Twelve years ago, forecasters at the North Central River Forecast Center in Chanhassen thought the Red River would just top the 50 foot mark in Grand Forks.
They were off by 3.8 feet - about the height of your average first grader.
In dollar terms, the forecast was about three and a half billion dollars off. That's the amount of damage done when the water started pouring over levees that were built to a little over 51 feet on either side of the river. The river eventually rose to over 54 feet.
Federal, state and local governments have since spent about $400,000,000 shoring up the levees and flood walls in Grand Forks and East Grand Forks.
And forecasters have made some changes, too. Three big ones, really.
Here's the Cliff's Notes version.
Change number one is on a complex chart, known as a "rating curve." If you calculate the total volume of water in a landscape and chart it against the depth of that water collecting together running down a river, you typically get a nice smooth arc that looks like about half a rainbow. And that's how the Red River seemed to work for most of recorded history.
But it turns out that as the volume of water peaks in very flat rivers, like the Red, funny things start to happen. The flooding gets higher, even though the total amount of water is decreasing. If you're plotting it on a graph, the rating curve actually loops up at the end, like a whip snap, before it goes down.
"It's due to a big word, a hysteresis effect," explained Steve Buan, a coordinating hydrologist with the National Weather Service.
He worked on the 1997 flood forecast. The weather service now figures this effect, known as "looping", accounted for about two feet of flooding in Grand Forks.
"It's some complex phenomenon of friction and velocity that are not easily modelable," Buen said. "But we can actually see that through the Geological Survey measurements near the peak that this is going on. So we sort of have the idea physically what's going on, but modeling that in real time has proven to be very difficult."
But at least now it doesn't come as a surprise, and flood forecasts take it into account.
That brings us to change number two, which is also a mouthful.
In weather parlance, it's known as the Kansas City Antecedent Precipitation Index, or API. It's really just a mathematical map that forecasters used to plot a flood forecast -- at one time, literally with a ruler and pencil. The trouble is the API could only handle two scenarios, one with just melting snow, and the other with melting snow and average spring rains. So forecasters had a low and a high number. Or high-ish, at least.
Buan said this left the system unable to predict many other circumstances.
"It didn't give us any indication how much higher it would be if we get above normal precipitation, or what would fall in between there if we got some precipitation," Buan said. "We had this climate record, but we had no way of quantifying that and putting that into that system at the time. We just didn't have the tools to do that."
Now, they do. In fact, National Weather Service forecasters now have computer runs that account for spring weather each year since 1949, including 57 different forecasts for 2009. If it suddenly starts pouring like it did in 1975, the Red River in Moorhead would top out at close to 47 feet and cause outlandish devastation. If the weather suddenly dries up, like it did in 1952, everybody in Fargo could put their boots away right now.
Next week's 38 to 40 foot crest is the statistically most likely scenario.
And finally, there's the third big change to flood forecasting. Which is actually a bunch of changes.
The Upper Midwest has been in what's known as a wet period for almost 20 years now. That means snowmelt is lingering longer in the fields. Water is pouring from one river basin to another like it never did before. And people are fundamentally changing the landscape. Weather records before the 1997 flood mentioned just a single railroad bridge in the Grand Forks area, but that year, there were five spans across the river, backing up the water.
"These bridges are interacting with the water a lot more," Buen said. "We actually saw that in the 2008 floods down in Cedar Rapids, with all the bridges there and the impediment they became and how the stage got pushed up there so high. That's becoming to be more understood over time."
Satellites, unmanned observation planes, GPS technology and bigger computers are all drawing better pictures of where the water goes than the images forecasters had when devastating floods last hit the Red River Valley.
And all of that is going to get a real-world test in about a week.