[geeks] Global warming, was Mr Bill?

[Dr Robert Pasken]

Valid question, My wife says I spend too much time with weather/computer 
ubergeeks to be understood by the ordinary geek. We have two sets of 
clusters; one is the production cluster designed to produce the products 
we need now. The other is the development cluster where all the software
development and testing is done. The development cluster is built of the
cast off parts from the production cluster. Since the microscale model is 
still being verified against observations it cann't run on the production 
machines. Two sets of test cases are being run. One is a hurricane 
forecasts where I collected the verification data via dropsondes 
dropped from the NASA DC-8. In this case a 4 day forecast takes about 3 days
with the 1000 meter resolution forecast domain extending from western Africa
to the mid-Atlantic. Closer to home a 6 hour 333 meter resolution forecast 
covering a mid-western city take about 4 hours. The single test on the 
production cluster took a little less than 1 hour to generate the 6 hour 
forecast.

On Sun, 21 Sep 2008, Lionel Peterson wrote:

>   Dr. Pasken,
>
>   You said "Even the more complex problem of forecasting on very small
>   scales (300 meter resolution) is working in slightly better than real
>   time."
>
>   Could you explain what you mean by "slightly better than real time"?
>
>   I don't doubt what you say, I can't figure out what you mean - as I
>   look at it, that means that if you wanted to predict what the weather
>   was going to be like in a certain neighborhood tomorrow at say 3:00 PM
>   local time, it would take until almost 3:00 PM local time to figure it
>   out, giving you a "prediction" of what the weather was going to be
>   like, say, an hour or two before it happens?
>
>   Again, I'm confused by the statement, not questioning it.
>
>   Thanks,
>
>   Lionel
>   Sep 20, 2008 11:17:18 PM, geeks at sunhelp.org wrote:
>
>     As a meteorologist specializing in numerical weather prediction
>     models I
>     can assure you that not only do climate models exist, but they work
>     very
>     well. Even the more complex problem of forecasting on very small
>     scales
>     (300 meter resolution) is working in slightly better than real
>     time.
>     Current generation microscale models are undergoing field
>     verification
>     tests. There are several DOE/DOD/DHS sponsored micro scale
>     observing
>     networks that are being used to verify the numerical forecasts from
>     meso/micro scale forecast models. There are alot of reasons why
>     this
>     is important. Think of the problem of a gas/particulate whose
>     dispersion
>     and chemical composition is dependent of the temperature, humidity
>     and winds.
>     In this case not only do I need to worry about the meteorology, but
>     the
>     transport, diffusion, dispersion and deposition as well. Current
>     generation meso/micro scale models can accurately predict the
>     meteorology
>     and dispersion/transport processes in better than real time.
>     Although
>     climate prediction is complex, it doesn't need to be concerned with
>     things
>     like turbulence around buildings which is ill-posed mathematically
>     (the
>     turbulence closure problem).
>     Most climate models don't start with "smoothed values" rather they
>     start
>     with the observed conditions at specified time in the past. The
>     typical
>     initial conditions for a short run (75 years) of a climate model is
>     January 1, 1950. The model is run forward for 50 years and then
>     compared
>     against conditions on January 1, 2000. Another simple alternative
>     is to
>     take current observed conditions and run the model backwards to a a
>     point
>     in the past and compare model "forecast" conditions against
>     observed
>     conditions.
>     There are two reasons why supercomputers and now Beowulf clusters
>     exist.
>     One is bomb testing and the other is weather/climate prediction.
>     _______________________________________________
>     GEEKS: http://www.sunhelp.org/mailman/listinfo/geeks
> _______________________________________________
> GEEKS:  http://www.sunhelp.org/mailman/listinfo/geeks