T31E-03
Fast Geostatistical Inversion using Randomized Matrix Decompositions and Sketchings for Heterogeneous Aquifer Characterization
Wednesday, 16 December 2015: 08:30
304 (Moscone South)
Ellen B Le1,2, Daniel O'Malley2 and Velimir V Vesselinov3, (1)University of Texas at Austin, Institute for Computational Engineering and Sciences, Austin, TX, United States, (2)Los Alamos National Laboratory, Computational Earth Sciences (EES-16), Los Alamos, NM, United States, (3)Los Alamos National Laboratory, Los Alamos, NM, United States
Abstract:
We present a fast, scalable, and highly-implementable stochastic inverse method for characterization of aquifer heterogeneity. The method utilizes recent advances in randomized matrix algebra and exploits the structure of the Quasi-Linear Geostatistical Approach (QLGA), without requiring a structured grid like Fast-Fourier Transform (FFT) methods. The QLGA framework is a more stable version of Gauss-Newton iterates for a large number of unknown model parameters, but provides unbiased estimates. The methods are matrix-free and do not require derivatives or adjoints, and are thus ideal for complex models and black-box implementation. We also incorporate randomized least-square solvers and data-reduction methods, which speed up computation and simulate missing data points. The new inverse methodology is coded in Julia and implemented in the MADS computational framework (
http://mads.lanl.gov). Julia is an advanced high-level scientific programing language that allows for efficient memory management and utilization of high-performance computational resources. Inversion results based on series of synthetic problems with steady-state and transient calibration data are presented.
