By: Dorian J. Burnette
Note: The following discussion becomes "meteorologically intense" in places. Basic knowledge of synoptic meteorology and severe thunderstorms and tornadoes is assumed.
Basics on severe thunderstorms and tornadoes can be found in Thunderstorms, Tornadoes and Hail! by Peter R. Chaston. Forecasting education (including the basics) can be found at Jeff Haby's The Weather Prediction.com. Advanced discussion about severe thunderstorms and tornadoes can be found in Severe Convective Storms edited by Charles A. Doswell III, which is available as a monograph from the American Meteorological Society.
I issued thousands of storm warnings to clients across Canada, the United States, and Mexico during my tenure at WeatherData, Inc. The discussions and experience working along side other meteorologists at WeatherData and elsewhere were an immense help in tweaking my warning methodology. I was very big on reducing my false alarm rate (FAR). Overwarning is a disservice to the folks who receive the warning. For example, a tornado warning from WeatherData, Inc. will shutdown plants potentially costing them millions of dollars. The lower the FAR, the better, as long as it does not compromise safety of course. You do need to be cautious with how you use new research in an operational setting though. In other words, do not assume the latest research is the latest and greatest technique that works. However, the literature on storm warnings in operational meteorology seems to indicate that a hook echo by itself is not necessarily enough to justify the issuance of a tornado warning. All of this has led to some of the ideas I have on forecasting severe thunderstorms and tornadoes that I will now present.
Use of Computer Models:
Computer modeling has definitely improved weather forecasting, but weather forecasting is not as simple as looking at the latest model run and calling the forecast done. All of the latest hard data (satellite, radar, upper air, surface) should be analyzed first. I also recommend printing out the upper air and surface data and analyzing them by hand because this is the best way to really get a feel for what is going on at the different levels of the atmosphere. Thereafter, the computer models can be critically evaluated based on all of these data. I often start by looking at the computer model's initialization and then I move into their individual forecasts. I compare the computer model forecasts with reality as I work through their individual forecasts, so certain elements that a particular model is not handling well can be noted. I am not partial to any comptuer model in particular because there are strengths and biases with each one that I will not go into here. Good science involves using the right tool for the right thing and being "biased" to one model is not recommended. I do look at the Model Output Statistics (MOS), but I do not treat it any differently than any of the model panels. Do keep in mind that MOS has a climatology bias, and does not perform well in predicting extreme events. You should have no problem picking out problems in the model runs if you have analyzed all of the available non-model data thoroughly, and you will end up with a much better forecast.
There is an art to weather forecasting too. You cannot just jump into it and expect to do better than a veteran forecaster...experience and continued education are keys to improvement. I still have plenty of things to learn! See this essay by Dr. Chuck Doswell for further discussion on the use of computer models.
A Sweet Spot?:
The discussion that follows has been on my mind for several years, and there may be something here that needs to be fixed in our forecasting of severe thunderstorms and tornadoes. Since the atmosphere is run by non-linear equations, we have to set limits on the variables until someone comes along that can invent another math beyond calculus. Given the need for both lower and upper limits, a question can be raised...Is there a sweet spot that when crossed on the side of too unstable or too much upper air support actually decreases the probability of organized severe thunderstorms and tornadoes? It seems that the answer to this question could be yes. Here are the two specifics I look for that immediately draw flags and the reasoning behind them:
1) Synoptic-scale surface winds greater than 35 knots
2) Storm system with a widespread area of lift and little cap and/or support yielding storm movements of greater than 50 mph
In 1996, Mike Smith suggested a correlation between violent tornadoes and rain prior to the arrival of the tornadic supercell. The theory here is that the prior rain will produce outflow boundaries with localized maxima of storm relative helicity and CAPE. It is well-known that severe thunderstorms can become quite nasty when they run into boundaries. Another possible aspect of this is the change in stability of the atmosphere when it rains in the morning only to become sunny in the afternoon on a potential severe thunderstorm day...the rate of change in instability with respect to time dramatically increases. In other words, the atmosphere becomes unstable quickly and becomes more prone to explosive development. The prior rain aspect seems to be quite important, and is supported by additional peer-reviewed studies (e.g., Markowski et al. 1998; Rasmussen et al. 1998).
In the Field:
It is quite clear that only using radar data for storm warnings is not necessarily sufficient. There have been numerous accounts when strong rotation or a hook echo has been observed on radar but yet storm spotters and storm chasers report nothing of note with the storm. It is interesting to note that the theory of tornadogenesis descending from a Tornado Vortex Signature aloft may not be as universal as previously thought (see this discussion by Dr. Chuck Doswell). Instead, it appears that the overall environmental conditions the storm is moving through play a crucial role in its overall intensity (e.g., Rasmussen et al. 2003; Davies 2004). Thus, I always check the latest surface observations and satellite observations in addition to the radar data, so a complete picture on the overall storm environment can be obtained. As an example, on 7 May 2002, a tornado warning was issued on a storm in Harper County, KS (and it was justified), but given that storm would soon move into a more stable environment in vicinity of Wichita, I had my chase team ignore the storm and continue farther west into southwestern KS where there had been plenty of afternoon sunshine. This was the correct move though I admit that it was tough to leave that nice looking storm behind. Here are the things I look for that seem to have operational value:
1) Strong gate-to-gate rotation on storm relative velocity data
2) Hook echo/kidney bean shape on base reflectivity data
3) Positive cloud-to-ground lightning bolts in vicinity of the "tornado area" of the storm
4) Decrease in storm foward speed or storm turning toward the right of the mean flow
5) Storm in vicinity of a boundary
6) Storm in or moving into an area of locally backed surface winds
7) Storm in or moving into an area where the level of free convection is low
8) Storm in or moving into an area where the lifted condensation level is low
9) Storm in or moving into an area with high amounts of instability (high CAPE, low lifted indices, etc.)
10) Storm in or moving into an area with high amounts of surface moisture convergence
See this discussion by Jon Davies for some specific details on the environmental values needed as well as some other useful tidbits regarding ingredients and patterns in tornado forecasting.
The Storm Prediction Center's Mesoanalysis Page is useful in order to analyze the various stability indices and other surface and upper-air conditions. Do note though that these data are the result of observations merged with the RAP computer model. Thus, it assumes the RAP model has a good handle on the situation. I also will look at satellite derived stability indices (links to which are also available via the link above). Although, they do not register values over areas that are cloudy and I have found that they tend to overestimate the amount of instability (CAPE will be too high, lifted indices will be too low). Of course, I also look at various plots of real surface data (e.g., surface map, theta-e, moisture convergence, streamlines, etc.).
Storm chasing is not simple! Even if you make the perfect forecast, there is still no guarantee you will see anything because there are other problems that can occur (e.g., all tornadoes being rain-wrapped and obscured from view). Then there are the bust days, but let's not go there. Hopefully some of the comments here will help in the forecast process. I know this stuff has been a tremendous help to me, and it is my intent to update these comments over time. Any comments on this may be sent to Thanks to WeatherData, KSNW-TV, the Wichita National Weather Service Forecast Office, Jon Davies, Dr. Rich Sleezer, and many others for discussions on this stuff over the years.