CMG Collaborative Research: Model Integration and Joint Inversion for Large-Scale Multi-Modal Geophysical Data

地球物理数据分析和反演是一个涉及建模、数据处理、反演和可视化的高度定量化的领域。 在大多数情况下，进行地球物理实验是为了收集对地球的特定物理性质敏感的数据。 数据被处理和反演以生成或测试所讨论的物理性质的地球模型。 为了更好地了解地球的结构，使用不同的成像模式进行不同的实验。 通常，每个实验的数据被单独反演以生成地球模型的集合。 共享所有物理属性的地球模型通常称为通用地球模型。通用地球模型在科学和商业应用中非常重要，因为集成所有物理信息可以让地球科学家更好地了解重要的地质和地球物理过程。 由于可以理解，利用不同的模态可以改善反演结果，因此许多算法依赖于不同物理模型之间的经验本构关系。 然而，这样的关系通常是站点依赖的，不精确的，并且难以获得。 这就阻碍了共同地球模型的使用和从中获得的理解，这是一个跨学科的研究项目，旨在为多模态地球物理数据的联合反演创建一个更系统的框架。 我们追求两种不同的和互补的方法：一个基于统计，另一个基于几何。 虽然我们的方法具有广泛的反演模式的适用性，我们专注于地震和电磁数据的联合反演。 我们开发的方法将具有更广泛的适用性，超出了联合地震电磁反演问题的目标在这里，更普遍地超出了地球科学领域，如医学成像。

Geophysical data analysis and inversion is a highly quantitative field that involves modeling, data processing, inversion, and visualization. In most cases, a geophysical experiment is conducted to collect data that are sensitive to a particular physical property of the earth. The data are processed and inverted to generate or test an earth model of the physical property in question. To better understand the earth's structure, different experiments are conducted using different imaging modalities. Usually the data of each experiment are inverted separately to generate a collection of earth models. An earth model that shares all physical attributes is usually called a common earth model.Common earth models are very important in scientific and commercial applications because integrating all physical information allows earth scientists to better understand important geological and geophysical processes. Since it is understood that utilizing different modalities may improve inversion results, many algorithms rely on empirical constitutive relationships between the different physical models. However, such relations are typically site-dependent, inexact, and hard to obtain. This hinders the use of common earth models and the understanding that could be obtained from them.This is an interdisciplinary research project to create a more systematic framework for joint inversion of multi-modal geophysical data. We pursue two different and complementary approaches: one based on statistics, and the other on geometry. While our methods have applicability across a wide spectrum of inversion modalities, we focus on the joint inversion of seismic and electromagnetic data. The methods we develop will have wider applicability beyond the joint seismic-electromagnetic inversion problems targeted here, and more generally beyond the geosciences in areas such as medical imaging.