Large-Scale Inverse Model Analyses Employing Fast Randomized Data Reduction
Y. Lin, E.B. Le, D. O'Malley, V.V. Vesselinov, T. Bui-Thanh
Water Resources Research2017DOI 10.1002/2016WR020299RRR
Summary
This paper introduces a randomized geostatistical approach for inverse problems with very large observational datasets. By combining principal-component geostatistical inversion with randomized sketching for data reduction, the method improves computational efficiency and memory use while retaining the information needed for large-scale model calibration.