A Seasonal Probabilistic Outlook for Tornadoes (SPOTter) in the Contiguous United States Based on the Leading Patterns of Large-Scale Atmospheric Anomalies

Abstract

AbstractThis study presents an experimental model for Seasonal Probabilistic Outlook for Tornadoes (SPOTter) in the contiguous United States for March, April, and May and evaluates its forecast skill. This forecast model uses the leading empirical orthogonal function modes of regional variability in tornadic environmental parameters (i.e., low-level vertical wind shear and convective available potential energy), derived from the NCEP Coupled Forecast System, version 2, as the primary predictors. A multiple linear regression is applied to the predicted modes of tornadic environmental parameters to estimate U.S. tornado activity, which is presented as the probability for above-, near-, and below-normal categories. The initial forecast is carried out in late February for March–April U.S. tornado activity and then is updated in late March for April–May activity. A series of reforecast skill tests, including the jackknife cross-validation test, shows that the probabilistic reforecast is overall skillful for predicting the above- and below-normal area-averaged activity in the contiguous United States for the target months of both March–April and April–May. The forecast model also successfully reforecasts the 2011 super-tornado-outbreak season and the other three most active U.S. tornado seasons in 1982, 1991, and 2008, and thus it may be suitable for an operational use for predicting future active and inactive U.S. tornado seasons. However, additional tests show that the regional reforecast is skillful for March–April activity only in the Ohio Valley and South and for April–May activity only in the Southeast and Upper Midwest. These and other limitations of the current model, along with the future advances needed to improve the U.S. regional-scale tornado forecast, are discussed.