地质学
卷积(计算机科学)
地震反演
计算机科学
环境地质学
地震记录
反演(地质)
算法
地震学
地球物理学
变质岩石学
数学
构造学
机器学习
几何学
方位角
人工神经网络
作者
Isabelle Lecomte,Paul Lubrano Lavadera,Charlotte Botter,Ingrid Anell,Simon John Buckley,Christian Haug Eide,Antonio Grippa,Valentina Mascolo,Sigurd Kjoberg
出处
期刊:First Break
[Wiley]
日期:2016-05-01
卷期号:34 (5)
被引量:58
标识
DOI:10.3997/1365-2397.34.5.84451
摘要
Seismic modelling is necessary to understand elasticwave propagation in the subsurface. Modelling is costeffective and insightful, as long as adequate methods are used. An ideal seismic-modelling strategy is to generate complete synthetic seismograms for realistic earth models, then process them as performed with real seismic data. These complete seismograms are best obtained by full-wavefield (FW) approaches. FW methods are therefore used in extensive benchmarking studies, which may require the joint effort of several institutions owing to high resource costs. Though computer power continues to increase, we are still a long way from applying ideal modelling to all cases where synthetic data is necessary to constrain results. This is especially true in interpretation and sensitivity analysis, where the influence of multiple parameters must be assessed. In many situations, ray-based (RB) methods are suitable alternatives, especially when rays, traveltimes, etc., are useful information for the problem at hand (Gjøystdal et al., 2007). However, standard RB methods do not allow modelling of detailed target structures owing to smoothness requirements. Geoscientists needing seismic modelling of such targets will then often resort to 1D convolution (Lecomte et al., 2015). 1D convolution has been successfully used for decades, and is still the method behind most standard well calibration, seismic inversion, etc. However, its conceptual validity is in reality very limited.
科研通智能强力驱动
Strongly Powered by AbleSci AI