Better Tool for Floodcasting

Dean Braatz knows what it's like to get sandbagged by spring flooding.

Mr. Braatz, who oversees federal flood forecasts for nine Midwestern states from his office outside Minneapolis, recalls how he and his colleagues underestimated last spring's flood in Grand Forks, N.D. The forecasting tools "were pretty capable," he says. But when his office had to predict how high the river would crest in certain areas, the river's quirks "ate our lunch."

For years, local emergency officials have had to rely on a single estimate from flood forecasters. When the forecasters said the river would crest at 32 feet, for example, that would be the final word, and plans to order sandbags or evacuate residents would be based on that estimate.

But today, a new system that is gradually being implemented throughout the Midwest will help forecasters provide local officials with a more thorough range of scenarios, giving emergency personnel the ability to make more informed decisions. And 13 months after the Grand Forks disaster, the system is giving some Midwesterners a more options for defense during the spring flooding season.

Based on comprehensive computer models that map out the details of river basins, the system helps forecasters predict a crest - like the old system - but it also gives them a sense of how accurate that prediction will be. Now, forecasters will also be able to say that there's a 30 percent chance that the river will exceed the predicted crest by 3 feet, for instance. And these scenarios are forecast as far out as three months, rather than a few days.

Currently, the system is only in place for the Des Moines River Basin in Iowa, and the preparations are under way in Grand Forks. National Weather Service (NWS) officials say that once the system is installed in the 13 river forecast offices nationwide, it will save the US at least $600 million a year through more timely flood-response, clean-up, and water-use planning

Pieces of the new system have been under development for about 15 years, according to Lee Larson, chief of the weather service's Hydrological Research Laboratory in Silver Spring, Md. The issue received increased focus following the Great Flood of 1993, which turned large parts of the Midwest into standing lakes. The NWS came under fire from Congress over shortcomings in the flood-forecast efforts.

"We'd put out a number, one number, and all the key decisions were made on that number," recalls John Feldt, who served in the Des Moines weather forecast office and now is chief hydrologist in the Southeast Regional River Forecast Office in Atlanta. "But a half-a-foot difference could inundate a water treatment plant. There was no way we could be that accurate."

Indeed, accounts at the time of the '93 flood cite local emergency managers who told their clients to ignore the weather service forecasts and plan for more water. Since then, local planners have said they might have opted to add sandbags to levees in critical locations if the weather service had issued a range of possible crests.

Although the system was unveiled in Des Moines last spring, it hasn't been tested much, because the weather failed to cooperate. "We've used it quite a bit this year" on minor flooding, says Jeff Johnson, a hydrologist at the NWS forecast office in Des Moines. "But we've had no severe flood to test it."

Still, some forecasters are eager to see it installed. Mr. Feldt notes that during the floods in Georgia in March, the new system might have saved officials in Albany, Ga., some wasted effort.

"Albany did a whole lot of preparation based on our forecast [with the old system]," he says. "We predicted that the river would crest twice, and that the second would be higher than the first. But the extra height never materialized."

The major challenge now is deploying the $100 million system. Dr. Larson points out that while it requires no major hardware purchases, the system is labor intensive to set up. Historical data on each river need to be built into a database, and a range of other conditions have to be described in sufficient detail for the new software to use it.