The Power of Researcher-Forecaster Collaboration: How WoFS-PHI Evolved Based on Feedback from Multiple Hazardous Weather Testbed Experiments

Warn-on-Forecast System-Probabilistic Hazard Information (WoFS-PHI) is a machine-learning-based product that predicts the probability of severe weather hazards based on a blend of observational and numerical weather prediction datasets. The product has been under development since 2022 and has evolved substantially since then based on forecaster feedback from four Hazardous Weather Testbed (HWT) experiments across 2023 and 2024.

The initial version of WoFS-PHI considered predictors from ProbSevere Version 2 (PS2) and WoFS and was visualized on the WoFS Viewer. It predicted in 30-minute windows; did not run in real-time; and used diffuse, hazard-specific color bars. After initial evaluation in the 2023 Spring Forecasting Experiment (SFE), WoFS-PHI was incorporated into the Prototype PHI Tool for forecaster interrogation. New color bars were created, and the product updated every 5 minutes using the most recently available WoFS and PS2 data. During the 2023 Watch-to-Warning Experiment (W2WE), which brought together Storm Prediction Center and local office forecasters to test the creation of probabilistic products and communication between watches and warnings, forecasters found WoFS-PHI useful for situational awareness. However, they felt probabilities were too low and had difficulty understanding and using the probabilities. To help contextualize the product, the next iteration of WoFS-PHI was trained on warnings and storm reports instead of just reports, mapped raw probabilities to success ratios, incorporated new color bars, and used 1-hour prediction windows. This new version was tested during a controlled real-time severe weather forecasting activity during the 2024 SFE, in which forecasters made 0-1- and 1-2-h probability forecasts with and without WoFS-PHI. Feedback on the updates was generally positive, although forecasters still felt probabilities were too broad, and some expressed a desire for 2-h time windows. As a result, after the SFE, WoFS-PHI added ProbSevere Version 3 and Tornado Probability Algorithm predictors and included options for 15- and 39-km radii and 1- and 2-h time windows. This newest version is currently being tested in the 2024 W2WE to determine if it helps forecasters issue probabilistic watch-to-warning products for the public and emergency managers. Early feedback from the 2024 W2WE suggests that WoFS-PHI can be further improved with more spatiotemporal specificity, contextualization, and documentation of past performance. The iterative improvements of WoFS-PHI reflect the power of forecaster-researcher collaboration through the HWT.