Validation and Uncertainty Quantification of Hyperspectral Image Modeling

摘要

In this paper, we introduce a hierarchical approach to validating a multi-scale model for predicting HSI observables from solid particulate materials, and illustrate techniques for quantifying uncertainties in the complex modeling process. We describe an iterative process of using comparative differences between experimental observations and model results to evaluate model validity. Using a single- scale model as a proof-of-concept example, we demonstrate the feasibility of the Bayesian approach to modeling uncertainty at the micro scale and briefly discuss using the results for estimating its impact on model uncertainty at higher scales. We also show that sensitivity and uncertainty analysis, as part of our approach, generated added insight into anomalies in the modeling, which can provide valuable feedback to model developers and experimentalists. Our work helps address UQ within the HSI model application domain and, with continuing efforts, will help researchers and stakeholders gain greater confidence in the utility and defensibility of multi-scale physical models and model outputs.