From 3D GPR facies and structures toward 3D petrophysical parameter models

探地雷达（GPR）是一种成熟的近地面地球物理工具，在各种应用中得到越来越多的应用。虽然我们看到改进GPR数据采集和处理的方法稳步进步，但在过去几十年中，用于解释处理过的GPR反射图像的概念和技术并没有显著变化。在勘探近地表沉积环境时，2D和3D GPR反射测量通常用于根据需要绘制和成像地下建筑，例如，用于表征沉积环境或辅助水文地质调查。即使在今天，这种GPR数据集的解释在很大程度上仍依赖于人工策略，如反射器选择（以描绘地平线）和特征反射模式的圈定单元（以识别不同的GPR相）。为了以更客观、更高效、更自动化的方式解释GPR数据，我们考虑了基于属性的解释策略，并在之前的项目中开发了一套用于GPR相自动描述的工作流程。该工作流依赖于生成由各种几何和纹理属性组成的属性数据库。经过属性过滤和统计分析（例如，减少冗余信息），我们从数据库中获得3D相图像和分类相模型。综合和现场数据实例表明，这些相模型勾勒出了特征构造单元，有助于对地下埋藏层的地质认识。然而，对于许多（例如，水文地质或岩土工程）应用，结构相模型往往是不够的，因为需要对地下特征和过程有更定量的了解。例如，在水文地质学中，需要对相关地下结构和控制材料特性（即孔隙率和水力导电性）的详细模型来全面了解地下水的流动和输送过程。因此，我们的目标是通过包含额外的基于点的信息和数据来扩展迄今为止开发的解释策略，例如，通过独立的直推式测深和井眼测井。我们提出了将这些点数据纳入工作流的不同想法，包括基于高级统计和机器学习的方法。这些方法将相互比较，并使用不同的综合和现场数据实例进行评估。因此，该后续项目的主要目标是：首先，确定适合建立不同点基岩石物理参数链接的探地雷达属性；其次，确定合适的策略和方法，以获得包含近地表沉积环境构造和岩石物理特征的有意义且可靠的3D模型。

Ground-penetrating radar (GPR) is an established near-surface geophysical tool, which is increasingly employed in variety of applications. Although we see steady methodological progress for improved GPR data acquisition and processing, the concepts and techniques typically employed for interpreting processed GPR reflection images have not significantly changed within the past decades. When exploring near-surface sedimentary environments, 2D and 3D GPR reflection surveying is routinely used to map and image subsurface architecture as needed, for example, to characterize depositional environments or to aid hydrogeological investigations. Even today, the interpretation of such GPR data sets largely relies on manual strategies such as reflector picking (to outline horizons) and delineating units of characteristic reflection patterns (to identify different GPR facies). To interpret GPR data in a more objective, efficient, and also more automated fashion, we have considered attribute-based interpretation strategies and developed, in the preceding project, a workflow for a largely automated delineation of GPR facies. This workflow relies on generating an attribute database consisting of various geometric and texture attributes. After attribute filtering and statistical analyses (e.g., to reduce redundancies information), we derive 3D facies images and classified facies models from our database. As demonstrated by synthetic and field data examples, these facies models outline characteristic structural units being helpful to develop a geological understanding of the buried subsurface. However, for many (e.g., hydrogeological or geotechnical) applications structural facies models are often not sufficient because a more quantitative understanding of subsurface characteristics and processes is needed. For example, in hydrogeology detailed models of the relevant subsurface structures and the governing material properties (i.e., porosity and hydraulic conductivity) are needed to develop a comprehensive understanding of groundwater flow and transport processes. Thus, we aim at extending our so far developed interpretation strategy by including additional point-based information and data as provided, for example, by independent direct-push soundings and borehole logs. We propose different ideas for incorporating such point data into the workflow including approaches based on advanced statistics and machine learning. These approaches will be compared to each other and evaluated using different synthetic and field data examples. Thus, the major objectives of this follow-up project are, firstly, to identify GPR attributes suitable for establishing links to different point-based petrophysical parameters and, secondly, to identify suitable strategies and methods to derive meaningful and reliable 3D models comprising structural and petrophysical characteristics of near-surface sedimentary environments.