Using GREMLIN and OCTANE to Supplement 88D Radar Data

 SYNOPSIS - Scattered to numerous thunderstorms developed along a slow-moving frontal boundary from Illinois to Arkansas. The environment was weakly sheared (20-25kt effective) and moderately unstable (1000-1500j/kg MLCAPE). Of note, along the frontal boundary there was a plume of higher 3CAPE (125-150j/kg) that developed near and to the southwest of the St. Louis metro.

OPERATIONAL NOTES AND FEEDBACK - Using GREMLIN and OCTANE in the absence of radar data

For today’s experiment, we were given satellite and lightning products, but no radar data. This presented a unique challenge, but also a unique opportunity to test the performance of new technology in a live warning environment with limited data.

Since GREMLIN is designed to mimic radar, it stands to reason that it could be a decent backup in situations where radar data is missing or significantly lacking. However, because it mimics composite reflectivity, which doesn’t tell the full story of a thunderstorm, additional information is needed for warning decisions. This is where OCTANE potentially comes in.

From a probabilistic standpoint, OCTANE products have shown skill in identifying which thunderstorms carry a higher probability of severe weather than others. My goal today was to determine if OCTANE and GREMLIN products could be combined to provide a warning forecaster a reasonable starting point for warning decisions and warning creation. OCTANE was used to help identify which thunderstorms had a higher probability of becoming severe, while GREMLIN data was used in tandem with lightning data to draw appropriate warnings.

In Figure 1 below, the OCTANE speed product showed an updraft over the central part of WFO LSX (SW of St. Louis) that was strong enough to impede the background flow aloft, suggesting a notably strong updraft. At the same time, there was a significant increase in lightning noted. This information seemed to suggest an increased risk of severe weather with the associated thunderstorm, and a Severe Thunderstorm Warning (yellow box) was issued at 1944z. GREMLIN data was used in tandem with GLM and ENTLN lightning data to help draw the most appropriate polygon possible (Figure 2). Seven minutes later, at 1951z, wind damage and a tornado were reported in the warned area. While this is only one case, it suggests that with satellite and lightning data alone, there is an opportunity to get lead time on severe weather, even in the absence of traditional radar data. It is also noteworthy that GREMLIN emulated radar reflectivity seemed to take on a bit of a bow shape which was later discovered to be not too dissimilar from actual WSR-88d radar data from WFO LSX. This is something I didn’t expect, and was a pleasant surprise. It also seems to give even more credence to the idea that GREMLIN could be a sufficient backup where radar data is lacking.

To me, this suggests an exciting and promising future for OCTANE and GREMLIN products, especially when used in tandem with each other, and especially in areas with limited, or no, radar coverage. For general comparison with GREMLIN data, Figure 3 shows a short radar loop (from WFO LSX) of the thunderstorm that was warned on. This radar data was analyzed after today’s experiment was over, and no radar data was used for warning decisions during today’s experiment.

Figure 1 - Octane Speed + convective warnings (top left), Octane CTD_no smoothing (top right), Octane CTD_medium smoothing + GLM lightning (bottom left), Octane CTD_high smoothing (bottom right)

Figure 2 - GREMLIN emulated reflectivity + convective warnings

Figure 3 - WSR 88D Radar Data from WFO LSX

It should be noted that while lead time for severe weather was achieved, the tornado report came without a Tornado Warning. While GREMLIN and OCTANE appear to have skill in identifying which thunderstorms may need a warning and which may not, there appears to be less skill in identifying specific hazards (tornadoes, large hail, damaging winds). This is where traditional 88D radar excels, especially since the advent of Dual Pol data. This is another great example of the need for skillful mesoanalysis during convective events, especially during radar outages. GREMLIN and OCTANE products can provide a probabilistic foundation for where to issue warnings, while skillful mesoanalysis can help provide a probabilistic foundation for what type of warning to issue and for what types of hazards.

- NW Flow

LightningCast v1 vs v2 in far southwest Texas

 Today forecasters had an opportunity to look at LightningCast v1 (ABI only) and v2 (ABI + Ref10 inputs) in far southwest Texas, where radar coverage is quite poor.

LightningCast v2 for convection in southwest Texas. Background is GOES-19 C13, foreground blue pixels is GLM flash-extent density. Contour legend: green = 10%, yellow = 30%, orange = 50%.

LightningCast contours v1 for convection in southwest Texas. Background is GOES-19 C13, foreground blue pixels is GLM flash-extent density. Contour legend: green = 10%, yellow = 30%, orange = 50%, red = 70%.

For this area of developing convection, v1 was more bullish on two areas of convection that indeed became thunderstorms.

The signals in the day-cloud-phase-distinction RGB, which are represented as inputs in both versions of LightningCast, were indeed indicative of glaciated convection and high lightning potential.

GOES-19 day-cloud-phase-distinction RGB for southwest Texas (courtesy of College of Dupage NEXLAB).

This region was on the edge of the valid MRMS domain for the Reflectivity -10C predictor (see below).

Reflectivity at -10C. Gray is invalid regions, while the rest of the domain is considered “valid” to the LightningCast v2 model.

It’s possible that LightningCast v2 was expecting good MRMS Reflectivity -10C in that region, but that the co-evolution of Ref -10C with the ABI image predictors was slower than the expected co-evolution based on countless training examples in areas of better radar coverage.

The developers of LightningCast theorize that this could be ameliorated by using the Radar Quality Index (RQI) from MRMS, which quantifies the data quality. Supplying this as a predictor to LightningCast could help in regions with moderate to very poor radar coverage, and provide better uniform guidance throughout the CONUS.

MRMS Radar Quality Index in southwest Texas.

- Hail yeah

Day 3- Using GREMLIN and OCTANE in Warning Operations

 Thunderstorms initiated over the Colorado Plateau, west of the City of Albuquerque, and over the Sangre de Cristo Mountains early in the afternoon on Wednesday, June 4, 2025.  Afternoon mesoanalysis showed instability of 1-2k J/kg and very steep mid-level lapse rates across the CWA. Effective shear of 40-45kts was present east of the Colorado Plateau. Supercell composite was low. The storm motion was roughly southwest-northeast as a shortwave trough moved across the four corners region.

Our warning team did not have any radar data, including the Albuquerque radar as well as MRMS data. This made satellite imagery and lightning data imperative tools in the warning decision process. Thankfully, a mesoscale sector was across the region for the duration of the event. The first storm of concern developed southwest of the City of Albuquerque just before 20Z on June 4, 2025.

I used GREMLIN EMeso-2 and ECONUS as an alternative to radar and MRMS data in the Albuquerque, NM CWA on Day3 of the HWT. Below is a 4-Panel of Gremlin highlighting the storm southwest of ABQ. This storm was alone and believed to have a tight reflectivity core. Personally, this is the greatest reflectivity core I have seen so far in the HWT this week. Reflectivity approached 55-60 DBZ and really caught my eye. GLM flashes were not as impressive as I would have thought, however ELNTN was jumping up during this time (not shown.)

The OCTANE speed-compressed product showed a developing updraft with cooling cloud tops and an expanding anvil. There was a tight gradient on the west side of the storm showing good speed divergence. Also, another developing thunderstorm was starting to catch our eye northeast was also growing with good speed divergence.

The combination of GREMLIN, OCTANE, CH-13 Clean IR, lightning, LightningCast (not shown) gave us confidence to issue the first severe thunderstorm warning of the day.

-Eagle

Anomalous Cloud-Top Divergence in Absence of Visible Convection

 The top panels show IR and VIS imagery, while the bottom panels display Octane CTD, and GREMLIN reflectivities. GREMLIN agrees well with the other fields/products, though I did not verify it against real radar data (as today is a no-radar day).

One feature that stood out was the Octane CTD product. I marked a blue rectangle highlighting high CTD values (between 2 and 4) in an area where no convection appears in the satellite imagery. I do not know if this is caused by the cloud motion around this area, but it appears to be an artifact caused by the algorithm.

- Iceman

Tornado Warning St. Louis Without Radar

 

Top left: GREMLIN simulated reflectivity

Top right: ELTLN lightning 1 min flash, 5 min CG flash, NLDN 15 min CG Flash 1km

Bottom left: GOES East Meso-1 Channel 13

Bottom right: GREMLIN radar with GLM flash extent density image overlaid

While the environment for this area was questionable (shear was <30kts) and we were unsure whether we wanted to even use this sector, it ended up being a fruitful area to observe signals and experience challenges associated with warning without radar in real-time.

Generally, the rhythm we fell into for warnings was using the GREMLIN product to track the storm cores and draw the polygon and cross-referencing with ground-based lightning networks to make sure parallax was accounted for.

Prior to making warning decisions, or when deciding to cancel an active warning, we would monitor the lightning (ground-based and GLM) to make sure it was increasing/intensifying or at least sustaining. LightningCast was consistent in probabilities for the warned area, so it did not provide much of a signal in terms of convective intensity and maintenance of severe storms.

Left: Day Cloud Phase Distinction satellite RGB with GLM and LightningCast version 1

Right: Day Cloud Phase Distinction satellite RGB with GLM and LightningCast version 2

Top left: GREMLIN simulated reflectivity

Top right: ELTLN lightning 1 min flash, 5 min CG flash, NLDN 15 min CG Flash 1km

Bottom left: GOES East Meso-1 Channel 13

Bottom right: GREMLIN radar with GLM flash extent density image overlaid

Around 19:45 UTC, lightning intensity trends (as observed on GLM) combined with cooling intensity and divergence of the cloud tops on OCTANE products led to the decision to issue a Severe Thunderstorm Warning with hazards of 60 mph winds, penny-sized hail, torrential rainfall, and lightning. Around 1951 UTC, a report came in indicating that a tornado was on the ground just north of the St. Louis metropolitan area, within the Severe Thunderstorm Warning polygon we had issued based on GREMLIN data. This prompted the issuance of a tornado warning, in combination with new SPC mesoanalysis data showing low-level hodograph curvature that would support low-level rotation using the “two out of three” framework (radar, reports, environment).

SPC Mesoanalysis 20 UTC June 4 2025: Hodographs

Top image: GREMLIN data with simulated Severe Thunderstorm Warning and Tornado Warning polygons

Bottom image: Graphic for posting to social media and Slack warning the public and partners about this high-profile hazard in a major metro area.

-prob30

Warning Without Radar Data in Northern New Mexico

 Day 3 of the testbed offered a unique opportunity to issue real-time warnings without access to radar data. Having never been in such a situation before, it was an eye-opening experience.

Mesoanalysis showed very steep lapse rates, a long and straight hodograph, high LCLs and a fairly deep mixed layer with relatively low surface dew points. This allowed us to key in on large hail and severe winds being the primary hazards. Additionally, equilibrium levels were only around 9km so hail much larger than 1” was determined to be unlikely.

We issued our first warning upon noticing a rapid uptick in updraft intensity on OCTANE, which showed cooling cloud tops, fast motions, and strong divergence aloft. This storm was located west of Albuquerque in a sparsely populated area so it is difficult to say if the warning verified or not. Gremlin showed a similar uptick in simulated reflectivity which added weight to our decision to warn.

We issued additional warnings as the storm traveled into the Albuquerque area. A second and third cell began to strengthen as well, and two more warnings were issued with the southwest storm looking the most intense on OCTANE. Our mesoanalysis determined that the low-level cumulus field east of the northeastern cell appeared flat and was therefore stable…so weakening was anticipated as it moved off the Raton Mesa.

As expected the northeastern cell began to weaken and dissipate, which was evident on OCTANE and Gremlin. The two cells further southwest continued to look strong, and warnings were maintained into the Albuquerque metro. A hail report of 1” was received at this point on the southern margin of the city.

Lastly, lightningCast showed the northeastern cell begin to weaken before its appearance on satellite degraded significantly. This allowed us to cancel the warning early, in conjunction with noticing the downward trends in OCTANE (weakening cloud top divergence).

During the course of the event we had to keep tabs on a fictitious DSS event, and used LightningCast and its associated dashboard to determine when lightning was approaching their critical threshold (10 miles). LighgtningCast did a good job with lead time as it had 70 percent or higher before any lightning was detected nearby.

- WxAnt

MAF Isolated Convection with LightningCast, Octane, and GREMLIN

 LightningCast

Towards the beginning of operations, there wasn’t much to look at in the MAF CWA, however just to our south, LightningCast was able to pick up on the cell shown in Figure 1 below pretty well before the first lightning strike. It was interesting to see that v2 increased the probabilities to 50% before v1, however v1 increased to 70-90% before v2 a couple frames before GLM depicted the first lightning flash. So in this case, both versions did well in detecting this cell’s lightning potential, with version 1 taking the lead in the higher probabilities right before the lightning occurred.

Figure 1: LightningCast v1 (left) and v2(right)

Another example of version 1 taking the lead is in a different cell just south of Redford, TX shown below in Figure 2. Both versions caught on to the cell at the same time with the 10% probabilities, however as the cell continued growing, version 1 seemed to hold on to the higher probabilities more so than version 2 before GLM showed the first lightning flash.

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

Octane & GREMLIN

I really liked assessing the cloud top cooling in the Octane 4-panel in the image below. You can really parse out that cell just south of Redford, which ended up also upticking in LightningCast probabilities (not shown). This was a great way to keep up the situational awareness and determine which cells needed more focus, especially being without radar to assist.

Figure 3: Octane 4 Panel

GREMLIN also picked up on this cell, which I thought did a pretty good job. It’s hard to assess whether or not it matched up with radar since we weren’t using radar today, however I think with the lack of lightning, and a newer cell, the GREMLIN imagery looked fairly good.

Figure 4: GREMLIN

This cell later went on to grow fairly tall, with GREMLIN actually depicting  a >60 dbZ echo and Octane showing pretty consistent divergence (not shown), so we ended up issuing a warning. I thought GREMLIN did really well, and led to higher confidence in issuing a warning without having actual radar data.

Comparing Octane Color Curves

With little convection in our CWA, I was able to take some time to compare the Octane colorcurves (Stoplight vs. Original). Before today, I tended to gravitate more towards the original colorcurve with Magenta hues as the divergence and the stoplight colors as the cloud top cooling. However the two images below show both color curves at 20:22Z - In this example, the magenta color curve in Figure 6 would lead me to believe the divergence was fairly good in this cell. But the stoplight color curve shows the divergence actually isn’t as good. Comparing this to lightning, GREMLIN, and IR satellite imagery, I like how the stoplight color curve “talked me down” to be more realistic of what was actually going on. So for day 3, the stoplight color curve took the lead.

Figure 5: Octane Stoplight Color Curve (for Divergence)

Figure 6: Octane Magenta Color Curve (for Divergence)

-Fropa

LightningCast Ending Time

 For Day 3, we were Midland TX. There were a few storms that formed off the higher terrain. We did not have radar, so we made use of the GREMLIN synthetic radar imagery, and Octane to make severe warning decisions.

First of all, the Gremlin showed increasing reflectivity, and at one point had 60+ dBZ. This factored into the warning decision.

The image above shows Gremlin reflectivity (top left) with LightningCast (bottom right). ENL was in the top right, but it wasn’t loading. The reflectivity of the GREMLIN factored into the “yes” warning decision.

Secondly, the OCTANE divergence showed a persistent area of cloud top divergence on the upshear side of the updraft. This also pushed our team to issue a severe thunderstorm warning.

The animated GIF above shows a persistent cloud top divergence signal with the storm of interest across southern Texas. There is another storm in the southeast part of the 4-panel imagery.

Unfortunately there were not any reports, but I did go back and look at the ProbSevere, and the MESH maxed out at around 1.13”. When evaluation these storms, they seemed small, so having ProbSevere in addition to the satellite imagery may have led me to hold off on the warning.

The image to the right shows ProbSevere. This was looked at after the fact, but does line up with the marginal severe storm, and warning that was issued for 1” hail.

- Updraft

Wednesday Observations 6/4: The Day with No Radar

 Today in the testbed all three forecaster groups were told to act as if they did not have any radar at their disposal, with offices in Albuquerque, Midland, and St. Louis.

Starting the day in Albuquerque CWA...The forecasters were quick to use OCTANE and GREMLIN to identify the strongest storms, and they picked out a storm just west of the city of Albuquerque. Increasing reflectivity values from GREMLIN, along with divergence signals from OCTANE Speed and the Cloud-Top Divergence products. Animation of GREMLIN and OCTANE (with radar included) are included below. The forecasters also leveraged the 12Z sounding from ABQ with its freezing levels, mid level lapse rates, and the wind profile.

GREMLIN 4 panel with the MESO scene (upper left), CONUS (lower right), MRMS composite reflectivity (upper right), and ABI MESO Clean-IR (lower left).

OCTANE with Speed (upper left), and the cloud top cooling and divergence flavors.

12Z Sounding from ABQ.

As the forecasters debated on issuing a warning, the 'real' NWS Albuquerque office issued a severe thunderstorm warning for 60 mph winds and 1 inch hail.

After some additional discussion, the HWT 'fake' Albuquerque office issued their own severe thunderstorm warning, also for 60 mph winds and 1 inch hail. They used LightningCast probabilities and ENTLN/NLDN flash locations to identify the thunderstorm 'core'.

Shifting to the Midland CWA...The forecasters in this office were watching for initiating convection just across the US-Mexico border in the higher elevations west of the Big-Bend region. There was discussion of the environment and how supportive it was for deep convection, especially related to the moisture/instability return from the Gulf during the forecast period.

OCTANE speed, cloud top divergence, and cloud top cooling products all showed the first robust updraft tapping into this instability with increasing speeds and strong cloud top cooling signatures followed by cloud top divergence.

The GREMLIN product from the MESO scene also reflected this cooling signature and strong gradients with increasing simulated reflectivity values. These values continued to increase until the HWT MAF office issued a severe thunderstorm warning. With 5-10 minutes of the warning, GREMLIN exceeded 60 dBZ. Another storm to the southeast also initiated and quickly intensified, with GREMLIN values also exceeding 60 dBZ.

We talked about the skill of GREMLIN to identify discrete convection and how realistic the values compared to our expectations from this machine-learning, satellite-based perspective. The takeaway from the forecasters was that GREMLIN has an easier time showing greater reflectivity values in discrete convective cores than multi-cell convection. 

I also looked at the ALPW product, and noted that the surface to 850mb layer showed the returning moisture, though I wonder if forecasters would like to see more frequent updates than hourly. I didn't have the chance to ask about this due to the active discussions during operations, but this seems like it would be the 'ideal' case since there was little cloud cover so sampling the lowest layers would be easier for the polar-orbiting sounders. How much this can help NWS forecasters searching for observations in remote areas is to be determined, but the product seemed to fit the conceptual model for the case.

-Dr. Thunder

Assessing GREMLIN skills

 Around 2000Z, a squall line was observed over Oklahoma and Kansas. GREMLIN successfully reproduces the overall reflectivity distribution seen by the KVNX radar, though it slightly underestimates the reflectivity of the convective cores (Fig. 1).

Fig 1: GREMLIN synthetic radar reflectivity (left) and KVNX NEXRAD actual radar reflectivity (right)

By 2016Z, convection intensifies, with convective cores covering larger areas. GREMLIN reproduces most of these features (Fig. 2), though some are heavily underestimated (black marks). The isolated convection—likely poorly resolved by satellite observations—appears to be particularly challenging for GREMLIN to accurately simulate.

Fig 2: GREMLIN synthetic radar reflectivity (left) and KVNX NEXRAD actual radar reflectivity (right). Black marks represent areas where GREMLIN did not reproduce correctly the intensity of the radar reflectivity field.

- Iceman