NWC REU 2021
May 24 - July 30



Photo of author

Comparing the Forecasting Performance of the UFS and WRF Models during High-Impact Severe Weather Events

Juan P. Mangual-Pagán, Larissa J. Reames, and David J. Bodine


What is already known:

  • For numerical weather prediction models, forecasting hail is a challenge due to the many microphysical and dynamical processes that influence its formation.
  • The Advanced Research WRF has been one of the main tools for researching severe weather in the past decade
  • The 2018 Spring Forecast Experiment analyzed the skill of FV3-based models to that of the HRRRv3, but other than this, the FV3 core is relatively untested at a convective-allowing scale.
  • The FV3 was chosen to be the core for the Unified Forecast System, which aims to replace all operational models under a unified framework. Therefore it is important to evaluate the FV3 model performance during high impact weather.

What this study adds:

  • This study assesses the performance of both the UFS-FV3 and the WRF-ARW models on April 28, 2021 hailstorms through inter-comparisons between models and verification of the model data with radar data and surface observations.
  • Both models were in agreement with the initiation and timing of the San Antonio hailstorm, but less so when it came to the placement of the storm at its peak.
  • Both models struggled with the placement of the Norman storm with respect to a cold front, therefore no hail was forecasted for/near Norman.
  • Documented the irregular and abnormally small values given by the UFS Hailcast in both locations, although further research is needed to explain these values.


The numerical weather prediction community is seeking a common dynamical core to improve forecasts across a range of space and time scales, which motivated the development of the Unified Forecast System (UFS). The community-driven UFS uses the Finite Volume Cubed-Sphere (FV3) as its core. The FV3-based models are relatively untested at a convective-allowing scale. This study compares the performance of the UFS-FV3 and the WRF-ARW and verifies their respective simulations with radar data and verification metrics, such as initiation, timing, movement, and placement of convection. The case being examined is 28 April 2021 where large hail impacted the metropolitan areas of Norman, OK and San Antonio, TX. Both models had similar initiation, timing, and movement in both locations. Both models struggled to predict the hail event in Norman, OK due to the relative positioning of the storm and a cold front. While in San Antonio, the WRF predicted a strong storm with better placement, in regards to the radar, than the UFS. The WRF hailcast also predicted the severe hail threat to the west of San Antonio. The UFS hailcast, on the other hand, predicted abnormally small hail to the east and southeast of San Antonio, where no hail was reported. These findings highlight the need to further test the UFS hailcast scheme with the FV3-core.

Full Paper [PDF]