Strong tornadoes in the Deep South

A seasonally strong shortwave trough tapped into abundant Gulf of Mexico moisture forcing severe and tornadic storms across Louisiana, Mississippi, and Alabama. The Storm Prediction Center issued a Moderate Risk outlook with tornadoes and strong wind gusts being the primary threats (see Figure 1).

Fig. 1: SPC outlook with 06Z 12/17 verification (dots).

ProbSevere/ProbTor models show the evolution of storms throughout the afternoon (Figure 2). The outer contours represent the ProbTor probability and are only present if ProbTor is ≥ 15%, here.

This environment was characterized by 1000 - 2500 J/kg of MLCAPE, 45-60 kts of effective bulk shear, and 1-3km AGL mean wind of 40-55 kts. The ProbSevere models generally track and discern the most dangerous threats well (see the NWS warning polygons). However, there are several storms with erroneously high ProbTor values (outer polygons with high probabilities) that quickly appear and disappear during the animation. Most of these false alarms are due to spurious MRMS azimuthal shear values which are produced by noisy Doppler velocity data. Work is ongoing to mitigate these errors in ProbTor.

Fig. 2: ProbSevere and ProbTor contours, MRMS MergedReflectivity, and NWS severe weather warnings.

A storm that formed in east Texas/west Louisiana spawned numerous tornadoes and prompted a tornado emergency for Alexandria, LA at 18:41 UTC. This storm was in a primed thermodynamic and kinematic environment, with 250-300 J/kg of 0-1km storm-relative helicity. The low-level and mid-level MRMS azimuthal shear values increased the ProbTor probabilities from 30% to 91% in about 20 minutes. You can see the saved time series of attributes for this storm here. Figure 3 shows how the ProbSevere products evolved for this storm in comparison to NWS severe weather warnings and local storm reports.

Fig. 3: Time series of ProbSevere probabilities for tornadic storm in Louisiana. The bottom axis plots durations for NWS severe weather warnings and times of preliminary LSRs.

At least one this this storm's tornadoes was expected to be rated as significant (EF3+).

Tornadoes in the desert

Early in the morning on the day after Thanksgiving, a long-wave upper-air trough with an embedded short wave disturbance and associated diffluent flow forced thunderstorms in the Phoenix, AZ region. There was enough low-level moisture return to provide adequate CAPE within a well-sheared kinematic environment, providing storm organization and maintenance.

ProbTor (from NOAA/CIMSS ProbSevere) captured the evolution of these storms, two of which spawned three tornadoes in the Phoenix metro area. The twisters uprooted trees, and caused damage to powerlines and roofs.

Fig. 1: ProbSevere contours (ProbTor is the outer contour), MRMS MergedRef, and NWS severe weather warnings. 

The tornado that traveled north of downtown Phoenix was rated EF1. From the time series below, ProbWind and ProbTor were about 30% when the NWS issued a severe thunderstorm warning. Then, ProbTor spiked to about 65% as the 0-2km MRMS AzShear increased markedly. The paltry lightning activity and weak 3-6km MRMS AzShear in this storm show that the rotation was shallow in the troposphere and this was not a supercellular storm. The 0-2km AzShear, along with very strong effective bulk shear and 1-3km mean wind helped the ProbTor values increase rapidly.

Fig. 2: ProbSevere time series for ProbTor, ProbWind, and ProbHail for the northern tornadic storm. NWS warnings and preliminary local storm reports are plotted on the lower axes.

Fig. 3: ProbSevere time series for ProbTor, ProbWind, and ProbHail for the northern tornadic storm. NWS warnings and preliminary local storm reports are plotted on the lower axes.

Further south along the line of storms, ProbTor values behaved in a more cyclic manner, first hitting 40% before dropping to 15%, then increasing rapidly to 30% (at the time of the first tornado LSR) and 60% (at the time of the second tornado LSR). ProbTor values then decreased to 20% and rebounded to 55%. This cyclic nature followed the 0-2km AzShear somewhat closely.

Fig. 4: ProbSevere time series for ProbTor, ProbWind, and ProbHail for the southern storm. NWS warnings and preliminary local storm reports are plotted on the lower axes. 

Fig. 5: ProbSevere time series for ProbTor, ProbWind, and ProbHail for the southern storm. NWS warnings and preliminary local storm reports are plotted on the lower axes.

Jefferson City tornado

A violent tornado hit the capital of Missouri last night at around 11:20pm local time. ProbTor showed rapid increases as well as some large fluctuations in probability prior to and during the tornado. After leaving Jefferson City, the storm continued to be warned and ProbTor again increased to probabilities ≥ 60% for about an hour; no reports of any kind were recorded during this time, however. You can see the archived time series of probabilities and predictors here.

Figure 1: Time series of ProbSevere products with NWS warnings and LSRs.

Figure 2: ProbSevere with MRMS MergedRef and NWS warnings in AWIPS2

Tornadic storm near Des Moines

As a surface low pressure system tracked northeast through central Iowa last night, one storm near the center of the low spawned at least one tornado southeast of Des Moines, IA. The first report came at 00:38 UTC, about 13 minutes after the first tornado warning. The ProbTor product from ProbSevere version2 showed elevated probability of tornado values (≥ 20%) as early as 23:48 UTC.

Figure 1: ProbTornado contours, MRMS MergedReflectivityComposite, and NWS severe weather warnings.

From the time series below, we see that increased rotation (0-2km MRMS AzShear; solid orange line) and increased total lightning density (solid green line) helped jump the probability of tornado from single digits to 40% at 00:00 UTC, and again from 20% to 50% between 00:26 and 00:38 UTC. The effective bulk shear and 0-1 km storm-relative helicity were clearly adequate, but not outrageous (~40 kts and ~150 m2/s2, respectively).

Figure 2: Time series of ProbTor and predictor values. The bottom axes contain NWS warning durations and LSRs.

One thing forecasters should look for when using ProbTor is when a storm's probability of tornado "sticks out" amongst the values of neighboring storms. This storm is a good example of that, where neighboring storms in a 4-county radius had probabilities in the single digits.

Forecasters should also be aware that the probability will fluctuate more with ProbTor than the probability of severe. This storm exhibited some fluctuation, whereby probabilities rose from 7% to 43%, dipped to 20%, and rose again to 55% during the tornado(es). ProbSevere developers are working on incorporating a visual time series function in AWIPS2 for ProbSevere storm objects (similar to Figure 2), so that forecasters can quickly see the history of a storm, which may help in warning decision making.

Lastly, there is no "magic" threshold of ProbTor for when tornadogenesis is imminent. Very potent environments characterized by SPC's "moderate" or "high" risk for tornado might see ProbTor values in the 70-90% range for tornadic storms, whereas for a more common tornadic environment like this, 20-50% can be expected for tornadic storms. In environments where tornadoes are not expected, ProbTor values of even 10% might be significant.

Users of the ProbTor product should keep these points in mind, but ultimately, forecasters should become more comfortable with the product as they gain experience with it in different regimes.

Please see our training module for more details on ProbHail, ProbWind, and ProbTor, and this fact sheet regarding the differences between ProbSevere version1 and version2.

Spartanburg, SC Tornadoes

A potent October shortwave traversed the southern U.S. on October 23rd, bringing with it a round of severe weather to the Southeast U.S. The Storm Prediction Center outlook had a 5% contour for tornadoes (Figure 1; probability of a tornado within 25 miles of a given point), citing very strong low-level shear and weak to moderate buoyancy.

Fig.1: SPC tornado outlook issued at 16Z 10/23/2017.

Convection took on a linear storm mode in South Carolina, with a surge in a line heading toward Spartanburg (Figure 2). The NOAA/CIMSS ProbTor model increased to over 80% by 18:20Z, owing to increases in low and mid level azimuthal shear, and a jump in total lightning density, while residing in an environment with excellent 0-1km storm relative helicity (Figure 3).

Fig. 2: ProbSevere (all hazards) contours, NWS warnings, MRMS MergedReflectivity, GOES visible reflectance. The ProbTor was the highest hazard probability for this storm; 88% at this time. 

The first tornado reported was at 18:54Z, coincident with the first NWS tornado warning. The first tornado damaged permanent and mobile homes, downed power lines, and destroyed a horse barn. Later, the storm produced another tornado between downtown Spartanburg and I-26, snapping large trees, as ProbTor was about 45%. After dropping below 20%, ProbTor again increased to 40% at 19:36Z, when another tornado was reported near the North/South Carolina border.

Fig. 3: Time series of ProbTor (thick red line), with constituent predictors. NWS warnings and preliminary LSR reports are on bottom axes.

In this case, ProbTor may have been able to provide an earlier "heads up" to forecasters monitoring the storms for potential tornadic activity. ProbTor is best utilized in conjunction with current warning practices and interrogation of base radar data. If a velocity couplet is outside of the ProbTor object polygon, ProbTor will be underestimated, as may have been the case for a second tornadic storm in North Carolina on this day.

Autumn storms on the Plains

An energetic, negatively-tilted shortwave trough traversed the middle of the country last week, bringing several bouts of severe weather to the Great Plains. By visualizing the accumulation of the ProbSevere output (storm centroids ≥ 50% [pink boxes]), the NWS warnings (orange and red polygons), and storm reports (blue, green, and red circles), we see that the ProbSevere model handled the first event on October 4th quite well, at least qualitatively.

Figure 1: A toggle between the 12Z 10/4 ->12Z 10/5 accumulation of NWS warnings and reports, with overlaid ProbSevere centroids greater than or equal to 50%.

A very long-lived storm that approached the Norman, OK area produced numerous hail reports. It was first identified at 23:47 UTC, with a probability exceeding 60%, due to high effective bulk shear, MUCAPE, and strong satellite growth rates. The MESH was 0.24" and total flash rate was only 1 fl/min. After quick jumps in the MESH (to ~0.5") and the flash rate (to almost 20 fl/min), the probability exceeded 80%, and was promptly warned at 23:52 UTC. We can see a time series of the ProbSevere probability (thick red line) and constituent predictors in the time series below. Note that the satellite growth rates were both 'Strong', and expired a little after 02:00Z. We can also see when NWS warnings were valid and when severe weather was reported in the lower subplot.

Figure 2: Time series of ProbSevere and its predictors (top), and NWS severe weather warnings and severe weather reports (bottom). The axes on the right are associated with the time series on the top subplot, while the legend in the lower right is associated with the bottom subplot.

The stunning time lapse of this storm was captured by Jim Ladue (NOAA/NWS/WDTD) as it approached from the west. The range of this video is from about 23:52 UTC (when the storm was first warned) to nearly 01:00 UTC.




A second short-wave ejected through the Plains on October 6-7 (as Hurricane Matthew threatened the southeastern seaboard), bringing with it another bout of storms, and this time, numerous tornadoes. ProbSevere again handled most storms quite well, with few false alarms and a couple of missed wind reports.

Figure 3: A toggle between the 12Z 10/6->12Z 10/7 accumulation of NWS warnings and reports, with overlaid ProbSevere centroids greater than or equal to 50%.

We can see the evolution of the storms and associated warnings below, over Kansas, Oklahoma, and far northern Texas. The animation is from 18:30UTC to 00:00UTC every 10 minutes. The pre-frontal storms and those close to the triple-point (in north central KS) remain discrete longer than those which are forced by the cold front. The red, orange, and green polygons denote NWS tornado, severe thunderstorm, and flash flood warnings.

Figure 4: Animation of ProbSevere contours, NWS polygons, and MRMS MergedReflectivity from 20161006-18:30Z to 20161007-00:00Z .

September storms near the Great Salt Lake

A number storms formed in the early afternoon of September 22nd in response to forcing associated with the North American Monsoon over the Intermountain West region of the U.S. The image below shows the accumulations of NWS severe weather warnings, storm reports from SPC, and the centroids of ProbSevere objects attaining 50%+, at each time. Each accumulation is over the timeframe of 12Z on 9/22 to 12Z on 9/23. You can get a quick-look at how the model performed using these accumulations (the previous day accumulations are here). On this day, the NOAA/CIMSS ProbSevere model performed reasonably well, with high probabilities corresponding to numerous wind, hail and tornado reports. There were a couple of false alarms to the south east of the Great Salt Lake and a couple wind reports missed to its west and south.

Figure 1: Accumulations of ProbSevere objects, reports, and NWS warnings for 9/22/2016.

The storm that produced hail, wind, and the one the tornado reported initiated well to the southwest of Salt Lake City. The time series below of its probability and constituent predictors in ProbSevere demonstrates its evolution.

The probability of severe is the thick red line, with the scale on the left. The six predictors in ProbSevere have varying scales on the right. The NWP predictors of effective bulk shear and MUCAPE are the dashed black and brown lines, respectively. The MESH is solid orange, and the total lightning flash rate is solid green. The lifetime max normalized satellite growth rate and glaciate rate are depicted by the solid blue and dashed cyan lines, respectively. Both satellite growth rates use the blue scale on the right with nominal 'Weak', 'Mod.' (moderate), and 'Strong' designations.

Figure 2: Time series of ProbSevere predictors and severe probability value for a long-lived storm affecting the Salt Lake City metro area.

We see that the normalized satellite growth rate from GOES-West was strong at 18:50Z, while the probability of severe jumped to 15%. The jump in MESH in a high shear environment also helped to jump the probability up to 50% at 19:12Z. Increasing MESH and flash rate helped the probability climb to over 90% by 19:50Z. The first severe thunderstorm warning was issued at 20:09Z. Golfball-sized hail was observed at the Antelope Island Marina at 21:37Z. The tornado in the city of Ogden was reported at 21:45Z, and left thousands without power.

Surprise Indiana tornadoes and total lightning in the ProbSevere model

A number of tornadoes spawned from storms in central and northern Indiana yesterday afternoon -- including strong ones -- in an area where tornadic activity was not expected. NOAA's Storm Prediction Center (SPC) forecasted a corridor of marginal to slight risk for general severe weather extending into Indiana, Ohio, and lower Michigan (Figure 1). However, the probability of a tornado within 25 miles of any given point was less than 2% in Indiana and Ohio (Figure 2).

Figure 1: SPC categorical outlook for August 24, 2016.

Figure 2: Tornado probability outlook for August 24, 2016.

Storms formed in the early afternoon in a juicy warm sector of an occluding system and put down multiple tornadoes, including an EF-3 tornado in Kokomo, Indiana, causing substantial damage to a shopping mall and leaving thousands without power.

Figure 3: SPC storm reports for August 25, 2016.

Though the NOAA/CIMSS ProbSevere model doesn't provide guidance to the type of severe weather, it could have given forecasters a heads up to some of the storms during this event. The animation below shows two storms developing in central Indiana (Figure 4). The first storm (near the Illinois border) had initially modest MESH (0.5-0.7"), but an excellent total lightning flash rate (50+ fl/min), before increasing to over 1" of MESH briefly. ProbSevere provided about 30 minutes of leadtime to the first severe thunderstorm warning from the 70% threshold. A tornado warning was issued at 18:37 UTC, and a damaging tornado reported at 18:55 UTC. One-inch hail was later reported in Indianapolis.

Further to the northeast, where MUCAPE was markedly less (~1600-1800 J/kg) but effective shear about the same (~45 kts), another storm formed west of the city of Kokomo, which produced the EF-3 tornado. This storm had low MESH (0.3-0.5"), but a rapid increase in flash rate (18->46->62 fl/min) in a well-sheared environment. This helped give the storm a probability of severe of 68% at 18:42 UTC, 8 minutes before a tornado warning was issued. The total lightning flash rate/effective shear predictor helped increase the probability of severe despite poor integrated radar reflectivity and satellite growth rates. A tornado emergency would later be issued for the community of Kokomo.

Figure 4: Two storms in central Indiana showcasing the utility of total lightning data in ProbSevere.

To show the effect of total lightning flash rate explicitly, ProbSevere was run with just radar, satellite and near-storm environment NWP data. As figure 5 shows, the probability of severe on the first storm only became elevated when the MESH neared or exceeded 1". In the Kokomo storm, the probability was 46% greater at the time of the initial tornado warning with lightning than without (70% vs 24%)! For both storms, the probability was generally 20-40% greater with the inclusion of total lightning data.

Figure 5: ProbSevere for the two storms in central Indiana, with total lightning data OMITTED from the probability computation.

The tornadic storm that traversed Kokomo traveled to the east side of Indiana, where it again re-intensified and became warned at 20:46 UTC. The probability of severe jumped up largely in response to an increasing flash rate. The storm would go on to produce numerous more tornado reports, beginning at 20:45 UTC.

This storm and another to it's north moved into Ohio, still producing tornadoes. However, the ProbSevere model only had low probabilities at this point for both storms, as the flash rates dropped into the single digits and MESH was largely below 0.33". So the flash rate did not contribute much with these two storms later on (see Figure 6). It is still uncertain what aspect of the environment in far northeastern Indiana modified the morphology of these storms.

Figure 6: Two storms in northeastern IN / northwestern OH that had poor ProbSevere values but produced tornadoes. These storms show that total lightning doesn't help every storm.

This event in Indiana and Ohio is interesting for a number reasons, including the unexpected number and severity of tornadoes, as well as the seemingly different morphologies of storms in reasonably close proximity. These examples highlight where total lightning flash rate improves ProbSevere probabilities, despite a meager reflectivity signature, as well as where total lightning doesn't contribute. I hope this case also shows the utility of an 'ingredients-based' approach to forecasting using observations (and near-storm environment data), where one data source may give insight to future storm severity when another is not, or when multiple observation sources may corroborate each other to enhance forecaster confidence and leadtime. This case also underscores NOAA/CIMSS's efforts to provide hazard specific guidance in future improvements to ProbSevere.

EDIT: This blog post by Jeff Frame gives a good post-mortem of the event. It appears the MCV in Illinois/Indiana played a key role, and that the environment itself was actually reasonably favorable for tornadoes, but that NWP guidance struggled depicting it as well as depicting the morphology of the storms.

2015 North American monsoonal storms

The NOAA/CIMSS ProbSevere model was reprocessed with total lightning data for several days in the fall of 2015 upon request from the National Weather Service. This post recaps a few of the interesting storms from these days.

October 18, 2015

A cluster of storms affected the Phoenix, AZ metro in the afternoon/evening of October 18, with one storm intensifying and producing severe weather in downtown Glendale. This storm only had weak satellite growth, but the good total lightning flash rate (up to 40 flashes/min) and strong MRMS MESH (1.08") generated a ProbSevere value of 51% at 22:48Z, despite rather weak effective bulk shear. The first wind report was at 22:50Z. So there wasn't much lead-time at all for this storm from the 50% threshold, but the ramp up in probabilities could have signaled to the forecaster that this was a storm to watch (3%-->11%-->17%-->38%-->41%-->51%), as well as how much higher the ProbSevere value was than neighboring storms. The storm produced multiple large hail (up to 1.25") and severe wind reports.

Figure 1: ProbSevere, MRMS composite reflectivity, and NWS warnings for storms near Phoenix, AZ.

Further southeast, northwest of Tucson, a very small storm produced big hail (1" diameter) at 21:04Z. A maximum MESH of 0.89", moderate satellite growth rate, and very low lightning (0-1 fl/min) combined for a max probability of only 21%, at the time of the first report. The very low lightning combined with modest effective bulk shear (~25 kts) certainly helped to keep the probability of severe low. This example shows that more training with western U.S. storms is necessary for the ProbSevere model, especially as far as total lightning is concerned.

Figure 2: ProbSevere and MRMS composite reflectivity for a small storm near Tucson. This storm had nearly zero observed lightning flashes (IC or CG). 

Strong storms also erupted in southeast California this day, with strong satellite growth rates, good total lightning flash rates, and strong MESH values, all combining to produce probabilities in excess of 90%. Only one storm was warned despite the high MESH values (as high as 1.45"), but no reports were recorded from these storms in the Mojave Desert region of California. The MESH might possibly have been biased due to rather sparse radar coverage in this region.

Figure 2: ProbSevere, MRMS composite reflectivity, and NWS warnings for storms in the Mojave Desert.

October 6, 2015

Numerous storms developed in southern Arizona in the early afternoon of October 6th. One storm stood out southwest of Phoenix, with a good flash rate (32 flashes/min), and good MESH (0.92"), but no satellite growth rates. The ProbSevere value ramped up from 17% to 66% in 10 min (from 18:58Z to 19:08Z). The probability then hovered in the 40-50% range before a tornado was reported at 19:34Z. Though ProbSevere doesn't have any predictors explicitly for tornadogenesis, this case demonstrates that it can highlight a strongly developing storm to the forecaster, which signals the need for him/her to further investigate it.

Figure 3: ProbSevere and MRMS composite reflectivity for a storm southwest of Phoenix, which produced a tornado. 

Further southeast in Tucson, a storm exhibited moderate glaciation and normalized satellite growth rates, modest lightning (< 20 fl/min), and modest MESH (lifetime max was 0.58"). The shear and MUCAPE were adequate (~35 kts and 1000 J/kg, respectively). The ProbSevere predictors all pointed to a garden variety thunderstorm (max probability was 24%), yet this warned storm went on to produce two 1" hail reports and a wind report in Tucson. The MESH may have been underestimated due to the storm being near the radar, and thus possibly partially in the "cone of silence". The next closest radar is in Phoenix, with it's lowest tilt being over 8,000 feet at the storm's location (possibly higher, depending on atmospheric conditions). It's also possible the storm may have been shallow, as well, with MESH not being as representative. The SPC mesoanalysis archive shows that the melting level was relatively low (~2500 m), which in the future might help correct the MESH in shallow storms.

Figure 4: A storm near Tucson, AZ, which produced severe hail and wind.

Finally, later in the afternoon, a storm quickly intensified (went from 10% at 21:50Z to 70% at 21:58Z), heading toward Casa Grande, AZ, and was promptly warned. The increasing MESH and total lightning caused the rapid increase in probability. The storm began producing golfball and silver dollar sized hail at 22:10Z.

Figure 5: A strong storm picked up by ProbSevere heading toward Casa Grande, AZ.

September 14, 2015

A couple of storms developed near the Phoenix, AZ metro area on the evening of Sept. 14th, with one storm producing multiple wind reports (e.g., trees and power poles down) in downtown Phoenix. The MUCAPE and effective bulk shear parameters for the wind-producing storm were good (~2200 J/kg and 30-35 kts, respectively). At 01:00Z, a moderate normalized satellite growth rate and MESH at 1.01" combined to generate a probability of 47% (the max in its lifetime). The flash rate was 7 flashes/min. About 10 min later, the storm diminished markedly, as the MESH went below 0.1" and flash rate below 5 fl/min. The ProbSevere values were in the single digits when the storm first began producing severe wind reports. So unless the radar operator was paying close attention to the probabilities nearly an hour prior, ProbSevere may not have helped much in this case. That being said, development is underway to incorporate other NWP and radar fields to better predict wet-microbursts. For instance, the low-level lapse rates were very good in this region (as shown by the SPC mesoanalysis archive), which helps in momentum transport. The ProbSevere developers will be investigating many fields, including low-level lapse rates to better predict severe wet-microbursts. The low ground-based total lightning also didn't help. It's not certain whether this is a detection efficiency or a meteorological cause, but it underscores the need for more training for western U.S. storms.

Figure 6: ProbSevere and MRMS composite reflectivity for a storm affecting the Phoenix metro.

These cases show that ProbSevere can help highlight storms for forecasters to watch and further interrogate, and that forecasters must also continue to bear in mind data problems (e.g., sparse radar coverage, possible low lightning detection efficiency), as well as environmental factors not captured in the ProbSevere model (e.g., shallow storms). We hope the ProbSevere model will constitute another piece of useful guidance to the forecaster and compliment the warning process.

Tornadic supercell near Paducah

A supercell quickly formed southwest of Paducah, KY, in an excellent environment (2500 J/kg of MUCAPE, 45+ kts of effective bulk shear). A strong normalized satellite growth rate and quickly increasing flash rate combined to increase the ProbSevere probability rapidly from 31% to 44% to 65% in 6 minutes. The 65% probability was exhibited 8 minutes before a severe thunderstorm warning was issued at 19:28Z, when the probability was 97%. Forecasters have noted at the HWT that a quick jump in probability (due to either increasing flash rate, MRMS MESH, or strong satellite growth rates) is typically a key indicator that a storm may soon become severe. This storm produced a tornado at 19:45Z.

Fig. 1: ProbSevere contours, MRMS composite reflectivity, and GOES-East visible reflectance, as shown from the ProbSevere website.

Fig. 2: Base reflectivity from KPAH.

Fig. 3: Base velocity from KPAH.

strong satellite growth rates boost ProbSevere value ahead of experimental warning

A storm south of Norman, OK exhibited strong glaciation and normalized satellite growth rates at 19:37Z, provided by GOES-East, to boost the probability of severe to over 60% by 20:34Z. Forecasters in Norman were shown the 2D flash rate/effective bulk shear predictor in ProbSevere, which gave them knowledge that well-sheared storms don't need super high flash rates to contribute postiviely to the overall probability. This knowledge helped forecasters quickly assess the storm and issue a severe thunderstorm warning at 20:41Z. The storm later produced a tornado.

Fig. 1: ProbSevere contours and MRMS MESH.

Tulsa-Claremore, OK Tornado

Most likely the strongest tornado of the day occurred just northeast of downtown Tulsa, Oklahoma, and moved east-northeastward toward Claremore and Justice, OK.

A strong GOES-derived normalized satellite growth rate, along with an increase in total flash rate from 22 to 35 flashes/min, increased the probability of severe to 84% at 22:42Z, in this high MUCAPE, high effective bulk shear environment, despite MRMS MESH only being 0.61 in. Fourteen minutes later (22:56Z), the probability increased to 95%, while MESH decreased to 0.56 in., and the flash rate increased to 84 fl/min. The probability generally stayed over 90% until the storm's demise. The first severe thunderstorm warning was issued at 23:35Z, about 40 min after the storm attained a >90% value from the ProbSevere model.

A tornadic debris ball signature was evident from KINX radar at 00:17Z, manifest in the base radar fields: a 'hook' signature in the reflectivity, strong velocity couplet, low correlation coefficient, and near zero or varying differential reflectivity; all coincident features are an indication that a tornado is creating and lifting debris. The storm also produced golfball-sized hail.

Future development of the ProbSevere model will investigate radar velocity products and dual-pol fields to give more accurate and hazard specific probabilities of severe (e.g., hail, wind, tornado).

KINX low-level scan ~00:17Z March 31, 2016. Images are from social media. The coincident signatures indicate a tornadic debris ball. Top-left: hook signature in horizontal reflectivity; top-right: strong inbound-outbound velocity couplet; bottom-left: a 'hole' of low correlation coefficient; bottom-right: near-zero differential reflectivity.