Envisioning Data Assimilation for Adaptive Mesh Computational Models and Intermittency
Abstract
Project Abstract SummaryApproved for Public ReleaseEnvisioning Data Assimilation for Adaptive Mesh Computational Models and IntermittencyPI: Erik Van Vleck, University of KansasThe use of adaptive meshing techniques for data assimilation (DA) presents both challengesand opportunities. The focus of the proposed research will be on the development of a flexibleframework for adaptive mesh DA (AMDA) based on the combination of metric tensors associatedwith different factors. Among the factors considered will be the accuracy of ensemble solutions, thelocation of observations, detection and increased resolution of intermittent extreme events, trackingof coherent structures, as well as incorporating adjoint sensitivity analysis into the development ofadaptive meshes for DA.The techniquesdeveloped will be applicable adaptively in time. Many of the factors beingemployed (accuracy and conditioning of ensemble forecasts, observation locations, Lagrangianflow, intermittent extreme events, adjoint sensitivities, etc.) to determine enhanced meshes areoften time dependent. A primary goal is to identify and weight different factors that will be usedto generate the forecast and analysis meshes in ways that enhance the DA skill. There is a tradeoffbetween the gains in accuracy (of both the forecasts and the analysis) and efficiency from adaptivemeshing and the loss of accuracy and efficiency from interpolating between meshes. As a result,we will further develop meshes (typically requiring greater computational resources) that performeffectively over larger ranges of time and/or solution behavior.An ultimate goal of this research is learning how to optimize different aspects of DA skill byweighting different factors and in particular their impact on adaptivity in the DA process. There isa natural competition for computational resources and this research can be viewed as developing aflexible means to allocate these resources to minimize what are ultimately errors in the DA process.The AMDA techniques we envision include the ability to put a magnifying glass on places andobjects to improve their resolution with respect to both the computational model for prediction andduring an analysis update.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 15, 2024
- Source ID
- N000142412218
Entities
People
- Erik Van Vleck
Organizations
- Office of Naval Research
- United States Navy
- University of Kansas