Using Multi-level Methods for Geospatial Data Representation to Enable Efficient Analytics of Large Scale Simulation Outputs

Abstract:

Uncertainty Quantification of large scale simulations of geophysical mass flows is essential to develop effective hazard analysis procedures. We will present work conducted by a multi-disciplinary group of scientists comprised of geo scientists, statisticians, mathematicians and engineers on a procedure for such analysis. Our methodology involves the use of Bayesian inference for constructing statistical surrogates. These "models" have to be carefulluy developed to enable efficient computation without loss of fidelity. Such a method required us to overcome many challenges in large data, complex and expensive computation, numerical errors and sparse observation data. We use the analysis of hazards from volcanos at several sites to illustrate our methodology.