EWP weekly debrief

Here is the breakdown from this week's EWP weekly debrief...

UWCI

This week was plagued by cirrus... can't remove it so we need to communicate well the limitations of the product. Forecasters were mentioning that they were constantly referring to the cloud type web page and suggested providing this into AWIPS. Will severely limit the applicability.

"If it's a good day to use it, the forecasters will use it... Most useful prior to development... right now things have to start developing before we issue products, this gives us some additional lead time."

Forecasters mentioned that the product would be useful in nighttime operations... possibly get more lead time... At night, stuff fires so quickly... perhaps providing a CI alarm in AWIPS would help situational awareness. Forecasters suggested providing a nighttime WES case for training. Lee is going to distribute web site and explain how to use it so they can look at it anytime.

Expressed interest in CI accumulated to follow CI signals through time in case they missed a scan.

Limitations in sensor scan time noted.

Forecasters mentioned the interest in providing more signals. they saw a lot of cases where CI was obvious but was not captured by the product. I asked if they were ok with the idea to sacrifice FAR for more detections...
"More times I expected to see something and didn't happened more often than not... not useful if things aren't showing where I expected to see it."
"Could you build different thresholds? Could you build one that was not so strict?"
"What about probabilistic detection? Then you can set your own threshold."
"Concerned about FAR getting too high because then it'll stop getting used."

There was a request for using additional bands... "You guys know the best bands to use... If you can get more information from other bands, go for it"

I asked if they saw the UWCI product being a precursor to lightning...
"Did not specifically check."
"I was more correlating the 35-40 dBZ"

During an event with a strong cap, the UWCI/CTC showed signals but no development occurred (or continued to occur)... perhaps provide a case of this in training.

From yesterday's IOP over Boulder, CO area... cell showed CI at 22:30 UTC... at 22:58 UTC a 30-40 dBZ occurred... showing 28 min lead time. Similar results were seen throughout the week.

A forecaster asked is testing on simulated satellite imagery planned? Yes, but currently the computation time is expensive so 5 min data would be rough.

Forecaster requested that it would be nice if the UWCI automatically loaded with satellite imagery in AWIPS.


OTTC

Forecasters mentioned they don't know what overshooting tops tell you... need more examples of this in training/literature.

Losing confidence by high threshold when you can see in visible... maybe a training issue... when you see a detection maybe it's more serious than a typical OT.

Combining with storm top divergence... "I think there's a lot of potential there."

Forecasters requested thermal couplet/overshoot detections in one product... hard to watch two different products... takes up a window in their D2D... "The more we can combine the better."


PGLM

Need to understand the importance of lightning "jumps"... show what would be significant... "What type of airmass dependence would we see with that?"

Need a rate of change product. Also, need to explain downward jumps... "What do we think is going on with the storm?"

Polarity information... needs to be mentioned that the ground based can be used in comparison.

Smoothing reducing peaks and faking a high res product.

Max-value track product would be useful... similar to rotation/hail track MRMS products.

"I think one of the problems with using lightning data in the WFO (currently) is that forecasters realize that it's (NLDN) such a small percentage of what's really going on."

"My guess is when the product becomes available it'll become a mainstream product that people are looking at all the time."

"Not necessarily going to be the main warning product, but it will be a good confirmation tool. If I had paid more attention and been more aware I could have issued my tornado warning one scan earlier." (24 May 2008 case event)

I asked if they noticed any lead time on CG from the PGLM... ~10 mins in general... canned case and real-time.


OVERALL/TRAINING

Training experience...
"Getting some simulations where the product does work out to field (WES cases) is important... hard to trust people with a Powerpoint. But be careful of putting out WES cases... 20 min articulate cases are fabulous... too long and we don't have time."

"Thankful for the canned case"

Asked forecasters how confident in the products they were after the training...
"With the UWCI I felt really good about, the other one (OTTC) was good too, I just probably don't experience it as much in operations"
"If you guys see something during real-time operations, don't hesitate to tell us something was there... don't hesitate to be the hands on instructor for the first couple days."

I asked what additional information they would like to see in the training...
"When you see a detection, provide a 'here's what it means' quicklook that we can refer to in operations."
Examples where it didn't work (cirrus, cap), and why are needed.

GOES-R products seem to be more useful before storm initiation.

Forecasters re-iterated need for object tracking and end to end multi-sensor tool... combine lightning, radar and satellite.

MRMS and GOES-R products are sparse grids... what could be done to improve your awareness that things are happening?
"Message or prompt to tell you when things are occurring."
"Some hesitation in alarms though... start to drown them out, especially when too many things trying to alert you."
"Number in lop left corner showing number of detections, like NLDN products, would be very useful."
"If there are a lot of false detections this would be hard."

Forecasters mentioned that you shouldn't be too worried about shipping the color curve information... forecasters are going to change it anyways.

"Glad you guys are here... good to have the interactions that we have and learn how these things work."

PGLM: Week in Review

I would like to offer my thanks to all of the forecasters who participated this week with the pseudo-GLM and other products this week. It was an educational experience for myself too, as I was provided with great suggestions for improved training and developing additional products to improve integration of future GLM data.

There were two big items I can take away from my participation this week. First, the PGLM is a strong support tool for the forecasters that helps provide confidence to the forecaster's thoughts on the current event. This matches what I have seen in previous assessments of total lightning data with NASA SPoRT's partners. Based on what was seen this week, one forecaster commented that the GLM data may eventually be a mainstream product with forecasters to provide situational awareness. Second, the utility of total lightning data will be greatly enhanced with some form of flash rate of change product. Currently, a lightning jump is a subjective analysis and there is a desire to create a more quantitative product. The first step is likely a maximum flash density "track" product that does not require a cell tracking algorithm. Ultimately, the work done by University of Alabama in Huntsville researchers to develop a lightning jump algorithm could be the best way to give forecasters valuable data in a form that does not overwhelm them during severe weather operations.

Early UWCI detection leads to golf-ball sized hail producing storm

At 2132 UTC UWCI made a detection of a 'Pre-CI Growth' with a convective element located withing the very southern region of the Boulder, CO CWA (See figure above). At approximately 2204 UTC, radar imagery indicated the first echo > 35 dBZ with this flagged convection giving the UWCI 32 minutes of lead time on the significant radar echo. At approximately 2309 UTC the ensuing storm had spotter reports of golf ball sized hail. A success!

Additionally, as a side note, UWCI has be performing well most of the afternoon, since our four-letter friends cirrus have not been creating problems for the algorithm. A lot of the early convection that has been flagged did not amount to much, since it would die as it hit the very large cap in place, thanks to ~ 19C temperature layer at 850 mb. Radar echoes would reach 40-45 dBZ to about 4000 ft and then die. Lead times on these initial convective elements were approximately 20-30 minutes. Later indications of cloud top cooling are showing more progress with vertical development.

Simulated lightning threat over western US for fire weather applications

NSSL-WRF Simulated lightning threat 3 21-hour forecast (contoured) and hourly NLDN lightning detections (red symbols) valid at 2100 UTC on 9 June 2010
Today I spoke with Mark Burger from NWS Eureka, CA who is participating in the EFP this week about the GOES-R products and our future plans for a fire weather and heavy rain experiment. One of his questions regarding the NSSL-WRF simulated lightning threat product particularly peaked my interest. He asked how the lightning threat performs over the west where it would be very useful in fire weather forecasting operations. I pulled up the output from yesterday's 00Z run and overlaid NLDN lightning detections over the OR/WA/ID area for some convection that occurred in the late afternoon. The attached image in this post shows an example of this output for the 2100 UTC time period (a 21-hour forecast), with NLDN lightning detections over the past hour shown as red + or - symbols. The lightning threat product, which predicts total lightning over each square km per 5 mins at the same time captured well the regions of peak lightning interest. This will be a very interesting thing to examine during this year's fire weather / heavy rain experiment taking place this August/September. I passed along the web link for this output, as well as some other GOES-R products, for him to use in the office and distribute amongst other forecasters.

Real-time PGLM: 9 June 2010

After a quiet first two days during our operation times, we had the chance to do real-time operations for the pseudo-GLM over the north Alabama domain Wednesday. The SPC outlook had a slight risk over the Memphis and Nashville, TN as well as Huntsville, AL county warning areas covering the northern half of our domain. Wind was the predominant threat, but severe hail was a possibility. Forecasters started in the Memphis county warning area to get a feel for the PGLM product in the far western part of the domain and then transferred to Nashville and Huntsville later in the evening. Overall, few warnings were issued, but this was a good event to introduce forecasters to the PGLM in real-time.

I will add a few general comments and then include two images from the event. Overall, like the 24 May 2008 archived case, the impression was that the PGLM served as a good situational awareness tool. I then asked about the resolution of the product. Again, the response is that forecasters always want better resolution. In this case, though, the PGLM was adequate for what it was being used to analyze. One comment was that the PGLM was useful to see the convective cores, particularly during initiation. The NLDN has been used in this role, but the PGLM can give a few extra minutes lead time on initiation since most storms initiate with intra-cloud lightning before cloud-to-ground strikes. Another interesting comment after the event was that the PGLM could be very useful for coastal WFOs where the offshore data is typically less reliable.

I will switch gears and go into some of the interesting things we saw yesterday. One item came from Huntsville's Hytop radar. The PGLM showed a jump in activity at 0016 UTC and then decreased. This was followed by a significant increase in the radar reflectivity at 0027 UTC at the -20 C isotherm level. This raised the question, "Why did was the lightning jump followed by an intensification in the radar signature?" This leads into the discussion of how lightning jumps precede severe weather. Typically, a jump will occur and then the lightning activity will decrease ahead of the severe weather event as the storm core descends. However, yesterday, we saw the echo tops increase after the lightning jump. This is likely a case of larger hail aloft developing, which results in weaker charging in the updraft. The result is less lightning, but stronger reflectivities.

Another fun example is in the image below.

This screen capture shows the PGLM flash extent density (as the 8 km blocks) along with the NLDN cloud-to-ground strike locations with the '-' and '+' symbols in yellow. Notice how the majority of storms have NLDN strikes co-located with PGLM flashes. The exception is the circled storm moving into the extreme western section of the Huntsville (HUN) county warning area. This storm was just developing and the PGLM began detecting lightning activity ahead of the NLDN strikes. The forecasters liked how this feature can be useful in detecting the initiation of convective cells and I pointed out the utility of the advanced lead time for the first cloud-to-ground strike for public products like airport weather warnings.

The second image, below, is an interesting combination of the PGLM with the IR satellite data that I have not seen used previously by forecasters.

I was interested in seeing this particular overlay and asked about its utility. The response was for monitoring the location of the convective cores. The IR shows the coldest cloud tops where convection is occurring, although cirrus can obscure what is happening below. By overlaying the PGLM over the IR imagery, the PGLM focuses the forecasters attention on the actual cores.

Overall, this was a good real-time event that generated good discussions throughout the evening.

EWP daily briefing... 6/10/2010

SPC Day-1 outlook for 10 June 2010
SPC has issued a moderate risk over the area formerly known as "Big-12 Country" (ie - Nebraska/Colorado border). The plan will be similar to what we have done for previous IOPs with the first half of the day spent monitoring the GOES-R products for convective initiation and the second half of the day monitoring the MRMS products during warning operations.

During the briefing, Lee discussed the previous day. Unfortunately the area we targeted yesterday did not have much happen in terms of UWCI due to a widespread area of cirrus. We were able to see some overshooting top detections over northern Alabama, but they were not really associated with a lot of severe weather. There was only one thermal couplet detection but was not used by the forecasters during warning operations because they did not notice it. Most of yesterday's IOP was focused on the PGLM product, which Geoffrey Stano will post on in more detail in an upcoming post, and we got some very good feedback from that, as well as in the case example that Geoffrey has already discussed. Lee did show an area over NM/CO where the UWCI was showing 30-35 minute lead times on the first occurrence of a 35 dBZ echo on composite radar.

One important thing I noted during the discussion is that the forecasters need some sort of tool to help them forecast, and/or diagnose areas of severe winds. Right now they mentioned they have nothing that really helps them and it is really hard for them to warn for severe wind. Something to keep in mind for possible applications of GOES-R data, but I know this would be a difficult task. Nonetheless, it is an area where a obvious void in forecaster tools are present.

Discussions from the 24 May 2008 Archived Case

As I mentioned in the previous pseudo-GLM post, there were several good lines of discussion during and after the event was run.

I had asked what the forecasters felt about the pseudo-GLM as a whole and how the WFOs use lightning now. One common response was that lightning is a "good to know" product, particularly for situational awareness. Lightning is good for outdoor events and monitoring lightning initiation. The pseudo-GLM improves the lead time on the first cloud-to-ground strike as most storms begin with intra-cloud lightning before the first cloud-to-ground strike. The pseudo-GLM also can assist in radar poor regions as well as gauge the intensity of the lightning activity.

Additionally, I was a little too concerned with influencing the forecasters' thought process during the event. As a result the lightning jump described in the previous post was not readily seen. During the post analysis, the discussion focused on the lightning jump and how it did help flag the southern cell a few minutes earlier than radar alone. The post-event analysis helped solidify the connection between the lightning jump and what was observed on radar.

I also asked about how the 8 km resolution worked for the forecasters. The main response is that everyone always prefers higher resolution. With that said, this resolution should be good enough to distinguish between the updraft and forward flank downdraft region of storms. It also can show the cores of individual storms.

Another comment by the forecasters was the interest in seeing total lightning during winter storms. They wish to determine if there is convective snow and they have found that the National Lightning Detection Network (that detects cloud-to-ground strikes) rarely indicates lightning. Knowing that a snow event is convective could lead to a doubling of the estimated snowfall amount.

One of the final topics I asked about was how to better include lightning data in public products. Currently, the WFOs do not issue "lightning warnings" like the 45th Weather Squadron does for Kennedy Space Center and Cape Canaveral Air Force Station. Also, the typical warning for severe thunderstorms or tornadoes starts with, "Doppler radar has indicated ..." We discussed how lightning wording could be added and I included some of my conversations with the Huntsville, AL and Melbourne, FL WFOs. There is the potential to state that, "A GLM lightning jump was detected ..." but this incurs problems with interpretation by the public. An excellent suggestion by the forecaster participants was to include the phrase, "... GLM lightning trends indicate a higher confidence that this storm is becoming severe or producing a tornado." This helps convey confidence in the forecast as well as adding additional lightning awareness.

Tomorrow I will update the blog with the discussions and comments from the real-time operations we performed over the Huntsville, Alabama domain today.

24 May 2008 Archived Lightning Case

With no lightning in any of our total lightning domains during operations Monday and Tuesday, the forecasters focused on the archived case from 24 May 2008 near Enid, Oklahoma. This case has served as a good introduction to the pseudo-GLM flash extent density product ahead of the real-time operations that have been taking place Wednesday afternoon. This event was interesting to go through, with a northern cell in Garfield County and a later, southern cell in Kingfisher County. The scenario starts with the northern cell already active. Both cells showed strong lightning jumps ahead of intensification.

For the northern cell, the jump preceded the formation of a strong rotation couplet. However, storm reports did not indicate an observed tornado or severe weather. I had remarked after the event that this was one of the stronger jumps I had seen without a severe weather report. Most of the forecasters warned on this storm due to the radar velocity signature, but the lightning jump helped add confidence. Potentially, severe weather occurred, but due to the location of the storm nothing was observed.

The southern cell demonstrated the utility of total lightning for both lightning safety and severe weather operations. As the southern cell began to develop, the pseudo-GLM indicated lightning activity 5 minutes before the first cloud-to-ground strike. Initially the pseudo-GLM was showing a flash density of 6 flashes at 1902 UTC. This gradually increased for the next 8 minutes and then demonstrated a rapid jump to 86 flashes by 1919 UTC. This occurred 8 minutes ahead of the first reported tornado just west of Lacey, Oklahoma at 1927 UTC. During the post analysis, this lead time was significant for the event as the radar velocity signature was weak at 1921 and was just showing signs of a couplet at 1926 UTC. The reflectivity was showing some signs of a hook at 1921.

I feel the discussions from going through this event with four forecasters were the best results. I will follow-up with this in a subsequent post.

EWP daily briefing... 6/9/2010

Before the official briefing started, one of the forecasters mentioned to Lee Cronce and myself the idea of creating a filter for the CI product where cirrus is present so that the forecasters will know when the product will not be able to make any nowcasts, similar to how the radars have a certain color for areas where range folding is occurring. He mentioned that at some times they cannot understand why the product is not producing nowcasts in obvious situations of CI unless someone looks up the cloud type output for them. This was noted by Lee as a possibility for future input into the product's output.


24-hour thermal couplet detections and severe reports for 7 June 2010
During the briefing Lee presented some scan by scan output from the UWCI product over Wyoming during the 6/7 IOP. He mentioned that the first signal in the UWCI was seen at 2032 UTC, with more widespread signals occurring at 2045 UTC and beyond. The area of interest was actually located between two radars where coverage is fairly limited and the UWCI provided some information on the initiation about 15 minutes prior to detection on radar. He also showed the 24-hour thermal couplet detections and severe reports from the same event (see images above). There were a lot of detections associated with the severe reports, including one that was co-located with and prior to a report of a tornado near Scottsbluff, NE.

One of the forecasters mentioned the possibility of a GOES-R storm top divergence product. He mentioned how this would be very useful in determining storm characteristics in warning operations. I mentioned Bob Rabin's work with current GOES WV winds in producing real-time winds at various levels by tracking features in the WV imagery. From these winds he is able to calculate divergence, vorticity and wind speed contours. I showed them real-time examples from the website provided by Bob Rabin at http://cimss.ssec.wisc.edu/mesoscale_winds/. The forecasters seemed very interested in this and mentioned interest in demonstrating this in future experiments. I also made sure to point out that with GOES-R's great improvement on spatial and temporal resolution, these products will be much more accurate and will be able to capture smaller scale features. I also talked about GOES-R's increased capability for object tracking.

During the weather discussion I showed the forecasters the UW-CIMSS simulated satellite/observed satellite comparison page to see how well the NSSL-WRF was representing the day's weather. It seemed to be doing fairly well so I showed the simulated lightning threat product to see what the NSSL-WRF was showing for the possibility of lightning over the CONUS region. There seemed to be the possibility of some lightning occurring over the Nashville area later on this evening, so following the weather discussion for the day we decided to localize near Huntsville, AL for the possibility of real-time PGLM operations.

Thermal Couplet detection prior to spotter funnel report

An interesting thermal couplet case not within the assigned WFO emulations. At 2202 UTC a thermal couplet was detected right near the border of northeastern KS/northwestern MO within the EAX CWA. At that time, no LSRs indicated any severe reports other than flooding. Approximately 15 minutes later a report came in of a funnel cloud via a spotter network. As an additional note, there is currently no severe thunderstorm warning for the storm with the detected thermal couplet. Perhaps this would be a good case of increasing forecaster confidence for warning with the associated thermal couplet detection?

Now 2239 UTC, and the cell that produced the thermal couplet is now tornado warned

TX Panhandle CI in the HWT-EWP 6-8-10

Today in the NOAA HWT EWP in Norman we have been watching a line of cu over the TX Panhandle. At 1932 UTC we first noticed the CI "pre-cloud growth" and the "likely" indications. It was interesting to note however that the "likely" was associated with cloud-top cooling rates in the -4 to -7 range, but the "pre-cloud growth" was associated with greater values of cloud-top cooling rates of around -9. Eventually the area of -9 became the first convective initiation spot, and those values increased to -14 before we saw the first 35 db echo. The first 35 db echo showed up at 1955 UTC, which was 23 minutes later. At 2025 UTC a 1" hail report was logged with that same storm.

One other point to note was that we were monitoring the products on the web site (http://cimss.ssec.wisc.edu/snaap/convinit/quicklooks2.php) and estimated that the CI indications were approximately 3 minutes faster than the LDM feed into AWIPS.

Identified Overshooting Top

The above overshooting top was identified at 2032 UTC outside the current IOP area of Wichita and Topeka, Kansas, with the associated severe thunderstorm warning issued at 2038 UTC by the Shreveport WFO. However, unsure what the forecasters were thinking when issuing the warning, and whether the OT product would have been helpful in their decision.

EWP daily briefing... The 2345 UTC hour-long satellite gap

At the beginning of the briefing we discussed the previous day's event. Unfortunately due to a few problems we were not able to draw any conclusions from the UWCI or OTTC products from yesterday's event. First, we were not able to display any satellite imagery, so overlaying the UWCI and OTTC products was not possible. Lee Cronce did suggest that we overlay the products on radar towards the middle of the IOP last night, but at this point the second problem came into play... cirrus. During the later half of the IOP the entire region was covered with thin/thick cirrus and thus made the UWCI products inoperable. The OTTC product were still working, but at this point we encountered the hour long void of satellite data from 2345 UTC to 0045 UTC. This hour long void has been causing a lot of trouble during our evening IOPs in the EWP. I have also been hearing complaints from SPC operations as this is a very important time-frame for severe weather and the forecasters don't understand why this time period was selected. Once this occurs we essentially have to shut down the GOES-R aspect of the IOP because the forecasters become disillusioned by the extremely poor temporal resolution of any satellite data during this time period. By the time the satellite data returns, there is no continuity in the products in the forecasters minds and they therefore stick to the radar products they have been comfortable looking at over the past hour. The silver lining from yesterday is that we did see our first thermal couplet detection in real-time last night and it did occur at the time of the issuance of a tornado warning by one of the forecaster groups. In this case it did not provide any lead time, but it did provide confidence in their warning decision.

During the weather briefing I showed the forecasters the NSSL-WRF simulated satellite imagery and lightning threat to help determine the location for tonight's IOP. The forecasters were very interested in the model output and asked to see it every day during the weather briefings. For tomorrow I will also show them the verification page that UW-CIMSS has created for the simulated satellite imagery. We decided to spend the first half of the IOP by splitting the forecasters up into two groups... one participating in a WES case for the PGLM, while the other monitors the location of interest for this evening's IOP for convective initiation. We have decided to localize over Kansas City, MO for the real-time IOP. Updates to follow.

Alternatives to satellite underlayment

As an alternative to overlaying the CI product on satellite imagery, I suggested that the forecasters overlay it on radar imagery for lead time purposes. The caveat in this is that the radar data was down for a while, too. They were able to get the mosaics back up and running, but the composite imagery could not be referenced.

As a side note, the Goodland, KS WFO emulation was transferred over to the North Platte, NE WFO.

EWP technical difficulties

As is the nature of experiments, things do not work all of the time. Today following the training we went into a real-time IOP with the groups localized over Cheyenne, WY and Goodland, KS. Unfortunately the feed for the satellite imagery into the AWIPS systems went down and cannot be fixed until the morning, so it makes it hard to evaluate the Proving Ground products today. Fortunately this is just a familiarization day so we didn't totally lose out on a day with the forecasters.

EWP back in action

Day 1 convective outlook for June 7 2010 issued at 1630 UTC
Today the EWP starts up again after a (refreshing) week off. Lee Cronce and Sarah Monette from UW-CIMSS are here to provide support for the UWCI and OTTC products, and Geoffrey Stano from NASA SPoRT is here to provide support for the PGLM product. Training is taking place for the first half of the day, but we plan to do a short familiarization IOP during the evening with some severe weather expected to occur near the CO/WY/NE border. We will be posting regularly throughout the week on forecaster interactions and some discussion on product performance.