11.6 Using Machine Learning Techniques to Predict Near-Term Severe Weather Trends

Thursday, 11 January 2018: 2:45 PM
Ballroom F (ACC) (Austin, Texas)
David Harrison, CIMMS/Univ. of Oklahoma/NOAA/SPC, Norman, OK; and C. Karstens and A. McGovern

The prototype probabilistic hazards information (PHI) warning system allows National Weather Service (NWS) forecasters to issue dynamically evolving severe weather warning and advisory products, which can provide end users with specific probabilities that a given location will see severe weather over a predicted time period. When issuing these products, forecasters are given a storm’s diagnostic probability of producing severe weather via the National Oceanic and Atmospheric Administration (NOAA) / Cooperative Institute for Meteorological Satellite Studies (CIMSS) ProbSevere model, and are then asked to forecast these probabilities along a trend graph in the warning creation software. However, feedback from several years of experimentation suggests that forecasters are unfamiliar with how these probabilities tend to evolve in different convective situations. To help address this issue, a machine learning model was developed to predict how the probability that a storm will be severe will prognostically change with time. This model uses the ensemble average of six members, consisting of ADA boosting regression, gradient boosting regression, and elastic net models trained on variables obtained from the initial ProbSevere model, environmental parameters, and a storm’s history, to predict future ProbSevere probabilities over a predicted duration of a storm. Preliminary verification suggests that the model has skill compared to a simple linear decay, and an early implementation into the PHI warning software was well received by forecasters during the 2017 Spring Experiment held at the NOAA Hazardous Weather Testbed (HWT).
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner