GLM Data Quality Under the Shadow of the Anvil

 

As we are watching a cluster of thunderstorms develop across Nebraska, we're in a region where GLM may not be able to most efficiently detect flashes in the region. However, underneath the shadow of the cirrus blow off, the flash detection efficiency increases, and the stoplight colormap begins to suggest that the data quality is better. Perhaps this offers a bit of hope when forecasting charge moving along with the anvil.

Kadic

A Tale of Two Storms From OCTANE, LightningCast, and GREMLIN

 

This is an interesting comparison of two storms that show two initially different satellite and LightningCast signals that produce very different results on radar. The southern cell shows an initially much more consistent cloud top divergence signal from OCTANE with a more robust looking anvil shield and an above anvil cirrus plume. That aside, the northern storm consistently had a higher probability of >10 GLM flashes from LightningCast and eventually developed a far stronger radar signature and eventual severe thunderstorm. The southern storm struggled to even develop a 40 dBZ core. The animation below shows the same progression but with the OCTANE speed and direction RGBs. In this case if a severe decision was to be made with just the satellite presentation, the wrong decision may have been made (at least initially). 

-Joaq

Finally, here is how GREMLIN handled the southern storm, which it understandably initially intruduced high reflectivity to the southern storm. 

-Joaq

OCTANE Trends for Ongoing Convection in ABR CWA

Taking a look at OCTANE - it seems like the IR/Nighttime data shows especially the directional details a bit more than the Visible/Daytime products. This has been a continuing trend this week. The CTD and CTC products have been useful with monitoring the strength of the storms (especially which may become severe) as well as with the newer convection/updrafts.

OCTANE - Visible (Daytime) vs IR (Nighttime)

OCTANE - Cloud Top Cooling and Cloud Top Divergence

Forecaster Cumulus

A Tale of Two Thunderstorms

Two thunderstorms developed with mixed signals between the variety of tools available. Satellite tools would've suggested that the southern storm was the area of interest, but the radar signature was much better for the northern storm. GREMLIN is shown below, with the GOES West on the left and GOES East on the right. LightningCast is contoured, and the probability of 10 flashes mainly favored the northern storm. The appearance on GREMLIN was much stronger for the southern storm.

When looking at the GLM RBG, the more frequent and shorter flashes were associated with the southern storm.

Looking at OCTANE, the southern storm appeared more impressive. Although towards the end, the northern storm began to exhibit stronger upper level divergence.

But again, if one were to look at radar, it would be readily apparent that the northern storm should be ranked as the biggest threat. In a situation involving satellite alone, I might have missed the event that did produce the severe event.

And of course remembering how significant parallax is. From GREMLIN with GOES West, my storms were neatly in their boxes, but from GOES East, it would've looked quite strange.

And despite the signal from satellite, the southern storm essentially collapsed in on itself. GREMLIN using GOES West does not seem to catch on to this fact, but GOES East has corrected to a stronger storm up north.
 

Kadic

Finding Boundaries in OCTANE Direction and Comparing to Satellite

As a forecaster, I'm most accustomed to pulse severe conditions, and I work in an area where there tends to be abundant cloud cover. So being able to find boundaries is really helpful. So I decided to review some of the low-level cumulus fields in OCTANE and looking how accurate they are compared to current derived GOES East winds. From what I sampled, they were within 5 degrees of OCTANE.

Now zooming out, we see that there will be an area of confluence in the wind field based on the low-level cu field. Further aloft, you can see some of the diffluence along the warm conveyor belt. Sometimes, wind barbs are really useful tools to assessing an environment, and I think a similar function in OCTANE could be useful for mesoanalysis.

 

Kadic

How is GREMLIN Doing? Storms Form Sooner Than Expected.

Took a look at GREMLIN to see how it was handling the convection, as storms have developed a bit earlier than expected. It’s doing a great job with the overall picture of the line of discrete storms (formation and placement), but perhaps not as much with the intensity of those individual storms.

Reflectivity from KUDX on the left, GREMLIN in the center, and MRMS on the right

Forecaster Cumulus

PHS and HRRR for WFO ABR

As of 2030Z, storms have already initiated near Pierre SD in the KABR forecast area.

Both PHS and the HRRR appear to be too slow on forecasting convective initiation in this area.

The first image is PHS comp reflectivity (overlaid on PHS SBCAPE) valid at 22Z.

The image below is HRRR comp reflectivity valid at 22Z. Both PHS and HRRR are depicting convective initiation between 21Z-22Z, but as mentioned, storms have already initiated as of 2030Z. PHS and HRRR both show initiation in slightly different locations, but in real life, initiation occurred near both indicated locations.

Both PHS and HRRR also depict that stronger cells will develop near the SD/ND border in the 23Z-00Z time frame. The next two images show this in the parameter space. The first image shows Sig Tor Parameter on PHS (valid 23Z) and the second image is the same thing, but with PHS comp reflectivity overlaid on top.

        

Of most note is the simulated storm northwest of Ashley ND, which not only has the highest simulated comp reflectivity value, but also significantly impacts the storm-scale environment (as depicted in the model by an unrealistic significant localized bullseye of STP). As a forecaster, although I recognize these STP values (~10) are likely spurious, I can use their presence as a proxy to the fact that the storm simulated by PHS in this location is likely very strong and modifying its environment in a way that only significant supercells are able to do.

-- Insolation

Reviewing Conditions Between the RAP and PHS

The PHS WRF will offer a unique means to compare RAP analysis and instability parameters. When observing what conditions were like in the PHS, it was noticed that the inclusion of satellite data resulted in drier conditions being observed at the lower levels across western Nebraska.

Satellite viewing of drier weather also corresponded with warmer temperatures across western Nebraska at the surface.

Now how does this look in comparison to various model fields in the RAP. On the bottom left are near surface dewpoints in the PHS, and on the bottom right is the RAP. Carrying forward into the simulation, the values for dewpoint were about 1-2 C below the RAP analyzed forecast. This corresponded better with surface observations within the area. Where the RAP had lower 60s lifting north, these have not been observed near surface stations at the same time. (Note color tables are different. Attempt to get this again was unsuccessful due to CAVE crash)

Here is the same post, but with the CAPE contours now shaded instead with the contours of dewpoint in Kelvin. The lack of a gradient in dewpoint conditions was more representative of the ground truth observed at different ASOS stations.

As far as the temperatures went and how it affected instability, you can also observe the RAP analyzing a cooler pocket, and it likely seems to be the result of the coarse model struggling to capture the terrain across western Nebraska.

What's interesting is that the instability parameters are not completely different. The reason could perhaps lie in the observed data suggesting a warmer layer aloft. However, the RAP analysis did not differ significantly from the PHS data at 00z.

If you also look, the PHS also has a broader area of instability to the east compared to the RAP. This is where the PHS indicated higher moisture content, and it may be able to destabilize the region more efficiently as a result.

Finally, despite the drier air in comparison to the RAP, it indicated convective initiation about an hour earlier. Using the the forecast helicity and updrafts at 500hPa, one can see that initially pulse convection with minimal rotation should encounter an environment of increasing helicity. We shall see what comes out of this.

For forecast areas that may not have the benefit of features like WoF or other tools to analyze environmental conditions, looking at how differences in the satellite incorporated data and RAP mesoanalysis can help forecasters weigh environmental conditions and what they see from the RAP.

Kadic

LightningCast and DSS

Monitoring convective development over the southwestern portion of the ABR CWA, where we are providing DSS support to a Fishing Derby (yellow range ring - 10 miles). From this data, I would be able to let an EM know that we're noticing an uptick in lightning probabilities due to storm cells developing to the south/southeast of the Derby and moving northeast. These are not severe at this time and no lightning has been observed as of yet. However, currently, probabilities of lightning within 10 miles are between 40-50%. If I were to use LC while at an on-site deployment, I would have both the map view (first image/loop below) and the graph (second image below) up to show what I am looking at, but especially the map view to give context to the graph.

LightningCast in "map view" showing probabilities of one or more flashes (as an image, not contoured)

LightningCast Dashboard showing lightning probabilities at DSS point (Fishing Derby, Aberdeen, DS)

UPDATE # 1 - Lightning has been observed! From when the 10% contour (10% chance of 10 or more flashes) first popped up (red contour north of Lyman) at 2010Z, it was 10 minutes until GLM and the ground networks observed flashes. It was 5 minutes later when ground networks observed CGs.

LightningCast - time of first 10% contour (red) 2010Z

LightningCast - GLM and ground networks observe flashes 2020Z

LightningCast - Ground networks observe CGs 2025Z

In terms of lead time, I crafted a DSS message between 310-315pm (seen in the first paragraph above this update) and the first flashes were observed at 320p and CGs at 325p. Therefore, this gave a 10-15 minute lead time.

UPDATE #2 - Below is a snapshot of the LightningCast Dashboard showing the above mentioned GLM flashes within 10 miles.

Forecaster Cumulus

Examining LightningCast Values Near Pierre SD

We have a Fishing Derby ongoing near Pierre, South Dakota. We observed some differences in the LightningCast probabilities within a 10 mile radius between the AWIPS contour plot and the dashboard.

Here's an image from AWIPS at 2004Z (1504 local time), with the DSS point marked by a 10-mile radius near Pierre SD (white circle), and the LightningCast probabilities (60-min for 1 flash) contoured in blue and green.

Next, here is a look at the dashboard for the same DSS event. Of interest is the pink line, which shows the maximum probability for a flash within 10 miles of the DSS point within the last 5 minutes.

At the 1506 (local time) time slot on the dashboard, it shows the maximum probability of 47 percent. However, the AWIPS image above shows the greatest contour values within the DSS point radius for that same time minute span (at 1504 local time) and the value only barely exceeds 30 percent. It appears that the value on the dashboard is reading about 15% higher than the value that would be implied by the contour display in AWIPS.

Functionally, as long as the trends are consistent, this may not make much of an impact on messaging or forecasting for this DSS point. However, having the values more closely matching between the two sources is something that would likely increase forecaster confidence.

--Insolation

Update to PHS versus HRRR Convective Initiation

 

Looking at an update of the 19Z HRRR run and the 17Z PHS run, the more recent HRRR run is catching up to a sooner initiation solution. The 19Z HRRR is initiating cells as early as 21Z with a series of cells displayed above by 22Z, similar to the 17Z PHS. Another note is that the newer HRRR run has a higher SBCAPE value than the PHS at 1700-2700 J/kg compared to 1000-2000 J/kg from the PHS. Recent visible satellite trends look to favor an earlier initiation as well.

-Joaq

Waiting for CI - Looking at PHS

Anticipating most of the activity, should it develop, to begin in the next couple of hours in the western/southwestern portion of ABR's CWA. However, I wanted to take a quick look at PHS to see some of the environmental parameters and how it relates to what's going on now (and maybe later). 

PHS Composite Reflectivity overlayed onto Visible Satellite (left) and MRMS Reflectivity (right)

For the above loop, PHS has perhaps the general idea of CI from the convection from earlier this afternoon, but struggled a bit with placement of some the stronger cells in the east. However, I think it may have a good handle of the lull we're expecting until later this afternoon. It will be interesting to see whether we get storms in the line of pearls like it's suggesting. 

PHS MUCAPE at 20Z

SPC (RAP) MUCAPE at 19Z

Comparing the MUCAPE from SPC's Mesoanalysis page (RAP) and PHS, it seems like both agree on ~2000 J/kg nosing into the southwestern portion of ABR's CWA. This is the area that we are monitoring for initiation over the next couple of hours.

Forecaster Cumulus

OCTANE Cloud Top Divergence and Cooling view of an Orphan Anvil

 Here is a good example of the OCTANE product showing the transition from an updraft to a divergent anvil cloud, which then cuts off from the parent updraft. This "orphan" anvil is an indication of a convective initiation attempt, which can be a signal for upcoming successful initiation in the near future. The OCTANE product does well to highlight both the updraft and then the anvil in different color shades through the entire orphan anviling process. 

Below is an additional animation of deepening cumulus development with the OCTANE cloud top cooling product highlighting the taller towers.

-Joaq

Convective Initiation Timing Between the PHS and HRRR

Here we compare convective initation in the PHS model (right) versus the HRRR (left). Both models have relatively similar SBCAPE (around 1000-20000 J/kg) but the HRRR does carry CIN longer through the afternoon. The 16Z (and 17Z, not shown) PHS model produces convection by 21Z to 22Z, while the 17Z HRRR waits until 23Z. 

Below is the 1854Z GOES-East day cloud phase RGB which shows some shallow cumulus growth in western to north central areas of the CWA. This may lead to more confidence in the earlier convective development from the PHS model.

-Joaq

Tracking Storm Strength With GLM and OCTANE

The GLM RGB, combining flash extent density and minimum flash area, highlighted the intensity trend of a cell in the northeast corner of the FWD CWA. The yellows of the RGB also corresponded with the uptick of cloud top cooling signatures shown from the OCTANE product. Using these products together I was able to track the intensity of thunderstorm, which took another uptick towards the eastern border of the CWA. The GLM RGB is definitely a useful tool in reading both the characteristics of the flash length and the flash density. 

The OCTANE cloud top divergence product here is overlaid atop the visible satellite imagery with the cloud top cooling product, which may look a little messy to look at at first glance. After some practice with the product I was able to learn to pick out both the cooling and the divergence in a strong convective cell. In the third image I did remove the divergence product to have a good look at the cloud top cooling and visible satelitte signatures. After going back to the combination of the two however I found it easy to read what was happening among both the divergence and cloud top cooling with both displayed. I did like having cloud top cooling displayed on top of the divergence product as the divergence product was broader spacially and it made more sense to have the smaller scale cooling signals pop up above the divergence display. 

-Joaq

Analyzing Clusters, Anvils Carrying Charge North, Final Checkup on LightningCast

Working on the DSS for the PGA tournament, one of the more frustrating features was how thunderstorms well outside the range of the event produced lightning. About the time one DSS image was sent, there was a lightning flash that occurred well north of the primary cluster and near a very weak area of reflectivity.

However, this has been the story for much of the day, where intense convection has been to the south and weaker cells to the north are still managing to produce lightning. Below is the probability of exceeding 10 flashes on GLM, with GLM and MRMS at -10 C to highlight how weak the convection was to the north. The fact that reflectivity was barely at 25-35 dBZ would suggest little potential for lightning.

Analyzing the RGB channels for lightning, one can see this evolution well. With more intense updrafts producing several flashes on GLM, they appear yellow. To the north, where it appears the anvil is carrying charge north, the flashes are very long. From a DSS perspective, this can be frustrating when communicating the potential for high impact weather when all that one gets are sprinkles and rumbles of thunder. Still, the RGB channel can be very helpful in delineating these features, but would also be a helpful means to suggest that the northern convection may not develop quite as much. The 50dBZ echo tops are intended to help highlight the stronger storms. Note how a few pixels of 50dBZ echo tops at best appear in the blue, while the larger cluster of taller storms have the younger convection. This also helped me consider parallax as well. Overall, I really like the potential for lightning characteristics divided into this RGB would be helpful in pulse convection.

And then later, the LightningCast began to behave a bit more oddly. Perhaps these situations cause it to become bouncy. Although, you can almost see these dips in the flashes on the chart as well. At this stage, I feel like I could tell the poor folks playing above par at the PGA tournament and taking forever that they can pack up their clubs and head home, because at this stage, the lightning is here to stay.

Outside of the one flash of lightning that took place over the event about when values crept upwards towards 50 percent, there was a flash. However, values had been hovering around 30-40 percent for much of the day. Values crept even higher, and yet there were no flashes nearby. It seems whatever convective debris left the region, and then the forecast became better overall.

Kadic

OCTANE... the GOAT for Warning Ops Today!

On the warning desk today, I've noticed that I have been using OCTANE the most out of the newer tools (besides an adjacent radar). It has really helped with confidence with warning decisions - are the storms maintaining strength... are newer ones growing? The storm in the eastern portion of this loop (below) has been going and going for a handful of hours (originally SVR in our area). OCTANE has kept right along with it and showing its evolution aloft (in both IR and Visible).

 

 

Also learning to use the CTP and CTD portion of OCTANE - it's been helpful to keep track of the newer updrafts (shows up really well to the west with the outflow, but also in the green where there is more anvil/cirrus overflow). Also, it's been a great tool to see where storms are maintaining strength (where the divergence signature in the pinks/purples hold on). 

These products have aided in confidence in warning ops, especially with lack of primary radar.

Forecaster Cumulus