Herein we present prelimary results from our numerical study of thunderstorm cell interaction. Much of what is shown below was presented at the Severe Local Storms conference in August, 2002 in San Antonio, Texas.
Interaction between convective cells may alter cell:
|
The convective cells in question could be:
|
Cell interaction could take the form of one cell:
Given a line of (distinct) cells, the orientation of the convective line to the vertical shear has been shown to be highly important (Bluestein and Weisman, 2000).
We are examining the behavior of cells forming in close proximity to one another. No pre-existing boundary is specified, and only a pair of convective cells is considered (rather than a line). This allows a range of orientations of the pair of developing storms. Splitting always occurred, followed in some cases by merging.
This work was motivated by observations of storm behavior in the 1996 tornado outbreak in Illinois. A brief description of that event, and our modeling results to date, follow.
| Fig. 1 |
Fig. 2 |
Fig. 3 |
Fig. 4 |
On 19 April 1996, 36 tornadoes struck parts of Illinois and nearby states (Fig. 1). Chase crews from the University of Illinois Dept. of Atmospheric Sciences witnessed and photographed one of the supercells as it produced tornadoes along interstate highway I-72 from western to eastern IL.
Later examination of the radar data from this case by Bruce Lee (now an associate professor at the University of Northern Colorado in Greeley) revealed a pattern of cell splitting and merging. While tornadic storms formed both along the dryline/drytrough and warm front, the drytrough storms were particularly notable for splitting and subsequent merging (Fig. 2) early in the life of the MCS. As the squall line moved eastward, more merging events were noted, with tornadoes occurring immediately after cell mergers in some cases.
The evolution of one set of the tornadic storms along the drytrough is shown in Fig. 3, above. The cells merged, becoming storms "A" and "B" in the figure. These storms went on to become tornadic (Fig. 4); the southern cell ("B") struck Jacksonville, Springfield, Decatur, Champaign and Ogden, IL, and was photographed and filmed by U.I. Atmospheric Sciences crews.
Several questions immediately arose after the event and subsequent radar interrogation. One foremost in our minds: was splitting and merging necessary for tornadogenesis? Another question, following the recent work of Bluestein and Weisman (2000): shouldn't the developing squall line, the orientation of which was nearly perpendicular to the shear, have decayed, perhaps with the exception of end cells? We are carrying out numerical studies of this event to help understand cell morphology and the role of splitting and merging, in this and more general cases.
Fig. 5 |
Fig. 6 |
Fig. 7 |
This work follows and is a companion to prior work in which we simulated the April 1996 event. In that (real-data) study, the parent cyclone and convection initiation and evolution were modeled with the MM5 model initialized with data from 00z the previous evening (Fig. 5; Jewett et al. 2000). In the current work, we have used results from the MM5 work to initialize the WRF model to carry out idealized simulations of thunderstorm cell interaction.
Details of the current experiment:
Cell interaction studies were carried out as follows. Two warm thermals were used to initiate two storm cells. The orientation and spacing of these two thermals was varied and the subsequent storm properties catalogued in terms of the original cell placement. In each case, the "primary" 3K thermal was placed at the center of the domain (Fig. 7), while the "secondary" 2K thermal was placed at one of the boxed locations in the figure - within a +/- 15 km subdomain centered on the primary thermal. Note that the overall domain size was much larger - 90x90 km. The mesh of secondary cell positions defined a total of 232 model runs. Two additional "control" runs, each with a single 3K or 2K thermal, were made for comparison with the 2-cell cases.
In Fig. 7, the R=7.5 km circle showed the region over which the two initial thermals overlapped. Examination of the early model rainwater structure showed that cells 10 km apart were distinct, while those 7.5 km or less apart were progressively less distinct.
Carrying out and analyzing a large number of storm simulations can be prohibitively time-intensive. As a result we have automated/scripted most of the tasks required in making such a large number of runs. Once the initial 2-cell location mesh (Fig. 7) was defined, a job management system handled submission of new simulations as others finished.
Data was saved at 1-minute intervals. Statistics gathered each minute included:
In addition, surface vorticity maxima were tracked in time. From this data, the vorticity duration was computed. This was a measure of how long each vorticity max exceeded one of three threshholds: 0.010, 0.015, 0.020 s-1. This information was of interest because of the abundance of short-lived (but strong) tornadoes in the April 19, 1996 case.
Fig. 8 |
Fig. 9 |
Fig. 10 |
The three figures above summarize the overall characteristics of our simulations. Fig. 8 shows the peak surface vorticity (top, x10-4 s-1) and 0.015 s-1 vorticity duration (bottom, minutes) for all simulations, sorted in increasing order. For comparison, the 3K single-cell control case values appear with a bold horizontal line on each scale. The control is in the upper 90% of each figure, indicating that cell interaction as configured in this study was most often destructive. Only a narrow range of cell orientation and spacing led to more intense rotation than occurred with only a single cell. This is not inconsistent with the study by Bluestein and Weisman (2000), who found that most cells in a line were weakened by interaction with their neighbors, while cells at the end of a line were more likely to be strong.
Fig. 9 shows the lack of any clear relationship between surface rotation and peak updraft speed. Evidently all cases were strong (~60 m s-1 updrafts) and exhibited significant rotation at some time. There was no clear association between stronger surface rotation and updraft strength.
Fig. 10 shows the relationship between peak surface vorticity and 0.015 s-1 vorticity duration. The clustering near duration times of T=70 min is attributed to the simulation cutoff time of 2.5 hours. There is a wide range of vorticity duration and peak surface intensity, again with no clear relationship. Significantly, this suggests that cells with long-lived rotation need not be those with the strongest rotation, and that some cases of strong rotation are short-lived. The latter is of interest given strong but short-lived tornado events on 19 April 1996. Because these statistics apply to the entire (horizontal) domain and length of each simulation, further interrogation is needed to understand the cell morphology resulting in the statistics seen here.
Fig. 11: Max Vorticity |
Fig. 12 |
Fig. 13: Vort. Duration |
Fig. 14 |
Fig. 15: Vert. Velocity |
The run characteristics summarized in Figs. 8-10 were then categorized by initial cell orientation and spacing. This allowed us to construct horizontal contour maps of max surface vorticity, duration and vertical velocity (Figs. 11-13). In each case, the figure domain represents the cell placement mesh seen earlier in Fig. 7, rather than any actual 90x90 km simulation domain.
The peak surface vorticity as a function of initial cell placement is seen in Fig. 11. One clear asymmetry is evident from southwest-to-northeast. Secondary cells initially southwest of the primary cell result in mature storms with greater surface rotation, while those initially northeast of the primary storm result in weaker rotation. The reasons for this are under investigation, but the southwest-vs.-northeast surface vorticity traces are shown in Fig. 12 for secondary storms of similar distance but opposite (SW, NE) orientation. Both have significant rotation, but the SW case is clearly stronger. The contour enclosing the southwest maximum contains six simulations. Snapshots of SW and NE cell evolution are available here.
The peak surface vorticity also shows that cells forming in close proximity (within the R=7.5 km, say) are characterized by stronger rotation when mature. This result is more problematic since overlapping initial thermals can, at the limit of R=0, consist of a stronger initial cell (i.e. the initialization may be artificially biasing the result). Also, the 1-km horizontal grid spacing limits our confidence in this result, but it warrants further investigation.
The positive "anomaly" of stronger rotation transitions to a local minimum for secondary cells farther east, or more nearly south of the primary cell. This may reflect the results also seen in Bluestein and Weisman (2000), where cells representative of a convective line nearly normal to the shear experience more destructive interaction as split cell pairs move laterally apart, colliding with their neighbors. Further displacing the secondary cell to the southwest - representing line orientation more nearly 45 degrees to the shear - results in stronger rotation. However, the details in Fig. 11 suggest a more complicated interaction is going on.
Finally, we note the vertical velocity statistics. Note that the contour interval, as noted in the Preprints, was chosen carefully; the vertical velocity values differed very little between the maxima and minima shown in Fig. 15. Clearly, initial cells placed in an east-west orientation had slightly less intense updrafts, compared to those in a north-south orientation. With the nearly straight hodograph, the low-level storm-relative winds were primarily easterly, so we expect that in an E-W orientation the eastern cell would interrupt the inflow to the western one. Beyond that, no conclusions have yet been drawn. Fig. 9 makes clear that a stronger (peak) updraft is no guarantee of greater rotation in the storm, for our case.