10.4 IBM GRAF - Scale-aware convective forecast evaluation and improvements.
Wilt, Brett, The Weather Company, an IBM Business and Wei Wang, National Center for Atmospheric Research
In August 2018,
The Weather Company (TWC), an IBM Business, implemented a custom version of the
NCAR Model for Prediction Across Scales (MPAS). The initial model configuration
utilized MPAS 5.2 Kain-Fritsch convection physics, operating on a uniform 15-km
resolution mesh with 72 forecast hours (experimentally 144 hrs for research
purposes).
With the recently announced IBM GRAF (Global High-Resolution Atmospheric
Forecasting System) development goals, the hourly-updating variable resolution
(15/3-km) MPAS mesh introduces the requirement for scale-aware convection
physics. Upon review of several convective parameterizations, a Tiedtke-based
approach was selected to compliment numerous convective aspects of the GRAF
application, including data-assimilation, statistics, visual representation,
tropical, pattern recognition, etc.
An overview of GRAF-Tiedtke scale-aware progress will be discussed, along with
supporting results and comparisons from a 50 case reforecast lab environment.
As verification examples of the lab environment, a baseline of 2-meter
temperature, 10-meter wind RMSE and bias statistics are computed for
approximately 3000 METAR sites across the continental United States (CONUS).
General precipitation metrics, such as Equitable Threat Score (ETS), Fractions
Skill Score (FSS), and Critical Success Index (CSI) will also be presented. In
addition to the lab statistics, applying basic scale-aware logic to Tiedtke
convection schemes provides valuable information for other critical scale-aware
aspects, including tropical convection across mesh boundaries, tropical cyclone
tracks, and 5-6 day synoptic scale pattern comparisons.