An Evaluation of Stochastic Physics Within The High Resolution Rapid Refresh (HRRR) Ensemble and The Impacts of High Performance Computing (HPC)

Abstract

Lack of spread is a common deficiency in ensemble models and results in over-confident and inaccurate forecasts. A number of options exist to increase the spread of an ensemble, including the use of multiple physics parameterizations or dynamic cores. However, implementation of these options can be cumbersome, costly, and suffer from certain theoretical deficiencies related to ensemble model climatology. In an attempt to generate a more unified and streamlined system, stochastic physics can be used within a single dynamic core and physics suite to represent model-related uncertainty. To this end, an effort is underway to implement and test the following three different stochastic physics schemes within the High-Resolution Rapid Refresh (HRRR) ensemble: Stochastic Parameter Perturbations (SPP), Stochastic Kinetic Energy Backscatter (SKEB), and Stochastic Perturbation of Physics Tendencies (SPPT). A combination of these methods is also being tested. Ensemble members are generated through the application of stochastic methods to a single physics suite. For SPP, hydraulic conductivity was perturbed in the Rapid Update Cycle (RUC) land surface model (LSM) as well as a number of mass flux parameter perturbations in the Mellor-Yamada-Nakanishi-Niino (MYNN) Planetary Boundary Layer (PBL) scheme. Verification and analysis will be conducted for the four-week period over which these tests were run. Given the computing resources necessary to run a high-resolution (3-km) HRRR ensemble over the CONUS with hourly cycling, a large allocation was requested and provided through the National Center for Atmospheric Research (NCAR) Strategic Capability (NSC) project. These resources provided enough computing power for an extended period of testing over multiple weeks, which is necessary to provide statistical significance. With the help of high performance computing, the findings from this study will provide further insights into the use of stochastic physics within a convection-allowing ensemble.