Faculty Recruiting Support CICS

Better weather prediction through data mining (Alum Amy McGovern's research)

Alum Research Focus: Amy McGovern

Improving the understanding and prediction of severe weather through spatiotemporal data mining

Severe weather including tornadoes, thunderstorms, hail and wind caused $32 billion dollars of damage in 2011 and annually cause significant loss of life.  Although forecasting the path and severity of hurricanes and tropical storms has improved significantly in recent years, tornadoes and other severe events on a smaller scale than hurricanes remain quite difficult to predict. While forecasters can identify conditions favorable for major tornado outbreaks several days in advance, short-term forecasting of individual storms, providing additional advanced notice, and predicting probable tornado paths remain a challenge.

The goal of much of Amy McGovern's (PhD 2002, MS 1998) research as an associate professor in the School of Computer Science at the University of Oklahoma has been to revolutionize tornado prediction and other forms of severe weather. She does this using artificial intelligence, data mining, machine learning and storm simulations. McGovern received a National Science Foundation (NSF) CAREER award in 2008 to jumpstart her research.  She collaborates with the National Oceanic and Atmospheric Administration's (NOAA) National Severe Storm Laboratory (NSSL) and researchers in the School of Meteorology at the University of Oklahoma. She is also working on improving the prediction of aircraft turbulence in collaboration with the National Center for Atmospheric Research (NCAR).

Severe weather poses a very challenging prediction and simulation problem. "Radars provide an incomplete picture of the atmosphere," says McGovern. "Although they can sense the intensity of the precipitation and a single dimension of the wind vector, there are many other important variables such as the full three-dimensional wind field, pressure, temperature, etc. that are important to prediction. Although simulations are an answer to this, fully simulating the atmosphere is not computationally feasible." McGovern is developing a unique set of high-resolution simulations of supercell thunderstorms. These are the most severe type of thunderstorms and cause the most destructive tornadoes. These simulations also provide an unprecedented view of atmospheric turbulence.

Mining the simulations is also challenging. At the resolutions McGovern is simulating, each simulation generates over 1 terabyte (TB) of data. Statistical relational learning is used to identify high-level concepts and relationships in the data that can be used to predict tornadoes. Meteorologists already study existing storm data using conceptual models. They identify high-level concepts and regions in a storm such as updrafts, a region of air blowing upward, and downdrafts, a region of air blowing downward. McGovern's models provide the ability to identify spatiotemporal relationships between these regions that can be used to predict the severe weather events. She has developed novel data mining models that make use of the spatiotemporal nature of the data because neither space nor time can be ignored for weather prediction. In addition, weather is three-dimensional and her models can identify arbitrary shapes and relationships between the shapes.

McGovern's three-dimensional weather modeling of spatiotemporal hazards will be valuable to aviation weather forecasting in support of the future U.S. National Airspace System, known as NextGen. The current Federal Aviation Administration (FAA) system provides guidelines about how close an aircraft can fly to a thunderstorm. Working with researchers at NCAR and using observations collected from aircraft flying over the continental United States to study convectively-induced turbulence, McGovern is improving the prediction of how far turbulence can spread from a thunderstorm. "This can be used to save money by flying more efficient routes and to prevent injuries by flying safer routes," says McGovern.

Another goal of McGovern's work as a professor is to engage, retain, and graduate more underrepresented students. She focuses on developing authentic CS and ML applications, especially those involving severe weather. Since her teaching environment is literally located in the proverbial "tornado alley" or "tornado capital of the world," it provides many real world severe weather experiences in students' actual lives. "They can more easily see the practicality of their CS and ML classes and related labs, homework, projects and research throughout their college experience," notes McGovern. She also engages students through a variety of active learning projects and has won several teaching awards while at the University of Oklahoma.

McGovern's volunteer and community service activities are varied and many relate to the encouragement of undergraduate minorities and younger students to study CS. For the past five years, she chaired the Oklahoma EPSCoR (NSF's Experimental Program to Stimulate Competitive Research) Women in Science conference held for middle and high school students and counselors across the state of Oklahoma. She also serves as faculty advisor to OU's chapter of Alpha Sigma Kappa, a sorority for women in technical studies. As chair of the American Meteorological Society's Committee on Artificial Intelligence Applications to the Environment, McGovern both brings together researchers in the environmental sciences and artificial intelligence and is also reaching out to students in grades kindergarten through sixth. Current development includes an iPad application to demonstrate the uses of artificial intelligence for weather applications.

McGovern joined the University of Oklahoma in 2005 and is currently the Director of the Interaction, Discovery, Exploration, Adaptation (IDEA) Laboratory. While at UMass Amherst, she was advised by Professor Andrew Barto.