National Weather Service testing new computer system

[Eric Friedebach]

By Bryon Okada, Star-Telegram Staff Writer, Apr. 19, 2005

Any North Texan can testify: A single cloud could turn into a raging thunderstorm in a matter of minutes.

Trying to predict where a thunderstorm will form an hour later, and how strong it will be, is a tall order even for veteran weather experts.

With the peak of thunderstorm season here, the National Weather Service in Fort Worth is testing a prototype computer system designed to increase the warning time before a storm forms.

"Even an hour in advance is useful," said Bill Bunting, meteorologist-in-charge. Getting severe-weather information out faster can keep residents, who are more likely to be outdoors this time of year, away from dangerous weather.

The Auto-nowcaster, developed by the National Center for Atmospheric Research, automatically compiles and processes more data than a human forecaster could to study the creation, life and eventual decay of thunderstorms

http://www.dfw.com/mld/startelegram/news/l...al/11431915.htm

 

[Nic Wilson]

I'm doing my master's research on this project so if anyone has any questions feel free to ask me. To avoid the registration process I have posted a copy of the article on my web server.

http://weather.ou.edu/~nwilson/anc-article.doc


More specifically, I will be incorporating the "in-cloud lightning" data it mentions into NCAR's software package known as the Auto-nowcaster. The "in-cloud lightning" at Dallas-Ft. Worth is identified using Vaisala's LDAR II, which is a type of lightning mapping array (More info on LMA's at my website... http://weather.ou.edu/~nwilson/lma.html ).

The Auto-nowcaster currently inputs radar derived products, satellite data, etc., but has yet to use the cutting edge total lightning information. Dan Breed from NCAR and I are hoping that the strong correlations between flash rates and storm growth/decay will further improve the Auto-nowcaster into a viable technology that will benefit forecasters across the world. Vaisala is my gracious sponsor of the research and they are hoping to find further applications for its up and coming LDAR II system, which is an adaptation of research systems currently in place at KSC in Florida, Central Oklahoma, Langmuir in New Mexico and Huntsville, AL.

The Auto-nowcaster has lacked funding the past few years but in previous experiments at Sterling, VA WFO, the 2000 Olympics in Sydney and White Sands, NM it performed very well. The DFW Metroplex will benefit greatly from its implementation there (funded by the FAA). It is not meant to replace forecasters, but is a very complex software package that even identifies boundary convergence and integrates multiple data sets that are not feasible for forecasters to look at when dealing with time constraints.

[/promotion for my research haha]

 

[jaybarnsmith]

I find this very interesting and I am wondering if you can describe the algorithm used for these predicitions a bit more?

What factors are taken into account? How precise is it? Hour? Half Hour? 10 minutes?

 

[Nic Wilson]

I find this very interesting and I am wondering if you can describe the algorithm used for these predicitions a bit more?

What factors are taken into account? How precise is it? Hour? Half Hour? 10 minutes?

The Auto-nowcaster (ANC) software is based on a fuzzy logic raster grid. It initially runs algorithms on the data sets (includes radar, satellite, sounding, mesonet, wind profilers and eventually total lightning) and creates prediction fields. Membership functions (determined by physical insight of the processes observed in the data sets) and weights of importance are applied to each data set input and combined to create a dimensionless likelihood field from -1 to 1, where a positive number indicates an area of increasing likelihood for storms, while a negative number denotes a decreasing likelihood. From this a threshold is established (usually > 0.75) that corresponds to predicted > 35 dBZ reflectivity (the minimum reflectivity typically associated with convection).

The ANC updates every 5 minutes to be in sync with the volume scan period of the WSR-88D's. It outputs predictions of convection for 15, 30, 45 and 60 minutes into the future. Eventually the goal is to run it out to two hours, but this may not be possible without the integration of a short-term numerical model since the feedback of storms not yet initiated can't be taken into account in the ANC.

The ANC falls into an area of meteorology gaining in interest as of late known as nowcasting. In the past this has merely taken radar observed storms and predicted their future position based on its motion (SCIT: storm cell identifcation and tracking). What makes the ANC unique is it performs better with forecaster interaction. Forecasters can draw in boundaries such as dry lines, outflow, etc. to enhance the ANC's performance since these are taken into account in the algorithms. The weights of the data set inputs can be adjusted for the specific geographic region it is in as well to provide better performance.

The last WFO in Sterling, VA that had this version of the ANC improved its forecasting skills scores significantly, so it will be interesting to see what occurs in Ft. Worth once the forecasters become comfortable using it.

I am going to be visiting Ft. Worth in a couple weeks to learn more of the set-up down there, but here is a chart taken from Mueller et al.'s 2003 paper in Wea. and Forecasting that describes the inputs used at the White Sands, NM when it was in use back in 2002.

http://weather.ou.edu/~nwilson/ANC%20Predi...or%20Fields.gif
Typical Input Fields for the Auto-nowcaster

 

[jaybarnsmith]

Awesome thanks for the information.