Integrated inversion using combined wave-equation tomography and full waveform inversion
Integrated inversion using combined wave-equation tomography and full waveform inversion
复制标题
DOI：
10.1093/gji/ggu138
复制
发表时间：
2014-07
影响因子：
2.8
通讯作者：
Haiyang Wang;Satish C. Singh;H. Calandra
中科院分区：
文献类型：
--
作者：
Haiyang Wang;Satish C. Singh;H. Calandra
摘要通过混合误差函数将波动方程层析成像（WT）和全波形反演（FWI）相结合，以从约束较差的初始速度模型开始估计高分辨率地下结构。这两种方法共享相同的波场正演模拟和反演方案，而它们的区别仅在于计算误差函数的方式以及在模型空间中采样的方式。小波变换以减小合成数据与真实的数据之间的互相关相位延迟为目标，可以提取FWI所必需的长、中波长模型分量。与基于渐近高频近似的射线走时层析成像方法相比，小波变换能够更好地利用地震波场的带限特性，从而获得更高的分辨率。另一方面，FWI能够分辨短波长模型分量，补充WT。在这项研究中，我们将WT应用于地表初至折射数据，并将FWI应用于折射和反射数据。我们为两个不同的误差测量分配自适应权重，并建立一个渐进的反演策略。为了说明我们的战略比传统的“射线层析成像+ FWI”的方法的优势，我们在合成透镜测试，WT可以提供额外的地下信息，这是一个成功的FWI应用程序的关键。为了进一步展示效率，我们在2-D Marmousi模型上测试了我们的策略，在没有太多人工干预的情况下，获得了令人满意的反演结果。最后，我们将反演策略应用于苏门答腊近海12公里长的拖缆采集的深水地震数据集。为了解决深水环境下的实际问题，我们采用向下延拓（DC）方法在反演前生成虚拟海底实验数据集。DC之后的新几何结构增强了浅层折射，并避免了通过厚水柱进行繁琐的建模，从而将计算成本降低了85%，新数据集的反演结果显示了高分辨率的浅层沉积物结构，偏移图像证明了反演模型优于常规层析成像模型。
SUMMARY Wave-equation tomography (WT) and full waveform inversion (FWI) are combined through a hybrid misfit function to estimate high-resolution subsurface structures starting from a poorly constrained initial velocity model. Both methods share the same wavefield forward modelling and inversion schemes, while they differ only on the ways to calculate misfit functions and hence the ways to sample in the model space. Aiming at minimizing the cross-correlation phase delay between synthetic and real data, WT can be used to retrieve the long- and middle-wavelength model components, which are essential to FWI. Compared to ray-based traveltime tomography that is based on asymptotic high-frequency approximation, WT provides a better resolution by exploring the band-limited feature of seismic wavefield. On the other hand, FWI is capable of resolving the short-wavelength model component, complementing the WT. In this study, we apply WT to surface first-arrival refraction data, and apply FWI to both refraction and reflection data. We assign adaptive weights to the two different misfit measurements and build a progressive inversion strategy. To illustrate the advantage of our strategy over conventional ‘ray tomography + FWI’ approach, we show in a synthetic lens test that WT can provide extra subsurface information that is critical for a successful FWI application. To further show the efficiency, we test our strategy on the 2-D Marmousi model where satisfactory inversion results are achieved without much manual intervention. Finally, we apply the inversion strategy to a deep-water seismic data set acquired offshore Sumatra with a 12-km-long streamer. In order to alleviate several practical problems posed by the deep-water setting, we apply downward continuation (DC) to generate a virtual ocean bottom experiment data set prior to inversion. The new geometry after DC boosts up the shallow refractions, as well as avoiding cumbersome modelling through the thick water column, thus reducing the computation cost by 85 per cent. The inversion result from the new data set shows high-resolution shallow sediment structures and the migration images prove the superiority of the inverted model over a conventional tomography model.