EWP Blog – Page 3 – Spring Experiments & Beyond

Operational Feedback of Gremlin, Octane, and LightningCast during a Severe Weather Outbreak in Central Oklahoma

Posted in GREMLIN, LightningCast, OCTANE Speed Sandwich, SPG2025 on June 3, 2025 by Kevin Thiel.

I tested the OCTANE, GREMLIN, and LightningCast products during an actual severe weather event on 6/3/2025. My role during this testbed was that of the mesoanalyst.

Initial environmental analysis shows weak to moderate shear, which was determined via ARARS soundings and SPC Mesoanalysis, along with OCTANE imagery showing divergent / accelerating speeds within the storm anvils. VAD hodographs were used as convection developed to see rapid changes within the shear profile during the course of the event (as convection altered the broader environment). Shear increased as the event progressed. OCTANE and LightningCast were both useful showing the uptick in storm intensity as shear increased.

LightningCast was very useful picking out developing updrafts and embedded updrafts within broader areas of convection. We used this product to gauge which updrafts had the greatest potential to become severe in the near term. A strong uptick in lightning would indicate a rapidly strengthening updraft which would warrant further interrogation.

Similar to LightningCast, OCTANE was useful in determining which updrafts were trending towards severe. While in the mesoanalyst role, I would check to see which updrafts looked most intense (warmer colors paired with a very bubbly/convective appearance) and showed strong divergence. Radar analysis would then help us determine which individual cells to warn on, especially if the area of convection is multicellular and warning the entire thing isn’t ideal.

I didn’t use GREMLIN as much, since this area had good radar coverage. However, I did use it to keep tabs on its performance. The product seems to do well with picking out the strongest discrete/semi-discrete cells and potentially struggles with smaller/shallower storms and mergers.

Using these products, and working as a team with good communication, we were able to successfully warn a tornado in the Norman area along with various severe wind and hail.

– WxAnt

PUB LightningCast and GREMLIN Nowcasting

Posted in DSS, GREMLIN, LightningCast, SPG2025 on June 2, 2025 by Kevin Thiel.

LightningCast

For this first day, I started out looking at Lightning Cast to gain familiarity with version 2 and see how it compares to version 1. The first thing I noticed was in southwest Pueblo County, where there seemed to be fairly frequent lightning. Version 1 in the top left panel (Figure 1 below) actually decreased in probability from 70% to 50%, whereas Version 2 in the top right panel remained at 70%. With both GLM and ENTLN depicting ongoing lightning, I think both versions should be showing higher probabilities. I’m wondering if it’s because both versions are so focused on the convection moving into southeast Pueblo County that they’re less focused on the stratiform lightning/less mature convection?

Figure 1: Four panel comparing LightningCast v1 (left panels) and LightningCast v2 (right panels)

Additionally, I tested out using the LightningCast dashboard for Fowler, CO beginning at 3PM MDT. One interesting thing to note was that it seemed to match better with the version 2 LightningCast in AWIPS versus with version 1, however both versions weren’t too far off. In the Figure 2 below, the left panel (version 1) shows between 30-50% probability of lightning, whereas the right panel (version 2) shows Fowler (purple dot in the image) right on the border of the 70% probability. Comparing that to the dashboard (Figure 3) for the same time, the yellow line (version 1) depicts a 54% probability, with the green line (version 2) showing an 84% probability for 21:18Z. With MRMS reflectivity at the -10C level showing a cell up to 42 dBz just southeast of Fowler, I would tend to lean towards utilizing version 2.

Figure 2: LightningCast v1 (left panel) and LightningCast v2 (right panel)

Figure 3: LightningCast Dashboard

One final note on the LightningCast Dashboard – I thought it was interesting to see that version 1 in Figure 4 below, the yellow line (version 1) shows two separate upticks in lightning probability versus the green line (version 2) showing a steady decline in probability.

Figure 4: LightningCast Dashboard

GREMLIN

I was also able to look at GREMLIN, which was my first time assessing this product. Figure 5 below shows a four-panel, with GREMLIN (top left), MRMS Reflectivity (top right), Satellite IR sandwich (bottom left), and GLM Flash Extent Density (bottom right). Just looking at MRMS and IR, the first cell that draws my attention is the cell in southeast Pueblo County as it has higher reflectivities and cooler cloud tops. The cell in southern Otero county looks like the cloud tops are slightly warming with time. However once we start looking at GREMLIN, those two cells look to go back and forth in reflectivity, leading to less confidence in overall intensity. If I were located in an area with poor radar coverage, or if a radar was down and I had to rely on GREMLIN, it may not be straightforward as to which cell could eventually warrant a warning.

Figure 5: Four Panel comparing GREMLIN (top left), MRMS Reflectivity (top right), Satellite IR Sandwich (bottom left), and GLM (bottom right).

That being said, Figure 6 below shows a screenshot of the same four-panel at 21:41Z, which shows GREMLIN having a pretty good grasp on the convection in Stanton and Morton counties (just outside of the PUB CWA). So in this instance, confidence in the GREMLIN product would at least be higher than the previous example shown.

Figure 6: Four Panel comparing GREMLIN (top left), MRMS Reflectivity (top right), Satellite IR Sandwich (bottom left), and GLM (bottom right).

Final Thoughts for Day 1

Overall I enjoyed testing out both of these products. I definitely want to get more hands-on experience with GREMLIN as well as the LightningCast dashboard in order to see these in different scenarios/environments.

– Fropa

LightningCast for Convective Initiation and IDSS

Posted in DSS, LightningCast, SPG2025 on June 2, 2025 by Kevin Thiel.

LightningCast V2 did a great job predicting lighting development with developing convection along a frontal boundary in northwest Iowa. It outperformed version 1, as shown by the loop and images below.

Animated GIF showing LightningCast V1 (top) and V2 (bottom) with the day cloud phase darkened to show detail. The ENI total lighting (yellow CTG flashes, white cloud flashes) is also displayed.

At 1946Z, V2 has a higher probability of lightning (50%) than V1 (30%).

This trend continued throughout, and at 2016Z the first lighting strike was detected. That’s 30 minutes of lead time, which would be helpful for outdoor event IDSS.

LightningCast at 2016Z with initial cloud to ground strike shown in the yellow dash.

– Updraft

Cloud Top Cooling, Cloud Top Divergence, and MESH

Posted in OCTANE Speed Sandwich, SPG2025 on June 2, 2025 by Kevin Thiel.

After looking at the Cloud Top Cooling and Divergence, there was a case from MPX where a severe thunder warning was issued across southwest Minnesota. The cloud top divergence overlaid with visible satellite did a really nice job showing areas of deep convection (texture) with areas of divergence. This signal preceded the MESH severe hail by 5 to 10 minutes.

Animated GIF showing a 4-panel plow with visible satellite (channel 2) underlaid on all 4 images. The contrast between the cold and warm hues in the top left was one of the strongest seen of the day. The other 3 panels show cloud top divergence (CTD) with small smooth (top right), medium smooth (lower left) and high smooth (lower right). MESH greater than 1 inch is shown in the lower left as well using the default MESH color curve. Green is > 1 inch hail, an the occasional yellow pixels are >2 inch hail. The white text on the lower right is just a placeholder to mark areas of high CTD at the start of the loop.

Notice the area of high cloud top divergence at the start of the loop (red, center of panels). However, there is not a coarse texture of the clouds indicating a strong updraft. Later on in the loop, you’ll see that same red area overlaid with deep, coarse texture of rigorous updrafts on the southwest part of the storm. This is followed by MESH output of severe hail.

A SVR was issued at 2146Z, MESH shows 2” hail at 2152Z, and the satellite signature preceded both of those.

At 2145Z, 1 minute prior to severe TS warning issuance. There is a strong signal of cloud top divergence colocated with a rigorous updraft. The MESH (green, lower right) had a hail greater than an inch at this time.

At 2152Z, MESH (green, lower right) had a hail greater than 2 inches at this time.

At 2110Z, MESH indicated another small area of hail greater than 2 inches. There were two reports (green dots) of 1.25” hail with this storm near this time.

– Updraft

GREMLIN and Lightning Cast – Observational Notes and Feedback

Posted in GREMLIN, LightningCast, SPG2025 on June 2, 2025 by Kevin Thiel.

SYNOPSIS – A broken line of thunderstorms lifted north through SE Colorado in a weakly sheared, high LCL environment with modest instability (1000-2000j/kg MUCAPE) and high DCAPE (1000+ j/kg). This environment appears to favor pulse severe potential, with primarily a gusty/damaging wind risk.

OPERATIONAL NOTES AND FEEDBACK – Using GREMLIN and Lightning Cast Together

I used a 4-panel to compare GREMLIN, satellite, radar, MRMS, and LTG Cast data. I’ve not typically used LTG Cast to nowcast the severity of convection, but when combined with GREMLIN, it kind of reminds me of looking for signals in model data. For the most sustained convection, for example, GREMLIN had a fairly consistent signal of 40-50dBZ echoes in tandem with consistently high LTG probabilities. In the past, I’ve typically just focused on GLM lightning data on its own separate from LTG probs. Overlaying LTG Cast probs with GLM data seems to provide a more uniform / smoothed view of the evolution of lightning within convection as opposed to using GLM on its own. GLM can be jumpy at times, which can give the impression that a thunderstorm is weakening. However, if LTG cast probabilities remain high, it may give the forecaster more confidence that a thunderstorm is not weakening. This seemed to be the case with multiple different thunderstorms in SE CO today.

OPERATIONAL NOTES AND FEEDBACK – GREMLIN

It was interesting to note how closely the increase and decrease in GREMLIN reflectivity was tied to the increase and decrease in lightning. The developers noted that this is to be expected. Since GLM data can sometimes be jumpy, and isn’t always reflective of the severity of a storm at a given moment in time, it might be interesting to see if there is a way to offset this. Perhaps there is some way to mesh GLM data with Lightning Cast data (reference the notes in the observation section about nowcasting convective strengths) or through some other means (longer averaging time, etc.). When GLM data isn’t jumpy, GREMLIN seemed to compare very nicely with MRMS. But, when GLM data was jumpy, GREMLIN seemed to struggle some, showing more rapid increases and decreases in reflectivity that what MRMS showed. As an alternative, I could see where simply overlaying LightningCast data on top of GREMLIN data could provide a more “smoothed” and uniform trend in convection over time, in a way that could still provide useful information for warning decisions.

From an operations standpoint, GREMLIN seemed to provide a great overview of convective evolution, especially when overlaid with LightningCast data. It’s possible this could translate to warning decisions, but this initial runthrough with the product suggests its biggest advantage may be nowcasting the general evolution of convection as opposed to making specific warning decisions. Admittedly this is my first use of the product, and I’m looking forward to trying it in future days of the HWT to see if anything different stands out.

– NW Flow

Day 1 – Analyzing GREMLIN

Posted in GREMLIN, SPG2025 on June 2, 2025 by Kevin Thiel.

GREMLIN

An area of showers and thunderstorms formed across southern Colorado this afternoon. A dryline was just to the east and coincided with a gradient of mixed-layer instability. Thunderstorms developed along an arc from Pueblo county to Baca county and slowly moved north-northeastward before weakening in the late afternoon hours. The Clean IR-CH13 band depicted a strengthening storm to the west while warming cloud tops were occurring over Otero county. GREMLIN picked up the deepening convection while quickly weakening convection to the east.

-Eagle

Cloud Top Cooling and Lightning

Posted in OCTANE Speed Sandwich, SPG2025 on June 2, 2025 by Kevin Thiel.

This was a case from MPX that showed two areas of intense cloud top cooling that corresponded to lightning jumps. Overall this is more of an observational post than a clear example of how OCTANE could be used for decision making.

Animated GIF showing two distinct areas of cloud top cooling (red, midway through the loop)

Animated GIF zoomed in on the area of interest, with ENL lightning overlaid. The cloud top cooling lines up well, and slightly precedes, the electrification of the developing cumulonimbus.

– Updraft

DSS LightningCast Dashboard

Posted in LightningCast, SPG2025 on June 2, 2025 by Kevin Thiel.

GOES-East LightningCast for DSS event in CYS CWA

GOES-West LightningCast for same DSS event in CYS CWA over same time period

When using the DSS event/stadium GLM dashboard on the web, with an event that is located in the CYS CWA in the mid-CONUS, there was a significant difference in the probability of lightning from GOES-West compared to GOES-East. The GOES-West data was ultimately better and more reflective of actual lightning trends in that area, despite GOES-East having two mesosectors located over the point in question.

Top panel, LightningCast version 1. Bottom panel, LightningCast version 2.

Meanwhile, in a different area (BOU), comparing LightningCast v1 to v2, it appears that v1 does better in areas with poor radar coverage, while v2 does better in areas with better radar coverage. In the image above, version 1 has a better handle on the isolated first GLM pixel (50%) than version 2 (10%). Meanwhile, the more robust lightning area is more accurately represented on version 2 (which happens to have better radar coverage) compared to version 1.

– prob30

Storm Initiation for FWD and Gremlin gets Going

Posted in GREMLIN, OCTANE Speed Sandwich, SPG2025 on May 22, 2025 by Kevin Thiel.

From a top down perspective looking at Octane products down to Gremlin and actual radar data from KFWS, it provides a clear picture of what you might want to be looking at as far as what the Octane products can show you and to take a targeted approach to potential warning operations. Taking a look at some satellite data, we can see two areas of interest to the north and then a second one to the west where strong Cloud top Cooling and Cloud top Divergence is seen. In addition to all of this, the greatest gradients from the Octane Speed/Direction images show a rather robust cell to the north and west of the Dallas Fort Worth CWA and this is the cell we will be focusing on.Figure 1: Four panel with Octane Speed Sandwich in upper left corner, Octane CTC/CTD with no smoothing upper right, Octane CTC/CTD High smoothing lower right, and Octance CTC/CTD Mod smoothing lower left. Times were between 1902Z-1932Z

Taking a look at the Gremlin data we can clearly see the MesoSector began to pick up on the previously mentioned cells and further strengthening of these cells located to the north and west of the Forth Worth CWA a few minutes prior to MRMS showing some reflectivities, as this initial CI began to strengthen further. We can also see a bit of a lightning jump with these cells as well.

Figure 2: Four panel with GREMLIN Mesosector on the top right, MRMS reflectivity on the lot left, ECONUS Gremlin with GLM flash density on bottom right and finally GOES Meso sector 3.9 Micron Imagery on bottom left. Times were between 1921Z-1945Z

Finally with radar returns finally noting a strengthening trend with the cell of note to the northwest of the Dallas Fort Worth Metro, and with the previous signals from the stronger cloud top cooling and resultant Divergence, we could infer if these storms did develop in our CWA that they had an increasingly likely chance to become severe within the next 30 minutes or so.

Filling in LightningCast Contours in AWIPS

Posted in LightningCast, SPG2025 on May 22, 2025 by Kevin Thiel.

TFX was focused on DSS messaging since it became evident fairly early on in the day that we were not expecting severe convection. The event we had was a State Track Meet with a range ring of 10 miles. Since there were a lot of contours to look at, our group decided to load them as an image and play around with the fill value of the LightningCast probabilities for easier visualization of imminent lightning threat for our partners. To do this, we loaded LightningCast as an image, went into the Change Colors option of the Img LightningCast and set the 10/30/50/70/90 thresholds to match with the colors, including setting 0-10% as transparent. Then we overlaid MRMS on top of it and set everything below 20 dbZ to transparent so we didn’t get any noise from the light showers since we were more focused on the thunderstorms with higher dBZs.

Initial attempt at filling in the LightningCast contours.

Later on in the day, we settled on a less opaque version of the colorbars and we were able to save them such that others in the TFX group could use them on the AWIPS user account as “LightningCastFilled”. This allowed the reflectivity above 20dBZ to stand out more so partners knew where the heaviest rain was without it blending into the bright filled LightningCast.

Final decision on the colormap filling in the LightningCast contours overlaid with MRMS composite reflectivity above 20 dBZ.

Our group members also noticed the default Max and Min for both versions of LightningCast (when loaded as an image) were originally set in AWIPS to random numbers like -20 and 113. Version 1’s default range was different than Version 2’s which added to the visual discrepancy. Before we figured this out, the contours and images did not match up in space (i.e. the image went outside the contour for the same value), but turning on samples revealed they were the same value. In theory, these should be set to 0 and 100 given that LightningCast is a probability. Once we changed these values on the LightningCast Img product in AWIPS to be set to a range of 0 to 100 and reset the colorbar levels according to this scale, they matched up perfectly with the contours. Our suggestion for developers was to ensure the default for these is 0 to 100 in AWIPS if they were ever to be loaded as an image.