About the forecasts

The probabilistic convective hazard forecasts are generated for operational HRRR forecasts using a dense neural network (NN) trained with ~500 00 UTC HRRR forecasts between 15 July 2018 and 31 December 2019. A smoothed mid-level updraft helicity (UH)-based HRRR forecast is also available.

The probabilities produced using the NN or UH are defined for an event occurring within 2 h of a forecast hour and 40 km or 120 km of a grid point. The web interface provides options for visualizing the 4-h hazard probabilities or aggregating the 4-h hazard probabilities across forecast hours to provide a summary of where severe weather is expected on a given day (e.g., “Day 1 Max Prob”).

Other visualization options include showing the hazard that produces the largest probabilities at a given grid point (select “Max Prob Hazard”) and the forecast hour when the largest probabilities occur (select “Max Prob Timing”).

Any feedback on the forecasts, the configuration choices, or the visualization interface can be directed to sobash (at) ucar.edu. Thanks for visiting!