Week 4 Summary: 18 – 22 May 2009

2009 Spring Season – Week 4 – May 18-22

This was my second opportunity in as many years to serve as an Experimental Warning Program Weekly Coordinator. I would like to thank all of our National Severe Storms Laboratory facilitators and support staff, as well as the project scientists for CASA, PAR, LMA, and MRMS. It is not easy to share two teams of meteorologist evaluators across four projects – some using real-time data in rapidly evolving time windows. All involved approached the week with common goals, however, to pursue as much data as possible and draw upon our four evaluators’ experience and feedback. Everyone is motivated by a clear desire to improve the severe weather warning process.

Our week 4 visitors brought with them a wealth of experience gained from years in the National Weather Service. They were Les Lemon, Research Associate at the NWS Warning Decision Training Branch, Matthew Kramar, Senior Forecaster at NWSFO Sterling, Virginia, Rob Handel, Senior Forecaster at NWSFO Peachtree City/Atlanta, Georgia, and Mike Vescio, Meteorologist-in-Charge at NWSFO Pendleton, Oregon.

It seems (and at least one dataset – significant tornadoes in the NWSFO Norman, OK, forecast area – shows) that a relative minimum in severe weather often occurs in mid May, surrounded by more active periods at the beginning and end of the month. But no one could have anticipated the degree to which this lull would affect the entire continental United States in 2009. We missed an opportunity to study a small area of severe storms that occurred in Mike’s home County Warning Area of northeast Oregon on Monday, instead deciding to push through the seminars and training on our four project areas. We attempted (unsuccessfully) real time operations for the MRMS project in the Intermountain West on Tuesday, and with some rotating storms in Florida on Thursday. Our week was made, however, by a severe thunderstorm event that occurred in western Nebraska on Wednesday, and fell nicely into the 4-9 pm time frame.

We spent the rest of our time viewing archive datasets across CASA, PAR, and LMA. This set of forecasters impressed me with their attention to base data and their ability to place much of their conversation in the context of proven methodologies and peer-reviewed research. Many of their recommendations for display methods and product definitions went into the real time blog entries. Still more of their feedback went directly to the project scientists during archive case studies. Greg Stumpf and I tried to pull more of that feedback out of them during the weekly debrief on Friday, May 22nd. A summary of that discussion is included below.

CASA Wind Prediction Exercise

Background: In addition to archive events in which forecasters simulate warning operations, CASA scientists are conducting an experiment in which forecasters – some given CASA and WSR-88D data and some given only WSR-88D data – are asked to estimate surface wind speeds. Oklahoma Mesonet data are used to verify the predictions.


We Need larger data window in time…leading up to the time at which we are asked to make a prediction (trends important)


There were numerous amorphous looking cells. I was looking for tighter refl gradients/more vigorous storms for wind.


Looking for outflow boundaries/cold pools to undercut and decrease the wind potential


Knocking down wind speeds based on 88D experience. Always seems to run high. CASA was even higher.


Should we warn for possible gusts, or an organized event?


With WSR-88D time scales, we are usually waiting for data… now it is always coming in very quickly with CASA. Do we really need the sector scan strategy? We like to know what to expect and to be comfortable knowing we will see the full picture with each scan


Willing to give up faster updates to get the full sector scans.

CASA Scientist Don Rule

The trade-off to eliminating sector scans is poorer performance of 3D VAR


Don’t put much effort into getting volume scans less than 1 minute. That should be sufficient.


At what point is there diminishing return?


Live! I want the Data to eventually be a live movie loop…refresh every second or less.


Lots of transient features, many are false


But some are real! Draper Lake tornado on 5/13/09, only had one 40 second volume scan of a TVS!

Note: Out of 27 forecasters who’ve gone through the experiment, best performer’s average error was + 5.5 knots. Worst performer + 17 kts (met grad student w/o operational experience).

Generally, those given CASA data would warn while those given only 88D data would not warn

When there are a lot of small radars, chances of getting the right viewing angle for radial velocity is much greater.

Other CASA Discussion

5/13/09 Case: Not just a tornado case, but a good sig wind case (RFD).

Two chasers who were in middle of Anadarko came in to look at data. Estimated at 120 mph (Greg: 2/3!), and a second surge. Tornado went to their SE. Since OEC power grid went down, sirens didn’t work. After generator, sirens went off 10 minutes late.

Multi-Doppler 3DVAR, compared to mesonet winds, very useful.

Classic case of occlusion, new meso did not produce tornado.

120 mph est., Doppler got 130mph (60-70 m.s).

When we see this amount of data, visualization capabilities become more critical, decision support. Data refreshes when you come off a data source, and the picture is different.


same for all the rapid-refresh data sources.


Display is hindering, not the immense quantity of data. Big take-away!

The eventual concept is that the user can add input to the MC&C to “override” some of the automation.


Scale issues – looking at 88D data, sometime you don’t look at the bigger picture (mesoscale). Same with CASA/PAR – may lose track of “88D scale” (storm scale). May take complete retooling of warning ops if we go to data at these scales.  From an operational perspective there is only so much you can process in real-time.


Need some algorithms to process some of the data, but still need person in the loop.


Need to study the precursors and their differences between false and true signatures.

Lightning Mapping Array Discussion

Forecasters looked at 5/15/09 squall line.

First time they looked at 3D visualization of LMA data. Dots. Isosurfaces.


2D VILMA product seems most operationally practical. This could help you prioritize storms. 3D is “cool” though.


Like the 3D information. I am used to 3D with GR Analyst.

Need a team of two. What can I see in 3D that I can’t get out of VILMA 2D product? Combined isosurfaces of dBZ and LMA density (Patrick’s Note: Not sure if this was said because Matt did this, or If he would like to see that capability).


We need to learn what the lightning signatures are, and how they relate to severe reports. Les found from this one case May 15, 2009 – that lightning max density matched the location of strongest winds


Training is also needed.


Trends were very valuable for LMA data using WDSSII.

Jim Wilson

Trends only work if you couple the information with knowledge of the NSE.

Early trends broke more often. WDSSII version is more robust.

GLM proxy: Need to use it without the ground-based data to do a fair assessment.

Phased Array Radar (PAR) Discussion

Early May 2009 Storm… Isolated Supercell at Foss Reservoir (long range from 88Ds and PAR):

One group compared PAR to KTLX, one compared PAR to KFDR. The latter group gravitated toward the KFDR data due to the better spatial resolution, traded-off with the better time resolution.


Very good clarity of data… spatially and temporally.

Oversampling… we like that.

Reflectivity seemed much lower on PAR than 88D.

That has a big effect on warning ops and Could make a difference in precip rates.

(Note: This is a known problem… vertical polarized and not as sensitive. Next step is to make PAR dual pol to address this problem.)


In what events/environments is the PAR best suited?


Events near the radar


Shorter Lifespan storms… rapid evolution

Or smaller scale features within a long lived storm


Was the vertical detail sufficient?


For TS Erin case, all we needed was low level data. Give us even faster updates at low levels. Give us sub 0.5 degree tilts

5/13/09 OKC (Draper Lake) storm:


In NWSFO Warning Operations… The initial Tornado Warning for Oklahoma/Cleveland Counties was based on fact that TDWR signature of convergent rotational RFD was minimally undercut by outflow. Then TVS developed.

8/19/07 T. S. Erin mini supercells:


Without PAR, wouldn’t have issued as many warnings. Felt we overwarned with the PAR data.

7/16/06 downburst case done by two of the four forecasters.


Mostly looking at rapid-evolution of descending cores.

Multi Radar / Multi Sensor Discussion

Real Time Data at http://wgserver.nssl.noaa.gov


What other products would you like to see?


User-definable interface for height of dbz products

Choose your dbz value and temperature level, etc.

Excerpts from the live blog during our MRMS real-time IOP on May 20:

Matthew and Mike note that for MESH the associated color tables are slightly different between Google Earth and AWIPS.  These need to be the same to avoid confusion about product times, and help the forecasters to “trust” the data on both platforms.

Rob has been looking at the Legacy Hail Algorithm on AWIPS to compare with the experimental products.  He discovered that the locally run Legacy Algorithm is not synced to updated environmental data.

The radar presentation had become quite impressive around 8 pm CDT. MESH indicated 2.88 inch diameter hail.  Les argued for baseball size hail in the warning.  Arthur and Patrick argued for something closer to 2 inch, based on reports (phone calls from the HWT and Local Storm Reports from CYS and LBF) throughout the evening that have been consistently lower than the MESH values. The reports have also indicated very heavy rain.  The storm does not have a very impressive mid level mesocyclone, and Les agreed that was a good reason to undercut MESH in the warning. Rob issued the warning, mentioning golf balls. The largest hail reported that day in this low population area was 1 inch.

All of this week’s forecasters advocate intense use of base data in warning decision making.  Mike says “We are overloaded with derived radar products. Just give me the base data, and a better understanding of the environment (to improve warning decision making).”  Others agree that environmental data is key to warning decision making, but they also point out that there may be some tasks that algorithms can perform very well, thus freeing the human forecaster to tackle other problems.  The group also discussed certain scenarios in which algorithms can aide the forecaster during low staffing or broad geographical outbreaks.

Les and Rob particularly like the LMA and lightning trend data, as it is “Base data of a different kind.”  The forecasters, as a whole, liked the trend graphs for all types of data.  That was the most useful function of the data in Google Earth.

The group noted several areas, however, where the Google Earth data needs improving.  There is no time stamp, images are smoothed, and there is no cursor readout.  They would like the color tables to be synced to the respective products with which they were intended to be used, rather than requiring the forecaster to click on a product and then click on an appropriate color curve.  In its current state, forecasters feel the images in Google Earth are better suited to verification efforts by mapping MESH and rotation tracks to GIS data, than they are suited to warning decision making.

HWT Feedback


You are trying to run AWIPS because that’s what we have in the field, but the computational hurdles associated with AWIPS hindered real-time IOPs this week. Do we need that in this experimental environment?


Do you like the simulated warning ops, or the free-style discussion?


We like both. Mike, though, prefers warning IOPs


Good to do warning ops and then go back through it in a post-mortem

Google Earth:

Will there be more functionality in Google Earth? Add time of image on screen? Cursor readout?

Can do some stuff with the w2 GE plugin. Queries the server to get a data readout. Can draw a line to build a cross-section. Areal alarms if a parameter reaches a threshold.


Data Issues and Recommendations:

Need Near Storm Environment grids in Google Earth (and AWIPS2).

Consider the color curves so they match in all places – (AWIPS, WDSSII, GE).

Would like some sharp cutoffs at some of the values (catch forecaster’s attention rather than requiring them to sample the data all the time while juggling multiple data sets).

Recommend 60 dBZ, 62 dBZ Echo Tops

User-definable on-demand queriable interface for products. Should be a requirement before MRMS concept is introduced in AWIPS2.

Lot of the products geared toward up draft strength and hail.

Convergence/divergence products for wind, precursors for tornadoes too?

Need lots of 3D.

Need better interface to choose which radar to look at for a multi-radar display.

AWIPS was problematic this week, did we need it? Could we issue warnings in WDSSII?

Future HWT Operations:

Expand the training. Too much information to cram into the Monday that people arrive. Split the training across multiple days.

Were four projects too much stressed our time?

Most of the value is in the discussion…regardless of number of cases accomplished. (e.g., GOES-R – had a 1 hour discussion with them. Should be more next year.)

Need more than one week. Is 2-week stint ok? Individual forecasters would get to perform more case studies and IOPs. Overlap incoming with outgoing forecasters each week. What kind of shifts? If they are in town for two weeks, could people cover some night and weekend events? But we’d need more staff!

Random “out of season” HWT idea: Use collaboration tools, and run an IOP virtually using volunteer X-shift forecasters? That way, we can get more forecasters looking at products in real-time. How could we do this?

Approach to Case Studies:

Lot of cases were the same from the different platforms. How did that affect the investigation?

Didn’t hurt unless you’re not supposed to know what happened.

Amount of Near Storm Environment data available to forecasters prior to looking at archive data was increased in 2009… yet this set of forecasters wants MORE! Near Storm Environment Data.

Need something like SPC mesoanalysis grids in WDSSII, AWIPS, during spin-up.

Weather and Society * Integrated Studies (WAS*IS):

On May 21, CASA invited McClain County, OK, Emergency Manager, Ed Craven, to the HWT. He was very excited at the opportunity to participate and view data. Later the same day, Two chasers, one an emergency manager and one a former television meteorologist and aspiring emergency manager, who had a close encounter with a nighttime tornado that occurred in the CASA network, also visited the HWT to view data and provide ground truth.

Miscellaneous Comments:

Nice to have the interaction between the researchers and forecasters.

Lots of kudos were given to our weekly coordinator.

Patrick Burke (EWP Weekly Coordinator, 18-22 May 2009)

Tags: None