbations. Generated patterns will be used to perturb parameters. We plan to apply stochastic approach in the Grell-Freitas convective scheme, RUC Land Surface Model scheme and MYNN PBL scheme. The choice of what physics parameterization is stochastically perturbed will determine the design of the mixed-physics ensemble that would be used as a baseline. For example, if an ensemble is created by stochastically perturbing a convective scheme, this ensemble will be compared to an ensemble that includes different convective schemes. Preliminary results will be presented at the conference.