Integrated Foundations of Sensing, Modeling, and Data Assimilation for Sea Ice Prediction
Abstract
DoD operations gain significant tactical advantage from accurate and detailed environmental information. While some environmental variables are well captured and forecast by current numerical weather prediction capabilities, many important aspects of the operational environment remain poorly described. Inadequately described environmental variables tend to share several characteristics. They are generally (1) discontinuous in space or time, (2) sparsely observed, and (3) governed by poorly described physics. We refer to these Discontinuous, Sparse, Underdescribed environmental state variables as DSU variables. Traditional approaches to improving the description of these DSU variables largely center on making advances in model resolution (to better capture discontinuities), improving observing technology (to reduce data sparsity or uncertainty), and developing process-based understanding (to improve algorithms describing the phenomena). While all are valid for improving environmental awareness of a specific DSU variable, we hypothesize that major, generalizable advances in the prediction of DSU variables will come from developing rigorous frameworks for data-model synthesis that can fully utilize the rapid increase of computational power. In particular, emerging mathematical tools permit describing discontinuous fields and extracting information from the sparse data and uncertain physical models. The goal of this project, therefore, is to develop a computational toolkit that enhances the information we can extract from currently existing environmental data and physical knowledge to improve DSU variable prediction. Our approach will be to develop, advance, adapt, and combine computational tools that address four general issues hindering accurate prediction of DSU variable fields, including (i) functional representation of discontinuous data; (ii) infilling of sparse fields; (iii) detection of fine features in observations; and (iv) managing conflicting or erroneous data and propagating uncertainty. The team will combine their expertise in discrete element methods, Bayesian inference and likelihood modeling, numerical optimization, and generalized inflation and likelihood reformulation to ensure a rigorous and comprehensive approach for addressing these issues. Together these efforts will formalize a framework for combining available data and physical understanding to produce estimates of DSU variable fields. The project will be executed with parallel development of tools during the first 2 years. In years 3 and 4, the tools will be integrated and tested in a series of increasingly complex and realistic environmental prediction/awareness problems. For the purposes of this project we have identified prediction of Arctic sea ice thickness and stress state with an eye toward navigation of a ship in heavy ice as our goal. The specific test case is largely irrelevant to the generalizable tools we seek to create. We choose this particular application because it will benefit from nexus with related ONR-funded projects, including the ONR-MURI Advanced Analytic and Computational Modeling of Arctic Sea Ice and the ONR-SIDEx program, both of which we expect will be sources of real data for tests. Our team includes an expert in the observation and modeling of sea ice who will facilitate data and physical models transitions and provide general understanding of the prediction problem. The ultimate outcome of the project will be a published DSU data toolkit a generalized set of software tools for creating optimal predictions of any DSU environmental variable from available data and models.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 20, 2020
- Source ID
- N000142012595
Entities
People
- Anne Gelb
Organizations
- Board of Trustees of Dartmouth College
- Office of Naval Research
- United States Navy