CAREER: Developing Dynamic Relational Models to Anticipate Tornado Formation

点击翻译按钮获取中文摘要

The goal of this research is to revolutionize the ability to anticipate tornadoes by developing advanced techniques for statistical pattern discovery in spatially and temporally varying relational data. These models are applied to complete fields of meteorological quantities obtained through data assimilation and simulation. Doppler radar data is limited and, while modern data assimilation techniques allow the unobserved quantities to be estimated, the resulting four- dimensional fields are too complicated for the extraction of meaningful, repeatable patterns by either humans or current data mining techniques. By studying a full field of variables, the models can identify critical interactions among high level features. The models are developed and verified in close collaboration with domain experts.The interdisciplinary research is used to improve retention and recruitment in computer science (CS). This draws on recent evidence that underrepresented groups are not drawn to computing careers because they do not appreciate how computing can be used to solve real world problems. Introducing authentic projects into both early CS and meteorology classes will improve the number of technically trained students in both majors.The primary broader impact of this research is to society, through the potential for reduction in loss of human life, property, and money. Models will be made available to operational meteorologists as they are verified. Another broader impact will come from increasing the number of computing oriented minors and majors through authentic projects. All data and results will be disseminated through peer reviewed publications and via open source online repositories accessible on the project Web site (http://www.cs.ou.edu/~amy/career/).