More or Less Scientific Tornado Thoughts


C. Doswell

NOTICE! The sole responsibility for the content of this essay is mine, and so the thoughts contained herein should not even remotely be considered official scientific output from the VORTEX Project, NSSL, ERL, NOAA, Dept of Commerce, or the President of the United States ... nor from any university-affiliated person or organization. Please consult the scientific literature for the latest in the "official" science.

Last update: 18 May 2003: added some new material and moved this page onto my "home" website.

I. Mesocyclonic storms

Current conceptual models of tornadoes focus on mesocyclones [a misnomer, by the way! See my essay about the definition of a tornado] Many (not all) have come to accept the notion that a supercell is a storm with a mesocyclone and a mesocyclone can be defined as a vortex meeting some vorticity (or shear) magnitude and temporal/spatial continuity thresholds. Because the thresholds are basically arbitrary, this create the usual run of classification problems, but it seems that the definitions can be "engineered" so that most of the storms which nearly everyone would call supercells are included, and most of the non-supercells are excluded. This arbitrary definition is a problem, naturally, but not a terribly serious one; a line has to be drawn somewhere, even if we realize the dangers of classification thresholds.

Assuming that a supercell is a storm with a mesocyclone, the generally accepted mechanism for mesocyclogenesis at mid-levels in supercells is the tilting of streamwise environmental vorticity into the vertical by the updraft of the convective storm. This mechanism seems to make an effective predictor of supercells in spite of a number of limitations tied to trying to define what is the "environment" of a convective storm. The storm itself alters the near-storm "environment" and pre-existing convection and other mesoscale structures make sampling of the environment a problem. Once again, although these difficulties have to be acknowledged, they do not appear to be serious impediments to a general acceptance of the process of supercell mid-level mesocyclogenesis through tilting of streamwise environmental vorticity.

This process results in a storm with a mesocyclone aloft, say around 3 km AGL initially. Such a storm with a sufficiently strong, deep, and persistent mesocyclone aloft is a supercell by definition. This sort of storm produces a readily-detectable signature with a single-Doppler radar at substantial ranges, because such a signature is relative untroubled by radar horizon (and beam resolution) problems at distance. The WSR-88D may not resolve the details of such a circulation and a long ranges may well underestimate the magnitude of the horizontal shears owing to spatial sampling errors, but the presence of the mesocyclone can be detected in many cases more than 200 km from the radar.


II. Tornadoes and mesocyclones

As our detection capability expands with the implementation of the WSR-88D radar network, we are coming to the awareness that the fraction of storms with radar-detectable mesocyclones (by most reasonable criteria) is larger than we thought. This is associated with a concomitant decrease in our perception of the fraction of supercells producing tornadoes. Whereas the JDOP and DOPLIGHT experiments produced figures of around 50% of mesocyclonic storms producing tornadoes, the figure eventually may go as low as 10-20%. Even at 50%, we had a problem of trying to anticipate which mesocyclonic storms would produce tornadoes. Should the proportion of tornadic mesocyclones go down to 20% or so, as it now appears it might, this problem is correspondingly exacerbated.

During the early development of Doppler radars, it was found that tornadic vortices (as evidenced by shears meeting TVS criteria) can develop aloft as much as 20-30 min before a tornado (by definition, a tornado is a tornadic vortex reaching the surface) develops. This gave rise to some optimism about the role of Doppler radar in providing lead time for tornado warnings. The TVS developed within the mesocyclone and descended with time, leading to a notion of the downward development of the tornadic vortex. This has become the paradigm (and is sometimes referred to as the cascade paradigm), as it proceeds downscale), and at one time, it seemed that the only task confronting us was to understand how the downward development process operated.

Of course, this left the non-supercell tornado out of the picture. A new conceptual model for at least one type of non-supercell tornado grew out of the Colorado front range observations of tornadoes arising in ways quite reminiscent of the waterspout, leading to the term "landspout" as a way of denoting this model. In this model, "misocyclones" (using Fujita's term) existed at low levels along some sort of boundary before convection even developed. In the conceptual model, these pre-existing, low-level misocyclones would be overridden by developing updrafts, stretching the existing low-level vorticity to tornadic proportions. This suggests an upward development, seemingly quite distinct from the supercell paradigm. The landspout model left other non-supercell tornado mechanisms (e.g., gustnadoes) up in the air (pun intended!).

Moreover, recent work by Trapp and Mitchell has suggested that the downward development paradigm is not as universal as once thought. Their observations underscore the continuing failures of the cascade paradigm that once was widely felt to explain what was going on rather well (at least by some, including me!)

Recently, numerical modeling experiments and some limited observations have led to a recognition that the development of mesocyclonic vertical vorticity values at low levels occurs via a different process from that producing mesocyclones aloft. The tilting of horizontal vorticity by an updraft cannot explain amplification of low-level (surface and near-surface) cyclonic vertical vorticity because cyclonic spin develops as the air is rising.


III. New Ideas

New conceptual models have developed from these studies, suggesting that the production of low-level cyclonic vertical vorticity is associated with baroclinic generation of horizontal vorticity that is tilted upward in the downdraft (or by the outflow boundary generated by the downdraft). The resulting cyclonic vertical vorticity is generated near, and is entrained by, the updraft which, in turn, amplifies it further by stretching.

In these conceptual models, the development of a low-level mesocyclone depends on baroclinic generation due to evaporatively-chilled outflow. This outflow may or may not "undercut" the mesocyclone aloft, depending on the storm relative flow aloft in the mesocyclone-bearing levels of the storm. The relative balance between (1) precipitation export into the anvil by the storm-relative flow at those mid- and upper levels and (2) the tendency for the mesocyclonic flow to wrap itself in precipitation determines the rate of low-level horizontal vorticity production and its relationship to the mid-level mesocyclone. In the conceptual model, long-lived, low-level mesocyclones are the result of a vertical alignment of the mid-level and low-level mesocyclones, leading to a deep vortex from the surface to well up into the convective storm. Too little precipitation near the updraft and the low-level mesocyclone never develops; too much precipitation near the updraft and strong cyclonic vertical vorticity develops rapidly at low levels but is swept out from under the mid-level mesocyclone by the outflow and elongated along the gust front. In the former case, tornadoes are unlikely whereas in the latter case, tornadoes may develop early in the storm's life cycle, but are much less likely during the outflow-dominated phase of the storm (HP supercell phase).

In the years since 1995, some continuing observational experiments (much more modest than VORTEX) have followed up on some observations made during VORTEX: the rear flank downdrafts (RFDs) in tornadic storms are still relatively warm and have some buoyancy, rather than being cold and stable. When RFDs become cold and stable, any ongoing tornadoes dissipate, but during formation and intensification of tornadoes, the RFDs are not very baroclinic near the surface. Further, mobile Doppler observations have shown the presence of cyclonic-anticyclonic vortex couplets in the RFDs, with any tornadoes typically associated with the cyclonic member, and obviously, any anticyclonic tornadoes associated with the anticyclonic member. This strongly suggests tornadogenensis is tied to vortex tilting in the RFD. This research, largely associated with Erik Rasmussen and Paul Markowski, looks VERY promising to me. The details have yet to be worked out, but I think it comprises very important new insights into tornadogenesis. It appears that boundary layer relative humidity (RH) is an important factor in determining the buoyancy of the RFD ... with low boundary layer RH (and correspondingly high lifted condensation levels [LCLs]) in the inflow to a supercell, the RFD tends to come down cold and stable, whereas with high RHs and low LCLs, the RFD tends to come down relatively warm and with some residual buoyancy. This indicates that at least some of the air in the RFD has origins in the inflow, at least near the axis of the mesocyclone.

Continuing research should clarify these details and may eventually lead to some useful new forecasting tools, finally. Thus, my relative excitement regarding the initial frustrations following VORTEX-95 may have been justified: apparent failure of existing paradigms through observations-based work is simply an invitation to rethink the science, and this healthy process is alive and well!


IV. Problems

Storm chase experience, some unexplained observations, and some new WSR-88D tornado cases, along with the VORTEX observations have begun to point at some inadequacies of the existing paradigms:

1. Chasing indicates that dust whirls and brief landspout events can occur along the flanking convective lines, especially early in a supercell's existence, well before the occurrence of major tornadoes associated with the mesocyclone and wall cloud. Such events do occur at almost any time in the life of a supercell, however, even as a tornado associated with the mesocyclone is occurring.

2. The have been observations of tornadoes well away from mesocyclones in bowing squall lines. The mesocyclone in such cases may be relatively distant from the tornadoes and can be thoroughly undercut by outflow. The tornadoes occur between the bowing line segment and the undercut mid-level mesocyclone, along the outflow boundary in regions of enhanced low-level cyclonic vertical vorticity. These events have been documented by WSR-88D radars. Such events have a landspout-like character, as developing updrafts along the outflow override enhanced low-level cyclonic vorticity.

3. The 29 May 1994 VORTEX case indicates that an F3 or possible F4 tornado occurred with a rapidly developing new updraft well removed from a pre-existing nontornadic mesocyclone, along a flanking line. This tornado developed rapidly and did not descend from a TVS aloft. There was some limited evidence of a mesocyclone aloft near the tornado, but the intensity dropped below thresholds before the tornado. The new cell developed a small-scale hook echo structure, but this morphology co-evolved with the tornado. The Doppler data prior to the tornado are not particularly compelling regarding the mesocyclone aloft, with a much more pronounced mesocyclone well-removed from the tornado. The tornado fails to fit the "cascade" paradigm, irrespective of whether or not a bona fide mesocyclone was present aloft prior to the tornado.

4. VORTEX-95 observations have demonstrated that many tornadic storms develop their tornadoes through complex interactions with pre-existing environmental features, meso- or even miso-scale structures about which little or nothing is known (e.g., 16 May 1995). It may be that most tornadoes are the result of this sort of complex process rather than the "cascade" toward a tornado envisioned in the paradigmatic conceptual model (mid-level mesocyclone to low-level mesocyclone to TVS aloft to tornado). If the process often includes interactions with low-level boundaries [most of which have unknown origins and dynamics], then it seems obvious that the challenges are multiplied .. we need to learn more about such boundaries if they are as important as they seem to be. The cascade may be the exception, rather than the rule. When we look closely, it seems that supercells arising in "pristine" environments [i.e., without previous convection and associated boundaries] have trouble producing significant tornadoes in spite of well-developed mesocyclones aloft. It appears that supercells in "synoptically evident" tornado outbreak events may be the most likely to follow the paradigm, and even then there are questions (e.g., Hesston, and storm interactions on 3 April 1974). Also, it is becoming evident that the tornadic storms may not always be the largest storms with the strongest mesocyclonic signatures (e.g., 17 and 22 May 1995).

5. Subsequent observations have continued to show that tornadoes often develop rapidly (in 5 min or less) within storms that shortly before had little or no indication of a tornadic circulation. They often develop from low levels upward, rather than following the cascade paradigm, or develop rapidly through considerable depth more or less all at once.


V. What does this all mean?

These observations suggest that tornadoes occasionally reaching strong to violent intensities can occur that are not clearly associated with a well-defined pre-existing mesocyclone aloft. These can pose a challenge to the NWS, since the WSR-88D is going to have a great deal of difficulty discriminating tornadic from nontornadic events, or even detecting them in some instances. WSR-88D radars do not "see" tornadoes ... rather, they see the mesocyclonic circulations that usually precede mesocyclonic tornado events (obviously, not present with nonmesocyclonic tornadoes). Radar operators may be focusing on the wrong part of a storm even to notice and accept observed events as credible indicators of tornadic activity. The rapid development and dissipation of tornadoes makes it hard to imagine having much lead time based on radar observations alone, except in those apparently relatively rare cases when the storm fits the classical "cascade" conceptual model. Radar horizon problems virtually preclude any chance to detect low-level tornado precursors beyond a range of a few tens of km.

From a science viewpoint, this is an exciting collection of observations. The new radars and other observing tools are making possible a revision of existing tornadogenesis paradigms to encompass the full spectrum of tornadoes and to elucidate the relationship of the tornado the parent storm, be it mesocyclonic or not. The issue of tornadogenesis has always focused, in some sense, on the development of low-level vorticity. Without intense low-level vorticity there is no tornado; one might have at best a tornadic vortex aloft. [see my essay on "What is a tornado?"] Hence the emphasis in tornadogenesis properly belongs on the processes enhancing low-level vorticity. The rapid onset of tornadoes is a well-known observation from chasing, but the 29 May 1994 VORTEX case, among others, makes it clear that tornadogenesis must, in general, always be associated with vortex stretching; the only term in the vorticity equation capable of explaining this rapid intensification is the exponentially-growing stretching term.

To some extent, then, we can shift our attention away from the question of how the final amplification occurs, to the origins of the low-level cyclonic vertical vorticity that is stretched to tornadic magnitudes. Again in a general sense, the origins of this low-level vorticity must almost certainly lie in baroclinic generation (the only true "source" of low-level vorticity in the inviscid vorticity equation) and its subsequent tilting into the vertical. Although it can be argued that very strong horizontal vorticity (certainly of order 10-3 s-1 and approaching 10-2 s-1 on occasion) already exists in the environment owing to vertical wind shear, this pre-existing vorticity alone cannot be enough, or tornadoes would be much more frequent. Moreover, getting this vorticity to low levels is a problem, as noted earlier. It now appears that high values of horizontal vorticity can be generated on the boundaries of the RFD and these vortex rings may be the source of (mesocyclonic) tornado vorticity.


VI. Tornadogenesis

It seems that convective storms in general and supercells in particular make tornadoes possible by providing outflows that generate enhanced horizontal vorticity. Recall that observations and modeling suggest that low-level mesocyclogenesis and mid-level mesocyclogenesis appear to be physically distinct processes. Supercells might well "condition" the low-levels by wrapping precipitation around the mesocyclone aloft, forming the hook echo and creating an "open wave" low-level gust front structure. Moreover, it is recognized that outflow boundaries typically are associated with enhanced low-level vertical vorticity, perhaps through tilting. The supercell might be able to take advantage of this low-level vertical vorticity to develop low-level mesocyclones. At this time, it is not known definitively how supercells produce low-level mesocyclones; numerical modeling studies to date have not necessarily spanned the space of possibilities.

Recent research associated with Jeff Trapp has suggested that the fraction of storms with low-level mesocyclones (according to any reasonable criteria) that go on to produce tornadoes is also not 100% ... perhaps on the order of 50%. .The mechanism(s) by which a storm with a low-level mesocyclone goes on to produce a tornado is (are) not known precisely. VORTEX observations have indicated that on some occasions, tornadogenesis appears to be contingent on interactions between a supercell and other processes (e.g., radar-detected "fine lines" and satellite-observed cloud lines about which little is known). If it is indeed the case that many, if not most, tornadoes arise through unique processes, then forecasting and warning for tornadoes would be much more difficult than heretofore imagined. That is, it would be difficult to generalize about tornadogenesis because each situation would be unique and dependent on processes we find it hard to observe quantitatively.

Another situation highlighted during VORTEX is where a storm with a mesocyclone aloft is "undercut" by surface-based cold air. Such "elevated supercells" are unlikely to become tornadic; although some tornadic examples may well have occurred, our experience suggests they must be rare, so the probability of tornadoes is low in such events even though the storms may exhibit dramatic mesocyclone signatures above the surface. These seem to be especially troublesome to operational warning decisions currently. What is difficult to imagine is to see a way out of this dilemma, except in cases where the storm is very near the radar. We do not understand how undercut mesocyclones might on rare occasions produce a tornado, but it is not hard to picture how elevated supercells might arise: typically, there is very strong veering of the wind profile above a layer of surface-based cold air, producing very high helicity, even though the surface-based cold pool is quite statically stable. Undercutting seems to be associated with (a) storms that produce overwhelming outflow, or (b) storms that tap the surface-based cold pool associated with nearby thermal boundaries. Storms may form where there is warm, unstable surface-based stratification and then move over thermal boundaries, thereby becoming undercut, even though they are not so initially. The helicity enhancement just described over the cold air might result in an initially non-mesocyclonic storm commencing vigorous rotation only after being undercut! This might not be apparent when storms are sufficiently far from a radar, necessitating a detailed analysis of surface conditions in the vicinity of storms, if that is possible.

Even though it is a tautology, understanding that tornadogenesis is the product of enhancing low-level vorticity means that it is important to recognize that there are many different processes that accomplish this. Moreover, this means that we should examine the processes going on during tornadoes in detail from inception to demise. If some process enhances low-level vorticity during tornadogenesis, is it the absence of that process that leads to "tornadolysis" or does some counteracting process cause low-level vorticity to decrease. And are all tornadoes alike in this aspect (and others)?


VII. Some things that need to be done:

1. We need to have observations of what fraction of tornadoes occur with and without mid-level mesocyclones, especially nationwide. No matter how a mesocyclone is defined (within some reasonable limits), this is a critical observation owing to its clear connection to NWS warning operations. This means we have to develop a uniform set of criteria for defining a mesocyclone and apply those criteria to the emerging nationwide WSR-88D data base. Clearly, this requires detecting and recording non-tornadic mesocyclones. The advent of the WSR-88Ds make this project feasible, but it is a substantial effort and it requires archiving Level II data continuously! As already noted, we need to know the fraction of tornado-producing low-level mesocyclones, as well. Some early efforts have shown that the WSR-88D vortex detection algorithms produce a host of false detections and various artifacts that need to be eliminated from the data before a reasonable picture of the distribution of mesocyclones will emerge. Such research is ongoing, finally. There is even less known about how many low-level mesocyclones produce tornadoes than for mid-level mesocyclones.

2. There is a desperate need for information about processes operating in the lowest 1-2 km of storms, with high temporal resolution. Given that tornadoes can develop within 5 min or perhaps even less, volume scans every 15 min could be woefully inadequate. Even 5 min intervals would be inadequate to provide much-needed detail. VORTEX has done as well as can be expected in this regard, and the data must be evaluated over the next several years, but it is probable that even the VORTEX observations will prove to be inadequately detailed! Other strategies would need to be developed, if this turns out to be the case. In this vein, it appears that a critical issue about the tornadic potential of a storm with a mid-level mesocyclone may be whether or not that mesocyclone is "undercut" by cold air, either from storm-induced outflow or from tapping the cold air north of a frontal boundary. Methods for recognition of "undercutting" at the lowest levels in the storm are needed, especially high-resolution surface observations.

3. Numerical simulations will continue to be an important component in the process of understanding tornadogenesis. A prime concern is the ability to simulate tornadogenesis clearly and unambiguously. If it turns out that real tornadogenesis involves interactions between the storm and pre-existing processes, then the modelers finally must move vigorously into non-horizontally homogeneous initializations.


VIII. Some NWS Considerations

Given the potential difficulties associated with what we have observed with respect to operational warnings, the NWS needs to consider how it operates in convective situations:

1. Any algorithm tuned to recognize tornadoes using the "cascade" paradigm might well produce an unacceptable number of false alarms, have limited lead time capability in many situations, and might even have an unacceptably low probability of detection! The optimum approach almost certainly is one that uses algorithms to detect intense upward motions in regions of enhanced low-level vorticity. It is difficult to imagine how this might be done with existing radar technology alone, since vertical motion is not observed. It is critical to understand that the WSR-88D radars do not detect tornadoes! Therefore, spotters and other non-radar approaches (e.g., tracking power line breaks) will continue to be a critical aspect of a competent tornado warning system. Integration of a broad spectrum of meteorological information into the warning process is needed; depending on radar alone is not likely to be an effective strategy.

2. Detection of tornadoes also will probably require knowledge of the relevant mesoscale processes, well beyond that provided by existing information; unfortunately, the science of meteorology knows very little about those apparently relevant mesoscale processes. This means that considerable research would be needed to make mesoscale information useful in NWS warning operations.

3. Spotter training will have to be revised and updated continually as we learn more and as technology changes. At the moment, creating spotter training material is a long process, in part because it is always done using limited resources. When forecasters and scientists working on such programs are always resource-bound, then the training materials end up being less than what is possible and take longer to develop than is desirable. Spotters will continue to be an important component in the warning system into the indefinite future [Doppler radar will not solve the tornado warning problem!], and their important voluntary contributions need to be supported as best we can. Thus, we need to devote non-trivial resources continually toward improving the training programs for spotters!

4. Forecaster training is becoming more important with each new scientific finding. There is virtually certainty now that tornadoes are not always the simple end-product of a linear "cascade" beginning with an easily-detected mid-level mesocyclone. Effective warning will depend on recognition of potential in a wide-ranging variety of situations. New parameters for forecasting tornado potential, based on new findings, are being explored. Results appear promising, but not without challenges ... no "magic bullets" are likely to be found. Forecasters must be trained to recognize these things, to whatever extent the science permits.

5. Since there is as yet only limited credible knowledge of tornado precursors, one way to develop that knowledge is a partnership between forecasters and researchers. The observations described above include contributions from Ron Przybylinski and Al Moller, forecasters who have recognized the need for an operations-research partnership. We need many more like them and an environment that nurtures such collaborations with real resources.

6. That portion of tornado research which is truly "solid" is what is useful to operations. Normally, there are disagreements among scientists about what is really "solid" in the research results. Forecasters should be careful about taking research results from limited data sets too literally ... I've seen indications that some well-intentioned forecasters are pushing the limits of what science offers about tornadic storms and tornadoes too far. Introducing new techniques based on speculation can do more harm than good. New techniques need to be tested thoroughly in operations, and researchers should work very closely over an extended period of testing and check-out with the intended operational testing team. This should be a real partnership where the forecasters have a chance for benefits to their operations and not be simply "used" by the researchers. In their turn, forecasters should not simply "use" researchers, in the sense of expecting them to solve all their operational problems.

Forecasters need to resist the temptation to take the latest paper they've read, or the newest concept they've encountered and consider it to be the most essential thing about tornadogenesis. As new ideas are presented, they need to be considered (a) tentative, and needing thorough validation (especially in operational practice), and (b) in the overall context of the science. Basically, the operational use of science in forecast and warning operations requires careful thought and a certain amount of restraint ... scientific ideas change and should never be taken as "cookie cutters" that map out categorical operational strategies (i.e., if a real event doesn't look like the model, then there's nothing to worry about, and if it does look like the model, then the event is inevitable).

Anyone wishing to discuss, argue, or otherwise harangue me on the foregoing is encouraged to respond via e-mail. Send your thoughts to and let's get it on!