Friday, June 13, 2025

LightningCast v1 vs. v2 in Texas and Alabama

The LightningCast model predicts next-hour probabilities of lightning. Version 1 (LCv1) of the model uses solely GOES-R ABI inputs (C02, C05, C13, and C15), whereas the experimental version 2 (LCv2) incorporates MRMS Reflectivity -10C in addition to the four ABI image-inputs.

While LCv1 provides accurate predictions in most circumstances, LCv2 significantly improves predictions under thick ice. This case in Texas demonstrates much higher probabilities from LCv2 (left, animation below) under a plume of thick ice with very little cloud-top texture or temperature contrast. The convection underneath was likely somewhat shallow, as storm tops were not poking out from the residual thick ice, but still produced lightning for several hours. 

LCv2 (left) and LCv1 (right) probabilities (contours), GOES-19 ABI day-cloud-phase-distinction RGB, and GOES-19 GLM flash-extent density observations over north Texas and southern Oklahoma.

Meanwhile in Alabama, cumulus clouds bubbled up along what appeared to be a residual thunderstorm outflow boundary. LCv2 consistently produced probabilities 10-20% higher than LCv1 for the congestus clouds that would go on to produce frequent lightning.

LCv2 (left) and LCv1 (right) probabilities (contours), GOES-19 ABI day-cloud-phase-distinction RGB, and GOES-19 GLM flash-extent density observations over eastern Alabama.

This convection happened to develop right over Anniston Metropolitan Airport in eastern Alabama. The lightning dashboard at that site plots both the LCv1 (red) and LCv2 (green) probabilities produced by the 5-minute CONUS sector (below). We can see that LCv2 consistently held higher probabilities 10-15 minutes prior to when LCv1 achieved the save likelihood. LCv2 had an elevated probability of lightning (>30%) about 50 minutes prior to the first GLM flash within 5 miles of the airport.

Lightning dashboard for Anniston Metropolitan Airport, with LCv1 (red) and LCv2 (green) probabilities, and GLM flash rates within 10 miles (small blue dots) and 5 miles (large blue dots) of the airport.

Scientists are working to evaluate and improve LightningCast where we have good MRMS data, while maintaining good skill and first-flash lead time where only GOES is available. 


Thursday, June 5, 2025

Southwest Texas Lightning Product Performance

 Lightning, and GOES-East vs. GOES-West

Observation 1: GLM discrepancies between satellites

Left: GOES-East LightningCast v1 & GLM FED        Right: GOES-East LightningCast v2 & GLM FED 1836 UTC - 1931 UTC 5 June 2025 in southwest Texas

GOES-West GLM 1911 UTC - 1933 UTC 5 June 2025 in southwest Texas (LightningCast outside the domain)

The Midland-Odessa (MAF) forecast area (and nearby upstream areas in Mexico) sits in a weird position where it is well within the GOES-East CONUS domain, but on the edge of the GOES-West CONUS domain (and thus outside the CONUS LightningCast domain), yet within the GOES-West full disk domain. The above images show GLM observations in southwest Texas from both satellites, where GOES-East shows far less lightning (and a downward trend) while GOES-West showed significantly more lightning at the same time (also with a downward trend, but still indicating a stronger storm).

New Mexico and Texas Tornado Outbreak

 We did warning operations today across the Lubbock TX CWA. Storms were beginning to fire up as we started, but took a while to mature. Multiple convective attempts were needed before a primary updraft took root. Seen below, is the transition from a sputtering narrow convective column to a deep mature storm. OCTANE helped pin-point this motion showing warmer/faster colors and stronger divergence.

An old outflow boundary, from yesterday’s activity across New Mexico and Texas, served as focus for convection. Soundings showed steep lapse rates through the column south of the boundary with a moist boundary layer.  North of the boundary, the boundary layer was stable as indicated by the billow clouds / undulating smooth stratus. Hodographs were long and straight, but we deduced that a tornado threat existed immediately along the boundary and perhaps a bit to the north where low-level helicity was more enhanced. We also surmised that the tornado threat would increase with time as the low-level jet ramped up and elongated the lowest portions of the hodograph.

Storms were initially in ABQ’s CWA and we issued our first warning around 2130z with a duration of one hour. I heavily used storm top divergence within my warning strategy as OCTANE showed 100+ knots and radar showed 145 knots at times. Combined with a BWER to 35k ft (-20C level was 22k ft) and 50dbz to 45k ft…I went with 2” diameter hail and 70mph winds along with possible tornado. I did not go tornado warning simply because that part of the storm wasn’t close enough to our CWA yet.

It was interesting to  note that the massive anvil from each of the two supercells merged causing GREMLIN to show only one primary core. The extreme DBZ values within gremlin added confidence to the severity of the cells within the anvil. Another interesting thing is that the northern cell died down shortly after this image was captured resulting in only one primary cell reminiscent of the above image.

Towards the end of operations the now mature primary supercell showed strong divergence within OCTANE and a wake near and immediately downstream of the overshooting top, indicative of a very strong and likely severe storm.

We ended up issuing a tornado warning as the storm neared the state line, which verified as broadcast media reported a tornado just east of the Texas line. Our warning had about 17 minutes of lead time. Additionally, 1.75 to 2 inch hail reports were on the New Mexico side of the state line during the time of our SVR warning…with a 3 inch hail report coming in on the Texas side as the testbed came to an end.

Overall, OCTANE aided our warning decisions significantly…the addition of satellite-derived storm top divergence was a good confidence nudger and helped us get half an hour to an hour lead time on large hail and over 15 minutes on a tornado. LightningCast and OCTANE helped guide us toward which cell would eventually become dominant during the initiation phase.

- WxAnt

OCTANE Speed and Direction Differences

 

OCTANE Direction

OCTANE Speed

One thing I noticed when flipping back and forth between the OCTANE speed and direction products is that there were areas where the Direction dropped out completely while Speed remained fully intact.

Top left: OCTANE Speed, top right: OCTANE Divergence (no smooth) & Cloud Top Cooling, Bottom left: OCTANE Divergence (med smooth) & Cloud Top Cooling, Bottom right: OCTANE Divergence (high smooth) & Cloud Top Cooling

Top left: OCTANE Speed

top right: OCTANE Divergence (no smooth)

bottom left: OCTANE Divergence (med smooth)

Bottom right: OCTANE Divergence (high smooth)

An additional observation and preference I had was that having the cloud top cooling and divergence overlaid could be helpful in some situations (initiation), but when monitoring for divergence signals, there were several occasions where the strongest divergence signals were far harder to spot or completely covered up by a cooling signal that wasn’t even that strong. This was especially true for when convection was ongoing/mature, when the cooling signals weren’t particularly strong because the tops couldn’t cool much more than they were. My personal preference would be to have separate procedures for convective initiation and mature/ongoing convection to have a more clear visualization of relevant features/products in the procedure when there are several areas to monitor for convective development. Perhaps this would be something I would remember to toggle on/off with time based on the situation, but wasn’t intuitive for learning how the products work and what scenarios each product could/should be used in.

-prob30



False alarm area reduction at DSS event with LightningCast v2

 A simulated DSS event was placed in Garden City, KS, ahead of expected convection forced via cyclogenesis in eastern Colorado and a strong mid-level jet streak.

As severe storms began to approach the vicinity, LightningCast v2 (with Ref. -10C predictor) held probabilities in the 20-30% range, whereas LightningCast v1 (with ABI-only predictors) had probability of lightning in the 50-60% range for an hour, topping out at 75%. Forecasters could see the differences on the DSS lightning dashboard (Figure 1). No flashes were observed within 10 miles of the DSS location.

Figure 1: Lightning dashboard showing LCv1 (red) and LCv2 (green) time series of next-hour probabilities of lightning. No GLM flashes were observed during this time.

When we spatially compare LCv1 and LCv2 probabilities, we see that v1 contours extend much further northeast into the large anvil cloud towards Garden City, KS, whereas v2 probability contours are correctly more conservative.

This example demonstrates the benefit of data fusion: The ABI channels cannot “see” under that very thick ice, but the radar was showing no new convective development. Therefore, the LightningCast v2 model could reduce false alarm area for this DSS event, only needing to factor in anvil lightning potential and (to some extent) the motion of the storms.

Figure 2: Toggle of LCv1 and LCv2 probabilities. LCv1 probabilities extend further northeast towards Garden City, KS, creating more false alarm areas. Note well that these contours are NOT parallax corrected, but the probabilities in Figure 1 are. GOES-19 C02 reflectance and C13 brightness temperatures are plotted in the background, whereas GOES-19 GLM flash-extent density are the blue-to-yellow foreground pixels.

- Hail yeah

Supercells in Southwest Kansas

 Convection crossing into Hamilton County

Pretty much right at the start of operations, convection over southeast Colorado began pushing east into Hamilton County, KS. Looking at the first gif below (Figure 1 with the Octane 4-panel), you can really see a persistent divergence signal as the storm continued into the county. Additionally, probsevere (located in the top left panel) was pretty high, maxing out above 90%. Next, looking at the LightningCast panels in Figure 2, you can see the lightning jump occurring right as it crosses over the county line. Utilizing these products together and noting radar showing a mid-level meso with 50 dBZ over 35kft, we felt confident to go ahead and issue a Severe Thunderstorm Warning for this cell (shown in Figure 3).

Figure 1: Octane 4-panel with ProbSevere overlaid in the top left panel

Figure 2: LightningCast v1(left) and v2(right) with GLM

Figure 3: Radar Reflectivity with the Severe Thunderstorm Warning

GREMLIN (top left panel in Figure 4 below) did a pretty good job showcasing this cell, as well as another strong cell just to the southwest, however compared to MRMS (top right panel), it didn’t capture the stronger reflectivities as well, and was approximately 5-10+ dBZ off. So if I didn’t have radar access and could only rely on GREMLIN, I may not have felt as strongly about issuing a SVR.

Figure 4: GREMLIN 4-panel

I didn’t grab images of this, however later on, there was a clear decrease in the lightning activity, with noticeably lower divergence and cloud top cooling. This gave me the confidence to cancel my warning early.

Octane and ProbSevere

Later on, the same cell over Hamilton County began slightly cooling again, with ProbSevere noting 59% probabilities (top left panel in Figure 5). Additionally, there was a very clear mesocyclone noted in Figure 6. Using just these two products, I may have been inclined to issue at least a SVR warning. However Octane wasn’t noting much, if any, cloud top divergence or cloud top cooling. Lightning also didn’t look very impressive either. Noting this, I strayed away from any warning issuance (especially considering radar was sampling this storm at 12.5kft), which I think was a good call, at least for this time.

Figure 5: Octane 4-panel with ProbSevere overlaid in the top left panel

Figure 6: Storm Relative Velocity

LightningCast Dashboard for the DSS Event in Dodge City

Closer to the end of operations, the LightningCast dashboard (Figure 7) started showing higher probabilities of lightning, with the Max prob for a 10-mile radius showing 77%, v1 at 45%, and v2 at 30%. There were two supercells several counties west of the event that were expected to track southeast, missing the venue, however with the dashboard and CAMs showcasing the potential for lightning to reach the event, we were confident enough to fill out a DSS form and graphic (shown in Figure 8) with this information.

Figure 7: LightningCast Dashboard

Figure 8: DSS Graphic

It should be noted that v2 only highlights a 30% probability, and later on was ~44% lower than v1 (v1 was at 76%, with v2 at 32%. With this being at the end of operations, we couldn’t see if lightning actually occurred at the event, but I’d be interested to see if v2 actually had a better grasp on the convection tracking southeast and missing the event altogether, or if v1 showcasing the higher probabilities was the better option.

Final Thoughts from Day 4:

The Octane product really shined today, both in increasing my confidence to issue a SVR warning, and in talking me down from issuing a separate warning. I’ve been sold on the Stoplight colorcurve, with no smoothing (top right panel) coming out on top.
- Fropa

Octane Warnings

 For Day 4, we were DDC. The storm environment was favorable for tornadoes. Our group issued four tornado warnings. This was based on the Octane storm top divergence, ProbSevere, radar, and the near storm environment. Radar confirmed a supercell with a mid-level meso, but the beam was +10Kft above ground, so it wasn’t helpful for the low levels. We kept the tornado warning going based on the environment, and did add 2.5” hail to the warnings. This storm ended up producing 2-3” hail, and multiple tornadoes. The tornado reports were delayed.

The animated GIF above shows the tornado warnings (red polygons) along with MESH (top right) and no smooth cloud top divergence (bottom left) and high smooth cloud top divergence (bottom right). This was a strong supercell in a favorable environment for tornadoes, so warnings were issued despite the lack of reports or low level radar data.

In addition to issuing the first warning with about 30 min lead time for our forecast area, the Octane was also able to identify a cell that developed a couple hours later behind the main storm, and it showed it as sub-severe. At quick glance, these updrafts look similar on the Day/Cloud (see below), but Octane did not have strong divergence with that second cell, and MESH had only small hail. For that reason, we didn’t issue a second warning, which I think is a big feat. Sometimes not issuing can be just as important as issuing.

The image above shows two updraft cores on the Day Cloud Phase at 2124Z. These updrafts look similar but the Octane Divergence was able to differentiate and show that updraft 1 had less divergence than updraft 2. (See image below)


The image above shows Octane Cloud Top Divergence (no smooth left, high smooth right) at 2125Z, corresponding to the Day Cloud Phase image above. This product, in addition to MESH, gave confidence to hold off on a warning for that trailing storm to the west.

The animated GIF above shows the two updrafts mentioned above, along with the tornado warning (red box) toward the end of the event.  MESH is also shown in the top right, with a filter for values less than an inch. Storms are showing a weakening trend in the Cloud Top Divergence (bottom panels) and MESH (top right).

- Updraft




Lightning Dashboard IDSS

 For Day 4, we were DDC. There was an outdoor event ahead of the convection, and prob lightning did a great job 1) detecting some convection ahead of it and 2) showing that it would be east of our IDSS event.

The animated GIF above shows both versions of the lightning cast, with the outdoor event marked by the “Home”. In this case, both versions accurately predicted lighting (white dashes), and also showed it staying east of the event.

- Updraft

Day 4- Large Hail in Texas

 Conditions were favorable for severe thunderstorms across west Texas on Thursday, June 5, 2025. High instability and shear led to rapid development thunderstorms across the high plains. Initially, a cluster of thunderstorms congealed into one which then split into two. The right mover went on to produce at least 3” in diameter hail and wind gusts in excess of 72mph. Due to the poor radar coverage in west Texas, GREMLIN was useful in supplementing radar and MRMS data. GREMLIN showed features I hadn’t seen in the start of the week, including bean-shaped storms and double updrafts that later split into two storms.

GREMLIN had some odd things occur too. ECONUS GLM did not represent the lightning that was occurring with these storms and it is believed that this may have degraded GREMLIN.  In the loop below, ECONUS GREMLIN produced a fictitious cell northeast of the left and right movers. It also lagged a bit before it showed two strong cells, especially the southern storm.

Four Panel-GREMLIN EMESO-1 (top left,) MRMS -10c (top right,) GREMLIN ECONUS (bottom right,) and CH07 (bottom left.) In this example you can see the stronger storm to the south

Here is a single image of the bean shaped cell that produced at least 3” hail.

LightningCast was useful for situational awareness and messaging our partners for the DSS event. We noticed V2 was a little better at maintaining the high probability of lightning (greater than 90%) than V1 in mature convection.

LightningCast proved to be useful for our fictitious DSS event in west Texas. V2 was faster to increase the probability of lightning prior to lighting occurring at the event (below.) It was also faster to show lightning cessation. There was a brief lull in lightning mid-way through the operational period and both V1 and V2 showed about a 20 min lead time of the probability of lightning decreasing in the next hour. V2 stayed slightly elevated compared to V1, but both highlighted that there was still a high probability of lightning in the next hour.  

Lastly, OCTANE (below) proved to be useful again in warning operations. Robust, mature convection was ongoing and while it was “off to the races” in west Texas, the speed and divergence products continued to the tight gradient of speed divergence. We noticed that the compressed color scale was more “eye catching” to show the tight gradient. Below is a picture of the speed/direction divergence product with sampling turned on. The overshooting top was at an impressive -85C with winds out of the west at 50 mph. In this example, the gradient on the west side of the storm helped maintain our confidence of a powerful storm.

- Eagle

Nowcasting Convective Initiation and Evolution with OCTANE, GREMLIN, and LightningCast

 SYNOPSIS - A remnant outflow boundary from previous convection was laid out across portions of west Texas. Heating and destabilization occurred along this boundary, setting the stage for scattered supercell development where convergence was maximized. Below are some observations and notes regarding data seen in OCTANE, GREMLIN, and LightningCast.

OPERATIONAL NOTES AND FEEDBACK - Nowcasting Convective Initiation with OCTANE and LightningCast.

As is so often the case in setups like this, there was a fairly large area of towering cumulus clouds and occasional attempts at convective initiation (CI) along the outflow boundary. For today’s operational period, I tried a 4-panel combination that included OCTANE products, LightningCast, and traditional satellite imagery (Figure 1).

Figure 1 - OCTANE Speed (top left); OCTANE Cloud Top Cooling/Divergence-no smoothing (top right); OCTANE Cloud Top Cooling/Divergence-medium smoothing (bottom left); VIS/IR Sandwich + LightningCast (bottom right)

As was seen in previous days of this experiment, OCTANE products helped to identify areas of towering cumulus that had the most vertical growth and, consequently, the greatest near-term potential for CI. This is evidenced by the green shading in both the OCTANE Speed and OCTANE Divergence products in Figure 1. One of the potential enhancements to nowcasting CI is LightningCast. While OCTANE appears to excel at depicting where towing cumulus has the most vertical development, LightningCast appears to excel at pinpointing which of those areas of more developed cumulus have a better chance at producing lightning. This essentially provides additional probabilistic information for trying to determine where CI may soon occur, which trickles down to knowing where to focus pre-CI IDSS.

In Figure 1, for example, an increase in lightning probabilities was observed just west of the WFO Lubbock forecast area, where OCTANE was highlighting increased vertical development of towering cumulus clouds. The increase in probabilities, plus consistently high probabilities, provided lead time on where CI would eventually occur. Note the consistently high probability of lightning in that area, as well as increasing probabilities to the south. Two separate supercells eventually developed in this area, and later went on to produce severe weather.

OPERATIONAL NOTES AND FEEDBACK - Nowcasting Convective Evolution and Intensity with GREMLIN.

For today’s operations, radar coverage was not a challenge, but GREMLIN data was still analyzed to determine if there is utility even in cases where radar coverage is sufficient. Today’s event highlighted several important observations about GREMLIN.

1) This is the first event during this week’s experiment in which 60+ dBZ echoes showed up on GREMLIN. This closely matched the 60+ dBZ cores seen on MRMS (as shown in Figure 2 below). It was noted by the developers that 60+ dBZ echoes offer a high likelihood of severe thunderstorms and this was, indeed, the case in today’s event. This suggests that GREMLIN may offer increased confidence in the occurrence of severe weather regardless of good or poor radar coverage.


Figure 2 - GREMLIN Emulated Reflectivity (top left); MRMS Reflectivity (top right)

2) For sustained, deep convection with cold cloud top temperatures and lightning in the anvil, GREMLIN may sometimes erroneously equate this to precipitation occurring at the surface. Note in Figure 3 below the area of 40+ dBZ echoes depicted by GREMLIN over the northwestern part of the Lubbock CWA. This area corresponds to nearby GLM lightning, but note that no precipitation is depicted by MRMS reflectivity or WSR-88D reflectivity (not shown).

3) When thunderstorms are close together with merging anvils, be aware that GREMLIN may treat this as one thunderstorm, instead of multiple thunderstorms, as seen in Figure 3. Note the one cell seen in GREMLIN data near the far western edge of the Lubbock CWA, then compare that with the two cells that actually existed (as seen in MRMS data). It should be noted, though, that GREMLIN did well to capture the most dominant of the two thunderstorms.

Figure 3 - GREMLIN Emulated Reflectivity (top left); MRMS Reflectivity (top right); GLM Lightning (bottom right)

OPERATIONAL NOTES AND FEEDBACK - Nowcasting Updraft Strength with OCTANE

Today’s event provided an interesting opportunity to nowcast two strong updrafts in close proximity to each other. In Figure 4 below, the OCTANE Speed product reveals a strong updraft developing within the divergent area of a nearby thunderstorm’s updraft. The developers noted that this is impressive because essentially the former updraft is strong enough to oppose the already enhanced, and modified, flow around the latter updraft, suggesting an increased risk of severe weather. The OCTANE Speed product appears to better depict what is happening to the flow aloft around both updrafts when compared to the VIS/IR sandwich product alone. I suggest using OCTANE and the VIS/IR sandwich product in tandem with each other when nowcasting convective updraft evolution.

Figure 4 - OCTANE Speed (top left); OCTANE Cloud Top Cooling/Divergence-no smoothing (top right); OCTANE Cloud Top Cooling/Divergence-medium smoothing (bottom left); VIS/IR Sandwich (bottom right)

- NW Flow