The EWP2011 Thank You Post

Here is our Thank You post for EWP2011, conveying our appreciation to the hard work and long hours put in by our forecasters, developers, and other participants for our spring experiment.  Even though we had a short experiment this year owing to a “slight decrease” in funds, the four weeks were successful nonetheless.  We had several noteworthy events, the biggest probably being the 24 May 2011 Central Oklahoma tornado outbreak, when our participants were asked to leave the HWT area to go to the NWS storm shelters on the first floor as a tornado dissipated just 2 miles away and debris rained on the building.   Our AWIPS1 software worked very well this year, owing to the fact that AWIPSII was still not yet ready for primetime.  And the software worked better than ever – of course with most bugs finally being fixed by the end of the experiment.

The biggest expression of thanks goes to our two AWIPS/WES gurus “on loan” from the NWS Warning Decision Training Branch, especially Darrel Kingfield, as well as Ben Baranowski early on (who left WDTB before the start of the experiment).   In addition, we had help from the Norman NWS forecast office from Matt Foster.  Both Darrel and Matt put in tons of effort getting our AWIPSII system up and running until we found the fatal memory leak that put our AWIPSII aspirations on hold.  Greg Stumpf provided AWIPS1 support to format and import the experimental data sets from various sources.

These scientists brought their expertise to the experiment to help guide live operations and playback of archive cases for each of the experiments:

For the Warn-On-Forecast 3D Radar Data Assimilation project we’d like to thank the principle scientists, Travis Smith and Jidong Gao, as well as their support team of Kristin Kuhlman and Kevin Manross (all from CIMMS/NSSL), and David Dowell (GSD).

For the OUN WRF project, they included principle investigators Gabe Garfield (CIMMS/NWS WFO OUN) and David Andra (NWS WFO OUN).

For the GOES-R Proving Ground experimental warning activities, including the Pseudo- Geostationary Lightning Mapping (pglm) array experiment, our thanks go to principle scientists Chris Siewert (CIMMS/SPC) and Kristin Kuhlman (CIMMS/NSSL), along with Wayne Feltz (UW-CIMSS), John Walker (UAH), Geoffrey Stano (NASA/SPoRT), Ralph Petersen (UW-CIMSS), Dan Lindsey (CSU-CIRA), John Mecikalski (UAH), Jason Otkin (UW-CIMSS), Chris Jewett (UAH), Scott Rudlosky (UMD), Lee Cronce (UW-CIMSS), Bob Aune (UW-CIMSS), Jordan Gerth (UW-CIMSS), Lori Schultz (UAH), and Jim Gurka (NESDIS).

We had undergraduate students helping out in some real-time support roles including monitor real-time severe weather reports.  They included Alex Wovrosh (Ohio University), as well as Ben Herzog, Brandon Smith, and Sarah Stough (all CIMMS/NSSL).

Next, we’d like to thank out four Weekly Coordinators for keeping operations on track: Kristin Kuhlman, Kevin Manross, Travis Smith, and Greg Stumpf.

We had much IT help from Kevin Manross, Jeff Brogden, Charles Kerr, Vicki Farmer, Karen Cooper, Paul Griffin, Brad Sagowitz, and Greg Stumpf.

The EWP leadership team of Travis Smith and David Andra, along with the other HWT management committee members (Steve Weiss, Jack Kain, Mike Foster, Russ Schneider, and Steve Koch), Stephan Smith, chief of the MDL Decision Assistance Branch, and Steve Goodman of the GOES-R program office, were instrumental in providing the necessary resources to make the EWP spring experiment happen.

Finally, we express a multitude of gratitude to our National Weather Service and international operational meteorologists who traveled to Norman to participate as evaluators in this experiment (and we also thank their local and regional management for providing the personnel). They are:

Jerilyn Billings (WFO Wichita, KS)

Scott Blair (WFO Topeka, KS)

Brian Curran (WFO Midland/Odessa, TX)

Andy Taylor (WFO Norman, OK)

Brandon Vincent (WFO Raleigh, NC)

Kevin Brown (WFO Norman, OK)

Kevin Donofrio (WFO Portland, OR)

Bill Goodman (WFO New York, NY)

Steve Keighton (WFO Blacksburg, VA)

Jessica Schultz (NEXRAD Radar Operations Center)

Jason Jordan (WFO Lubbock, TX)

Daniel Leins (WFO Phoenix, AZ)

Robert Prentice (Warning Decision Training Branch)

Pablo Santos (WFO Miami, FL)

Kevin Smith (WFO Paducah, KY)

Rudolf Kaltenböck (Astrocontrol, Vienna, Austria)

Bill Bunting (WFO Fort Worth, TX)

Chris Buonanno (WFO Little Rock, AR)

Justin Lane (WFO Greenville, SC)

Chris Sohl (WFO Norman, OK)

Pieter Groenemeijer (European Severe Storms Laboratory, Munich, Germany)

Many thanks to everyone, including those we may have inadvertently left off this list. Please let us know if we missed anyone. We can certainly edit this post and include their names later.

The EWP2011 Team

Tags: None

Forecaster Thoughts – Chris Sohl (2011 Week 4)

I think that both operational forecasters and program developers benefit from the opportunity to interact with each that the EWP 2011 program provided. Forecaster participants are introduced to new tools that are becoming available. Not only do they have an opportunity to make a preliminary evaluation of each tool but also to explore how they might be incorporated into an operational setting. It was a plus having folks knowledgeable about the new tools available to answer questions and to suggest possible ways in which forecasters might use the tools. This interaction should result in a better product by the time the new tools are delivered to the entire field.

Some of the datasets explored in EWP 2011 included convective initiation schemes and storm top growth. Based on my initial impressions gained over a period of working only a few days with the data, the UAH CI product seemed to have a greater FAR with CI compared to the UW product which itself seemed to be too conservative. While a high FAR with the UAH product might at first glance seem like a poorer performance, I think it may still provide  useful information (for example, getting a sense on how the cap strength might be evolving).

In the short amount of time that I had to look at the satellite-derived theta-e/moisture fields, I saw enough to keep me interested in spending more time evaluating with these products. The opportunity to discuss possible product display methodologies with Ralph Petersen was helpful.

The 3D-VAR dataset looked very interesting and seems to have potential to provide useful information. There were some issues where the strongest updrafts appeared to be in the trailing part of the storm and it might be interesting to see if that behavior was strictly an artifact of the algorithm or a function of the variability of the updraft strength at various levels in the storm. I would also like to have more opportunity to examine some of the other fields (vorticity, etc.) in several different storms to see if there might be a signal which could provide the forecaster a heads-up regarding what kind of  short-term storm evolution might be expected.

I appreciate that some of the participating organizations continue make much of their data available on-line following the conclusion of the spring experiment. Not only does this help me not forget about the new product some 6 months later, but rather allows me to further explore how I might better include the new datasets  into my shift operations. It is possible that a further review of  a product that initially seemed to have minimal value to me in an operational sense ends up providing more utility than I originally thought.

Chris Sohl (Senior Forecaster, NWS Norman OK – EWP2011 Week 4 Participant)

Tags: None

Week 4 Summary: 6-10 June 2011

EWP2011 PROJECT OVERVIEW:

The National Oceanic and Atmospheric Administration (NOAA) Hazardous Weather Testbed (HWT) in Norman, Oklahoma, is a joint project of the National Weather Service (NWS) and the National Severe Storms Laboratory (NSSL).  The HWT provides a conceptual framework and a physical space to foster collaboration between research and operations to test and evaluate emerging technologies and science for NWS operations.  The Experimental Warning Program (EWP) at the HWT is hosting the 2011 Spring Program (EWP2011).  This is the fifth year for EWP activities in the testbed.  EWP2011 takes place across four weeks (Monday – Friday), from 9 May through 10 June.  There are no operations during Memorial Day week (30 May – 3 June).

EWP2011 is designed to test and evaluate new applications, techniques, and products to support Weather Forecast Office (WFO) severe convective weather warning operations.  There will be three primary projects geared toward WFO applications this spring, 1) evaluation of 3DVAR multi-radar real-time data assimilation fields being developed for the Warn-On-Forecast initiative, 2)  evaluation of multiple CONUS GOES-R convective applications, including pseudo-geostationary lightning mapper products when operations are expected within the Lightning Mapping Array domains (OK, AL, DC, FL), and 3) evaluation of model performance and forecast utility of the OUN WRF when operations are expected in the Southern Plains.

More information is available on the EWP Blog:  https://hwt.nssl.noaa.gov/ewp/internal/blog/

WEEK 4 SUMMARY:

Week #4 of EWP2011 was conducted during the week of 6-10 June and was the final week of the spring experiment.  It was another pretty “average” week for severe weather, certainly paling in comparison to Week #3.  During this week, NSSL and the GOES-R program hosted the following National Weather Service participants:  Bill Bunting (WFO Fort Worth, TX), Chris Buonanno (WFO Little Rock, AR), Justin Lane (WFO Greenville, SC), and Chris Sohl (WFO Norman, OK).  We also hosted special guest Dr. Pieter Groenemeijer, Director of the European Severe Storms Laboratory near Munich, Germany, for several of the days.  Pieter was visiting both sides of the HWT to learn about the process in order to develop a similar testbed for the ESSL in 2012.

The real-time event overview:

7 June: Failure of CI over eastern ND and northern MN; late action on post-frontal storms in central ND.

8 June: Squall line with embedded supercell and bow elements over eastern IA and southern WI.

9 June: Afternoon squall line over southern New England and NY; evening supercells western OK and southern KS.

The following is a collection of comments and thoughts from the Friday debriefing.

NSSL 3D-VAR DATA ASSIMILATION:

One major technical issue was noted but not diagnosed.  It appeared that at times, the analysis grids were offset from the actual storms, so it is possible that there were some larger-than-expected latency issues with the grids.

It was suggested to add a “Height of maximum vertical velocity” product.  However, we hope to have the entire 3D wind field available in AWIPSII.  We also hope to have a model grid volume browser, similar to the radar “All-Tilts” feature within AWIPSII.  We used the WDSSII display for the wind vector displays.  The forecasters noted that the arrows were plotted such that the tail of the arrow was centered on the grid point.  It should be changed to the middle of the arrow.

The vorticity product was the deciding factor on issuing a Tornado Warning for the Thursday storm north of Wichita.

Bad data quality leads to bad 3DVAR.  In particular, it was noted several times that side-lobe contamination in the inflow of storms was giving false updraft strengths.  Improper velocity dealiasing is also detrimental to good 3DVAR analysis.  There is an intensive data quality improvement effort ongoing as part of the WOF project.

The downdraft product occasionally took a maximum “downdraft” at the upper levels of the storm and projected it to the surface.  There’s not a lot of continuity, and it is difficult to discern consistent features associated with the storms.

Would also like to use the products with less classic type storms, like low-topped convection, microbursts, etc.

David Dowell, who was visiting from GSD this week, is working on a next-generation assimilation using Kalman filtering, but requires more CPU power.  Jidong Gao at NSSL might create a blended technique (with Dowell) that requires less CPU power, something like a 3.5DVAR, which uses 3DVAR for hot-start analysis with radar and model analysis background, and then runs a cloud model out 5 minutes based on that and use it for the first guess on the next analysis, and so on.  This means we would be able to get more fields like T and P for cold pools, downdraft intensity and location, for storm types other than supercells.

OUN-WRF:

There were very few opportunities to evaluate OUN WRF data this week.  Our only event within the domain was on Thursday with a late domain switch for evening half of activities within OK and KS, but convection was already on-going and evaluation concentrated on other experimental EWP products.  The model suggested a few more storms that weren’t there.  One of our forecasters who use the data during regular warning operations in their WFO commented that the updraft helicity product helps with predicting storm type, but that it tends to overproduce cold pools and outflow.

GOES-R Nearcast:

The Nearcast principle investigator, Ralph Peterson, was on hand this week.  He posed the following questions to the forecasters:  Did you find the Nearcast products useful to ID the areas likely for convection initiation, and to predict the timing and location in pre-convective atmosphere.

The Nearcast products were primarily used during the early parts of the day to facilitate the area forecast discussion and afternoon/evening warning location decisions.  One forecaster noted that the Nearcast data behind squall lines becomes less useful with time due to intervening cloud cover.

The forecasters were asked if it would be useful to provide extended forecast hours but at the expense of greater data smoothing.  They liked to have the higher-resolution data to as far out as it is useful to have it.

The forecasters were also asked if they would have used the observation/analysis alone without the forward extrapolation, and the answer was that it wouldn’t have been as useful, since it is better to see how the current environment will evolve.

Showed an arch of destabilization between 2200-0300 across the eastern halves of OK and KS… storms formed on the western edge of this gradient and forecaster did not expect the storms to diminish anytime soon and thus increased warning confidence… stronger wording regarding hail/wind potential in warning was issued.

There seemed to be small scale features in the fields, areas of relative maximum that were moving around… would be nice to compare to radar evolution and see how those areas affected the storm structure.

Helped understand why convection occurred and where it would occur… definitely the 1-6 or 1-9 hour timeframe was the most useful aspect of it.

Having a 4-panel set up of the individual layers in addition to the difference field to help increase the understanding of the product.

The color-table in AWIPS was poor… Also, the values were reversed from those in NAWIPS and on the web. The individual layers of PW were also not available in AWIPS.

GOES-R Convective Initiation (UW and UAH):

The forecasters were asked if they compared the two CI products side-by-side.  The UAH product is more liberal in detections, has a higher resolution (1 km), and uses visible satellite data during daytime mode.  The UW product is more conservative in detections, has a 4 km resolution, uses only IR data, and masks output where there is cirrus contamination.

During the daytime, most forecasters were able to spot initiation in the visible satellite data, and thus the CI products were not all that useful for heads-up.  They did mention there could be value during nocturnal events, but the EWP doesn’t operate after dark, so we couldn’t test.

The notion of probabilistic output was once again brought up.  Instead of a product that was “somewhere in the middle” of good detections and false alarms, a probabilistic product could be more useful.  And a comment was made to bring both groups together to product a single probabilistic product.

In some cases, the products failed to trigger on clumps of cumulus that looked similar to other clumps that were receiving detections.

One forecaster raised a concern about consistency with respect to the FAA using the product for air routing.  If the CI product was automated and used by FAA, how would that conflict with human-created TAFs and other products?

A forecaster found that the UW cloud-top cooling rates useful to look for the timing of the next area of developing convection.

Even though CI didn’t always occur… false hits were useful in identifying clouds trying to break the cap.

GOES-R OTTC:

The one day we would have expected a lot of Overshooting Top detections, Thursday over Kansas, there were lots of missed detections.  Otherwise, the forecasters felt that they could ID the overshooting tops well before the algorithms, except perhaps at night (when we don’t operate).  Chris Siewert mentioned that the spatial resolution of current imager is too great (4x4km), and OT detection works better on higher-res data sets.  The temporal refresh rate also affects detection; sometimes feature show up between scans.

GOES-R pGLM:

We only had one half of an event day to view real-time pGLM data, the Thursday evening OK portion of our operations. Some of the storms to the east had higher flash rates, but this was an artifact of the LMA network’s detection efficiencies.  Flash rates would pick up a short time before increases in reflectivity.

One forecaster has access to real-time LMA data in the WFO and had some comments. They get a lot of calls wanting to know about lightning danger for first and last flash and stratiform rain regions.  It is also good for extremely long channel lightning – might get a rogue hit well away from main core, and sometimes anvils well downstream of main core can get electrically active.

There are more GOES-R details on the GOES-R HWT Blog Weekly Summary.

OVERALL COMMENTS:

The challenge, which was good, was integrating that info with all the other data sets, but also on how to set up the workstations, and best practices to use it.  Need six monitors!

Need pre-defined procedures.  Forecasters used the “ultimate CI” procedure heavily and liked to see what we think they should be combining to help enhance the utility of the products. (However, it is not always clear to the PIs which procedures would be best, as the experimental data has not yet been tested in real-time).

Like the two shifts.  Get to experience both types, a nice change.

I sometimes got too tied into warning operations rather than looking at experimental products.  It’s Pavlovian to think about the “issuing warnings” paradigm. (We tried to emphasize that getting the warning out on time wasn’t a priority this year, but using the warning decision making process to determine how best to use the experimental data sets, but “comfort zone” issues inevitably rise up.)

Training would have been better if done prior to visit, using VisitView or Articulate, and spend training day on how to use products rather than coming in cold.

Two weeks is nice, but April-May is a tough time to add another week, or even one or two X shifts for pre-visit training.

I went through most of training on web before visiting, it was abstract.  But once here, went through it again in a different light.

EFP interaction was tough – it was too jammed at the CI desk.  We felt more like an “add-on” rather than an active participant.

The joint EFP/EWP briefings were too long, and covered aspects we didn’t care about.  There were competing goals.  We should have done it in 15 minutes and moved on.  Need microphones for briefing.  Didn’t need to hear hydro part.  Need to set a time guideline at briefing for all groups.  Also, the information being provided was more academic than pure weather discussion.

The HWT needs more chairs.  Also, two separate equal sized rooms would be better than the current layout.

A LOOK AHEAD:

EWP2011 spring experiment operations are now completed.

CONTRIBUTORS:

Greg Stumpf, EWP2011 Operations Coordinator and Week 4 Weekly Coordinator

Chris Siewert, EWP2011 GOES-R Liaison (from the GOES-R blog)

Tags: None

3DVAR: the good and the bad in Wichita, KS

As one might expect the output from the 3DVAR data assimilation is only as good as the radar data that goes into it.  This was seen clearly with tonight’s storms near the Wichita radar (KICT).

Below are displays of both the vorticity (at 2.25km) and vertical velocity (at 8km) at 2345 UTC.  At this time, the storm was just to the north of KICT and was well-sampled by the radar including good data from the inflow region as well.

KICT and 3DVAR derived vorticity
9 June 2011 2345 UTC KICT reflectivity and 3DVAR derived vorticity
KICT and 3dvar vertical velocity
9 June 2011 2345 UTC KICT reflectivity and vertical velocity from 3DVAR data assimilation

However, just 20 min later at 0005 UTC, the 3DVAR vertical velocity field developed a suspicious updraft southeast of the radar and the storm cluster:

KICT reflectivity and 3dvar W
KICT reflectivity and 3DVAR vertical velocity at 0005 UTC on 10 June 2011

The reason for this updraft may actually have been side-lobe contamination due to anvil spread SE over the radar.  Velocity data from KICT shows this area of suspect data — the signal was continuous in both height (through almost all tilts) and time (for about 15 min).  The below image is just one example from this time:

KICT velocity and 3dvar W
KICT velocity from 8 deg elevation and 3DVAR vertical velocity field at 0010 UTC on 10 June 2011

The solution for the velocity data lead to convergence in the area and a false updraft.

Unfortunately, the output from data assimilation is only as good as the radar data brought in and this specific type error is extremely hard for even the best quality control algorithms to remove…

-Kristin Kuhlman (3DVAR scientist week 4)

Tags: None

Update: 2011-06-09 2355 UTC

This was somewhat unexpected – a very well-formed classic supercell has developed just north of ICT.  Here are various images:

3DVAR and KICT reflectivity
Derived reflectivity from 3DVAR and KICT reflectivity with 3DVAR windfield overlaid at 2345 UTC
3DVAR vertical velocity and KICT reflectivity
Vertical Velocity field from 3DVAR data assimilation and KICT reflectivity with 3DVAR windfield overlaid

Tornado reported near Colwich KS!!!

Tags: None

Update: 2011-06-09 2340 UTC

Our early shift participants (Chris #1 and Chris #2) have broken off warnings to fill out their surveys, and Bill and Justin have taken over the NYC WFO (OKX) as the squall line moves over Long Island Sound.  It will soon be offshore, and at 6pm, we are planning to take a dinner break and move our domain to Oklahoma to watch the high-based storms and monitor OUN WRF, PGLM, and 3DVAR products.

The 3DVAR products, particularly the updraft products, seemed to underperform today.  Not worthy of an image on the blog.

Greg Stumpf, EWP2011 Week 4 Coordinator

Tags: None

Update: 2011-06-09 2100 UTC

In other news…the SPC just issued a blue box for western Oklahoma.  Since this is in our LMA domain, and OUN WRF, we may consider moving our operations to there.  More later….

Greg Stumpf, EWP2011 Week 4 Coordinator

Tags: None

Update: 2011-06-09 2150 UTC

Tech woes are slowly resolving.  The 3DVAR issues are fixed for now, although our domain is a bit too small to capture the breadth of the severe weather event.  AWIPS issues on higgins are causing problems with warning issuance.  We discovered that the Boston warning headers are ‘BOX’ and need to be manually changed to ‘BOS’.  In addition, the warning VTEC counters are not incrementing with each new warning, so we’ll need to edit these manually.

So, what is going on weatherwise?  We have a line of severe storms moving through southern New England and southeast New York, with a few stray severe cells out ahead of the line.  And we just got a report of 1.75″ hail from Wappingers Falls, NY, my home town!  The event looks like this right now (Reflectivity at -20C):

Greg Stumpf, EWP2011 Week 4 Cooridinator

Tags: None

Update: 2011-06-09 1950 UTC

We started early today, as severe convection is ongoing in New England and Eastern New York State.  However, our systems decided to finish the experiment a day early.  We are in the process of relocalizing AWIPS and restarting radars and the ingest, as well as fixing some 3DVAR issues right now.  One team is working the Boston CWA, and the other started out as Albany, but have now moved to New York City CWA.  More soon….

Tags: None

2011-06-09: Area Forecast Discussion

Cold front…analyzed at 17z from northern NY…to northwest Ohio will continue to move southeast through the afternoon hours. Convection is currently developing fairly across northwest VT to near Binghamton. Expect continued new convective development along the boundary towards the southwest…as forcing along the boundary quickly overwhelms limited CINH.

Observations of UAH CI products and CIMMS cloud top cooling products has confirmed the rapid nature of convective development.

Objectively analyzed instability fields indicates moderate to extreme surface based energy exists ahead of the cold front…with an axis of 4000+ j/kg from just west of PHL to west of ALB. Forecast soundings ahead of the boundary suggest damaging wind gusts are the primary…and most widespread…severe weather threat.

Convection ahead of cold front is expected to exit New England by 2300z which may require shifting the area of interest farther to the west later in the operations period.

Frontal boundary becomes stationary…and more poorly defined westward across the western Ohio valley….to the central plains…as several areas of convectively induced outflow adds complexity to subjective analysis. Much of the area near the front/outflow boundaries through the Ohio valley will likely continue to be active with strong/severe storms through the afternoon and evening. A series of weak waves moving through the region may result in some modulation in the activity. Interest also points to portions of eastern KS… IA and northern half of MO… depending on exact location of boundaries and the front by late afternoon. Adequate deep layer shear would support a few supercells.

Farther southwest in OK…

The 15 UTC surface analysis shows a 1005 low south of Dodge City, KS,
and a frontal zone extending from near Kansas City to the low center
toward the SW into the Texas panhandle. A 30-35 kt Sly low level jet
is located over W Texas into W Kansas per RUC analysis. A zone of
elevated convection stretches from NE Kansas into N-central and SW
Oklahoma. East of the front, surface dew points are in the mid 60’s
over much of central and E OK, and the upper 50’s and low 60’s in W
Oklahoma. Within the warm air mass steep lapse rates are present in
the 800-500 hPa layer. A thermal axis at 850 hPa more or less
collocated with the zone of elevated convection is associated with a
rather robust cap that will probably require lift in addition to
diurnal heating in order to be broken.

Model guidance suggests that CAPE should rise a little bit to ~2000
J/kg across most of C & E OK, E KS during the day, but there is
consensus among RUC/NAM and GFS that the boundary layer will tend to
dry out during the day, as a result of vertical mixing and northward
advection of drier air, analysed over north-central Texas. A shortwave
trough over E Colorado will move E/NE ward, but most forcing
associated with this feature should pass into Kansas.

GFS, RUC and NAM produce little to no precip over OK until 00 UTC
strong CAP. HRRR has a little bit of precipitation developing around
21 to 22 UTC in several runs.  Yet, a few storms will probably form
over W Oklahoma in vicinity of the front, probably in the late
afternoon or early evening (22-02 UTC). Somewhat higher coverage is
expected across Kansas. Given that 0-6 wind shear increases to about
40 kt, well-organized multicells and perhaps one or two rotating
storms will be possible. The few storms that develop will probably
produce some large hail, and, given the deep dry boundary layer, some
severe wind gusts.

Buonanno/Sohl/Groenemeijer

Tags: None