Can Deep Learning Predict Complete Ruptures in Numerical Megathrust Faults?
Can Deep Learning Predict Complete Ruptures in Numerical Megathrust Faults?
复制标题
DOI：
10.1029/2021gl092607
复制
发表时间：
2021-09-28
影响因子：
5.2
通讯作者：
Morgan, Julia
中科院分区：
文献类型：
--
作者：
Blank, David;Morgan, Julia
我们提出了一种植根于最先进的深度学习技术的二元分类模型，用于预测沿数值巨型逆冲断层是否即将发生完全界面破裂。这些模型根据从合成故障系统中获取的标记二维时空输入特征进行训练。我们对比了在三种类型的数据上训练的两个神经网络的性能，以确定每个神经网络的相对预测能力。神经网络能够将即将发生的完全破裂前兆与其他事物区分开来，从而提供相对大小和时间的预测。沿断层的垂直位移表现出相对较好的预测能力。结果证实了先前的定性观察，即前兆变形随即将发生的事件大小而变化，与地震成核的预滑移模型一致。我们提出的方法具有很强的适应性，并且可以在将来进行修改以使用 3D 数据。 通俗易懂的语言摘要 与许多其他自然灾害不同，大地震总是让我们措手不及。尽管进行了多次尝试，但地震预测历史上均未成功。一些人认为这是由于数据匮乏造成的。由于许多地震发生在仪器仪表较差的断层带上，因此很难找到前兆行为的趋势。随着技术进步允许沿着地震断层收集大量数据，这种限制正在迅速改变。深度学习算法非常适合在巨大的数据集中寻找模式。在这项研究中，我们使用计算机模型生成合成地震数据集，并使用深度学习来预测大地震是否即将发生。我们希望这个概念验证模型能够指导未来的数据收集，并最终进行修改以预测真实断层上的大地震。
We propose a binary classification model rooted in state-of-the-art deep learning techniques to predict whether or not complete-interface rupture is imminent along a numerical megathrust fault. The models are trained on labeled 2D space-time input features taken from the synthetic fault system. We contrast the performance of two neural networks trained on three types of data, to determine the relative predictive power of each. The neural networks are able to discriminate imminent complete rupture precursors from everything else, thus providing a relative size and time forecast. Vertical displacements along the fault demonstrate relatively good predictive power. The results confirm previous qualitative observations that precursory deformation scales with upcoming event size, consistent with the preslip model for earthquake nucleation. The methods we propose are adaptable and can be modified to use 3D data in the future.Plain Language Summary Unlike many other natural disasters, big earthquakes always take us by surprise. Despite many attempts, earthquake prediction has been historically unsuccessful. Some argue that this is due to a scarcity of data. Because many earthquakes take place along poorly instrumented fault zones, trends in precursory behavior are hard to find. This limitation is quickly changing, as technological advancements allow for the collection of massive quantities of data along earthquake-generating faults. Deep learning algorithms are well suited to find patterns in huge data sets. In this study, we use computer models to generate synthetic earthquake data sets and use deep learning to predict whether or not a big earthquake is imminent. We hope that this proof-of-concept model can guide future data collection and eventually be modified to forecast big earthquakes on real faults.