NWC REU 2014
May 21 - July 30



Photo of author

Sensitivity of Simulated Supercell Thunderstorms to Horizontal Grid Resolution

Montgomery Flora and Corey Potvin


What is already known:

  • Supercell thunderstorms produce the majority of significant severe weather (1”+ diameter hail and EF-2+ tornadoes) in the United States and are difficult to forecast numerically.
  • Sensitivity of supercell development, particularly operationally significant impacts, to grid spacing has not been thoroughly explored.
  • A major challenge confronting Warn-on-Forecast is the tradeoff between increased model resolution and increased computational cost.
  • How well do horizontal grid spacings of 1, 2, 3 and 4 km capture operationally significant features of supercells?

What this study adds:

  • A 4 km grid spacing appears too coarse to well represent supercell dynamics and evolution.
  • Grid spacing sensitivity varies strongly with storm environment.
  • In general, more benefit gained moving from 2 km to 1 km spacing than from 3 km to 2 km.


The effects of horizontal grid spacing on idealized supercell simulations are investigated. Motivation for the study largely stems from the NOAA Warn-on-Forecast program, which envisions a paradigm shift from “warn-on-detection”, where convective hazard warning decisions are primarily based on current observations, to a new paradigm where storm-scale numerical weather models play a greater role in generating forecasts and warnings. Unlike most previous grid spacing sensitivity studies, we focus on impacts to operationally significant features of supercells. Using the WRF-ARW model, idealized supercell simulations are run for 2 hours using three different environments and horizontal grid spacings of 333 m and 1, 2, 3, and 4 km. Given that forecasts under the Warn-on-Forecast paradigm will be initialized after several radar data assimilation cycles, we initialize our coarser simulations with filtered versions of the 333m “truth” simulation valid at t = 30 min. To isolate differences in storm evolution arising from finer-scale processes being unrepresented in the coarser simulations, the latter are compared to appropriately filtered versions of the truth simulations.


Results show that operationally significant errors in supercell evolution arise as the grid spacing is increased. Furthermore, the grid spacing sensitivity is strongly influenced by the environment. The 4 km grid spacing is too coarse to even qualitatively reproduce the supercell evolution, with the storm dying before the end of the simulation in one of the three environments. The improvement as grid spacing decreases from 2 to 1 km is greater than that from 3 to 2 km. Implications of this and other findings for Warn-on-Forecast are discussed.

Full Paper [PDF]