Seismic full-waveform inversion (FWI) is a powerful method used to estimate the elastic properties of the subsurface. To mitigate the nonlinearity and cycle-skipping problems, in a hierarchical manner, one first inverts the low-frequency content to determine long- and medium-wavelength structures and then increases the frequency content to obtain detailed information. However, the inversion of higher frequencies can be computationally very expensive, especially when the target of interest, such as oil/gas reservoirs and axial melt lens, is at a great depth, far away from source and receiver arrays. To address this problem, we have developed a localized FWI algorithm in which iterative modeling is performed locally, allowing us to extend inversions for higher frequencies with little computation effort. Our method is particularly useful for time-lapse seismic, where the changes in elastic parameters are local due to fluid extraction and injection in the subsurface. In our method, the sources and receivers are extrapolated to a region close to the target area, allowing forward modeling and inversion to be performed locally after low-frequency full-model inversion for the background model, which by nature only represents long- to medium-wavelength features. Numerical tests show that the inversion of low-frequency data for the overburden is sufficient to provide an accurate high-frequency estimation of elastic parameters of the target region.

中文翻译：

---

时移地震勘探的高频局部弹性全波形反演

地震全波形反演（FWI）是一种强大的方法，可用于估计地下的弹性特性。为了减轻非线性和循环跳跃问题，人们首先以一种分层的方式将低频成分求反以确定长波和中波结构，然后增加频率成分以获得详细的信息。但是，较高频率的反演在计算上可能会非常昂贵，尤其是当感兴趣的目标（例如油气储层和轴向熔融透镜）处于很深的深度，并且远离源阵列和接收器阵列时，尤其如此。为了解决这个问题，我们开发了一种局部FWI算法，其中在本地执行迭代建模，从而使我们能够以很少的计算工作来扩展更高频率的反演。我们的方法对于延时地震特别有用，在延时地震中，由于地下流体的抽取和注入，弹性参数的变化是局部的。在我们的方法中，将源和接收器外推到靠近目标区域的区域，从而在对背景模型进行低频全模型反演后，可以在本地执行前向建模和反演，因为自然背景下，该模型仅代表中长距离波长功能。数值测试表明，覆盖层低频数据的反演足以提供目标区域弹性参数的准确高频估计。源和接收器被外推到靠近目标区域的区域，从而允许在对背景模型进行低频全模型反演后本地进行正向建模和反演，因为自然背景仅代表长波到中波特征。数值测试表明，覆盖层低频数据的反演足以提供目标区域弹性参数的准确高频估计。源和接收器被外推到靠近目标区域的区域，从而允许在对背景模型进行低频全模型反演后本地进行正向建模和反演，因为自然背景仅代表长波到中波特征。数值测试表明，覆盖层低频数据的反演足以提供目标区域弹性参数的准确高频估计。