A Tree-based Direct Sampling Method for Stochastic Modeling in Surface and Subsurface Hydrology
Abstract
Direct sampling (DS) is a high-order stochastic modeling framework with extensive applications including hydrostratigraphic modeling, rainfall modeling, gap filling of remote sensing, etc. Using a training image as an explicit prior model, DS can generate realistic spatial models, honor hard and soft data, tackle multivariate, and quantify spatial uncertainty. However, one limitation of DS is the running time because the spatial information of training images is not well organized. It is very time-consuming for geometrically complex, large size and high-dimensional modeling cases.
In this work, we develop a tree-based direct sampling (TDS) method to address the above challenges. First, patterns in the training image are grouped according to their similarities. With the objective to rapid query, we introduce a tree structure to organize the clustering results. Second, we propose a tree-driven search approach to efficiently retrieve patterns. Not only similarity but also diversity is considered to avoid local optima. Third, an adaptive simulation patch is designed to balance efficiency and hard data conditioning. We validate the proposed method using a classical 2D channel simulation. Compared to the original DS, our TDS significantly improves computational efficiency without simulation quality degradation. We apply TDS to two real cases in glaciology and hydrogeology. The first one is to solve a gap-filling problem in Antarctica topography modeling, while the other is for groundwater modeling with borehole data and geophysics data in Denmark. Our TDS efficiently creates a group of 2D glacial bed elevation models and 3D hydrostratigraphic models to observe spatial uncertainty. The experiment results indicate that generating realistic models have a significant impact on understanding subglacial flow behavior and predicting groundwater resources.- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2019
- Bibcode:
- 2019AGUFM.H13Q1991Y
- Keywords:
-
- 1847 Modeling;
- HYDROLOGY;
- 1873 Uncertainty assessment;
- HYDROLOGY;
- 1880 Water management;
- HYDROLOGY;
- 1916 Data and information discovery;
- INFORMATICS