First, let me say…

GOES-E visible imagery at 2115 UTC on 2 June 1995. The outflow boundary associated with the significant tornadoes is shown in red.

I have inserted small boxes here and there guiding you to additional useful information.  Many of the links in these boxes are to the online archive of formal articles provided as a service by the American Meteorological Society.  You or your institution must subscribe to this service.  Contact the AMS for more details

Where I have inserted an asterisk (*) below, I intend to later add more material on another related page, and add a hyperlink.  This is a work in progress!

Tornadic storm ingredients

We commonly hear the popular media say that the “clashing of airmasses” caused a tornado.  That’s not exactly it.  For a simpler explanation of the ideas of this web page, go here*(link to follow).  It won’t be as simple as clashing of airmasses, but hopefully it is understandable.

The ingredients for tornadoes are not the ones that we have commonly identified for several decades of tornado forecasting… a jet stream, CAPE, a trigger, etc., although all of these are related to the process of tornado formation.

To forecast tornadoes, we need to understand something about the tornadogenesis process itself.  That’s covered on another page*.  The reader needs to be aware that this discussion refers to the dominant class of tornadoes that form because of the reorientation of initially quasi-horizontal rotation (like a rolling pin).  This is the way that supercells produce tornadoes.  We are excluding discussion of the “landspout” tornado in which the rotation is originally vertically oriented and extending to the ground before being acted upon by storm processes.

Tornadogenesis in a nutshell …

Many aspects of the tornadogenesis process are poorly understood, if at all. However, based on our limited observations, especially those obtained in VORTEX and subsequent efforts, we have begun to see some common processes occurring. These are briefly summarized here so that we can attempt to relate them to things that can be observed and forecast at scales larger than the storm.

The first step in tornadogenesis is the development of a rotating updraft. We believe this can occur on a variety of scales, from the typical radar-observed “mesocyclone” that occurs on a scale of several kilometers, down to scales closer to that of a tornado itself and not readily observed on Doppler radar, except at very close range. At this stage, we have rotation aloft… a precursor of tornado formation (although it sometimes is there only a few minutes before a tornado forms!).

Then, a downdraft must develop within the rotating column of air. It seems that this downdraft is usually at the upshear periphery of the mesocyclone. There is no other way to get rotation to the ground (except through turbulent diffusion of rotation from aloft which is a highly unlikely process). The details of how this downdraft is thought to transport rotation toward the ground are left to another page*. Not only does the downdraft bring the rotation downward, there is good evidence that in some cases it actually intensifies the overall mesocyclone as well. At this stage, we now have rotation extending to the ground, but we do not yet have a tornado.

In the third stage, some of the downdraft air that descends in the outer portion of the rotating region spreads upon reaching the ground. It may spread inward toward the axis of rotation, as well as outward away from the axis. To the extent that the inward-spreading air contains buoyancy, CAPE, and does not contain CIN, it can then accelerate upward near the axis of rotation. The associated convergence and stretching leads to the development of a tornado.

Three ingredients for tornado formation …

As just described, we believe supercell tornado formation follows a process of 1) the development of a persistent rotating updraft, 2) the development of a downdraft partially embedded in the rotation that aids in the development of rotation to the ground, and 3) the focusing of that rotation through convergence if the downdraft reaches the ground with some very uncommon properties.  Based on this description, it is our hypothesis that tornado formation involves the intersection of three ingredients, which are described below.  When thinking about the ingredients, the reader should become aware that …

• Because all three ingredients must occur in concert, tornado formation is much less frequent than the occurrence of any one or two of these processes.
•  These ingredients themselves are rather infrequent in the atmosphere.
•  Tornadoes can be expected to occur even on some occasions when forecast parameters are generally unimpressive. The three ingredients that must occur to allow tornado formation are illustrated in this diagram.  None of these ingredients is common in the atmosphere; tornadoes in this forecast model only occur when all three occur jointly. 

The three ingredients, and why they are relatively uncommon, are described in subsequent sections.

Persistent updraft

What does persistent mean in the context of tornado forecasting?  Consider the brief explanation above regarding the tornadogenesis process.  An updraft must persist long enough to ingest air containing enhanced SRH and develop rotation.  It must continue to persist as the RFD develops, focuses, and concentrates rotation below it.  Then it must persist, even though it might be shrinking and weakening, through the tornado life cycle*.

Rasmussen and Straka (1998) discussed the role of upper-level storm-relative flow in the supercell precipitation distribution. We have attached a page of graphs from the 1992 climatology that tends to refute the hypothesized role of deep shear in promoting tornadoes.

Thus, the longer an updraft persists, the more likely tornado formation becomes. What sort of updraft are we talking about?  In order for the tornadogenesis process to occur, the updraft must process near-ground, SRH-rich air.  This happens when the updraft extends below the region of buoyant ascent toward the ground, a result of vertical pressure gradient forces related to the interaction of the updraft with lower-tropospheric shear.  Further, because tornadoes form beneath the updraft, the low-level ascent should not be shallowly sloped as it often is when the storm is associated with a vigorous gust front.

A strong updraft probably can ingest SRH, and organize it into a mesocyclone (of some scale) faster than a weak updraft.  Further, if ambient SRH is very large, the process can probably occur more rapidly than if SRH is small.  In any case, the amount of time for the process to occur can be as short as perhaps, 10 minutes and as long as,

The detrimental effect of cold pools was demostrated by Brooks (1994) using a cloud model and through theobservational work of Paul Markowski

A typical “airmass” thunderstorm updraft persists for about 40 minutes, but perhaps this is misleading.  It may not persist for that entire period with the low-level strength and configuration needed to tilt horizontal vorticity and develop a mesocyclone. 

See Rotunno and Klemp (1982) to learn more about how low-level updrafts are caused by shear.

This is an important issue: updrafts in shear can produce low-level lifting at levels where there is no parcel buoyancy, and it is this low-level processing of inflow that brings low-level SRH into the updraft.The long-recognized supercell storm is more likely than other storm types to produce tornadoes largely because it has relatively long updraft persistence.  Updraft persistence is related most strongly to shear through the lowest one-half of the troposphere (because this forces low-level lifting as just mentioned), as well as a combination of precipitation distribution and low-level humidity.  The latter two factors are important in controlling the nature and vigor of the pool of evaporatively cooled air that may or may not form beneath the storm.  Vigorous low-level cold pools beneath the updraft are detrimental to tornado formation.  If the near-ground air is relatively dry, relatively little precipitation falling around the updraft could produce a vigorous cold pool.  On the other hand, if the near-ground air is nearly saturated, cooling will be weak even if there is a lot of precipitation around the updraft

. 

When I began to write this, I thought I could neatly show that if the low-levels were dry, more deep shear was required for supercells to keep precipitation from falling into, and cooling, the sub-updraft inflow.  Conversely, I thought that if low-levels were moist, any deep shear was allowable for supercells because precipitation falling near the updraft could not cause the generation of cold pools. I looked at the 1992 climatological data set (see below)

Rasmussen and Straka (1998) discussed the role of upper-level storm-relative flow in the supercell precipitation distribution. We have attached a page of graphs from the 1992 climatology that tends to refute the hypothesized role of deep shear in promoting tornadoes.

in a variety of ways, and I just cannot support this conjecture.  It seems that the processes responsible for cold pool generation are complicated.

The one thing we are fairly certain of at this time is that large low-level humidity (as evidenced by low LCLs) is protective of the supercell and promotes tornado formation, most likely by inhibiting cold pool formation.  In drier environments, we feel it is quiet likely that storms must have a lack of precipitation near the updraft, but this is a complicated issue.

The topic of supercell structure and precipitation distribution will be covered in another part of this web site*.  We are actively exploring this topic in 2001-2003 with the generous support of the National Science Foundation.

In summary, the first ingredient in the tornadogenesis process is a persistent updraft, rooted in the boundary layer immediately below the updraft.  This is not a typical thunderstorm updraft.  It occurs when outflow is inhibited owing to large low-level humidity and/or (possibly) precipitation being deposited well away from the updraft because of sufficient deep shear, and when the subcloud updraft is present owing to shear/updraft interaction producing an upward-directed pressure gradient.

Sufficient SRH

As mentioned above, we’ve done some interesting studies in the last few years using all of the evening soundings in the U.S. in 1992.  A recent finding that is not yet in the formal published literature is that the 0-1 km layer SRH is more important for tornado forecasting than the historically

used 0-3 km layer. In fact, we can’t determine just how shallow the layer is that is the best predictor because of the way sounding winds are reported and archived.  You can read this entire short paper, with the newest findings,

This climatological paper, Rasmussen and Blanchard (1998), can be read via the AMS online service.

In the study, we found that SRH measured in the 0-3 km layer was one of the stronger predictors of tornadoes.  At the same time that this study was being conducted, Paul Markowski was investigating the variability of SRH.  He found that SRH varies quite strongly on scales that are never sampled with conventional sensor network (soundings, profilers, VAD), and even inadequately sampled with special networks such as in VORTEX. 

Related  Papers:

          ·    Markowski et al. (1998) 
           SRH Variability
       ·    Markowski et al. (1998) 
           Tornado/Boundary 
       ·    Markowski  et al. (1998)
           Anvil Shadows
       ·    Rasmussen et al. (1998) 
           Boundary Case Study

So, we had two new pieces of information that seem to be somewhat at odds.  On the one hand, we found that SRH was augmented on larger scales when significant tornadoes occurred.  On the other hand, we had good evidence that SRH is quite augmented on nearly undetectable scales where supercells are tornadic.  Indeed, several of our case studies show inflow values of SRH exceeding 1000 J/kg in supercell inflow! An example is the hodograph shown below, with SRH > 1100 J/kg.  

Taken together…

• the fact that SRH often (not always) is elevated on the synoptic scale when significant tornadoes occur;

• SRH is much larger than the sounding climatology values when actually measured in 10’s-km scale regions and in the inflow of tornadic supercells

…implies to us that nature has augmentation processes for SRH and that these typically operate when SRH is already elevated. That makes the question of “how much SRH is required” for significant tornadoes difficult.  I think we can best answer it by saying that the forecaster should be aware of synoptic values of SRH, as provided by soundings, profilers, NEXRAD, etc.  If these are large, then local augmentation is easier and less is required by the augmentation mechanisms (below).  If large-scale SRH is small, then attention to mesoscale augmentation processes becomes even more crucial.

From theoretical arguments, SRH (for certain storm motions) increases when horizontal vorticity increases.  This happens whenever a buoyancy (temperature) gradient exists, and parcels spend time in the gradient instead of passing quickly through.  It takes relatively little time to generate horizontal vorticity large enough to support a tornadic supercell. For typical temperature gradients associated with outflow boundaries, parcel residence of about 20 minutes leads to horizontal vorticity of this magnitude.  Low-level outflow boundaries are probably the dominant source of narrow zones of enhanced SRH. 

Although we do not yet have the sensors we need in order to detect enhanced SRH (a mesonet of lower-tropospheric profilers would be nice!), it may be possible to infer its presence.  Areas with earlier outflow, wetted soil, and satellite-indicated boundaries are suspect.  And assuming that the flow is much more uniform at the 1-3 km level than it is at the ground (assuming the temperature contrasts are largest at the ground), then monitoring the surface winds for local backing is also an important tool for detecting augmented SRH.

We do not know what the details of the low-level horizontal distribution of SRH look like.  We suspect that on days after rain has fallen, or days with active convection, there could be a spaghetti-like distribution of enhanced SRH, and predicting which storm will encounter one of these spaghetti is nearly impossible.  Sometimes, it is simpler, with one dominant outflow boundary associated with most of the tornadic storms. It is one of our goals to measure the SRH distribution in the pre-storm environment using dual mobile Doppler radar during this first decade of 2000.

In summary, we cannot say how much SRH is required for a tornadic supercell to occur.  This threshold may vary with updraft strength, persistence, measurement scale, and other factors. We can say that if synoptic-scale SRH is augmented, significant tornadoes are more likely, probably because the additional augmentation is more likely and less needs to occur.  But because low-level baroclinic zones, if parcels reside in them for sufficient lengths of time, are such strong generators of SRH, forecasters must be cognizant of these zones even when background SRH is small.

Special RFD

One of the most exciting findings of VORTEX and its successors is that rear-flank downdrafts in tornadic supercells seem to have a very unusual character compared to non-tornadic supercells and thunderstorm downdrafts in general.  This finding comes from the Ph.D. dissertation research of Paul Markowski and collaborators, which should be submitted for formal publication during 2001.  By examining mobile mesonet observations from beneath about 18 tornadic and 12 non-tornadic mesocyclones, the following was found:

• Tornado cyclones… ~2-3 km diameter vortices… extended to the ground in all but one of the storms.  The sample is biased towards storms that appeared to have good tornado potential, but suggests that mesocyclones that fail to produce tornadoes do not fail, in general, because of an “undercutting” by outflow.  The notion of “outflow dominated” supercells may be much overused, and it is possible that many supercells have vortices extending to the ground embedded in the outflow below the updraft.

• In tornado cyclones that produce tornadoes, the RFD reaches the ground with more CAPE, less CIN, larger equivalent potential, wet bulb potential, and virtual potential temperature than in tornado cyclones that do not produce tornadoes.  In fact, equivalent potential temperature in tornadic circulations is about the same as the supercell inflow, while in non-tornadic circulations it is colder.  Note that all three potential temperatures are correlated with each other.

These are very significant findings in our effort to understand and forecast tornadogenesis.  Unfortunately, when Markowski examined proximity soundings to all of these events, there were only very weak signals at best.  The one environmental measurable that was reasonably well correlated with RFD character and with tornado production was the surface dewpoint depression in the airmass the storm was moving through.  This is completely consistent with the strongest predictor found in the 1992 climatological study of Rasmussen and Blanchard: LCL height. 

We speculate that the following description summarizes why buoyancy may be important in an RFD.  The RFD is known to descend in an annular, or semi-annular region roughly centered on the axis of maximum low-level rotation.  I.e., it descends around the developing vortex.  (The degree to which the RFD is driven by thermodynamic/microphysical forcing, and/or dynamic forcing through vertical pressure forces, remains to be resolved.)  Upon reaching the ground, some of the RFD air flows toward the axis, and some flows away from the annular region and thus away from the vortex.  It appears that the vigor of the down-in-up flow vs. the down-out flow is related to the buoyancy present in the RFD air.  If it is relatively buoyant, more air flows toward the axis with subsequent convergence and stretching leading to tornado formation

From a forecasting perspective, large low-level humidity (i.e., small dewpoint depressions, low LCLs) in the presence of sufficient CAPE is a red flag that the threat of significant tornadoes is enhanced.  We must point out that it is a rare occurrence in the atmosphere to have small dewpoint depressions and still have CIN small enough, and CAPE large enough, for supercells.  It is much more common to have humid low-level conditions in which CIN is large and CAPE is small or nonexistent.  We also note that humidity is higher on the cool side mesoscale outflow boundaries, where SRH is enhances as discussed previously.  This means that boundaries may play a role in tornado production beyond the enhancement of SRH.

The situation of tornado threat in relatively drier low-level environments is much more complicated and will require additional research into the conditions in which the RFD can reach the surface with sufficient CAPE and reduced CIN for tornado formation.  Right now, we think that a dry environment means that the precipitation in the hook echo must be “just right” to prevent too much evaporation, while a humid environment affords much more latitude in the amount/type of precipitation in the RFD.

Deterministic forecasting

It follows from our research that tornadoes cannot be forecast deterministically.  There are two primary factors that make this so.

First, we cannot observe the actual distribution of SRH in the atmospheric boundary layer.  At times, this is not important: in some cases, large values of SRH may be widespread and fairly uniformly distributed.  But typically, SRH is probably concentrated into structures that we have not yet identified.  It may be concentrated in boundary layer rolls.  It may be in ribbons or spaghetti following the passage of convective cells that produce cool outflow.  It may be in narrow, long ribbons along mesoscale baroclinic boundaries.  Regardless of the distribution, it is generally impossible to observe, and only somewhat more possible to infer based on the presence of boundaries, cloud bands, etc.  Much research remains to be done on this topic, and much more dense observing systems must be developed.  (My favorite scheme is surface mesonetworks with boundary layer profilers, such as SODAR, at many/most of the sites.)  Because we cannot know where enhanced SRH lurks, we cannot deterministically forecast where updrafts might interact with the enhanced SRH.  All is not lost: forecasters must remain cognizant of the possibility of enhanced SRH that we can anticipate, such as baroclinic boundaries.

Second, we cannot yet forecast the character of the RFD.  We do know that in near-saturated lower-levels, the amount of type of the precipitation in the RFD is much less important because not much evaporation can occur.  But with the more typically observed low-level humidities, it becomes more important to know what kinds of hydrometeors compose the precipitation that contributes to the RFD.  A few sparse, large drops or hailstones will not cause much cooling; large populations of small drops could very well cause too much cooling.  And in the case of a dry boundary layer, it is vitally important that the RFD contain only sparse large precipitation particles.  Until we have dual-polarization capability on NEXRAD, and surface mesonetworks with which to characterize low-level humidity, we cannot measure what we need to measure.  Hence, we can only make probabilistic statements about whether the character of the RFD will limit or enhance tornado potential.

At this point, it is appropriate to comment on the goal of the National Weather Service to greatly improve tornado warning statistics.  Our findings argue strongly against the idea that we can improve statistics with the present tools and warning methodology.  Tornado warnings are not crafted to reflect our knowledge and the inherent probabilistic nature of the problem.  Further, significant improvements in tornado warnings will require both new knowledge and new tools.  It is not enough to detect vortices with NEXRAD radar: our work shows that many (if not most) tornadic supercells have vortices extending to the ground, and it is not clear that these are stronger in tornadic supercells than non-tornadic.  These vortices may or may not be detectable depending on range from the radar.  The best hope for improving tornado warnings is to incorporate knowledge of the storm environment (especially local SRH… again, unmeasurable… and low-level humidity) with NEXRAD information.  It can be argued that surface mesonetworks with the capability of wind profiling through at least the lowest 1 km are badly needed if significant improvement in tornado warnings is to be had. 

Finally, we are crippled by the lack of investment in new knowledge compared to the investment in new technology.  We do believe that tornado and other severe weather warnings can be made much more valuable to the public, but it is not clear that we are on the path to achieving that goal.

It is quite likely that the ideas in this section are very incomplete.  We may find that there are other processes, not yet identified, that contribute to, or inhibit tornado formation.  So even with perfect observations, and perfect knowledge of the role of SRH, updraft persistence, and RFD character, coupled with perfect observations, there could still appear to be a strong element of randomness to tornado occurrence.

Once we gain more knowledge, and the necessary routine observations, we can be more confident of our probability estimates, but tornado forecasting will naturally best done probabilistically for the near future.  The issue of how we convey probabilities to the users of forecasts is an entirely different issue that deserves much more attention than it presently receives in the meteorological community.

Further discussion

We welcome further discussion of the material in this essay.  If you have thoughts to share concerning tornado forecasting, send an email to ras@ou.edu.  At my discretion, and with the permission of the sender, I will post these emails and some comments below.