A Comparison of Multi-Dynamic Core vs Single-Dynamic Core Multi-Physics Ensemble Design for Convection-Allowing Forecasting Initialized By the Multiscale GSI-Based EnVar Data Assimilation System

The Gridpoint Statistical Interpolation (GSI)-based hybrid ensemble-variational (EnVar) data assimilation (DA) scheme has been extended for convective scales including the direct assimilation of radar radial velocity and reflectivity by Johnson et al (2015) and Wang and Wang (2017). This system has been tested within the operational NAMRR and HRRR model contexts for the 2017 and 2018 NOAA Hazardous Weather Testbed (HWT) spring forecasting experiments by the Multiscale data Assimilation and Predictability (MAP) lab at the University of Oklahoma. Ten retrospective cases from 2015 and 2016 are used to facilitate development and testing of potential improvements to the convection allowing assimilation and ensemble forecast system. A single 3-km horizontal resolution domain covering the CONUS is used throughout, with hourly assimilation of conventional observations as well as 20-min sub-hourly assimilation of radar DA for the final hour of cycling.

In this study, the ten retrospective cases are used to explore the optimal design of convection allowing forecasts. The performance of an ensemble forecasting system depends upon the ability of the ensemble to represent all sources of uncertainty, including initial condition errors and model error. The ensemble of analyses produced by the hybrid DA system can automatically sample the errors of the initial conditions. How to optimally sample the errors in the model still remains to be addressed. In this study, we look to address the following questions: how is a multi-core ensemble compared to a single core multi-physics or stochastic physics ensemble?

One approach to address model error is by combining forecast members from two or more model dynamical cores into a multi-model ensemble. NAMRR and HRRR models’ own GSI hybrid DA system automatically generate their respective single-core single physics ensemble. Here, we randomly combine 5 members each from the NAMRR and HRRR models’ single-core single-physics ensembles to comprise a multi-model ensemble. Another approach is to employ a multi-physics ensemble setup, which can be done within a single dynamic core, e.g. the WRF-ARW core model that the HRRR uses. Finally, recent studies have employed stochastic perturbations to address model uncertainty associated with sub-grid scale processes. The stochastic kinetic energy backscatter scheme (SKEBS; Berner et al. 2011, Duda et al. 2017) was developed to represent the dynamic net forcing of unresolved flow cascading energy into resolved scales. The SKEBS scheme is implemented here using optimized parameters from Duda et al. (2017) for convective scale prediction; it is tested with the HRRR single-physics and multi-physics configurations. These five ensemble configurations are compared subjectively and quantitatively within the 18-h free forecast period. Verification metrics include RMSE-spread plots as well as contingency table-based scores such as Fractions Skill Score on both 1-h accumulated precipitation and composite reflectivity fields at various thresholds, compared to observations from the Multi-Radar Multi-Sensor (MRMS) system. Initial results have shown the HRRR configuration to be more skillful than the NAMRR configuration, particularly at lower thresholds of precipitation and composite reflectivity. The multi-model ensemble helps to subjectively fill in gaps that a single-model single-physics configuration misses; initial skill scores of multi-model are comparable to the optimal HRRR configuration. The SKEB configuration is able to increase the spread of the ensemble forecast and in some cases produce a better forecast than its respective single-physics single-core ensemble without SKEB. Further detailed results will be presented at the conference.