Washington ? Government and university scientists announced Monday a major study of why big, violent storms occur and how they can be predicted.

The study will cover the Midwest thunderstorm zone from South Dakota to Ohio, using radar-bearing aircraft and ground-based mobile laboratories. It begins May 20 and runs through July 6 under the direction of the National Center for Atmospheric Research in Boulder, Colo.

The scientists will investigate huge thunderstorm complexes that can spread hurricane-force winds and torrential rains for hundreds of miles. These monsters sometimes spin off tornadoes, like those that killed at least 38 people in Kansas, Missouri and Tennessee on Sunday. The $4 million study was in the works long before this week.

Most ordinary summer storms are born and die in an hour or two and span about 12 miles. But these large systems — known as mesoscale convective vortices — can be 500 miles long and 90 miles wide. They typically develop in the late afternoon and can last all night, bringing winds up to 100 miles an hour and intense rainfall.

“The effects are like a land hurricane,” said Christopher Davis, co-leader of the project at the research center.

Even after one of these storms decays, its low-pressure vortex may persist and give rise to a new storm the next day.

These storms sometimes give rise to tornadoes at their leading or trailing edge, as happened Sunday and Monday. More typically, damage from superstorms is due to high winds and floods.

Between January 1995 and July 2000, high winds from mesoscale storms caused $1.4 billion in damages, 72 deaths and 1,000 injuries, according to the National Oceanic and Atmospheric Administration, which operates the Weather Service.

Researchers have been using computer simulations to try to understand how these storm clusters work, but until now there have been no large-scale data-gathering projects.

“We’d gone about as far as we could with the idealized simulations,” said Morris Weisman, co-leader of the project at the atmospheric research center. “We needed to get good data.”