Comparing the Impact of Ensemble Design in the Rapid Refresh Forecast System Using Object-Based Methods

As NOAA transitions to the Unified Forecast System, operational regional forecast systems will be replaced with the Finite Volume on a Cubed Sphere (FV3) based convection-allowing Rapid Refresh Forecast System (RRFS). The current convection-allowing ensemble model, the High Resolution Ensemble Forecast system (HREFv3), is a multi-dynamical core (dycore) and multi-physics ensemble that employs time lagging. It contains sufficient spread, but the members tend to cluster by dycore and physics pairings which produces multi-modal statistics. The RRFS however, is planned to be a single-dycore and single-physics ensemble. Often, these single-dycore, single-physics ensembles are under dispersive, or lack sufficient spread. This project aims to optimize ensemble design of the RRFS by examining the use of time-lagging, neighborhood methods, initial condition perturbations, and stochastic physics.

Previous work has been done comparing these different ensemble design configurations using both traditional verification metrics such as reliability, spread/skill, and bias, and qualitative evaluations of storm mode, structure, and convective evolution. Here we extend the traditional verification metrics and qualitative evaluation to include object-based verification methods. Specifically, we will evaluate the impact of initial condition perturbations and stochastic physics on model performance using discontinuous fields. These object-based methods can provide additional information such as identifying differences in storm size, spatial displacement, and orientation errors. In this study, RRFS ensemble verification is performed for the period between 30 April and 12 May 2022 using the Method for Object Based Diagnostic Evaluation (MODE), which is part of the Model Evaluation Tools Enhanced Framework (METplus).