# 3282
Imagery courtesy of Timothy C. Germann, Kai Kadau, Catherine A. Macken (Los Alamos National Laboratory); Ira M. Longini Jr. (Emory University)]
A computer model of a the spread of a pandemic virus from the Los Alamos National Laboratory.
The old saying (well, not that old, as it is attributed to George E. P. Box, Professor Emeritus of Statistics at the University of Wisconsin) is that:
“All models are wrong, but some models are useful.”
We use models to try to mathematically simulate real-life events, with probably the most commonly used model coming from weather forecasting.
And we know that despite the use of multi-million dollar Cray Supercomputers, inputting hundreds of thousands of daily data points, and decades of tweaking algorithms, that weather forecasting is still an inexact science.
Models are never going to be perfect. But the more times that we can compare a model’s output to a real life event, the better (in theory) we can make future models.
Over the past 30 years we’ve seen major improvements in the National Hurricane Center’s ability to predict landfall of a tropical system. Since they get a dozen or so storms to track and study each year, their models (and there are several in use) get better with each season.
Models that attempt to simulate events further out in time, or which are based on rare events without much historical data, become much more difficult to devise.
One such difficult scenario to model is the spread of a novel infectious disease.
According to this article ihealthbeat.org (hat tip Shiloh on Flutrackers), two statistical models that attempted to quantify the spread of the H1N1 Swine Flu virus fell short of the mark, although the knowledge currently being gained by watching H1N1 should allow for refinements in these, and other models over time.
Computer Model Predictions of H1N1 Flu Spread Way Off
Two rival supercomputer teams significantly underestimated the spread of H1N1 influenza when they predicted there would be only 2,000 to 2,500 cases in the United States by the end of May, the New York Times reports.
On May 15, CDC estimated that there were "upwards of 100,000" U.S. cases, although only 7,415 had been confirmed at that point. CDC has not yet updated its estimate.
Tim Germann -- a computational scientist who worked on a 2006 flu forecast model at Los Alamos National Laboratory -- said he believed there are now "a few hundred thousand" cases (McNeil, New York Times, 6/2).
Computer Model Predictions
An Indiana University team used air traffic and commuter traffic patterns for the entire U.S. to make its projections. Meanwhile, a team at Northwestern University used a Web site -- called Where's George? -- that tracks how far and how fast $1 bills travel to make its predictions (iHealthBeat, 5/4).
Dirk Brockmann, an engineering professor who led the Northwestern team, said, "The number of reported cases in Mexico that we plugged in at the beginning of our model were orders of magnitude lower."
He added the computer model still was fairly accurate in predicting the geographical spread of the H1N1 flu and identifying California, Texas, Illinois and Florida as hot spots.
