The Development of an Experimental Warn-on-Forecast System Using the MPAS dynamic Core and the DART system

The Development of an Experimental Warn-on-Forecast System Using the MPAS Dynamical Core and the DART System

Yunheng Wang^1,2, Larissa Reames^1,2, Thomas Jones^1,2, Nusrat Yussouf ^1,2, Lou Wicker¹

¹NOAA/National Severe Storm Laboratory, Norman, OK 73072, and

²Cooperative Institute for Severe and High-Impact Weather Research and Operations (CIWRO),

University of Oklahoma, Norman, OK 73072

The experimental Warn-on-Forecast system (WoFS) aims to extend warning lead times for severe weather hazards using on-demand, adaptable domain and rapidly updating high-resolution ensemble analysis and forecasts. The current prototype of WoFS uses the Advanced Research WRF (ARW) configuration of the Weather Research and Forecasting model as dynamic core and the EnKF implementation within the Gridpoint Statistical Interpolation (GSI-EnKF) system for high-frequency ensemble data assimilation (DA) cycles. To fit within the framework of NOAA’s Unified Forecast System (UFS) initiative, in collaboration with NCAR, NSSL explored the Model Prediction Across Scales (MPAS) model as a dynamic core and the Data Assimilation Research testbed (DART EAKF) as the ensemble DA system for its next-generation WoFS. An experimental MPAS-based WoFS is developed at the National Severe Storms Laboratory. Besides using the regional settings of the MPAS model, the MPAS-based WoFS also uses DART EAKF software to perform high frequency data analysis cycles for ingesting MRMS reflectivity, WSR-88D radial velocity, satellite observations (CWP), Mesonet and other conventional observations. The high frequency data assimilation (DA) cycles are conducted every 15 minutes staring from 1500 UTC on each event date until 0300 UTC the next day. Then, short-term (6-hour) forecasts initialized from the corresponding DA analysis are launched hourly staring from 1700 UTC on the event day to 0300 UTC the next day. The storm-scale MPAS ensemble is designed at approximately 3-km horizontal cell spacing with multi-physical configurations that introduces diversity in the planetary boundary layer schemes (YSU scheme, Mellor-Yamada-Janjic TKE scheme and MYNN 2.5 level TKE scheme) and the surface layer schemes (Monin-Obukhov scheme, revised MM5 Monin-Obukhov scheme and MYNN surface layer). The NSSL double-moment microphysics scheme and the RUC land surface model are used for all ensemble members, and the “convection permitting” physics suite from the MPAS package is used for other physical options (radiation, convection, gravity wave drag etc.).

To develop the workflow for this MPAS-based WoFS, several workflow engines are explored, including cylc, Rocoto, AirFlow etc. Eventually, we found that those workflow engines just migrate the code complexity to a low-level configuration step. For simplicity and compatibility, the Bash shell language is chosen for the workflow development with the following concerns: 1) flexibility, the system should be initialized from a set of external models, use various physical options for each member and work with on-demand moving domains; 2) portability, the system is configurable with a set of job schedulers (SLRUM or PBS) and can be run interactively for retrospective studies and operationally via crontab for real-time applications. 3). Fault tolerance, the system can resume at any check/break point for any member without wasting of resources for duplicated tasks. Furthermore, the system separates the configurations of machine-dependent components from the analysis and forecast configurations with modular design. So that each part of the workflow can be customized and tuned individually following the scientists’ needs and the engineering requirements. To test the performance of this MPAS-based WoFS, several high-impact convective-storm events in 2023 will be simulated. The purpose is to share the capability, the design principle, the challenges and the remaining issues of this MPAS-based WoFS and some of its preliminary results.