Dataset - Simulation Results Ensembles of synthetic simulations of 2D stationary saturated flow at the Hanford Site

This repository contains three datasets consisting of 20,000 randomly generated two-dimensional log-transmissivity field realizations over the Hanford Site's geometry, and the corresponding stationary saturated flow pressure field. These datasets were generated using the code hosted on https://github.com/yeungyh/hanford_data. The log-hydraulic transmissivity fields are constructed by computing and sampling the 1,000-dimensional truncated Kosambi-Karhunen-Loève (KKL) expansion of a stationary Gaussian Process Regression (GPR)/Kriging model calibrated using spatially sparse observations of a reference log-hydraulic transmissivity field. The pressure fields are computed by solving the corresponding stationary flow problem for fixed reference boundary conditions using the two-point flux approximation-based finite volume method. The reference log-transmissivity field and boundary conditions are computed using data from the report "Transient Inverse Calibration of Hanford Site-Wide Groundwater Model to Hanford Operational Impacts - 1943 to 1996" (https://doi.org/10.2172/786781).

This repository contains two "unconditional" sets and one "conditional" set. For "unconditional" sets, log-transmissivity measurements are only used for estimating the parameters of the GPR model. For "conditional" datasets we condition the GPR model using log-transmissivity observations before computing the KKL. Therefore, for unconditional datasets, the variance of the log-transmissivity realizations is constant and equal to the calibrated GPR variance, whereas for conditional models the variance at the observation locations is close to zero (but not equal to zero due to truncation).