Here are some important bullet points from this week's EWP debrief...
UWCI and Cloud-top Cooling
- Forecasters wondering how well it will work in the Northeast... so often cirrus is in the area... Saw very few signals for yesterdays case.
- A little ahead of total lightning with yesterday's case... not like case event when it occurred at same time
- Cloud-top cooling more valuable for situational awareness than CI
- "Wonder if this type of product would be useful in something like guardian or alert type software, because not all people will be looking at it all the time."
- "Would be neat to look at if it saved all the locations where CI was occurring... would like to have the 60-min accumulated.. make it more like the hail tracks" (mentioned this was available)
- "I was watching a storm over Amarillo with 60+ dBZ after CI and there was no CG... I was waiting for lightning to happen because that's what we were told the research used to verify it... You can't use CG as a discriminator... you need total lightning."
OTTC
- All week continued to lag radar signals of severe.
- No thermal couplets saw in real time all week
- Forecasters understand it should work better with 2 km data.
- Similar comment a UWCI about accumulated product
- Overall was not very useful in warning operations.
Pseudo-GLM
- During real-time it was used a little bit, but the events were not as significant (37 flashes/2 min vs over 100 flashes/2 min in archive case)
- "What do these numbers mean?" training needed... need to see it several times over the summer to get used to it.
- "Might be useful to have a grad student or someone track particular features in radar with lightning flash rates to make some correlations... In theory you saw increases in reflectivity with jumps in lightning rates."..."You would have to prove to me that this jump rate is a precursor to tornado occurrence before I would be willing to use it in operations... more research."
- Regarding archive case... lightning jumped way up 5 mins before lightning... dropped significantly as tornado occurred.
- Need for trends (like a max VIL) noted... then they don't have to sample the whole thing every scan... Eric Bruning showed Schultz et al. 2009 work... Also mentioned flash rate trend swath... "Let's do that for next year, I would love to see that."
- 8km resolution made hard to see individual updraft areas... smoothing looked nice and confused forecasters making them think it was higher resolution, but in fact it was removing the peaks.
- "That's a neat product, but by having 1km visible and 0.5km visible on GOES-R, having an 8km product seems like it wouldn't provide as much information as we could be getting from other sources."
- Noticed that when it came to warning decisions, forecasters went straight to standard radar tools (reflectivity tilts, velocity)... "It's what we're trained to do... it's the best tool for those things."
- Issue with identifying polarity for flashes... Told that you can compare NLDN to the GLM
- May be more of an aviation, winter weather thing... "I see it really useful for convective snow events for picking up areas of convective snow."... "Winter storms are more costly for our area." Felt comfortable throwing in winter weather archive cases in Spring if necessary.
Overall / Training
- Felt very comfortable with using the products off the bat... "Eventually as the week progressed, I had my 4-panels set for the products that I found more useful to me."
- "Need a thorough review of how you come up with the values for these data."
- "I think you will find few forecasters that will readily abandon base data interpretation."
- "Try to make the experiment as real as possible as we would experience in the WFO... give us what we're used to."
- "Do more WES cases... don't go into real-time ops for marginal severe events... but it was nice to have live data since we didn't know what or if anything was going to happen."
- "Having some pre-set procedures would be good"
- "Maybe having a checklist would be good to make sure we check out all the products."
- First day come in early and do training day... felt like they were doing "hurry up and wait"
Friday, May 28, 2010
SATCAST within the GOES-R Proving Ground
Over the past two weeks, SATCAST has been brought into the Hazardous Weather Testbed for evaluation by the University of Alabama in Huntsville, the developer of the GOES-R Algorithm Working Group Convective Initiation (CI) algorithm. John Walker and myself have participated in the Experimental Forecast Program and we have learned a great deal on how to best prepare the CI algorithm for forecaster use and prepare for GOES-R.
Over the past two weeks, the SATCAST algorithm performed very well with average lead-times ranging from 15 to 45 minutes. In high CAPE environments, the algorithm was more diagnostic, however, GOES-R will be able to give high temporal trends to allow for more frequent cloud-top trends and give more lead-time for the algorithm. There were some things learned that need some improvement and we plan to work on those weaknesses over the coming months.
Thursday, May 27, 2010
Graphical plots of lightning trends
In our evaluation of the 24 May 2008 archive case today, forecasters highly recommended implementation of a gridded map display of lightning trends in addition to counts. The idea is as follows:
Work by Schultz et al (2009) has shown the applicability of a two-standard-deviation jump of the flash rate derivative above a running mean as predicator of hail, wind, and tornado events. About a year ago, I prototyped a visualization of a cell-entity plot that could graphically show how close the current flash rate was to being considered a lightning jump. These data are courtesy Chris Schultz, for a case from July 16, 2007.
In the video below, the top two panels show time versus (top) flash rate) and (bottom) the flash rate derivative with respect to time (DFRDT), and various threshold- and standard deviation-based thresholds for lightning jump. The bottom panel shows a plan view of a thunderstorm cell colored according to the ratio of DFRDT to the 2-sigma trigger (thick blue line vs. pink line). Yellow colors show that the cell is fliriting with jumping, and the discontinuity to red in the color scale at a ratio above 1.0 calls attention to the jump without requiring an additional symbolic flag.
A real storm cell derived from the WDSSII k-means tracking approach would have a non-circular entity shape, which would obviously be much more realistic than the simplistic circle used here. This demo doesn't show the swath idea, but it is clear that it could definitely improve awareness of flash rate trend history at a glance.
Also, a forecaster just suggested to me that a plot of IC:CG ratio would be interesting, perhaps implemented on a cell basis like the trend plot discussed above. He noted interest in continued availability of the NLDN ground strike (and polarity) data.
Reference:
Schultz, C. J., W. A. Petersen, and L. D. Carey, 2009: Preliminary development and evaluation of lightning jump algorithms for the real-time detection of severe weather. J. Appl. Meteor. Climatol., 48, 2543–2563.
Need to have trend plots, because it's too hard to figure out by querying flash count grids. The forecasters preferred a gridded plan view / map mode (not a line graph), and either a plot of (1) the flash rate derivative or, in the lightning jump sense, (2) the number of standard deviations (possibly fractional, e.g., 1.5) relative to the running mean. This could be implemented using the WDSSII k-means cell shape colored according to the above trend metrics. 30 min time lapse trend swath would also be helpful.
Work by Schultz et al (2009) has shown the applicability of a two-standard-deviation jump of the flash rate derivative above a running mean as predicator of hail, wind, and tornado events. About a year ago, I prototyped a visualization of a cell-entity plot that could graphically show how close the current flash rate was to being considered a lightning jump. These data are courtesy Chris Schultz, for a case from July 16, 2007.
In the video below, the top two panels show time versus (top) flash rate) and (bottom) the flash rate derivative with respect to time (DFRDT), and various threshold- and standard deviation-based thresholds for lightning jump. The bottom panel shows a plan view of a thunderstorm cell colored according to the ratio of DFRDT to the 2-sigma trigger (thick blue line vs. pink line). Yellow colors show that the cell is fliriting with jumping, and the discontinuity to red in the color scale at a ratio above 1.0 calls attention to the jump without requiring an additional symbolic flag.
A real storm cell derived from the WDSSII k-means tracking approach would have a non-circular entity shape, which would obviously be much more realistic than the simplistic circle used here. This demo doesn't show the swath idea, but it is clear that it could definitely improve awareness of flash rate trend history at a glance.
Also, a forecaster just suggested to me that a plot of IC:CG ratio would be interesting, perhaps implemented on a cell basis like the trend plot discussed above. He noted interest in continued availability of the NLDN ground strike (and polarity) data.
Reference:
Schultz, C. J., W. A. Petersen, and L. D. Carey, 2009: Preliminary development and evaluation of lightning jump algorithms for the real-time detection of severe weather. J. Appl. Meteor. Climatol., 48, 2543–2563.
First real-time pseudo-GLM images on AWIPS
Above is the first real-time image we have seen from the psuedo-GLM product over the DCLMA at 2104 UTC. We are currently participating in a real-time IOP over the region due to the severe weather threat and a few severe thunderstorm warnings have already been issued by the forecasters who are localized over State College, PA and Philadelphia, PA.
EFP daily map discussion
Today's EFP map discussion included some very interesting interactions regarding GOES-R. Bruce Entwistle from the Aviation Weather Center who was participating in the EFP's aviation forecast group brought up their afternoon forecast on a loop of visible satellite imagery with cloud-top cooling rates overlaid during the briefing. He described how the cloud-top cooling rates helped focus them on area of interest for deepening convection and helped them nudge their forecast outlook lines slightly. He said that he was seeing rates exceeding 22 K/15-min at some times. He noted that for this sort of application where they are expected to issue their forecasts hours in advance, the cloud-top cooling and convective initiation products may not be as useful, but regional forecast facilities that are required to issue more short term forecasts may find this invaluable.
Louie Grasso brought up a separate discussion regarding the fire activity up north of Montreal, Canada. He talked to the participants about how the 3.9 micron channel was able to detect fire hotpots and showed a very good example of that in real-time over Canada. He mentioned how since GOES-R will have higher spectral resolution, we will be better able to determine the intensity and distribution of the fires. He pointed out that the smoke plumes associated with these fires were hard to distinguish from the clouds around them, and may even become hard to see at all when the sun angle is high since they can be very thin. He told the participants that when GOES-R is available, additional visible and near-IR channels will be able to distinguish different cloud and aerosol types through techniques such as RGB composites. He also pointed out to the participants that the 3.9 micron channel was composed of a reflected and emitted component and showed them an example of a MCS during day and night hours, which showed the cloud tops "cooling" substantially during the switch from night to day. Since the reflected solar component of the 3.9 micron channel is very sensistive to ice particles, the cloud seem's "cooler" when the sun is up. The participants seemed very interested in this and seemed to capture their interest into the wide variety of uses of satellite data outside of visible and IR images alone.
Louie Grasso brought up a separate discussion regarding the fire activity up north of Montreal, Canada. He talked to the participants about how the 3.9 micron channel was able to detect fire hotpots and showed a very good example of that in real-time over Canada. He mentioned how since GOES-R will have higher spectral resolution, we will be better able to determine the intensity and distribution of the fires. He pointed out that the smoke plumes associated with these fires were hard to distinguish from the clouds around them, and may even become hard to see at all when the sun angle is high since they can be very thin. He told the participants that when GOES-R is available, additional visible and near-IR channels will be able to distinguish different cloud and aerosol types through techniques such as RGB composites. He also pointed out to the participants that the 3.9 micron channel was composed of a reflected and emitted component and showed them an example of a MCS during day and night hours, which showed the cloud tops "cooling" substantially during the switch from night to day. Since the reflected solar component of the 3.9 micron channel is very sensistive to ice particles, the cloud seem's "cooler" when the sun is up. The participants seemed very interested in this and seemed to capture their interest into the wide variety of uses of satellite data outside of visible and IR images alone.
24 May 2008 Archive Pseudo-GLM: More feedback
Today two new forecasters will examine the 24 May 2008 Oklahoma LMA case that was also done yesterday.
See the end of this post for several concrete recommendations of enhanced plot types suggested by the forecasters.
-------
A first reaction from the first few frames of total lightning data: Learning curve associated with the change from ground strike lightning data to total lightning data. The larger flash rates require recalibration on what rates are typical.
18:05 - Weakening trend noted in both ground and cloud flash rates. Down to 18/min in Psuedo-GLM. Also tracking total flash rates vs. 50dBZ height.
Request for max flash rate display (like 88D max dBZ readout) - reduces clicks / exploration to get readout.
18:22 - Starting trend back up in lightning, up to 26/min peak.
"If this comes in every minute, it will definitely beat the radar, since we can spot strengthening/weakening trends prior to the radar data." Conducting real-time test of the that hypothesis via the 50 dBZ comparison.
Again noting dropouts in the archived CG and LAPS data.
"Anything that can give me a few minutes of extra time, I want it. Give it to me." - referring to the total lightning data. Has been switching back and forth between lightning and radar, and carefully monitoring every lightning frame.
18:46 - Southern cell initiated, seen in satellite cooling trends and elevated weak reflectivity core prior to ocurrence of lightning. Explained that since lightning requires precipitation-sized ice, this is expected, but is helpful in tracking stages of convective initiation through mature thunderstorm.
18:28 - up to 53/min
19:00 - SVR issued on both storms
19:08 - Lightning in second cell noted as catching up to the first
Again it seems that once the warning issuance process has begun, radar interrogation occupies most of the forecaster screen time. The radar provides evidence of the actual physical manifestation of hail, so this is more worthy of time than a secondary indicator of convective strength in a less-specific sense.
Might consider using sparklines(miniature trend lines displaced slightly from the cell center, with peak and minimum label values) as a low-impedence way to do non-context-switched examination of total lightning trends.
19:19 - "Broad rotation the whole time in northern storm, but hasn't tightened up."
Near the end of the case, lightning was noted as perhaps being more important in pulse storm situations where lightning would be more clearly indicative of important updraft fluctuations. Not so helpful in this sort of case, with little indication in lightning relative to the specific tornado times (or hail presence, which was already well-determined by radar). But great as situational awareness (like MESH) in making sure no dangerous cells are being missed.
Need to have trend plots, because it's too hard to figure out by querying flash count grids. The forecasters preferred a gridded plan view / map mode (not a line graph), and either a plot of (1) the flash rate derivative or, in the lightning jump sense, (2) the number of standard deviations (possibly fractional, e.g., 1.5) relative to the running mean. This could be implemented using the WDSSII k-means cell shape colored according to the above trend metrics. 30 min time lapse trend swath would also be helpful.
See the end of this post for several concrete recommendations of enhanced plot types suggested by the forecasters.
-------
A first reaction from the first few frames of total lightning data: Learning curve associated with the change from ground strike lightning data to total lightning data. The larger flash rates require recalibration on what rates are typical.
18:05 - Weakening trend noted in both ground and cloud flash rates. Down to 18/min in Psuedo-GLM. Also tracking total flash rates vs. 50dBZ height.
Request for max flash rate display (like 88D max dBZ readout) - reduces clicks / exploration to get readout.
18:22 - Starting trend back up in lightning, up to 26/min peak.
"If this comes in every minute, it will definitely beat the radar, since we can spot strengthening/weakening trends prior to the radar data." Conducting real-time test of the that hypothesis via the 50 dBZ comparison.
Again noting dropouts in the archived CG and LAPS data.
"Anything that can give me a few minutes of extra time, I want it. Give it to me." - referring to the total lightning data. Has been switching back and forth between lightning and radar, and carefully monitoring every lightning frame.
18:46 - Southern cell initiated, seen in satellite cooling trends and elevated weak reflectivity core prior to ocurrence of lightning. Explained that since lightning requires precipitation-sized ice, this is expected, but is helpful in tracking stages of convective initiation through mature thunderstorm.
18:28 - up to 53/min
19:00 - SVR issued on both storms
19:08 - Lightning in second cell noted as catching up to the first
Again it seems that once the warning issuance process has begun, radar interrogation occupies most of the forecaster screen time. The radar provides evidence of the actual physical manifestation of hail, so this is more worthy of time than a secondary indicator of convective strength in a less-specific sense.
Might consider using sparklines(miniature trend lines displaced slightly from the cell center, with peak and minimum label values) as a low-impedence way to do non-context-switched examination of total lightning trends.
19:19 - "Broad rotation the whole time in northern storm, but hasn't tightened up."
Near the end of the case, lightning was noted as perhaps being more important in pulse storm situations where lightning would be more clearly indicative of important updraft fluctuations. Not so helpful in this sort of case, with little indication in lightning relative to the specific tornado times (or hail presence, which was already well-determined by radar). But great as situational awareness (like MESH) in making sure no dangerous cells are being missed.
Need to have trend plots, because it's too hard to figure out by querying flash count grids. The forecasters preferred a gridded plan view / map mode (not a line graph), and either a plot of (1) the flash rate derivative or, in the lightning jump sense, (2) the number of standard deviations (possibly fractional, e.g., 1.5) relative to the running mean. This could be implemented using the WDSSII k-means cell shape colored according to the above trend metrics. 30 min time lapse trend swath would also be helpful.
EWP daily briefing... 5/27/2010
During the weather briefing we showed the NSSL-WRF lightning threat and simulated satellite imagery since it seemed as though the NSSL-WRF was handling the weather pattern fairly well today. The decision was made to focus over the DC area since there is a slight risk and severe thunderstorm watch out at the time of the briefing, as well as the DCLMA for pseudo-GLM demonstrations. Unfortunately it looks as though the lightning threat decreases fairly rapidly as the storms move south towards the DCLMA network according to the NSSL-WRF lightning threat output, but we hope that we will be able to capture some information.
Synthetic ABI brightness temperatures
Synthetic ABI brightness temperatures are being generated
at CIMSS and CIRA using output from the 00 UTC NSSL-WRF
4-km model simulations. Discussions have been held during
the past several days describing how infrared imagery can
be used to quickly evaluate the structural evolution of
the simulated thunderstorms and other cloud features as
well as to examine the overall accuracy of model
forecasts. Simulated IR imagery provides a powerful means
to quickly evaluate model output since various
characteristics such as cloud top height, water vapor
distribution, and thunderstorm coverage and intensity can
be inferred from a single image rather than having to look
at many different fields. As an example, the synthetic
imagery was used today to determine how quickly a
low-level stratus cloud deck present over the high plains
of NE Colorado and SE Wyoming would dissipate this
morning. Real GOES imagery indicated that the cloud cover
was not dissipating very quickly, which raised some
concerns that this would reduce surface warming and impede
the development of thunderstorms later in the day. A
quick examination of the synthetic 11.2 micron imagery,
however, quickly showed that this cloud cover would
dissipate by late in the morning. This provided greater
confidence that a risk for severe thunderstorms was
warranted over that region. The clouds did indeed
dissipate during the morning and several severe
thunderstorms, including one near the Denver airport,
developed during the afternoon.
-Jason Otkin
at CIMSS and CIRA using output from the 00 UTC NSSL-WRF
4-km model simulations. Discussions have been held during
the past several days describing how infrared imagery can
be used to quickly evaluate the structural evolution of
the simulated thunderstorms and other cloud features as
well as to examine the overall accuracy of model
forecasts. Simulated IR imagery provides a powerful means
to quickly evaluate model output since various
characteristics such as cloud top height, water vapor
distribution, and thunderstorm coverage and intensity can
be inferred from a single image rather than having to look
at many different fields. As an example, the synthetic
imagery was used today to determine how quickly a
low-level stratus cloud deck present over the high plains
of NE Colorado and SE Wyoming would dissipate this
morning. Real GOES imagery indicated that the cloud cover
was not dissipating very quickly, which raised some
concerns that this would reduce surface warming and impede
the development of thunderstorms later in the day. A
quick examination of the synthetic 11.2 micron imagery,
however, quickly showed that this cloud cover would
dissipate by late in the morning. This provided greater
confidence that a risk for severe thunderstorms was
warranted over that region. The clouds did indeed
dissipate during the morning and several severe
thunderstorms, including one near the Denver airport,
developed during the afternoon.
-Jason Otkin
Wednesday, May 26, 2010
SATCAST within Aviation Applications
The GOES-R Convective Initiation algorithm being developed by the University of Alabama in Huntsville within the GOES-R Algorithm Working Group Aviation Applications Team. Nowcasting convective initiation is very important to aviation interests for airport operations and airline planning. If there is a good indication that a storm will develop 30 minutes to 1 hour, the airlines can plan for a particular aircraft takeoff/landing route to close causing an airport to reduce capacity for take-offs and landings or shutdown airport operations. The airlines would be able to hold back some aircraft at other airports, preventing a backlog at a particular airport and cause airplane diversions which can be costly.
In addition, knowledge of thunderstorm growth can help airport operators plan for the potential of lightning strikes and clear the ramp preventing any possible injury/deaths from lightning strikes.
There were significant airmass thunderstorm development over Illinois, eastern TN, and over the Atlanta area today. Below is an example of SATCAST over the Atlanta region which is a major Delta and Airtran Airline hub.
SATCAST Nowcast valid 1702 UTC
Radar valid 1709 UTC
Radar valid 1903 UTC
Radar valid 2000 UTC
Below are examples over the Ohio valley. Notice the convection which developed over central Illinois and around the southern eastern KY area.
SATCAST Nowcast valid 1732 UTC
Radar valid 1733 UTC
Radar valid 1811 UTC
Radar valid 1830 UTCGOES-R Proxy products used in archive mode from 24 May 2010
The four forecasters split into two groups, one group focused on real-time storm development in western Oklahoma area while the other is looking at GOES-R proxy GLM lightning, convective initiation, overshooting-top, and enhanced-V signal from May 24, 2010 convective outbreak in Oklahoma area to take advantage of OKC LMA data. The figure below shows pseudo GLM (upper left), UW convective initiation (upper right), UW cloud top cooling (lower left), and visible lower right. This allowed forecasters to access GOES-R GLM proxy cloud-to-cloud lightning in conjunction with GOES imager based convective initiation/cloud-top cooling within AWIPS. Cloud-to-cloud lightning was shown to first occur at first indication of cloud-top cooling in this case.
Subscribe to:
Posts (Atom)