Outer Banks tornadoes

An energetic shortwave traversing the Ohio River Valley took on a negative tilt and forced several storms on the North Carolina coast. As is somewhat common for the time of year, the kinematic parameters for this environment were very strong -- effective bulk shear ≥ 60 kts, meanwind 1-3km AGL ≥ 40 kts, and storm-relative helicity 0-1km AGL ~200 J/kg.

One storm in particular spawned two tornadoes: an EF0 at 19:06 UTC, and an EF2 from 19:10 to 19:16 UTC, with large sections of roofs removed from homes and powerlines down. The EF2 twister occurred on Emerald Isle.

ProbTornado (a product of NOAA/CIMSS ProbSevere) ramped up quickly to nearly 70% before the first tornado touchdown, but decreased  to 30% as reports came in. In the animation below (Figure 1), you can see the ProbSevere contours (inner contour) and the ProbTor contour (outer contour) colored using the same colorbar (top of image). Increasing rotation evident in MRMS azimuthal shear products, as well as an increase in ENI total lightning density in a very favorable environment led to the soaring ProbTor probabilities.

Fig. 1: ProbSevere contours with MRMS MergedReflectivity and NWS severe weather warnings.

A time series of this storm shows how the ProbWind, ProbHail, and ProbTor probabilities evolved (Figure 2). After a 70mph straight-line wind report at 19:50 UTC, ProbTor increased to 80% when a second tornado warning was issued, but there were no reports. The storm traveled over mostly water at this point, and was starting to reside in a region of high radar beam height, even at the lowest tilt, which potentially may have affected the azimuthal shears.

Nevertheless, this event shows how monitoring the ProbSevere products (namely ProbTor in this case), may help increase situational awareness, forecaster confidence, and ultimately lead-time to severe convective hazards.

October tornadoes

A seasonably potent, negatively tilted trough traversed the central U.S. with an embedded strong 850mb jet, spawning severe weather from Texas to Wisconsin. 

Early in the day, storms surged through central Oklahoma, where NOAA's SPC had a 2% chance of tornadoes within 25 miles of a point (Figure 1). The tornadic nature of these storms was perhaps a bit of a surprise, as a tornado watch was issued at 13:50 UTC, after numerous twisters touched down in the OKC metro area. Structural damage to homes and overturned cars were reported.

Fig. 1: SPC Day 1 tornado outlook at 1300 UTC with tornado reports.

The NOAA/CIMSS ProbSevere model (version2, v2) picked up on this tornado threat shortly before the first tornadoes touched down. In Figure 2, a new AWIPS-2 plug-in demonstrates how the probability of tornado product (ProbTor) is displayed with the probability of any severe. The outer contour is colored by the ProbTor value, whereas the inner contour is colored by the probability of any severe (i.e., the maximum of ProbHail, ProbWind, and ProbTor). Both contours use the same colorbar. The outer contour only appears if ProbTor is ≥ 3% by default, but this can be changed to any value by editing a menu file (note that this feature can also be turned off if the value is set to an impossible value, such as 101%). The developers hope that this outer contour will give forecasters a better visual and quantitative indication for a potential tornadic threat, while still being able to see the overall probability of severe.

In this line of storms, we can see that the two contours appear to have the same color, indicating that the ProbTor value is greater than both ProbHail and ProbWind. Note that by sampling the storm, you can still see the normal readout of probabilities and predictors for ProbSevere v2 (Figure 3). 

Fig. 2: ProbSevere output, with ProbTor contours (outer contours), along with MRMS MergedReflectivity and NWS severe weather warnings. 

Figure 3: ProbSevere with sampling.

Figure 4 shows the rapid increase in ProbTor from 22% at 13:20 UTC to 60% 6 minutes later. This was largely due to increasing MRMS azimuthal shear in an environment characterized by 45 kts of 1-3 km AGL mean windspeed. Interestingly, these storms had no lightning activity.

Fig. 4: Time series for ProbSevere v2 probabilities, with preliminary severe LSRs and NWS severe weather warnings.

In the afternoon, storms spawned tornadoes in Missouri (Figure 5, Figure 6) and Iowa. A storm in Missouri exhibited large fluctuations of ProbTor before producing a tornado. The fluctuations in this case were tied to changes in the MRMS azimuthal shear. Sometimes noisy doppler velocity data contributes to rapid azimuthal shear increases or decreases, so forecasters should always be monitoring base velocity data as well.

In Iowa, numerous tornadoes were reported, but ProbTor was generally < 20% for many of them. Despite a conducive environment, azimuthal shear values were not very high for these storms.

Fig. 5: ProbSevere, MRMS MergedReflectivity, and NWS severe weather warnings for a storm in Missouri.

Fig. 6: Time series of ProbSevere v2 probabilities for the highlighted storm in Figure 5.

No tornadoes were reported in Wisconsin, but there were numerous wind and hail reports. The still image (Figure 7) and time series (Figure 8) show a strong satellite growth rate, moderate MRMS azimuthal shear, and a brisk 1-3 km AGL meanwind (~30 kts) contributing to a ProbWind of 53%, shortly before a barn and power lines were blown down in Iowa Co., WI. ProbWind later increased to over 60% and a tornado warning was issued (ProbTor = 3%), but no reports were received.

Fig. 7: ProbSevere, MRMS MergedReflectivity, and NWS severe weather warnings for storms in southwest Wisconsin.

Fig. 8: Time series of ProbSevere v2 probabilities for the storm in Figure 7, in southwest WI.

If a tornado blows through the Maine woods but no one’s there to see it, did it really happen?

The title of this post is exactly what the Bangor Daily News asked after a supercell snapped trees in the Maine northwoods on August 29th. The average radar beam height of the lowest tilt was over 2 km for much of this storm's lifetime, but the storm exhibited a very strong increase in rotation in the low-levels and mid-levels at around 19:54Z, a few minutes before the first tornado warning was issued. The storm also had markedly weak lightning activity from the ENTLN.

The NWS in Caribou and the Maine Forest Service did a fly over and looked for ground damage but apparently found none. So in this case, it probably didn't happen, but at least any lumberjacks in the area were warned for this potent storm.

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

Figure 2: Time series of the probability of tornado and constituent predictors for the tornado-warned storm in Figure 1.

Wilkes-Barre tornado

A pronounced shortwave traversed the eastern Great Lakes with a trailing cold front spawning severe storms in New York and Pennsylvania. The Storm Prediction Center forecasted a 2% outlook for tornadoes within 25 miles of a given point (Figure 1).

Figure 1: SPC tornado outlook from 06/13/2018 2000Z. 

Wednesday evening, an embedded supercell emerged from a linear storm segment over north-central PA. Strong low-level rotation, adequate effective bulk shear and meanwind in the 1-3km layer (both 35-40 kts), as well as very high 0-1km storm-relative helicity (> 200 J/kg) produced ProbTor model output over 40% when the storm was first tornado-warned by the NWS. As 0-2km MRMS AzShear decreased, so did the probability of tornado. Then, from 0152Z to 0202Z, the ProbTor value jumped from 20% to over 80% as both low-level and mid-level rotation increased in this storm. A wind report (but likely tornado damage) was received from Wilkes-Barre at 0215Z, with multiple injuries and cars flipped over. See Figure 2 and Figure 3 for a depiction of the evolution of this storm.

Figure 2: ProbTor contours, NWS warnings, and MRMS MergedReflectivity from 0100Z to 0230Z.

Figure 3: Time series of ProbTor and constituent predictors. NWS warnings and preliminary LSRs are plotted as well.

An accumulation of the 0-2km MRMS AzShear nicely shows the cycling nature of the strong low-level rotation in the storm, with white pixels exceeding 0.015 s^-1 over Nordmont and Pennsylvania state lands, and then over the city of Wilkes-Barre (Figure 4).

Figure 4: MRMS low-level rotation track over the Wilkes-Barre, PA region for the evening of June 13, 2018.

The GLM and ABI instruments from GOES-16 also captured the evolution of the storm (Figures 5 and 6).

Figure 5 shows a 4-panel of new GLM products (produced via Eric Bruning's GOES-R Geostationary Lightning Mapper Tools), along with MRMS MergedReflectivity. Here, the FlashExtentDensity is the count of flashes in 10-km boxes over 3 min, updated every min; the TotalEnergy is the accumulated energy of all GLM flashes in each box over 3 min; and the FlashAvgArea is the average area per flash over 3 min in each 10-km box. We see increases in the FlashExtentDensity and the TotalEnergy fields from 01:15Z to 01:20Z, 01:40Z to 02:00Z, and then a smaller increase from about 02:10Z to 02:20Z. The second jump in total lightning activity corresponds well with increased rotation in the storm and increased probability of tornado.

In Figure 6, note the cooling cloud tops as the storm enters Luzerne county. The cooling starting at about 01:40Z corroborates the increased lightning activity from GLM. Evolution of products from infrared and optical imagers (i.e., ABI and GLM) with high temporal resolution data is an active area of research for severe storm nowcasting at CIMSS.

Figure 6: 10.35µm channel from GOES-16 mesoscale sector for 0130Z to 0230Z. Note the cooling cloud-tops prior to and during tornado occurrence.

EDIT:

The National Weather Service in Binghamton, NY has confirmed an EF2 tornado, beginning approximately at 10:00pm EDT.

Location...Wilkes-Barre Township in Luzerne County Pennsylvania
Date...June 13 2018
Estimated Time...1000 pm EDT
Maximum EF-Scale Rating...EF2
Estimated Maximum Wind Speed...130 mph
Maximum Path Width...200 yards
Path Length...0.75 mile
Beginning Lat/Lon...41.2436/-75.8467
Ending Lat/Lon...41.2390/-75.8392
* Fatalities...0
* Injuries...6

GLM parallax issues?

We've been watching t-storms across central/southern SD and northeast ND today. A few storms were severe with reports of hail 1" or greater. I was kind of
expecting to see some GLM FED lightning jumps with these storms as they became severe, but that just wasn't the case.

Could this be a parallax issue?

This far N (roughly a similar latitude to Toronto, only farther W, so that there are even MORE potential parallax problems), the GLM is likely not sampling storms very well (at more of a side angle), so perhaps it can't actually "see" all of the lightning occurring within a storm.

If that is the case, watching for lightning jumps within a storm might not be a good "warning determination method" for a forecaster across the Northern High Plains and Northern Mississippi Valley...

Or...

Maybe we can still see lightning jumps in more northern/western storms, but due to parallax, we should expect more storms to have lower lightning values, hence lower lightning jump thresholds?

Other parallax-related questions I have include:

How does parallax affect other GLM products like the Total Energy and Avg Flash/Group Areas?

If the GLM is sampling more of the side of a storm across the northern US, can we expect storms to look brighter or darker?

Should we expect the areal extent of lightning flashes within a storm appear to increase or decrease as parallax increases?

How parallax affects GLM data is a mystery to me at this point and certainly another area of research that GLM developers should explore.

- Thomas Bell

DLH - GFS CAPE forecast degrading all-sky CAPE

During my mesoscale discussion I blogged about how the all-sky CAPE did not seem to match up well at all compared to SPC mesoanalysis. I compared the all-sky CAPE to the clear-sky only CAPE & noticed that the most questionable areas were blacked out on the clear-sky product, & thus filled in with GFS data. Looking at the 12Z GFS CAPE data it was obviously too high with 2500 J/kg SBCAPE over NW MN & over 3000 J/kg over SE MN. It took me a while to identify that the GFS CAPE was too high & thus the all-sky LAP CAPE would also be too high because I really do not use the GFS much when doing a short-term mesoscale analysis. I think it would be much more useful to fill in the missing clear-sky data with RAP/HRRR/or some other hourly updating field vs the every 6-hour GFS.

22Z LAP all-sky CAPE:

22Z clear-sky CAPE:

Peter Sunday

Highest Flash Event Densities Collocated With Updraft Location

[23:00 UTC] Have tried to overlay semi-transparent GLM data over GOES-East meso-sector visible imagery and the results are pretty useful. For one, the highest flash event densities seem to be collocated well with perceived updraft location (overshooting tops are being used as a proxy), which fits the conceptual model of what one would expect. It should be noted that the values themselves were not quite as useful as seeing the change in the values and the "peaks" in values which were highlighted a bit more by altering the color scale. This certainly is useful in an operational setting in terms of isolating the most important (rapidly developing) updrafts in a quick and timely fashion.

Fig. 1: GLM FED data overlaid on GOES-East meso-sector Ch. 2 data

Rosie Red

CI With Radar

Couldn't tell if there was something weird with the CI probs...it looked like the storms had already developed.  Unless it picked up that more towering CU was developing right next to the new storms.  Which in this case it seems possible as the storms were building off each other and morphing into a line.

-Penny Gardens

*Later note. I tried using this to overlap outflow boundaries with the CI or CI Severe to see if any CU developed along the boundary.  Didn't see anything, but that may be because of the cirrus.
And is there a probability tool for a gust front? I'm always reminded of the Indiana State Fair situation.  I kept seeing gust fronts ahead of our storms today...however the peak wind seemed to be only about 30 mph.  

GLM Average Flash Area - Updraft vs. Non-updraft/Anvil

[22:00 UTC] After an idea was shared with me from a colleague of mine of increasing the transparency of GLM data, I decided to explore its usability overlaid on visible satellite.

More specifically, I took a look at average flash area to see if there was any notable difference between the perceived updraft and anvil areas of mature convection moving through the Duluth CWA.

While data still remains a bit jumpy, I did notice that it seems that the average flash area near the overshooting tops (used here as a proxy for approximate updraft location) seems to be lower/smaller than in surrounding anvil areas or in areas where updrafts do not appear quite as strong (or have become weaker). The average flash area appears to increase as the convection becomes "older" (i.e. new/strengthening updrafts not apparently seen on visible imagery). Fig. 1 shows an animation through time while Fig. 2 shows a still frame at 21:40 UTC.

Fig. 1: animation of 5-min/1-min GLM Average Flash Area overlaid on GOES-East Ch. 2

Fig. 2: still of 5-min/1-min GLM Average Flash Area overlaid on GOES-East Ch. 2 at 21:40 UTC

Rosie Red

DLH - Nifty Convection Monitoring Procedure

2150Z: I decided to try looking at smoothed GLM data overlaid on visible satellite by using the "interpolate image" & "interpolate colors" options along with some transparency in AWIPS. On the same image I also overlaid ProbSevere polygons & GOES CI probabilities. It was much easier for me to interpret GLM trends while having some of the texture from the visible satellite imagery. The identification of new updraft growth in the GLM flash extent density data was much more evident with this procedure vs just using the GLM data alone. ProbSevere polygons were useful in monitoring the intensity trends of the storm at a glance. Finally having the GOES CI probability helped in a zoomed-out view to identify potential area of new convection. I'm impressed at the amount of data I'm able to view in this procedure without the whole thing looking too cluttered.

Loop of above procedure showing storm evolution in SW MN:

Peter Sunday