Three days before Hurricane Andrew struck South Florida, not a single computer model used by forecasters indicated that the storm would hit where it did, and the art of predicting a storm's path remains just as chancy, experts warn.
Today, thanks to modern technology, scientists at the National Hurricane Center know a storm is brewing days before it is strong enough to have a name.
They can track every shift of its path as it crosses thousands of miles of ocean and can measure its size and strength on an hourly basis.
But predicting where it will go next is as much an art as a science, even in 1993, says Jerry Jarrell, the center's deputy director.
"Our forecasting models are improving, but they're not improving very fast," he said.
Warning time is a key not only to preventing destruction, but also to saving lives.
Hurricane watches are issued 36 hours before a storm is expected to hit land, and hurricane warnings only 24 hours.
That's not enough to evacuate congested areas with limited access, such as the Florida Keys or Miami Beach, experts say.
"We really need to make the watch and warning period longer," Jarrell said. But the current uncertainty makes that impossible.
When the Andrew warning went up the day before the Aug. 24 storm, the degree of uncertainty was so great that an evacuation had to be ordered in flood zones from Key West to Vero Beach, a stretch of about 300 miles of coastline where 1.1-million people reside.
Twelve hours earlier, the hurricane warning went as far north as Daytona Beach, covering almost 450 miles. The storm, however, hit along only about a 30-mile-wide swath in south Dade County.
To predict its course, Hurricane Center computers include data on water temperature, air pressure, wind direction and speed, as well as other variables.
But scientists have only a rudimentary understanding of some factors, and even a difference of a hundredth of a degree can throw a computer model way off.
There is some hope of improving predictions, however. The Geophysical Fluid Dynamics Laboratory at Princeton, which looks at broader conditions around the storm, has been used on 14 weather systems so far and appears to be a significant improvement.
It shows that weather to the north of hurricanes may be more significant in steering storms than weather inside the systems.