'Hurricane' Schwartz: Why European computer model is best

(First of two parts on computer modeling and the weather)

When I was starting out as a meteorologist back in the early '70s, forecasts were less accurate and it took 22 minutes out of every hour just to plot and analyze a weather map by hand.

Now it takes one click.

Back then, we got two grainy satellite pictures each day. Now we can see "rapid scan" movies with pictures 1 minute apart. And, in the old days, we got to see only one computer model to help us make the forecast. It was called the LFM. I used to call it the LUM-the Limited Usefulness Model.
Now there are numerous computer models from all over the world, and they are thousands of times better than the LFM.

What is a computer model?

The atmosphere is made up of air. And the air moves, a lot like a river (except we can't see it). The more we know about those rivers, the better we can predict how they'll act in the future. Imagine each little particle of air, not only across the whole globe, but also miles up into the atmosphere. Unless we know about each of those trillion gadzillion particles*, we can't perfectly simulate the current conditions of the world. And if we can't do that, there's no way any computer can make a perfect forecast. Ever.

There are rules the atmosphere must obey. They're called the "Laws of Motion" (discovered by Isaac Newton in 1687). And there are mathematical equations that explain how the atmosphere works. It's all very complicated, which is why "real" meteorologists have to know at least some calculus and physics.

A child can build a "model" of actual pieces of something. Scientists build "models" of the atmosphere. Since even the fastest computers can't simulate every particle on earth, we use a "grid" system to make things simpler. Early, slow computers needed those grid points to be hundreds of miles apart. Each computer model would use different methods to calculate how the weather at those grid points would change in the future.

As computer power increased, the number of grid points could increase. Now they might be separated by 10 miles instead of hundreds, and some even get down to less than a mile! The closer the grid points are, the more accurate the forecast will be.

What makes the European model so good?

Computer models for operational use have been around for more than 40 years. There has been increased competition for the crown of most useful model (remember, none is perfect - some are just more useful than others). It takes some of the fastest and most expensive computers to do all the trillion gadzillion calculations need to make those grid points closer together. The European Center for Medium Range Weather Forecasting (ECMWF) has been the leader for more than 20 years. The U.S. models have been getting much better. But so has the "European Model," or "EURO" - it has kept hold of its lead.

The simplest advantage the EURO has is that it "initializes" the atmosphere better. There are debates about the reason why, but not about this basic fact. If we use the EURO to initialize, the U.S. models do much better. If you are interested in some provocative, controversial theories about why the EURO is better, I suggest you search for blogs from Cliff Mass, a Professor of Atmospheric Sciences at the University of Washington.

Every computer model starts off with an "initialization."  Data from ground stations all over the world,  ocean buoys, and satellite measurements of air, ground, and ocean are all included. It's all thrown into the computer model as a starting point. The purpose: make as perfect a start as possible. The old computer saying says: "Garbage in; garbage out". If the initialization isn't good, that computer model has no hope of being the best.

Just because the European model has the better initialization doesn't mean it's going to be right all the time. Yes, the EURO did a much better job than the main U.S. model (GFS) for Sandy. It was predicting a New Jersey hit days ahead of time, while the GFS had Sandy moving out to sea. But the EURO was way off in our non-snowstorm in late January. It was indeed suggesting amounts around 30 inches in Philadelphia, while the GFS had only a few inches. The overall superiority of the EURO backfired that time.

*I made that number up. There is no number big enough.

Later this week: how forecasters use computer models

Glenn "Hurricane" Schwartz
Chief Meteorologist
NBC10 Philadelphia