UNCERTAINTY QUANTIFICATION AND DECOMPOSITION THROUGH BAYESIAN DEEP LEARNING FOR BIG DATA SATELLITE REMOTE SENSING PROBLEMS
Loading...
Authors
Ortiz, Pedro
Subjects
Bayesian deep learning
uncertainty quantification
uncertainty decomposition
passive microwave
infrared
remote sensing
big data
uncertainty quantification
uncertainty decomposition
passive microwave
infrared
remote sensing
big data
Advisors
Orescanin, Marko
Date of Issue
2023-06
Date
Publisher
Monterey, CA: Naval Postgraduate School
Language
Abstract
This dissertation demonstrates the value of uncertainty quantification and decomposition for big data satellite remote sensing problems in both classification and regression settings. Bayesian deep learning methods are applied to a classification and a regression problem with datasets in excess of 14 million samples, quantifying total uncertainty and decomposing the total uncertainty into separate components. In all cases, Bayesian probabilistic models perform comparably to their deterministic counterparts but provide valuable additional information in the form of quantified uncertainty that can be decomposed by source. The value of the quantified uncertainty is demonstrated by analyzing the relationship between model errors, uncertainty and underlying physical properties of the observational data and the atmosphere. Quantitative analysis of accuracy, error, and uncertainty metrics is illustrated through use cases for quantified uncertainty that informs decisions concerning high uncertainty predictions, model selection, targeted data analysis, data collection/processing and virtual concept drift (distribution shift) detection. These methods apply wherever deterministic deep learning is currently being applied or where it might be applied in the future. The work presented in this dissertation has the potential to positively affect joint all-domain command and control (JADC2) in both present-day and future operations.
Type
Thesis
Description
Series/Report No
Department
Organization
Identifiers
NPS Report Number
Sponsors
Funding
Format
150 p.
Citation
Distribution Statement
Distribution Statement A. Approved for public release: Distribution is unlimited.
Rights
This publication is a work of the U.S. Government as defined in Title 17, United States Code, Section 101. Copyright protection is not available for this work in the United States.