Locating change regions of seismic velocities and seismic scattering properties associated with volcanic activities and earthquakes is important for structural monitoring. To increase such applications, we propose to use sparse modeling to estimate spatial distributions of seismic scattering property changes. The sparse modeling is an inversion technique that enables us to estimate model parameters from a small data set with sparsity condition such as ℓ ¹ norm regularization. We apply this technique to seismic ambient noise cross‐correlation functions from 17 Hi‐net stations around the epicenter of the 2008 Iwate‐Miyagi Nairiku, Japan, earthquake (M w =6.9). We compute waveform decoherences at the 0.5–1 Hz band and invert the waveform decoherences for the spatial distributions of seismic scattering property changes. Just after the main shock, the largest change occurred at the south of the epicenter, and the maximum change of the scattering coefficient in this region is estimated to be 0.032 km⁻¹. The result from an ordinary linear least squares inversion with the ℓ ² norm regularization is almost consistent with that from the sparse modeling. Moreover, we confirm the superiority of sparse modeling in imaging with smaller data sets. Only five seismic stations that are deployed near the epicenter so as to surround the change regions are necessary to retrieve the result from 17 stations. On the other hand, in the case of the ℓ ² norm regularization, we need at least 15 stations. The sparse modeling will be helpful to estimate the spatial distribution of seismic scattering property changes from a small data set.

中文翻译：

---

通过地震环境噪声互相关函数的稀疏模型来定位地震散射特性的空间变化：在日本2008年岩手宫城奈里库（Mw 6.9）中的应用

定位与火山活动和地震有关的地震速度和地震散射特性的变化区域对于结构监测很重要。为了增加这种应用，我们建议使用稀疏模型来估计地震散射特性变化的空间分布。稀疏建模是反演技术，使我们能够从与稀疏性条件的小的数据集估计模型参数如ℓ ¹范数正则。我们将该技术应用于日本2008年岩手县宫城县内里区震中附近17个高台站的地震环境噪声互相关函数中（M w= 6.9）。我们计算0.5-1 Hz频带上的波形退相干，并针对地震散射特性变化的空间分布反转波形退相干。在主震之后，最大的变化发生在震中南部，该区域散射系数的最大变化估计为0.032 km ^-1。从一个普通的线性最小二乘反演与结果ℓ ²规范正则化与稀疏建模几乎一致。此外，我们确认了稀疏建模在较小数据集成像中的优越性。只有五个地震台站部署在震中附近，以便围绕变化区域，才能从17个台站获取结果。在另一方面，在的情况下，ℓ ²范数正，我们至少需要15个站。稀疏建模将有助于从较小的数据集中估计地震散射特性变化的空间分布。