U.S. Climate Versus Weather Computers: Climate Wins

Why is the U.S. government provide hugely more computer resources for climate prediction than weather prediction?  And why is far more emphasis given to climate prediction research than weather prediction research and development?

In some past blogs I talked about the unfortunate lack of computer resources available to the National Weather Service (NWS), resulting in the U.S. trailing behind many international numerical weather prediction centers.  This lack of computer resources undermines the ability of the NWS to run high-resolution weather models, to move effectively into probabilistic weather prediction, to enhance hurricane prediction, and much, much more.  Far smaller nations, with much more benign weather, such as England and South Korea possess weather supercomputer facilities that dwarf ours.
The cost to the American people  of  inferior weather computer resources is substantial, both economically and in saving lives. 

While  U.S. operational weather prediction is provided inadequate computer resources, climate prediction, including studies of potential human-forced global warming, ienjoys the availability of huge supercomputers, with capacities hundreds of times larger than that available for weather prediction.

It is not a little ironic that great emphasis is placed on acquiring state-of-the-art petaflop supercomputers for climate change, while in a year of a dozen billion-dollar weather disasters, the NWS is not given critical tools needed to protect the American people.   Someone has their priorities wrong.

Let me give you a few examples.    As noted in an earlier blog, the National Weather Service operational computer system has 4992 processors for a total computational capacity of .07 petaflops (a petaflop is one quadrillion floating point operations per second).   Keep this .07 petaflop number in mind.

The new 1.1 petaflop GAIA computer just acquired by NOAA

Now consider a few of the climate computers purchased by the U.S. government and associated agencies.

Lets begin with the 1.1 petaflop GAIA computer recently acquired by NOAA, a machine sixteen times more capable than the NWS weather prediction computer.  This machine will be dedicated to climate research (see an article here on this machine).

The National Center for Atmospheric Research, an entity mainly funded by the National Science Foundation, is now completing a 70 million dollar facility that includes the new Yellowstone computer, capable of 1.5 petaflops.    This machine will be used mainly for climate research (article here).  A great irony is that this machine uses HUGE amounts of power, power that will come from coal-fired power plants that emit lots of CO2.

NASA Pleiades supercomputer

Or consider the new "Discover" supercomputer acquired by NASA's Center for Climate Simulation (NCCS) that has .32 petaflop capacity, nearly 5 times that of the NWS forecast machine.  The Discover machine is dwarfed by NASA's new "Pleiades" supercomputer (1.3 petaflops) that is being used for climate simulations and other NASA needs.

I could go on and on naming other supercomputers owned by the U.S. Department of Energy and others that are used for climate research--it would be a very long list.  The bottom line is that the computational power available for climate simulations, for understanding and predicting climate change over the next few decades to a century, absolutely dwarfs what is available for predicting the weather and for understanding how weather systems work.  

This makes no sense.

Imagine if one of the petaflop machines was made available for weather prediction.  The forecast skill of the U.S. global weather models could be substantially increased--providing skillful forecasts further out in time.  We could run models with enough resolution to get the fine scale structures of hurricanes and other storms.   A new age of probabilistic weather prediction could begin, with high-resolution ensembles providing uncertainty information for local weather features.  The impact would be huge, saving hundreds of millions or billions of dollars in economic impacts from severe weather, protecting lives, and improving the functioning of our air traffic control and highway systems.   Real and profound benefits.

Now don't get me wrong.  Understanding and modeling climate change is important.  But there are dozens of supercomputers in the U.S. that are quite capable of this task--and remember  that climate simulations don't have to be done within a set schedule like weather predictions.    And there are many groups around the world doing the same type of global climate simulations--and quite frankly all the better models get essentially the same results.  There is a vast overkill in pushing computer resources for climate prediction, while weather prediction is a very poor cousin.  And consider the fact that with all the supercomputers available for climate prediction, the uncertainty of the predictions for the next century has remained essentially unchanged.  The reason...adding more physics and interactions in the models adds uncertainty since the problem is getting more and more complex.

Why is this gross imbalance happening?   That is something I will leave to the comment section of this blog.  But it is clear that leadership in NOAA, the Department of Commerce, and in other Federal agencies have let this go on too long, to the detriment of the American people. Our congressional representatives and others need to intervene.  The U.S. Office of Management and Budget (OMB) needs to evaluate this situation more carefully.

And don't forget that improved weather prediction is critical for dealing with climate change.

Mankind is doing very little to stop anthropogenic global warming--we are going to do the experiment.   We are ALREADY doing the experiment.  Thus, adaptation will be critical and what is more important for adaptation than improved weather forecasts?  If climate extremes will increase under global warming we need to be able to predict them in the short-term to protect people and property.   Furthermore, there is no better way to improve climate models than to improve weather models, since essentially they are the same.  You learn about model weaknesses from daily forecast errors in a way you can't do in climate predictions.

Climate Computer Support

Weather Computer Support

We can do BOTH climate and weather prediction, but more balance in resource allocation is needed.

PS:  Probcast, the UW high-tech probabilistic prediction system (www.probcast.com), is back up!  The software is now on my department servers, so it should be far more dependable.

PS:: My lost dog was spotted in Mountlake Terrace (the person was pretty sure about this)-- near the intersection of 238th place SW and 52nd Ave.  If you live around, let me know if you see her!  (see right panel for more information)