Monday, November 7 2022
1:00pm - 1:30pm
Advancing the Data-to-Distribution Pipeline for Scalable Data-Consistent Inversion to Quantify Uncertainties in Coastal Hazards

Abstract: We present an overview of the research activities and preliminary results for a recently awarded NSF grant. The proposed research is based upon a novel data-consistent inversion (DCI) approach to identify, quantify, and reduce sources of uncertainty for inputs (parameters) of complex physics-based models of physical/engineered systems using model outputs associated with observational data. This approach is original, data-oriented, rooted in measure theory, and applicable to a wide range of modeling questions of interest to the broader scientific community. Due to their societal importance, the project focuses on three inter-related applications for coastal hazards for which we will apply the proposed mathematical research that advances DCI to improve the inference and prediction of: (i) storm surge and flooding due to hurricanes in coastal communities stretching from the Gulf of Mexico to the western North Atlantic; (ii) arctic storms and evolving sea ice coverage impacting North American coastal communities; (iii) oil spill spread from various sources such as offshore and deepwater drilling rigs. The proposed research builds upon a rigorous measure-theoretic foundation to tackle significant mathematical, statistical, and computational challenges hindering the application of DCI to a wide range of complex physical systems. We propose: (i) a deep learning based data-to-distribution pipeline to transform spatial-temporal data clouds into non-parametric distributions for DCI; (ii) scalable approaches to DCI that simultaneously address computational issues arising from high-dimensional feature spaces as well as limited simulated data availability due to computationally expensive models; (iii) iterative approaches to DCI to identify high-probability feasibility regions for critical model parameters; (iv) continued development and implementation of algorithmic developments in public domain codes for DCI and the data-to-distribution pipeline.
Speaker:Dr. Troy Butler
Affiliation:Department of Mathematical and Statistical Sciences, University of Colorado Denver
Location:StuCo 4017

Download as iCalendar