Producing new extended-range probabilistic tornado forecasts through statistical post-processing using a reforecast dataset

Improved lead-time for probabilistic tornado prediction is important for the welfare of the general public as well as hydrological, agricultural, energy and emergency managers. While statistical methods have been used in order to forecast severe weather events with a short lead-time, little research has been done in the way of using those methods to predict tornadic activity on an extended range time scale (e.g. Days 3-14), leaving many issues unexplored.

Recently, the skill in forecasts for extreme events, such as heavy precipitation, has been improved through statistical post-processing calibrated with a set of retrospective forecasts using a static ensemble forecast system. Calibration of statistical post-processing methods with reforecast datasets combine the benefits of statistical correction techniques with the advantages that come from the use of a very large sample set. Previous research of forecasting rare events involving a reforecast dataset used a wide variety of statistical techniques, producing forecasts superior than raw ensemble forecasts. However, it is not clear if these methods will produce skillful, reliable extended range forecast of severe weather.

The objective of this research is to determine the usefulness of a reforecast dataset in the statistical calibration of extended range severe weather forecasts. A new, 25+ -year reforecast dataset based on a recently updated version of NCEP's Global Ensemble Forecast System is used to calibrate tested statistical methods with SPC's tornado reports database used for verification. Forecast skill is calculated and reliability diagrams are created in order to determine the usefulness of the forecasts. Preliminary results suggest that skillful tornado forecasts at leads of 1 week and longer may be possible. The talk will discuss both the methodology used for post-processing and show some preliminary results, including statistics over many cases and some particular case study data.