CAREER: Improving Convective-Scale Weather Prediction through Advanced Bayesian Filtering, Verification, and Uncertainty Quantification

University Of Maryland, College Park, College Park MD

Investigators

Abstract

This research project is motivated by the large impact severe convective storms, hurricanes, and flooding have on life and property each year in the United States. The research will address fundamental limitations in current data assimilation (DA) and uncertainty quantification for weather models. The outcome from the research will have a tremendous impact on multi-disciplinary efforts that focus on these weather hazards, which will lead to sustained long-term reductions in forecast errors. Intellectual Merit: The project will focus on 1) the development and testing of an advanced DA framework designed to eliminate specific assumptions currently used in practice; 2) the adoption of sophisticated uncertainty visualization schemes for exploring probabilistic information estimated from ensembles; and 3) the development of a DA research and educational module for class curriculum at the University of Maryland and external summer schools. The project will adopt new Bayesian filtering techniques based on "particle filters". The method uses samples of model simulations to represent probabilistic properties of model state variables conditioned on current and past observations. In addition to providing the most thorough investigation of particle filters for weather prediction, the research will apply the method for isolating sources of bias in models and observing systems. Another unique aspect of this work is its use of novel visualization techniques developed by statisticians and computer scientists. These techniques form a set of analysis tools based on "data depth," which allows for an insightful look at multivariate ensemble output via contour and curve boxplots. They also provide a means of verifying probabilistic quantities from ensembles with no assumptions for the underlying error distribution, thus aiding in the verification of non-parametric DA techniques, like particle filters. Broader Impacts: This research is motivated directly by hazardous weather events that affect the well-being of individuals in the United States and around the world. In addition to advancing predictive skill in numerical predictions, a portion of this work focuses on uncertainty quantification and visualization of ensemble datasets, which aims to improve the communication of severe weather risk to the public. DA advancements made during this work will be committed to the National Center for Atmospheric Research's Data Assimilation Research Testbed, a community software infrastructure for linking DA research to geoscientists. The project also includes a detailed strategy for developing an educational DA module for geoscience students and researchers. The module will evolve with the work plan and result in a valuable learning tool for class exercises and summer school activities used to promote diversity in STEM fields. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.