Monday, September 10, 2012

The Need for Speed

Like Maverik and Goose in Top Gun, meteorologists feel the need for speed.

The political conventions are over and predictably, both major-party candidates talked a good game about innovation, education, and the economy.  Well, I have something that either of them could do to immediately help in these areas: Infuse the National Centers for Environmental Prediction with the funding needed to put the United States back on top again in numerical weather prediction (NWP).

NWP involves the use of mathematic models of the atmosphere and other Earth-system components to predict the weather (this is why meteorologists call them models).  First envisioned by Vilhelm Bjerknes, a Norwegian physicist and meteorologist in the early 20th century, NWP became a reality in 1950 when a group formed by John von Neuman and led by Jule Charney produced the first one-day forecast on ENIAC, the first electronic computer (see Platzman 1979 for a historical review).

From left to right, Harry Wexler, John von Neuman, M. H. Frankel,
Jerome Namias, John Freeman, Ragnar Fjortoft, Francis Reichelderfer and
Jule Charney in front of ENIAC in 1950. © MIT Museum. Source:
In 1954, the US formed the Joint Numerical Weather Prediction Unit, which began producing twice daily NWP predictions on an IBM 701 in 1955.  Since then, as computer power has increased, atmospheric scientists have refined their mathematical models and forecasts have improved markedly.  In 2006, the 72-hour NWP forecast produced by the National Centers for Environmental Prediction is as good as the 36-hour forecast was around 1985.  This relentless improvement has continued in the past few years as well.

Source: Harper et al. (2007)
Evident in the graph above are regular supercomputer upgrades that, along with advances in model formulations and observational technologies, have contributed to this advancement.  The first supercomputer that I used in graduate school (circa 1990) was a Cray Y-MP, which had a speed of 2.3 gigaflops (think of this as 2.3 billion calculations per second).  The iPad2 clocks in at about 1.5 gigaflops, so you have about that much power in your hand today.

Currently, the National Centers for Environmental Prediction has two IBM Power 575 compute clusters (as is the case at most forecast centers, one is for operational use, the other for backups and development), each with 4,992 cores and a peak output of about 73 teraflops (~73 trillion calculations per second).  These computers are a few years old, however, and are being replaced by two newer IBM compute clusters that are each expected to be rated at 149 teraflops and will become operational next summer.

For comparison, Environment Canada has two IBM Power 775 computer clusters, each with 8,192 cores and a peak output of 185 teraflops.  The European Center for Medium Range Weather Forecasting (ECMWF) has two IBM Power 575 computer clusters, each with 8,320 cores and a peak output of 116 teraflops.

However, what really caught my attention is the latest (June 2012) list of the world's 500 fastest supercomputers at  In that ranking the ECMWF has two supercomputers listed, each with 24,576 cores and a maximum output of 549 teraflops.  I haven't been able to ascertain if these are their future operational computers, but if they are, they will be taking a quantum leap forward in operational weather prediction.

This is the kind of investment that the US can and should be making in operational NWP.  Recent studies indicate that variations in the US economy that are attributable to weather variability amount to 3.4%  of the gross domestic product (Lazo et al. 2011).  That's $485 billion.  Weather forecasts are used for everything from commodities trading to picnic planning.  Evacuating coastline for hurricane landfall costs $1 million per mile.  Better forecasts add up to real money.  They help protect lives and property.  They help container ships that transport hundreds of millions of dollars of goods cross the high seas as safely and quickly as possible.  They help you plan for your morning commute.  Simply put, ensuring that the US is at the cutting edge for computer horsepower and NWP modeling is a good investment for the nation.

Addendum@2:50 PM:

Bob Maddox informs me that Kristine Harper's book Weather by the Numbers provides a detailed history of the early NWP efforts and says that Carl Rossby, who is not mentioned above, played a pivotal role in assembling the group that ultimately built the first successful NWP model.  Further, the ENIAC they used was at Aberdeen Proving Grounds.  

1 comment: