Proceedings, 20th Conference on Severe Local Storms
Orlando, FL, 9/2000
April 19, 1996 study
Convective modeling group



Brian F. Jewetta*, Bruce D. Leeb and Robert B. Wilhelmsona

aDepartment of Atmospheric Sciences and
National Center for Supercomputing Applications (NCSA)
University of Illinois at Urbana-Champaign

bDept. of Earth Sciences, University of Northern Colorado


On 19 April 1996 an outbreak of tornadic storms occurred in central and northern Illinois and nearby regions of neighboring states. The total of 36 tornadoes in Illinois set a single day record and exceeded the number expected in an average year. The storms formed along both a surface drytrough and warm front. The convection along the drytrough was observed, after a period of splitting and merging, to move off the boundary into central Illinois and become tornadic. In Illinois, 31 counties sustained damage. One fatality, and $30 million in damage was reported (Storm Data, April 1996).

Several aspects of this case are of interest. The presence of many supercells with a large number of tornadoes was significant, and six, or 20%, of the tornadoes were rated F2 or F3. Multiple surface boundaries were active regions for convective initiation, while cell movement off the pressure trough suggested the influence of an upper-level front. Early stages of organization were marked by cell interaction and consolidation, resulting in 9-12 supercells later (see Lee et al. 2000, this volume). Finally, the sensitivity of the simulated convection to resolution, boundary layer physics and cumulus parameterization will be discussed.


We are modeling the April 19 cyclone and squall line with the NCAR/Penn. State MM5 model (Grell et al., 1995). Sixty-six vertical layers were used, with up to five grids (four nests) employed (Fig. 1). The horizontal resolution ranged from 81 km (outer grid) to 1 km, with the innermost grid movable. Simple ice microphysics, Blackadar boundary layer physics, a 5-layer soil model and 3-km resolution were used for most results presented here. The Grell cumulus parameterization was employed on the 27 and 81-km grids, with explicit precipitation (only) for resolutions of 9, 3 and (when applicable) 1-km. Initial conditions were provided on both 81- and 27-km domains. Data assimilation was deemed necessary to adequately represent the pre-convective environment. Analysis nudging was applied to the outermost grid for the entire integration and to the 27-km domain for the first 12 hours. Simulations were begun at 0000 UTC 19 April 1996, and run up to 27 hours. In all cases, three domains (up to 9 km resolution) were used at the initial time, with the fourth domain started at 1800 UTC and the fifth at 2000 UTC, when used. In the discussion that follows, all times are UTC.

Fig. 1. Left: Grid layout used in this study. Right: 00z 20 April midwest composite radar reflectivity. Click on either image for full-sized version.


On April 19, a 300 hPa jet streak (not shown) was located over eastern KS and southwest MO. The left exit region of this jet streak was over northern MO and western IL much of the day. In response, surface pressure falls exceeding 4 hPa (3hr)-1 moved from northwest MO to central IL between 1500 and 2100. These pressure falls helped back and strengthen the surface winds in IL, increasing the shear available to the afternoon convection.

At 2100, the surface cyclone was located over southwest IA, with a warm front extending eastward into western and central IL. The NCEP surface analysis (not shown) indicated a north-south cold front over MO. This was likely the occlusion of a drytrough (pressure trough and dryline - Martin et al., 1995) and cold front. Time series data indicated that a slight windshift and drying preceded the arrival of the pressure trough. This is consistent with a cold front aloft slightly ahead (east) of the surface occluded front. In eastern MO and western IL, dewpoint temperatures of 16-18deg.C characterized the moisture return ahead of the windshift. Convective initiation (first radar echo) in eastern MO occurred near 2020, with the first severe thunderstorm warning issued at 2115. By 0000 that evening (Fig. 1), severe storms were found along the warm front in northern IL, isolated in central IL, and in a broken squall line extending from southern IL into southern MO and on to TX. Tornado reports were most numerous in central and northern IL.


MM5 was initialized with data from the previous evening, 0000 19 April. Three grids (9 km inner resolution) were used for the first 18 hours, with a fourth (3 km) thereafter. Convective initiation in eastern MO, in terms of modeled surface rainwater, was shortly before 2200. Straight hodographs characterized the model wind fields there.

A snapshot of the early convection is shown in Fig. 2. Cells at this time were organized northwest-southeast in eastern MO. The paths taken by individual storms are also indicated. Most initially tracked northeast, paralleling the 700 hPa flow. After initiation, several cells turned right, as expected with the development of rotation. Splitting and merging was also noted. The cells labeled "A" in Fig. 2 merged and turned almost due east. Cell "B" split, with the left mover propagating north-northeast and the right-moving cell traveling due east ahead of the merged cell "A". These modeled storms were long-lived cells which later propagated into central IL north of the squall line and south of the warm frontal convection (Fig. 2), similar to observed storms on this day.

Compared to observations, convective initiation was delayed, warm frontal convective initiation followed that on the drytrough (rather than simultaneously), and additional storms developed in the model north of the central IL storms. Nonetheless, many of the observed morphological characteristics have been reproduced. We note that the number of initial convective cells is, as expected, strongly a function of model resolution. In 1-km simulations to date (Fig. 2), nearly 4 times the number of convective cells are seen. This represents more opportunities for cell interaction, whether leading to long-lived sustained convection (as in cells "A") or not.

The mesoscale convective system propagated ahead of the surface drytrough/cold front in the model and observations (see 0000 radar in Fig. 1, and MM5 rainwater at 0200 inset in Fig. 2). One possible explanation for movement off the surface boundary is the influence of a cold front aloft (CFA). The CFA has been shown to be present in some cases when precipitation is far removed from surface boundaries (Hobbs et al. 1990, 1996; Locatelli et al. 1997). Convective initiation has also been attributed to the coincidence of the CFA over the surface boundary location (e.g. Locatelli et al. 1998).

Fig. 2. Left: 3-km model surface rainwater after 22.2 hours, with cell tracks (from 22.2-24.2 hours). Inset: rainwater at 26 hours. Right: surface rainwater at 22.2 hours for case with 1-km resolution.
Click on image for full-sized version, or here for color.

Fig. 3. Left: Cross section of model thetae (every 2K) after 26 hours. Bold line outlines area of precipitation. Right: 600 hPa wind vectors and thetae (every 1K), with surface boundaries superimposed.
Click on image for full-sized version.

Cold advection was indicated in 700-hPa observations (not shown) on the evening of April 19. We have examined the simulation results for evidence of a CFA and its role in convective initiation. The 26-hour fields (valid 0200 April 20) are shown in Fig. 3. Convection is discernible in the 600 hPa thetae field (right half of figure), ahead of the occluded drytrough/frontal boundary. The back edge of the precipitation parallels the thetae contours. In the cross section, the western extent of the precipitation is near the leading edge of the 600 hPa drying/cooling.

While the association at later times seems clear, the role of the CFA in initiation is less so. Analyses at earlier times (not shown) indicate that convective initiation in MM5 occurred in the axis of maximum thetae and maximum low-level moisture convergence along the drytrough. While slight mid-level cooling might not trigger convection, the combination of surface heating and convergence could do so. However, it is notable that cells developed immediately after passage (SW to NE) of the 600 hPa thetae ridge, and thus the onset of cooling and drying. Further diagnosis is needed to determine if, e.g., CFA-enhanced surface convergence (Locatelli et al. 1997) was important in the initiation of convection.


Fig. 2 makes clear the sensitivity of the simulated cell formation to model resolution. Cloud formation was explicit, rather than parameterized, in the region of interest. It is known that resolution coarser than 4 km is inadequate for explicit modeling of deep convection, particularly in higher shear environments (Weisman et al. 1991). At the same time, cumulus parameterization schemes are not designed for, and are of questionable utility, at scales under 10 km (e.g. Wang and Seaman, 1997). Thus models may perform poorly (in the warm season) at a resolution of 5-10 km, a range likely of interest to operational prediction centers. While we do see reasonable convective behavior at 3 km, the resolution is barely adequate to capture the details of cell formation and thus subsequent splitting and merging. From the two panels in Fig. 2, there are clearly fewer opportunities for cell merging at 3 km. Our future work will include mapping cell morphology at 1 km and determining what is "lost" at coarser resolution beyond the number of initial cells.

Fig. 4 demonstrates further the uncertainties in modeling convection. All fields are valid at 2330, an early stage in the model convective evolution (which lagged observed evolution by up to 2 hours). Significant sensitivities were found to the choice of boundary layer parameterization. The Gayno-Seaman and MRF schemes resulted in convective development in central Illinois, while cells failed to form on the drytrough.

Experiments were also performed with parameterized convection at 9 km resolution. With the Blackadar PBL, the broad precipitation coverage in western IL was similar to explicit simulations at 3 km. Convection with the Kain-Fritsch scheme was bowed eastward, compared to that produced using the Grell parameterization. From a predictability point of view (or use of different model physics as part of an ensemble modeling study), there is significant sensitivity to the physical parameterizations as applied to modeling deep convection.


We have studied convective initiation and evolution on April 19, 1996 using MM5. The general precipitation coverage and surface boundary movement compare favorably to observations. Convective cell splitting and merging was also modeled, and observed on this day. The details of cell formation were quite sensitive to model PBL parameterization (for location of initiation) and horizontal resolution (for number of initial convective cells).

Fig. 4. Precipitation at 23.5 hours, for various PBL parameterizations and (bottom right) cumulus parameterization schemes.
Click on image for full-sized version.
One case, however, was characterized by convective formation along the drytrough in eastern MO and resulted in long-lived right-moving cells propagating through central IL. Convective initiation in the model appears coincident with the arrival and collocation of a cold front aloft over the surface drytrough.

We plan to analyze the interaction between the cold front, the drytrough, the CFA and subsequent cell formation. 1-km simulations will be analyzed and compared to the 3-km results. The pattern of splitting and merging will be further studied with the aid of idealized simulations with the COMMAS model, utilizing soundings taken from the MM5 simulations. The idealized studies are intended to study the cell interaction in a systematic way and determine the role of merging in cell organization, rotation and longevity. Results will be presented at the conference.


Simulations were carried out on the NCSA CRAY Origin2000. Additional computing and other support was provided by NCSA. This work was supported by the National Science Foundation under grants ATM 96-33228 and ATM 99-86672.


Grell, G. A., J. Dudhia, and D. R. Stauffer, 1995: A description of the fifth-generation Penn State/NCAR Mesoscale Model (MM5). NCAR Technical Note TN-398+STR, 122 pp.

Hobbs, P.V., J.D. Locatelli, and J.E. Martin, 1996: A new conceptual model for cyclones generated in the lee of the Rocky Mountains. Bull. Amer. Meteor. Soc., 77, 1169-1178.

Lee, B.D., B.F. Jewett and R.B. Wilhelmson, 2000: Supercell differentiation and organization for the 19 April 1996 Illinois Tornado Outbreak. P6.13, this volume.

Locatelli, J.D., M.T.Stoelinga, R.D.Schwartz, P.V.Hobbs, 1997: Surface convergence induced by cold fronts aloft and prefrontal surges. Mon. Wea. Rev., 125, 2808-2820.

Locatelli, J.D., M.T. Stoelinga, P.V. Hobbs, 1998: Structure and evolution of winter cyclones in the central United States and their effects on the distribution of precipitation. Part V: Thermodynamic and dual-Doppler radar analysis of a squall line associated with a cold front aloft. Mon. Wea. Rev., 126, 860-875.

Martin, J.E., J.D. Locatelli, P.V. Hobbs, P.-Y. Wang, and J.A. Castle, 1995: Structure and evolution of winter cyclones in the central United States and their effects on the distribution of precipitation. Part I: A synoptic-scale rainband associated with a dryline and lee trough. Mon. Wea. Rev., 123, 241-264.

National Climatic Data Center, 1997: Storm Data, 38, No. 4 (4/96).

Wang, W., and N.L. Seaman, 1997: A comparison study of convective parameterization schemes in a mesoscale model. Mon. Wea. Rev., 125, 252-278.

Weisman, M.L., J.B. Klemp, and W.C. Skamarock, 1991: The resolution-dependence of explicitly-modeled convection. Preprints, 9th Conf.. on Numerical Weather Prediction, Denver, American Meteorological Society, Boston, MA, 38-41.

*Corresponding author address: Dr. Brian F. Jewett, 216 Atmos Sci Bldg, 105 S. Gregory St., Urbana, IL 61801. Email:


More information: April 19, 1996 study / observations / simulations / Convective modeling group / Publications