NTDA and NMDA performed well with a strong QLCS as it tracked through central OK. Issued a SVR at 0254Z (Figure 2) as the QLCS was still in it’s developing stage. The NMDA confirmed what I was seeing in base data, in that several decent meso’s were embedded in the line. NMDA detection and tracking were consistent and persistent with the meso features prior to issuance (Figure 1), and increased my confidence that a severe thunderstorm warning was warranted.
NTDA picked up on a TDS signature embedded within the QLCS at 0229Z, with this feature persisting through at least the next hour. Confidence became high enough to issue a tornado warning at 0305Z as the circulation tightened with strong gate to gate shear evident in base SRM data. The algorithm tripped on the high side at 0250Z with probs of >80% for several scans, but no gate to gate shear appeared until 0305Z when the tor warning was issued. The NTDA maxed out with tor probs of 87% at 0310Z (Figure 3). A tornado was reported at 0313Z 3 miles NE Konawa, OK.
Figure 1. NMDA objects were persistent and consistent with detection and tracking of mesocyclones within a developing QLCS at 0233Z. Note detected duration of the southernmost mesocyclone is 30 minutes.Figure 2. Severe warned QLCS with embedded QLCS tornado warned storm. NTDA is overlaid.Figure 3. NTDA peaked at 87% at 0310Z.
A storm moving northeast through the southeastern corner of the OUN CWA began to take on QLCS-like characteristics (or perhaps hybrid supercell). This is at a fair distance from the radar (between 45 and 50 nm), and at a marginal viewing angle. The NMDA first showed a moderate strength meso on the 0403Z scan, followed by a NTDA detection with a 31% probability on the following scan. Both algorithms tracked this well, with the NTDA showing increasing tornado probabilities as a mesovort became more evident on radar. This would be a challenging warning situation due to the distance from the radar, marginal viewing angle, and hybrid nature of the storm. I suspect that in “real life”, both algorithms could have resulted in a faster lead time on the storm, or perhaps nudged someone to move from a SVR to TOR.
Initial NMDA moderate strength detection. NTDA trips with probs >30% on the next scan.A few scans later showing stronger mesovort.
Two intermingled supercells combined into a LEWPy QLCS as they moved out of Pottawatomie County OK into Seminole County OK. The southern mesocyclone became quite strong, with NTDA probabilities well into the 80% range, successfully drawing your eye to the feature (and tracking it fairly consistently in terms of geography).
KTLX 0.5° V with NTDA/NMDA detections at 0308Z. Note the purple triangle with the dominant circulation on the QLCS.
In the following 10-15 minutes, it appeared that a tornado developed, or at least a stronger/tighter circulation. This circulation then occluded, falling back one or two miles behind the gust front / updraft-downdraft convergence zone (UDCZ). It was noted that NTDA continued to show >50% probabilities, but was tracking the UDCZ instead of the likely-tornadic circulation. A similar behavior was seen in other recent cases, including with the end of the life cycle of the Montgomery County (Dayton) OH EF4. We looked at the plots of the azimuthal shear on this particular storm, and the highest AzShear values were indeed located with the UDCZ instead of the tight circulation. Both features are, of course, in very close proximity to each other.
KTLX 0.5° V with NTDA/NMDA detections at 0322Z. Note the red triangle showing the location of the NTDA detection on the UDCZ, while the occluded likely-tornadic circulation is just to the southwest of the red triangle.
In the next 10-15 minutes after that, after the occluded circulation washed out, the entire line became much more outflow-based — with few significant signs of wrapping up. I think NTDA handled this transition quite nicely. It continued to track a low-probability object through Hughes County just north of a slight surge in the line, which is great to be aware of, but correctly identified that this line did not appear particularly likely to be tornadic at this time. In the experiment, a decision was made to transition warnings to SVR for this QLCS segment at this time.
KTLX 0.5° V with NTDA/NMDA detections at 0347Z. Note the green triangle showing a low-end NTDA detection just north of a not-very-impressive surge in the line.
A supercell merger took place northwest of the Tulsa (INX) RDA. Mergers are always tricky; often there is no way to know whether it will be constructive or destructive in nature, but one should always “beware the merger!” when making warning decisions. In this case, the merger takes place, and a small area of enhanced rotation becomes evident on velocity. It would be tempting to issue a Tornado Warning in this case, but the weaker NMDA and NTDA output might give me pause to wait another scan or two. If I waited, the next few scans would show a weakening trend in the velocity fields. I suspect that if I had gone with a warning, I would quickly come to regret it.
The NMDA failed to be of use on a tornadic cell east of Des Moines. A strong meso was detected by radar, but it was above the tilts that the NMDA was sampling to generate its output, so it generated ‘weak’ strength of the meso. It should be noted that the cell was only 21 nm from the RDA at this time (2013Z).
Strong AzShear is present on mid level cuts of the KDMX radar (3.1 degrees and above). The NMDA was sampling below these levels however to generate its output, and therefore missed the strong rotation aloft.
In southern Jasper County, IA, a supercell is moving ESE. A small/tight circulation developed, associated with a tornado that was moving at a bit of a deviant motion (more like SE or even SSE).
NTDA correctly identified this tornado at first with ID 20, which had previously been used spuriously to identify some transient rotational features in the same general area. NTDA continued to briefly track this tornado as it moved toward the Marion County line, but suddenly, at 2049Z, ID 20 was erroneously re-assigned to another (weaker) circulation a few miles to the north. In fact, no NTDA detection continued for the tornadic circulation entering Marion County at all.
Ideally, you would want to continue to use the same ID to track the tornado until it dissipated, and perhaps assign a new NTDA element to the other circulation further north. I might speculate that the tornado’s deviant motion, compared to the overall storm-scale motion, might have had something to do with this issue.
KDMX 0.5° V and NTDA detections at 2046Z. Note that NTDA ID 20 is correctly located with an obvious tornadic circulation.KDMX 0.5° V and NTDA detections at 2049Z. Note that NTDA ID 20 has suddenly re-located further north (with an erroneous speed/direction) and the tornadic circulation is no longer covered by any NTDA detections.
–Insolation
2136Z
A tornadic supercell in Marshall County IA has developed a tight, intense couplet on KDMX radar. The NTDA algorithm has performed fairly well from a qualitative sense, but some of the underlying numbers are not exactly as expected for such an obvious tornadic signature. The first image is from the 2132Z scan, which was the first scan to show a tight gate-to-gate tornadic couplet.
KDMX 0.5° V and NTDA detections at 2132Z.
It is notable that despite high/extreme categorizations on several variables, the probability is only about 35%.
This changes significantly just one minute later.
KDMX 0.5° V and NTDA detections at 2133Z.
At 2133Z, the probability has increased to about 95%. In the opinion of our group, this storm absolutely warrants very high probabilities, so the 2133Z NTDA detection appears to be the better of the two.
The big question here is why the NTDA changed so much from 2132Z to 2133Z, because the underlying statistics (as displayed in the sampled readout) do not change significantly from one minute to the next. We were very curious as to why small changes in these calculations led to such a big change (yet a very good change) in the tornado probability.
Another item of note from this example is that the Abs Vel Max and Max-Min Vel Diff statistics do not seem to be capturing the highest values of velocity when sampled by hand bin-by-bin. Using the 2133Z time step as an example, there is a 69.0 knot outbound velocity, but the Abs Vel Max only lists a maximum value of 47.6 knots. The Max-Min Vel Diff (Vr) calculations are similarly affected.
Early on in convection, funnel clouds and tornadoes were being reported in small storms with virtually no rotation on radar. Once there was sufficient rotation on radar, it appeared that funnels/tornadoes were being reported with NTDA probabilities as little as 30%. Normally I would take notice, but not warn, on such low numbers. Given the earlier storm reports, when an NTDA marker first appeared for the Marshall County cell I had a feeling that it would eventually produce a funnel or tornado. The first detections were very low, only about 20%, first around 2059Z and then again around 2109Z. By 2110Z it had jumped to 30% with a hook evident on radar and rotation on the few lowest tilts. It was at this point that I would have warned on this storm. The radar data may have been enough alone at this point, but the first detection at 2059Z definitely drew my eye to this area and I knew it was something I needed to keep an eye on, despite being distracted by the much larger supercell in Marion County.
By 2113Z the NTDA probability had increased again to almost 44%, and at 2116Z it was up to 56%. A funnel cloud was ported at 2117Z, with tornado reports following around 2122 and 2130Z.
-At the beginning of the event, while the TDA didn’t trigger high values (as tornadoes were being observed), it still ultimately added a lot of value. As a forecaster walking into that event, I probably would’ve thought that the chance for tornadoes as the cells were initially going up was around zero, so having guidance that showed that “hey look, something’s here” provided a lot of value added. So while the verification stats might not look great for this phase of the event, it potentially added more value than during sure-fire high end tornado warning scenarios. In those scenarios, the post-processed stats for the algorithms would look great, but they might not add as much value, since you’d have warnings out either way.
-It was also encouraging to see the high probs from both algorithms later in the event with the strongest supercell. As was brought up in discussions in our group, in that case those super high probs in both algorithms might have given you confidence to consider a higher tag on the warning (potentially the considerable tag).
At this time, the northern supercell (moving through northern Marshall County) is beginning to rapidly strengthen. At the same time, the storm near Pella is extremely strong and likely doing significant damage, and could be taking the bulk of the attention of the forecasters. Here, both the NMDA and NTDA have correctly identified the strengthening of the northern storm. I think the algorithms could have been helpful in bringing attention back to the Marshall County supercell before it became any stronger. This situation (two storms, one very strong and damaging while the other is rapidly strengthening) could be a “high risk” situation for loss of SA. This is especially the case on a day like this, when the overall situation was not well anticipated, and officing staffing is likely not at the level it would ideally be for such a set up.
At this point the tornado is moving through the Pella area. There’s not much of a yes/no warning decision to be made here…it’s obvious that a tornado is ongoing. However, I think it’s noteworthy to see both algorithms outputting extreme values. This could aid forecaster confidence in “upgrading” a warning to the “Considerable” threat tag, or, depending on reports, a “Tornado Emergency”. While we have several ways to ascertain tornado strength in real-time (rotational velocity, TDS height, etc.), having an extra nudger is very useful. It can also be challenging to know when to discontinue the higher end wording in the warning; seeing the trends from these algorithms will be valuable data points.
