| Building numerical models requires to integrate all available data. In the earth sciences, data typically come from different sources with different volume supports. Some are fine scale quasi-point support, such as well or core data; others are coarse scale data averaged over large block support, such as remote sensing and seismic travel time data. Both point and block support data are valuable information and should be incorporated into the final models. In addition, prior information, such as spatial correlation and property statistics (mean, variance, etc), should also be considered. This thesis aims at building high resolution models conditioned to both point and linear average block support data and accounting for prior structural information.; Three algorithms, kriging, direction sequential simulation ( DSSIM) and error simulation, are proposed to integrate point and block data. They are extended to account fully for both point and block supports. Estimation and simulation are performed in the original data space, which allows preserving the linearity of block data. These three algorithms are developed into three SGeMS plug-ins: BKRIG (block kriging estimation), BSSIM (block sequential simulation) and BESIM (block error simulation). BKRIG provides a deterministic least square type estimation map. BSSIM provides a stochastic solution based on DSSIM with histogram reproduction. BESIM is a faster stochastic algorithm based an error simulation. In the process of implementation, focus has been given to make the algorithms fast, general enough to handle blocks with any geometry, and applicable to large 3D cases.; A critical issue relates to the computation efficiency of the block covariances defining the correlations between data at different scales. Three approaches, traditional integration, FFT-integration hybrid and analytical approaches, are proposed and investigated. The traditional and the hybrid approaches have been implemented as SGeMS libraries. A SGeMS utility program BCOVAR is developed to compute and visualize these block covariances. These programs are applied to two types case studies: downscaling coarse data conditional to the fine scale data, and integrating travel-time tomography data with well data. The results show that point and block data are honored; histogram and spatial correlation are reproduced; uncertainty is assessed from the different realizations. Approved for publication: By: Yongshe Liu For (Energy Resources Engineering Dept.)... |
