Scientists at CIMSS and NOAA/NSSL have been working on what we are calling ProbSevere v4. The goals are the same as previous version of ProbSevere: short-term prediction of severe convective hazards (large hail, damaging wind gusts, and tornadoes). The approach is very different. ProbSevere v4 does not use storm-object identification or tracking. Instead, it uses a more holistic approach, and uses images of meteorological data to make predictions at every output point or pixel. The output has a resolution of 4 km. The output is the probability of any severe hazard within 20 km and within the next 45 minutes. These spatiotemporal parameters may change in the future.

ProbSevere v4 is part of a greater effort to create accurate, calibrated guidance for convective hazards from 0-6 hours, using the best data available (e.g., use Warn-on-Forecast data or GOES-R mesoscale sectors when available).

We have found that the model has very good spatial skill — that is, maximizing Critical Success Index on a pixel-by-pixel basis. See this conference paper for an overview of the evaluation.

Here are few recent examples of model predictions.