Transfer learning: improving neural network based prediction of earthquake ground shaking for an area with insufficient training data

Abstract

Summary In a recent study we showed that convolutional neural networks (CNNs) applied to network seismic traces can be used for rapid prediction of earthquake peak ground motion intensity measures (IMs) at distant stations using only recordings from stations near the epicenter. The predictions are made without any previous knowledge concerning the earthquake location and magnitude. This approach differs significantly from the standard procedure adopted by earthquake early warning systems (EEWSs) that rely on location and magnitude information. In the previous study, we used 10 s, raw, multistation (39 stations) waveforms for the 2016 earthquake sequence in central Italy for 915 M ≥ 3.0 events (CI dataset). The CI dataset has a large number of spatially concentrated earthquakes and a dense network of stations. In this work, we applied the same CNN model to an area of central western Italy. In our initial application of the technique, we used a dataset consisting of 266 M ≥ 3.0 earthquakes recorded by 39 stations. We found that the CNN model trained using this smaller-sized dataset performed worse compared to the results presented in the previously published study. To counter the lack of data, we explored the adoption of ‘transfer learning’ (TL) methodologies using two approaches: first, by using a pre-trained model built on the CI dataset and, next, by using a pre-trained model built on a different (seismological) problem that has a larger dataset available for training. We show that the use of TL improves the results in terms of outliers, bias, and variability of the residuals between predicted and true IMs values. We also demonstrate that adding knowledge of station relative positions as an additional layer in the neural network improves the results. The improvements achieved through the experiments were demonstrated by the reduction of the number of outliers by 5 per cent, the residuals R median by 39 per cent and their standard deviation by 11 per cent.

Graph neural networks for multivariate time series regression with application to seismic data

“…We perform regression on two datasets recorded by the Italian national seismic network [39,40], described fully in [15,16]. GNNs are an ideal candidate for the analysis of seismic data, since seismic measurements contain (1) an enormous amount of data and (2) sensors that are geographically grounded.…”

“…Therefore, the feature vector of each node now consists of time series features from the convolutional layers and classical node features. This addition of this node metadata has showed to improve performance in [16].…”

“…We compare our work to the model presented by [15]. This baseline model was given the exact same input data except for the graph, i. e., both models received the time series per station for every earthquake combined with the latitude and longitude node features of every station, as proven to be effective in [16]. In addition, we also examine the performance of GAT [17] layers and an adjusted version of [14] for our task.…”

“…We test our proposed model on network-based seismic data, which serve as an intuitive domain where GNN models could be operated due to the naturally geographical grounded sensors. The model is inspired by the work presented in [15] and [16], which functions as our most prominent baseline. In their work, the maximum ground shaking at a set of seismic stations is predicted by tackling this as a regression problem.…”

CREIME—A Convolutional Recurrent Model for Earthquake Identification and Magnitude Estimation

“…The advent of deep learning, coupled with the availability of large volumes of data and affordable computational power in the form of GPUs, have led to state‐of‐the‐art results in image recognition (He et al., 2016; Krizhevsky et al., 2017), speech recognition (Hinton et al., 2012; Mikolov et al., 2011), and natural language processing (Collobert et al., 2011; Peters et al., 2018). In fields such as seismology, which have been data‐intensive since their very origin and are witnessing an exponential increase in the volume of data (Kong et al., 2018), deep learning has proven successful in several tasks such as event detection (Fenner et al., 2022; Li et al., 2018, 2022; Meier et al., 2019; Perol et al., 2018) and phase picking (Li et al., 2021; Liao et al., 2021; Mousavi et al., 2020; Zhou et al., 2019; Zhu & Beroza, 2019), event location characterization (Kuyuk & Susumu, 2018; Panakkat & Adeli, 2009; Perol et al., 2018), first motion polarity detection (Hara et al., 2019; Ross et al., 2018), and ground motion estimation (Datta et al., 2022; Fayaz & Galasso, 2022; Jozinović et al., 2020, 2021) among others.…”