Explosive growth in Oklahoma

Convection is growing rapidly in southern Oklahoma, in the midst of a very unstable environment with ≥4000 J/kg of MLCAPE.

LightningCast version 2, with MRMS Reflectivity -10C (left, below) appears to have a 1-2 minute jump on version 1 (right) with the developing convection. Importantly, LightningCast v2 also diminished some false alarm probabilities to the southeast about 5 minutes prior to v1.

We have already received reports of confirmed tornadoes and significant hail (≥ 2” diameters).

LightningCast v2 contours (left) and v1 contours (right), with GOES-East ABI day-cloud-phase-distinction RGB (background) and GLM flash-extent density (blue, foreground).

- Hail yeah

LightningCast V1 vs V2

 When initially getting into the Lightningcast product I noticed just some subtle differences between Lightningcast V1 (left window) vs V2 (right window) around the 1851Z timeframe to 1921Z. Near the Choctow and Pushmataha Counties had jumped up to 30% with even a small area of 50%, while the version 2 had a lower probability of lightning closer to 10-30% during that same timeframe. Will note on both, given the anvil cirrus from a strong storm off to the west may have been obscuring the area which looks like it lead to abnormally higher lightning probabilities than what otherwise would have occurred without the cirrus overhead. That being said, felt the Lightningcast V2 captured the threat better overall and kept the probabilities lower which would have lead me to not mentioning the area for thunderstorm probabilities just yet,

-Sting Jet

LightningCast Comparison

 While the training touted cases where Version 2 (V2) of LightningCast had a clear lead time advantage over Version 1 (V1), many of the new cells I tracked today showed the opposite, with V1 showing higher probabilities sooner than V2.

It seemed like many of the cells were producing lightning very quickly after achieving even modest returns at the -10C level. In this case, the significant instability in the region today led to very rapid vertical cloud growth, which likely outpaced significant radar returns aloft. And while it is purely an educated guess on my part, that might have something to do with the weight given to the -10C reflectivity - if it needs a certain reflectivity threshold to really boost the probability, then the V2 product would be artificially slowing the increase in probabilities. Then in addition to waiting for the -10C reflectivities, you have to add in processing and dissemination lag time.

The V1 product, being based entirely on satellite data, was able to key in on just the rapid vertical growth and boost probabilities based on that alone, and not have to wait for the -10C returns to show up in MRMS.

Fig 1: Loop comparing LightningCast V1 (left) to LightningCast V2 (right)

Fig 2: Comparison of LightningCast V1 (left) and V2 (right) at 2041Z on 19 May 2025, showing V1 being first to have a 30% contour over the cell of interest in NW Arkansas.

Fig 3: Comparison of LightningCast V1 (left) and V2 (right) at 2046Z on 19 May 2025, showing both having a 50% at the same time, though V1 is larger in area.

Fig 4: Comparison of LightningCast V1 (left) and V2 (right) at 2051Z on 19 May 2025, showing V1 being first to have a 70% contour, while V2 still only has a small 50% area.

Fig 5: Comparison of LightningCast V1 (left) and V2 (right) at 2056Z on 19 May 2025, with the first GLM Flash Extent Density return (blue square) noted at 2059Z. Note that V1 had a 70% contour nearly coincident with the GLM Flash square, while V2’s highest return was still only 50%, and well displaced from where the lightning actually happened.

While I don’t doubt that the -10C reflectivity can help in many scenarios, on days like today it didn’t seem to help much, if at all, and in many cases the satellite-only V1 seemed to do a bit better. As I said, it’s my hypothesis that this is due to the rapid vertical development outstripping the production of -10C returns. Additionally, I would be curious if testing out MRMS Vertically Integrated Ice (VII) instead of -10C reflectivity would produce better results (if it hasn’t been tried already)

- Marko Ramius

LightningCast and GREMLIN Identifying Threats Fast w/ OCTANE Being Great with Initiation

 LightningCast seems to be doing its job quite well as it honed in on a small area in southeast KS with ample lead time (2042Z with more strikes appearing at 2056Z west of Jasper). This would be extremely useful for event deployments and getting ample lead time for our partners as seen below.

Regarding GREMLIN, it does a spectacular job identifying CI along boundaries (in this case, the dryline), somewhat ahead of time compared to MRMS reflectivity.

In this case, it appears to have picked up on more robust CI well ahead of time in TOP’s area. However, it does not appear to be super consistent later on with really capturing how robust some of these cells ended up later on, likely due to overall resolution of the product as well as lightning activity as these cells matured.

Lastly, regarding OCTANE, it did a great job on picking up a cell with ample cooling initially (reds and yellows) followed by strong divergence aloft (purples and pinks). This storm would go on to further intensify down the road.

This cluster of storms would then go on to produce multiple 60mph+ severe gusts.=

- Ryan Cooper

GREMLIN vs MRMS vs Z

 When initially digging into the GREMLIN product (since I had never seen or worked with it before), I wanted to see how it looked in comparison to reflectivity products that I use daily. Loading a 4P with MRMS RALA in the top left, GREMLIN in the top right, and KTLX/KSRX base reflectivity in the bottom left and right, respectively, helped me visualize the differences. One thing I noticed is that the GREMLIN product had an odd jump in dBZ into Okfuskee County at 19:50Z Mon 19-May-25 that was not reflected by the base reflectivity products. This appears to be because GREMLIN’s view is from the top down rather than the top up, so it’s sampling more of the lightning in the anvil blowing off downstream to the north, causing an apparent downstream jump in reflectivity. I think this would be useful in anticipating where downstream convection is going to grow based on the lightning presence and anvil presentation on the satellite view with GREMLIN.

- millibar

GREMLIN Comparison

 While it was noted in training materials that the GREMLIN based on 1-minute mesosector imagery would be a bit ‘noisier’, I wasn’t quite prepared for just how much that could show up. This is likely due to the use of the 1-minute lightning data, which will naturally be noisier than the 5-minute data for the CONUS-based GREMLIN. As you can see in the GIF below, this shows up especially strongly in the anvil regions of mature storms, where lightning in the anvil may be less frequent, and thus more noisy in the 1-min data, but the noisy data does show up somewhat elsewhere, too.

Comparison of GREMLIN based on GOES East Meso-1 (left) and GOES CONUS (right)

This would be a case where one would want to have the image looping to make it obvious what was going on, so you could know to ignore those simulated ‘returns’.

The mesosector GREMLIN also tends to have a bit ‘sharper’ resolution in most of its features, and also tends to have higher peak simulated reflectivity, again possibly/probably owing to the 1-min lightning data, which should be a bit less broadbrushed compared to the 5-min data used for the CONUS GREMLIN.

- Marko Ramius

Day 1 GREMLIN and LightningCast

 Noticed that the meso GREMLIN was a little jumpy when it came to the apparent ‘strength’ of the storm. Tried to get a gif of it, but this isn’t the best…

Noticed that the CONUS did a better job of matching what the radar was showing (at this time frame). Same issues as above with the ‘jumpy-ness’.  GREMLIN seemed to pick up more on the new convection (cell furthest to the south in image below).

When using ProbSevere with GREMLIN, can more easily see the slight shift in what it shows vs radar

LightningCast V2 appeared to consistently do a better job

- Lightning McQueen

4-Panel GREMLIN Satellite

 We discussed how we would handle warning this cluster of storms around 20:30Z Mon 19-May-2025, whether we would do one big SVR warning encompassing the whole cluster or concentrate into individual storms with different hail sizes and/or wind speeds. If I was warning this and saw a similar presentation in the radar for both clusters, I would begin with a larger SVR encompassing both storms.

The satellite product I use most often for cloud top cooling/warming for diagnosing convection growth is the Ch 13 IR. By using Ch 13 IR satellite and GLM Flash Extent Density, that helped me determine which updraft was the strongest and/or tallest. The overshooting top visible in the bottom left corner with the northern storm cluster suggests a quickly growing updraft that may start to produce large hail (and/or damaging winds) soon given the lapse rates and the explosive environment. Corroborating this with GLM FED and discussing with the group, we also came to the conclusion that the southern cluster of storms is probably broader but not as strong given its broader but less concentrated lightning presence, and might even be weakening below severe limits. With this information, I would have probably SVSed my warning to only include the rapidly growing northern cluster and maybe upping the hail size.

GREMLIN also tends to agree that they start off with relatively similar intensities, but eventually the northern storms take precedence. Using all four of these products together from the start to finish of the storms’ lifecycle would inform my warning decisions for initial issuance, SVSing, and eventual EXPing or reissuing downstream.

4 Panel loop below.

Loop of GREMLIN (top right), Composite Reflectivity (top right), Ch 13 IR (bottom left), GLM FED (bottom right)

- millibar

Large GREMLIN Differences in Western NC

GOES-E Mesoscale and GOES-E CONUS GREMLIN output displayed large differences for a storm in western NC.

GOES-East Meso (left) and GOES-East CONUS (right) GREMLIN output at 1951 UTC 08 May 2025.

A GREMLIN developer attributed this difference to most likely be related to the integration of lightning (1-minute for the mesosector, 5-minute for CONUS).

Odd LightningCast Dashboard Reading

 An odd or false reading from LightningCast occurred on the Dashboard readout on May 8, 2025.  The IDSS event was the Southern Skies Music Festival in Knoxville, TN. The screenshot below shows the Max Lightning Potential (10-mile radius) increasing dramatically after 14:40 UTC while the other LightningCast options don’t suggest any potential for lightning (in next 60 minutes) until after 15:45 UTC and even then, the others didn't have any probability over 30%. On the plainview map comparing LightningCast 1 vs LightningCast 2 (Figures 1, 2, 3), the contours finally overlap the 10-mile event ring by 20:31 UTC and match up with the Dashboard probability at the same time (See Figure 2).  Discussions within the HWT noted the issue could likely be a domain or pixel issue for the Max P.  

Screenshot of the LightningCast Dashboard for Southern Skies Music Festival on May 8, 2025 from 19:40 UTC to 21:25 UTC. Interesting to note the blue arrows pointing to high probability of lightning depicted by the Max while the others remained less than 10% until 15:45 UTC.

Figure 1: At 19:46 UTC, the lightning potential approaches the 10-mile radius when the Max probability suggests nearly a 50% probability of lightning in the next 60 minutes.

Figure 2: At 20:31 UTC, the 10 and 25 probability of lightning enter the 10-mile radius.

Figure 3: At 21:26 UTC is when the probability of lightning really tapers off, even the Max P version.

Loop of the LightningCast version 1 and 2 surrounding the Southern Skies Music Festival.

- Podium