Severe thunderstorms and the damaging weather they produce (including large hail, strong straight-line winds and tornadoes) are by nature small scale phenomena occupying narrow segments of space and time. As such, local extreme events such as tornadoes are not resolved even by advanced mesoscale numerical models. Recent evidence nonetheless suggests that clusters of such severe weather events (termed "outbreaks") may be anticipated on a regional basis with lead times of 2-3 days or more, although the ability of models to discriminate tornadic outbreaks from primarily non-tornadic ones is apparently diminished during the spring season when damaging storms are most frequent. This study will investigate the limits of predictability (viz. maximum lead times) over which contrasting outbreak types may be reliably forecast by the MM5 and WRF (Weather Research & Forecasting) community models when initialized using only spatially coarse synoptic-scale information on atmospheric structure. Derived meteorological covariates (summary parameters used to evaluate the potential for contrasting outbreak types) will be analyzed through a combination of subjective and objective statistical techniques. This analysis will cover the period 1970-2003. Initial results suggest that dynamical measures including shear-related quantities such as storm-relative helicity are favored over thermodynamically-based measures such as Convective Available Potential Energy (CAPE) in these long-range forecasts. This invites further exploration of why certain pathways may be preferred for communicating large-scale atmospheric influences down to more the scale of individual thunderstorms. Evidence of seasonal modulation of ability to discern tornadic vs. non-tornadic outbreaks will be further explored, as will forecast skill based both upon single model runs and more computationally demanding ensemble runs. The utility of model-resolved storm properties such as updraft strength in discriminating outbreak type will also be evaluated.
The intellectual merit of this work rests on improved ability to run and interpret mesoscale forecast models to reliably anticipate major outbreaks of severe weather, to discern the nature of these outbreaks with lead times of several days, and to gain improved understanding of atmospheric processes connecting large and small scales. Broader impacts of this work will include benefits from increased lead time for the public to prepare for hazardous weather events, through immediate communication of these advances through cooperation and collaboration with operational forecasters at NOAA's Storm Prediction Center and Australian Bureau of Meteorology, as well as more traditional links and graduate student education.
A team of scientists from The University of Oklahoma, Mississippi State University and The University of South Alabama has developed a weather-prediction system that will allow forecasters to assess the likelihood of tornadic and other severe storm outbreaks occurring up to three days before the storms form. Tornadoes and other severe weather outbreaks annually cause loss of life and massive property damage to many portions of the U.S. Scientists have long sought to extend the prediction skill and warning lead time of tornado and severe weather outbreaks. Prior to this study, it was generally accepted that a purely objective forecast system could not skillfully discriminate storm type, and forecasters have emphasized short-term prediction of individual storms. The research team, combining scientists from the aforementioned universities, based their system on new knowledge about the relationship between the large-scale state of the atmosphere and the processes that initiate storm outbreaks. Examining the large scale allows them to extract more signals in the atmosphere that are associated with outbreaks. Key findings of intellectual merit include: 1. Forecast degradation was pronounced with forecasts of outbreaks occurring in the spring and fall, with severely reduced capability (but still showing skill) of distinguishing outbreak type three days in advance compared to one day in advance. 2. Large scale parameters, such as mean sea level pressure, low-level geopotential heights, and low-level wind speeds/directions, were commonly as effective as low-level shear variables in distinguishing outbreak type. 3. Three basic categories of severe weather outbreaks (in terms of outbreak severity) were apparent: high-end severe weather outbreaks, intermediate outbreaks, and outbreak days with large geographical scatter or relatively few reports. 4. As reanalysis data set used in this project may not measure the boundary layer sufficiently to provide accurate approximations of conditions that inhibit formation of these storms. This increases the number of false alarms. 5. There is relatively good performance of high-risk outlooks with large scale tornadic outbreaks. 6. Non-tornadic outbreaks in the spring and fall have vastly different synoptic structures than those occurring in the summer. No such difference was observed with the tornado outbreaks due to the lack of tornado outbreaks occurring in the summer. 7. There is considerable deterioration in the skillful discrimination of outbreak types from 1-day to 3-day forecasts when the PNTOs selected are constrained to the spring and fall seasons, which differs from the relatively consistent skillful discrimination of summer PNTOs from TOs. 8. Although substantial skill is observed, a noticeable false alarm problem is present. There are a large number of minor outbreaks that are improperly diagnosed as major outbreaks, owing to similar synoptic- and subsynoptic-scale environments. 9. The meteorological variables (known as covariates) most helpful in distinguishing major and minor outbreaks include indices derived from a combination of buoyant instability, deep-layer shear, and low-level helicity (e.g., the supercell composite parameter, the significant tornado parameter, and the energy-helicity index). However, there are no statistically significant differences among these parameters. 10. A probabilistic framework is being extended to numerical model simulations of a subset of the outbreaks to compare the model values to the analyzed probabilities. Extensions to ensemble simulations for model runs of the outbreaks are currently being tested. These research findings can help weather prediction centers improve computer modeling of storms and increase warning lead times, potentially mitigating such devastating effects. The findings have a broad impact for severe-weather forecasting across the eastern two-thirds of the United States. Forecasters at the Storm Prediction Center in Norman, Oklahoma, which is in the region of U.S. known as "Tornado Alley" can use these findings to increase the accuracy of their severe weather outlooks.
