Bayesian inference of earthquake source parameters: kinematic and dynamic finite fault models

正确认识控制地震破裂传播的地球物理过程是减轻地震灾害的关键。滑移反演通过模拟地震记录提供了破裂传播的细节。然而，它们本质上是非唯一的，需要正则化，并且通常缺乏适当的不确定性估计。我们的目标是i)利用动态破裂建模来正则化解决方案，ii)使用贝叶斯方法来正确捕获检索参数的不确定性。为了进一步提高我们对影响破裂的物理过程的了解，我们将利用SeisSol软件包将模型从有限源反演扩展到更现实的条件。这种增强的方法将主要应用于选择的M~6级地震，这些地震在破裂传播过程中具有不同的行为，很难用流变复杂性（分段跳跃、大浅层滑动、断层摩擦屏障）来解释，也可能应用于研究期间在全球范围内发生的一些记录良好的重大地震（M bbb6级）。拟议的研究将需要为地震断层破裂过程的数学建模开发新的代码，修改现有框架并利用高性能计算基础设施。为了在贝叶斯框架下发展运动学和动力滑动反演，我们将扩展布拉格查尔斯大学（CUP）的运动学震源反演代码，并开发新的动态震源反演代码。为了从观测中约束震源过程，我们将把推断的震源模型扩展到选定真实事件的大尺度（高分辨率）动态破裂情景。为此，我们将利用ludwig - maximilian - university Munich （LMU）开发的SeisSol软件包，该软件包允许合并复杂的自然断层带。这项双边资助提案旨在补充两名年轻研究人员在震源物理和地面运动建模方面的专业知识。两组研究人员的专业知识的结合将允许超越标准的运动学源反演和基于场景的正演动态破裂模拟，从而开发完全动态的有限范围源模型，以执行物理一致的有限范围源反演，包括通过贝叶斯推理评估参数不确定性。新开发的代码结合了解决正演和反演问题的尖端方法，将利用现代和未来的高性能计算基础设施，并将完全提供给全世界的地震学界。

Proper understanding of geophysical processes controlling earthquake rupture propagation is a key point in earthquake hazard mitigation. Slip inversions provide details on rupture propagation by modeling seismic records. However, they are inherently non-unique requiring regularization and typically lacking proper uncertainty estimation. We aim at i) utilizing dynamic rupture modeling to regularize the solution and ii) use a Bayesian approach to capture correctly the uncertainty of the retrieved parameters. To further improve our knowledge about the physical processes affecting the rupture, we will extend the models from finite source inversions to more realistic conditions utilizing the SeisSol software package. This enhanced approach will be applied primarily to selected M~6 events with distinct behaviors in their rupture propagation that are challenging to explain in terms of rheological complexity (segment jumps, large shallow slip, on-fault frictional barriers), and possibly also to a few well-recorded significant (M>6) earthquakes that would occur worldwide during the research period. The proposed research will require the development of new codes for mathematical modeling of the earthquake fault rupture process, modification of existing frameworks and utilization of high-performance computing infrastructure. For the development of kinematic and dynamic slip inversions in the Bayesian framework we will extend kinematic earthquake source inversion codes of the Charles University in Prague (CUP) and develop new ones for dynamic source inversions. To constrain earthquake source processes from observations, we will extend the inferred earthquake source models to large-scale (high-resolution) dynamic rupture scenarios of selected real events. To this aim, we will utilize the SeisSol software package developed at the Ludwig-Maximilians-University Munich (LMU) which allows to incorporate complex, natural fault zones. This proposal for bilateral funding aims to complement expertise of two young researchers in earthquake source physics and ground motion modeling. Combination of expertise of the researchers of the two groups will permit to go beyond standard kinematic source inversions and beyond scenario-based forward dynamic rupture simulations towards the development of fully dynamic finite-extent source models, to perform physically consistent, finite-extent source inversions including evaluation of parameter uncertainties by means of Bayesian inference. The newly developed codes, combining cutting-edge methods for solving the forward and inverse problem, will make use of modern and future high-performance computing infrastructure and will be made fully available to the seismological community worldwide.