Training image selection for development of subsurface geological cross-section by conditional simulations

Research output: Journal Publications and Reviews (RGC: 21, 22, 62)21_Publication in refereed journalpeer-review

5 Scopus Citations
View graph of relations

Author(s)

Detail(s)

Original languageEnglish
Article number106415
Journal / PublicationEngineering Geology
Volume295
Online published21 Oct 2021
Publication statusPublished - 20 Dec 2021

Abstract

Delineation of a subsurface geological cross-section can be accomplished by conditional simulations, which can effectively combine prior geological knowledge as training images with site-specific measurements for spatial interpolation. Valuable prior geological knowledge may be concisely represented by a single training image, and the evaluation and optimal selection of a representative training image is crucial for a successful application of conditional simulation methods. In this study, a data-driven method based on edge orientation detection is proposed for selection of the optimal training image. The stratigraphic soil boundaries of all candidate training images (CTIs) and site-specific measurements are scanned by an edge detector. The derived edge orientations are quantified and compared between CTIs and site-specific measurements. Among all CTIs, the CTI that has the minimal difference of edge orientation distribution with respect to the site-specific measurements is selected as the optimal one. The proposed method is validated using both an illustrative example and a real case. It is demonstrated that edge orientation successfully differentiates soil stratigraphic patterns between different training images, and the derived edge orientation distribution can be used as a quantitative indicator for selection of the optimal training image.

Research Area(s)

  • Convolutional neural network, Edge orientation detection, Image-based method, Sparse measurements, XGBoost