Week 3 Wrap Up

Wrap up feedback from our week 3 discussion is posted below. It was another great week of forecaster interaction with a lot of valuable discussion time about the various products. Next week will be our final week for this year's evaluation in the HWT.

GLM

Lightning definitely behaved differently in some of the different environments, particularly in west
Texas where the storms seemed to be less electrified

Good at identifying the intense storms in South Dakota yesterday, seemed to ramp up when storms
were just starting to go severe

The 2 minutes update product seemed to be a pretty good compromise for viewing the data

The parallax shift was more noticeable up in South Dakota than in other areas from throughout the
week.

Had a hard time picking out any features in the core using the GLM products while it was good for
finding the large flashes out in the anvil

The smaller flashes were more difficult to pick out and interpret

ProbSevere

Continue to try to find a way to discriminate the storms better, even at the expense of longer lag time

Did seem to be a lot better calibrated and didn’t overdue probabilities as much as I’m used to

See the use in having the option to have the individual probability contours

ProbTor especially needs to be kept separate to pick out the relative increases

NUCAPS

Better than the sounding that we never get

The modified seemed to be very situation dependent on how much modification took place

Did underdo the lapse rates yesterday quite a bit even in the mid to upper levels

There were definitely some discrepancies between the direct broadcast and the operational

Focus training on the “gotchas”

The awareness is the main point to focus on to get others to use it

The availability of the soundings can help fill a lot of holes

Having some best practice starting procedures would be useful particularly for sending out the
gridded data

All-Sky LAP and CI

CI was still just not very useful and frustrating to put much confidence in

Did reasonably well in the times when it wasn’t really needed, times that it was expected

Didn’t do well at times when it would be most needed

Interested to see some improvement with the GOES-16 training data

There was nothing yesterday in South Dakota leading up to initiation time

LAP did pretty well yesterday, and the GFS fill in areas seemed to do a lot better yesterday than the
day before

Echo the fact that getting a higher resolution or faster updating model would still be ideal

There was a more smooth transition between the clear and cloudy regions than previous days

Values also seemed to be better than the baseline GOES data

Good to be aware of which data type is filling in the retrieval at that point

The PW products seemed to make terrain features stand out in certain cases

The Layered PWAT color tables should probably be adjusted to pick out the smaller scale gradients
easier

General

It was good to be immersed in the science and learn the new tools in a familiar setting

Turn the data on and use it

Getting into the nuts and bolts has been invigorating

Very helpful having experts here for explanation and question answering

-Michael

Severe Thunderstorms North and South of Rapid City

Clusters of severe thunderstorms developed north and south of Rapid City, SD. Eventually, a line of thunderstorms formed once the clusters began to move east-northeastward. The line filled in between the clusters and became severe at times, although the storms wee persistently severe at the northern and southern ends of the line.

The ProbSevere data clearly highlighted the hail threat, with the wind threat also increasing as the line formed and a well defined outflow moved east of the storms. I did notice an increase in the ProbTor percentages for a while with the northern cluster; subsequent analysis showed a brief period of significant rotation at about 6000 feet AGL with lesser rotation on a few scan levels above the 0.5 degree scan. The rotation weakened in about 15 minutes, and the ProbTor levels fell quickly. No tornado was reported nor was a warning issued.

GLM lightning data showed definite flash clusters near the strongest updrafts and overshooting cloud tops, although the parallax in western South Dakota is pretty significant from GOES-16, so I had to use the existing lightning network data and NEXRAD data to get a better idea of storm placement. It will be interesting to see how the eventual GOES-17 data would improve things in that part of the country.

Gary Cannalte

ProbSevere way too large

We talked briefly about using the MRMS isothermal (-10C, I think) reflectivity to differentiate cells for the ProbSevere algorithm. This would certain improve the smearing of echoes that occurs when storms are nearby. In this particular example, At least 4-5 separate updrafts were observed but the ProbSevere included all as one cell, basically covering much of 2 counties, plus a few more. Certainly the algorithm is useful for the strongest storm, but it doesn't provide much spatial value like this. I feel a longer processing time is useful if it can prevent this from occurring.

Fig 1. Multi-county ProbSevere on a non-linear cluster of storms.

wall_cloud

Local "Higher" Flash Extent Density on Low Lightning Day

The majority of the storms across the AMA forecast area today did not have very impressive lightning signatures/numbers on GLM, despite having storms with hail reported as large as half dollars. Most of the storms were slow developing intense cores. However, starting around 2232Z, one storm rapidly intensified and developed a forward speed more than double the average of storms so far today. During this ramp-up in intensity, the Flash Extent Density also spiked up. While not an especially high value (max 14 flashes on 2 minute sampling/1 minute update product), it was the most impressive seen today.

0.5 Degree AMA Reflectivity 2222Z-2253Z

GLM Flash Extent Density (2 minute sample/1 minute update) 2232Z-2254Z

Alex Brown

Flash Extent Density and Flash Centroids

Here's another comparison of GLM flash extent density with ENI flashes and pulses. GLM parallax displacement is again apparent with the line of storms extending southwestward across western SD into eastern WY. Allowing for that, there is good general agreement between the two data sources regarding lightning location and concentration. The GLM FED clearly gives a more complete picture of the area extent of lightning.

The next image shows GLM flash centroid density (blue boxes) instead of FED. Most of the values are 1 indicating a single flash centroid. (This product is probably not destined for operations.) Interestingly, these GLM flash centroids match up fairly well with the ENI flashes and also with the MRMS RALA reflectivity cores. A.Cope

Supercell SW of Rapid City

Thunderstorm development began shortly after we began. Cells north of Rapid City became severe, with hail as the main threat.

Shortly after 2000Z, a supercell thunderstorm developed rapidly along the Wyoming/South Dakota state line. Visible satellite imagery showed a nice flanking line of towering cumulus feeding into the southern part of the supercell, and the storm was showing signs of rotation in the lower levels, although the radar beam was at about 6500 feet AGL, and about 10,000 ASL.

In monitoring the 2-minute (1-minute update) GLM event density lightning, there was a noticable increase in flash events in the 10 minutes or so leading to where low-level rotation was observed, with a quick drop off in lightning from that point on for the next 20 minutes as the storm probably began producing large hail.

In checking the ProbSevere products, the hail showed a large increase into the 85% range (the MESH hail size was around 1.25") and a smaller increase into the 30% range of wind. Interestingly, even with the low leve rotation and some evidence of a hook echo, the ProbTor probabilities stayed less than 2%.

Gary Cannalte

AllSky SFC-900mb PW Nicely Captures Dryline

The AllSky SFC-900mb PW nicely captured the eastward advance of the dryline into the AMA forecast area this afternoon.

Alex Brown

CI Probs Along WY-SD Border

Another example of the GOES-16 general convective initiation probabilities. Loop below shows CI probs overlaid on VIS imagery. From 1700Z to 1900Z, storms developed N-S along the WY-SD border. Prior to deep convection developing, the CI probs were no higher than about 30 percent. Seems like they should be higher than that. A. Cope

NUCAPS Lapse Rates Not Steep Enough

The afternoon NUCAPS 700-500mb lapse rates were about 1-2C/km too low across the AMA forecast area this afternoon when compared to models. This was also confirmed by the 18Z AMA RAOB.

19Z NUCAPS 4 Panel 700-500mb Lapse Rate Procedure

18Z AMA RAOB

Alex Brown

Parallax error on CWA border

We are all familiar with the parallax issues that arise with the satellite data. Sometimes that error can be in a location that may confuse the forecaster if this isn't in the front of his/her mind. I posted this mainly as a suggestion that we really hit the parallax issue hard in the training for operational staff, especially since lightning location is a significant aspect of IDSS.

Fig 1. Storm located over Parmer County, TX (in LUB CWA). GLM indicates lightning extent into southern Deaf Smith County (AMA CWA) due to parallax issue.

wall_cloud

Terrain showing in Layered PWAT

The image below shows the AllSkyLAP total PWAT and lower/upper layer PWAT, plus topo image in the lower left. Interesting that higher terrain, such as the Black Hills and mountains of CO/WY, can be seen in the PWAT images, even the upper-level 700-300mb image. A.Cope

Lightning with first severe? storm

The first strong/severe storm developed around 1900z along the WY/SD border, at least according to ProbSevere.  Four-panel loop below shows GLM lightning products,  plus radar and ProbSvr in the lower right panel, from 1850 to 1910Z. In this case the GLM lightning alone doesn't really stand out for this storm, as far as I can tell. WFO UNR did issue a SVR at 1917z.

Also in the upper left panel the NLDN CG lightning is plotted along with GLM FED. The slight NW displacement of GLM is evident.  A.Cope

Afternoon Convective Outlook for AMA

Quiet conditions at the moment across the AMA forecast area.  A cluster of storms that was over the northeast portion of the region has recently pushed off to the east. Minimal cloud cover remains, although there is a small CU field left in the vicinity of outflow from the previous storms. Instability is steadily building across the region with AllSkyLAP CAPE values from 2000-2500 J/kg in many areas.  An ongoing 18Z balloon launch from AMA still shows a weak cap around 770 mb, but this should be overcome later this afternoon as diurnal instability increases. Guidance suggests instability will peak later today near 3500 J/kg. Lapse rates are also very steep, with 700-500mb lapse rates near 9 C/km. An upper-level impulse tracking through the area today will increase bulk shear values to around 40 kts. The surface dryline, which currently lies just west of the forecast area, should advance east into the area later this afternoon.

Expect severe storms to develop along/ahead of the advancing dryline within the next 1-2 hours. The moderately strong instability, steep lapse rates, and moderate shear support large hail with the initial development. The storms should eventually grow upscale into a line and more of a damaging wind threat.

19Z Surface Plot/1929Z Visible image

1828Z AllSkyLAP CAPE

18Z AMA RAOB - CAPE values comparable to AllSkyLAP

Alex Brown

AllSkyLAP over Northern High Plains

Working today as the Rapid City SD office. This area is under a slight risk, with convection expected to develop through a north/south corridor over the high plains.

Looking at AllSkyLAP stability indices to help evaluate the environment. The 4-pan image below shows  the north-south axis of instability over the High Plains, with CAPE (upper left) in the 1500-2000 J/kg range over SW SD. Upper right is lifted index and lower right is K index. The lower left panel shows whether the values are influenced by GOES data (blue/yellow) or if they are just GFS data (gray). Differences due to data type are noticeable in the CAPE image over western NE, but not so evident in the LI or K images. A.Cope

Forecast 5/17/18

A strong shortwave trough will continue to move across the Great Basin this afternoon, providing increasing southwesterly flow aloft downstream over the High Plains. The water vapor loop implies increasing forcing for ascent across the Southern Rockies with increasing mid/high clouds approaching the TX/OK Panhandles. Strong diabatic heating will allow scattered strong to severe thunderstorms to develop this afternoon and evening.

Fig 1. Increasing forcing for ascent noted over the Southern Rockies.

Rapid surface destabilization is taking place this afternoon with 18z KAMA RAOB indicating 700-500mb lapse rates of 9.1 C/km and surface temperatures near 90 degrees. The AllSkyLAP product suggests CAPE near 2500 J/kg with model progs increasing to near 3500 J/kg by peak heating. Deep-layer shear is a bit strong than previously anticipate with 0-6km BWD generally between 35-45 kts. This will be sufficient to sustain organized updrafts with some storms becoming supercellular initially before growing upscale into linear segments.

Fig 2. AllSkyLAP CAPE of 2000-2500 J/kg as of 1830 UTC.

Storms are expected to initiate along the dryline on the Caprock, moving in a general eastward direction. Lift may be enhanced across portions of the Panhandles by an outflow boundary originating from the ongoing convection over northwest OK. The Bunkers RM storm motion from the W and NW only 5-7 kts, suggesting a potential for heavy rainfall with any right movers. The very steep lapse rates and strong instability will support a large hail threat, with isolated sig hail reports possible this afternoon. There is also a potential for isolated damaging wind early in the storms' lifecycles, becoming more widespread as storms grow upscale.

wall_cloud

Week 3 Day 4 Operations

Today we will be operating in the Rapid City, SD and Amarillo, TX CWAs monitoring the area for storm development later in the afternoon.

-Michael

Day 3 Wrap Up

Wrap-Up and feedback from the day yesterday is listed below. Yesterday was a pretty slow day weather wise which gave the forecasters a good opportunity to explore more of the products.

GLM

A summation somewhere between 1 and 5 minutes would be perfect, not sure of the 2 minutes is the
perfect answer

Struggled to find a need to have the FED and the event density. FED and flash area seem to provide
the most use when using it within warning operations

Flashes and subsequently FED seems to be the most intuitive product

Hands on training would be the most imperative to get people up to speed on using GLM data in
operations and what and when the products are useful

Need to stick with simplicity, especially when presenting the information to the public or to EMs

It is important to use in conjunction with other lightning data to see CG strokes

GLM seems to provide a much larger coverage in stratiform areas or areas behind a line

ProbSevere

Environment yesterday was not conducive for ProbSevere to be real useful

Had some pulse storms and microbursts that weren’t detected

Still took a little while to catch on to the hail in the Texas storm as the background environment again
wasn’t real good

The object tracking was a little junky yesterday. It would combine cells and then split apart and then
combine again and so on.

Might be worthwhile to try to use reflectivity at -10C, even if it was a minute or so later

NUCAPS

Liked the lapse rate procedures to compare to models and see how well things were handled by the
models

Soundings would definitely be useful outside of the balloon times

Yesterday the modified had a substantially better boundary layer representation compared to the
unmodified version

Useful to have a background satellite image to understand which soundings are useful

The DB soundings seem to have a lot more yellow points than the operational version

CI

Continue to struggle with the fact that the severe CI is higher in a lot of cases that the non severe

The jumps in probabilities are inconsistent between the two products and don’t instill much confidence
in using the product

To use the product it needs to really draw my attention with a consistent higher value in a field of lower
values, especially in the summertime southeast thunderstorm environments

The data is very noisy and jumpy especially in the more discrete cloud areas

All-Sky LAP

Had some problems in the Arkansas area where there was quite a bit of cloud cover and the GFS data
seemed to be convectively contaminated and undercut the CAPE values

Did show a nose of more unstable air coming up into southern AR where storms did intensify slightly

Would be worth trying to have a CONUS “nest” using a little more high resolution model data to fill in the
cloudy regions

Like having the All-sky compared to the baseline product to fill in the cloudy areas, but some
improvements would be desired

Worked pretty well in Texas where there was very little cloud cover

GOES-16 Products

Loaded the Day Convection RGB, but need the higher resolution to be useful

Love using the Day Cloud Phase Distinction in initiation stages

Looking at the three WV channels on their own is still the most useful

Generally, the RGBs are difficult to understand and interpret without a lot of experience, even with
quick guides and training

The 1 minute updates are amazing

Lightning Shift

A severe thunderstorm developed over western Sterling County just within the SJT CWA. Hail was the main threat, with 1.5" hail expected. At the same time, and additional thunderstorm was developing over eastern Glasscock County immediately to the west of Sterling County.

A few minutes later, the most intense lightning began shifting to the west into Glasscock County. The storm began intensifying rapidly at that point.

Within a few more minutes, the Glasscock County storm became the dominant storm. hail up to 2" in diameter was expected. The Sterling County storm weakened a little as the Glasscock County storm took over.


Gary Cannalte

GLM Total Power

We had a brief discussion on the utility of Total Energy and what a "large" value may be. I noted this storm in southern Arkansas was indicating a 2 minute Flash Extent Density near 45  which is near the high end of the default color table. The Total Energy registered 240 fJ at this time, which is double the top end of that respective color table.

It makes me wonder what all factors contribute to high energy for a particular storm over another storm in a similar environment. What processes are at work in this instance?  It's all very fascinating and could have potential benefits in the warning environment should a correlation become evident.

Fig 1. Note the Total Energy in the bottom left panel is 2 times the color table maximum.

wall_cloud

NUCAPS Soundings and Imagery

The first image below shows the afternoon NUCAPS soundings over AR and vicinity, displayed over the GOES CH2 Vis and NUCAPS 850-500mb lapse rate imagery. The image has data only for the "green" soundings, i.e., those for which both the IR and MW retrievals were successful. This results in a lot of "holes" in the NUCAPS images. Taking away the NUCAPS image (second screen capture below), it's not exactly clear why some of the soundings are green and some adjacent ones are yellow, e.g., over southern AR, as the cloud fields below look very similar. Perhaps the QC process could be modified so that more sounding are included in the creation of the NUCAPS images, as these are much quicker to examine than looking at each individual sounding. BTW, the NUCAPS lapse rates over southern AR are only about 6.5 deg C per km, not particularly impressive. The t-storms in AR today have been mostly sub-severe.  A.Cope

Odd ProbSevere Cell ID

A cluster of several distinct storm cores were identified by ProbSevere moving across west Texas this afternoon. However, for a brief period of time, the cores were combined into one storm ID, event though the cores still looked distinct.

2254Z-2312Z ProbSevere/MRMS RALA Loop

Alex Brown

CI/CI-Severe in West Texas Convection

I was able to watch the CI and CI-Severe products in some convection that developed over Glasscock county just west of the San Angelo CWA.

This image was taken at 2147Z on Wednesday, with CI in the upper left, CI-Severe upper right, Goes-16 Visible lower right, and SJT 0.5 degrees reflectivity lower left. While satellite has started to show buildups in the cumulus field in the county, no precipitation or lightning has developed. CI is in the 30-50% range over southeastern parts of the county, while CI-Severe was higher for the central area.

At 2157Z, precipitation developed over the northwestern part of the county, with low CI probability over the eastern part. The CI-Severe jumped up to 80%+ over the northeastern part of the county.

By 2212Z, 4 areas of precipitation had developed. Some lightning was occurring withe the easternmost and northwesternmost cells. Since convection was occurring, the CI and CI-Severe dropped almost completely.

The CI product was useful for anticipating where the showers and thunderstorms would develop, but did not provide a clue as to their intensity. All of the cells were producing lightning by around 2245Z.

Gary Cannalte

Convective Initiation over Arkansas

Looking at another example of the Convective Initiation Product over western Arkansas, from earlier this afternoon. The loop below shows the general CI product overlaid on the "sandwich" VIS/IR imagery, for the period approximately 18Z to 19Z. Looking at the cu or towering cu field over SW AR, the CI values vary considerably from one time to the next. Values are generally around 50 percent or less (green/blue), but sometimes jump up to 60-70 percent (yellow). Thunderstorms did develop over this area shortly after 21Z.  A. Cope

Overshooting top with Flash Extent Density

I found it interesting to watch the Flash Extent Density increase concurrently with the development of an overshooting top.The FED increased from a count of 2 at the onset of the overshooting top to a maximum of 11 at its peak.

Fig 1. Overshooting top detected by Day Cloud Phase Distinction RGB, overlayed with Flash Extent Density (5 minute with 1 minute update) from the GLM.

wall_cloud

Update to Earlier Blog Post on CAPE spread between AllSky/NUCAPS

In an earlier blog post today I showed that considerable spread existed between the AllSkyLAP CAPE and the NUCAPS Improved Latency CAPE values over northern SJT forecast area. While it was not available at the time of the previous blog post, the NUCAPS Modified sounding showed CAPE values that were in the ballpark of what AllSkyLAP was showing.

1941Z NUCAPS Modified sounding

Alex Brown

Other GLM Products

The 4-panel image below shows lightning over western AR. Upper-left is flash extent density (FED), which is scheduled to be the initial operational GLM product in AWIPS. Upper right is Event density which seems to show similar patterns to FED, but the values are 5 to 10 times higher. Not sure if this adds any significant value to FED.

The bottom two panels are Group Centroid Density and Flash Centroid Density. All the centroids are just single points, so each one highlights a single grid cell in the 2x2 km ABI grid, leading to a much finer "pixelated" appearance. At this point it's hard to see what operational use they might have.  A.Cope

Convection in SW Texas

An isolated thunderstorm developed over Pecos County in southwestern Texas. While the storm wasn't producing a lot of lightning, it was interesting to see how the increase in lightning activity corralated to brief increases in radar reflectivity, resulting in the production of some small hail. This thunderstorm pulsed up and down a few times in intensity and was proceeded by brief increases in the flash rate of lightning.

Also of interest was watching the anvil of the thunderstorm blow off to the east-southeast pretty rapidly, while a field of surface cumulus clouds steadily moved to the north-northwest in an area just to the east of the thunderstorm. It will be interesting to see if development takes place this evening in the zone of higher shear to the east of the afternoon thunderstorm.

Gary Cannalte

CI regular vs severe

I have noticed on severe occasions where the CI (for regular convective initiation) is much lower than the probability depicted by the Severe CI product.  Intuitively, one would think the probability for convective initiation should ALWAYS be higher than the probability for convective initiation to lead to severe weather.

Fig 1. "Regular" CI product at 2037 UTC. Note the green 50% CI probability toward central AR.

Fig 2. Severe CI product at 2037 UTC. There is a 90% probability for severe convection in the same location as the above image.

wall_cloud

Comparing 1-min, 2-min and 5-min running FED totals

For the GLM products such as Flash Extent Density (FED), we are looking at looking at running accumulations over different time periods of 1 minute, 2 minutes and 5 minutes (and maybe others?). The 5-min (1-min update) loops are smoother, but may miss or be slower to show sudden changes. The 1-min loops tend to be rather jumpy and may obscure trends. The 2-min (1-min update) loops might be a good compromise, but we are still comparing.

The 3-panel  below shows a 30-min loop comparing all three accumulation times for a storm over northwest MS. Comparing the frame-to-frame variability among the three options, the 5-min loop is certainly smoother, but has higher max values than the 2-min or 1-min loops, which may require a different color scale.

Also here is a plot from the Tracking Meteogram which compares the max values from all three accumulation types. Note the 5-min accumulation is in the middle. The 2-min running accumulation (bottom curve) might be the best combination of smoothness and sensitivity.  A.Cope

RGB of convective initiation

I found this capture of convective initiation to be interesting and cool to watch. As dry air on the south side of the MCV advected east into far western AR, storms would fire near the state line. The Day Cloud Phase Distinction shows this cycle very well and provides a nice optical depth to the satellite loop.

Fig 1. Mid-level water vapor loop showing dry air advecting into western Arkansas on the south side of the MCV.

Fig 2. Day Cloud Phase Distinction RGB depicting convective initiation on OK/AR state line on the leading edge of the dry punch.

I was looking for any obvious signal in the lightning data for that point but nothing significant seemed to stand out for this case. That said, it is producting consistently high Flash Event Density with relatively high Total Energy values and low to moderate Average Flash Area.

Fig 3. Clockwise beginning with upper-left image:  Flash Event Density 2 minute data (1 minute update), Average Flash Area 2 minute data (1 minute update), Total Power 2 minute data (1 minute update), ENTLN 5 minute data (1 minute update).

It was interesting, however, that the AllSkyLAP CAPE for this area was quite muted. Makes me wonder if this isn't a function of the model first guess. Since the sky was cloudy and the satellite data relied more heavily on the GFS (which was convectively contaminated), I didn't find this data particularly useful.

Fig 3. AllSkyLAP CAPE product showing CAPE values < 350 J/kg over eastern OK and western AR. This is using primarily the GFS background information.

Fig 4. GFS QPF for 18Z. The convective contamination likely resulted in erroneous CAPE values on the AllSkyLAP product.

wall_cloud

AllSkyLAP CAPE missing frames

The AllSkyLAP CAPE product was coming in at sporadic times. A good frame would arrive at a give time (example 1758 UTC) but then an empty frame would populate in AWIPS at 1800 UTC. This really limited the usefulness of looping the product.

Fig 1. AllSkyLAP CAPE loop with missing frames.

wall_cloud