Lightning under thick ice

 Just documenting another case of LightningCast improving predictions of lightning under thick ice, where ABI can't "see" the convection. Storms were diminishing in southern Wisconsin and northern Illinois during the morning of July 29, but there was still a lightning threat, which LCv2 (ABI+MRMS predictors) correctly highlighted whereas LCv1 had much diminished probabilities. 

Figure 1: Animation of LCv2 probabilities (contours, left), LCv1 probabilities (contours, right), GOES-19 ABI Ash RGB (background), and GOES-19 GLM flash-extent density (foreground blue pixels).

We see that MRMS Reflectivity at -10C, a predictor in LCv2, showed pockets of ≥ 35 dBZ in southern Wisconsin, indicative of a lightning threat. Without the reflective bands in ABI contributing much at this time in the morning, the Reflectivity at -10C was impactful at correctly elevating lightning probabilities from about 09:30 to 10:20 UTC.

Figure 2: MRMS Reflectivity -10C at 10:00 UTC

LightningCast v2 performs better with thick ice obscuration

Here is another example---this time in western Wisconsin---where LightningCast v2 (LCv2) outperforms LCv1.  The models are trained to predict the probability of next-hour lightning, with LCv1 using ABI inputs and LCv2 using ABI and MRMS inputs.

There is thick ice obscuration from an anvil could blow-off entering western Wisconsin, south of La Crosse. Both models had about 10% probability of lightning 20 minutes prior to the first flashes (not shown). Fifteen minutes prior to the first flash, LCv2 shot up to 50% probability of lightning, whereas LCv1 hovered near 10% (top row, Figure 1). LCv1 exceeded 50% 5 minutes prior to the first flash produced by the multi-celled convection (not shown). 

The MRMS Reflectivity -10C in LCv2 (Figure 2), which observed 35-dBZ pixels in the convection, together with ABI signatures (e.g., cold cloud tops, nearby boundary layer cumulus) was able to provide an additional 10 minutes of lead time to the first flashes compared to the ABI-only LCv1 model. 

Figure 1: Top panel: LCv2 (left) and LCv1 (right) contours for 6/16/2025 13:31 UTC. Bottom panel: LCv2 (left) and LCv1 (right) contours for 6/16/2025 13:46 UTC. Background on all panels is the GOES-19 ABI day-cloud-phase-distinction RGB, while the foreground blue pixels are GOES-19 GLM flash-extent density.

Figure 2: MRMS Reflectivity at -10C over the nexus of Minnesota, Iowa, and Wisconsin at 13:32 UTC. 

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. 

Comparing LightningCast v1 and v2 in benign thunderstorms

Lightning is deadly even in "benign" thunderstorms. Some morning convection in Arkansas demonstrates the everyday value of having MRMS Reflectivity at -10C in LightningCast v2 (LCv2), compared to satellite-only LightningCast v1 (LCv1). 

The GOES ABI data is invaluable in the probabilistic LightningCast model, especially prior to convective initiation. But radar data sometimes (not always) provides a signal of convective weakening or decay prior to satellite data.

In this example, thunderstorms ongoing in southern Missouri and northern Arkansas weaken as the sun is coming up. In the animation below, LCv2 is on the left and LCv1 is on the right.

 
Towards the end of the period, when only one cell is present, we see that LCv2 reduces probabilities of lightning about 10-15 minutes prior to LCv1. This cell did not produce any additional lightning.

Figure 2: LCv2 (left) and LCv1 (right) contours, along with GOES-19 day-cloud-phase-distinction RGB and GLM flash-extent density (large blue pixels).

Other features that stand out are "stronger" false alarms in LCv1, compared to LCv2. The images below show these false alarm regions to the south (Figure 3) and south and east of the thunderstorms (Figure 4).

Figure 3: LCv2 (left) and LCv1 (right) contours, along with GOES-19 day-cloud-phase-distinction RGB and GLM flash-extent density (large blue pixels). Note the stronger false alarm in LCv1 south of the ongoing thunderstorms.

Figure 4: LCv2 (left) and LCv1 (right) contours, along with GOES-19 day-cloud-phase-distinction RGB and GLM flash-extent density (large blue pixels). Note the stronger false alarms in LCv1 south and east of the ongoing thunderstorms.

Overall, the radar predictor in LCv2 helps to reduce false alarms that occurred in LCv1.

 

Lightning cessation

Lightning cessation typically occurs when a storm decays to the point of no longer producing lightning. It is indicative of a weakened updraft. This is an important event for forecasters to predict well, as event managers and other users need good guidance on when they can expect a lightning threat to be over.

LightningCast predicts the probability of lightning in the next hour, including before first-flash events in storms, and for lightning cessation. LightningCast v1 uses solely ABI inputs. This sometimes makes it difficult to discern lightning cessation, particularly when the reflective bands are not present (or strongly illuminated) and when there is thick ice present with a "smooth" look at cloud-top, which often is the case when storms decay, leaving an anvil cloud.

The animation below demonstrates LightningCast v2 (on the left) vs. LightningCast v1 (on the right). LightningCast v2 incorporates the MRMS Reflectivity -10C as a predictor at 1-km resolution. This helps the LCv2 to "see" the diminishing lightning signal sooner than in LCv1, due to a collapsing updraft and reflectivity core, which often precedes warming cloud tops in satellite imagery.

LCv2 (left) reduces lightning probabilities for the storm on the Oklahoma/Kansas border much more rapidly than LCv1 (right).

Toggling of LCv1 vs. LCv2 at 11:46 UTC helps us see the false-alarm-area reduction in south-central Kansas for LCv2. Predicting lightning cessation is one important scenario where we feel that LCv2 improves upon LCv1.

 

LightningCast v1 vs v2

ProbSevere LightningCast v2 adds a new predictor---MRMS Reflectivity -10C. This predictor has been well documented to correspond to electrification in storms, as it observes water particles in the mixed-phase region, where charge separation begins. We use it in LightningCast v2 at approximately 1-km resolution, along with the GOES-R ABI 0.64 µm reflectance (C02, 0.5 km), 1.6 µm reflectance (C05, 1 km), 10.3 µm brightness temperature (C13, 2 km), and 12.3 µm brightness temperature (C15, 2 km).

In the example below, LCv2 is on the bottom left, LCv1 is on the bottom right, and the MRMS Ref. -10C is on top. You can see that there is lightning (as detected from GLM) in a mid-level cloud deck, but obscured by some high-level ice. LCv1 drops probabilities < 10% at numerous times in the animation, due to the uncertain signal from the relatively smooth but not overly cold cloud tops. On the other hand, LCv2 picks up on the re-developing convective cores, as shown in the Ref. -10C pockets of ≥ 35 dBZ (yellow-orange), and correctly maintains a higher probability of lightning throughout the animation.

This is one way in which LCv2 is an improvement over LCv1---in convection that is obscured by moderate-to-thick ice. We've also seen improvement for lightning cessation situations and more "marginal" situations such as for lightning in low-topped convection.