IntenseStormNet in Kansas and Oklahoma

ProbSevere IntenseStormNet is not being evaluated at HWT this week, but it is a unique application of satellite data for severe weather. On the eve of the 2nd week of the 2024 HWT, severe storms were ravaging the Plains.

IntenseStormNet is an AI model that uses patterns in ABI and GLM image data to predict a probability of "intense" convection. The strongest probabilities are often correlated with strong overshooting tops, bubbly texture in the visible band, storm-top divergence, and lightning cores.

In the movie above, you can see several MCSs traversing the state, producing severe hail, wind, and tornado reports. One the convection becomes cold-pool driven, the probabilities often diminish, sometimes significantly (see eastern Kansas at the end of the period).

Meanwhile in Oklahoma, a monster supercell progressed steadily through the western half of the state, dropping tornadoes, hail, and producing severe wind gusts. One interesting aspect about the product was the drop in probabilities from about 01:40 - 02:10 UTC. This drop corresponded to a short gap in severe weather reports produced by the supercell. Visually, the main overshooting top appeared to diminish.

IntenseStormNet is used in ProbSevere v3, feeding in satellite information at the mature stage to the severe-weather models.

Midwest storms...in February!

 Record-breaking high temperatures for February preceded and helped spawn storms in eastern Iowa and northern Illinois yesterday.

ProbSevere LightningCast gave a heads up on the developing convection. LightningCast uses deep-learning methods and GOES ABI data to predict the probability of lightning up to an hour in advance. LightningCast version 1 (LCv1) is being transitioned to NOAA operations in 2024. 

Development and improvement of LightningCast continues, however, based on forecaster feedback. One new product that is being evaluated at the Hazardous Weather Testbed this year is the probability of ≥ 10 fl in 60 min. Forecasters remarked that in some situations, having guidance on a more robust level of lightning would be very helpful. 

In the animation below, the P(≥ 1fl in 60 min) are depicted by contours in shades of blue (at 10%, 25%, and 50% levels). The P(≥ 10 fl in 60 min) are in red-shaded contours (10% and 25% levels).

For the three storms in Illinois, there was 13 min, 6 min, and 13 min of lead time to ≥10 flashes (when measured from the 10% contour). The storms in Iowa never got to 10 flashes, but the LightningCast probabilities only touched 25%. Users should recognize that lead time for the 10-flash product will be lower than the 1-flash product. When probabilities above 10% begin to appear (and especially above 25%), forecasters should anticipate an intensification in the lightning activity. 

Figure 1: LightningCast contours (P[≥ 1 fl] in blues, P[≥ 10 fl] in reds), GOES-16 ABI imagery, and GOES-16 GLM flash-extent density for storms in eastern Iowa and northern Illinois.

These storms developed rapidly into supercells and produced hail, severe wind gusts, and several tornadoes. ProbSevere v3 corresponded with NWS warnings quite well, even in this explosive environment. Notably, ProbTor v3 was much higher (30-45%) than ProbTor v2. While more investigation is needed to confirm this, it is likely that the HRRR model inputs in v3 were more impactful than the RAP model inputs in v2, signifying that the HRRR better depicted the environment. 

ProbSevere v3 for a couple of early autumn storms

ProbSevere v3 (PSv3) was able to provide an earlier heads-up and increased lead time to the initial severe hazards in several recent autumn storms on the Southern Great Plains. 

On September 23, there were a number of powerful supercells ravaging east-central Oklahoma. The storm in Figure 1 quickly split into two cells, with the right split producing 2"-diameter hail at 23:55 UTC and 3" hail at 00:15 UTC.

Figure 1: ProbSevere v3, MRMS MergedReflectivity, and NWS severe weather warnings for storms in central Oklahoma.

Figure 2: ProbSevere v3 for a storm in Pottawatomie county at 22:56 UTC.

As this supercell was developing (see Figure 2), PSv3 was 37%, whereas PSv2 was only 3%. Hail appeared to be the main threat. At this time, the very weak lightning signal (only 5 fl/min) and low MESH (0.33") were keeping the PSv2 probability very low. However, the top 5 contributing predictors in PSv3 were:

• Lapse rate 0-3 km (8.4 C/km)
• MLCAPE (3685 J/kg)
• Eff. bulk shear (46 kt)
• Sat growth rate (2.7%/min -- "moderate")
• MRMS MESH (0.33")

ProbSevere v3, compared to v2, is more adept at extracting salient signals in the combined NWP, satellite, radar, and lightning phase space. In the developing stage of this storm, the environment and the satellite were more important than the radar predictors.  

The next evening, it was Texas's turn. One lone supercell in west Texas produced golfball-sized hail and wind reports of 59 and 62 mph (Figure 3).

Figure 3: ProbSevere v3, MRMS MergedReflectivity, and NWS severe weather warnings for a lone supercell in west Texas.

Figure 4: The storm in Figure 3, at 01:44 UTC.

ProbSevere v3 again had a jump on this storm before PSv2. At 01:44 UTC, right after the initial severe wind report, PSv3 was 55% vs. v2's 9% (Figure 4). PSv3 was showing both hail and wind as potential threats. The very weak low-level mean wind (5 kt) was keeping ProbWind v2 low (5%), and the modest flash rate and MESH, and very low hail CAPE (~250 J/kg) were combining for a ProbHail v2 of only 9%.

In contrast, the top contributors in ProbSevere v3 were the modest flash rate (18 fl/min), modest MESH (0.72"), the effective shear (47 kt), and the low-level lapse rate (8.1 C/km). PSv3 was able to integrate the marginal radar signature with favorable NWP data (from HRRR) to provide a better indication of severe probability.

Figure 5 highlights the higher probability of severe in PSv3 well before PSv2 shoots up, which was after the initial wind report. This storm later produced severe hail and another wind report.

Figure 5: Time series comparing PSv3 and PSv2 probabilities during the developing stage of the supcercell in Figures 3 and 4.

Tennessee storm seeing huge difference in ProbSevere v3 vs. v2

 We're in the Great Plains today, but I just wanted to document this storm where ProbSevere v3 was much higher than version 2. Powerlines were reported down on I-65, south of Nashville, shortly after the image time in Figure 1 (at 18:28 UTC). PSv2 was 9% while PSv3 was 51%. The kinematic fields were likely dampening PSv2 probabilities, while the azimuthal shear and flash rates were only modest. The low-level lapse rate was also strong, contributing to an elevated probability of severe (in version 3). PSv3 was higher than PSv2 for quite a while, until the azshear increased and the version 2 probabilities (see Figure 2).

Figure 1: ProbSevere contour and MRMS MergedReflectivity for a storm in central Tennessee.

Figure 2: Trends in ProbSevere probabilities for PSv2 and PSv3, for the storm in Figure 1.

Tornadic Panhandle storm emblematic of the differences between ProbTor v3 and v2

A powerful storm that saw it's genesis in northeast New Mexico and it's demise in south-central Oklahoma, spanned nearly 11 hours. It produced tornado reports near the Oklahoma-Texas border in the Panhandle region, as well as numerous gigantic hail (5" diameter) and wind reports (up to 75 mph) (Figure 1).

Figure 1: ProbSevere v3 contours (outer contour is colored by ProbTor v3), MRMS MergedReflectivity, and NWS severe weather warnings for a supercell in the OK/TX Panhandles.

This storm clearly demonstrates some important differences between ProbTor version 3 (PTv3) and version 2 (PTv2). Early on in the storm's lifecycle, PTv3 was much greater than PTv2. At 21:24 UTC (see Figure 2), PTv3 = 34%, vs. PTv2 = 1%. The first tornado report was at 21:51 UTC, when PTv2 finally increased to 14% (PTv2 later increased to about 70%). Looking into this further, we found several things contributing to the elevated PTv3 and much diminished PTv2:

• In PTv2, which uses RAP, the SRH 0-1 km AGL was only 50 m2/s2, vs. 100 m2/s2 in the HRRR, which PTv3 utilizes.
• Similarly, PTv2 strongly relies on the mean wind in the 1-3 km AGL layer, which was only 16 kt. This also was dampening the PTv3 probability.
• The 0-2 km azimuthal shear (i.e., low-level storm rotation) was modest. This was not enough to overcome the poor low-level kinematics in PTv2 (until much later), while at the same time, it was not too harsh on PTv3.
• The top contributing predictors for PTv3 were the 0-2 km azimuthal shear, the effective bulk shear (~ 65 kt), the very high MESH (3.43") and surprisingly, a strong low-level laspe rate (8.8 C/km). 

Figure 2: ProbSevere v3 contours (outer contour is colored by ProbTor v3), MRMS MergedReflectivity, and NWS severe weather warnings for the tornadic supercell. Note that PTv3 is 34% here, while PTv2 was only 1%. 

Later on, as the azimuthal shear went off the charts (see Figure 3), PTv2 was much higher than PTv3, but also much more erratic. PTv3 consistently remained in the 30-40% range for awhile, despite very high azshear. This indicates that the PTv3 model has learned that azshear can be quite noisy (due to things such as radar sidelobe contamination), and it learned not to overemphasize the radar-based rotation predictors.

The PTv3 model correctly ramped up probabilities to 30-35% when rotation increased in a decent environment (from about 20:20 to 21:00 UTC), but hedged when rotation (as observed by MRMS azshear) was very high, as evidenced by probabilities below 40%. As of now, there were no other tornado reports after 22 UTC, so this seems like a sensible hedge (the storm was tornado-warned until 02:15 UTC).

In practice, we hope that forecasters find value in earlier ramp-ups in the probability of tornado, while understanding the uncertainty of using scalar-based predictors leads to lower probabilities overall, compared to PTv2 (and fewer erratic swings). We hope that using image-based methods will improve the guidance even further. 

Interactive version of the plots in Figure 3. 

Figure 3: Time series of PSv2 and PSv3 probabilities and certain predictors for this storm.

Storms surging in northeast Texas

As we begin today's HWT shift, storms are already developing quickly. ProbSevere LightningCast was able to readily give objective probabilistic guidance on the next-hour probability of lightning for these storms, which developed under clear skies.

Figure 1: LightningCast probabilities (contours), GOES-16 ABI daytime cloud phase distinction RGB (background), and GOES-16 GLM flash-extent density (orange-foreground). 

The Dallas / Fort Worth metro is seeing severe hail from these rapidly developing storms. The GOES-R satellite growth rates are contributing to high probabilities of severe for both ProbSevere v2 and v3 (huge CAPE and effective shear doesn't hurt, either). 

Figure 2: ProbSevere v3 contours, MRMS MergedReflectivity, and NWS severe weather warnings for storms near the Dallas / Fort Worth metro.

ProbSevere v3 showing improvement

While the HWT forecasters didn't operate in North Dakota or South Carolina yesterday, I wanted to document several cases where Probsevere v3 (PSv3) was making improved predictions of severe weather, relative to ProbSevere v2 (PSv2).

In eastern North Dakota, very modest shear under a ridge helped these storms to get a little organization. Though PHv3 and PWv3 are rather low, the overall PSv3 probability was 44%, compared to PSv2 of 14%. Soon after the image in Figure 1, golf ball-sized hail was reported. In this storm, the low effective shear and lightning flash rate (12 fl/min) were hurting PSv2, while the MESH, ENI flash rate, low-level lapse rate (8.7 C/km), and decent mid-level azimuthal shear were the top contributors to the probability of severe. 

Figure 1: ProbSevere contours, MRMS MergedReflectivity, and NWS severe weather warnings.

A little further west and about an hour later, another storm showed a nearly 40% difference between PSv3 and PSv2 (50% vs. 13%). Despite a better flash rate and 1" of MESH, PSv2 was again hampered by low effective shear. Similar to the previous storm, the low-level lapse rate, MESH, and flash rate were helping PSv3. But in this storm, a strong satellite growth rate (not displayed in Figure 2) was the 5th leading positive predictor. Soon after the image below, there were two reports of semi-trucks blow over. Interestingly in this storm, the GLM flash rate was very low (1 fl/5 min) compared to the ENI flash rate (34 fl/min). This is another example showing how using data fusion helps create a more robust probabilistic model---when one data source is suspect (for whatever reason), others pick up the slack. We have also seen storms where GLM flash rates are much higher than ENI flash rates.

Figure 2: As Figure 1, but for a second North Dakota storm.

Several storms in South Carolina were showing some good improvement as well. A storm entering western South Carolina was pegged at 35% in PSv3, while only 3% in PSv2, owing to low MESH (0.47 in), low shear (~20 kt), and low/modest ENI flash rate (16 fl/min). In PSv3, a combination of good satellite growth (the growth rate was older, so not displayed in Figure 3 readout below, but still used in the model computations), modest reflectivity-based parameters, and a solid low-level lapse rate (9.1 C/km) combined to produce the 35% probability of severe. Reports of multiple trees down came in soon after this image.

Figure 3:  As Figure 1, but for a storm on the South Carolina /  Georgia border.

Further east, a similar story, with PSv3 13% greater than PSv2 (43% vs 30%). Dozens of trees and powerlines were down from this storm. The probability of wind was much higher than the probability of hail in v3.

Figure 4:  As Figure 1, but for a storm in central South Carolina

An analysis that was provided in the training slides for PSv3 showed that improvement was greatest in the moderate MLCAPE and modest/moderate effective shear regimes, which these storms fit into. One other point to note is that while ProbSevere v2 is often higher on very strong and mature storms, we've seen that quite frequently, ProbSevere v3 ramps up in probability sooner than v2, which we feel is a very important distinction and improvement. While not explicitly shown in the animation in Figure 5, v3 was again a little ahead of v2 in these storms in southwest Texas, where one team of HWT forecasters was working.

Figure 5: An animation of ProbSevere v3 contours, MRMS MergedReflectivity, and NWS severe weather warnings.

The value of data fusion

We had an interesting storm develop in a radar gap in far eastern New Mexico yesterday. This is a great case study to demonstrate the value of data fusion in ProbSevere.

Figure 1 shows where the storm developed (the red circle), which was in a region of very poor "radar quality", as the eastern New Mexico KFDR radar was down. Thus, the closest radar was KAMA in Amarillo, TX.

Figure 1: Radar Quality Index for eastern New Mexico yesterday. The red circle is the approximate location of where the storm first developed. 

Figure 2: ProbSevere IntenseStormNet contours with GOES-16 ABI vis-IR sandwich product for a rapidly developing storm in eastern New Mexico.

One input into ProbSevere v3 is the probability of "intense" convection, as computed from IntenseStormNet. This is a deep-learning model that uses images of ABI 0.64-µm reflectance, 10.3-µm brightness temperature, and GLM flash-extent density to compute a probability of how "intense" the storm looks from a satellite perspective [paper]. 

The rapidly increasing IntenseStormNet probability, along with a favorable environment, and increasing total lightning flash rates helped jump the probability of severe despite poor radar reflectivity.

As the storm moved south and east into better radar coverage, radar reflectivity increased and the probabilities of severe further increased to above 70%. 

Figure 3: ProbSevere contours, MRMS MergedReflectivity, and NWS severe weather warnings.

Forecasters at the HWT have noted numerous times how ProbSevere v3 has increased before v2, particularly noticeable in the regime we've experienced this week, where the storms have had a dearth of lightning at the developing stages. At the time in Figure 4, this storm had PSv3 of 36% vs PSv2 of 12%. 

Figure 4: ProbSevere and MRMS MergedReflectivity for a developing storm in a radar gap in eastern New Mexico.

Later on, this storm produced numerous large hail, severe wind, and several tornado reports. Interestingly, the ProbTor v3 was much higher than ProbTor v2 prior to the first tornado report. In Figure 5, we can see PTv3 is 47% while PTv2 is only 9%. Looking into this deeper, we found that the environmental information such as the 0-1 km storm-relative helicity (~ 30 m^2/s^2) and the 1-3 km mean wind (~15 kt) were very low. The HRRR values in PTv3 were much better (~100 m^2/s^2 for SRH and 27 kt for the low-level mean wind). I believe this is an indication that PTv2 was too dependent on environmental information, compared to PTv2. This also demonstrates that the HRRR had a better handle on the environment than the RAP. You can see the low 0-1 km storm-relative helicity in the SPC mesoanalysis (Figure 6).

Figure 5: ProbSevere contours (the outer contour is colored by the probability of tornado), MRMS MergedReflectivity, and NWS severe weather warnings.

Figure 6: 0-1 km SRH (contours) and storm motion (vectors) prior to tornadogenesis. The red circle shows where the approximate location of the storm prior to producing tornadoes. 

Figure 7 demonstrates how ProbTor v3 was much higher than ProbTor v2 early on. The vertical black lines in the top-left two panels represent the times of the first and last tornado reports. The interactive version of these time series have been saved off and are available here.  

Figure 7: Time series of ProbSevere probabilities and radar, satellite, lightning, and HRRR attributes for the tornadic storm in Figure 6. 

Just a couple of central Texas hailers

Forecasters in moisture-poor Midland, TX yesterday were peeking into central Texas with envy, as surface dewpoint temperatures and resultant CAPE were much higher. One testbed forecaster noted that ProbSevere LightningCast was his first indication of strengthening convection to their east (Figure 1).

Figure 1: ProbSevere LightningCast contours, GOES-16 day-cloud phase distinction RGB, and GOES-16 GLM flash extent density for two storms in central Texas.

ProbSevere v3 later had a handle on these storms, which developed in an environment of strong CAPE (> 3000 J/kg) and modest effective shear (35-40 kt). Strengthening radar signatures, moderate satellite growth rates, and increasing total lightning flash rates helped to boost the PSv3 probabilities. PSv3 again ramped up probabilities before PSv2, whereas PSv2 probabilities were higher at peak maturity. Each storm produced severe hail (1.5" and 2" diameters, respectively). 

Figure 2: ProbSevere v3 contours, MRMS MergedReflectivity, and NWS severe weather warnings for two storms in central Texas. 

Two-inch hail on New Mexico/Texas border

Two-inch hail was reported on a storm on the New Mexico/Texas border, just on the edge of the Lubbock, TX CWA (Figure 1). At 21:04 UTC, 56 minutes prior to the report, ProbSevere v3 (PSv3) was 58% while PSv2 was 8% (Figure 2). At that point, the low ENI and GLM flash rates (≤ 4 fl/min) and modest environmental shear (≤ 30 kt) was dramatically reducing the PSv2 probabilities, whereas the shear was higher in PSv3 (36 kt) and the PSv3 models are less affected by low lightning (note: this storm eventually became well-electrified). The team in Lubbock today has noted how high PSv3 was prior to PSv2 early in the storms' development. While this is not always the case, it was true in Texas today because of the dearth of lightning early on.

Figure 1: ProbSevere v3 contours, MRMS MergedReflectivity, and NWS severe weather warnings for a storm on the NM/TX border.

 

Figure 2: Time series comparison of PSv2 and PSv3 for the storm in Figure 1. The annotation is valid for 21:04 UTC, when PSv2 

HWT 2023 underway

The HWT is back in person this week, after a 3-year hiatus. Forecasters have been working in Lubbock TX, Amarillo TX, and Great Falls MT. ProbSevere LightningCast and ProbSevere v3 are being evaluated this year. Here are few notes from our conversations regarding ProbSevere:

• Forecasters find the time series capabilities for each product to be very effective, and desire them in operations.
• Forecasters look mainly at the parallax-corrected LightningCast contours in AWIPS
  
• Forecasters have appreciate the improved calibration over ProbSevere v3 over v2. This often means that PSv3 is less than PSv2, but this comes with a much lower false alarm rate.
• Furthermore, we've seen a number of storms with higher PSv3 probabilities early on, compared to PSv2.
• Forecasters use both ProbSevere v3 and LightningCast in GRLevel software in their offices, as well as in AWIPS.
  
• We had a discussion about making LightningCast contours more configurable, such as custom contours that could be individually toggled on and off. 

Figure 1: LightningCast contours, GOES-16 ABI day-cloud phase distinction RGB, and GOES-16 GLM flash-extent density.

Figure 2: ProbSevere v3 contours, MRMS MergedReflectivity, and NWS severe weather warnings in west Texas.

ProbSevere v3 less affected by meager flash rates and effective shear

A number of severe storms on May 17, 2023 illustrated improvements of ProbSevere version 3 (PSv3) over version 2 (PSv2). 

Several storms in South Dakota exhibited low total-lightning flash rates (6-10 fl/min). The dearth of lightning greatly diminished the PSv2 probability of severe weather in the next 60 minutes, while not hurting the probabilities for PSv3 nearly as much.

In this storm near Rapid City, SD, PSv3 = 48% while PSv2 = 17%. This was shortly after the first 1-inch diameter hail report. The low effective shear (21 kt) was also diminishing the probability for both models.  The MESH (1.62"), VIL (40 kg/m^2), ENI flash rate (6 fl/min), and lapse rate 0-3 km (9.3 C/km) were the strongest contributors here. One observation we've noticed repeatedly is that total lightning information is still important in PSv3 (even low flash rates), but does adversely diminish the probability of severe to nearly the same degree as PSv2. 

Figure 1: ProbSevere, MRMS MergedRef, and NWS severe thunderstorm warning for a storm near Rapid City, SD.

Further south, the effective shear was a bit more favorable (33 kt). At this time (see Figure 2), PSv3 = 73% and PSv2 = 41%, and a 1-inch diameter hail report was recorded. More hail reports were recorded an hour later. The main difference between this storm and the storm in Figure 1 is that the effective shear was slightly favorable, while the low ENI flash rate was unfavorable in PSv2 (but not so in PSv3).

Figure 2: ProbSevere, MRMS MergedRef, and NWS severe thunderstorm warning for a storm west of Pine Ridge, SD.

Traveling eastward, another severe storm produced a 60-mph wind gust at 21:46 UTC (Figure 3). At this time, PSv3 was 23% higher than PSv2 (though PWv3 was only 11%). The high MESH (1.37") and composite reflectivity (70 dBZ) help compensate for the very low shear (18 kt), while again we see that the low flash rate does not harm PSv3 as much as PSv2. 

Figure 3: ProbSevere, MRMS MergedRef, and NWS severe thunderstorm warning for a storm near Belvidere, SD

Heading to the eastern U.S., numerous trees were reported down at 18:55 UTC in eastern Alabama (Figure 4). At this time, PSv3 = 21% and PSv2 = 2%. The flash rate was decent (25 fl/min), with a MESH of 0.57" and VIL of 31 kg/m^2. The environmental flow, however, was quite poor (eff. bulk shear = 19 kt; mean wind 1-3 km = 16 kt). An enhanced low-level lapse rate (7.5 C/km) and lower wetbulb 0C higher (11.1 kft) also nudged up the probability of severe in PSv3. Even though PSv3 was only 21%, that is a marked improvement over PSv2, and could have helped forecasters identify a possible severe weather threat sooner, while a PSv2 value of 2% would likely be ignored. 

Figure 4: ProbSevere and MRMS MergedRef for a severe storm near Mount Olive, AL.

A storm near Waynesboro, MS was increasing in probability prior to a hail report at 22:08 UTC. At this time, PSv3 was 31% higher than PSv2 (though v3 was highlighting wind as the main threat). The MESH, flash rate, and VIL were the main contributors here, while the paltry effective shear was the top detractor (Figure 5). 

Figure 5: ProbSevere and MRMS MergedRef for a severe storm near Waynesboro, MS.

Finally, in Charleston, SC, this storm produced a hail report and multiple wind reports from 21:40 to 21:50 UTC. At this time, PSv3 = 45% and PSv2 = 17%. Again, the flash rate, MESH, and lapse rate 0-3 km were the top contributors, whereas the lack of satellite growth was the main detractor (Figure 6). 

Figure 6: ProbSevere and MRMS MergedRef for a severe storm over Charleston, SC.

Overall, we've found that PSv3 improves severe weather detection and prediction over its predecessor, often highlighting threats sooner. For mature severe storms with very strong storm attributes, PSv2 is often higher than PSv2, sometimes by 20% or more (e.g., 95% vs. 75%). However, we believe that a little worse performance on the "high end" of storms is a good trade-off for being able to capture severe weather in more marginal (but still impactful) situations, and doing so sooner, overall. 

ProbSevere v3 in west Texas

Severe convection was rather marginal and sparse on 3 May, 2023. However, a few storms of note demonstrated better predictions for ProbSevere v3 (PSv3) over ProbSevere v2 (PSv2). 

A storm in the Texas Panhandle, southwest of Dumas, TX, was in an environment with moderate MLCAPE (1200 J/kg) and low-to-moderate effective bulk shear (30 kt). This storm was around for over an hour before it finally produced severe hail reports (up to 1.5" in diameter). Approximately an hour prior, however, PSv3 was much greater than PSv2; 62% vs. 4% (see Figure 1). In this case, PSv2 was too reliant on the ENI flash rate, which was only 4 fl/min. While lightning is important in PSv3 models, it is less affected by storms with a dearth of lightning than version 2, especially when other parameters show increased severe potential (in this case, MESH and mid-level azshear). 

Figure 1: ProbSevere contours, MRMS MergedRef, and NWS severe thunderstorm warning for a storm near Dumas, TX.

Figure 2: Time series of PSv3 and PSv2 probabilities for the storm near Dumas, TX, which produced 1.5" hail. 

The 00Z sounding from Amarillo, TX (Figure 3) might be able to shed some light on why the MESH was so high (up to 1.6") yet lightning was quite low (~6 fl/min) from both ENI and GLM sensors. In this case, the precipitable water in the column was quite low (about 0.68"). This, combined with the low environmental RH in the mixed phase region of the troposphere probably combined to produce relatively few collisions of liquid and frozen water particles needed to produce charge separation necessary for lightning production. Ice nuclei, on the other hand, seemed to accrete supercooled droplets efficiently, producing large hail stones.

Figure 2: Observed sounding from Amarillo, TX at 00Z on 4 May 2023.

Another storm in southwest Texas, near Pecos, also produced severe hail (1" diameter).  The lightning was also low for this storm, though PSv3 had 51% while PSv2 has 37%. The MRMS azimuthal shear parameters were zero here. This was likely because the average beam height of the lowest tilt of the closest radar was 10-11 km above ground level! The MRMS products compute azimuthal shear between 0-2 km and 3-6 km AGL. The missing azimuthal shear values for this storm reduced the probabilities of severe, but not quite as much in PSv3 as in PSv2. 

Figure 3: ProbSevere contours, MRMS MergedRef, and NWS severe thunderstorm warning for a storm near Pecos, TX.

East Coast Tornadoes

The past few days have seen several strong tornadoes along the U.S. east coast. A shortwave trough with ample upper-level diffluence provided a forcing mechanism for severe storms from Florida to Virginia.

Near Juno Beach, FL, a tornado damaged power lines, homes, buildings, and cars. Maximum wind speeds were estimated at 130 mph (rated EF2). Oddly enough, this tornado was only about 20 miles north of a weaker tornado from the day before.

Figure 1: ProbSevere v3 contours, MRMS MergedReflectivity, and NWS severe weather warnings for a storm near Juno Beach, FL. Outer contours on ProbSevere objects are colored by the probability of tornado.

ProbTor v3 (PTv3) is better calibrated than its v2 counterpart. There was a distinct ramp up in the tornado probability for this storm prior to tornadogenesis, compared to PTv2 (Figure 2). Part of this ramp up was due to higher 0-1 km storm-relative helicity depicted in the HRRR (~160 J/kg), which was much higher than the RAP. Storm rotation was also slowly increasing. Interestingly, this occurred at the same time that lightning and reflectivity-based parameters were decreasing. Despite low overall probability for tornado (20-30%), the ramp up, coupled with the fact that PTv3 remains on the low end overall (max of ~60%) could perhaps have tipped off users to look more closely at this developing storm.

Figure 2: Time series of PTv3 and PTv2 for a tornadic storm on the Florida coast, along with severe reports and NWS severe weather warnings.

The next day, Virginia Beach, VA was hit with an EF3 tornado, with peak winds estimated between 140 and 150 mph. Remarkably, no injuries were reported despite damage to 100 homes. In this case, PTv3 exceeded PTv2, and even hit 60%, which is a very high value for v3. The dip in probability shortly before the tornado was likely due to a pronounced reduction in mid-level azimuthal shear, which quickly rebounded (the 1-3 km mean wind also dropped from 37 kt to 30 kt during that time). 

Figure 3: ProbSevere v3 contours, MRMS MergedReflectivity, and NWS severe weather warnings for a tornadic storm near Virigina Beach, VA. Outer contours on ProbSevere objects are colored by the probability of tornado.

Figure 4: Time series for ProbSevere v3 probabilities, along with reports and NWS severe weather warnings.

Eastern Oregon storm

An elongated trough drew in enough elevated CAPE for a few storms to pop up in eastern Oregon. Very strong lapse rates (≥ 8-9.5 C/km) mixed down strong momentum from roughly 4-5 km AGL, producing downed trees and measured gusts to 65 mph. ProbSevere v3 (PSv3) showed rapid increases in the probability of severe at around 00:20 UTC and 00:50 UTC (Figures 1 and 2). 

Figure 1: ProbSevere v3, MRMS MergedReflectivity, and NWS severe weather warnings for a storm in eastern Oregon.

Figure 2: Time series of PSv3 probabilities for the eastern Oregon storm

The increasing MESH (up to around 1"), composite reflectivity (up to 63 dBZ), and ENI lightning (up to 10 fl/min) contributed to the first rapid rise in the probability of severe. A very strong satellite growth rate at 00:46 led to the next jump, up from 49% to 67%. One can see the cooling cloud tops in the GOES-18 visible/IR sandwich imagery that preceded 3 of the 4 severe wind reports (the first wind report was recorded at 00:41 UTC). 

Figure 3: GOES-18 10.3-µm brightness temperature and 0.64-µm reflectance for the storm in eastern Oregon.

In this example, the probability of severe hail seemed to be the strongest value; however, wind was the only hazard reported. While PSv3 generally has improved discernment among severe weather hazards, this would be a good case to look into to see if we can better incorporate model predictors in PWv3 that take into account the mixing of momentum very high in the atmosphere down to the surface, which is relatively common in the western U.S. 

Elevated hailers

Some elevated storms in Wisconsin have been producing one-inch-diameter hail today. One such storm traveled just west and north of Madison, Wisconsin (Figure 1).

Figure 1: ProbSevere contours and MRMS MergedReflectivity for an elevated storm in Dane Co., Wisconsin.

The storms have been forming in a low-CAPE environment (all of the CAPE is from the most-unstable parcel), with decent effective bulk shear. In ProbSevere version 3, ProbHail has been handling this event better than version 2. While ProbWind v2 is elevated (40%), this isn't really a wind threat, but solely a hail threat. We've seen that ProbSevere v3 consistently discerns the most prominent hazard type better than v2, generally speaking.

One-inch-diameter hail was reported at 15:28 UTC, just a few minutes after a large increase in ProbHail v3 (Figure 2), which was partially due to an increase in the maximum composite reflectivity.

Midwest mayhem

Warm muggy weather under northwest flow churned out a number of supercell thunderstorms last Saturday, June 11th, producing huge hail (up to 3.75" in diameter), damaging winds, and tornadoes. 

ProbSevere LightningCast gave a heads up on the quickly developing convection in Nebraska and Iowa (see Figure 1), with lead-times ranging from 10 to 40 minutes prior to the first flashes in several storms (measured from the 25% probability threshold). LightningCast is adept at predicting lightning in the next hour using GOES-R ABI images as predictors.

Figure 1: LightningCast contours, GOES-16 ABI day land cloud convection RGB, and GOES-16 GLM flash-extent density over the Missouri River valley.

ProbSevere v3, being evaluated at the HWT this month, quickly picked up on the developing convection, using the juicy environmental parameters, increases in total lightning flash rates, and strong satellite growth rates to predict increasing probabilities of severe weather. Numerous forecasters at the HWT have noted that increases in ProbSevere helps them identify potential threats more quickly.

Figure 2: ProbSevere v3 contours, MRMS MergedReflectivity, and NWS severe weather warnings.

Several supercells that spawned in Nebraska were soon tornado-warned and at least one produced multiple tornadoes, destroying trees, power lines, outbuildings, as well as industrial buildings in Maryville, KS. ProbTor v3, which is generally more conservative than version 2, had elevated probabilities of tornado before version 2 did. At 21:26 UTC, 6 minutes before the first tornado warning was issued, ProbTor v3 was 25%, whereas ProbTor v2 was only 1% (for the storm in Figure 3). In this case, the strong STP (2.6) and effective storm-relative helicity (SRH, 200 m2/s2), as well as the high MRMS MESH, were contributing to the enhanced probability of tornado for version 3, whereas the 0-1 km SRH in version 2 was relatively low (60 m2/s2). Forecasters should take note when ProbTor v3 is markedly higher than ProbTor v2.

Figure 3: ProbSevere contours (outer contours is colored by ProbTor v3 probability), MRMS MergedReflectivity, and NWS severe weather warnings for supercells in southeast Nebraska.

Cherry County Storm

A storm in Cherry County, NE was moving southeast, in an environment characterized by very high effective shear (70 kt) and lower MLCAPE (700 J/kg). ProbSevere v3 (PSv3) was maintaining a higher probability of severe than v2 (by about 20%). This is likely due to version 2's over-reliance on ENI lightning data. In this case, the storm was very lightning deficient. However, the PSv3 models were able to overcome the dearth of lightning with all of the other data that it uses. The ML models of PSv3 do this better than those of v2. This is just another example of why using fused observations and datasets really help optimize your use of the data.

Figure 1: ProbSevere v3, MRMS MergedRef, and NWS severe thunderstorm warning for a storm entering Cherry Co., Nebraska.

Scottsbluff storm

ProbSevere v3 (PSv3) has shown a consistently higher signal than ProbSevere v2 (PSv2) on storm near Scottsbluff, NE. PSv3 was 20-40% higher during the development stage and as it matured. PSv3 also had consistently (and correctly) lower tornado probabilities than PSv2. The elevated tornado probabilities was likely a result of very high MRMS azimuthal shear, which was a result of sidelobe contamination. In this case, ProbTor v3 correctly didn't emphasize the azimuthal shear as much as PTv2. ProbHail v3 has been the highest hazard probability, whereas ProbWind v2 was the highest for version 2. One forecaster noted that v3 is definitely picking up on the hail threat better than v2. Hail up to the size of 1.5"-diameter has been reported thus far. Forecasters have noted that the higher PSv3 values seems to correlate better with their expectations of the storm.

Figure 1: ProbSevere v3, MRMS MergedRef, and NWS severe weather warnings for a storm near Scottsbluff, NE. 

Storm on the High Plains

A storm developed quickly in eastern Wyoming, but the lowest tilts from the KUDX radar were partially blocked by the Black Hills in western South Dakota. Fortunately, ProbSevere utilizes MRMS, which merges nearby radars observations. However, in this part of the country, the "nearby" radars are not all that close. ProbSevere v3 also uses GOES-R satellite growth rates and output from the GOES ABI+GLM-based IntenseStormNet, which capitalizes on storm-top patterns in visible and long-wave IR bands, and internal electrification. 

In this storm, the GOES-R satellite growth rate helped contribute to higher PSv3 early on this storm's life. You can see the growing towers in GOES ABI imagery (Figure 1).

Figure 1: Growing Cb in northeast Wyoming

ProbSevere v3 (PSv3) was 15-30% greater than ProbSevere v2 for this storm, 15-20 minutes before the first warning was issued. Part of the reason for this was the 4.4%/min satellite growth rate, which was designated as "strong". An HWT forecaster noted that PSv3 had a good handle on this storm.

Figure 2: ProbSevere v3 at the time of the first warning.