Sequential data consistent inversion algorithms for parameter estimation in coastal storm surge models using high performance computing applications

Computational methods have gained significant prominence in addressing complex challenges across science and industry. These methods are grounded in physical principles and expressed in mathematical models, usually as a set of partial differential equations (PDEs), which are then discretized and solved, often times requiring High-Performance Computing (HPC) resources. Nevertheless, the efficacy of these computational methods is inherently tied to their ability to align predictions with observations and quantify uncertainties. This dissertation explores the critical intersection of three domains: Uncertainty Quantification (UQ), Storm-Surge Modeling, and High-Performance Computing (HPC) Applications for Ensemble Simulations (ES). The primary contribution is a set of novel algorithms for generating sequential parameter estimates and quantifying epistemic uncertainty in dynamical systems within a data-consistent (DC) framework. In particular, data-constructed Quantity of Interest (QoI) maps are introduced using observed and simulated data to reduce uncertainty in model parameters and learn the optimal space to perform parameter inversion. A Maximal Updated Density (MUD) parameter estimate, similar to MAP estimate in Bayesian frameworks, is constructed using the DC update in a sequential context, with previous iterations informing subsequent ones, and with computational diagnostics within the DC framework providing critical information to both evaluate the quality of the DC update and MUD estimate as well as helping detect potential parameter value drifts. The results presented showcase the potential of DC methods and MUD estimation in operational settings for analyzing and quantifying uncertainties in real- or near-real-time as packets of noisy observational data are obtained at discrete times from a network of sensors. In order to apply the sequential parameter estimation techniques presented, Ensemble Simulations (ES) of the high-fidelity forward models must be coordinated and executed in a reliable and reproducible manner. Thus this dissertation also delves into the issue of HPC Applications for Ensemble Simulations, recognizing the emergence of complex workflows as the dominant form of research applications. Developing these workflows presents its own set of challenges, including data staging, HPC system configuration, and job management. To address these challenges, the TACC Job Manager (taccjm) and related tools are introduced, enabling efficient interaction with HPC resources and the development of complex HPC workflows at the Texas Advanced Computing Center (TACC). Additionally, two versatile ES applications, PySLURM Task Queue (pyslurmtq) and tapis-pylauncher, are introduced as general purpose parametric job launchers for running ES. The culmination of this work is the application of the sequential DC algorithms to estimate wind drag parameters for a simulated extreme weather event using the high-fidelity storm-surge model ADCIRC. The results demonstrate how sequential DC algorithms cannot only effectively estimate parameters that match observations, but also inform decision makers on the quality of these parameter estimates and their sensitivity towards the dynamics of the system. Furthermore the application to reproduce the simulations on HPC systems is published via the DesignSafe cyber-infrastructure platform - a cloud based platform providing access to HPC compute resources and data storage - making the application openly available and reproducible.