Stochastic Downscaling Algorithm for Geological Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional downscaling methods for geological models fail to capture fine-scale heterogeneities, preserve continuity across coarse grid areas, quantify static property uncertainty, and condition to available hard data, while also being limited by excessive memory requirements.
Innovation Solution
A computer-aided downscaling algorithm that generates numerical stochastic fine-scale models by creating point and block covariance maps, storing them efficiently based on symmetry, and using kriging systems to simulate values, thereby reducing memory usage and enabling the incorporation of data from different scales.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional downscaling methods are used to maintain data value consistency between coarse and fine grids, then property value consistency is preserved, but fine-scale heterogeneities cannot be captured and continuity across coarse grid boundaries is lost
Solution Approach 1:
The algorithm pre-calculates and stores covariance maps at multiple scales before the actual downscaling process. These pre-computed covariance structures (point-to-point, point-to-block, and block-to-block) are stored in memory for rapid retrieval during simulation, enabling the method to honor both coarse-scale averages and fine-scale variations without recalculating during the main downscaling loop
Solution Approach 2:
The method transforms the downscaling problem by changing the parameter representation from direct property values to covariance-based statistical parameters. By working with covariance maps that encode spatial correlations at different scales, the algorithm can generate fine-scale realizations that match both the statistical properties and the average values of the coarse-scale model, resolving the contradiction between maintaining consistency and capturing heterogeneity
2Adaptability or versatility
If block sequential simulation algorithm is used to integrate coarse-scale and fine-scale data, then data integration capability is improved, but memory requirements become excessive limiting model size to less than 1000 coarse grid cells
Solution Approach 1:
The algorithm extracts and stores only the essential covariance information needed for downscaling into pre-computed maps. By separating the covariance calculation step from the simulation step and storing only the necessary statistical parameters rather than full covariance matrices for all cell pairs, the method dramatically reduces memory requirements while maintaining the ability to integrate multi-scale data
Solution Approach 2:
All computationally intensive covariance calculations are performed in advance and stored in compact map structures. During the actual downscaling simulation, only simple lookups and linear combinations of these pre-stored values are needed, eliminating the need to maintain large in-memory covariance structures throughout the simulation process
3Ease of manufacture
If conventional downscaling methods are used to resample property values from coarse to fine grids, then computational simplicity is maintained, but continuity across coarse grid areas is not preserved and sharp changes are created at boundaries
Solution Approach 1:
The method changes from direct property value copying to a statistical parameter-based approach using covariance maps. By representing spatial variations through covariance structures that encode continuity information, the algorithm naturally produces continuous transitions across coarse grid boundaries while maintaining computational efficiency through pre-computed statistical parameters
Data Source
AI summary
A computer-aided method of downscaling a three-dimensional geological model by generating numerical stochastic fine-scale models conditioning to data of different scales and capturing spatial uncertainties which involves a downscaling algorithm.