What happened
NOAA and the Cooperative Institute for Meteorological Satellite Studies in Wisconsin developed ProbSevere. Its second version uses naive Bayesian classifiers to estimate the probability that a storm will produce severe hail, wind or a tornado. In a technology assessment dated 14 December 2023, the U.S. Government Accountability Office (GAO) states that the model has improved forecasters' ability to warn of severe thunderstorms and tornadoes. GAO writes that forecaster feedback suggests that warnings come about 5 to 10 minutes earlier. It reports that forecasters who evaluated the model said that it increased their confidence in issuing warnings 95 percent of the time and lead time 76 percent of the time. GAO did not test the model or analyze survey data. It relays the developers' 2020 paper and gives no source for the two percentages. In that paper, 68 forecasters in NOAA's Hazardous Weather Testbed answered daily surveys from 2017 to 2019. In about 80% to 90% of their answers, forecasters said that ProbSevere increased confidence. GAO and the developers write that more research is needed to quantify the effect on lead time.
Sources
-
Peer-reviewed study, by the makers.
Cintineo, J. L., M. J. Pavolonis, J. M. Sieglaff, L. Cronce and J. Brunner. NOAA ProbSevere v2.0—ProbHail, ProbWind, and ProbTor. Weather and Forecasting 35(4): 1523–1543, 2020.
From the forecaster feedback, ~80%–90% of forecaster responses said that ProbSevere increases confidence in their warning decision making, while ~60%–70% indicated that ProbSevere increases lead time to severe hazards.
-
Descriptive report, by authors with no tie to the system.
U.S. Government Accountability Office. Artificial Intelligence in Natural Hazard Modeling: Severe Storms, Hurricanes, Floods, and Wildfires. Technology assessment GAO-24-106213, 14 December 2023, p. 24.
One exception is ProbSevere (version 2), which is a machine learning model that has improved forecasters' ability to issue warnings of severe thunderstorms and tornadoes.
Details
AI system
- Reported configuration
-
Version 2
- Workflow
-
ProbSevere combined radar, satellite, lightning and weather model data for each storm that it tracked. Naive Bayesian classifiers trained to each hazard gave the probability of severe hail, wind and tornadoes. National Weather Service forecasters used ProbSevere as an additional tool when predicting severe weather.
Follow-up
- Conditions for replication
-
Forecasters must use the model alongside radar and other information, since it is one source among several. Probability thresholds must suit the region, since the best threshold varies across the United States.
Entry details
- Sector
-
Weather forecasting / severe storm warnings.
- Purpose
-
NOAA and the Cooperative Institute for Meteorological Satellite Studies built ProbSevere to estimate the probability that a developing storm will become severe in about the next hour.
30 more questions not yet checked
model, provider, role, configuration, inputs and outputs, autonomy, human oversight, AI contribution, beneficiaries, reach, distribution and access, duration, every measurement question, every money question, ownership and control, dependencies, cross-border connections, geopolitical implications, replication attempts, availability, institutional decision, grounds, action and result, success criterion, every evidence question, observed or intended, alternative assessments, related entries, other identifiers, category mappings.
Cite this record
Peter H. Vartanian. (2026). ProbSevere helps US forecasters warn of severe storms, GAO reports (AIBD-1127). Artificial Intelligence Benefits Database, edition of 1 November 2026. https://petervartanian.xyz/aibd/records/AIBD-1127.html (archived at https://doi.org/10.7910/DVN/AADSII, version 7.0)