Tornadoes in a Deceptively Small CAPE Setting:  The "Surprise" 4/20/04 Outbreak in Illinois and Indiana
by Jon Davies

     4/20/04 tor tracks>042004tortracks.gif (13510 bytes) boonecotor_sml.gif (21932 bytes)<Boone Co. IN

The outbreak of more than 20 tornadoes in Illinois and Indiana on 4/20/04, including an F3 tornado that killed 8 people (see NWS Chicago, NWS Indianapolis, and adjacent NWS office sites for detailed information.), was one of the most difficult such outbreaks in recent history to forecast based on model data less than 12 hours prior to the event.  The tornadic supercells occurred along an advancing warm front in an environment where CAPE was relatively small compared to typical tornadic supercells.  The Eta and RUC models had trouble forecasting and depicting this instability until very close to the event, but there were subtle clues in advance for forecasters who monitored trends and evolution of observed data in combination with the model data.  This case highlights some important issues regarding tornado events that occur with relatively small CAPE, and that is the motivation for this study and overview.

In Thompson et al.'s recent paper about supercell and tornado environments from RUC soundings in Weather and Forecasting (WAF Dec. 2003), their Fig. 6 showed that significant tornadoes are most associated with MLCAPE (lowest 100 mb mean lifted parcel) notably greater than 1000 J/kg.  But occasionally, significant tornadoes occur in environments with lesser MLCAPE ("small CAPE", generally < 1000 J/kg).   The 4/20/04 case is an excellent example.

I've done some recent study with "small CAPE" tornado environments using a large database of RUC profiles associated with supercells. The graph and table below strongly suggest that tornadic storms in "small CAPE" settings with notable shear are associated with CAPE that is mostly below the mid levels of the atmosphere: 

smallcpe_graph.gif (8604 bytes)<MLCAPE distributions for "small CAPE" supercell cases   

supercell cases w/CAPE < 1000 J/kg  (110 of 518 cases in RUC database): nontornadic & F0 (79 cases) F1-F4 tornadoes (31 cases)
  mean value mean value
level of maximum buoyancy    488 mb   544 mb  ***
total MLCAPE (lowest 100 mb parcel)    584 J/kg   626 J/kg
0-3 km MLCAPE    13 J/kg   86 J/kg
MLCAPE below 500 mb    178 J/kg   294 J/kg  ***
% of total CAPE below 500 mb    32 %   47 %       ***
0-1 km SRH (storm-rel helicity)    211 m2/s2   245 m2/s2
0-6 km shear    48 kts   49 kts
0-1 km EHI (energy-helicity index)    0.67   0.98
STP (significant tornado parameter)    1.0   1.8    

You can see that, in small CAPE tornado cases, around 50 % or more of the CAPE is typically located below 500 mb, and the level of maximum buoyancy (most unstable lifted index) is centered below 500 mb.  This matches nicely a detailed modelling study by McCaul and Weisman (Monthly Weather Review, April 2001) that found when shear was sizable and held constant, surface vorticity in simulated storms increased as CAPE was compressed into lower levels of the environment profile. 

We'll refer back to the information above when looking at CAPE profiles later on for the 4/20/04 event. Because small CAPE tornado cases tend to be very sensitive to the depth of the mixed layer lifted, we will examine that as well.

Within a few hours prior to a major tornado event, sizable CAPE is usually evident or forecast over the area where the event occurs.  However, in the 4/20/04 case, CAPE appeared small or largely absent over Illinois and Indiana from model forecasts and early analyses:

042004ruccpe21f06_anno.gif (29916 bytes)<RUC 6 hr MLCAPE fcst 21 UTC   042004rucml100cpe19a.gif (5172 bytes)<RUC analysis of MLCAPE at 19 UTC

The above maps used mixed-layer parcels in the lowest 100 mb (MLCAPE).  Although surface-based (SB) computations often overstate the case for instability because they do not account for low-level mixing of lifted parcels, they may have been more appropriate for this case because of the near-saturated environment with reduced mixing along the warm front in this event.  Unfortunately, I did not save specific forecasts of SBCAPE for the afternoon of 4/20/04, but I did keep some RUC forecast graphics showing combinations of SBCAPE and storm-relative helicity / low-level shear (0-1 km EHI and VGP computations using SBCAPE):

042004rucsbeh121f06_anno.gif (25321 bytes)<RUC 6 hr fcst SBEHI 21 UTC   042004rucvg121f06_anno.gif (37234 bytes)< RUC 6 hr fcst SBVGP 21 UTC

On both graphics, the SBCAPE/shear combinations were most impressive in eastern Oklahoma, but an area of enhanced CAPE and low-level shear was also forecast at 21 UTC over southern Indiana into central Illinois. While the model forecast details did not turn out to be particularly good regarding location placement, the forecast of SBCAPE combined with measures of low-level shear did at least highlight an area along the warm front in Indiana and Illinois suggesting that this general area should be monitored for trends and evolution of local environment and potential for storm rotation during the afternoon.  

Other RUC forecasts showing 0-1 km storm-relative helcity (SRH) and 0-3 km SBCAPE roughly highlighted the same area.  Large SRH and low-level SBCAPE were suggested to be somewhat co-located over parts of Illinois and Indiana, hinting at possible concern for low-level storm rotation and tornadic supercells::

042004rucsrh121f06_anno.gif (34872 bytes)<RUC 6 hr fcst SRH 21 UTC   042004ruccp321f06_anno.gif (55455 bytes)< RUC 6 hr fcst 0-3 km CAPE 21 UTC

Moving to observed data, the surface map at mid morning showed a warm front stretching from Missouri to West Virginia that was forecast by the RUC to advance significantly northward by mid-afternoon:

042004sfc14_anno.gif (45656 bytes)<surface map 14 UTC   042004rucsfc21f06_anno.gif (59490 bytes)<RUC 6 hr sfc fcst for 21 UTC

The RUC 6 hr forecast 500mb forecast also showed significant divergence suggesting notable upper forcing during the afternoon over the Illinois/Indiana area:

042004ruc500mb21f06_anno.gif (52766 bytes)<RUC 6 hr 500mb wind forecast for 21 UTC

By 18 UTC, the warm front had moved 70-80 miles northward over Illinois, with the T/Td at Springfield (SPI) jumping from low-mid 50s F to low-mid 60s F:

042004sfc18_anno.gif (42908 bytes)<18 UTC surface map

At this rate of translation with a strong low-level jet, the front would probably advance into northern Illinois by late afternoon.  Note the precipitation and small T/Td spreads just north of the warm front, suggesting a very humid near-saturated environment with little mixing regarding low-level air parcels.  The backed winds also reflected strong low-level shear present just north of the front.

As noted earlier, the RUC analysis of MLCAPE at 19 UTC was not impressive, with most CAPE values notably less than 500 J/kg.  However, using a shallower mixed-layer parcel (lowest 50 mb) based on the near-saturated conditions near the warm front resulted in the area of CAPE extending further northward, with a larger area of CAPE values greater than 500 J/kg:

042004rucml100cpe19a.gif (5172 bytes)<19 UTC MLCAPE (lowest 100mb)   042004rucml50cpe19a.gif (5548 bytes)<19 UTC MLCAPE (lowest.50mb)

One could modify the 19 UTC RUC analysis profile at Peoria (PIA, just north of the front) using the Springfield 19 UTC surface ob (SPI, south of the front) to get an idea of how the local thermodynamic profile might evolve during the coming few hours with the approaching warm front:

042004pia19ruca_anno.gif (31133 bytes)<PIA 19 UTC RUC   042004pia19rucamod_anno.gif (32761 bytes)<PIA 19 UTC modifed by SPI sfc T/Td

This modification shows how quickly the environment could change from no CAPE to positive CAPE as the warm front approached.  Just as important, notice on the modified profile how much of the CAPE would be located relatively low in the profile.   Based on the small CAPE tornado case study mentioned earlier, this would be a favorable "small CAPE" tornado profile with nearly all the CAPE below 500 mb and the level of maximum buoyancy centered below 600 mb.   This is very different from typical spring tornado environments in the plains where only around a third of the CAPE is below 500 mb, and the maximum buoyancy is centered much higher (usually near 400 mb).

As the afternoon progressed, the CAPE at Peoria actually jumped to near 1500 J/kg based on RUC analysis profiles (not shown). Just north of the warm front in better low-level shear with backed surface winds, values were in the 500-700 J/kg range.  For example, see the RUC analysis profile for Pontiac (PNT) at 22 UTC below:

042004pnt22ruca_anno.gif (32544 bytes)<PNT 22 UTC RUC analysis profile

Notice how, like the PIA modified profile, nearly all the CAPE on the PNT profile was below 500 mb and the maximum buoyancy was centered below 600 mb. Notice also the very favorable wind profile and hodograph co-located vertically with the main CAPE.

To check how well the RUC model handled the general thermodynamic environment in this case, we can use the 00 UTC Davenport (DVN) RAOB later in the evening located not far south of where a tornado was occurring in Clinton county of extreme eastern Iowa (the RAOB at Lincoln IL was located too far south of the warm front): 

042104dvn00raob_anno.gif (31329 bytes)<DVN observed sounding at 00 UTC

Note how the CAPE at DVN was located mainly below 500 mb and the maximum buoyancy was centered below 500 mb, suggesting that the earlier RUC profiles in Illinois were effective in reflecting important characteristics of the thermodynamic environment.

Compare the sounding profiles above to an estimated  profile at late afternoon in northeast Oklahoma (at Bartlesville/BVO) where forecasters like myself were more focussed on severe weather potential with developing storms and supercells that for some reason had trouble producing tornadoes:

042004bvo22ruca_anno.gif (31115 bytes)<Bartlesville OK (BVO) 22 UTC RUC analysis profile

Notice that, in contrast to many profiles in the Illinois/Indiana area, the total CAPE in northeast Oklahoma was larger, but was elongated vertically and located much higher in the profile, not as well co-located in the vertical with the best turning and increase in speed of winds with height.  That may offer some clues as to why the Illinois/Indiana environment in this particular case, though subtle, turned out to be the better one for supporting significant supercell tornadoes. 

Let's examine the evolving CAPE environment at 22 UTC using plane view fields.   The SPC mesoanalysis page was not sectored on the Illinois/Indiana area until 23 UTC after tornadoes had begun, so I downloaded roughly 60 RUC analysis profiles from the FSL RUC web site for the local IL/IN area at 19 UTC (see the CAPE fields shown earlier) and 22 UTC, just before tornadoes began over northern Illinois and central Indiana.   These were used to generate CAPE fields at 22 UTC using lowest 100 mb mean parcels, and also lowest 50 mb mean parcels (a shallower mixed layer).

042004sfc22_anno.gif (44266 bytes)<22 UTC surface map

042004rucml100cpe22a.gif (7259 bytes)<19 UTC MLCAPE (lowest 100mb)  042004rucml50cpe22a.gif (8346 bytes)<19 UTC MLCAPE (lowest 50mb)

Similar to the 19 UTC fields displayed earlier, the lifted parcel choice in this "small CAPE/saturated environment" made a big difference, and the graphics above suggest that using the lowest 100 mb as the mixed layer in such situations can result in misleading information.  Notice how, with the shallower mixed parcel, sizable CAPE extended much further eastward into eastern Illinois and Indiana, and in some locations more than doubled in value!  The amount of increase in CAPE when using the shallower mixed parcel is shown below:

042004rucml50v100cpe22a.gif (5461 bytes)<increase in CAPE lowest 50 mb vs. lowest 100 mb mixed parcel

In most plains warm season supercell cases, differences in parcel selection would not be so significant as to greatly impact a forecast, but in this and other "small CAPE" cases I've studied, such differences can have a big impact. Compounding the situation on 4/20/04 was the fact that model profiles in eastern Illinois and central Indiana tended to show a small dry layer just above the surface that was somewhat questionable given the observed near-saturated surface environment near the advancing warm front.  A shallower parcel assumption helped to eliminate some of the "false" mixing that possibly contaminated the deeper mixed layer computations.

Shown below is the significant tornado parameter (STP) based on the RUC 22 UTC analysis incorporating CAPE using the two different mixed parcels:

042004rucml100stp22a.gif (6278 bytes)<22 UTC RUC STP (lowest 100mb)  042004rucml50stp22a.gif (6938 bytes)<19 UTC RUC STP (lowest 50mb) 

Notice the big impact that parcel selection had on the depiction of this parameter.

The tornado track map below shows where tornadoes occurred between 22 UTC and 01 UTC, progressing from southwest to northeast in both Illinois and Indiana (storms initiated at mid afternoon near the warm front in west-central sections of the two states).  SPC graphics of selected parameters are also shown below to be used with the track map and late afternoon radar images to see how the environment appeared relative to the where tornadoes occurred. (NOTE: the SPC mesoanalysis page did not focus on the Illinois/Indiana area until 23 UTC, so earlier graphics were not available).  See the NWS Chicago site for Al Pietrycha's summary of this case, including additional radar, satellite, and surface information:

042004tortracks.gif (13510 bytes)<tornadoes 23-01 UTC  042004rd22.gif (33909 bytes) 042004rd2330.gif (34411 bytes)<radar mosiacs

(These SPC mesoanalysis page graphics are courtesy of Daniel Nietfeld at NWS Omaha and Andy Kula at NWS Des Moines...)
042004spcmlcpe23.gif (39052 bytes)<SPC 23 UTC MLCAPE          042004spcsbcpe23.gif (39746 bytes)<SPC 23 UTC SBCAPE
042004spcsrh123.gif (35926 bytes)<SPC 23 UTC 0-1 km SRH       uticator_sml.gif (15781 bytes)<Utica IL tornado(F3) near 23 UTC
042004spcstp23.gif (28758 bytes)<SPC 23 UTC STP                   042004spceh123.gif (21355 bytes)<SPC 23 UTC 0-1 km EHI

At 23 UTC above, notice the differences between MLCAPE and SBCAPE during the heart of the outbreak, similar to the 22 UTC parcel comparisons earlier.  This impacted STP (the significant tornado parameter) and resulted in "negligible" values over Indiana where tornadoes were occurring.  Interestingly, 0-1 km EHI did a better job highlighting northeast Illinois and central Indiana for tornado potential than STP in this case.  A key feature, of course, was the large 0-1 km storm-relative helicity (SRH) along and north of the front, also reflected in the EHI field.

042104spceh100.gif (21141 bytes)<SPC 00 UTC 0-1 km EHI      042104spclcl00.gif (27426 bytes)<SPC 00 UTC MLLCL
042104spccp300.gif (40996 bytes)<SPC 00 UTC 0-3 km CAPE  042104spclfc00.gif (33149 bytes)<SPC 00 UTC MLLFC

At 00 UTC above, notice the large 0-3 km CAPE and low LFC heights, reflective to some extent of the large amount of CAPE located unusually low in the profiles near the warm front.  LCL heights were also quite low, showing the very humid conditions near the front.  0-1 km EHI continued favorable as well in the same area. 

From the 23 UTC and 00 UTC data above, it can be seen that by late afternoon, environment characteristics had become more evident and favorable for tornadoes over northern Illinois and central Indiana.  However, the environment evolution was not simple to forecast from earlier in the day, and depended on careful attention to surface trends and modification of model analysis profiles, along with appropriate lifted parcel selections given the humid low-level environment.  In the aftermath of this case, new ideas and products toward making it easier for forecasters to detect and become aware of deceptive "small CAPE" environments that have increased potential for tornadoes might be desirable and useful.

Another case involving relatively small CAPE, an advancing warm front, and surprise tornadoes occurred on 5/21/01 in Michigan (see Rob Dale's case study at   A good example of a tornadic small CAPE environment without the involvement of a larger scale warm front can be found on 9/20/02 when a long track F3 tornado occurred in southwest and central Indiana.


To summarize, here are key points from the above discussion regarding the evolving environment on 4/20/04:

1)  Model forecasts using lowest 100 mb CAPE largely overlooked and misrepresented the instability in this event. 

2)  Surface-based or shallower mixed parcels worked better when assessing relevant CAPE from model-derived products..

3)  CAPE was located unusually low in the environment profiles for this event, suggesting the setting to be thermodynamically more favorable for tornadoes than total CAPE alone would indicate.

4)  The "low-profile" CAPE in this event appeared to be ideally co-located in the vertical with available shear, possibly increasing tornado potential.

5)  Monitoring of the rapidly advancing warm front was of key importance in anticipating rapid thermodynamic environment changes favorable for tornadoes.


Here are some ideas for future research work or test applications in anticipating small CAPE tornado cases:

-  Making available surface-based or lowest 50 mb lifted parcel products for assessing tornado potential in near-saturated low-level environments where mixing may be reduced.

-  Development of CAPE assessments utilizing level of maximum buoyancy and amount/percentage of CAPE below 500 mb to highlight "small CAPE" and "low-profile" CAPE environments that may have potential for tornadoes.

Development of CAPE/shear products that might highlight "small CAPE/large shear" environments favorably distributed in the vertical for tornadoes (e.g., parameters combining low-level shear and CAPE below 3-5 km above ground?).


The 4/20/04 tornado outbreak was an event that most forecasters (including myself) did not anticipate well.  It certainly makes an interesting and informative case study.

Jon Davies  4/25/04

(Thanks to Daniel Nietfeld, NWS Omaha, and Andy Kula, NWS Des Moines, for their help obtaining data!)

back to Cases Studies page