Lower Calculated Probability in Higher Confidence Scenario

This was an interesting find during the Archived Case from ILN. Focus was on western portions of Indiana/eastern Ohio at several areas of rotation associated with a multicellular cluster of convection. A tool commonly used to interrogate a potentially tornadic storm’s rotational velocity is the VRshear tool. This tool calculates the Vrot, line distance (end of points), shear (in ms-1) and distance from the RDA. WDTD mentions from their Impact Based Warnings Guidance of around 30kt of rotational velocity being the initial supercell tornado warning threshold. Of course, this has some gray area but is generally expressed as an approximation.

At 0207Z from KILN, a cell was identified to show increasing cyclonic rotation. Vrot was calculated to be at 37kt estimated. NDTA probabilities began relatively low at 21.48% which at this early indication, a warning would have already been required. The next scan at 0208Z showed an increase of Vrot to 38kt, with NTDA probability increasing to 24.85%. One more minute followed at 0209Z showed an obvious increase in Vrot at 44.7kt with NTDA Probability of 32.41%.

KILN 0.5º BR at 0207Z. Distance from RDA: 55nm.
KILN 0.5º BR at 0208Z. Distance from RDA: 55nm.
KILN 0.5º BR at 0208Z. Distance from RDA: 54nm.

This is one of several examples where lowering the current default threshold of 30% was necessary. While tracking of both NMDA and NDTA was very accurate, I was surprised at the lower probabilities. A few calculated parameters were identified to be a potential hindrance at a higher probabilistic value, including high SW (and trending higher) and AzShear Max value actually showing a temporary lowering trend at 0207Z. An interesting situation where having to lower the probabilistic threshold was necessary, and was a bit lower than what was expected but still showed consistency with tracking.

– Dusty Davis

Example of High NTDA Prob in Low Confidence Case

In this case the NTDA was able to detect strong azShear in a messy reflectivity signature. It was difficult to detect any reflectivity signatures with a cell that was approaching New Madison due to a cell merger. There was also evidence that the cell was weakening from 1:30 Z to 2:00Z. Cloud tops were warming on IR, the reflectivity core was falling and lightning count and VII were also much lower at 1:45Z than 1:30Z. But the NTDA picked up on a quickly strengthening velocity couplet. The 0.5° beam was also about 5kft agl which lowered forecaster confidence in a tornado warning.

At 1:55Z NTDA probs were at 20% and quickly strengthened to around 50% at 2Z then peaked out at 84% at 202Z. Looking at the velocity data at 201Z I may have assumed that there was a daliasing issue since there was such a rapid increase in the outbound velocities at 0.5°. Had I not already had a warning in place from when the storm was stronger I’m not sure I would have issued one for this couplet. But we had an LSR at 205Z near New Madison with a tornado possible tag. The NTDA was very useful to pick up on this signature when most of the other data showed the storm weakening overall and the reflectivity was not very helpful.

There were two other interesting features of the NTDA for this case. At 202Z the azShear, divShear and Abs Velocity max were all decreasing from the previous scan but the probability actually went up from 72% to 84%. Another interesting thing was how quickly the probability decreased after the peak of 84%. 2 min after the peak, the prob dropped to 3.3%. I was not expecting to see probability drop this rapidly. Albeit the velocity signature was much weaker.

0205 Z

– Gerry Bertier

NMDA aids in detecting 1” hail :20 before occurrence

In the archived ILN event, a LSR of 1-inch hail piqued my interest and I wanted to see how the NMDA product handled it.  I used the ILN WSR-88D and the storm at that time was 66NM NW of the Radar. The NMDA product was consistent for the 20 minutes prior to the storm report and I would have felt confident using the product (along with interrogating other base products) to effectively warn/issue Severe TS for this storm.

Wayne County, IN storm producing 1” hail


NTDA Threshold vs QC Suppression

A variety of non-meteorological returns seem to trigger the NTDA. In the attached image there are 3 NTDA icons within ~25 miles from the RDA. All have “Prob” values less than 30%. There were other examples from earlier in the day where bad radials and elevated roadways caused detections. All of these were also less than 30% prob. So I can see why the default threshold for the NTDA is >= 30%. I was assuming that the probability from the NTDA represents the actual probability of a tornado occurring based on the training cases. To me I would want the NTDA to pick up on anything that would be even a slight potential for a tornado. That way I could use it as SA or a first alert. I would be concerned if the NTDA was predicting a 22.7% probability of a tornado, like in the screen capture below. That to me means similar detections from the training cases produced a tornado about 1 out of 5 times. To see a non-meteorological return produce that high of values would lower my confidence in real returns. Rather than setting the threshold at >=30% I think that filtering out the bad detections  would make me trust the NTDA more.

Three NDTA detections from near the KLBB radar on non-meteorological returns.

– Gerry Bertier

Tags: None

Mon Archived Case Blog

0140Z – Nice Heads Up on New Updraft

The NMDA gave me a nice heads up to the new updraft that was forming on SE flank and already showing signs of rotation. At first I thought it was a “false alarm”, but upon further inspection, I could see new rotating updraft on higher tilts.

0.5 deg tilt
4.0 tilt

It looks like this updraft ultimately played pivotal role (or was at least in the area of) new tornadogenesis about 15-20 min later.



0220Z – Two NTDAs in Close Proximity

ID20 was the main object. Lat/Lon is 32.491/-96.916
ID33 came along second and is at 32.514/-96.889
This is about 4km apart

These seem a little close together and could clutter up display. Also, these didn’t seem to be two separate areas of low level rotation, but both part of the parent circulation.


Blog Post for Thursday PM Real-Time Event (ILX/IND)


A line segment coming into the far NW corner of the Indianapolis CWA (Warren and Fountain counties) has shown consistent weak detections on both the NTDA/NMDA algorithms. While not terribly impressive on reflectivity, the cell at the northern end of this line does show signs of a rear inflow jet on velocity data, and the NTDA/NMDA detections have been on the northern gradient of this area of slightly enhanced inbound flow. The 0217Z scan has shown the highest NTDA probability so far, with 42%.

KIND 0.5° V and NTDA/NMDA detections at 2017Z in Fountain County, Indiana.

The lack of a tight circulation, or any higher wind speeds (abs vel max of only 30 knots) is likely keeping this feature from receiving a higher probability. Given the mesoscale environment and the way this storm is behaving, this seems reasonable, as a tornado does not appear imminently likely. With that said, it is definitely good performance by the algorithms to be continuously and (mostly) consistently tracking this feature on its way inbound into Indiana.





The line segment moving ESE through the Indianapolis CWA has become a mature outflow-dominant feature. Given the environment, which is characterized by somewhat dry boundary layer conditions, the overall tornado potential is not expected to be very high. With that said, this line segment has occasionally had areas where there have been small-scale surges, in addition to a fairly persistent area of weak/broad rotation at the northern end of the segment. The NMDA/NTDA algorithms have done a very good job tracking these features as they appear and disappear, with generally low probabilities (no greater than 40%) which seem reasonable for the environment.

KIND 0.5° V and NTDA/NMDA detections at 2059Z in Montgomery County, Indiana. This detection (orange on both NTDA/NMDA) seems to be a reasonable depiction based on the longevity of the feature and its position with respect to the gust front / precipitation.
KIND 0.5° V and NTDA/NMDA detections at 2117Z in Montgomery County, Indiana. A small nub / surge has developed along the outflow / UDCZ, and has been identified by both algorithms with lower-end probabilities (NTDA) and AzShear values (NMDA).





The convection over the IND forecast area has weakened significantly, and the velocity fields are characterized by a long arc of outflow with no strong values. It is likely that this gust front is producing marginally strong winds (40 MPH) based on radar appearance.

Nonetheless, there has been one consistent kink on the line, which has persisted since the 2117Z image posted earlier in this blog. While this feature is very weak, and certainly not worthy of a tornado warning, both algorithms have fairly impressively continued to track it with low-end detections.

KIND 0.5° V and NTDA/NMDA detections at 2201Z in Boone County, Indiana. The velocity signature here is very weak, but there is a very slight increase in inbounds/outbounds at this portion of the line. This feature has persisted for almost an hour, and has tracked well with both algorithms.


A good no-call for the NMDA

Here is a storm with respectable storm top divergence shown to the left (HPX at 2105 UTC today). The right 2 panels show evidence of weak 2D shear if anything else. Glad to see there is no NMDA detection on this. However, at 2106 UTC there was a very weak detection with this at VWX that lasted 2 scans then disappeared.


Transition from Balanced to Outflow Dominant Detections

We got to do a real-time case this afternoon. A fairly classic cold core aloft with steep lapse rates, a deep mixed layer and dewpoint depressions of 15F or greater. Based on cloud heights in the surface observations we’re probably looking at LCLs on the order of 5000-6000 ft AGL. Not terribly favorable for tornadoes, but perhaps a chance for the NMDA to shine before thermodynamics takes over with very strong downdrafts that might congeal into a cold pool.

At the left is coincident NTDA and NMDA detections coincident with a respectable kink in the convergence line. The 0.5 degree cut corresponds to an approximate height of just over 2000 ft AGL. The NMDA is triggered off the 0.5 to 1.3 degree cuts, even though there is still a weak couplet above this at 2.4 degrees (that’s probably just a bit below the AzShear threshold).

The right column is much later when the outflow has begun to outpace the convection. As seen in other cases, we see a lot more NMDA and NTDA detections overall, but the sliders are very helpful by allowing us to filter out the false detections (lower 20s percent chance threshold in this case for NTDA).

It should be noted that the ProbTor showing zero percent during this time and I think this has a lot to do with its awareness of the unfavorable thermodynamic environment.



Another small supercell formed along the line moving through central Illinois, taking on characteristics very similar to the storm which produced a tornado to the north earlier in the day.  As of this writing, the SPC storm report page indicates two storm reports with the new storm; a tornado near Altamont and another tornado near Mason.  Both are located in Effingham County, IL.

The NTDA began showing low probability detections about the time of the first tornado, and then continued to track the storm with increasing tornado probabilities as it moved to the southeast. Probabilities maxed out around 40% around the time of the second tornado.  The NMDA also showed several detections through this time.

This performance is noteworthy because the quality of the radar data wasn’t great.  The storm was located around 60 nm away from the radar (KVWX–Evansville), with the radar having to shoot through a hail and rain core.  Thus, TBSS and slide lobe contamination were an issue.  Given that forecasters were aware of the tornado potential from the earlier storm, it’s very possible that these directions could have resulted in a Warning being issued when it otherwise would not have.

– Rabbit

Marginal Severe Hail with NMDA/NTDA

A stronger cell began to move into Jefferson County, IL around 2044Z. There were some indications of hail on the base data, including a core of 70 dBZ and a slight drop in CC to 94%. Based on the lat/lon of the Local Storm Report, the 1” diameter hail would have occurred around the time of the two 0.5 degree scans at 2100Z and 2106Z (though the time of the LSR stated 2113Z). While the NMDA was not present prior to the storm passing over the hail report location, it was an indication of a stronger updraft and would have been a nudge to issue a SVR as I was still on the fence looking at the base data alone.

4 panel of 0.5 deg Z, ZDR, KDP, and CC at 2100Z. Point J is the location of the hail report
4 panel of 0.5 deg Z, ZDR, KDP, and CC at 2106Z. Point J is the location of the hail report. Weak NMDA present.
Loop of lowest 3 tilts of SRM, 0.5 Z, and NMDA, NTDA as the storm passed through Jefferson County

– Angelica Schuyler